From 4dba5f546a7d04d7087c846ecfdb895bbc6fa874 Mon Sep 17 00:00:00 2001 From: hcj <1224667075@qq.com> Date: Fri, 6 Feb 2026 10:16:11 +0800 Subject: [PATCH] =?utf8?q?=E6=8F=90=E4=BA=A4mqtt=E7=9B=B8=E5=85=B3?= =?utf8?q?=E5=86=85=E5=AE=B9?= MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit --- .../brandy/brandy_standard/service_init.c | 714 ++-- .../config/target_config/brandy/config.py | 3 + .../middleware/utils/fs/vfs/CMakeLists.txt | 1 + code/sdk/zhihui/MQTT/MQTT_server_mian.c | 2 +- code/sdk/zhihui/MQTT/uapi_mqtt_at.c | 21 +- .../zhihui/leiniao/imsl_ln/LN_MQTT/ln_UDP.c | 283 +- .../leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.c | 448 ++- .../leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.h | 13 +- .../leiniao/imsl_ln/LN_MQTT/ln_mqtt_api_at.c | 2 +- code/sdk/zhihui/leiniao/imsl_ln/ln_api.c | 7 + .../zhihui/leiniao/imsl_ln/ln_api_websocket.c | 28 + .../test_opus_frame_to_pcm/CMakeLists.txt | 14 +- .../test_opus_frame_to_pcm/ENCODER_SUPPORT.md | 160 + .../zhihui/test_opus_frame_to_pcm/Makefile | 54 +- .../test_opus_frame_to_pcm/Makefile.libopus | 116 +- .../celt/celt_encoder.c | 3161 ++++++++++++++++ .../copy_encoder_sources.ps1 | 195 + .../copy_encoder_sources.sh | 166 + .../copy_missing_headers.sh | 23 + .../include/opus_frame_to_pcm.h | 36 + .../test_opus_frame_to_pcm/silk/NLSF_VQ.c | 76 + .../silk/NLSF_del_dec_quant.c | 215 ++ .../test_opus_frame_to_pcm/silk/NLSF_encode.c | 124 + .../zhihui/test_opus_frame_to_pcm/silk/NSQ.c | 437 +++ .../test_opus_frame_to_pcm/silk/NSQ_del_dec.c | 730 ++++ .../zhihui/test_opus_frame_to_pcm/silk/VAD.c | 360 ++ .../test_opus_frame_to_pcm/silk/VQ_WMat_EC.c | 131 + .../silk/ana_filt_bank_1.c | 74 + .../silk/check_control_input.c | 109 + .../test_opus_frame_to_pcm/silk/control_SNR.c | 113 + .../silk/control_audio_bandwidth.c | 132 + .../silk/control_codec.c | 423 +++ .../test_opus_frame_to_pcm/silk/enc_API.c | 601 +++ .../silk/encode_indices.c | 181 + .../silk/encode_pulses.c | 206 + .../silk/fixed/LTP_analysis_filter_FIX.c | 89 + .../silk/fixed/LTP_scale_ctrl_FIX.c | 58 + .../silk/fixed/apply_sine_window_FIX.c | 101 + .../silk/fixed/autocorr_FIX.c | 48 + .../silk/fixed/burg_modified_FIX.c | 280 ++ .../silk/fixed/burg_modified_FIX_sse4_1.c | 400 ++ .../silk/fixed/corrMatrix_FIX.c | 149 + .../silk/fixed/encode_frame_FIX.c | 454 +++ .../silk/fixed/find_LPC_FIX.c | 151 + .../silk/fixed/find_LTP_FIX.c | 99 + .../silk/fixed/find_pitch_lags_FIX.c | 143 + .../silk/fixed/find_pred_coefs_FIX.c | 147 + .../silk/fixed/k2a_FIX.c | 54 + .../silk/fixed/k2a_Q16_FIX.c | 54 + .../silk/fixed/main_FIX.h | 233 ++ .../silk/fixed/noise_shape_analysis_FIX.c | 400 ++ .../silk/fixed/pitch_analysis_core_FIX.c | 721 ++++ .../silk/fixed/process_gains_FIX.c | 117 + .../silk/fixed/regularize_correlations_FIX.c | 47 + .../silk/fixed/residual_energy16_FIX.c | 103 + .../silk/fixed/residual_energy_FIX.c | 98 + .../silk/fixed/schur64_FIX.c | 93 + .../silk/fixed/schur_FIX.c | 107 + .../silk/fixed/structs_FIX.h | 109 + .../silk/fixed/vector_ops_FIX.c | 102 + .../silk/fixed/vector_ops_FIX_sse4_1.c | 93 + .../silk/fixed/warped_autocorrelation_FIX.c | 92 + .../fixed/warped_autocorrelation_FIX_arm.h | 68 + .../fixed/warped_autocorrelation_FIX_mipsr1.h | 189 + .../warped_autocorrelation_FIX_neon_intr.c | 265 ++ .../silk/float/SigProc_FLP.h | 197 + .../silk/float/main_FLP.h | 282 ++ .../silk/float/structs_FLP.h | 105 + .../silk/init_encoder.c | 68 + .../silk/process_NLSFs.c | 107 + .../silk/quant_LTP_gains.c | 132 + .../silk/stereo_LR_to_MS.c | 229 ++ .../silk/stereo_encode_pred.c | 62 + .../silk/stereo_quant_pred.c | 73 + .../src/opus_pcm_to_frame.c | 228 ++ .../test_opus_frame_to_pcm/src_lib/analysis.c | 982 +++++ .../test_opus_frame_to_pcm/src_lib/mlp.c | 131 + .../test_opus_frame_to_pcm/src_lib/mlp_data.c | 671 ++++ .../src_lib/opus_encoder.c | 3365 +++++++++++++++++ .../src_lib/opus_multistream_encoder.c | 1363 +++++++ .../src_lib/opus_projection_encoder.c | 521 +++ 81 files changed, 22433 insertions(+), 506 deletions(-) create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/ENCODER_SUPPORT.md create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/celt/celt_encoder.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.ps1 create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.sh create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/copy_missing_headers.sh create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_VQ.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_del_dec_quant.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_encode.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ_del_dec.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/VAD.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/VQ_WMat_EC.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/ana_filt_bank_1.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/check_control_input.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_SNR.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_audio_bandwidth.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_codec.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/enc_API.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_indices.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_pulses.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_analysis_filter_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_scale_ctrl_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/apply_sine_window_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/autocorr_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX_sse4_1.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/corrMatrix_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/encode_frame_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LPC_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LTP_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pitch_lags_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pred_coefs_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_Q16_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/main_FIX.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/noise_shape_analysis_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/pitch_analysis_core_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/process_gains_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/regularize_correlations_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy16_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur64_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/structs_FIX.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX_sse4_1.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_arm.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_mipsr1.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_neon_intr.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/SigProc_FLP.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/main_FLP.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/structs_FLP.h create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/init_encoder.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/process_NLSFs.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/quant_LTP_gains.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_LR_to_MS.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_encode_pred.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_quant_pred.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src/opus_pcm_to_frame.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/analysis.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/mlp.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/mlp_data.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_encoder.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_multistream_encoder.c create mode 100644 code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_projection_encoder.c diff --git a/code/sdk/application/brandy/brandy_standard/service_init.c b/code/sdk/application/brandy/brandy_standard/service_init.c index 6ed3503d3..5061be55c 100644 --- a/code/sdk/application/brandy/brandy_standard/service_init.c +++ b/code/sdk/application/brandy/brandy_standard/service_init.c @@ -1,346 +1,368 @@ - -/* - * Copyright (c) CompanyNameMagicTag 2021-2022. All rights reserved. - * Description: Application core board init function for standard - * Author: - * Create: - */ - -#ifdef __LITEOS__ -#include "partition.h" -#endif -#include "pm.h" -#include "cpu_utils.h" -#include "log_common.h" -#include "log_uart.h" -#include "exception.h" -#include "log_oml_exception.h" -#ifdef HSO_SUPPORT -#include "dfx_system_init.h" -#endif -#include "debug_print.h" -#ifdef SUPPORT_AT_GRAPHIC -#include "graphic_cmd_manager.h" -#endif -#ifdef SUPPORT_AT_JSINSTALL -#include "ohosfwk_at_process.h" -#endif -#ifdef SUPPORT_AT_APP -#include "app_cmd_manager.h" -#endif -#ifdef SUPPORT_AT_AUDIO -#include "audio_cmd_manager.h" -#endif -#ifdef SUPPORT_AT_BT_MANAGER -#include "bt_manager_cmd_manager.h" -#endif -#ifdef SUPPORT_AT_MEDIA -#include "media_cmd_manager.h" -#endif -#ifdef SUPPORT_AT_PPG -#include "ppg_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_VIBRATOR -#include "vibrator_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_HX3507 -#include "hx3507_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_BATTERY -#include "battery_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_LIS2DOC -#include "lis2doc_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_CYWEE -#include "cywee_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_MAG -#include "mag_at_cmd.h" -#endif - -#ifdef SUPPORT_AT_CUSTOMUTILS -#include "custom_utils_at_cmd.h" -#endif -#ifdef SUPPORT_AT_CAT1 -#include "cat1_at_cmd.h" -#endif -#ifdef SUPPORT_AT_PPP -#include "ppp_at_cmd.h" -#endif -#include "log_file.h" -#ifdef CONFIG_OTA_UPDATE_SUPPORT -#include "upg.h" -#endif -#ifdef SUPPORT_AT_HX3508 -#include "hx3508_at_cmd.h" -#endif -#ifdef SUPPORT_AT_LQBP002TA -#include "lqbp_at_cmd.h" -#endif -#ifdef SUPPORT_AT_CHARGE -#include "charge_at_cmd.h" -#endif -#ifdef SUPPORT_AT_CROWN -#include "crown_at_cmd.h" -#endif -#ifdef SUPPORT_AT_MIXO -#include "mixo_at_cmd.h" -#endif -#ifdef SUPPORT_AT_GYRO -#include "gyro_at_cmd.h" -#endif -#ifdef SUPPORT_AT_BP -#include "bp_at_cmd.h" -#endif -#ifdef SUPPORT_AT_LEINIAO_API -#include "leiniao_api_at.h" -#endif -#ifdef SUPPORT_ZHIHUI_AI_SERVICE -#include "ai_at_cmd.h" -#endif -#ifdef SUPPORT_ZHIHUI_AI_OLD -#include "ai_old_at_cmd.h" -#endif - -#ifdef SUPPORT_ZHIHUI_AI_NEW -#include "ai_V2_5_at_cmd.h" -#endif - -#ifdef ENABLE_VAD_SERVER -#include "imsl_at_cmd.h" -#endif - -#ifdef SUPPORT_ZHIHUI_DMP -#include "dmp_api_at.h" -#endif - -static void register_at_service(void) -{ -#ifdef SUPPORT_AT_GRAPHIC - uint32_t graphic_cmd_count; - at_cmd_table_t *graphic_table = get_graphic_cmd_table(&graphic_cmd_count); - at_cmd_add_moudle(graphic_table, graphic_cmd_count); -#endif -#ifdef SUPPORT_AT_JSINSTALL - uint32_t ohosfwk_cmd_count; - at_cmd_table_t *ohosfwk_table = get_ohosfwk_cmd_table(&ohosfwk_cmd_count); - at_cmd_add_moudle(ohosfwk_table, ohosfwk_cmd_count); -#endif -#ifdef SUPPORT_AT_APP - uint32_t app_cmd_count; - at_cmd_table_t *app_table = get_app_cmd_table(&app_cmd_count); - at_cmd_add_moudle(app_table, app_cmd_count); -#endif -#ifdef SUPPORT_AT_AUDIO - uint32_t audio_cmd_count; - at_cmd_table_t *audio_table = get_audio_cmd_table(&audio_cmd_count); - at_cmd_add_moudle(audio_table, audio_cmd_count); -#endif -#ifdef SUPPORT_AT_BT_MANAGER - uint32_t bt_manager_cmd_count; - at_cmd_table_t *bt_manager_table = get_bt_manager_cmd_table(&bt_manager_cmd_count); - at_cmd_add_moudle(bt_manager_table, bt_manager_cmd_count); -#endif -#ifdef SUPPORT_AT_MEDIA - uint32_t media_cmd_count; - at_cmd_table_t *media_table = get_media_cmd_table(&media_cmd_count); - at_cmd_add_moudle(media_table, media_cmd_count); -#endif - -#ifdef SUPPORT_AT_PPG - uint32_t hx3695_cmd_count; - at_cmd_table_t *hx3695_table = get_ppg_cmd_table(&hx3695_cmd_count); - at_cmd_add_moudle(hx3695_table, hx3695_cmd_count); -#endif - -#ifdef SUPPORT_AT_VIBRATOR - uint32_t vibrator_cmd_count; - at_cmd_table_t *vibrator_table = get_vibrator_cmd_table(&vibrator_cmd_count); - at_cmd_add_moudle(vibrator_table, vibrator_cmd_count); -#endif - -#ifdef SUPPORT_AT_HX3507 - uint32_t hx3507_cmd_count; - at_cmd_table_t *hx3507_table = get_hx3507_cmd_table(&hx3507_cmd_count); - at_cmd_add_moudle(hx3507_table, hx3507_cmd_count); -#endif - -#ifdef SUPPORT_AT_BATTERY - uint32_t battery_cmd_count; - at_cmd_table_t *battery_table = get_battery_cmd_table(&battery_cmd_count); - at_cmd_add_moudle(battery_table, battery_cmd_count); -#endif - -#ifdef SUPPORT_AT_CUSTOMUTILS - uint32_t custom_utils_cmd_count; - at_cmd_table_t *custom_utils_table = get_custom_utils_cmd_table(&custom_utils_cmd_count); - at_cmd_add_moudle(custom_utils_table, custom_utils_cmd_count); -#endif -#ifdef SUPPORT_AT_CAT1 - uint32_t cat1_cmd_count; - at_cmd_table_t *cat1_table = get_cat1_cmd_table(&cat1_cmd_count); - at_cmd_add_moudle(cat1_table, cat1_cmd_count); -#endif - -#ifdef SUPPORT_AT_PPP - uint32_t ppp_cmd_count; - at_cmd_table_t *ppp_table = get_ppp_cmd_table(&ppp_cmd_count); - at_cmd_add_moudle(ppp_table, ppp_cmd_count); -#endif - -#ifdef SUPPORT_ZHIHUI_AI_OLD - uint32_t ai_old_cmd_count; - at_cmd_table_t *ai_old_table = get_ai_old_cmd_table(&ai_old_cmd_count); - at_cmd_add_moudle(ai_old_table, ai_old_cmd_count); -#endif - -#ifdef SUPPORT_ZHIHUI_AI_NEW - uint32_t ai_new_cmd_count; - at_cmd_table_t *ai_new_table = get_ai_2_5_cmd_table(&ai_new_cmd_count); - at_cmd_add_moudle(ai_new_table, ai_new_cmd_count); -#endif - -#ifdef SUPPORT_AT_LIS2DOC - uint32_t lis2doc_cmd_count; - at_cmd_table_t *lis2doc_table = get_lis2doc_cmd_table(&lis2doc_cmd_count); - at_cmd_add_moudle(lis2doc_table, lis2doc_cmd_count); -#endif - -#ifdef SUPPORT_AT_CYWEE - uint32_t cywee_cmd_count; - at_cmd_table_t *cywee_table = get_cywee_cmd_table(&cywee_cmd_count); - at_cmd_add_moudle(cywee_table, cywee_cmd_count); -#endif - -#ifdef SUPPORT_AT_MAG - uint32_t mag_cmd_count; - at_cmd_table_t *mag_table = get_mag_cmd_table(&mag_cmd_count); - at_cmd_add_moudle(mag_table, mag_cmd_count); -#endif - -#ifdef SUPPORT_AT_RTCTEST - uint32_t rtc_cmd_count; - at_cmd_table_t *rtc_table = get_rtc_cmd_table(&rtc_cmd_count); - at_cmd_add_moudle(rtc_table, rtc_cmd_count); -#endif -#ifdef SUPPORT_AT_HX3508 - uint32_t hx3508_cmd_count; - at_cmd_table_t *hx3508_table = get_hx3508_cmd_table(&hx3508_cmd_count); - at_cmd_add_moudle(hx3508_table, hx3508_cmd_count); -#endif -#ifdef SUPPORT_AT_LQBP002TA - uint32_t lqbp002ta_cmd_count; - at_cmd_table_t *lqbp002ta_table = get_lqbp002ta_cmd_table(&lqbp002ta_cmd_count); - at_cmd_add_moudle(lqbp002ta_table, lqbp002ta_cmd_count); -#endif -#ifdef SUPPORT_AT_CHARGE - uint32_t charge_cmd_count; - at_cmd_table_t *charge_table = get_charge_cmd_table(&charge_cmd_count); - at_cmd_add_moudle(charge_table, charge_cmd_count); -#endif -#ifdef SUPPORT_AT_CROWN - uint32_t crown_cmd_count; - at_cmd_table_t *crown_table = get_crown_cmd_table(&crown_cmd_count); - at_cmd_add_moudle(crown_table, crown_cmd_count); -#endif -#ifdef SUPPORT_AT_MIXO - uint32_t mixo_cmd_count; - at_cmd_table_t *mixo_table = get_mixo_cmd_table(&mixo_cmd_count); - at_cmd_add_moudle(mixo_table, mixo_cmd_count); -#endif -#ifdef SUPPORT_AT_GYRO - uint32_t gyro_cmd_count; - at_cmd_table_t *gyro_table = get_gyro_cmd_table(&gyro_cmd_count); - at_cmd_add_moudle(gyro_table, gyro_cmd_count); -#endif -#ifdef SUPPORT_AT_BP - uint32_t bp_cmd_count; - at_cmd_table_t *bp_table = get_bp_cmd_table(&bp_cmd_count); - at_cmd_add_moudle(bp_table, bp_cmd_count); -#endif -#ifdef SUPPORT_AT_LEINIAO_API - uint32_t leiniao_api_cmd_count; - at_cmd_table_t *leiniao_api_table = get_leiniao_api_cmd_table(&leiniao_api_cmd_count); - at_cmd_add_moudle(leiniao_api_table, leiniao_api_cmd_count); -#endif - -#ifdef SUPPORT_ZHIHUI_AI_SERVICE - uint32_t ai_cmd_count; - at_cmd_table_t *ai_table = get_ai_cmd_table(&ai_cmd_count); - at_cmd_add_moudle(ai_table, ai_cmd_count); -#endif - -#ifdef ENABLE_VAD_SERVER - uint32_t imsl_vad_cmd_count; - at_cmd_table_t *imsl_vad_table = get_imsl_vad_at_cmd_table(&imsl_vad_cmd_count); - at_cmd_add_moudle(imsl_vad_table, imsl_vad_cmd_count); -#endif - - -#ifdef SUPPORT_ZHIHUI_DMP - uint32_t dmp_api_cmd_count; - at_cmd_table_t *dmp_api_table = get_dmp_api_cmd_table(&dmp_api_cmd_count); - at_cmd_add_moudle(dmp_api_table, dmp_api_cmd_count); -#endif - -} - -#ifdef CONFIG_OTA_UPDATE_SUPPORT -static void upg_putc(const char c) -{ - printf("%c", c); -} -#endif - -void service_init(void) -{ -#if defined(FREERTOS_TEST) - // freertos demo test - extern int main_full(void); - (void)main_full(); - for (;;) { - } -#endif - - /* dfx init step 1: log init */ - log_init(); - log_init_after_rtos(); - log_uart_init_after_rtos(); - - cpu_utils_init(); - hal_register_exception_dump_callback(log_exception_dump); - uapi_partition_init(); - PRINT("partition init success"NEWLINE); - -#ifdef CONFIG_OTA_UPDATE_SUPPORT - upg_func_t upg_func = {0}; - upg_func.malloc = malloc; - upg_func.free = free; - upg_func.serial_putc = upg_putc; - (void)uapi_upg_init(&upg_func); - PRINT("upg init success\r\n"); -#endif - crypto_drv_init(); - uapi_nv_init(); -#ifdef SUPPORT_AT_BT_MANAGER - set_calibration_from_nv(); -#endif - /* dfx init step 2: dfx init */ -#ifdef HSO_SUPPORT - dfx_system_init(); -#endif - /* power manager init */ - uapi_pm_init(); -#ifdef SUPPORT_POWER_MANAGER - uapi_power_display_init(); -#endif - - register_at_service(); -} + +/* + * Copyright (c) CompanyNameMagicTag 2021-2022. All rights reserved. + * Description: Application core board init function for standard + * Author: + * Create: + */ + +#ifdef __LITEOS__ +#include "partition.h" +#endif +#include "pm.h" +#include "cpu_utils.h" +#include "log_common.h" +#include "log_uart.h" +#include "exception.h" +#include "log_oml_exception.h" +#ifdef HSO_SUPPORT +#include "dfx_system_init.h" +#endif +#include "debug_print.h" +#ifdef SUPPORT_AT_GRAPHIC +#include "graphic_cmd_manager.h" +#endif +#ifdef SUPPORT_AT_JSINSTALL +#include "ohosfwk_at_process.h" +#endif +#ifdef SUPPORT_AT_APP +#include "app_cmd_manager.h" +#endif +#ifdef SUPPORT_AT_AUDIO +#include "audio_cmd_manager.h" +#endif +#ifdef SUPPORT_AT_BT_MANAGER +#include "bt_manager_cmd_manager.h" +#endif +#ifdef SUPPORT_AT_MEDIA +#include "media_cmd_manager.h" +#endif +#ifdef SUPPORT_AT_PPG +#include "ppg_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_VIBRATOR +#include "vibrator_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_HX3507 +#include "hx3507_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_BATTERY +#include "battery_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_LIS2DOC +#include "lis2doc_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_CYWEE +#include "cywee_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_MAG +#include "mag_at_cmd.h" +#endif + +#ifdef SUPPORT_AT_CUSTOMUTILS +#include "custom_utils_at_cmd.h" +#endif +#ifdef SUPPORT_AT_CAT1 +#include "cat1_at_cmd.h" +#endif +#ifdef SUPPORT_AT_PPP +#include "ppp_at_cmd.h" +#endif +#include "log_file.h" +#ifdef CONFIG_OTA_UPDATE_SUPPORT +#include "upg.h" +#endif +#ifdef SUPPORT_AT_HX3508 +#include "hx3508_at_cmd.h" +#endif +#ifdef SUPPORT_AT_LQBP002TA +#include "lqbp_at_cmd.h" +#endif +#ifdef SUPPORT_AT_CHARGE +#include "charge_at_cmd.h" +#endif +#ifdef SUPPORT_AT_CROWN +#include "crown_at_cmd.h" +#endif +#ifdef SUPPORT_AT_MIXO +#include "mixo_at_cmd.h" +#endif +#ifdef SUPPORT_AT_GYRO +#include "gyro_at_cmd.h" +#endif +#ifdef SUPPORT_AT_BP +#include "bp_at_cmd.h" +#endif +#ifdef SUPPORT_AT_LEINIAO_API +#include "leiniao_api_at.h" +#include "uapi_mqtt_at.h" +#include "ln_mqtt_api_at.h" +#endif +#ifdef SUPPORT_ZHIHUI_AI_SERVICE +#include "ai_at_cmd.h" +#endif +#ifdef SUPPORT_ZHIHUI_AI_OLD +#include "ai_old_at_cmd.h" +#endif + +#ifdef SUPPORT_ZHIHUI_AI_NEW +#include "ai_V2_5_at_cmd.h" +#endif + +#ifdef ENABLE_VAD_SERVER +#include "imsl_at_cmd.h" +#endif + +#ifdef SUPPORT_ZHIHUI_DMP +#include "dmp_api_at.h" +#endif + +#ifdef SUPPORT_AT_AI_API +#include "ai_api_at.h" +#endif + +static void register_at_service(void) +{ +#ifdef SUPPORT_AT_GRAPHIC + uint32_t graphic_cmd_count; + at_cmd_table_t *graphic_table = get_graphic_cmd_table(&graphic_cmd_count); + at_cmd_add_moudle(graphic_table, graphic_cmd_count); +#endif +#ifdef SUPPORT_AT_JSINSTALL + uint32_t ohosfwk_cmd_count; + at_cmd_table_t *ohosfwk_table = get_ohosfwk_cmd_table(&ohosfwk_cmd_count); + at_cmd_add_moudle(ohosfwk_table, ohosfwk_cmd_count); +#endif +#ifdef SUPPORT_AT_APP + uint32_t app_cmd_count; + at_cmd_table_t *app_table = get_app_cmd_table(&app_cmd_count); + at_cmd_add_moudle(app_table, app_cmd_count); +#endif +#ifdef SUPPORT_AT_AUDIO + uint32_t audio_cmd_count; + at_cmd_table_t *audio_table = get_audio_cmd_table(&audio_cmd_count); + at_cmd_add_moudle(audio_table, audio_cmd_count); +#endif +#ifdef SUPPORT_AT_BT_MANAGER + uint32_t bt_manager_cmd_count; + at_cmd_table_t *bt_manager_table = get_bt_manager_cmd_table(&bt_manager_cmd_count); + at_cmd_add_moudle(bt_manager_table, bt_manager_cmd_count); +#endif +#ifdef SUPPORT_AT_MEDIA + uint32_t media_cmd_count; + at_cmd_table_t *media_table = get_media_cmd_table(&media_cmd_count); + at_cmd_add_moudle(media_table, media_cmd_count); +#endif + +#ifdef SUPPORT_AT_PPG + uint32_t hx3695_cmd_count; + at_cmd_table_t *hx3695_table = get_ppg_cmd_table(&hx3695_cmd_count); + at_cmd_add_moudle(hx3695_table, hx3695_cmd_count); +#endif + +#ifdef SUPPORT_AT_VIBRATOR + uint32_t vibrator_cmd_count; + at_cmd_table_t *vibrator_table = get_vibrator_cmd_table(&vibrator_cmd_count); + at_cmd_add_moudle(vibrator_table, vibrator_cmd_count); +#endif + +#ifdef SUPPORT_AT_HX3507 + uint32_t hx3507_cmd_count; + at_cmd_table_t *hx3507_table = get_hx3507_cmd_table(&hx3507_cmd_count); + at_cmd_add_moudle(hx3507_table, hx3507_cmd_count); +#endif + +#ifdef SUPPORT_AT_BATTERY + uint32_t battery_cmd_count; + at_cmd_table_t *battery_table = get_battery_cmd_table(&battery_cmd_count); + at_cmd_add_moudle(battery_table, battery_cmd_count); +#endif + +#ifdef SUPPORT_AT_CUSTOMUTILS + uint32_t custom_utils_cmd_count; + at_cmd_table_t *custom_utils_table = get_custom_utils_cmd_table(&custom_utils_cmd_count); + at_cmd_add_moudle(custom_utils_table, custom_utils_cmd_count); +#endif +#ifdef SUPPORT_AT_CAT1 + uint32_t cat1_cmd_count; + at_cmd_table_t *cat1_table = get_cat1_cmd_table(&cat1_cmd_count); + at_cmd_add_moudle(cat1_table, cat1_cmd_count); +#endif + +#ifdef SUPPORT_AT_PPP + uint32_t ppp_cmd_count; + at_cmd_table_t *ppp_table = get_ppp_cmd_table(&ppp_cmd_count); + at_cmd_add_moudle(ppp_table, ppp_cmd_count); +#endif + +#ifdef SUPPORT_ZHIHUI_AI_OLD + uint32_t ai_old_cmd_count; + at_cmd_table_t *ai_old_table = get_ai_old_cmd_table(&ai_old_cmd_count); + at_cmd_add_moudle(ai_old_table, ai_old_cmd_count); +#endif + +#ifdef SUPPORT_ZHIHUI_AI_NEW + uint32_t ai_new_cmd_count; + at_cmd_table_t *ai_new_table = get_ai_2_5_cmd_table(&ai_new_cmd_count); + at_cmd_add_moudle(ai_new_table, ai_new_cmd_count); +#endif + +#ifdef SUPPORT_AT_LIS2DOC + uint32_t lis2doc_cmd_count; + at_cmd_table_t *lis2doc_table = get_lis2doc_cmd_table(&lis2doc_cmd_count); + at_cmd_add_moudle(lis2doc_table, lis2doc_cmd_count); +#endif + +#ifdef SUPPORT_AT_CYWEE + uint32_t cywee_cmd_count; + at_cmd_table_t *cywee_table = get_cywee_cmd_table(&cywee_cmd_count); + at_cmd_add_moudle(cywee_table, cywee_cmd_count); +#endif + +#ifdef SUPPORT_AT_MAG + uint32_t mag_cmd_count; + at_cmd_table_t *mag_table = get_mag_cmd_table(&mag_cmd_count); + at_cmd_add_moudle(mag_table, mag_cmd_count); +#endif + +#ifdef SUPPORT_AT_RTCTEST + uint32_t rtc_cmd_count; + at_cmd_table_t *rtc_table = get_rtc_cmd_table(&rtc_cmd_count); + at_cmd_add_moudle(rtc_table, rtc_cmd_count); +#endif +#ifdef SUPPORT_AT_HX3508 + uint32_t hx3508_cmd_count; + at_cmd_table_t *hx3508_table = get_hx3508_cmd_table(&hx3508_cmd_count); + at_cmd_add_moudle(hx3508_table, hx3508_cmd_count); +#endif +#ifdef SUPPORT_AT_LQBP002TA + uint32_t lqbp002ta_cmd_count; + at_cmd_table_t *lqbp002ta_table = get_lqbp002ta_cmd_table(&lqbp002ta_cmd_count); + at_cmd_add_moudle(lqbp002ta_table, lqbp002ta_cmd_count); +#endif +#ifdef SUPPORT_AT_CHARGE + uint32_t charge_cmd_count; + at_cmd_table_t *charge_table = get_charge_cmd_table(&charge_cmd_count); + at_cmd_add_moudle(charge_table, charge_cmd_count); +#endif +#ifdef SUPPORT_AT_CROWN + uint32_t crown_cmd_count; + at_cmd_table_t *crown_table = get_crown_cmd_table(&crown_cmd_count); + at_cmd_add_moudle(crown_table, crown_cmd_count); +#endif +#ifdef SUPPORT_AT_MIXO + uint32_t mixo_cmd_count; + at_cmd_table_t *mixo_table = get_mixo_cmd_table(&mixo_cmd_count); + at_cmd_add_moudle(mixo_table, mixo_cmd_count); +#endif +#ifdef SUPPORT_AT_GYRO + uint32_t gyro_cmd_count; + at_cmd_table_t *gyro_table = get_gyro_cmd_table(&gyro_cmd_count); + at_cmd_add_moudle(gyro_table, gyro_cmd_count); +#endif +#ifdef SUPPORT_AT_BP + uint32_t bp_cmd_count; + at_cmd_table_t *bp_table = get_bp_cmd_table(&bp_cmd_count); + at_cmd_add_moudle(bp_table, bp_cmd_count); +#endif +#ifdef SUPPORT_AT_LEINIAO_API + uint32_t leiniao_api_cmd_count; + at_cmd_table_t *leiniao_api_table = get_leiniao_api_cmd_table(&leiniao_api_cmd_count); + at_cmd_add_moudle(leiniao_api_table, leiniao_api_cmd_count); + + uint32_t uapi_mqtt_at_cmd_count; + at_cmd_table_t *uapi_mqtt_at_table = get_uapi_mqtt_at_cmd_table(&uapi_mqtt_at_cmd_count); + at_cmd_add_moudle(uapi_mqtt_at_table, uapi_mqtt_at_cmd_count); +#ifdef SUPPORT_LN_MQTT_SERVICE + uint32_t ln_mqtt_at_cmd_count; + at_cmd_table_t *ln_mqtt_at_table = get_ln_mqtt_at_cmd_table(&ln_mqtt_at_cmd_count); + at_cmd_add_moudle(ln_mqtt_at_table, ln_mqtt_at_cmd_count); +#endif + +#endif + +#ifdef SUPPORT_ZHIHUI_AI_SERVICE + uint32_t ai_cmd_count; + at_cmd_table_t *ai_table = get_ai_cmd_table(&ai_cmd_count); + at_cmd_add_moudle(ai_table, ai_cmd_count); +#endif + +#ifdef ENABLE_VAD_SERVER + uint32_t imsl_vad_cmd_count; + at_cmd_table_t *imsl_vad_table = get_imsl_vad_at_cmd_table(&imsl_vad_cmd_count); + at_cmd_add_moudle(imsl_vad_table, imsl_vad_cmd_count); +#endif + + +#ifdef SUPPORT_ZHIHUI_DMP + uint32_t dmp_api_cmd_count; + at_cmd_table_t *dmp_api_table = get_dmp_api_cmd_table(&dmp_api_cmd_count); + at_cmd_add_moudle(dmp_api_table, dmp_api_cmd_count); +#endif + +// #ifdef SUPPORT_AT_AI_API +// uint32_t ai_api_cmd_count; +// at_cmd_table_t *ai_api_table = get_ai_api_cmd_table(&ai_api_cmd_count); +// at_cmd_add_moudle(ai_api_table, ai_api_cmd_count); +// #endif + +} + +#ifdef CONFIG_OTA_UPDATE_SUPPORT +static void upg_putc(const char c) +{ + printf("%c", c); +} +#endif + +void service_init(void) +{ +#if defined(FREERTOS_TEST) + // freertos demo test + extern int main_full(void); + (void)main_full(); + for (;;) { + } +#endif + + /* dfx init step 1: log init */ + log_init(); + log_init_after_rtos(); + log_uart_init_after_rtos(); + + cpu_utils_init(); + hal_register_exception_dump_callback(log_exception_dump); + uapi_partition_init(); + PRINT("partition init success"NEWLINE); + +#ifdef CONFIG_OTA_UPDATE_SUPPORT + upg_func_t upg_func = {0}; + upg_func.malloc = malloc; + upg_func.free = free; + upg_func.serial_putc = upg_putc; + (void)uapi_upg_init(&upg_func); + PRINT("upg init success\r\n"); +#endif + crypto_drv_init(); + uapi_nv_init(); +#ifdef SUPPORT_AT_BT_MANAGER + set_calibration_from_nv(); +#endif + /* dfx init step 2: dfx init */ +#ifdef HSO_SUPPORT + dfx_system_init(); +#endif + /* power manager init */ + uapi_pm_init(); +#ifdef SUPPORT_POWER_MANAGER + uapi_power_display_init(); +#endif + + register_at_service(); +} diff --git a/code/sdk/build/config/target_config/brandy/config.py b/code/sdk/build/config/target_config/brandy/config.py index 124949c39..605ee074f 100644 --- a/code/sdk/build/config/target_config/brandy/config.py +++ b/code/sdk/build/config/target_config/brandy/config.py @@ -1293,6 +1293,7 @@ target = { # 'CYWEE_MAC_ACTIVE', #'SUPPORT_HID_SERVICE', # 'SUPPORT_BAIDU_MAP', + # 'SUPPORT_LN_MQTT_SERVICE', 'CFG_FLASH_SIZE=16', 'CONFIG_NORFLASH_SIZE=16', 'SINGLE_USER_PARTITION', @@ -1331,6 +1332,7 @@ target = { #cat1 相关配置 #有方需要开启 PPP # "ppp", + # "mqtt_service", #EG810 需要开启 cat1Service 和 cat1Test "cat1Service", @@ -1352,6 +1354,7 @@ target = { #VAD 相关配置 "imsl", "minimp3", + # "opus_frame_to_pcm", # "CustomUIcomponentTool", "nativeabilityfwk", "custom_utils", diff --git a/code/sdk/middleware/utils/fs/vfs/CMakeLists.txt b/code/sdk/middleware/utils/fs/vfs/CMakeLists.txt index 16047700a..163619ef9 100644 --- a/code/sdk/middleware/utils/fs/vfs/CMakeLists.txt +++ b/code/sdk/middleware/utils/fs/vfs/CMakeLists.txt @@ -102,6 +102,7 @@ set(PUBLIC_HEADER ${ROOT_DIR}/kernel/dpal/device/include ${ROOT_DIR}/kernel/osal/include ${FREERTOSTOPDIR}/sys/dev/mmc + ${ROOT_DIR}/open_source/openharmony/third_party/lwip/lwip_adapter/liteos/src/include ) aux_source_directory(${FS_PATH}/yaffs2/os_adapt/adapt_freertos PUBLIC_HEADER) diff --git a/code/sdk/zhihui/MQTT/MQTT_server_mian.c b/code/sdk/zhihui/MQTT/MQTT_server_mian.c index 771ce0a27..9a9bb9567 100644 --- a/code/sdk/zhihui/MQTT/MQTT_server_mian.c +++ b/code/sdk/zhihui/MQTT/MQTT_server_mian.c @@ -144,7 +144,7 @@ static void mqtt_queue_init(void) static void mqtt_connection_status_cb(void *arg, int status) { (void)arg; - MQTT_DEBUG("Connection status callback: %d\n", status); + MQTT_ERROR("Connection status callback: %d\n", status); // 更新连接状态码 g_mqtt_connection_status = status; diff --git a/code/sdk/zhihui/MQTT/uapi_mqtt_at.c b/code/sdk/zhihui/MQTT/uapi_mqtt_at.c index fbd4a7711..b98910bcb 100644 --- a/code/sdk/zhihui/MQTT/uapi_mqtt_at.c +++ b/code/sdk/zhihui/MQTT/uapi_mqtt_at.c @@ -81,6 +81,25 @@ static void mqtt_at_message_cb(void *arg, const char *topic, const uint8_t *data } } +/* OTA-发送设备信息到服务器：AT^LN_OTA_SEND_DEVICE_INFO */ +// static uint32_t ln_ota_send_device_info_at(uint8_t *para, uint32_t paraLen, char *strCommonCmd) +// { +// (void)para; +// (void)paraLen; +// (void)strCommonCmd; + +// printf("[LN][OTA_SEND_DEVICE_INFO]\n"); + +// // 直接调用封装函数，使用默认URL（NULL表示使用默认URL） +// int ret = ln_ota_send_device_info_json(NULL, NULL); +// if (ret == 0) { +// printf("+LN_OTA_SEND_DEVICE_INFO: OK\n"); +// return ERRCODE_SUCC; +// } else { +// printf("+LN_OTA_SEND_DEVICE_INFO: FAIL\n"); +// return ERRCODE_FAIL; +// } +// } /* MQTT-初始化：AT^LN_MQTT_INIT */ static uint32_t uapi_mqtt_init_at(uint8_t *para, uint32_t paraLen, char *strCommonCmd) @@ -177,7 +196,7 @@ static uint32_t uapi_mqtt_AudioCaptureSample_at(uint8_t *para, uint32_t paraLen, (void)paraLen; (void)strCommonCmd; - const char *defaultArgs[6] = {"01", "opus", "16000", "1", "16", "user/userdata/test.ogg"}; + const char *defaultArgs[6] = {"01", "sbc", "16000", "1", "16", "user/userdata/test.sbc"}; int ret = AudioCaptureSample(6, defaultArgs); LOS_Msleep(1000*10); diff --git a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_UDP.c b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_UDP.c index 074c67721..bab9283f5 100644 --- a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_UDP.c +++ b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_UDP.c @@ -22,12 +22,6 @@ #include "opus_frame_to_pcm.h" #include "ai_media.h" -/* 双流接口开关：定义 USE_DUAL_STREAM_INTERFACE 则使用 old_Start_AIVoice 双流接口，否则使用 MQTT_StartRecord */ -/* 可在 CMakeLists.txt 或编译选项中定义，例如：-DUSE_DUAL_STREAM_INTERFACE=1 */ -#ifndef USE_DUAL_STREAM_INTERFACE -#define USE_DUAL_STREAM_INTERFACE 0 -#endif - /* OGG 封装相关前向声明（不透明句柄，避免依赖 ogg.h） */ /* OGG 上下文句柄（不透明指针，实际类型在 ai_media_service_mqtt.cpp 中定义） */ typedef void* MqttOggContextHandle; @@ -107,10 +101,33 @@ static uint32_t g_ogg_out_of_order_count = 0; // 连续错序计数器（用于 #define OGG_MAX_OUT_OF_ORDER_COUNT 5 // 最大连续错序次数，超过后跳过一包 #define AUDIO_SEND_MAX_DATA_SIZE 0x800 // 最大音频数据大小（与 LN_MQTT_RECORD_MAX_FRAME 一致） +/* 音频编码类型：入队时 data 的格式 */ +#define AUDIO_SEND_ENCODING_OPUS 0 // 已是 Opus，直接发送 +#define AUDIO_SEND_ENCODING_PCM 1 // PCM，发送前用 opus_pcm_to_frame_encode_packet 编码为 Opus + +/* PCM 编码默认参数（encoding==PCM 时使用）：输入为 16kHz 单声道、16-bit 小端 */ +#define AUDIO_SEND_PCM_SAMPLE_RATE 16000 /* 采样率 16kHz */ +#define AUDIO_SEND_PCM_CHANNELS 1 /* 单声道 */ +#define AUDIO_SEND_PCM_BITRATE 64000 +#define AUDIO_SEND_PCM_APPLICATION 2049 /* Opus: OPUS_APPLICATION_AUDIO，必须为 2048/2049/2051 之一 */ +#define AUDIO_SEND_PCM_FRAME_MS 60 // 每帧 60ms，切割固定长度再编码 +/* 60ms 对应字节数：sample_rate * frame_ms/1000 * channels * 2(bytes/sample) */ +#define AUDIO_SEND_PCM_FRAME_BYTES ((uint32_t)((AUDIO_SEND_PCM_SAMPLE_RATE) * (AUDIO_SEND_PCM_FRAME_MS) / 1000 * (AUDIO_SEND_PCM_CHANNELS) * 2)) +#define AUDIO_SEND_PCM_FRAME_SAMPLES ((int)((AUDIO_SEND_PCM_SAMPLE_RATE) * (AUDIO_SEND_PCM_FRAME_MS) / 1000)) +/* 动态分配容量：帧数可调以增大缓存（约 6s @ 60ms/帧） */ +#define AUDIO_SEND_PCM_CACHE_ALLOC_FRAMES 100 +#define AUDIO_SEND_PCM_CACHE_ALLOC_SIZE ((AUDIO_SEND_PCM_CACHE_ALLOC_FRAMES) * (AUDIO_SEND_PCM_FRAME_BYTES) + (AUDIO_SEND_MAX_DATA_SIZE)) + +/* PCM 发送缓存：动态分配，新数据追加，满 60ms 切割编码，余量保留 */ +static uint8_t *g_pcm_send_cache = NULL; +static uint32_t g_pcm_send_cache_capacity = 0; /* 已分配容量（字节） */ +static uint32_t g_pcm_send_cache_size = 0; /* 当前有效数据长度 */ + /* 音频数据消息结构体 */ typedef struct { uint8_t *data; // 音频数据指针（动态分配） - uint32_t size; // 数据大小 + uint32_t size; // 数据大小（字节） + uint8_t encoding; // 0=Opus，1=PCM（发送前编码为 Opus） } audio_send_msg_t; static unsigned long g_audio_send_queue = 0xFFFFFFFF; // 消息队列句柄 @@ -1077,6 +1094,28 @@ static int ln_udp_dual_stream_handle_packet(uint32_t sequence, const uint8_t *da } #endif /* USE_DUAL_STREAM_INTERFACE */ +/* 接收线程内固定时间间隔（秒）后重新创建 socket，0 表示不按时间重建 */ +#define UDP_RECV_RECREATE_INTERVAL_SEC 0 + +/** + * @brief 仅创建并配置 UDP socket（不启动线程），供创建与接收线程内重建复用 + * @return socket fd，失败返回 -1 + */ +static int ln_udp_create_socket_fd_only(void) +{ + ln_mqtt_udp_info_t udp_info; + if (ln_mqtt_api_get_udp_info(&udp_info) != 0 || !udp_info.valid) { + return -1; + } + int fd = lwip_socket(AF_INET, SOCK_DGRAM, 0); + if (fd < 0) { + return -1; + } + int opt = 1; + lwip_setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)); + return fd; +} + /** * @brief UDP接收线程 */ @@ -1086,9 +1125,11 @@ static void ln_udp_recv_thread(void *arg) uint8_t recv_buffer[UDP_RECV_BUFFER_SIZE]; struct sockaddr_in server_addr; socklen_t addr_len = sizeof(server_addr); - + uint32_t last_recreate_tick = 0; + uint32_t recreate_interval_ticks = 0; + printf("[LN_UDP] Receive thread started\n"); - + // 获取服务器地址信息 ln_mqtt_udp_info_t udp_info; if (ln_mqtt_api_get_udp_info(&udp_info) != 0 || !udp_info.valid) { @@ -1096,40 +1137,68 @@ static void ln_udp_recv_thread(void *arg) g_udp_thread_running = false; return; } - + // 设置服务器地址 memset(&server_addr, 0, sizeof(server_addr)); server_addr.sin_family = AF_INET; server_addr.sin_port = lwip_htons(udp_info.port); server_addr.sin_addr.s_addr = inet_addr(udp_info.server); - - // printf("[LN_UDP] Ready to receive UDP packets from %s:%d\n", udp_info.server, udp_info.port); - + if (UDP_RECV_RECREATE_INTERVAL_SEC > 0) { + recreate_interval_ticks = (uint32_t)osKernelGetTickFreq() * UDP_RECV_RECREATE_INTERVAL_SEC; + last_recreate_tick = osKernelGetTickCount(); + } + while (g_udp_thread_running && g_udp_socket >= 0) { + printf("g_udp_thread_running %d g_udp_socket %d\r\n",g_udp_thread_running,g_udp_socket); + // 固定时间间隔到达则重新创建 socket + if (UDP_RECV_RECREATE_INTERVAL_SEC > 0 && recreate_interval_ticks > 0) { + uint32_t now = osKernelGetTickCount(); + if ((now - last_recreate_tick) >= recreate_interval_ticks) { + printf("[LN_UDP] Recreate interval reached (%u s), recreating socket\n", (unsigned)UDP_RECV_RECREATE_INTERVAL_SEC); + lwip_close(g_udp_socket); + g_udp_socket = ln_udp_create_socket_fd_only(); + if (g_udp_socket < 0) { + printf("[LN_UDP] Failed to recreate socket: %s\n", strerror(errno)); + break; + } + /* 重新获取 UDP 地址信息（服务端地址/端口可能变化） */ + if (ln_mqtt_api_get_udp_info(&udp_info) != 0 || !udp_info.valid) { + printf("[LN_UDP] UDP info not available after recreate, keep last address\n"); + } else { + memset(&server_addr, 0, sizeof(server_addr)); + server_addr.sin_family = AF_INET; + server_addr.sin_port = lwip_htons(udp_info.port); + server_addr.sin_addr.s_addr = inet_addr(udp_info.server); + addr_len = sizeof(server_addr); + } + last_recreate_tick = osKernelGetTickCount(); + } + } + // 接收UDP数据（阻塞，超时1秒） fd_set read_fds; struct timeval timeout; - + FD_ZERO(&read_fds); FD_SET(g_udp_socket, &read_fds); timeout.tv_sec = 1; timeout.tv_usec = 0; - + int select_ret = lwip_select(g_udp_socket + 1, &read_fds, NULL, NULL, &timeout); if (select_ret > 0 && FD_ISSET(g_udp_socket, &read_fds)) { // 有数据可读 ssize_t recv_len = lwip_recvfrom(g_udp_socket, recv_buffer, sizeof(recv_buffer) - 1, 0, (struct sockaddr *)&server_addr, &addr_len); - + if (recv_len > 0) { - printf("[LN_UDP] Received %d bytes\r\n", + printf("[LN_UDP] Received %d bytes\r\n", (int)recv_len ); - + // 处理接收到的数据（解包、解密） ln_udp_process_received_data(recv_buffer, (uint16_t)recv_len); - + } else if (recv_len == 0) { printf("[LN_UDP] Connection closed by peer\n"); break; @@ -1139,6 +1208,7 @@ static void ln_udp_recv_thread(void *arg) } } } else if (select_ret < 0) { + printf("select_ret %d\r\n",select_ret); if (errno != EINTR) { printf("[LN_UDP] Select error: %s\n", strerror(errno)); break; @@ -1146,7 +1216,7 @@ static void ln_udp_recv_thread(void *arg) } // select_ret == 0 表示超时，继续循环 } - + printf("[LN_UDP] Receive thread exited\n"); g_udp_thread_running = false; } @@ -1170,28 +1240,20 @@ int ln_udp_create_socket(void) } printf("[LN_UDP] Creating UDP socket for server: %s:%d\n", udp_info.server, udp_info.port); - + // 如果socket已存在，先关闭 if (g_udp_socket >= 0) { printf("[LN_UDP] Closing existing UDP socket\n"); ln_udp_close_socket(); } - - // 创建UDP套接字 - g_udp_socket = lwip_socket(AF_INET, SOCK_DGRAM, 0); + + // 创建UDP套接字（复用仅创建 fd 的逻辑） + g_udp_socket = ln_udp_create_socket_fd_only(); if (g_udp_socket < 0) { printf("[LN_UDP] Failed to create UDP socket: %s\n", strerror(errno)); return -1; } - - // 设置socket选项（允许地址重用） - int opt = 1; - lwip_setsockopt(g_udp_socket, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)); - - // 设置为非阻塞模式（可选，用于select超时） - // int flags = lwip_fcntl(g_udp_socket, F_GETFL, 0); - // lwip_fcntl(g_udp_socket, F_SETFL, flags | O_NONBLOCK); - + printf("[LN_UDP] UDP socket created successfully, fd=%d\n", g_udp_socket); // 创建接收线程 @@ -1578,6 +1640,50 @@ static void create_audio_send_queue(void) } } +/** + * @brief 发送单帧音频数据（打包/加密 + 发送） + * @note 复用原有 ln_udp_pack_audio_data 加密逻辑 + * @return 0 成功，-1 失败 + */ +static int ln_udp_send_audio_frame(int sock, const uint8_t *audio_data, uint32_t audio_size) +{ + if (audio_data == NULL || audio_size == 0 || audio_size > AUDIO_SEND_MAX_DATA_SIZE) { + return -1; + } + + // 打包（加密+UDP头） + uint8_t send_buf[UDP_PACKET_HEADER_SIZE + AUDIO_SEND_MAX_DATA_SIZE]; + uint16_t out_len = 0; + if (ln_udp_pack_audio_data(audio_data, (uint16_t)audio_size, send_buf, + (uint16_t)sizeof(send_buf), &out_len) != 0) { + printf("[LN_UDP] Failed to pack audio frame\n"); + return -1; + } + + // 获取UDP信息（用于发送） + ln_mqtt_udp_info_t udp_info; + if (ln_mqtt_api_get_udp_info(&udp_info) != 0 || !udp_info.valid) { + printf("[LN_UDP] UDP info not available, drop audio frame\n"); + return -1; + } + + // 发送到服务器 + struct sockaddr_in server_addr; + memset(&server_addr, 0, sizeof(server_addr)); + server_addr.sin_family = AF_INET; + server_addr.sin_port = lwip_htons(udp_info.port); + server_addr.sin_addr.s_addr = inet_addr(udp_info.server); + + ssize_t sent = lwip_sendto(sock, send_buf, out_len, 0, + (struct sockaddr *)&server_addr, sizeof(server_addr)); + if (sent < 0) { + printf("[LN_UDP] UDP sendto failed: %s\n", strerror(errno)); + return -1; + } + + return 0; +} + /** * @brief 音频数据发送线程 */ @@ -1622,37 +1728,72 @@ static void ln_udp_audio_send_thread(void *arg) continue; } - // 打包（加密+UDP头） - uint8_t send_buf[UDP_PACKET_HEADER_SIZE + AUDIO_SEND_MAX_DATA_SIZE]; - uint16_t out_len = 0; - if (ln_udp_pack_audio_data(msg.data, (uint16_t)msg.size, send_buf, - (uint16_t)sizeof(send_buf), &out_len) != 0) { - printf("[LN_UDP] Failed to pack audio frame\n"); + const uint8_t *send_data = msg.data; + uint32_t send_size = msg.size; + if (msg.encoding == AUDIO_SEND_ENCODING_PCM) { + /* PCM：追加到缓存，按 60ms 切割编码，余量保留 */ + if (g_pcm_send_cache == NULL) { + printf("[LN_UDP] PCM cache not allocated, drop %u bytes\n", msg.size); + free(msg.data); + continue; + } + while (g_pcm_send_cache_size + msg.size > g_pcm_send_cache_capacity && + g_pcm_send_cache_size >= AUDIO_SEND_PCM_FRAME_BYTES) { + /* 缓存满时丢弃最旧一帧，避免持续溢出 */ + g_pcm_send_cache_size -= AUDIO_SEND_PCM_FRAME_BYTES; + memmove(g_pcm_send_cache, g_pcm_send_cache + AUDIO_SEND_PCM_FRAME_BYTES, g_pcm_send_cache_size); + } + if (g_pcm_send_cache_size + msg.size > g_pcm_send_cache_capacity) { + printf("[LN_UDP] PCM cache full, drop %u bytes (cache=%u, cap=%u)\n", msg.size, g_pcm_send_cache_size, g_pcm_send_cache_capacity); + free(msg.data); + continue; + } + memcpy(g_pcm_send_cache + g_pcm_send_cache_size, msg.data, msg.size); + g_pcm_send_cache_size += msg.size; free(msg.data); + msg.data = NULL; + + /* 每次取 60ms 固定长度编码并发送，多余数据留在缓存 */ + uint8_t opus_buf[1500]; + + while (g_pcm_send_cache_size >= AUDIO_SEND_PCM_FRAME_BYTES) { + int encoded_len = 0; + printf("[LN_UDP] PCM send cache size: %d\n", g_pcm_send_cache_size); + for (int i = 0; i < AUDIO_SEND_PCM_FRAME_BYTES; i++) { + printf("%02x ", g_pcm_send_cache[i]); + } + printf("\n"); + + if (opus_pcm_to_frame_encode_packet( + (const short *)g_pcm_send_cache, AUDIO_SEND_PCM_FRAME_SAMPLES, + AUDIO_SEND_PCM_SAMPLE_RATE, AUDIO_SEND_PCM_CHANNELS, + AUDIO_SEND_PCM_BITRATE, AUDIO_SEND_PCM_APPLICATION, + opus_buf, (int)sizeof(opus_buf), &encoded_len) != 0) { + printf("[LN_UDP] PCM to Opus encode failed, drop 60ms frame\n"); + break; + } + printf("[LN_UDP] Opus encoded len: %d\n", encoded_len); + printf("[LN_UDP] Opus encoded data: "); + for (int i = 0; i < encoded_len; i++) { + printf("%02x ", opus_buf[i]); + } + printf("\n"); + if (ln_udp_send_audio_frame(sock, opus_buf, (uint32_t)encoded_len) != 0) { + break; + } + /* 移除已编码的一帧，保留余量 */ + g_pcm_send_cache_size -= AUDIO_SEND_PCM_FRAME_BYTES; + memmove(g_pcm_send_cache, g_pcm_send_cache + AUDIO_SEND_PCM_FRAME_BYTES, g_pcm_send_cache_size); + } continue; } - // 获取UDP信息（用于发送） - ln_mqtt_udp_info_t udp_info; - if (ln_mqtt_api_get_udp_info(&udp_info) != 0 || !udp_info.valid) { - printf("[LN_UDP] UDP info not available, drop audio frame\n"); + (void)send_data; + if (ln_udp_send_audio_frame(sock, send_data, send_size) != 0) { free(msg.data); continue; } - // 发送到服务器 - struct sockaddr_in server_addr; - memset(&server_addr, 0, sizeof(server_addr)); - server_addr.sin_family = AF_INET; - server_addr.sin_port = lwip_htons(udp_info.port); - server_addr.sin_addr.s_addr = inet_addr(udp_info.server); - - ssize_t sent = lwip_sendto(sock, send_buf, out_len, 0, - (struct sockaddr *)&server_addr, sizeof(server_addr)); - if (sent < 0) { - printf("[LN_UDP] UDP sendto failed: %s\n", strerror(errno)); - } - // 释放音频数据内存 free(msg.data); msg.data = NULL; @@ -1668,10 +1809,27 @@ static void ln_udp_audio_send_thread(void *arg) */ int ln_udp_audio_send_init(void) { + // 动态分配 PCM 发送缓存（若未分配或需扩容可在此统一申请） + if (g_pcm_send_cache == NULL) { + g_pcm_send_cache = (uint8_t *)malloc(AUDIO_SEND_PCM_CACHE_ALLOC_SIZE); + if (g_pcm_send_cache == NULL) { + printf("[LN_UDP] Failed to allocate PCM send cache (%u bytes)\n", (unsigned)AUDIO_SEND_PCM_CACHE_ALLOC_SIZE); + return -1; + } + g_pcm_send_cache_capacity = AUDIO_SEND_PCM_CACHE_ALLOC_SIZE; + g_pcm_send_cache_size = 0; + printf("[LN_UDP] PCM send cache allocated: %u bytes\n", g_pcm_send_cache_capacity); + } + // 创建消息队列 create_audio_send_queue(); if (g_audio_send_queue == 0xFFFFFFFF) { printf("[LN_UDP] Failed to create audio send queue\n"); + if (g_pcm_send_cache != NULL) { + free(g_pcm_send_cache); + g_pcm_send_cache = NULL; + g_pcm_send_cache_capacity = 0; + } return -1; } @@ -1680,7 +1838,7 @@ int ln_udp_audio_send_init(void) app_task_definition_ext_t send_task = { .attr = { .name = "ln_udp_audio_send", - .stack_size = 0x2000, + .stack_size = 0x8000, .priority = osPriorityNormal }, .func = ln_udp_audio_send_thread, @@ -1708,6 +1866,12 @@ int ln_udp_audio_send_init(void) */ void ln_udp_audio_send_deinit(void) { + g_pcm_send_cache_size = 0; + g_pcm_send_cache_capacity = 0; + if (g_pcm_send_cache != NULL) { + free(g_pcm_send_cache); + g_pcm_send_cache = NULL; + } // 停止发送线程 if (g_audio_send_thread_running) { printf("[LN_UDP] Stopping audio send thread\n"); @@ -1718,6 +1882,7 @@ void ln_udp_audio_send_deinit(void) audio_send_msg_t wake_msg = {0}; wake_msg.data = NULL; wake_msg.size = 0; + wake_msg.encoding = AUDIO_SEND_ENCODING_OPUS; osal_msg_queue_write_copy(g_audio_send_queue, (void *)&wake_msg, sizeof(audio_send_msg_t), 0); // 等待线程退出（最多等待2秒） @@ -1759,8 +1924,9 @@ void ln_udp_audio_send_deinit(void) * @param size 音频数据大小 * @return 0 成功，-1 失败 * @note 函数内部会动态分配内存并复制数据，发送线程负责释放内存 + * @param encoding AUDIO_SEND_ENCODING_OPUS(0)=已是 Opus；AUDIO_SEND_ENCODING_PCM(1)=PCM，发送前编码为 Opus */ -int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size) +int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size, uint8_t encoding) { if (!data || size == 0 || size > AUDIO_SEND_MAX_DATA_SIZE) { printf("[LN_UDP] Invalid parameters for audio send enqueue\n"); @@ -1784,6 +1950,7 @@ int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size) audio_send_msg_t msg; msg.data = data_copy; msg.size = size; + msg.encoding = (encoding != AUDIO_SEND_ENCODING_PCM) ? AUDIO_SEND_ENCODING_OPUS : AUDIO_SEND_ENCODING_PCM; // 写入队列（非阻塞，如果队列满则丢弃并释放内存） int32_t ret = osal_msg_queue_write_copy(g_audio_send_queue, (void *)&msg, sizeof(audio_send_msg_t), 0); diff --git a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.c b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.c index 6638bd5fa..ecbb17457 100644 --- a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.c +++ b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.c @@ -16,14 +16,26 @@ #include "lwip/sockets.h" #include "lwip/inet.h" #include "lwip/netdb.h" +#include +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER +#include "vad_server.h" +#include "errcode.h" +#include "cmsis_os2.h" +#endif #define LN_MQTT_SEND_TOPIC "device-server-enhanced" +/* 测试性录音数据保存到文件：1=启用，0=禁用。启用后每次开始录音创建新文件，停止时关闭。 */ +#ifndef LN_MQTT_RECORD_SAVE_TO_FILE_TEST +#define LN_MQTT_RECORD_SAVE_TO_FILE_TEST 1 +#endif + /* MQTT API 全局状态 */ static bool g_ln_mqtt_service_started = false; /* UDP信息全局变量 */ static ln_mqtt_udp_info_t g_udp_info = {0}; +static uint8_t g_audio_format_input = SBC; /* hello 消息中的 session_id（用于后续 listen 控制消息） */ static char g_ln_mqtt_session_id[64] = {0}; @@ -32,6 +44,9 @@ static bool g_ln_mqtt_session_id_valid = false; /* 音频初始化状态标志位 */ static volatile bool g_mqtt_audio_initialized = false; +/* 消息处理回调函数（参考 ln_ws_start，可为 NULL） */ +static ws_message_callback_t_ln g_mqtt_message_callback = NULL; + int ln_mqtt_api_send_hello(void); // --- MQTT 录音（来自 ai_media_service_mqtt.cpp，无对应头文件时在此声明） --- @@ -53,11 +68,11 @@ extern int ln_udp_pack_audio_data(const uint8_t *audio_data, uint16_t audio_len, uint8_t *buffer, uint16_t buffer_size, uint16_t *out_len); extern int ln_udp_audio_send_init(void); extern void ln_udp_audio_send_deinit(void); -extern int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size); +extern int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size, uint8_t encoding); #define LN_UDP_PACKET_HEADER_SIZE 16 #define LN_MQTT_RECORD_MAX_FRAME 0x800 -#define LN_MQTT_RECORD_BATCH_SIZE 5 // 每5包数据合并后入队 +#define LN_MQTT_RECORD_BATCH_SIZE 2 // 每5包数据合并后入队 /* 双流接口开关：定义 USE_DUAL_STREAM_INTERFACE 则使用 old_Start_AIVoice 双流接口，否则使用 MQTT_StartRecord */ /* 可在 CMakeLists.txt 或编译选项中定义，例如：-DUSE_DUAL_STREAM_INTERFACE=1 */ @@ -79,6 +94,251 @@ static ln_mqtt_record_cache_t g_record_cache = {0}; static volatile bool g_recording_upload_started = false; #endif /* USE_DUAL_STREAM_INTERFACE */ +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER +/* VAD 服务器状态标志位（参考 ln_api_websocket.c） */ +static volatile bool g_mqtt_vad_server_enabled = false; +static volatile bool g_mqtt_vad_server_initialized = false; +/* VAD 冷却时间（参考 ln_ws，避免频繁触发对话） */ +static uint32_t g_mqtt_vad_cooldown_start_tick = 0; +static volatile bool g_mqtt_vad_cooldown_active = false; +static uint32_t g_mqtt_vad_cooldown_duration_ms = 3000; /* 10秒 */ + +static bool ln_mqtt_is_vad_cooldown_active(void); +static void ln_mqtt_start_vad_cooldown(void); +#endif /* ENABLE_VAD_SERVER */ + +static void ln_mqtt_record_data_cb(const uint8_t* data, uint32_t size, void* userData); + +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER +/** + * @brief 检查 VAD 冷却是否激活（参考 ln_ws_is_vad_cooldown_active） + */ +static bool ln_mqtt_is_vad_cooldown_active(void) +{ + if (!g_mqtt_vad_cooldown_active) { + return false; + } + uint32_t current_tick = osKernelGetTickCount(); + uint32_t elapsed_ms = current_tick - g_mqtt_vad_cooldown_start_tick; + if (elapsed_ms >= g_mqtt_vad_cooldown_duration_ms) { + g_mqtt_vad_cooldown_active = false; + printf("[LN_MQTT_API] VAD cooldown period ended\n"); + return false; + } + return true; +} + +/** + * @brief 开始 VAD 冷却计时（参考 ln_ws_start_vad_cooldown） + */ +static void ln_mqtt_start_vad_cooldown(void) +{ + g_mqtt_vad_cooldown_start_tick = osKernelGetTickCount(); + g_mqtt_vad_cooldown_active = true; + printf("[LN_MQTT_API] VAD cooldown started, duration: %u ms\n", g_mqtt_vad_cooldown_duration_ms); +} + +/** + * @brief VAD 状态回调（参考 vad_status_callback_ln_ws，含冷却机制） + * @param status 1=检测到说话开始，2=检测到说话结束 + */ +static void vad_status_callback_ln_mqtt(int status) +{ + printf("[LN_MQTT_API] VAD status callback: %d\n", status); + if (!g_mqtt_vad_server_enabled) { + return; + } + if (status == 1) { + if (ln_mqtt_is_vad_cooldown_active()) { + printf("[LN_MQTT_API] VAD cooldown active, ignoring start request\n"); + return; + } + (void)ln_mqtt_api_send_listen_state(true); + } else if (status == 2) { + (void)ln_mqtt_api_send_listen_state(false); + /* 对话结束后启动冷却，避免频繁触发 */ + ln_mqtt_start_vad_cooldown(); + } +} +#endif /* ENABLE_VAD_SERVER */ + +/** + * @brief MQTT 音频辅助初始化（参考 ln_ws_audio_init，独立可调用） + * @note 仅在 USE_DUAL_STREAM_INTERFACE 时初始化 old_Start_AIVoice 双流接口 + */ +void ln_mqtt_audio_init(void) +{ +#if USE_DUAL_STREAM_INTERFACE + if (g_mqtt_audio_initialized) { + printf("[LN_MQTT_API] Audio already initialized, skipping duplicate initialization\n"); + return; + } + + printf("[LN_MQTT_API] Initializing audio with old_Start_AIVoice dual stream interface\n"); + + if (AI_IsVoiceServiceRunning()) { + printf("[LN_MQTT_API] AI voice service already running, stopping it first\n"); + StopAIVoice(); + } + + AIVoiceAudioParams inputParams = { + .audioFormat =g_audio_format_input , + .channelCount = 1, + .sampleRate = 16000 + }; + + AIVoiceAudioParams outputParams = { + .audioFormat = PCM, + .channelCount = 1, + .sampleRate = 16000 + }; + +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER + AIVoiceAudioParams audioParams_in_vad = { + .audioFormat = PCM, + .channelCount = 1, + .sampleRate = 16000 + }; +#endif + + int32_t start_ret = old_Start_AIVoice( + (AIVoiceDataCallback)ln_mqtt_record_data_cb, + NULL, + &inputParams, + &outputParams, +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER + NULL, /* VAD 上行流（PCM） */ +#else + NULL, /* 第二个上行流参数（NULL 表示不启用双流） */ +#endif + false + ); + + if (start_ret != 0) { + printf("[LN_MQTT_API] WARNING: old_Start_AIVoice failed: %d\n", (int)start_ret); + return; + } + + g_mqtt_audio_initialized = true; + printf("[LN_MQTT_API] old_Start_AIVoice initialized successfully (SBC, 16kHz, mono)\n"); + +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER + /* 初始化 VAD 服务器（参考 ln_api_websocket.c ln_ws_audio_init） */ + g_mqtt_vad_server_enabled = true; + if (!g_mqtt_vad_server_initialized) { + printf("[LN_MQTT_API] Initializing VAD server...\n"); + errcode_t ret = vad_server_init(vad_status_callback_ln_mqtt); + if (ret == ERRCODE_SUCC) { + g_mqtt_vad_server_initialized = true; + printf("[LN_MQTT_API] VAD server initialized successfully\n"); + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_VAD_INITIALIZED); + } + } else { + printf("[LN_MQTT_API] Failed to initialize VAD server: %d\n", (int)ret); + } + } +#endif /* ENABLE_VAD_SERVER */ + + #ifdef SUPPORT_CAT1_SERVICE + cat1_api_keep_wakeup(1); + cat1_api_wakeup_cat1(1); + + #endif + uapi_cat1_add_cat1_vote(); + +#else + (void)0; /* 非双流模式使用 MQTT_StartRecord，无需预初始化 */ +#endif /* USE_DUAL_STREAM_INTERFACE */ +} + +/** + * @brief MQTT 音频辅助反初始化（参考 ln_ws_audio_deinit） + */ +void ln_mqtt_audio_deinit(void) +{ +#if USE_DUAL_STREAM_INTERFACE + if (!g_mqtt_audio_initialized) { + return; + } + +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER + if (g_mqtt_vad_server_initialized) { + printf("[LN_MQTT_API] Deinitializing VAD server...\n"); + errcode_t ret = vad_server_deinit(); + if (ret == ERRCODE_SUCC) { + g_mqtt_vad_server_initialized = false; + printf("[LN_MQTT_API] VAD server deinitialized successfully\n"); + } else { + printf("[LN_MQTT_API] Failed to deinitialize VAD server: %d\n", (int)ret); + } + } + g_mqtt_vad_server_enabled = false; +#endif /* ENABLE_VAD_SERVER */ + + g_recording_upload_started = false; + if (AI_IsVoiceServiceRunning()) { + (void)StopAIVoice(); + printf("[LN_MQTT_API] AI voice service stopped\n"); + } + g_mqtt_audio_initialized = false; + printf("[LN_MQTT_API] Audio deinitialized\n"); + + #ifdef SUPPORT_CAT1_SERVICE + cat1_api_keep_wakeup(0); + cat1_api_wakeup_cat1(0); + + #endif + uapi_cat1_remove_cat1_vote(); +#else + (void)0; +#endif /* USE_DUAL_STREAM_INTERFACE */ +} + +/** + * @brief 获取 MQTT 音频初始化状态 + */ +bool ln_mqtt_is_audio_initialized(void) +{ + return g_mqtt_audio_initialized; +} + +#if LN_MQTT_RECORD_SAVE_TO_FILE_TEST +static FILE *g_record_test_file = NULL; + +static void ln_mqtt_record_test_file_open(void) +{ + if (g_record_test_file != NULL) { + return; /* 已打开 */ + } + time_t now = time(NULL); + struct tm *tm_info = (now >= 0) ? localtime(&now) : NULL; + char filename[96]; + if (tm_info) { + snprintf(filename, sizeof(filename), "/user/userdata/ln_mqtt_record_%04d%02d%02d_%02d%02d%02d.sbc", + tm_info->tm_year + 1900, tm_info->tm_mon + 1, tm_info->tm_mday, + tm_info->tm_hour, tm_info->tm_min, tm_info->tm_sec); + } else { + snprintf(filename, sizeof(filename), "/user/userdata/ln_mqtt_record_test.raw"); + } + g_record_test_file = fopen(filename, "wb"); + if (g_record_test_file) { + printf("[LN_MQTT_API] Test save file opened: %s\n", filename); + } else { + printf("[LN_MQTT_API] Failed to open test save file: %s\n", filename); + } +} + +static void ln_mqtt_record_test_file_close(void) +{ + if (g_record_test_file) { + fclose(g_record_test_file); + g_record_test_file = NULL; + printf("[LN_MQTT_API] Test save file closed\n"); + } +} +#endif /* LN_MQTT_RECORD_SAVE_TO_FILE_TEST */ + static void ln_mqtt_record_data_cb(const uint8_t* data, uint32_t size, void* userData) { (void)userData; @@ -90,17 +350,55 @@ static void ln_mqtt_record_data_cb(const uint8_t* data, uint32_t size, void* use return; } +#if defined(ENABLE_VAD_SERVER) && ENABLE_VAD_SERVER && USE_DUAL_STREAM_INTERFACE + /* VAD 音频送入（须在 early return 之前，VAD 需持续收流以检测唤醒） */ + if (g_mqtt_vad_server_enabled && g_mqtt_vad_server_initialized && vad_server_is_running() ) { + if(g_audio_format_input == PCM) + { + // printf("vad_server_send_audio size %d\r\n",size); + (void)vad_server_send_audio(VAD_AUDIO_FORMAT_PCM, data, size, false); + } + else if(AI_GetSecondStreamInStatus()) + { + uint8_t *pcm_data=(uint8_t *)malloc(1024*3); + if(pcm_data) + { + uint32_t len=1024*3; + AI_SecondStreamInObtainBuffer(pcm_data, &len); + printf("vad_server_send_audio size %d\r\n",len); + (void)vad_server_send_audio(VAD_AUDIO_FORMAT_PCM, pcm_data, len, false); + free(pcm_data); + } + printf("vad_server_send_audio size %d\r\n",size); + (void)vad_server_send_audio(VAD_AUDIO_FORMAT_PCM, data, size, false); + } + + } +#endif /* ENABLE_VAD_SERVER && USE_DUAL_STREAM_INTERFACE */ + #if USE_DUAL_STREAM_INTERFACE - // 如果录音上传未开始，只缓存数据，不处理上传 + /* 如果录音上传未开始，只做 VAD，不处理上传 */ if (!g_recording_upload_started) { - // 可以在这里缓存数据，等待开始标志位设置后再处理 - // 目前先直接返回，不处理数据 return; } #endif /* USE_DUAL_STREAM_INTERFACE */ printf("ln_mqtt_record_data_cb size %d\r\n",size); +#if LN_MQTT_RECORD_SAVE_TO_FILE_TEST + /* 测试性保存到文件：首次写入时打开，之后追加 */ + if (g_record_test_file == NULL) { + ln_mqtt_record_test_file_open(); + } + if (g_record_test_file != NULL && size > 0) { + size_t n = fwrite(data, 1, (size_t)size, g_record_test_file); + if (n != (size_t)size) { + printf("[LN_MQTT_API] Test file write incomplete: %zu/%u\n", n, (unsigned)size); + } + } +#endif /* LN_MQTT_RECORD_SAVE_TO_FILE_TEST */ + +#ifdef USE_RECORD_CACHE // 如果缓冲区未分配，先分配 if (g_record_cache.buffer == NULL) { uint32_t initial_size = LN_MQTT_RECORD_MAX_FRAME * LN_MQTT_RECORD_BATCH_SIZE; @@ -121,7 +419,7 @@ static void ln_mqtt_record_data_cb(const uint8_t* data, uint32_t size, void* use g_record_cache.current_size, size, g_record_cache.buffer_size); // 如果缓存已满，强制刷新当前缓存 if (g_record_cache.current_size > 0) { - if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size) != 0) { + if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size, AUDIO_SEND_ENCODING_OPUS) != 0) { printf("[LN_MQTT_API] Failed to enqueue cached audio data\n"); } g_record_cache.current_size = 0; @@ -136,7 +434,8 @@ static void ln_mqtt_record_data_cb(const uint8_t* data, uint32_t size, void* use // 如果累积到5包，将缓存数据入队 if (g_record_cache.packet_count >= LN_MQTT_RECORD_BATCH_SIZE) { - if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size) != 0) { + if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size, AUDIO_SEND_ENCODING_OPUS) != 0) + { printf("[LN_MQTT_API] Failed to enqueue batched audio data, drop frame\n"); } else { printf("[LN_MQTT_API] Enqueued %u packets, total size: %u bytes\n", @@ -146,8 +445,16 @@ static void ln_mqtt_record_data_cb(const uint8_t* data, uint32_t size, void* use g_record_cache.current_size = 0; g_record_cache.packet_count = 0; } - - msleep(10); +#else + if (ln_udp_audio_send_enqueue(data, size, AUDIO_SEND_ENCODING_OPUS) != 0) { + printf("[LN_MQTT_API] Failed to enqueue audio data, drop frame\n"); + } else { + printf("[LN_MQTT_API] Enqueued %u bytes of audio data\n", size); + } +#endif +#if USE_DUAL_STREAM_INTERFACE + msleep(50); +#endif } /** @@ -341,6 +648,10 @@ static int ln_mqtt_handle_hello_message(const char *data) } else { printf("[LN_MQTT_API] No UDP information in hello message\n"); } + + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_CONNECTED); + } cJSON_Delete(root); return 0; @@ -396,6 +707,9 @@ static void ln_mqtt_internal_connect_cb(void *arg, int status) } else { + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_DISCONNECTED); + } extern void ln_udp_close_socket(void); ln_udp_close_socket(); } @@ -437,14 +751,20 @@ static void ln_mqtt_internal_subscribe_cb(void *arg, const char *topic, uint8_t /** * @brief 启动MQTT服务（启动线程、自动连接等） + * @param callback 消息处理回调函数，参考 ln_ws_start，可为 NULL * @return 0 成功，-1 失败 */ -int ln_mqtt_api_start_service(void) +int ln_mqtt_api_start_service(ws_message_callback_t_ln callback) { if (g_ln_mqtt_service_started) { WEARABLE_LOGW(WEARABLE_LOG_MODULE_APP, "[LN_MQTT_API] Service already started"); return 0; } + + g_mqtt_message_callback = callback; + if (callback != NULL) { + printf("[LN_MQTT_API] Message callback registered\n"); + } printf("[LN_MQTT_API] Starting MQTT service\n"); @@ -468,58 +788,10 @@ int ln_mqtt_api_start_service(void) printf("[LN_MQTT_API] WARNING: Failed to init audio send queue/thread\n"); // 不影响服务启动，继续执行 } - -#if USE_DUAL_STREAM_INTERFACE - // 使用 old_Start_AIVoice 接口时，在服务初始化时就启动音频 - printf("[LN_MQTT_API] Initializing audio with old_Start_AIVoice dual stream interface\n"); - - // 检查 AI 语音服务是否已经在运行（如果已运行则先停止） - if (AI_IsVoiceServiceRunning()) { - printf("[LN_MQTT_API] AI voice service already running, stopping it first\n"); - StopAIVoice(); - } - - // 设置第二个上行流参数：SBC 格式，16kHz，单声道 - AIVoiceAudioParams secondStreamParams = { - .audioFormat = SBC, // SBC 格式（对应 AudioCodecFormat） - .channelCount = 1, // 单声道 - .sampleRate = 16000 // 16kHz 采样率 - }; - - // 输入流参数（主输入流） - AIVoiceAudioParams inputParams = { - .audioFormat = SBC, - .channelCount = 1, - .sampleRate = 16000 - }; + // 音频辅助初始化（双流模式下启动 old_Start_AIVoice） + // ln_mqtt_audio_init(); - - // 输出流参数（设置为NULL，因为我们只关心录音输入） - AIVoiceAudioParams outputParams = { - .audioFormat = PCM, - .channelCount = 1, - .sampleRate = 16000 - }; - - // 使用 old_Start_AIVoice 启动，传入第二个上行流参数启用双流 - int32_t start_ret = old_Start_AIVoice( - (AIVoiceDataCallback)ln_mqtt_record_data_cb, // 数据回调函数（类型转换） - NULL, // 用户数据 - &inputParams, // 输入流参数（主输入流） - &outputParams, // 输出流参数（NULL使用默认） - NULL, // 第二个上行流参数（启用双流） - false // 不使用OGG封装 - ); - - if (start_ret != 0) { - printf("[LN_MQTT_API] WARNING: old_Start_AIVoice failed during service init: %d\n", (int)start_ret); - // 不影响服务启动，继续执行 - } else { - printf("[LN_MQTT_API] old_Start_AIVoice initialized successfully (SBC, 16kHz, mono)\n"); - } -#endif /* USE_DUAL_STREAM_INTERFACE */ - g_ln_mqtt_service_started = true; printf("[LN_MQTT_API] MQTT service started successfully\n"); @@ -545,7 +817,7 @@ void ln_mqtt_api_stop_service(void) // 如果有剩余数据，先刷新 if (g_record_cache.current_size > 0) { printf("[LN_MQTT_API] Flushing remaining cached audio data before service stop\n"); - if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size) != 0) { + if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size, AUDIO_SEND_ENCODING_OPUS) != 0) { printf("[LN_MQTT_API] Failed to enqueue remaining cached audio data\n"); } } @@ -556,18 +828,15 @@ void ln_mqtt_api_stop_service(void) g_record_cache.packet_count = 0; printf("[LN_MQTT_API] Cache buffer freed\n"); } - -#if USE_DUAL_STREAM_INTERFACE - // 清理双流接口资源 - // 清除录音上传标志位 - g_recording_upload_started = false; - // 停止 AI 语音服务（在服务初始化时已启动） - if (AI_IsVoiceServiceRunning()) { - (void)StopAIVoice(); - printf("[LN_MQTT_API] AI voice service stopped\n"); + + // 音频辅助反初始化（双流模式下停止 AI 语音服务） + ln_mqtt_audio_deinit(); + + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_DISCONNECTED); } -#endif /* USE_DUAL_STREAM_INTERFACE */ - + g_mqtt_message_callback = NULL; + // 停止MQTT主线程（会自动断开连接和清理资源） uapi_mqtt_main_stop(); @@ -724,7 +993,15 @@ int ln_mqtt_api_send_hello(void) cJSON_Delete(root); return -1; } - cJSON_AddStringToObject(audio_params, "format", "sbc"); + if(g_audio_format_input == PCM ) + { + cJSON_AddStringToObject(audio_params, "format", "pcm"); + } + else if(g_audio_format_input == SBC) + { + cJSON_AddStringToObject(audio_params, "format", "sbc"); + } + cJSON_AddNumberToObject(audio_params, "sample_rate", 16000); cJSON_AddNumberToObject(audio_params, "channels", 1); cJSON_AddNumberToObject(audio_params, "frame_duration", 60); @@ -785,6 +1062,11 @@ int ln_mqtt_api_send_listen_state(bool start) return -1; } + if(start==false && g_recording_upload_started==false) + { + start=true; + } + // 构建JSON数据 cJSON *root = cJSON_CreateObject(); if (!root) { @@ -843,6 +1125,9 @@ int ln_mqtt_api_send_listen_state(bool start) // 设置录音上传标志位，开始处理并上传音频数据 g_recording_upload_started = true; + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_AUDIO_RECORDING_START); + } printf("[LN_MQTT_API] Dual stream recording started, upload enabled\n"); #else // 使用原有 MQTT_StartRecord 接口 @@ -856,6 +1141,9 @@ int ln_mqtt_api_send_listen_state(bool start) if (rec_ret != 0) { printf("[LN_MQTT_API] WARNING: MQTT_StartRecord failed: %d\n", (int)rec_ret); } else { + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_AUDIO_RECORDING_START); + } printf("[LN_MQTT_API] MQTT_StartRecord started with SBC format (16kHz, mono)\n"); } #endif /* USE_DUAL_STREAM_INTERFACE */ @@ -864,7 +1152,7 @@ int ln_mqtt_api_send_listen_state(bool start) if (g_record_cache.buffer != NULL && g_record_cache.current_size > 0) { printf("[LN_MQTT_API] Flushing remaining cached audio data: %u bytes, %u packets\n", g_record_cache.current_size, g_record_cache.packet_count); - if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size) != 0) { + if (ln_udp_audio_send_enqueue(g_record_cache.buffer, g_record_cache.current_size, AUDIO_SEND_ENCODING_OPUS) != 0) { printf("[LN_MQTT_API] Failed to enqueue remaining cached audio data\n"); } g_record_cache.current_size = 0; @@ -877,12 +1165,22 @@ int ln_mqtt_api_send_listen_state(bool start) g_record_cache.buffer_size = 0; printf("[LN_MQTT_API] Cache buffer freed\n"); } - + +#if LN_MQTT_RECORD_SAVE_TO_FILE_TEST + ln_mqtt_record_test_file_close(); +#endif /* LN_MQTT_RECORD_SAVE_TO_FILE_TEST */ + #if USE_DUAL_STREAM_INTERFACE // 清除录音上传标志位，停止处理并上传音频数据 g_recording_upload_started = false; + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_AUDIO_RECORDING_END); + } printf("[LN_MQTT_API] Dual stream recording stopped, upload disabled (AI voice service still running)\n"); #else + if (g_mqtt_message_callback != NULL) { + g_mqtt_message_callback(NULL, WS_CMD_AUDIO_RECORDING_END); + } // 停止原有 MQTT_StartRecord 接口 (void)MQTT_StopRecord(); #endif /* USE_DUAL_STREAM_INTERFACE */ diff --git a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.h b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.h index 277d29231..b0518dcd1 100644 --- a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.h +++ b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api.h @@ -4,6 +4,7 @@ #include #include #include "uapi_Mqtt_api.h" +#include "ln_api.h" #ifdef __cplusplus extern "C" { @@ -16,11 +17,12 @@ extern "C" { /** * @brief 启动MQTT服务（启动线程、自动连接等） + * @param callback 消息处理回调函数，参考 ln_ws_start，可为 NULL 使用默认处理 * @note 启动服务时会自动进行初始化，无需单独调用初始化函数 * @note 服务模式下使用内部回调，自动处理连接状态和消息接收 * @return 0 成功，-1 失败 */ -int ln_mqtt_api_start_service(void); +int ln_mqtt_api_start_service(ws_message_callback_t_ln callback); /** * @brief 停止MQTT服务（停止线程、断开连接等） @@ -96,13 +98,18 @@ int ln_udp_audio_send_init(void); */ void ln_udp_audio_send_deinit(void); +/* 音频发送编码类型：与 ln_UDP.c 中 AUDIO_SEND_ENCODING_* 一致 */ +#define AUDIO_SEND_ENCODING_OPUS 0 /* 已是 Opus，直接发送 */ +#define AUDIO_SEND_ENCODING_PCM 1 /* PCM，发送前编码为 Opus */ + /** * @brief 将音频数据放入发送队列（供录音回调函数调用） * @param data 音频数据指针 - * @param size 音频数据大小 + * @param size 音频数据大小（字节） + * @param encoding AUDIO_SEND_ENCODING_OPUS(0)=已是 Opus；AUDIO_SEND_ENCODING_PCM(1)=PCM，发送前编码为 Opus * @return 0 成功，-1 失败 */ -int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size); +int ln_udp_audio_send_enqueue(const uint8_t *data, uint32_t size, uint8_t encoding); #ifdef __cplusplus diff --git a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api_at.c b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api_at.c index d7a899023..862a3723c 100644 --- a/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api_at.c +++ b/code/sdk/zhihui/leiniao/imsl_ln/LN_MQTT/ln_mqtt_api_at.c @@ -150,7 +150,7 @@ static uint32_t ln_mqtt_start_service_at(uint8_t *para, uint32_t paraLen, char * (void)strCommonCmd; printf("[LN_MQTT_AT] Starting MQTT service...\n"); - int ret = ln_mqtt_api_start_service(); + int ret = ln_mqtt_api_start_service(NULL); if (ret == 0) { printf("+LN_MQTT_START_SERVICE: OK\n"); return ERRCODE_SUCC; diff --git a/code/sdk/zhihui/leiniao/imsl_ln/ln_api.c b/code/sdk/zhihui/leiniao/imsl_ln/ln_api.c index 2183140c8..dccb58e58 100644 --- a/code/sdk/zhihui/leiniao/imsl_ln/ln_api.c +++ b/code/sdk/zhihui/leiniao/imsl_ln/ln_api.c @@ -1,4 +1,7 @@ #include "ln_api.h" +#ifdef SUPPORT_LN_MQTT_SERVICE +#include "ln_mqtt_api.h" +#endif /* for https/http client */ #include "netstack/http/include/httpclient.h" /* for HMAC-SHA256 */ @@ -1496,6 +1499,10 @@ errcode_t uapi_ln_ws_start(void) // 雷鸟http服务初始化 errcode_t uapi_ln_http_server_init(void) { + #ifdef SUPPORT_LN_MQTT_SERVICE + ln_mqtt_api_start_service(ln_ws_process_message); + return ERRCODE_SUCC; + #endif printf("uapi_ln_http_server_init !\r\n"); ln_send_taskcfg.create_queue_handle = create_ln_http_send_queue; diff --git a/code/sdk/zhihui/leiniao/imsl_ln/ln_api_websocket.c b/code/sdk/zhihui/leiniao/imsl_ln/ln_api_websocket.c index 96e867fdd..776e47504 100644 --- a/code/sdk/zhihui/leiniao/imsl_ln/ln_api_websocket.c +++ b/code/sdk/zhihui/leiniao/imsl_ln/ln_api_websocket.c @@ -801,6 +801,15 @@ int ln_ws_send_audio_data(const uint8_t *audio_data, uint32_t audio_size, bool f // 开始新的音频聊天 int ln_ws_start_audio_chat(ln_audio_chat_command_t command_type) { + #ifdef SUPPORT_LN_MQTT_SERVICE + + ln_mqtt_api_send_listen_state(true); + + return 0; + + + #else + // 检查音频是否已初始化 if (!g_audio_initialized) { LN_WS_ERROR("Audio not initialized, please call ln_ws_audio_init() first\n"); @@ -845,12 +854,18 @@ int ln_ws_start_audio_chat(ln_audio_chat_command_t command_type) LN_WS_DEBUG("Audio chat started: %s (command: %s)\n", g_audio_chat_ctx.chat_id, command_name); + #endif return 0; } // 结束当前音频聊天 int ln_ws_end_audio_chat(void) { + + #ifdef SUPPORT_LN_MQTT_SERVICE + ln_mqtt_api_send_listen_state(false); + return 0; + #else // 检查音频是否已初始化 if (!g_audio_initialized) { LN_WS_ERROR("Audio not initialized, please call ln_ws_audio_init() first\n"); @@ -874,6 +889,7 @@ int ln_ws_end_audio_chat(void) g_audio_chat_ending = true; return 0; + #endif } // 获取当前聊天ID @@ -1410,6 +1426,11 @@ void vad_status_callback_ln_ws(int status) void ln_ws_audio_init(bool is_vad_server_enabled) { + + #ifdef SUPPORT_LN_MQTT_SERVICE + ln_mqtt_audio_init(); + #else + // is_vad_server_enabled=false; // 检查是否已经初始化 if (g_audio_initialized) { @@ -1511,6 +1532,8 @@ void ln_ws_audio_init(bool is_vad_server_enabled) } } #endif + +#endif } @@ -2009,7 +2032,12 @@ int ln_ws_stop(void) /* 获取当前连接状态 */ int ln_ws_get_connection_state(void) { + #ifdef SUPPORT_LN_MQTT_SERVICE + // return ln_mqtt_api_get_connection_state(); + return WS_STATE_CONNECTED; + #else return g_ws_connected; + #endif } /* 获取连接状态字符串 */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/CMakeLists.txt b/code/sdk/zhihui/test_opus_frame_to_pcm/CMakeLists.txt index 60a1d6aca..c571978ba 100644 --- a/code/sdk/zhihui/test_opus_frame_to_pcm/CMakeLists.txt +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/CMakeLists.txt @@ -9,6 +9,7 @@ set(COMPONENT_NAME "opus_frame_to_pcm") # 应用主程序 set(SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/src/opus_frame_to_pcm.c + ${CMAKE_CURRENT_SOURCE_DIR}/src/opus_pcm_to_frame.c ) # 将 opus 库的 .c 源文件全部链接进来（替代原有静态库） @@ -16,16 +17,22 @@ file(GLOB OPUS_SRC_LIB "${CMAKE_CURRENT_SOURCE_DIR}/src_lib/*.c") file(GLOB OPUS_CELT "${CMAKE_CURRENT_SOURCE_DIR}/celt/*.c") file(GLOB OPUS_SILK "${CMAKE_CURRENT_SOURCE_DIR}/silk/*.c") file(GLOB OPUS_SILK_FLOAT "${CMAKE_CURRENT_SOURCE_DIR}/silk/float/*.c") +file(GLOB OPUS_SILK_FIXED "${CMAKE_CURRENT_SOURCE_DIR}/silk/fixed/*.c") # 仅 x86 平台编译 SIMD 优化源；非 x86（如 ARM/Cortex/RISC-V）不包含，使用 C 通用实现 if(CMAKE_SYSTEM_PROCESSOR MATCHES "[xX]86|i686|AMD64|x86_64") file(GLOB OPUS_CELT_X86 "${CMAKE_CURRENT_SOURCE_DIR}/celt/x86/*.c") file(GLOB OPUS_SILK_X86 "${CMAKE_CURRENT_SOURCE_DIR}/silk/x86/*.c") - list(APPEND SOURCES ${OPUS_SRC_LIB} ${OPUS_CELT} ${OPUS_CELT_X86} ${OPUS_SILK} ${OPUS_SILK_FLOAT} ${OPUS_SILK_X86}) + # x86 使用 fixed 的 C/SSE4.1 实现，排除 ARM NEON 文件 + list(REMOVE_ITEM OPUS_SILK_FIXED "${CMAKE_CURRENT_SOURCE_DIR}/silk/fixed/warped_autocorrelation_FIX_neon_intr.c") + list(APPEND SOURCES ${OPUS_SRC_LIB} ${OPUS_CELT} ${OPUS_CELT_X86} ${OPUS_SILK} ${OPUS_SILK_FLOAT} ${OPUS_SILK_FIXED} ${OPUS_SILK_X86}) else() - # 非 x86 平台：排除 x86 特定的 AVX2 文件 + # 非 x86 平台：排除 float 的 AVX2、fixed 的 SSE4.1/NEON 架构特定文件 list(REMOVE_ITEM OPUS_SILK_FLOAT "${CMAKE_CURRENT_SOURCE_DIR}/silk/float/inner_product_FLP_avx2.c") - list(APPEND SOURCES ${OPUS_SRC_LIB} ${OPUS_CELT} ${OPUS_SILK} ${OPUS_SILK_FLOAT}) + list(REMOVE_ITEM OPUS_SILK_FIXED "${CMAKE_CURRENT_SOURCE_DIR}/silk/fixed/vector_ops_FIX_sse4_1.c") + list(REMOVE_ITEM OPUS_SILK_FIXED "${CMAKE_CURRENT_SOURCE_DIR}/silk/fixed/burg_modified_FIX_sse4_1.c") + list(REMOVE_ITEM OPUS_SILK_FIXED "${CMAKE_CURRENT_SOURCE_DIR}/silk/fixed/warped_autocorrelation_FIX_neon_intr.c") + list(APPEND SOURCES ${OPUS_SRC_LIB} ${OPUS_CELT} ${OPUS_SILK} ${OPUS_SILK_FLOAT} ${OPUS_SILK_FIXED}) endif() # 头文件目录 @@ -38,6 +45,7 @@ set(PRIVATE_HEADER ${CMAKE_CURRENT_SOURCE_DIR}/src_lib ${CMAKE_CURRENT_SOURCE_DIR}/silk ${CMAKE_CURRENT_SOURCE_DIR}/silk/float + ${CMAKE_CURRENT_SOURCE_DIR}/silk/fixed ${CMAKE_CURRENT_SOURCE_DIR}/celt ) diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/ENCODER_SUPPORT.md b/code/sdk/zhihui/test_opus_frame_to_pcm/ENCODER_SUPPORT.md new file mode 100644 index 000000000..baf3c2290 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/ENCODER_SUPPORT.md @@ -0,0 +1,160 @@ +# Opus 编码器支持说明 + +本文档说明如何在 `test_opus_frame_to_pcm` 目录中添加编码器支持。 + +## 概述 + +默认情况下，`test_opus_frame_to_pcm` 目录只包含解码器相关的源码，用于构建最小化的解码库。如果需要编码功能，需要额外复制编码器相关的源文件。 + +## 快速开始 + +### 1. 复制解码器源码（如果还没有） + +首先确保已经复制了解码器源码： + +**Linux/macOS:** +```bash +cd test_opus_frame_to_pcm +./copy_sources.sh +``` + +**Windows PowerShell:** +```powershell +cd test_opus_frame_to_pcm +.\copy_sources.ps1 # 如果有的话，或者使用 Git Bash 运行 copy_sources.sh +``` + +### 2. 复制编码器源码 + +运行编码器源码复制脚本： + +**Linux/macOS:** +```bash +./copy_encoder_sources.sh +``` + +**Windows PowerShell:** +```powershell +.\copy_encoder_sources.ps1 +``` + +### 3. 编译库 + +使用更新后的 Makefile 编译包含编码器的库： + +```bash +make -f Makefile.libopus clean +make -f Makefile.libopus +``` + +这将生成包含编码器和解码器功能的 `libopus_min.a`。 + +### 4. 测试编码功能 + +编译编码工具并测试： + +```bash +make clean +make +``` + +然后测试编码功能： + +```bash +# 编码：PCM -> Opus +./opus_frame_to_pcm -e input.pcm -o output.opus -r 16000 -c 1 -b 64000 -a 1 + +# 或使用独立的编码工具 +./opus_pcm_to_frame input.pcm -o output.opus -r 16000 -c 1 -b 64000 -a 1 +``` + +## 添加的编码器源文件 + +### Opus API 层（6 个文件） +- `src/opus_encoder.c` - Opus 编码器主实现 +- `src/opus_multistream_encoder.c` - 多流编码器 +- `src/opus_projection_encoder.c` - 投影编码器 +- `src/analysis.c` - 编码分析（VAD 和带宽检测） +- `src/mlp.c` - MLP 神经网络 +- `src/mlp_data.c` - MLP 数据 + +### CELT 编码器（1 个文件） +- `celt/celt_encoder.c` - CELT 编码器核心 + +### SILK 编码器（约 30 个文件） +- **核心编码文件（8 个）** + - `silk/enc_API.c` + - `silk/init_encoder.c` + - `silk/encode_indices.c` + - `silk/encode_pulses.c` + - `silk/NLSF_encode.c` + - `silk/NLSF_VQ.c` + - `silk/NLSF_del_dec_quant.c` + - `silk/process_NLSFs.c` + +- **立体声编码（3 个）** + - `silk/stereo_LR_to_MS.c` + - `silk/stereo_encode_pred.c` + - `silk/stereo_quant_pred.c` + +- **NSQ（噪声整形量化）（2 个）** + - `silk/NSQ.c` + - `silk/NSQ_del_dec.c` + +- **控制和分析（8 个）** + - `silk/quant_LTP_gains.c` + - `silk/VQ_WMat_EC.c` + - `silk/VAD.c` + - `silk/control_audio_bandwidth.c` + - `silk/control_SNR.c` + - `silk/control_codec.c` + - `silk/check_control_input.c` + - `silk/ana_filt_bank_1.c` + +- **浮点实现（1 个）** + - `silk/float/encode_frame_FLP.c` + +- **定点实现（约 20 个）** + - `silk/fixed/*.c` 中的所有编码相关文件 + +## 库大小对比 + +- **仅解码器**: 约 500-600 KB +- **编码器+解码器**: 约 800-1000 KB + +编码器会增加约 300-400 KB 的库大小。 + +## 注意事项 + +1. **必须先复制解码器源码**：编码器脚本假设解码器源码已经存在。 + +2. **Makefile 已更新**：`Makefile.libopus` 已经更新以支持编码器源文件。如果只想要解码器，可以注释掉编码器相关的源文件定义。 + +3. **编译时间**：包含编码器后，编译时间会显著增加（约 2-3 倍）。 + +4. **依赖关系**：编码器需要 `entenc.c`（熵编码器），这个文件在解码器构建中已经包含。 + +## 故障排除 + +### 编译错误：找不到编码器源文件 + +确保已经运行了 `copy_encoder_sources.sh` 或 `copy_encoder_sources.ps1`。 + +### 链接错误：未定义的编码器函数 + +检查 `Makefile.libopus` 中是否包含了所有编码器源文件。 + +### 运行时错误：编码失败 + +检查输入 PCM 文件的格式： +- 必须是 16 位 PCM +- 小端序（little-endian） +- 正确的采样率和声道数 + +## 相关文件 + +- `copy_encoder_sources.sh` - Linux/macOS 编码器源码复制脚本 +- `copy_encoder_sources.ps1` - Windows PowerShell 编码器源码复制脚本 +- `Makefile.libopus` - 库构建 Makefile（已更新支持编码器） +- `src/opus_pcm_to_frame.c` - 独立的编码工具 +- `src/opus_frame_to_pcm.c` - 统一的编解码工具（支持 `-e` 选项） diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile b/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile index 88f4c14fe..ca6315f6a 100644 --- a/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile @@ -9,17 +9,20 @@ LDFLAGS = -L. -lopus -lm SRC_DIR = src INCLUDE_DIR = include -# Target executable +# Target executables TARGET = opus_frame_to_pcm +TARGET_ENCODE = opus_pcm_to_frame -# Source file +# Source files SOURCES = $(SRC_DIR)/opus_frame_to_pcm.c +SOURCES_ENCODE = $(SRC_DIR)/opus_pcm_to_frame.c # Object files OBJECTS = $(SOURCES:.c=.o) +OBJECTS_ENCODE = $(SOURCES_ENCODE:.c=.o) # Default target -all: $(TARGET) +all: $(TARGET) $(TARGET_ENCODE) # Build executable LIBOPUS ?= libopus_min.a @@ -28,28 +31,47 @@ $(TARGET): $(OBJECTS) $(LIBOPUS) # Ensure we link against the chosen archive name @if [ "$(LIBOPUS)" != "libopus.a" ]; then \ rm -f libopus.a; \ - ln -s "$(LIBOPUS)" libopus.a; \ + if command -v ln >/dev/null 2>&1; then \ + ln -s "$(LIBOPUS)" libopus.a; \ + else \ + cp "$(LIBOPUS)" libopus.a; \ + fi; \ fi $(CC) $(OBJECTS) -o $(TARGET) $(LDFLAGS) +$(TARGET_ENCODE): $(OBJECTS_ENCODE) $(LIBOPUS) + # Ensure we link against the chosen archive name + @if [ "$(LIBOPUS)" != "libopus.a" ]; then \ + rm -f libopus.a; \ + if command -v ln >/dev/null 2>&1; then \ + ln -s "$(LIBOPUS)" libopus.a; \ + else \ + cp "$(LIBOPUS)" libopus.a; \ + fi; \ + fi + $(CC) $(OBJECTS_ENCODE) -o $(TARGET_ENCODE) $(LDFLAGS) + # Compile source files $(SRC_DIR)/%.o: $(SRC_DIR)/%.c $(CC) $(CFLAGS) -c $< -o $@ -# Check if libopus.a exists +# Check if libopus.a exists and build it if needed $(LIBOPUS): @if [ ! -f $(LIBOPUS) ]; then \ - echo "Error: libopus.a not found!"; \ - echo "Please create $(LIBOPUS) first."; \ - echo "Recommended (PowerShell):"; \ - echo " ./build_min_libopus.ps1"; \ - echo "Fallback: copy the full libopus.a as $(LIBOPUS) (NOT recommended for size validation)."; \ - exit 1; \ + echo "Error: $(LIBOPUS) not found!"; \ + echo "Building $(LIBOPUS) using Makefile.libopus..."; \ + $(MAKE) -f Makefile.libopus $(LIBOPUS) || ( \ + echo "Build failed. Please ensure:"; \ + echo " 1. Source files are copied (run copy_sources.sh)"; \ + echo " 2. Encoder sources are copied if needed (run copy_encoder_sources.sh)"; \ + echo " 3. config.h exists in this directory"; \ + exit 1 \ + ); \ fi # Clean build artifacts clean: - rm -f $(OBJECTS) $(TARGET) + rm -f $(OBJECTS) $(OBJECTS_ENCODE) $(TARGET) $(TARGET_ENCODE) # Test with a sample file (if available) test: $(TARGET) @@ -60,4 +82,10 @@ test: $(TARGET) echo "No test_frame.opus found. Please provide a test file."; \ fi -.PHONY: all clean test +.PHONY: all clean test encode-only decode-only + +# Build only decode tool +decode-only: $(TARGET) + +# Build only encode tool +encode-only: $(TARGET_ENCODE) diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile.libopus b/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile.libopus index 9b16a4bf6..28f89fd14 100644 --- a/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile.libopus +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/Makefile.libopus @@ -1,5 +1,5 @@ -# Makefile for building minimal Opus decoder library (libopus_min.a) -# This builds a decoder-only library from source files +# Makefile for building minimal Opus library (libopus_min.a) +# This builds a library with both encoder and decoder support CC = gcc AR = ar @@ -23,7 +23,7 @@ SRC_LIB_DIR = src_lib CELT_DIR = celt SILK_DIR = silk -# Opus API layer sources (decoder only) +# Opus API layer sources (decoder only, encoder added if available) OPUS_SOURCES = \ $(SRC_LIB_DIR)/opus.c \ $(SRC_LIB_DIR)/opus_decoder.c \ @@ -33,7 +33,21 @@ OPUS_SOURCES = \ $(SRC_LIB_DIR)/repacketizer.c \ $(SRC_LIB_DIR)/mapping_matrix.c -# CELT decoder sources +# Add encoder sources if they exist +ifeq ($(wildcard $(SRC_LIB_DIR)/opus_encoder.c),$(SRC_LIB_DIR)/opus_encoder.c) +OPUS_SOURCES += \ + $(SRC_LIB_DIR)/opus_encoder.c \ + $(SRC_LIB_DIR)/opus_multistream_encoder.c \ + $(SRC_LIB_DIR)/opus_projection_encoder.c \ + $(SRC_LIB_DIR)/analysis.c \ + $(SRC_LIB_DIR)/mlp.c \ + $(SRC_LIB_DIR)/mlp_data.c +ENCODER_SUPPORT = 1 +else +ENCODER_SUPPORT = 0 +endif + +# CELT decoder sources (encoder added if available) CELT_SOURCES = \ $(CELT_DIR)/bands.c \ $(CELT_DIR)/celt.c \ @@ -53,6 +67,11 @@ CELT_SOURCES = \ $(CELT_DIR)/rate.c \ $(CELT_DIR)/vq.c +# Add CELT encoder if available +ifeq ($(wildcard $(CELT_DIR)/celt_encoder.c),$(CELT_DIR)/celt_encoder.c) +CELT_SOURCES += $(CELT_DIR)/celt_encoder.c +endif + # CELT x86 optimization (if available) CELT_X86_SOURCES = \ $(CELT_DIR)/x86/x86cpu.c \ @@ -75,6 +94,20 @@ SILK_CORE_SOURCES = \ $(SILK_DIR)/decode_pulses.c \ $(SILK_DIR)/decoder_set_fs.c +# SILK encoder sources (core) - only if available +SILK_ENCODER_CORE_SOURCES = +ifeq ($(wildcard $(SILK_DIR)/enc_API.c),$(SILK_DIR)/enc_API.c) +SILK_ENCODER_CORE_SOURCES += \ + $(SILK_DIR)/enc_API.c \ + $(SILK_DIR)/init_encoder.c \ + $(SILK_DIR)/encode_indices.c \ + $(SILK_DIR)/encode_pulses.c \ + $(SILK_DIR)/NLSF_encode.c \ + $(SILK_DIR)/NLSF_VQ.c \ + $(SILK_DIR)/NLSF_del_dec_quant.c \ + $(SILK_DIR)/process_NLSFs.c +endif + # SILK signal processing (shared) SILK_SIGNAL_SOURCES = \ $(SILK_DIR)/CNG.c \ @@ -104,12 +137,20 @@ SILK_NLSF_SOURCES = \ $(SILK_DIR)/table_LSF_cos.c \ $(SILK_DIR)/LPC_inv_pred_gain.c -# SILK stereo processing +# SILK stereo processing (decoder, encoder added if available) SILK_STEREO_SOURCES = \ $(SILK_DIR)/stereo_MS_to_LR.c \ $(SILK_DIR)/stereo_decode_pred.c \ $(SILK_DIR)/stereo_find_predictor.c +# Add encoder stereo processing if available +ifeq ($(wildcard $(SILK_DIR)/stereo_LR_to_MS.c),$(SILK_DIR)/stereo_LR_to_MS.c) +SILK_STEREO_SOURCES += \ + $(SILK_DIR)/stereo_LR_to_MS.c \ + $(SILK_DIR)/stereo_encode_pred.c \ + $(SILK_DIR)/stereo_quant_pred.c +endif + # SILK tables SILK_TABLE_SOURCES = \ $(SILK_DIR)/tables_gain.c \ @@ -137,6 +178,28 @@ SILK_HELPER_SOURCES = \ $(SILK_DIR)/sum_sqr_shift.c \ $(SILK_DIR)/gain_quant.c +# SILK encoder NSQ (Noise Shaping Quantization) - only if available +SILK_ENCODER_NSQ_SOURCES = +ifeq ($(wildcard $(SILK_DIR)/NSQ.c),$(SILK_DIR)/NSQ.c) +SILK_ENCODER_NSQ_SOURCES += \ + $(SILK_DIR)/NSQ.c \ + $(SILK_DIR)/NSQ_del_dec.c +endif + +# SILK encoder control and analysis - only if available +SILK_ENCODER_CONTROL_SOURCES = +ifeq ($(wildcard $(SILK_DIR)/VAD.c),$(SILK_DIR)/VAD.c) +SILK_ENCODER_CONTROL_SOURCES += \ + $(SILK_DIR)/quant_LTP_gains.c \ + $(SILK_DIR)/VQ_WMat_EC.c \ + $(SILK_DIR)/VAD.c \ + $(SILK_DIR)/control_audio_bandwidth.c \ + $(SILK_DIR)/control_SNR.c \ + $(SILK_DIR)/control_codec.c \ + $(SILK_DIR)/check_control_input.c \ + $(SILK_DIR)/ana_filt_bank_1.c +endif + # SILK float implementation (default, comment out if using fixed-point) SILK_FLOAT_SOURCES = \ $(SILK_DIR)/float/apply_sine_window_FLP.c \ @@ -167,6 +230,40 @@ SILK_FLOAT_SOURCES = \ $(SILK_DIR)/float/schur_FLP.c \ $(SILK_DIR)/float/sort_FLP.c +# Add encode_frame_FLP.c if available +ifeq ($(wildcard $(SILK_DIR)/float/encode_frame_FLP.c),$(SILK_DIR)/float/encode_frame_FLP.c) +SILK_FLOAT_SOURCES += $(SILK_DIR)/float/encode_frame_FLP.c +endif + +# SILK fixed-point implementation (encoder) - only if available +SILK_FIXED_SOURCES = +ifeq ($(wildcard $(SILK_DIR)/fixed/encode_frame_FIX.c),$(SILK_DIR)/fixed/encode_frame_FIX.c) +SILK_FIXED_SOURCES += \ + $(SILK_DIR)/fixed/apply_sine_window_FIX.c \ + $(SILK_DIR)/fixed/autocorr_FIX.c \ + $(SILK_DIR)/fixed/burg_modified_FIX.c \ + $(SILK_DIR)/fixed/corrMatrix_FIX.c \ + $(SILK_DIR)/fixed/encode_frame_FIX.c \ + $(SILK_DIR)/fixed/find_LPC_FIX.c \ + $(SILK_DIR)/fixed/find_LTP_FIX.c \ + $(SILK_DIR)/fixed/find_pitch_lags_FIX.c \ + $(SILK_DIR)/fixed/find_pred_coefs_FIX.c \ + $(SILK_DIR)/fixed/k2a_FIX.c \ + $(SILK_DIR)/fixed/k2a_Q16_FIX.c \ + $(SILK_DIR)/fixed/LTP_analysis_filter_FIX.c \ + $(SILK_DIR)/fixed/LTP_scale_ctrl_FIX.c \ + $(SILK_DIR)/fixed/noise_shape_analysis_FIX.c \ + $(SILK_DIR)/fixed/pitch_analysis_core_FIX.c \ + $(SILK_DIR)/fixed/process_gains_FIX.c \ + $(SILK_DIR)/fixed/regularize_correlations_FIX.c \ + $(SILK_DIR)/fixed/residual_energy_FIX.c \ + $(SILK_DIR)/fixed/residual_energy16_FIX.c \ + $(SILK_DIR)/fixed/schur_FIX.c \ + $(SILK_DIR)/fixed/schur64_FIX.c \ + $(SILK_DIR)/fixed/vector_ops_FIX.c \ + $(SILK_DIR)/fixed/warped_autocorrelation_FIX.c +endif + # SILK x86 optimization (if available) SILK_X86_SOURCES = \ $(SILK_DIR)/x86/x86_silk_map.c @@ -177,12 +274,16 @@ ALL_SOURCES = \ $(CELT_SOURCES) \ $(CELT_X86_SOURCES) \ $(SILK_CORE_SOURCES) \ + $(SILK_ENCODER_CORE_SOURCES) \ $(SILK_SIGNAL_SOURCES) \ $(SILK_NLSF_SOURCES) \ $(SILK_STEREO_SOURCES) \ $(SILK_TABLE_SOURCES) \ $(SILK_HELPER_SOURCES) \ + $(SILK_ENCODER_NSQ_SOURCES) \ + $(SILK_ENCODER_CONTROL_SOURCES) \ $(SILK_FLOAT_SOURCES) \ + $(SILK_FIXED_SOURCES) \ $(SILK_X86_SOURCES) # Object files @@ -194,6 +295,11 @@ all: $(LIBRARY) # Build library $(LIBRARY): $(OBJECTS) @echo "Creating $(LIBRARY)..." + @if [ "$(ENCODER_SUPPORT)" = "1" ]; then \ + echo "Building with encoder support"; \ + else \ + echo "Building decoder-only library"; \ + fi $(AR) rcs $(LIBRARY) $(OBJECTS) $(RANLIB) $(LIBRARY) @echo "Library created: $(LIBRARY)" diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/celt/celt_encoder.c b/code/sdk/zhihui/test_opus_frame_to_pcm/celt/celt_encoder.c new file mode 100644 index 000000000..ab97ed677 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/celt/celt_encoder.c @@ -0,0 +1,3161 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_ENCODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include +#include "celt_lpc.h" +#include "vq.h" + + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + + +/** Encoder state + @brief Encoder state + */ +struct OpusCustomEncoder { + const OpusCustomMode *mode; /**< Mode used by the encoder */ + int channels; + int stream_channels; + + int force_intra; + int clip; + int disable_pf; + int complexity; + int upsample; + int start, end; + + opus_int32 bitrate; + int vbr; + int signalling; + int constrained_vbr; /* If zero, VBR can do whatever it likes with the rate */ + int loss_rate; + int lsb_depth; + int lfe; + int disable_inv; + int arch; +#ifdef ENABLE_QEXT + int enable_qext; + int qext_scale; +#endif + + /* Everything beyond this point gets cleared on a reset */ +#define ENCODER_RESET_START rng + + opus_uint32 rng; + int spread_decision; + opus_val32 delayedIntra; + int tonal_average; + int lastCodedBands; + int hf_average; + int tapset_decision; + + int prefilter_period; + opus_val16 prefilter_gain; + int prefilter_tapset; +#ifdef RESYNTH + int prefilter_period_old; + opus_val16 prefilter_gain_old; + int prefilter_tapset_old; +#endif + int consec_transient; + AnalysisInfo analysis; + SILKInfo silk_info; + + opus_val32 preemph_memE[2]; + opus_val32 preemph_memD[2]; + + /* VBR-related parameters */ + opus_int32 vbr_reservoir; + opus_int32 vbr_drift; + opus_int32 vbr_offset; + opus_int32 vbr_count; + opus_val32 overlap_max; + opus_val16 stereo_saving; + int intensity; + celt_glog *energy_mask; + celt_glog spec_avg; + +#ifdef RESYNTH +#ifdef ENABLE_QEXT + /* +MAX_PERIOD/2 to make space for overlap */ + celt_sig syn_mem[2][2*DEC_PITCH_BUF_SIZE+MAX_PERIOD]; +#else + /* +MAX_PERIOD/2 to make space for overlap */ + celt_sig syn_mem[2][DEC_PITCH_BUF_SIZE+MAX_PERIOD/2]; +#endif +#endif + + celt_sig in_mem[1]; /* Size = channels*mode->overlap */ + /* celt_sig prefilter_mem[], Size = channels*COMBFILTER_MAXPERIOD */ + /* celt_glog oldBandE[], Size = channels*mode->nbEBands */ + /* celt_glog oldLogE[], Size = channels*mode->nbEBands */ + /* celt_glog oldLogE2[], Size = channels*mode->nbEBands */ + /* celt_glog energyError[], Size = channels*mode->nbEBands */ +}; + +int celt_encoder_get_size(int channels) +{ +#ifdef ENABLE_QEXT + CELTMode *mode = opus_custom_mode_create(96000, 1920, NULL); +#else + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); +#endif + return opus_custom_encoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_get_size(const CELTMode *mode, int channels) +{ + int extra=0; + int size; +#ifdef ENABLE_QEXT + int qext_scale; + extra = channels*NB_QEXT_BANDS*sizeof(celt_glog); + if (mode->Fs == 96000 && (mode->shortMdctSize==240 || mode->shortMdctSize==180)) { + qext_scale = 2; + } else qext_scale = 1; +#endif + size = sizeof(struct CELTEncoder) + + (channels*mode->overlap-1)*sizeof(celt_sig) /* celt_sig in_mem[channels*mode->overlap]; */ + + channels*QEXT_SCALE(COMBFILTER_MAXPERIOD)*sizeof(celt_sig) /* celt_sig prefilter_mem[channels*COMBFILTER_MAXPERIOD]; */ + + 4*channels*mode->nbEBands*sizeof(celt_glog) /* celt_glog oldBandE[channels*mode->nbEBands]; */ + /* celt_glog oldLogE[channels*mode->nbEBands]; */ + /* celt_glog oldLogE2[channels*mode->nbEBands]; */ + /* celt_glog energyError[channels*mode->nbEBands]; */ + + extra; + return size; +} + +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) +CELTEncoder *opus_custom_encoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTEncoder *st = (CELTEncoder *)opus_alloc(opus_custom_encoder_get_size(mode, channels)); + /* init will handle the NULL case */ + ret = opus_custom_encoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_encoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode, + int channels, int arch) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL || mode==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels)); + + st->mode = mode; + st->stream_channels = st->channels = channels; + + st->upsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + st->arch = arch; + + st->constrained_vbr = 1; + st->clip = 1; + + st->bitrate = OPUS_BITRATE_MAX; + st->vbr = 0; + st->force_intra = 0; + st->complexity = 5; + st->lsb_depth=24; + +#ifdef ENABLE_QEXT + if (st->mode->Fs == 96000 && (mode->shortMdctSize==240 || mode->shortMdctSize==180)) st->qext_scale = 2; + else st->qext_scale = 1; +#endif + + opus_custom_encoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) +int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels) +{ + return opus_custom_encoder_init_arch(st, mode, channels, opus_select_arch()); +} +#endif + +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch) +{ + int ret; +#ifdef ENABLE_QEXT + if (sampling_rate==96000) { + st->upsample = 1; + return opus_custom_encoder_init_arch(st, + opus_custom_mode_create(96000, 1920, NULL), channels, arch); + } +#endif + ret = opus_custom_encoder_init_arch(st, + opus_custom_mode_create(48000, 960, NULL), channels, arch); + if (ret != OPUS_OK) + return ret; + st->upsample = resampling_factor(sampling_rate); + return OPUS_OK; +} + +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) +void opus_custom_encoder_destroy(CELTEncoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + + +static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int C, + opus_val16 *tf_estimate, int *tf_chan, int allow_weak_transients, + int *weak_transient, opus_val16 tone_freq, opus_val32 toneishness) +{ + int i; + VARDECL(opus_val16, tmp); + opus_val32 mem0,mem1; + int is_transient = 0; + opus_int32 mask_metric = 0; + int c; + opus_val16 tf_max; + int len2; + /* Forward masking: 6.7 dB/ms. */ +#ifdef FIXED_POINT + int forward_shift = 4; +#else + opus_val16 forward_decay = QCONST16(.0625f,15); +#endif + /* Table of 6*64/x, trained on real data to minimize the average error */ + static const unsigned char inv_table[128] = { + 255,255,156,110, 86, 70, 59, 51, 45, 40, 37, 33, 31, 28, 26, 25, + 23, 22, 21, 20, 19, 18, 17, 16, 16, 15, 15, 14, 13, 13, 12, 12, + 12, 12, 11, 11, 11, 10, 10, 10, 9, 9, 9, 9, 9, 9, 8, 8, + 8, 8, 8, 7, 7, 7, 7, 7, 7, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, + }; + SAVE_STACK; +#ifdef FIXED_POINT + int in_shift = IMAX(0, celt_ilog2(1+celt_maxabs32(in, C*len))-14); +#endif + ALLOC(tmp, len, opus_val16); + + *weak_transient = 0; + /* For lower bitrates, let's be more conservative and have a forward masking + decay of 3.3 dB/ms. This avoids having to code transients at very low + bitrate (mostly for hybrid), which can result in unstable energy and/or + partial collapse. */ + if (allow_weak_transients) + { +#ifdef FIXED_POINT + forward_shift = 5; +#else + forward_decay = QCONST16(.03125f,15); +#endif + } + len2=len/2; + for (c=0;c=0;i--) + { + /* Backward masking: 13.9 dB/ms. */ +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + mem0 = mem0 + PSHR32(SHL32(tmp[i],4)-mem0,3); + tmp[i] = PSHR32(mem0, 4); + maxE = MAX16(maxE, tmp[i]); +#else + mem0 = tmp[i] + 0.875f*mem0; + tmp[i] = 0.125f*mem0; + maxE = MAX16(maxE, 0.125f*mem0); +#endif + } + /*for (i=0;i>1))); +#else + mean = celt_sqrt(mean * maxE*.5*len2); +#endif + /* Inverse of the mean energy in Q15+6 */ + norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1)); + /* Compute harmonic mean discarding the unreliable boundaries + The data is smooth, so we only take 1/4th of the samples */ + unmask=0; + /* We should never see NaNs here. If we find any, then something really bad happened and we better abort + before it does any damage later on. If these asserts are disabled (no hardening), then the table + lookup a few lines below (id = ...) is likely to crash dur to an out-of-bounds read. DO NOT FIX + that crash on NaN since it could result in a worse issue later on. */ + celt_assert(!celt_isnan(tmp[0])); + celt_assert(!celt_isnan(norm)); + for (i=12;imask_metric) + { + *tf_chan = c; + mask_metric = unmask; + } + } + is_transient = mask_metric>200; + /* Prevent the transient detector from confusing the partial cycle of a + very low frequency tone with a transient. */ + if (toneishness > QCONST32(.98f, 29) && tone_freq < QCONST16(0.026f, 13)) + { + is_transient = 0; + mask_metric = 0; + } + /* For low bitrates, define "weak transients" that need to be + handled differently to avoid partial collapse. */ + if (allow_weak_transients && is_transient && mask_metric<600) { + is_transient = 0; + *weak_transient = 1; + } + /* Arbitrary metric for VBR boost */ + tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42); + /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */ + *tf_estimate = celt_sqrt(MAX32(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28))); + /*printf("%d %f\n", tf_max, mask_metric);*/ + RESTORE_STACK; +#ifdef FUZZING + is_transient = rand()&0x1; +#endif + /*printf("%d %f %d\n", is_transient, (float)*tf_estimate, tf_max);*/ + return is_transient; +} + +/* Looks for sudden increases of energy to decide whether we need to patch + the transient decision */ +static int patch_transient_decision(celt_glog *newE, celt_glog *oldE, int nbEBands, + int start, int end, int C) +{ + int i, c; + opus_val32 mean_diff=0; + celt_glog spread_old[26]; + /* Apply an aggressive (-6 dB/Bark) spreading function to the old frame to + avoid false detection caused by irrelevant bands */ + if (C==1) + { + spread_old[start] = oldE[start]; + for (i=start+1;i=start;i--) + spread_old[i] = MAXG(spread_old[i], spread_old[i+1]-GCONST(1.0f)); + /* Compute mean increase */ + c=0; do { + for (i=IMAX(2,start);i GCONST(1.f); +} + +/** Apply window and compute the MDCT for all sub-frames and + all channels in a frame */ +static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in, + celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample, + int arch) +{ + const int overlap = mode->overlap; + int N; + int B; + int shift; + int i, b, c; + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<maxLM-LM; + } + c=0; do { + for (b=0;bmdct, in+c*(B*N+overlap)+b*N, + &out[b+c*N*B], mode->window, overlap, shift, B, + arch); + } + } while (++ceBands[len]-m->eBands[len-1])<eBands[len]-m->eBands[len-1])<eBands[i+1]-m->eBands[i])<eBands[i+1]-m->eBands[i])==1; + OPUS_COPY(tmp, &X[tf_chan*N0 + (m->eBands[i]<eBands[i]<>LM, 1<>k, 1<=0;i--) + { + if (tf_res[i+1] == 1) + tf_res[i] = path1[i+1]; + else + tf_res[i] = path0[i+1]; + } + /*printf("%d %f\n", *tf_sum, tf_estimate);*/ + RESTORE_STACK; +#ifdef FUZZING + tf_select = rand()&0x1; + tf_res[0] = rand()&0x1; + for (i=1;istorage*8; + tell = ec_tell(enc); + logp = isTransient ? 2 : 4; + /* Reserve space to code the tf_select decision. */ + tf_select_rsv = LM>0 && tell+logp+1 <= budget; + budget -= tf_select_rsv; + curr = tf_changed = 0; + for (i=start;i> 10; + trim = QCONST16(4.f, 8) + QCONST16(1.f/16.f, 8)*frac; + } + if (C==2) + { + opus_val16 sum = 0; /* Q10 */ + opus_val16 minXC; /* Q10 */ + /* Compute inter-channel correlation for low frequencies */ + for (i=0;i<8;i++) + { + opus_val32 partial; + partial = celt_inner_prod_norm_shift(&X[m->eBands[i]<eBands[i]<eBands[i+1]-m->eBands[i])<eBands[i]<eBands[i]<eBands[i+1]-m->eBands[i])<nbEBands], 5)*(opus_int32)(2+2*i-end); + } + } while (++cvalid) + { + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), + (opus_val16)(QCONST16(2.f, 8)*(analysis->tonality_slope+.05f)))); + } +#else + (void)analysis; +#endif + +#ifdef FIXED_POINT + trim_index = PSHR32(trim, 8); +#else + trim_index = (int)floor(.5f+trim); +#endif + trim_index = IMAX(0, IMIN(10, trim_index)); + /*printf("%d\n", trim_index);*/ +#ifdef FUZZING + trim_index = rand()%11; +#endif + return trim_index; +} + +static int stereo_analysis(const CELTMode *m, const celt_norm *X, + int LM, int N0) +{ + int i; + int thetas; + opus_val32 sumLR = EPSILON, sumMS = EPSILON; + + /* Use the L1 norm to model the entropy of the L/R signal vs the M/S signal */ + for (i=0;i<13;i++) + { + int j; + for (j=m->eBands[i]<eBands[i+1]<eBands[13]<<(LM+1))+thetas, sumMS) + > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); +} + +#define MSWAP(a,b) do {celt_glog tmp = a;a=b;b=tmp;} while(0) +static celt_glog median_of_5(const celt_glog *x) +{ + celt_glog t0, t1, t2, t3, t4; + t2 = x[2]; + if (x[0] > x[1]) + { + t0 = x[1]; + t1 = x[0]; + } else { + t0 = x[0]; + t1 = x[1]; + } + if (x[3] > x[4]) + { + t3 = x[4]; + t4 = x[3]; + } else { + t3 = x[3]; + t4 = x[4]; + } + if (t0 > t3) + { + MSWAP(t0, t3); + MSWAP(t1, t4); + } + if (t2 > t1) + { + if (t1 < t3) + return MING(t2, t3); + else + return MING(t4, t1); + } else { + if (t2 < t3) + return MING(t1, t3); + else + return MING(t2, t4); + } +} + +static celt_glog median_of_3(const celt_glog *x) +{ + celt_glog t0, t1, t2; + if (x[0] > x[1]) + { + t0 = x[1]; + t1 = x[0]; + } else { + t0 = x[0]; + t1 = x[1]; + } + t2 = x[2]; + if (t1 < t2) + return t1; + else if (t0 < t2) + return t2; + else + return t0; +} + +static celt_glog dynalloc_analysis(const celt_glog *bandLogE, const celt_glog *bandLogE2, const celt_glog *oldBandE, + int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN, + int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM, + int effectiveBytes, opus_int32 *tot_boost_, int lfe, celt_glog *surround_dynalloc, + AnalysisInfo *analysis, int *importance, int *spread_weight, opus_val16 tone_freq, opus_val32 toneishness + ARG_QEXT(int qext_scale)) +{ + int i, c; + opus_int32 tot_boost=0; + celt_glog maxDepth; + VARDECL(celt_glog, follower); + VARDECL(celt_glog, noise_floor); + VARDECL(celt_glog, bandLogE3); + SAVE_STACK; + ALLOC(follower, C*nbEBands, celt_glog); + ALLOC(noise_floor, C*nbEBands, celt_glog); + ALLOC(bandLogE3, nbEBands, celt_glog); + OPUS_CLEAR(offsets, nbEBands); + /* Dynamic allocation code */ + maxDepth=-GCONST(31.9f); + for (i=0;i=0;i--) + mask[i] = MAXG(mask[i], mask[i+1] - GCONST(3.f)); + for (i=0;i> shift; + } + /*for (i=0;i= (30 + 5*LM) && !lfe) + { + int last=0; + c=0;do + { + celt_glog offset; + celt_glog tmp; + celt_glog *f; + OPUS_COPY(bandLogE3, &bandLogE2[c*nbEBands], end); + if (LM==0) { + /* For 2.5 ms frames, the first 8 bands have just one bin, so the + energy is highly unreliable (high variance). For that reason, + we take the max with the previous energy so that at least 2 bins + are getting used. */ + for (i=0;i bandLogE3[i-1]+GCONST(.5f)) + last=i; + f[i] = MING(f[i-1]+GCONST(1.5f), bandLogE3[i]); + } + for (i=last-1;i>=0;i--) + f[i] = MING(f[i], MING(f[i+1]+GCONST(2.f), bandLogE3[i])); + + /* Combine with a median filter to avoid dynalloc triggering unnecessarily. + The "offset" value controls how conservative we are -- a higher offset + reduces the impact of the median filter and makes dynalloc use more bits. */ + offset = GCONST(1.f); + for (i=2;i=12) + follower[i] = HALF32(follower[i]); + } + /* Compensate for Opus' under-allocation on tones. */ + if (toneishness > QCONST32(.98f, 29)) { +#ifdef FIXED_POINT + int freq_bin = PSHR32(QEXT_SCALE((opus_val32)tone_freq)*QCONST16(120/M_PI, 9), 13+9); +#else + int freq_bin = (int)floor(.5 + QEXT_SCALE(tone_freq)*120/M_PI); +#endif + for (i=start;i= eBands[i] && freq_bin <= eBands[i+1]) follower[i] += GCONST(2.f); + if (freq_bin >= eBands[i]-1 && freq_bin <= eBands[i+1]+1) follower[i] += GCONST(1.f); + if (freq_bin >= eBands[i]-2 && freq_bin <= eBands[i+1]+2) follower[i] += GCONST(1.f); + if (freq_bin >= eBands[i]-3 && freq_bin <= eBands[i+1]+3) follower[i] += GCONST(.5f); + } + if (freq_bin >= eBands[end]) { + follower[end-1] += GCONST(2.f); + follower[end-2] += GCONST(1.f); + } + } +#ifdef DISABLE_FLOAT_API + (void)analysis; +#else + if (analysis->valid) + { + for (i=start;ileak_boost[i]; + } +#endif + for (i=start;i 48) { + boost = (int)SHR32(follower[i]*8,DB_SHIFT-8); + boost_bits = (boost*width<>BITRES>>3 > 2*effectiveBytes/3) + { + opus_int32 cap = ((2*effectiveBytes/3)< 0) { + for (i=0;i>15; + tmp = (762*x14>>14)-3308; + tmp = (tmp*x14>>14)+25726; + tmp = tmp*celt_sqrt(IMAX(0, (1<<30) - (x<<1)))>>16; + if (flip) tmp = 25736 - tmp; + return tmp; +} +#endif + +/* Compute the LPC coefficients using a least-squares fit for both forward and backward prediction. */ +static int tone_lpc(const opus_val16 *x, int len, int delay, opus_val32 *lpc) { + int i; + opus_val32 r00=0, r01=0, r11=0, r02=0, r12=0, r22=0; + opus_val32 edges; + opus_val32 num0, num1, den; + celt_assert(len > 2*delay); + /* Compute correlations as if using the forward prediction covariance method. */ + for (i=0;i= den) lpc[1] = QCONST32(1.f, 29); + else if (num1 <= -den) lpc[1] = -QCONST32(1.f, 29); + else lpc[1] = frac_div32_q29(num1, den); + num0 = MULT32_32_Q31(r00,r12) - MULT32_32_Q31(r02,r01); + if (HALF32(num0) >= den) lpc[0] = QCONST32(1.999999f, 29); + else if (HALF32(num0) <= -den) lpc[0] = -QCONST32(1.999999f, 29); + else lpc[0] = frac_div32_q29(num0, den); + /*printf("%f %f\n", lpc[0], lpc[1]);*/ + return 0; +} + +/* Detects pure of nearly pure tones so we can prevent them from causing problems with the encoder. */ +static opus_val16 tone_detect(const celt_sig *in, int CC, int N, opus_val32 *toneishness, opus_int32 Fs) { + int i; + int delay = 1; + int fail; + opus_val32 lpc[2]; + opus_val16 freq; + VARDECL(opus_val16, x); + SAVE_STACK; + ALLOC(x, N, opus_val16); + /* Shift by SIG_SHIFT+2 (+3 for stereo) to account for HF gain of the preemphasis filter. */ + if (CC==2) { + for (i=0;i QCONST32(1.f, 29) && lpc[1] < 0))) { + delay *= 2; + fail = tone_lpc(x, N, delay, lpc); + } + /* Check that our filter has complex roots. */ + if (!fail && MULT32_32_Q31(lpc[0],lpc[0]) + MULT32_32_Q31(QCONST32(3.999999, 29), lpc[1]) < 0) { + /* Squared radius of the poles. */ + *toneishness = -lpc[1]; +#ifdef FIXED_POINT + freq = (acos_approx(lpc[0]>>1)+delay/2)/delay; +#else + freq = acos(.5f*lpc[0])/delay; +#endif + } else { + freq = -1; + *toneishness=0; + } + /*printf("%f %f %f %f\n", freq, lpc[0], lpc[1], *toneishness);*/ + RESTORE_STACK; + return freq; +} + +static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem, int CC, int N, + int prefilter_tapset, int *pitch, opus_val16 *gain, int *qgain, int enabled, int complexity, opus_val16 tf_estimate, + int nbAvailableBytes, AnalysisInfo *analysis, opus_val16 tone_freq, opus_val32 toneishness ARG_QEXT(int qext_scale)) +{ + int c; + VARDECL(celt_sig, _pre); + celt_sig *pre[2]; + const CELTMode *mode; + int pitch_index; + opus_val16 gain1; + opus_val16 pf_threshold; + int pf_on; + int qg; + int overlap; + int min_period, max_period; + opus_val32 before[2]={0}, after[2]={0}; + int cancel_pitch=0; + SAVE_STACK; + + max_period = QEXT_SCALE(COMBFILTER_MAXPERIOD); + min_period = QEXT_SCALE(COMBFILTER_MINPERIOD); + mode = st->mode; + overlap = mode->overlap; + ALLOC(_pre, CC*(N+max_period), celt_sig); + + pre[0] = _pre; + pre[1] = _pre + (N+max_period); + + + c=0; do { + OPUS_COPY(pre[c], prefilter_mem+c*max_period, max_period); + OPUS_COPY(pre[c]+max_period, in+c*(N+overlap)+overlap, N); + } while (++c QCONST32(.99f, 29)) { + int multiple=1; + /* Using aliased version of the postfilter above 24 kHz. + First value is purposely slightly above pi to avoid triggering for Fs=48kHz. */ + if (QEXT_SCALE(tone_freq) >= QCONST16(3.1416f, 13)) tone_freq = QCONST16(3.141593f, 13) - tone_freq; + /* If the pitch is too high for our post-filter, apply pitch doubling until + we can get something that fits (not ideal, but better than nothing). */ + while (QEXT_SCALE(tone_freq) >= multiple*QCONST16(0.39f, 13)) multiple++; + if (QEXT_SCALE(tone_freq) > QCONST16(0.006148f, 13)) { +#ifdef FIXED_POINT + pitch_index = IMIN((51472*multiple+QEXT_SCALE(tone_freq)/2)/QEXT_SCALE(tone_freq), COMBFILTER_MAXPERIOD-2); +#else + pitch_index = IMIN((int)floor(.5+2.f*M_PI*multiple/QEXT_SCALE(tone_freq)), COMBFILTER_MAXPERIOD-2); +#endif + } else { + /* If the pitch is too low, using a very high pitch will actually give us an improvement + due to the DC component of the filter that will be close to our tone. Again, not ideal, + but if we only have a single tone, it's better than nothing. */ + pitch_index = COMBFILTER_MINPERIOD; + } + gain1 = QCONST16(.75f, 15); + } else if (enabled && complexity >= 5) { + VARDECL(opus_val16, pitch_buf); + ALLOC(pitch_buf, (max_period+N)>>1, opus_val16); + + pitch_downsample(pre, pitch_buf, (max_period+N)>>1, CC, 2, st->arch); + /* Don't search for the fir last 1.5 octave of the range because + there's too many false-positives due to short-term correlation */ + pitch_search(pitch_buf+(max_period>>1), pitch_buf, N, + max_period-3*min_period, &pitch_index, + st->arch); + pitch_index = max_period-pitch_index; + + gain1 = remove_doubling(pitch_buf, max_period, min_period, + N, &pitch_index, st->prefilter_period, st->prefilter_gain, st->arch); + if (pitch_index > max_period-QEXT_SCALE(2)) + pitch_index = max_period-QEXT_SCALE(2); +#ifdef ENABLE_QEXT + pitch_index /= qext_scale; +#endif + gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1); + /*printf("%d %d %f %f\n", pitch_change, pitch_index, gain1, st->analysis.tonality);*/ + if (st->loss_rate>2) + gain1 = HALF32(gain1); + if (st->loss_rate>4) + gain1 = HALF32(gain1); + if (st->loss_rate>8) + gain1 = 0; + } else { + gain1 = 0; + pitch_index = COMBFILTER_MINPERIOD; + } +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + gain1 = (opus_val16)(gain1 * analysis->max_pitch_ratio); +#else + (void)analysis; +#endif + /* Gain threshold for enabling the prefilter/postfilter */ + pf_threshold = QCONST16(.2f,15); + + /* Adjusting the threshold based on rate and continuity */ + if (abs(pitch_index-st->prefilter_period)*10>pitch_index) + { + pf_threshold += QCONST16(.2f,15); + /* Completely disable the prefilter on strong transients without continuity. */ + if (tf_estimate > QCONST16(.98f, 14)) + gain1 = 0; + } + if (nbAvailableBytes<25) + pf_threshold += QCONST16(.1f,15); + if (nbAvailableBytes<35) + pf_threshold += QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.4f,15)) + pf_threshold -= QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.55f,15)) + pf_threshold -= QCONST16(.1f,15); + + /* Hard threshold at 0.2 */ + pf_threshold = MAX16(pf_threshold, QCONST16(.2f,15)); + if (gain1prefilter_gain)prefilter_gain; + +#ifdef FIXED_POINT + qg = ((gain1+1536)>>10)/3-1; +#else + qg = (int)floor(.5f+gain1*32/3)-1; +#endif + qg = IMAX(0, IMIN(7, qg)); + gain1 = QCONST16(0.09375f,15)*(qg+1); + pf_on = 1; + } + /*printf("%d %f\n", pitch_index, gain1);*/ + + c=0; do { + int i; + int offset = mode->shortMdctSize-overlap; + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + OPUS_COPY(in+c*(N+overlap), st->in_mem+c*(overlap), overlap); + for (i=0;iprefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain, + st->prefilter_tapset, st->prefilter_tapset, NULL, 0, st->arch); + + comb_filter(in+c*(N+overlap)+overlap+offset, pre[c]+max_period+offset, + st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1, + st->prefilter_tapset, prefilter_tapset, mode->window, overlap, st->arch); + for (i=0;i thresh[0] || after[1]-before[1] > thresh[1]) cancel_pitch = 1; + /* Use the filter only if at least one channel gets significantly better. */ + if (before[0]-after[0] < thresh[0] && before[1]-after[1] < thresh[1]) cancel_pitch = 1; + } else { + /* Check that the mono channel actually got better. */ + if (after[0] > before[0]) cancel_pitch = 1; + } + /* If needed, revert to a gain of zero. */ + if (cancel_pitch) { + c=0; do { + int offset = mode->shortMdctSize-overlap; + OPUS_COPY(in+c*(N+overlap)+overlap, pre[c]+max_period, N); + comb_filter(in+c*(N+overlap)+overlap+offset, pre[c]+max_period+offset, + st->prefilter_period, pitch_index, overlap, -st->prefilter_gain, -0, + st->prefilter_tapset, prefilter_tapset, mode->window, overlap, st->arch); + } while (++cin_mem+c*(overlap), in+c*(N+overlap)+N, overlap); + + if (N>max_period) + { + OPUS_COPY(prefilter_mem+c*max_period, pre[c]+N, max_period); + } else { + OPUS_MOVE(prefilter_mem+c*max_period, prefilter_mem+c*max_period+N, max_period-N); + OPUS_COPY(prefilter_mem+c*max_period+max_period-N, pre[c]+max_period, N); + } + } while (++cnbEBands; + eBands = mode->eBands; + + coded_bands = lastCodedBands ? lastCodedBands : nbEBands; + coded_bins = eBands[coded_bands]<analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/ +#ifndef DISABLE_FLOAT_API + if (analysis->valid && analysis->activity<.4) + target -= (opus_int32)((coded_bins<activity)); +#endif + /* Stereo savings */ + if (C==2) + { + int coded_stereo_bands; + int coded_stereo_dof; + opus_val16 max_frac; + coded_stereo_bands = IMIN(intensity, coded_bands); + coded_stereo_dof = (eBands[coded_stereo_bands]<valid && !lfe) + { + opus_int32 tonal_target; + float tonal; + + /* Tonality boost (compensating for the average). */ + tonal = MAX16(0.f,analysis->tonality-.15f)-0.12f; + tonal_target = target + (opus_int32)((coded_bins<tonality, tonal);*/ + target = tonal_target; + } +#else + (void)analysis; + (void)pitch_change; +#endif + + if (has_surround_mask&&!lfe) + { + opus_int32 surround_target = target + (opus_int32)SHR32(MULT16_16(SHR32(surround_masking,DB_SHIFT-10),coded_bins<end, st->intensity, surround_target, target, st->bitrate);*/ + target = IMAX(target/4, surround_target); + } + + { + opus_int32 floor_depth; + int bins; + bins = eBands[nbEBands-2]<shortMdctSize<>2); + target = IMIN(target, floor_depth); + /*printf("%f %d\n", maxDepth, floor_depth);*/ + } + + /* Make VBR less aggressive for constrained VBR because we can't keep a higher bitrate + for long. Needs tuning. */ + if ((!has_surround_mask||lfe) && constrained_vbr) + { + target = base_target + (opus_int32)MULT16_32_Q15(QCONST16(0.67f, 15), target-base_target); + } + + if (!has_surround_mask && tf_estimate < QCONST16(.2f, 14)) + { + opus_val16 amount; + opus_val16 tvbr_factor; + amount = MULT16_16_Q15(QCONST16(.0000031f, 30), IMAX(0, IMIN(32000, 96000-bitrate))); + tvbr_factor = SHR32(MULT16_16(SHR32(temporal_vbr, DB_SHIFT-10), amount), 10); + target += (opus_int32)MULT16_32_Q15(tvbr_factor, target); + } + + /* Don't allow more than doubling the rate */ + target = IMIN(2*base_target, target); + + return target; +} + +#ifdef ENABLE_QEXT +static void encode_qext_stereo_params(ec_enc *ec, int qext_end, int qext_intensity, int qext_dual_stereo) { + ec_enc_uint(ec, qext_intensity, qext_end+1); + if (qext_intensity != 0) ec_enc_bit_logp(ec, qext_dual_stereo, 1); +} +#endif + +int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_res * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc) +{ + int i, c, N; + opus_int32 bits; + ec_enc _enc; + VARDECL(celt_sig, in); + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(celt_ener, bandE); + VARDECL(celt_glog, bandLogE); + VARDECL(celt_glog, bandLogE2); + VARDECL(int, fine_quant); + VARDECL(celt_glog, error); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, importance); + VARDECL(int, spread_weight); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *prefilter_mem; + celt_glog *oldBandE, *oldLogE, *oldLogE2, *energyError; + int shortBlocks=0; + int isTransient=0; + const int CC = st->channels; + const int C = st->stream_channels; + int LM, M; + int tf_select; + int nbFilledBytes, nbAvailableBytes; + opus_int32 min_allowed; + int start; + int end; + int effEnd; + int codedBands; + int alloc_trim; + int pitch_index=COMBFILTER_MINPERIOD; + opus_val16 gain1 = 0; + int dual_stereo=0; + int effectiveBytes; + int dynalloc_logp; + opus_int32 vbr_rate; + opus_int32 total_bits; + opus_int32 total_boost; + opus_int32 balance; + opus_int32 tell; + opus_int32 tell0_frac; + int prefilter_tapset=0; + int pf_on; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence=0; + int tf_chan = 0; + opus_val16 tf_estimate; + int pitch_change=0; + opus_int32 tot_boost; + opus_val32 sample_max; + celt_glog maxDepth; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + int secondMdct; + int signalBandwidth; + int transient_got_disabled=0; + celt_glog surround_masking=0; + celt_glog temporal_vbr=0; + celt_glog surround_trim = 0; + opus_int32 equiv_rate; + int hybrid; + int weak_transient = 0; + int enable_tf_analysis; + opus_val16 tone_freq=-1; + opus_val32 toneishness=0; + VARDECL(celt_glog, surround_dynalloc); + int qext_bytes=0; + int packet_size_cap = 1275; +#ifdef ENABLE_QEXT + int qext_scale; + int qext_end=0; + int qext_intensity=0; + int qext_dual_stereo=0; + int padding_len_bytes=0; + unsigned char *ext_payload; + opus_int32 qext_bits; + ec_enc ext_enc; + VARDECL(int, extra_quant); + VARDECL(int, extra_pulses); + VARDECL(celt_glog, error_bak); + const CELTMode *qext_mode = NULL; + CELTMode qext_mode_struct; + celt_ener qext_bandE[2*NB_QEXT_BANDS]; + celt_glog qext_bandLogE[2*NB_QEXT_BANDS]; + celt_glog *qext_oldBandE=NULL; + celt_glog qext_error[2*NB_QEXT_BANDS]; +#endif + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + start = st->start; + end = st->end; + hybrid = start != 0; + tf_estimate = 0; + if (nbCompressedBytes<2 || pcm==NULL) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + frame_size *= st->upsample; + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<mode->maxLM) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + M=1<shortMdctSize; + +#ifdef ENABLE_QEXT + qext_scale = st->qext_scale; + if (st->enable_qext) packet_size_cap = QEXT_PACKET_SIZE_CAP; +#endif + + prefilter_mem = st->in_mem+CC*(overlap); + oldBandE = (celt_glog*)(st->in_mem+CC*(overlap+QEXT_SCALE(COMBFILTER_MAXPERIOD))); + oldLogE = oldBandE + CC*nbEBands; + oldLogE2 = oldLogE + CC*nbEBands; + energyError = oldLogE2 + CC*nbEBands; + + if (enc==NULL) + { + tell0_frac=tell=1; + nbFilledBytes=0; + } else { + tell0_frac=ec_tell_frac(enc); + tell=ec_tell(enc); + nbFilledBytes=(tell+4)>>3; + } + +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) + if (st->signalling && enc==NULL) + { + int tmp = (mode->effEBands-end)>>1; + end = st->end = IMAX(1, mode->effEBands-tmp); + compressed[0] = tmp<<5; + compressed[0] |= LM<<3; + compressed[0] |= (C==2)<<2; + /* Convert "standard mode" to Opus header */ +# ifndef ENABLE_QEXT + if (mode->Fs==48000 && mode->shortMdctSize==120) +# endif + { + int c0 = toOpus(compressed[0]); + if (c0<0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + compressed[0] = c0; + } + compressed++; + nbCompressedBytes--; + } +#else + celt_assert(st->signalling==0); +#endif + + /* Can't produce more than 1275 output bytes for the main payload, plus any QEXT extra data. */ + nbCompressedBytes = IMIN(nbCompressedBytes,packet_size_cap); + + if (st->vbr && st->bitrate!=OPUS_BITRATE_MAX) + { + vbr_rate = bitrate_to_bits(st->bitrate, mode->Fs, frame_size)<signalling) + vbr_rate -= 8<>(3+BITRES); + } else { + opus_int32 tmp; + vbr_rate = 0; + tmp = st->bitrate*frame_size; + if (tell>1) + tmp += tell*mode->Fs; + if (st->bitrate!=OPUS_BITRATE_MAX) + { + nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, + (tmp+4*mode->Fs)/(8*mode->Fs)-!!st->signalling)); + if (enc != NULL) + ec_enc_shrink(enc, nbCompressedBytes); + } + effectiveBytes = nbCompressedBytes - nbFilledBytes; + } + nbAvailableBytes = nbCompressedBytes - nbFilledBytes; + equiv_rate = ((opus_int32)nbCompressedBytes*8*50 << (3-LM)) - (40*C+20)*((400>>LM) - 50); + if (st->bitrate != OPUS_BITRATE_MAX) + equiv_rate = IMIN(equiv_rate, st->bitrate - (40*C+20)*((400>>LM) - 50)); + + if (enc==NULL) + { + ec_enc_init(&_enc, compressed, nbCompressedBytes); + enc = &_enc; + } + + if (vbr_rate>0) + { + /* Computes the max bit-rate allowed in VBR mode to avoid violating the + target rate and buffering. + We must do this up front so that bust-prevention logic triggers + correctly if we don't have enough bits. */ + if (st->constrained_vbr) + { + opus_int32 vbr_bound; + opus_int32 max_allowed; + /* We could use any multiple of vbr_rate as bound (depending on the + delay). + This is clamped to ensure we use at least two bytes if the encoder + was entirely empty, but to allow 0 in hybrid mode. */ + vbr_bound = vbr_rate; + max_allowed = IMIN(IMAX(tell==1?2:0, + (vbr_rate+vbr_bound-st->vbr_reservoir)>>(BITRES+3)), + nbAvailableBytes); + if(max_allowed < nbAvailableBytes) + { + nbCompressedBytes = nbFilledBytes+max_allowed; + nbAvailableBytes = max_allowed; + ec_enc_shrink(enc, nbCompressedBytes); + } + } + } + total_bits = nbCompressedBytes*8; + + effEnd = end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + ALLOC(in, CC*(N+overlap), celt_sig); + + sample_max=MAX32(st->overlap_max, celt_maxabs_res(pcm, C*(N-overlap)/st->upsample)); + st->overlap_max=celt_maxabs_res(pcm+C*(N-overlap)/st->upsample, C*overlap/st->upsample); + sample_max=MAX32(sample_max, st->overlap_max); +#ifdef FIXED_POINT + silence = (sample_max==0); +#else + silence = (sample_max <= (opus_val16)1/(1<lsb_depth)); +#endif +#ifdef FUZZING + if ((rand()&0x3F)==0) + silence = 1; +#endif + if (tell==1) + ec_enc_bit_logp(enc, silence, 15); + else + silence=0; + if (silence) + { + /*In VBR mode there is no need to send more than the minimum. */ + if (vbr_rate>0) + { + effectiveBytes=nbCompressedBytes=IMIN(nbCompressedBytes, nbFilledBytes+2); + total_bits=nbCompressedBytes*8; + nbAvailableBytes=2; + ec_enc_shrink(enc, nbCompressedBytes); + } +#ifdef ENABLE_QEXT + else if (st->enable_qext) { + nbCompressedBytes = IMIN(nbCompressedBytes, 1275); + nbAvailableBytes = nbCompressedBytes - nbFilledBytes; + total_bits = nbCompressedBytes*8; + ec_enc_shrink(enc, nbCompressedBytes); + } +#endif + /* Pretend we've filled all the remaining bits with zeros + (that's what the initialiser did anyway) */ + tell = nbCompressedBytes*8; + enc->nbits_total+=tell-ec_tell(enc); + } + c=0; do { + int need_clip=0; +#ifdef FIXED_POINT + need_clip = st->clip && sample_max>65536<clip && sample_max>65536.f; +#endif + celt_preemphasis(pcm+c, in+c*(N+overlap)+overlap, N, CC, st->upsample, + mode->preemph, st->preemph_memE+c, need_clip); + OPUS_COPY(in+c*(N+overlap), &prefilter_mem[(1+c)*QEXT_SCALE(COMBFILTER_MAXPERIOD)-overlap], overlap); + } while (++cFs); + isTransient = 0; + shortBlocks = 0; + if (st->complexity >= 1 && !st->lfe) + { + /* Reduces the likelihood of energy instability on fricatives at low bitrate + in hybrid mode. It seems like we still want to have real transients on vowels + though (small SILK quantization offset value). */ + int allow_weak_transients = hybrid && effectiveBytes<15 && st->silk_info.signalType != 2; + isTransient = transient_analysis(in, N+overlap, CC, + &tf_estimate, &tf_chan, allow_weak_transients, &weak_transient, tone_freq, toneishness); + } + toneishness = MIN32(toneishness, QCONST32(1.f, 29)-SHL32(tf_estimate, 15)); + /* Find pitch period and gain */ + { + int enabled; + int qg; + enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && !hybrid && !silence && tell+16<=total_bits && !st->disable_pf; + + prefilter_tapset = st->tapset_decision; + pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, st->complexity, tf_estimate, nbAvailableBytes, &st->analysis, tone_freq, toneishness ARG_QEXT(qext_scale)); + if ((gain1 > QCONST16(.4f,15) || st->prefilter_gain > QCONST16(.4f,15)) && (!st->analysis.valid || st->analysis.tonality > .3) + && (pitch_index > 1.26*st->prefilter_period || pitch_index < .79*st->prefilter_period)) + pitch_change = 1; + if (pf_on==0) + { + if(!hybrid && tell+16<=total_bits) + ec_enc_bit_logp(enc, 0, 1); + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + int octave; + ec_enc_bit_logp(enc, 1, 1); + pitch_index += 1; + octave = EC_ILOG(pitch_index)-5; + ec_enc_uint(enc, octave, 6); + ec_enc_bits(enc, pitch_index-(16<0 && ec_tell(enc)+3<=total_bits) + { + if (isTransient) + shortBlocks = M; + } else { + isTransient = 0; + transient_got_disabled=1; + } + + ALLOC(freq, CC*N, celt_sig); /**< Interleaved signal MDCTs */ + ALLOC(bandE,nbEBands*CC, celt_ener); + ALLOC(bandLogE,nbEBands*CC, celt_glog); + + secondMdct = shortBlocks && st->complexity>=8; + ALLOC(bandLogE2, C*nbEBands, celt_glog); + if (secondMdct) + { + compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample, st->arch); + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + amp2Log2(mode, effEnd, end, bandE, bandLogE2, C); + for (c=0;cupsample, st->arch); + /* This should catch any NaN in the CELT input. Since we're not supposed to see any (they're filtered + at the Opus layer), just abort. */ + celt_assert(!celt_isnan(freq[0]) && (C==1 || !celt_isnan(freq[N]))); + if (CC==2&&C==1) + tf_chan = 0; + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + + if (st->lfe) + { + for (i=2;ienergy_mask&&!st->lfe) + { + int mask_end; + int midband; + int count_dynalloc; + opus_val32 mask_avg=0; + opus_val32 diff=0; + int count=0; + mask_end = IMAX(2,st->lastCodedBands); + for (c=0;cenergy_mask[nbEBands*c+i], + GCONST(.25f)), -GCONST(2.0f)); + if (mask > 0) + mask = HALF32(mask); + mask16 = SHR32(mask, DB_SHIFT-10); + mask_avg += MULT16_16(mask16, eBands[i+1]-eBands[i]); + count += eBands[i+1]-eBands[i]; + diff += MULT16_16(mask16, 1+2*i-mask_end); + } + } + celt_assert(count>0); + mask_avg = SHL32(DIV32_16(mask_avg,count), DB_SHIFT-10); + mask_avg += GCONST(.2f); + diff = SHL32(diff*6/(C*(mask_end-1)*(mask_end+1)*mask_end), DB_SHIFT-10); + /* Again, being conservative */ + diff = HALF32(diff); + diff = MAX32(MIN32(diff, GCONST(.031f)), -GCONST(.031f)); + /* Find the band that's in the middle of the coded spectrum */ + for (midband=0;eBands[midband+1] < eBands[mask_end]/2;midband++); + count_dynalloc=0; + for(i=0;ienergy_mask[i], st->energy_mask[nbEBands+i]); + else + unmask = st->energy_mask[i]; + unmask = MING(unmask, GCONST(.0f)); + unmask -= lin; + if (unmask > GCONST(.25f)) + { + surround_dynalloc[i] = unmask - GCONST(.25f); + count_dynalloc++; + } + } + if (count_dynalloc>=3) + { + /* If we need dynalloc in many bands, it's probably because our + initial masking rate was too low. */ + mask_avg += GCONST(.25f); + if (mask_avg>0) + { + /* Something went really wrong in the original calculations, + disabling masking. */ + mask_avg = 0; + diff = 0; + OPUS_CLEAR(surround_dynalloc, mask_end); + } else { + for(i=0;ilfe) + { + celt_glog follow=-QCONST32(10.0f, DB_SHIFT-5); + opus_val32 frame_avg=0; + celt_glog offset = shortBlocks?HALF32(SHL32(LM, DB_SHIFT-5)):0; + for(i=start;ispec_avg); + temporal_vbr = MING(GCONST(3.f), MAXG(-GCONST(1.5f), temporal_vbr)); + st->spec_avg += MULT16_32_Q15(QCONST16(.02f, 15), temporal_vbr); + } + /*for (i=0;i<21;i++) + printf("%f ", bandLogE[i]); + printf("\n");*/ + + if (!secondMdct) + { + OPUS_COPY(bandLogE2, bandLogE, C*nbEBands); + } + + /* Last chance to catch any transient we might have missed in the + time-domain analysis */ + if (LM>0 && ec_tell(enc)+3<=total_bits && !isTransient && st->complexity>=5 && !st->lfe && !hybrid) + { + if (patch_transient_decision(bandLogE, oldBandE, nbEBands, start, end, C)) + { + isTransient = 1; + shortBlocks = M; + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch); + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + amp2Log2(mode, effEnd, end, bandE, bandLogE, C); + /* Compensate for the scaling of short vs long mdcts */ + for (c=0;c0 && ec_tell(enc)+3<=total_bits) + ec_enc_bit_logp(enc, isTransient, 3); + + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + /* Band normalisation */ + normalise_bands(mode, freq, X, bandE, effEnd, C, M); + + enable_tf_analysis = effectiveBytes>=15*C && !hybrid && st->complexity>=2 && !st->lfe && toneishness < QCONST32(.98f, 29); + + ALLOC(offsets, nbEBands, int); + ALLOC(importance, nbEBands, int); + ALLOC(spread_weight, nbEBands, int); + + maxDepth = dynalloc_analysis(bandLogE, bandLogE2, oldBandE, nbEBands, start, end, C, offsets, + st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, + eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc, &st->analysis, importance, spread_weight, tone_freq, toneishness ARG_QEXT(qext_scale)); + + ALLOC(tf_res, nbEBands, int); + /* Disable variable tf resolution for hybrid and at very low bitrate */ + if (enable_tf_analysis) + { + int lambda; + lambda = IMAX(80, 20480/effectiveBytes + 2); + tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, tf_estimate, tf_chan, importance); + for (i=effEnd;isilk_info.signalType != 2) + { + /* For low bitrate hybrid, we force temporal resolution to 5 ms rather than 2.5 ms. */ + for (i=0;iforce_intra, + &st->delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe); + + tf_encode(start, end, isTransient, tf_res, LM, tf_select, enc); + + if (ec_tell(enc)+4<=total_bits) + { + if (st->lfe) + { + st->tapset_decision = 0; + st->spread_decision = SPREAD_NORMAL; + } else if (hybrid) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else if (isTransient) + st->spread_decision = SPREAD_NORMAL; + else + st->spread_decision = SPREAD_AGGRESSIVE; + } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else + st->spread_decision = SPREAD_NORMAL; + } else { + /* Disable new spreading+tapset estimator until we can show it works + better than the old one. So far it seems like spreading_decision() + works best. */ +#if 0 + if (st->analysis.valid) + { + static const opus_val16 spread_thresholds[3] = {-QCONST16(.6f, 15), -QCONST16(.2f, 15), -QCONST16(.07f, 15)}; + static const opus_val16 spread_histeresis[3] = {QCONST16(.15f, 15), QCONST16(.07f, 15), QCONST16(.02f, 15)}; + static const opus_val16 tapset_thresholds[2] = {QCONST16(.0f, 15), QCONST16(.15f, 15)}; + static const opus_val16 tapset_histeresis[2] = {QCONST16(.1f, 15), QCONST16(.05f, 15)}; + st->spread_decision = hysteresis_decision(-st->analysis.tonality, spread_thresholds, spread_histeresis, 3, st->spread_decision); + st->tapset_decision = hysteresis_decision(st->analysis.tonality_slope, tapset_thresholds, tapset_histeresis, 2, st->tapset_decision); + } else +#endif + { + st->spread_decision = spreading_decision(mode, X, + &st->tonal_average, st->spread_decision, &st->hf_average, + &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M, spread_weight); + } + /*printf("%d %d\n", st->tapset_decision, st->spread_decision);*/ + /*printf("%f %d %f %d\n\n", st->analysis.tonality, st->spread_decision, st->analysis.tonality_slope, st->tapset_decision);*/ + } + ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); + } else { + st->spread_decision = SPREAD_NORMAL; + } + + /* For LFE, everything interesting is in the first band */ + if (st->lfe) + offsets[0] = IMIN(8, effectiveBytes/3); + ALLOC(cap, nbEBands, int); + init_caps(mode,cap,LM,C); + + dynalloc_logp = 6; + total_bits<<=BITRES; + total_boost = 0; + tell = ec_tell_frac(enc); + for (i=start;iintensity = hysteresis_decision((opus_val16)(equiv_rate/1000), + intensity_thresholds, intensity_histeresis, 21, st->intensity); + st->intensity = IMIN(end,IMAX(start, st->intensity)); + } + + alloc_trim = 5; + if (tell+(6< 0 || st->lfe) + { + st->stereo_saving = 0; + alloc_trim = 5; + } else { + alloc_trim = alloc_trim_analysis(mode, X, bandLogE, + end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate, + st->intensity, surround_trim, equiv_rate, st->arch); + } + ec_enc_icdf(enc, alloc_trim, trim_icdf, 7); + tell = ec_tell_frac(enc); + } + + /* In VBR mode the frame size must not be reduced so much that it would + result in the encoder running out of bits. + The margin of 2 bytes ensures that none of the bust-prevention logic + in the decoder will have triggered so far. */ + min_allowed = ((tell+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)) + 2; + /* Take into account the 37 bits we need to have left in the packet to + signal a redundant frame in hybrid mode. Creating a shorter packet would + create an entropy coder desync. */ + if (hybrid) + min_allowed = IMAX(min_allowed, (tell0_frac+(37<>(BITRES+3)); + /* Variable bitrate */ + if (vbr_rate>0) + { + opus_val16 alpha; + opus_int32 delta; + /* The target rate in 8th bits per frame */ + opus_int32 target, base_target; + int lm_diff = mode->maxLM - LM; + + /* Don't attempt to use more than 510 kb/s, even for frames smaller than 20 ms. + The CELT allocator will just not be able to use more than that anyway. */ + nbCompressedBytes = IMIN(nbCompressedBytes,packet_size_cap>>(3-LM)); + if (!hybrid) + { + base_target = vbr_rate - ((40*C+20)<constrained_vbr) + base_target += (st->vbr_offset>>lm_diff); + + if (!hybrid) + { + target = compute_vbr(mode, &st->analysis, base_target, LM, equiv_rate, + st->lastCodedBands, C, st->intensity, st->constrained_vbr, + st->stereo_saving, tot_boost, tf_estimate, pitch_change, maxDepth, + st->lfe, st->energy_mask!=NULL, surround_masking, + temporal_vbr ARG_QEXT(st->enable_qext)); + } else { + target = base_target; + /* Tonal frames (offset<100) need more bits than noisy (offset>100) ones. */ + if (st->silk_info.offset < 100) target += 12 << BITRES >> (3-LM); + if (st->silk_info.offset > 100) target -= 18 << BITRES >> (3-LM); + /* Boosting bitrate on transients and vowels with significant temporal + spikes. */ + target += (opus_int32)MULT16_16_Q14(tf_estimate-QCONST16(.25f,14), (50< QCONST16(.7f,14)) + target = IMAX(target, 50<>(BITRES+3); + nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes); + nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes); + + /* By how much did we "miss" the target on that frame */ + delta = target - vbr_rate; + + target=nbAvailableBytes<<(BITRES+3); + + /*If the frame is silent we don't adjust our drift, otherwise + the encoder will shoot to very high rates after hitting a + span of silence, but we do allow the bitres to refill. + This means that we'll undershoot our target in CVBR/VBR modes + on files with lots of silence. */ + if(silence) + { + nbAvailableBytes = 2; + target = 2*8<vbr_count < 970) + { + st->vbr_count++; + alpha = celt_rcp(SHL32(EXTEND32(st->vbr_count+20),16)); + } else + alpha = QCONST16(.001f,15); + /* How many bits have we used in excess of what we're allowed */ + if (st->constrained_vbr) + st->vbr_reservoir += target - vbr_rate; + /*printf ("%d\n", st->vbr_reservoir);*/ + + /* Compute the offset we need to apply in order to reach the target */ + if (st->constrained_vbr) + { + st->vbr_drift += (opus_int32)MULT16_32_Q15(alpha,(delta*(1<vbr_offset-st->vbr_drift); + st->vbr_offset = -st->vbr_drift; + } + /*printf ("%d\n", st->vbr_drift);*/ + + if (st->constrained_vbr && st->vbr_reservoir < 0) + { + /* We're under the min value -- increase rate */ + int adjust = (-st->vbr_reservoir)/(8<vbr_reservoir = 0; + /*printf ("+%d\n", adjust);*/ + } + nbCompressedBytes = IMIN(nbCompressedBytes,nbAvailableBytes); + /*printf("%d\n", nbCompressedBytes*50*8);*/ + /* This moves the raw bits to take into account the new compressed size */ + ec_enc_shrink(enc, nbCompressedBytes); + } +#ifdef ENABLE_QEXT + if (st->enable_qext) { + int new_compressedBytes; + /* Don't give any bits for the first 80 kb/s per channel. Then 80% of the excess. */ + opus_int32 offset = bitrate_to_bits(C*80000, mode->Fs, frame_size)/8; + qext_bytes = IMAX(nbCompressedBytes-1275, IMAX(0, (nbCompressedBytes-offset)*4/5)); + if (qext_bytes > 20) { + opus_int32 target; + opus_val16 scale; + target = ((nbCompressedBytes-qext_bytes/3)*8<analysis, target, LM, equiv_rate, + st->lastCodedBands, C, st->intensity, st->constrained_vbr, + st->stereo_saving, tot_boost, tf_estimate2, pitch_change, maxDepth, + st->lfe, st->energy_mask!=NULL, surround_masking, + temporal_vbr ARG_QEXT(st->enable_qext)); + target += tell; + } + scale = PSHR32(toneishness,14); + scale = Q15ONE - MULT16_16_Q15(scale, scale); + qext_bytes += MULT16_32_Q15(scale, (nbCompressedBytes-(target/(8< 20) { + new_compressedBytes = nbCompressedBytes-qext_bytes-padding_len_bytes-1; + ec_enc_shrink(enc, new_compressedBytes); + if (compressed == NULL) { + compressed = enc->buf; + } + compressed[-1] |= 0x03; /* Code 3 packet */ + enc->buf += 1+padding_len_bytes; + OPUS_MOVE(compressed+1+padding_len_bytes, compressed, new_compressedBytes); + compressed[0] = 0x41; /* Set padding */ + for (i=0;iFs == 48000 || mode->Fs == 96000) && (mode->shortMdctSize==120*qext_scale || mode->shortMdctSize==90*qext_scale)) { + compute_qext_mode(&qext_mode_struct, mode); + qext_mode = &qext_mode_struct; + qext_end = (qext_scale == 2) ? NB_QEXT_BANDS : 2; + ec_enc_bit_logp(&ext_enc, qext_end == NB_QEXT_BANDS, 1); + } + } else { + ec_enc_init(&ext_enc, NULL, 0); + qext_bytes = 0; + } + } else { + ec_enc_init(&ext_enc, NULL, 0); + } +#endif + + /* Bit allocation */ + ALLOC(fine_quant, nbEBands, int); + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + /* bits = packet size - where we are - safety*/ + bits = (((opus_int32)nbCompressedBytes*8)<=2&&bits>=((LM+2)<analysis.valid) + { + int min_bandwidth; + if (equiv_rate < (opus_int32)32000*C) + min_bandwidth = 13; + else if (equiv_rate < (opus_int32)48000*C) + min_bandwidth = 16; + else if (equiv_rate < (opus_int32)60000*C) + min_bandwidth = 18; + else if (equiv_rate < (opus_int32)80000*C) + min_bandwidth = 19; + else + min_bandwidth = 20; + signalBandwidth = IMAX(st->analysis.bandwidth, min_bandwidth); + } +#endif + if (st->lfe) + signalBandwidth = 1; + codedBands = clt_compute_allocation(mode, start, end, offsets, cap, + alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth); + if (st->lastCodedBands) + st->lastCodedBands = IMIN(st->lastCodedBands+1,IMAX(st->lastCodedBands-1,codedBands)); + else + st->lastCodedBands = codedBands; + + quant_fine_energy(mode, start, end, oldBandE, error, NULL, fine_quant, enc, C); + OPUS_CLEAR(energyError, nbEBands*CC); +#ifdef ENABLE_QEXT + if (qext_mode) + { + /* Don't bias for intra. */ + opus_val32 qext_delayedIntra=0; + qext_oldBandE = energyError + CC*nbEBands; + compute_band_energies(qext_mode, freq, qext_bandE, qext_end, C, LM, st->arch); + normalise_bands(qext_mode, freq, X, qext_bandE, qext_end, C, M); + amp2Log2(qext_mode, qext_end, qext_end, qext_bandE, qext_bandLogE, C); + if (C==2) { + qext_intensity = qext_end; + qext_dual_stereo = dual_stereo; + encode_qext_stereo_params(&ext_enc, qext_end, qext_intensity, qext_dual_stereo); + } + quant_coarse_energy(qext_mode, 0, qext_end, qext_end, qext_bandLogE, + qext_oldBandE, qext_bytes*8, qext_error, &ext_enc, + C, LM, qext_bytes, st->force_intra, + &qext_delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe); + } + ALLOC(extra_quant, nbEBands+NB_QEXT_BANDS, int); + ALLOC(extra_pulses, nbEBands+NB_QEXT_BANDS, int); + ALLOC(error_bak, C*nbEBands, celt_glog); + + qext_bits = ((opus_int32)qext_bytes*8< 0) { + quant_fine_energy(mode, start, end, oldBandE, error, fine_quant, extra_quant, &ext_enc, C); + } +#endif + + /* Residual quantisation */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + quant_all_bands(1, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks, + bandE, pulses, shortBlocks, st->spread_decision, + dual_stereo, st->intensity, tf_res, nbCompressedBytes*(8<rng, st->complexity, st->arch, st->disable_inv + ARG_QEXT(&ext_enc) ARG_QEXT(extra_pulses) + ARG_QEXT(qext_bytes*(8<spread_decision, + qext_dual_stereo, qext_intensity, zeros, qext_bytes*(8<rng, st->complexity, st->arch, st->disable_inv, &dummy_enc, zeros, 0, NULL); + } +#endif + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = st->consec_transient<2; +#ifdef FUZZING + anti_collapse_on = rand()&0x1; +#endif + ec_enc_bits(enc, anti_collapse_on, 1); + } + if (qext_bytes == 0) + quant_energy_finalise(mode, start, end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); + c=0; + do { + for (i=start;i 0) + quant_energy_finalise(mode, start, end, NULL, error_bak, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); +#endif + if (silence) + { + for (i=0;irng, 1, st->arch); + } + + c=0; do { + OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, QEXT_SCALE(DEC_PITCH_BUF_SIZE)-N+overlap/2); + } while (++csyn_mem[c]+QEXT_SCALE(DEC_PITCH_BUF_SIZE)-N; + } while (++cupsample, silence, st->arch ARG_QEXT(qext_mode) ARG_QEXT(qext_oldBandE) ARG_QEXT(qext_end)); + + c=0; do { + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize, + st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, + mode->window, overlap, st->arch); + if (LM!=0) + comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize, + st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, + mode->window, overlap, st->arch); + } while (++cupsample, mode->preemph, st->preemph_memD, 0); + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + st->prefilter_period = pitch_index; + st->prefilter_gain = gain1; + st->prefilter_tapset = prefilter_tapset; +#ifdef RESYNTH + if (LM!=0) + { + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + if (CC==2&&C==1) { + OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands); + } + + if (!isTransient) + { + OPUS_COPY(oldLogE2, oldLogE, CC*nbEBands); + OPUS_COPY(oldLogE, oldBandE, CC*nbEBands); + } else { + for (i=0;iconsec_transient++; + else + st->consec_transient=0; + st->rng = enc->rng; + + /* If there's any room left (can only happen for very high rates), + it's already filled with zeros */ + ec_enc_done(enc); +#ifdef ENABLE_QEXT + ec_enc_done(&ext_enc); + if (qext_bytes > 0) + nbCompressedBytes += padding_len_bytes+2+qext_bytes; + if (qext_bytes) st->rng = st->rng ^ ext_enc.rng; + if (ec_get_error(&ext_enc)) + return OPUS_INTERNAL_ERROR; +#endif +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) + if (st->signalling) + nbCompressedBytes++; +#endif + + RESTORE_STACK; + if (ec_get_error(enc)) + return OPUS_INTERNAL_ERROR; + else + return nbCompressedBytes; +} + + +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) + +#if defined(FIXED_POINT) && !defined(ENABLE_RES24) +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} +#else +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(opus_res, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(in, C*N, opus_res); + + for (j=0;jchannels; + N = frame_size; + ALLOC(in, C*N, opus_res); + + for (j=0;jchannels; + N = frame_size; + ALLOC(in, C*N, opus_res); + + for (j=0;j10) + goto bad_arg; + st->complexity = value; + } + break; + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_PREDICTION_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>2) + goto bad_arg; + st->disable_pf = value<=1; + st->force_intra = value==0; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>100) + goto bad_arg; + st->loss_rate = value; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->constrained_vbr = value; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->vbr = value; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<=500 && value!=OPUS_BITRATE_MAX) + goto bad_arg; + value = IMIN(value, 750000*st->channels); + st->bitrate = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + goto bad_arg; + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + *value=st->lsb_depth; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->disable_inv = value; + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->disable_inv; + } + break; +#ifdef ENABLE_QEXT + case OPUS_SET_QEXT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->enable_qext = value; + } + break; + case OPUS_GET_QEXT_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->enable_qext; + } + break; +#endif + case OPUS_RESET_STATE: + { + int i; + celt_glog *oldBandE, *oldLogE, *oldLogE2; + oldBandE = (celt_glog*)(st->in_mem+st->channels*(st->mode->overlap+QEXT_SCALE2(COMBFILTER_MAXPERIOD, st->qext_scale))); + oldLogE = oldBandE + st->channels*st->mode->nbEBands; + oldLogE2 = oldLogE + st->channels*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->ENCODER_RESET_START, + opus_custom_encoder_get_size(st->mode, st->channels)- + ((char*)&st->ENCODER_RESET_START - (char*)st)); + for (i=0;ichannels*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-GCONST(28.f); + st->vbr_offset = 0; + st->delayedIntra = 1; + st->spread_decision = SPREAD_NORMAL; + st->tonal_average = 256; + st->hf_average = 0; + st->tapset_decision = 0; + } + break; +#if defined(CUSTOM_MODES) || defined(ENABLE_OPUS_CUSTOM_API) + case CELT_SET_INPUT_CLIPPING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->clip = value; + } + break; +#endif + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case CELT_SET_ANALYSIS_REQUEST: + { + AnalysisInfo *info = va_arg(ap, AnalysisInfo *); + if (info) + OPUS_COPY(&st->analysis, info, 1); + } + break; + case CELT_SET_SILK_INFO_REQUEST: + { + SILKInfo *info = va_arg(ap, SILKInfo *); + if (info) + OPUS_COPY(&st->silk_info, info, 1); + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + celt_glog *value = va_arg(ap, celt_glog*); + st->energy_mask = value; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.ps1 b/code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.ps1 new file mode 100644 index 000000000..2e1101ea0 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.ps1 @@ -0,0 +1,195 @@ +# PowerShell script to copy encoder source files in addition to decoder files +# This script should be run after copy_sources.sh to add encoder support + +$ErrorActionPreference = "Stop" + +$ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path +$ProjectRoot = Split-Path -Parent $ScriptDir + +Write-Host "==========================================" +Write-Host "Copying Opus Encoder Source Files" +Write-Host "==========================================" +Write-Host "Project root: $ProjectRoot" +Write-Host "Target directory: $ScriptDir" +Write-Host "" + +# Create directories if they don't exist +$dirs = @( + "$ScriptDir\src_lib", + "$ScriptDir\celt", + "$ScriptDir\silk", + "$ScriptDir\silk\fixed" +) + +foreach ($dir in $dirs) { + if (-not (Test-Path $dir)) { + New-Item -ItemType Directory -Path $dir -Force | Out-Null + } +} + +Write-Host "Created directory structure" +Write-Host "" + +# Copy Opus API layer encoder sources +Write-Host "Copying Opus encoder API sources..." +$OpusEncoderFiles = @( + "src\opus_encoder.c", + "src\opus_multistream_encoder.c", + "src\opus_projection_encoder.c", + "src\analysis.c", + "src\mlp.c", + "src\mlp_data.c" +) + +foreach ($file in $OpusEncoderFiles) { + $srcPath = Join-Path $ProjectRoot $file + $dstPath = Join-Path $ScriptDir "src_lib" (Split-Path -Leaf $file) + if (Test-Path $srcPath) { + Copy-Item $srcPath $dstPath -Force + Write-Host " Copied: $file" + } else { + Write-Host " Warning: $file not found" + } +} + +Write-Host "" + +# Copy CELT encoder sources +Write-Host "Copying CELT encoder sources..." +$CeltEncoderFiles = @( + "celt\celt_encoder.c" +) + +foreach ($file in $CeltEncoderFiles) { + $srcPath = Join-Path $ProjectRoot $file + $dstPath = Join-Path $ScriptDir "celt" (Split-Path -Leaf $file) + if (Test-Path $srcPath) { + Copy-Item $srcPath $dstPath -Force + Write-Host " Copied: $file" + } else { + Write-Host " Warning: $file not found" + } +} + +Write-Host "" + +# Copy SILK encoder core sources +Write-Host "Copying SILK encoder core sources..." +$SilkEncoderCoreFiles = @( + "silk\enc_API.c", + "silk\init_encoder.c", + "silk\encode_indices.c", + "silk\encode_pulses.c", + "silk\NLSF_encode.c", + "silk\NLSF_VQ.c", + "silk\NLSF_del_dec_quant.c", + "silk\process_NLSFs.c" +) + +foreach ($file in $SilkEncoderCoreFiles) { + $srcPath = Join-Path $ProjectRoot $file + $dstPath = Join-Path $ScriptDir "silk" (Split-Path -Leaf $file) + if (Test-Path $srcPath) { + Copy-Item $srcPath $dstPath -Force + Write-Host " Copied: $file" + } else { + Write-Host " Warning: $file not found" + } +} + +# Copy SILK encoder stereo processing +Write-Host "Copying SILK encoder stereo processing..." +$SilkEncoderStereoFiles = @( + "silk\stereo_LR_to_MS.c", + "silk\stereo_encode_pred.c", + "silk\stereo_quant_pred.c" +) + +foreach ($file in $SilkEncoderStereoFiles) { + $srcPath = Join-Path $ProjectRoot $file + $dstPath = Join-Path $ScriptDir "silk" (Split-Path -Leaf $file) + if (Test-Path $srcPath) { + Copy-Item $srcPath $dstPath -Force + Write-Host " Copied: $file" + } else { + Write-Host " Warning: $file not found" + } +} + +# Copy SILK encoder NSQ (Noise Shaping Quantization) +Write-Host "Copying SILK encoder NSQ sources..." +$SilkEncoderNsqFiles = @( + "silk\NSQ.c", + "silk\NSQ_del_dec.c" +) + +foreach ($file in $SilkEncoderNsqFiles) { + $srcPath = Join-Path $ProjectRoot $file + $dstPath = Join-Path $ScriptDir "silk" (Split-Path -Leaf $file) + if (Test-Path $srcPath) { + Copy-Item $srcPath $dstPath -Force + Write-Host " Copied: $file" + } else { + Write-Host " Warning: $file not found" + } +} + +# Copy SILK encoder control and analysis +Write-Host "Copying SILK encoder control and analysis..." +$SilkEncoderControlFiles = @( + "silk\quant_LTP_gains.c", + "silk\VQ_WMat_EC.c", + "silk\VAD.c", + "silk\control_audio_bandwidth.c", + "silk\control_SNR.c", + "silk\control_codec.c", + "silk\check_control_input.c", + "silk\ana_filt_bank_1.c" +) + +foreach ($file in $SilkEncoderControlFiles) { + $srcPath = Join-Path $ProjectRoot $file + $dstPath = Join-Path $ScriptDir "silk" (Split-Path -Leaf $file) + if (Test-Path $srcPath) { + Copy-Item $srcPath $dstPath -Force + Write-Host " Copied: $file" + } else { + Write-Host " Warning: $file not found" + } +} + +# Copy SILK fixed-point encoder implementation +Write-Host "Copying SILK fixed-point encoder implementation..." +$fixedDir = Join-Path $ProjectRoot "silk\fixed" +$targetFixedDir = Join-Path $ScriptDir "silk\fixed" +if (Test-Path $fixedDir) { + $fixedFiles = Get-ChildItem -Path $fixedDir -Filter "*.c" -File + foreach ($file in $fixedFiles) { + Copy-Item $file.FullName $targetFixedDir -Force + } + $fixedHeaders = Get-ChildItem -Path $fixedDir -Filter "*.h" -File + foreach ($file in $fixedHeaders) { + Copy-Item $file.FullName $targetFixedDir -Force + } + Write-Host " Copied SILK fixed-point encoder sources and headers" +} + +# Copy SILK float encoder implementation +Write-Host "Copying SILK float encoder implementation..." +$encodeFrameFlp = Join-Path $ProjectRoot "silk\float\encode_frame_FLP.c" +$targetFlp = Join-Path $ScriptDir "silk\float\encode_frame_FLP.c" +if (Test-Path $encodeFrameFlp) { + Copy-Item $encodeFrameFlp $targetFlp -Force + Write-Host " Copied: silk\float\encode_frame_FLP.c" +} + +Write-Host "" +Write-Host "==========================================" +Write-Host "Encoder source files copied successfully!" +Write-Host "==========================================" +Write-Host "" +Write-Host "Next steps:" +Write-Host " 1. Review copied files" +Write-Host " 2. Run: make -f Makefile.libopus" +Write-Host " 3. This will create libopus_min.a with encoder support" +Write-Host "" diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.sh b/code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.sh new file mode 100644 index 000000000..b3fb2dcf6 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/copy_encoder_sources.sh @@ -0,0 +1,166 @@ +#!/bin/bash +# Script to copy encoder source files in addition to decoder files +# This script should be run after copy_sources.sh to add encoder support + +set -e + +SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" +PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)" + +echo "==========================================" +echo "Copying Opus Encoder Source Files" +echo "==========================================" +echo "Project root: $PROJECT_ROOT" +echo "Target directory: $SCRIPT_DIR" +echo "" + +# Create directories if they don't exist +mkdir -p "$SCRIPT_DIR/src_lib" +mkdir -p "$SCRIPT_DIR/celt" +mkdir -p "$SCRIPT_DIR/silk" +mkdir -p "$SCRIPT_DIR/silk/fixed" + +echo "Created directory structure" +echo "" + +# Copy Opus API layer encoder sources +echo "Copying Opus encoder API sources..." +OPUS_ENCODER_FILES=( + "src/opus_encoder.c" + "src/opus_multistream_encoder.c" + "src/opus_projection_encoder.c" + "src/analysis.c" + "src/mlp.c" + "src/mlp_data.c" +) + +for file in "${OPUS_ENCODER_FILES[@]}"; do + if [ -f "$PROJECT_ROOT/$file" ]; then + cp "$PROJECT_ROOT/$file" "$SCRIPT_DIR/src_lib/" + echo " Copied: $file" + else + echo " Warning: $file not found" + fi +done + +echo "" + +# Copy CELT encoder sources +echo "Copying CELT encoder sources..." +CELT_ENCODER_FILES=( + "celt/celt_encoder.c" +) + +for file in "${CELT_ENCODER_FILES[@]}"; do + if [ -f "$PROJECT_ROOT/$file" ]; then + cp "$PROJECT_ROOT/$file" "$SCRIPT_DIR/celt/" + echo " Copied: $file" + else + echo " Warning: $file not found" + fi +done + +echo "" + +# Copy SILK encoder core sources +echo "Copying SILK encoder core sources..." +SILK_ENCODER_CORE_FILES=( + "silk/enc_API.c" + "silk/init_encoder.c" + "silk/encode_indices.c" + "silk/encode_pulses.c" + "silk/NLSF_encode.c" + "silk/NLSF_VQ.c" + "silk/NLSF_del_dec_quant.c" + "silk/process_NLSFs.c" +) + +for file in "${SILK_ENCODER_CORE_FILES[@]}"; do + if [ -f "$PROJECT_ROOT/$file" ]; then + cp "$PROJECT_ROOT/$file" "$SCRIPT_DIR/silk/" + echo " Copied: $file" + else + echo " Warning: $file not found" + fi +done + +# Copy SILK encoder stereo processing +echo "Copying SILK encoder stereo processing..." +SILK_ENCODER_STEREO_FILES=( + "silk/stereo_LR_to_MS.c" + "silk/stereo_encode_pred.c" + "silk/stereo_quant_pred.c" +) + +for file in "${SILK_ENCODER_STEREO_FILES[@]}"; do + if [ -f "$PROJECT_ROOT/$file" ]; then + cp "$PROJECT_ROOT/$file" "$SCRIPT_DIR/silk/" + echo " Copied: $file" + else + echo " Warning: $file not found" + fi +done + +# Copy SILK encoder NSQ (Noise Shaping Quantization) +echo "Copying SILK encoder NSQ sources..." +SILK_ENCODER_NSQ_FILES=( + "silk/NSQ.c" + "silk/NSQ_del_dec.c" +) + +for file in "${SILK_ENCODER_NSQ_FILES[@]}"; do + if [ -f "$PROJECT_ROOT/$file" ]; then + cp "$PROJECT_ROOT/$file" "$SCRIPT_DIR/silk/" + echo " Copied: $file" + else + echo " Warning: $file not found" + fi +done + +# Copy SILK encoder control and analysis +echo "Copying SILK encoder control and analysis..." +SILK_ENCODER_CONTROL_FILES=( + "silk/quant_LTP_gains.c" + "silk/VQ_WMat_EC.c" + "silk/VAD.c" + "silk/control_audio_bandwidth.c" + "silk/control_SNR.c" + "silk/control_codec.c" + "silk/check_control_input.c" + "silk/ana_filt_bank_1.c" +) + +for file in "${SILK_ENCODER_CONTROL_FILES[@]}"; do + if [ -f "$PROJECT_ROOT/$file" ]; then + cp "$PROJECT_ROOT/$file" "$SCRIPT_DIR/silk/" + echo " Copied: $file" + else + echo " Warning: $file not found" + fi +done + +# Copy SILK fixed-point encoder implementation +echo "Copying SILK fixed-point encoder implementation..." +if [ -d "$PROJECT_ROOT/silk/fixed" ]; then + find "$PROJECT_ROOT/silk/fixed" -name "*.c" -exec cp {} "$SCRIPT_DIR/silk/fixed/" \; + find "$PROJECT_ROOT/silk/fixed" -name "*.h" -exec cp {} "$SCRIPT_DIR/silk/fixed/" \; + echo " Copied SILK fixed-point encoder sources and headers" +fi + +# Copy SILK float encoder implementation (encode_frame_FLP.c should already exist) +echo "Copying SILK float encoder implementation..." +if [ -f "$PROJECT_ROOT/silk/float/encode_frame_FLP.c" ]; then + cp "$PROJECT_ROOT/silk/float/encode_frame_FLP.c" "$SCRIPT_DIR/silk/float/" + echo " Copied: silk/float/encode_frame_FLP.c" +fi + +echo "" +echo "==========================================" +echo "Encoder source files copied successfully!" +echo "==========================================" +echo "" +echo "Next steps:" +echo " 1. Review copied files" +echo " 2. Run: make -f Makefile.libopus" +echo " 3. This will create libopus_min.a with encoder support" +echo "" diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/copy_missing_headers.sh b/code/sdk/zhihui/test_opus_frame_to_pcm/copy_missing_headers.sh new file mode 100644 index 000000000..1ba86a32a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/copy_missing_headers.sh @@ -0,0 +1,23 @@ +#!/bin/bash +# Script to copy missing header files needed for encoder support + +set -e + +SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" +PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)" + +echo "Copying missing header files..." + +# Copy SILK float headers +if [ -d "$PROJECT_ROOT/silk/float" ]; then + find "$PROJECT_ROOT/silk/float" -name "*.h" -exec cp {} "$SCRIPT_DIR/silk/float/" \; + echo " Copied SILK float headers" +fi + +# Copy SILK fixed headers +if [ -d "$PROJECT_ROOT/silk/fixed" ]; then + find "$PROJECT_ROOT/silk/fixed" -name "*.h" -exec cp {} "$SCRIPT_DIR/silk/fixed/" \; + echo " Copied SILK fixed headers" +fi + +echo "Done!" diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/include/opus_frame_to_pcm.h b/code/sdk/zhihui/test_opus_frame_to_pcm/include/opus_frame_to_pcm.h index 2ce2b7c4f..e87aad40f 100644 --- a/code/sdk/zhihui/test_opus_frame_to_pcm/include/opus_frame_to_pcm.h +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/include/opus_frame_to_pcm.h @@ -51,6 +51,42 @@ int opus_frame_to_pcm_decode_packet(const unsigned char *packet_data, int packet int opus_frame_to_pcm_decode_file(const char *input_file, int sample_rate, int channels, const char *output_file); +/** + * @brief Encode PCM buffer (in-memory) to a single Opus packet. No file I/O. + * + * @param pcm Input: interleaved 16-bit PCM (little-endian). + * @param pcm_samples_per_channel Number of samples per channel (e.g. 320 for 20ms @ 16kHz). + * @param sample_rate Sample rate in Hz (e.g. 16000/48000). + * @param channels Channel count (1=mono, 2=stereo). + * @param bitrate Target bitrate in bps (e.g. 64000). + * @param application Opus application: 0=VOIP, 1=AUDIO, 2=RESTRICTED_LOWDELAY. + * @param out_packet Output buffer for encoded Opus packet. + * @param out_packet_max_len Capacity of out_packet in bytes. + * @param out_encoded_len [out] Number of bytes written to out_packet. May be NULL. + * + * @return 0 on success, non-zero on failure. + */ +int opus_pcm_to_frame_encode_packet(const short *pcm, int pcm_samples_per_channel, + int sample_rate, int channels, + int bitrate, int application, + unsigned char *out_packet, int out_packet_max_len, + int *out_encoded_len); + +/** + * @brief Encode PCM file to a single Opus frame file. + * + * @param input_file Input file path containing 16-bit little-endian PCM. + * @param sample_rate Sample rate in Hz (e.g. 16000/48000). + * @param channels Channel count (1=mono, 2=stereo). + * @param bitrate Target bitrate in bps (e.g. 64000). + * @param application Opus application: 0=VOIP, 1=AUDIO, 2=RESTRICTED_LOWDELAY. + * @param output_file Output Opus file path. If NULL, write to stdout. + * + * @return 0 on success, non-zero on failure. + */ +int opus_pcm_to_frame_encode_file(const char *input_file, int sample_rate, int channels, + int bitrate, int application, const char *output_file); + #ifdef __cplusplus } #endif diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_VQ.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_VQ.c new file mode 100644 index 000000000..b83182a79 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_VQ.c @@ -0,0 +1,76 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Compute quantization errors for an LPC_order element input vector for a VQ codebook */ +void silk_NLSF_VQ( + opus_int32 err_Q24[], /* O Quantization errors [K] */ + const opus_int16 in_Q15[], /* I Input vectors to be quantized [LPC_order] */ + const opus_uint8 pCB_Q8[], /* I Codebook vectors [K*LPC_order] */ + const opus_int16 pWght_Q9[], /* I Codebook weights [K*LPC_order] */ + const opus_int K, /* I Number of codebook vectors */ + const opus_int LPC_order /* I Number of LPCs */ +) +{ + opus_int i, m; + opus_int32 diff_Q15, diffw_Q24, sum_error_Q24, pred_Q24; + const opus_int16 *w_Q9_ptr; + const opus_uint8 *cb_Q8_ptr; + + celt_assert( ( LPC_order & 1 ) == 0 ); + + /* Loop over codebook */ + cb_Q8_ptr = pCB_Q8; + w_Q9_ptr = pWght_Q9; + for( i = 0; i < K; i++ ) { + sum_error_Q24 = 0; + pred_Q24 = 0; + for( m = LPC_order-2; m >= 0; m -= 2 ) { + /* Compute weighted absolute predictive quantization error for index m + 1 */ + diff_Q15 = silk_SUB_LSHIFT32( in_Q15[ m + 1 ], (opus_int32)cb_Q8_ptr[ m + 1 ], 7 ); /* range: [ -32767 : 32767 ]*/ + diffw_Q24 = silk_SMULBB( diff_Q15, w_Q9_ptr[ m + 1 ] ); + sum_error_Q24 = silk_ADD32( sum_error_Q24, silk_abs( silk_SUB_RSHIFT32( diffw_Q24, pred_Q24, 1 ) ) ); + pred_Q24 = diffw_Q24; + + /* Compute weighted absolute predictive quantization error for index m */ + diff_Q15 = silk_SUB_LSHIFT32( in_Q15[ m ], (opus_int32)cb_Q8_ptr[ m ], 7 ); /* range: [ -32767 : 32767 ]*/ + diffw_Q24 = silk_SMULBB( diff_Q15, w_Q9_ptr[ m ] ); + sum_error_Q24 = silk_ADD32( sum_error_Q24, silk_abs( silk_SUB_RSHIFT32( diffw_Q24, pred_Q24, 1 ) ) ); + pred_Q24 = diffw_Q24; + + silk_assert( sum_error_Q24 >= 0 ); + } + err_Q24[ i ] = sum_error_Q24; + cb_Q8_ptr += LPC_order; + w_Q9_ptr += LPC_order; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_del_dec_quant.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_del_dec_quant.c new file mode 100644 index 000000000..44a16acd0 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_del_dec_quant.c @@ -0,0 +1,215 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Delayed-decision quantizer for NLSF residuals */ +opus_int32 silk_NLSF_del_dec_quant( /* O Returns RD value in Q25 */ + opus_int8 indices[], /* O Quantization indices [ order ] */ + const opus_int16 x_Q10[], /* I Input [ order ] */ + const opus_int16 w_Q5[], /* I Weights [ order ] */ + const opus_uint8 pred_coef_Q8[], /* I Backward predictor coefs [ order ] */ + const opus_int16 ec_ix[], /* I Indices to entropy coding tables [ order ] */ + const opus_uint8 ec_rates_Q5[], /* I Rates [] */ + const opus_int quant_step_size_Q16, /* I Quantization step size */ + const opus_int16 inv_quant_step_size_Q6, /* I Inverse quantization step size */ + const opus_int32 mu_Q20, /* I R/D tradeoff */ + const opus_int16 order /* I Number of input values */ +) +{ + opus_int i, j, nStates, ind_tmp, ind_min_max, ind_max_min, in_Q10, res_Q10; + opus_int pred_Q10, diff_Q10, rate0_Q5, rate1_Q5; + opus_int16 out0_Q10, out1_Q10; + opus_int32 RD_tmp_Q25, min_Q25, min_max_Q25, max_min_Q25; + opus_int ind_sort[ NLSF_QUANT_DEL_DEC_STATES ]; + opus_int8 ind[ NLSF_QUANT_DEL_DEC_STATES ][ MAX_LPC_ORDER ]; + opus_int16 prev_out_Q10[ 2 * NLSF_QUANT_DEL_DEC_STATES ]; + opus_int32 RD_Q25[ 2 * NLSF_QUANT_DEL_DEC_STATES ]; + opus_int32 RD_min_Q25[ NLSF_QUANT_DEL_DEC_STATES ]; + opus_int32 RD_max_Q25[ NLSF_QUANT_DEL_DEC_STATES ]; + const opus_uint8 *rates_Q5; + + opus_int out0_Q10_table[2 * NLSF_QUANT_MAX_AMPLITUDE_EXT]; + opus_int out1_Q10_table[2 * NLSF_QUANT_MAX_AMPLITUDE_EXT]; + + for (i = -NLSF_QUANT_MAX_AMPLITUDE_EXT; i <= NLSF_QUANT_MAX_AMPLITUDE_EXT-1; i++) + { + out0_Q10 = silk_LSHIFT( i, 10 ); + out1_Q10 = silk_ADD16( out0_Q10, 1024 ); + if( i > 0 ) { + out0_Q10 = silk_SUB16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + out1_Q10 = silk_SUB16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else if( i == 0 ) { + out1_Q10 = silk_SUB16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else if( i == -1 ) { + out0_Q10 = silk_ADD16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else { + out0_Q10 = silk_ADD16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + out1_Q10 = silk_ADD16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } + out0_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_RSHIFT( silk_SMULBB( out0_Q10, quant_step_size_Q16 ), 16 ); + out1_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_RSHIFT( silk_SMULBB( out1_Q10, quant_step_size_Q16 ), 16 ); + } + + silk_assert( (NLSF_QUANT_DEL_DEC_STATES & (NLSF_QUANT_DEL_DEC_STATES-1)) == 0 ); /* must be power of two */ + + nStates = 1; + RD_Q25[ 0 ] = 0; + prev_out_Q10[ 0 ] = 0; + for( i = order - 1; i >= 0; i-- ) { + rates_Q5 = &ec_rates_Q5[ ec_ix[ i ] ]; + in_Q10 = x_Q10[ i ]; + for( j = 0; j < nStates; j++ ) { + pred_Q10 = silk_RSHIFT( silk_SMULBB( (opus_int16)pred_coef_Q8[ i ], prev_out_Q10[ j ] ), 8 ); + res_Q10 = silk_SUB16( in_Q10, pred_Q10 ); + ind_tmp = silk_RSHIFT( silk_SMULBB( inv_quant_step_size_Q6, res_Q10 ), 16 ); + ind_tmp = silk_LIMIT( ind_tmp, -NLSF_QUANT_MAX_AMPLITUDE_EXT, NLSF_QUANT_MAX_AMPLITUDE_EXT-1 ); + ind[ j ][ i ] = (opus_int8)ind_tmp; + + /* compute outputs for ind_tmp and ind_tmp + 1 */ + out0_Q10 = out0_Q10_table[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE_EXT ]; + out1_Q10 = out1_Q10_table[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE_EXT ]; + + out0_Q10 = silk_ADD16( out0_Q10, pred_Q10 ); + out1_Q10 = silk_ADD16( out1_Q10, pred_Q10 ); + prev_out_Q10[ j ] = out0_Q10; + prev_out_Q10[ j + nStates ] = out1_Q10; + + /* compute RD for ind_tmp and ind_tmp + 1 */ + if( ind_tmp + 1 >= NLSF_QUANT_MAX_AMPLITUDE ) { + if( ind_tmp + 1 == NLSF_QUANT_MAX_AMPLITUDE ) { + rate0_Q5 = rates_Q5[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE ]; + rate1_Q5 = 280; + } else { + rate0_Q5 = silk_SMLABB( 280 - 43 * NLSF_QUANT_MAX_AMPLITUDE, 43, ind_tmp ); + rate1_Q5 = silk_ADD16( rate0_Q5, 43 ); + } + } else if( ind_tmp <= -NLSF_QUANT_MAX_AMPLITUDE ) { + if( ind_tmp == -NLSF_QUANT_MAX_AMPLITUDE ) { + rate0_Q5 = 280; + rate1_Q5 = rates_Q5[ ind_tmp + 1 + NLSF_QUANT_MAX_AMPLITUDE ]; + } else { + rate0_Q5 = silk_SMLABB( 280 - 43 * NLSF_QUANT_MAX_AMPLITUDE, -43, ind_tmp ); + rate1_Q5 = silk_SUB16( rate0_Q5, 43 ); + } + } else { + rate0_Q5 = rates_Q5[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE ]; + rate1_Q5 = rates_Q5[ ind_tmp + 1 + NLSF_QUANT_MAX_AMPLITUDE ]; + } + RD_tmp_Q25 = RD_Q25[ j ]; + diff_Q10 = silk_SUB16( in_Q10, out0_Q10 ); + RD_Q25[ j ] = silk_SMLABB( silk_MLA( RD_tmp_Q25, silk_SMULBB( diff_Q10, diff_Q10 ), w_Q5[ i ] ), mu_Q20, rate0_Q5 ); + diff_Q10 = silk_SUB16( in_Q10, out1_Q10 ); + RD_Q25[ j + nStates ] = silk_SMLABB( silk_MLA( RD_tmp_Q25, silk_SMULBB( diff_Q10, diff_Q10 ), w_Q5[ i ] ), mu_Q20, rate1_Q5 ); + } + + if( nStates <= NLSF_QUANT_DEL_DEC_STATES/2 ) { + /* double number of states and copy */ + for( j = 0; j < nStates; j++ ) { + ind[ j + nStates ][ i ] = ind[ j ][ i ] + 1; + } + nStates = silk_LSHIFT( nStates, 1 ); + for( j = nStates; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + ind[ j ][ i ] = ind[ j - nStates ][ i ]; + } + } else { + /* sort lower and upper half of RD_Q25, pairwise */ + for( j = 0; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + if( RD_Q25[ j ] > RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ] ) { + RD_max_Q25[ j ] = RD_Q25[ j ]; + RD_min_Q25[ j ] = RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ]; + RD_Q25[ j ] = RD_min_Q25[ j ]; + RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ] = RD_max_Q25[ j ]; + /* swap prev_out values */ + out0_Q10 = prev_out_Q10[ j ]; + prev_out_Q10[ j ] = prev_out_Q10[ j + NLSF_QUANT_DEL_DEC_STATES ]; + prev_out_Q10[ j + NLSF_QUANT_DEL_DEC_STATES ] = out0_Q10; + ind_sort[ j ] = j + NLSF_QUANT_DEL_DEC_STATES; + } else { + RD_min_Q25[ j ] = RD_Q25[ j ]; + RD_max_Q25[ j ] = RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ]; + ind_sort[ j ] = j; + } + } + /* compare the highest RD values of the winning half with the lowest one in the losing half, and copy if necessary */ + /* afterwards ind_sort[] will contain the indices of the NLSF_QUANT_DEL_DEC_STATES winning RD values */ + while( 1 ) { + min_max_Q25 = silk_int32_MAX; + max_min_Q25 = 0; + ind_min_max = 0; + ind_max_min = 0; + for( j = 0; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + if( min_max_Q25 > RD_max_Q25[ j ] ) { + min_max_Q25 = RD_max_Q25[ j ]; + ind_min_max = j; + } + if( max_min_Q25 < RD_min_Q25[ j ] ) { + max_min_Q25 = RD_min_Q25[ j ]; + ind_max_min = j; + } + } + if( min_max_Q25 >= max_min_Q25 ) { + break; + } + /* copy ind_min_max to ind_max_min */ + ind_sort[ ind_max_min ] = ind_sort[ ind_min_max ] ^ NLSF_QUANT_DEL_DEC_STATES; + RD_Q25[ ind_max_min ] = RD_Q25[ ind_min_max + NLSF_QUANT_DEL_DEC_STATES ]; + prev_out_Q10[ ind_max_min ] = prev_out_Q10[ ind_min_max + NLSF_QUANT_DEL_DEC_STATES ]; + RD_min_Q25[ ind_max_min ] = 0; + RD_max_Q25[ ind_min_max ] = silk_int32_MAX; + silk_memcpy( ind[ ind_max_min ], ind[ ind_min_max ], MAX_LPC_ORDER * sizeof( opus_int8 ) ); + } + /* increment index if it comes from the upper half */ + for( j = 0; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + ind[ j ][ i ] += silk_RSHIFT( ind_sort[ j ], NLSF_QUANT_DEL_DEC_STATES_LOG2 ); + } + } + } + + /* last sample: find winner, copy indices and return RD value */ + ind_tmp = 0; + min_Q25 = silk_int32_MAX; + for( j = 0; j < 2 * NLSF_QUANT_DEL_DEC_STATES; j++ ) { + if( min_Q25 > RD_Q25[ j ] ) { + min_Q25 = RD_Q25[ j ]; + ind_tmp = j; + } + } + for( j = 0; j < order; j++ ) { + indices[ j ] = ind[ ind_tmp & ( NLSF_QUANT_DEL_DEC_STATES - 1 ) ][ j ]; + silk_assert( indices[ j ] >= -NLSF_QUANT_MAX_AMPLITUDE_EXT ); + silk_assert( indices[ j ] <= NLSF_QUANT_MAX_AMPLITUDE_EXT ); + } + indices[ 0 ] += silk_RSHIFT( ind_tmp, NLSF_QUANT_DEL_DEC_STATES_LOG2 ); + silk_assert( indices[ 0 ] <= NLSF_QUANT_MAX_AMPLITUDE_EXT ); + silk_assert( min_Q25 >= 0 ); + return min_Q25; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_encode.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_encode.c new file mode 100644 index 000000000..01ac7db78 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NLSF_encode.c @@ -0,0 +1,124 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/***********************/ +/* NLSF vector encoder */ +/***********************/ +opus_int32 silk_NLSF_encode( /* O Returns RD value in Q25 */ + opus_int8 *NLSFIndices, /* I Codebook path vector [ LPC_ORDER + 1 ] */ + opus_int16 *pNLSF_Q15, /* I/O (Un)quantized NLSF vector [ LPC_ORDER ] */ + const silk_NLSF_CB_struct *psNLSF_CB, /* I Codebook object */ + const opus_int16 *pW_Q2, /* I NLSF weight vector [ LPC_ORDER ] */ + const opus_int NLSF_mu_Q20, /* I Rate weight for the RD optimization */ + const opus_int nSurvivors, /* I Max survivors after first stage */ + const opus_int signalType /* I Signal type: 0/1/2 */ +) +{ + opus_int i, s, ind1, bestIndex, prob_Q8, bits_q7; + opus_int32 W_tmp_Q9, ret; + VARDECL( opus_int32, err_Q24 ); + VARDECL( opus_int32, RD_Q25 ); + VARDECL( opus_int, tempIndices1 ); + VARDECL( opus_int8, tempIndices2 ); + opus_int16 res_Q10[ MAX_LPC_ORDER ]; + opus_int16 NLSF_tmp_Q15[ MAX_LPC_ORDER ]; + opus_int16 W_adj_Q5[ MAX_LPC_ORDER ]; + opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; + opus_int16 ec_ix[ MAX_LPC_ORDER ]; + const opus_uint8 *pCB_element, *iCDF_ptr; + const opus_int16 *pCB_Wght_Q9; + SAVE_STACK; + + celt_assert( signalType >= 0 && signalType <= 2 ); + silk_assert( NLSF_mu_Q20 <= 32767 && NLSF_mu_Q20 >= 0 ); + + /* NLSF stabilization */ + silk_NLSF_stabilize( pNLSF_Q15, psNLSF_CB->deltaMin_Q15, psNLSF_CB->order ); + + /* First stage: VQ */ + ALLOC( err_Q24, psNLSF_CB->nVectors, opus_int32 ); + silk_NLSF_VQ( err_Q24, pNLSF_Q15, psNLSF_CB->CB1_NLSF_Q8, psNLSF_CB->CB1_Wght_Q9, psNLSF_CB->nVectors, psNLSF_CB->order ); + + /* Sort the quantization errors */ + ALLOC( tempIndices1, nSurvivors, opus_int ); + silk_insertion_sort_increasing( err_Q24, tempIndices1, psNLSF_CB->nVectors, nSurvivors ); + + ALLOC( RD_Q25, nSurvivors, opus_int32 ); + ALLOC( tempIndices2, nSurvivors * MAX_LPC_ORDER, opus_int8 ); + + /* Loop over survivors */ + for( s = 0; s < nSurvivors; s++ ) { + ind1 = tempIndices1[ s ]; + + /* Residual after first stage */ + pCB_element = &psNLSF_CB->CB1_NLSF_Q8[ ind1 * psNLSF_CB->order ]; + pCB_Wght_Q9 = &psNLSF_CB->CB1_Wght_Q9[ ind1 * psNLSF_CB->order ]; + for( i = 0; i < psNLSF_CB->order; i++ ) { + NLSF_tmp_Q15[ i ] = silk_LSHIFT16( (opus_int16)pCB_element[ i ], 7 ); + W_tmp_Q9 = pCB_Wght_Q9[ i ]; + res_Q10[ i ] = (opus_int16)silk_RSHIFT( silk_SMULBB( pNLSF_Q15[ i ] - NLSF_tmp_Q15[ i ], W_tmp_Q9 ), 14 ); + W_adj_Q5[ i ] = silk_DIV32_varQ( (opus_int32)pW_Q2[ i ], silk_SMULBB( W_tmp_Q9, W_tmp_Q9 ), 21 ); + } + + /* Unpack entropy table indices and predictor for current CB1 index */ + silk_NLSF_unpack( ec_ix, pred_Q8, psNLSF_CB, ind1 ); + + /* Trellis quantizer */ + RD_Q25[ s ] = silk_NLSF_del_dec_quant( &tempIndices2[ s * MAX_LPC_ORDER ], res_Q10, W_adj_Q5, pred_Q8, ec_ix, + psNLSF_CB->ec_Rates_Q5, psNLSF_CB->quantStepSize_Q16, psNLSF_CB->invQuantStepSize_Q6, NLSF_mu_Q20, psNLSF_CB->order ); + + /* Add rate for first stage */ + iCDF_ptr = &psNLSF_CB->CB1_iCDF[ ( signalType >> 1 ) * psNLSF_CB->nVectors ]; + if( ind1 == 0 ) { + prob_Q8 = 256 - iCDF_ptr[ ind1 ]; + } else { + prob_Q8 = iCDF_ptr[ ind1 - 1 ] - iCDF_ptr[ ind1 ]; + } + bits_q7 = ( 8 << 7 ) - silk_lin2log( prob_Q8 ); + RD_Q25[ s ] = silk_SMLABB( RD_Q25[ s ], bits_q7, silk_RSHIFT( NLSF_mu_Q20, 2 ) ); + } + + /* Find the lowest rate-distortion error */ + silk_insertion_sort_increasing( RD_Q25, &bestIndex, nSurvivors, 1 ); + + NLSFIndices[ 0 ] = (opus_int8)tempIndices1[ bestIndex ]; + silk_memcpy( &NLSFIndices[ 1 ], &tempIndices2[ bestIndex * MAX_LPC_ORDER ], psNLSF_CB->order * sizeof( opus_int8 ) ); + + /* Decode */ + silk_NLSF_decode( pNLSF_Q15, NLSFIndices, psNLSF_CB ); + + ret = RD_Q25[ 0 ]; + RESTORE_STACK; + return ret; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ.c new file mode 100644 index 000000000..4ffdf3d9a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ.c @@ -0,0 +1,437 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" +#include "NSQ.h" + + +static OPUS_INLINE void silk_nsq_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ +); + +#if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE void silk_noise_shape_quantizer( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ +); +#endif + +void silk_NSQ_c +( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ + opus_int k, lag, start_idx, LSF_interpolation_flag; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + SAVE_STACK; + + NSQ->rand_seed = psIndices->Seed; + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + /* Set up pointers to start of sub frame */ + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ (( k >> 1 ) | ( 1 - LSF_interpolation_flag )) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->rewhite_flag = 1; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + } + } + + silk_nsq_scale_states( psEncC, NSQ, x16, x_sc_Q10, sLTP, sLTP_Q15, k, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType ); + + silk_noise_shape_quantizer( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, + AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, + offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->arch ); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Update lagPrev for next frame */ + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech and noise shaping signals */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; +} + +/******************************/ +/* silk_noise_shape_quantizer */ +/******************************/ + +#if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE +#endif +void silk_noise_shape_quantizer( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ +) +{ + opus_int i; + opus_int32 LTP_pred_Q13, LPC_pred_Q10, n_AR_Q12, n_LTP_Q13; + opus_int32 n_LF_Q12, r_Q10, rr_Q10, q1_Q0, q1_Q10, q2_Q10, rd1_Q20, rd2_Q20; + opus_int32 exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *psLPC_Q14, *shp_lag_ptr, *pred_lag_ptr; +#ifdef silk_short_prediction_create_arch_coef + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; +#endif + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + + /* Set up short term AR state */ + psLPC_Q14 = &NSQ->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 ]; + +#ifdef silk_short_prediction_create_arch_coef + silk_short_prediction_create_arch_coef(a_Q12_arch, a_Q12, predictLPCOrder); +#endif + + for( i = 0; i < length; i++ ) { + /* Generate dither */ + NSQ->rand_seed = silk_RAND( NSQ->rand_seed ); + + /* Short-term prediction */ + LPC_pred_Q10 = silk_noise_shape_quantizer_short_prediction(psLPC_Q14, a_Q12, a_Q12_arch, predictLPCOrder, arch); + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q13 = 2; + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ 0 ], b_Q14[ 0 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -1 ], b_Q14[ 1 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -2 ], b_Q14[ 2 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -3 ], b_Q14[ 3 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + pred_lag_ptr++; + } else { + LTP_pred_Q13 = 0; + } + + /* Noise shape feedback */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + n_AR_Q12 = silk_NSQ_noise_shape_feedback_loop(&NSQ->sDiff_shp_Q14, NSQ->sAR2_Q14, AR_shp_Q13, shapingLPCOrder, arch); + + n_AR_Q12 = silk_SMLAWB( n_AR_Q12, NSQ->sLF_AR_shp_Q14, Tilt_Q14 ); + + n_LF_Q12 = silk_SMULWB( NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - 1 ], LF_shp_Q14 ); + n_LF_Q12 = silk_SMLAWT( n_LF_Q12, NSQ->sLF_AR_shp_Q14, LF_shp_Q14 ); + + celt_assert( lag > 0 || signalType != TYPE_VOICED ); + + /* Combine prediction and noise shaping signals */ + tmp1 = silk_SUB32_ovflw( silk_LSHIFT32( LPC_pred_Q10, 2 ), n_AR_Q12 ); /* Q12 */ + tmp1 = silk_SUB32_ovflw( tmp1, n_LF_Q12 ); /* Q12 */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q13 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_SMLAWT( n_LTP_Q13, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_LSHIFT( n_LTP_Q13, 1 ); + shp_lag_ptr++; + + tmp2 = silk_SUB32( LTP_pred_Q13, n_LTP_Q13 ); /* Q13 */ + tmp1 = silk_ADD32_ovflw( tmp2, silk_LSHIFT32( tmp1, 1 ) ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 3 ); /* Q10 */ + } else { + tmp1 = silk_RSHIFT_ROUND( tmp1, 2 ); /* Q10 */ + } + + r_Q10 = silk_SUB32( x_sc_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + if( NSQ->rand_seed < 0 ) { + r_Q10 = -r_Q10; + } + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + int rdo_offset = Lambda_Q10/2 - 512; + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } + if( q1_Q0 > 0 ) { + q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q20 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == 0 ) { + q1_Q10 = offset_Q10; + q2_Q10 = silk_ADD32( q1_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q20 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == -1 ) { + q2_Q10 = offset_Q10; + q1_Q10 = silk_SUB32( q2_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q20 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else { /* Q1_Q0 < -1 */ + q1_Q10 = silk_ADD32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q20 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( -q2_Q10, Lambda_Q10 ); + } + rr_Q10 = silk_SUB32( r_Q10, q1_Q10 ); + rd1_Q20 = silk_SMLABB( rd1_Q20, rr_Q10, rr_Q10 ); + rr_Q10 = silk_SUB32( r_Q10, q2_Q10 ); + rd2_Q20 = silk_SMLABB( rd2_Q20, rr_Q10, rr_Q10 ); + + if( rd2_Q20 < rd1_Q20 ) { + q1_Q10 = q2_Q10; + } + + pulses[ i ] = (opus_int8)silk_RSHIFT_ROUND( q1_Q10, 10 ); + + /* Excitation */ + exc_Q14 = silk_LSHIFT( q1_Q10, 4 ); + if ( NSQ->rand_seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD_LSHIFT32( exc_Q14, LTP_pred_Q13, 1 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, silk_LSHIFT32( LPC_pred_Q10, 4 ) ); + + /* Scale XQ back to normal level before saving */ + xq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( xq_Q14, Gain_Q10 ), 8 ) ); + + /* Update states */ + psLPC_Q14++; + *psLPC_Q14 = xq_Q14; + NSQ->sDiff_shp_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_sc_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( NSQ->sDiff_shp_Q14, silk_LSHIFT32( n_AR_Q12, 2 ) ); + NSQ->sLF_AR_shp_Q14 = sLF_AR_shp_Q14; + + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx ] = silk_SUB32_ovflw(sLF_AR_shp_Q14, silk_LSHIFT32(n_LF_Q12, 2)); + sLTP_Q15[ NSQ->sLTP_buf_idx ] = silk_LSHIFT( LPC_exc_Q14, 1 ); + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Make dither dependent on quantized signal */ + NSQ->rand_seed = silk_ADD32_ovflw( NSQ->rand_seed, pulses[ i ] ); + } + + /* Update LPC synth buffer */ + silk_memcpy( NSQ->sLPC_Q14, &NSQ->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); +} + +static OPUS_INLINE void silk_nsq_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ +) +{ + opus_int i, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + for( i = 0; i < psEncC->subfr_length; i++ ) { + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + silk_assert( i < MAX_FRAME_LENGTH ); + sLTP_Q15[ i ] = silk_SMULWB( inv_gain_Q31, sLTP[ i ] ); + } + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + NSQ->sLF_AR_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sLF_AR_shp_Q14 ); + NSQ->sDiff_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sDiff_shp_Q14 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + NSQ->sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLPC_Q14[ i ] ); + } + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + NSQ->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sAR2_Q14[ i ] ); + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ_del_dec.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ_del_dec.c new file mode 100644 index 000000000..31ac11af4 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/NSQ_del_dec.c @@ -0,0 +1,730 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" +#include "NSQ.h" + + +typedef struct { + opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ]; + opus_int32 RandState[ DECISION_DELAY ]; + opus_int32 Q_Q10[ DECISION_DELAY ]; + opus_int32 Xq_Q14[ DECISION_DELAY ]; + opus_int32 Pred_Q15[ DECISION_DELAY ]; + opus_int32 Shape_Q14[ DECISION_DELAY ]; + opus_int32 sAR2_Q14[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; + opus_int32 Seed; + opus_int32 SeedInit; + opus_int32 RD_Q10; +} NSQ_del_dec_struct; + +typedef struct { + opus_int32 Q_Q10; + opus_int32 RD_Q10; + opus_int32 xq_Q14; + opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; + opus_int32 sLTP_shp_Q14; + opus_int32 LPC_exc_Q14; +} NSQ_sample_struct; + +typedef NSQ_sample_struct NSQ_sample_pair[ 2 ]; + +static OPUS_INLINE void silk_nsq_del_dec_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + opus_int nStatesDelayedDecision, /* I Number of del dec states */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +); + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay, /* I */ + int arch /* I */ +); + +void silk_NSQ_del_dec_c( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ + opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; + opus_int last_smple_idx, smpl_buf_idx, decisionDelay; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + opus_int32 RDmin_Q10, Gain_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + VARDECL( opus_int32, delayedGain_Q10 ); + VARDECL( NSQ_del_dec_struct, psDelDec ); + NSQ_del_dec_struct *psDD; + SAVE_STACK; + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + /* Initialize delayed decision states */ + ALLOC( psDelDec, psEncC->nStatesDelayedDecision, NSQ_del_dec_struct ); + silk_memset( psDelDec, 0, psEncC->nStatesDelayedDecision * sizeof( NSQ_del_dec_struct ) ); + for( k = 0; k < psEncC->nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psDD->Seed = ( k + psIndices->Seed ) & 3; + psDD->SeedInit = psDD->Seed; + psDD->RD_Q10 = 0; + psDD->LF_AR_Q14 = NSQ->sLF_AR_shp_Q14; + psDD->Diff_Q14 = NSQ->sDiff_shp_Q14; + psDD->Shape_Q14[ 0 ] = NSQ->sLTP_shp_Q14[ psEncC->ltp_mem_length - 1 ]; + silk_memcpy( psDD->sLPC_Q14, NSQ->sLPC_Q14, NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + silk_memcpy( psDD->sAR2_Q14, NSQ->sAR2_Q14, sizeof( NSQ->sAR2_Q14 ) ); + } + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + smpl_buf_idx = 0; /* index of oldest samples */ + + decisionDelay = silk_min_int( DECISION_DELAY, psEncC->subfr_length ); + + /* For voiced frames limit the decision delay to lower than the pitch lag */ + if( psIndices->signalType == TYPE_VOICED ) { + for( k = 0; k < psEncC->nb_subfr; k++ ) { + decisionDelay = silk_min_int( decisionDelay, pitchL[ k ] - LTP_ORDER / 2 - 1 ); + } + } else { + if( lag > 0 ) { + decisionDelay = silk_min_int( decisionDelay, lag - LTP_ORDER / 2 - 1 ); + } + } + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + ALLOC( delayedGain_Q10, DECISION_DELAY, opus_int32 ); + /* Set up pointers to start of sub frame */ + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + subfr = 0; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ ( ( k >> 1 ) | ( 1 - LSF_interpolation_flag ) ) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + if( k == 2 ) { + /* RESET DELAYED DECISIONS */ + /* Find winner */ + RDmin_Q10 = psDelDec[ 0 ].RD_Q10; + Winner_ind = 0; + for( i = 1; i < psEncC->nStatesDelayedDecision; i++ ) { + if( psDelDec[ i ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psDelDec[ i ].RD_Q10; + Winner_ind = i; + } + } + for( i = 0; i < psEncC->nStatesDelayedDecision; i++ ) { + if( i != Winner_ind ) { + psDelDec[ i ].RD_Q10 += ( silk_int32_MAX >> 4 ); + silk_assert( psDelDec[ i ].RD_Q10 >= 0 ); + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + last_smple_idx = smpl_buf_idx + decisionDelay; + for( i = 0; i < decisionDelay; i++ ) { + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gains_Q16[ 1 ] ), 14 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDD->Shape_Q14[ last_smple_idx ]; + } + + subfr = 0; + } + + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + NSQ->rewhite_flag = 1; + } + } + + silk_nsq_del_dec_scale_states( psEncC, NSQ, psDelDec, x16, x_sc_Q10, sLTP, sLTP_Q15, k, + psEncC->nStatesDelayedDecision, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType, decisionDelay ); + + silk_noise_shape_quantizer_del_dec( NSQ, psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, + delayedGain_Q10, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], + Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, + psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay, psEncC->arch ); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Find winner */ + RDmin_Q10 = psDelDec[ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < psEncC->nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psDelDec[ k ].RD_Q10; + Winner_ind = k; + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + psIndices->Seed = psDD->SeedInit; + last_smple_idx = smpl_buf_idx + decisionDelay; + Gain_Q10 = silk_RSHIFT32( Gains_Q16[ psEncC->nb_subfr - 1 ], 6 ); + for( i = 0; i < decisionDelay; i++ ) { + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gain_Q10 ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDD->Shape_Q14[ last_smple_idx ]; + } + silk_memcpy( NSQ->sLPC_Q14, &psDD->sLPC_Q14[ psEncC->subfr_length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + silk_memcpy( NSQ->sAR2_Q14, psDD->sAR2_Q14, sizeof( psDD->sAR2_Q14 ) ); + + /* Update states */ + NSQ->sLF_AR_shp_Q14 = psDD->LF_AR_Q14; + NSQ->sDiff_shp_Q14 = psDD->Diff_Q14; + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech signal */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; +} + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +#ifndef OVERRIDE_silk_noise_shape_quantizer_del_dec +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay, /* I */ + int arch /* I */ +) +{ + opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; + opus_int32 Winner_rand_state; + opus_int32 LTP_pred_Q14, LPC_pred_Q14, n_AR_Q14, n_LTP_Q14; + opus_int32 n_LF_Q14, r_Q10, rr_Q10, rd1_Q10, rd2_Q10, RDmin_Q10, RDmax_Q10; + opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; +#ifdef silk_short_prediction_create_arch_coef + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; +#endif + + VARDECL( NSQ_sample_pair, psSampleState ); + NSQ_del_dec_struct *psDD; + NSQ_sample_struct *psSS; + SAVE_STACK; + + celt_assert( nStatesDelayedDecision > 0 ); + ALLOC( psSampleState, nStatesDelayedDecision, NSQ_sample_pair ); + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + +#ifdef silk_short_prediction_create_arch_coef + silk_short_prediction_create_arch_coef(a_Q12_arch, a_Q12, predictLPCOrder); +#endif + + for( i = 0; i < length; i++ ) { + /* Perform common calculations used in all states */ + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q14 = 2; + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ 0 ], b_Q14[ 0 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -1 ], b_Q14[ 1 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -2 ], b_Q14[ 2 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -3 ], b_Q14[ 3 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + LTP_pred_Q14 = silk_LSHIFT( LTP_pred_Q14, 1 ); /* Q13 -> Q14 */ + pred_lag_ptr++; + } else { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ + shp_lag_ptr++; + } else { + n_LTP_Q14 = 0; + } + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + /* Delayed decision state */ + psDD = &psDelDec[ k ]; + + /* Sample state */ + psSS = psSampleState[ k ]; + + /* Generate dither */ + psDD->Seed = silk_RAND( psDD->Seed ); + + /* Pointer used in short term prediction and shaping */ + psLPC_Q14 = &psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 + i ]; + /* Short-term prediction */ + LPC_pred_Q14 = silk_noise_shape_quantizer_short_prediction(psLPC_Q14, a_Q12, a_Q12_arch, predictLPCOrder, arch); + LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ + + /* Noise shape feedback */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + /* Output of lowpass section */ + tmp2 = silk_SMLAWB( psDD->Diff_Q14, psDD->sAR2_Q14[ 0 ], warping_Q16 ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ 1 ], tmp2), warping_Q16 ); + psDD->sAR2_Q14[ 0 ] = tmp2; + n_AR_Q14 = silk_RSHIFT( shapingLPCOrder, 1 ); + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ 0 ] ); + /* Loop over allpass sections */ + for( j = 2; j < shapingLPCOrder; j += 2 ) { + /* Output of allpass section */ + tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ j + 0 ], tmp1), warping_Q16 ); + psDD->sAR2_Q14[ j - 1 ] = tmp1; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ j - 1 ] ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], silk_SUB32_ovflw(psDD->sAR2_Q14[ j + 1 ], tmp2), warping_Q16 ); + psDD->sAR2_Q14[ j + 0 ] = tmp2; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ j ] ); + } + psDD->sAR2_Q14[ shapingLPCOrder - 1 ] = tmp1; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ shapingLPCOrder - 1 ] ); + + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 1 ); /* Q11 -> Q12 */ + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, psDD->LF_AR_Q14, Tilt_Q14 ); /* Q12 */ + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 2 ); /* Q12 -> Q14 */ + + n_LF_Q14 = silk_SMULWB( psDD->Shape_Q14[ *smpl_buf_idx ], LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_SMLAWT( n_LF_Q14, psDD->LF_AR_Q14, LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_LSHIFT( n_LF_Q14, 2 ); /* Q12 -> Q14 */ + + /* Input minus prediction plus noise feedback */ + /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ + tmp1 = silk_ADD_SAT32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ + tmp2 = silk_ADD32_ovflw( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ + tmp1 = silk_SUB_SAT32( tmp2, tmp1 ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 4 ); /* Q10 */ + + r_Q10 = silk_SUB32( x_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + if ( psDD->Seed < 0 ) { + r_Q10 = -r_Q10; + } + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + int rdo_offset = Lambda_Q10/2 - 512; + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } + if( q1_Q0 > 0 ) { + q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == 0 ) { + q1_Q10 = offset_Q10; + q2_Q10 = silk_ADD32( q1_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == -1 ) { + q2_Q10 = offset_Q10; + q1_Q10 = silk_SUB32( q2_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else { /* q1_Q0 < -1 */ + q1_Q10 = silk_ADD32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( -q2_Q10, Lambda_Q10 ); + } + rr_Q10 = silk_SUB32( r_Q10, q1_Q10 ); + rd1_Q10 = silk_RSHIFT( silk_SMLABB( rd1_Q10, rr_Q10, rr_Q10 ), 10 ); + rr_Q10 = silk_SUB32( r_Q10, q2_Q10 ); + rd2_Q10 = silk_RSHIFT( silk_SMLABB( rd2_Q10, rr_Q10, rr_Q10 ), 10 ); + + if( rd1_Q10 < rd2_Q10 ) { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 0 ].Q_Q10 = q1_Q10; + psSS[ 1 ].Q_Q10 = q2_Q10; + } else { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 0 ].Q_Q10 = q2_Q10; + psSS[ 1 ].Q_Q10 = q1_Q10; + } + + /* Update states for best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 0 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + psSS[ 0 ].Diff_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( psSS[ 0 ].Diff_Q14, n_AR_Q14 ); + psSS[ 0 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 0 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 0 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 0 ].xq_Q14 = xq_Q14; + + /* Update states for second best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 1 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32_ovflw( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + psSS[ 1 ].Diff_Q14 = silk_SUB32_ovflw( xq_Q14, silk_LSHIFT32( x_Q10[ i ], 4 ) ); + sLF_AR_shp_Q14 = silk_SUB32_ovflw( psSS[ 1 ].Diff_Q14, n_AR_Q14 ); + psSS[ 1 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 1 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 1 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 1 ].xq_Q14 = xq_Q14; + } + + *smpl_buf_idx = ( *smpl_buf_idx - 1 ) % DECISION_DELAY; + if( *smpl_buf_idx < 0 ) *smpl_buf_idx += DECISION_DELAY; + last_smple_idx = ( *smpl_buf_idx + decisionDelay ) % DECISION_DELAY; + + /* Find winner */ + RDmin_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + if( psSampleState[ k ][ 0 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + Winner_ind = k; + } + } + + /* Increase RD values of expired states */ + Winner_rand_state = psDelDec[ Winner_ind ].RandState[ last_smple_idx ]; + for( k = 0; k < nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RandState[ last_smple_idx ] != Winner_rand_state ) { + psSampleState[ k ][ 0 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 0 ].RD_Q10, silk_int32_MAX >> 4 ); + psSampleState[ k ][ 1 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 1 ].RD_Q10, silk_int32_MAX >> 4 ); + silk_assert( psSampleState[ k ][ 0 ].RD_Q10 >= 0 ); + } + } + + /* Find worst in first set and best in second set */ + RDmax_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + RDmin_Q10 = psSampleState[ 0 ][ 1 ].RD_Q10; + RDmax_ind = 0; + RDmin_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + /* find worst in first set */ + if( psSampleState[ k ][ 0 ].RD_Q10 > RDmax_Q10 ) { + RDmax_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + RDmax_ind = k; + } + /* find best in second set */ + if( psSampleState[ k ][ 1 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 1 ].RD_Q10; + RDmin_ind = k; + } + } + + /* Replace a state if best from second set outperforms worst in first set */ + if( RDmin_Q10 < RDmax_Q10 ) { + silk_memcpy( ( (opus_int32 *)&psDelDec[ RDmax_ind ] ) + i, + ( (opus_int32 *)&psDelDec[ RDmin_ind ] ) + i, sizeof( NSQ_del_dec_struct ) - i * sizeof( opus_int32) ); + silk_memcpy( &psSampleState[ RDmax_ind ][ 0 ], &psSampleState[ RDmin_ind ][ 1 ], sizeof( NSQ_sample_struct ) ); + } + + /* Write samples from winner to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + if( subfr > 0 || i >= decisionDelay ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + xq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], delayedGain_Q10[ last_smple_idx ] ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay ] = psDD->Shape_Q14[ last_smple_idx ]; + sLTP_Q15[ NSQ->sLTP_buf_idx - decisionDelay ] = psDD->Pred_Q15[ last_smple_idx ]; + } + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Update states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psSS = &psSampleState[ k ][ 0 ]; + psDD->LF_AR_Q14 = psSS->LF_AR_Q14; + psDD->Diff_Q14 = psSS->Diff_Q14; + psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH + i ] = psSS->xq_Q14; + psDD->Xq_Q14[ *smpl_buf_idx ] = psSS->xq_Q14; + psDD->Q_Q10[ *smpl_buf_idx ] = psSS->Q_Q10; + psDD->Pred_Q15[ *smpl_buf_idx ] = silk_LSHIFT32( psSS->LPC_exc_Q14, 1 ); + psDD->Shape_Q14[ *smpl_buf_idx ] = psSS->sLTP_shp_Q14; + psDD->Seed = silk_ADD32_ovflw( psDD->Seed, silk_RSHIFT_ROUND( psSS->Q_Q10, 10 ) ); + psDD->RandState[ *smpl_buf_idx ] = psDD->Seed; + psDD->RD_Q10 = psSS->RD_Q10; + } + delayedGain_Q10[ *smpl_buf_idx ] = Gain_Q10; + } + /* Update LPC states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + silk_memcpy( psDD->sLPC_Q14, &psDD->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + } + RESTORE_STACK; +} +#endif /* OVERRIDE_silk_noise_shape_quantizer_del_dec */ + +static OPUS_INLINE void silk_nsq_del_dec_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + opus_int nStatesDelayedDecision, /* I Number of del dec states */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +) +{ + opus_int i, k, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + NSQ_del_dec_struct *psDD; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + for( i = 0; i < psEncC->subfr_length; i++ ) { + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + silk_assert( i < MAX_FRAME_LENGTH ); + sLTP_Q15[ i ] = silk_SMULWB( inv_gain_Q31, sLTP[ i ] ); + } + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx - decisionDelay; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + + /* Scale scalar states */ + psDD->LF_AR_Q14 = silk_SMULWW( gain_adj_Q16, psDD->LF_AR_Q14 ); + psDD->Diff_Q14 = silk_SMULWW( gain_adj_Q16, psDD->Diff_Q14 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + psDD->sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->sLPC_Q14[ i ] ); + } + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + psDD->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->sAR2_Q14[ i ] ); + } + for( i = 0; i < DECISION_DELAY; i++ ) { + psDD->Pred_Q15[ i ] = silk_SMULWW( gain_adj_Q16, psDD->Pred_Q15[ i ] ); + psDD->Shape_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->Shape_Q14[ i ] ); + } + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/VAD.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/VAD.c new file mode 100644 index 000000000..d0cda5216 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/VAD.c @@ -0,0 +1,360 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/* Silk VAD noise level estimation */ +# if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE void silk_VAD_GetNoiseLevels( + const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +); +#endif + +/**********************************/ +/* Initialization of the Silk VAD */ +/**********************************/ +opus_int silk_VAD_Init( /* O Return value, 0 if success */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +) +{ + opus_int b, ret = 0; + + /* reset state memory */ + silk_memset( psSilk_VAD, 0, sizeof( silk_VAD_state ) ); + + /* init noise levels */ + /* Initialize array with approx pink noise levels (psd proportional to inverse of frequency) */ + for( b = 0; b < VAD_N_BANDS; b++ ) { + psSilk_VAD->NoiseLevelBias[ b ] = silk_max_32( silk_DIV32_16( VAD_NOISE_LEVELS_BIAS, b + 1 ), 1 ); + } + + /* Initialize state */ + for( b = 0; b < VAD_N_BANDS; b++ ) { + psSilk_VAD->NL[ b ] = silk_MUL( 100, psSilk_VAD->NoiseLevelBias[ b ] ); + psSilk_VAD->inv_NL[ b ] = silk_DIV32( silk_int32_MAX, psSilk_VAD->NL[ b ] ); + } + psSilk_VAD->counter = 15; + + /* init smoothed energy-to-noise ratio*/ + for( b = 0; b < VAD_N_BANDS; b++ ) { + psSilk_VAD->NrgRatioSmth_Q8[ b ] = 100 * 256; /* 100 * 256 --> 20 dB SNR */ + } + + return( ret ); +} + +/* Weighting factors for tilt measure */ +static const opus_int32 tiltWeights[ VAD_N_BANDS ] = { 30000, 6000, -12000, -12000 }; + +/***************************************/ +/* Get the speech activity level in Q8 */ +/***************************************/ +opus_int silk_VAD_GetSA_Q8_c( /* O Return value, 0 if success */ + silk_encoder_state *psEncC, /* I/O Encoder state */ + const opus_int16 pIn[] /* I PCM input */ +) +{ + opus_int SA_Q15, pSNR_dB_Q7, input_tilt; + opus_int decimated_framelength1, decimated_framelength2; + opus_int decimated_framelength; + opus_int dec_subframe_length, dec_subframe_offset, SNR_Q7, i, b, s; + opus_int32 sumSquared, smooth_coef_Q16; + opus_int16 HPstateTmp; + VARDECL( opus_int16, X ); + opus_int32 Xnrg[ VAD_N_BANDS ]; + opus_int32 NrgToNoiseRatio_Q8[ VAD_N_BANDS ]; + opus_int32 speech_nrg, x_tmp; + opus_int X_offset[ VAD_N_BANDS ]; + opus_int ret = 0; + silk_VAD_state *psSilk_VAD = &psEncC->sVAD; + SAVE_STACK; + + /* Safety checks */ + silk_assert( VAD_N_BANDS == 4 ); + celt_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); + celt_assert( psEncC->frame_length <= 512 ); + celt_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); + + /***********************/ + /* Filter and Decimate */ + /***********************/ + decimated_framelength1 = silk_RSHIFT( psEncC->frame_length, 1 ); + decimated_framelength2 = silk_RSHIFT( psEncC->frame_length, 2 ); + decimated_framelength = silk_RSHIFT( psEncC->frame_length, 3 ); + /* Decimate into 4 bands: + 0 L 3L L 3L 5L + - -- - -- -- + 8 8 2 4 4 + + [0-1 kHz| temp. |1-2 kHz| 2-4 kHz | 4-8 kHz | + + They're arranged to allow the minimal ( frame_length / 4 ) extra + scratch space during the downsampling process */ + X_offset[ 0 ] = 0; + X_offset[ 1 ] = decimated_framelength + decimated_framelength2; + X_offset[ 2 ] = X_offset[ 1 ] + decimated_framelength; + X_offset[ 3 ] = X_offset[ 2 ] + decimated_framelength2; + ALLOC( X, X_offset[ 3 ] + decimated_framelength1, opus_int16 ); + + /* 0-8 kHz to 0-4 kHz and 4-8 kHz */ + silk_ana_filt_bank_1( pIn, &psSilk_VAD->AnaState[ 0 ], + X, &X[ X_offset[ 3 ] ], psEncC->frame_length ); + + /* 0-4 kHz to 0-2 kHz and 2-4 kHz */ + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState1[ 0 ], + X, &X[ X_offset[ 2 ] ], decimated_framelength1 ); + + /* 0-2 kHz to 0-1 kHz and 1-2 kHz */ + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState2[ 0 ], + X, &X[ X_offset[ 1 ] ], decimated_framelength2 ); + + /*********************************************/ + /* HP filter on lowest band (differentiator) */ + /*********************************************/ + X[ decimated_framelength - 1 ] = silk_RSHIFT( X[ decimated_framelength - 1 ], 1 ); + HPstateTmp = X[ decimated_framelength - 1 ]; + for( i = decimated_framelength - 1; i > 0; i-- ) { + X[ i - 1 ] = silk_RSHIFT( X[ i - 1 ], 1 ); + X[ i ] -= X[ i - 1 ]; + } + X[ 0 ] -= psSilk_VAD->HPstate; + psSilk_VAD->HPstate = HPstateTmp; + + /*************************************/ + /* Calculate the energy in each band */ + /*************************************/ + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* Find the decimated framelength in the non-uniformly divided bands */ + decimated_framelength = silk_RSHIFT( psEncC->frame_length, silk_min_int( VAD_N_BANDS - b, VAD_N_BANDS - 1 ) ); + + /* Split length into subframe lengths */ + dec_subframe_length = silk_RSHIFT( decimated_framelength, VAD_INTERNAL_SUBFRAMES_LOG2 ); + dec_subframe_offset = 0; + + /* Compute energy per sub-frame */ + /* initialize with summed energy of last subframe */ + Xnrg[ b ] = psSilk_VAD->XnrgSubfr[ b ]; + for( s = 0; s < VAD_INTERNAL_SUBFRAMES; s++ ) { + sumSquared = 0; + for( i = 0; i < dec_subframe_length; i++ ) { + /* The energy will be less than dec_subframe_length * ( silk_int16_MIN / 8 ) ^ 2. */ + /* Therefore we can accumulate with no risk of overflow (unless dec_subframe_length > 128) */ + x_tmp = silk_RSHIFT( + X[ X_offset[ b ] + i + dec_subframe_offset ], 3 ); + sumSquared = silk_SMLABB( sumSquared, x_tmp, x_tmp ); + + /* Safety check */ + silk_assert( sumSquared >= 0 ); + } + + /* Add/saturate summed energy of current subframe */ + if( s < VAD_INTERNAL_SUBFRAMES - 1 ) { + Xnrg[ b ] = silk_ADD_POS_SAT32( Xnrg[ b ], sumSquared ); + } else { + /* Look-ahead subframe */ + Xnrg[ b ] = silk_ADD_POS_SAT32( Xnrg[ b ], silk_RSHIFT( sumSquared, 1 ) ); + } + + dec_subframe_offset += dec_subframe_length; + } + psSilk_VAD->XnrgSubfr[ b ] = sumSquared; + } + + /********************/ + /* Noise estimation */ + /********************/ + silk_VAD_GetNoiseLevels( &Xnrg[ 0 ], psSilk_VAD ); + + /***********************************************/ + /* Signal-plus-noise to noise ratio estimation */ + /***********************************************/ + sumSquared = 0; + input_tilt = 0; + for( b = 0; b < VAD_N_BANDS; b++ ) { + speech_nrg = Xnrg[ b ] - psSilk_VAD->NL[ b ]; + if( speech_nrg > 0 ) { + /* Divide, with sufficient resolution */ + if( ( Xnrg[ b ] & 0xFF800000 ) == 0 ) { + NrgToNoiseRatio_Q8[ b ] = silk_DIV32( silk_LSHIFT( Xnrg[ b ], 8 ), psSilk_VAD->NL[ b ] + 1 ); + } else { + NrgToNoiseRatio_Q8[ b ] = silk_DIV32( Xnrg[ b ], silk_RSHIFT( psSilk_VAD->NL[ b ], 8 ) + 1 ); + } + + /* Convert to log domain */ + SNR_Q7 = silk_lin2log( NrgToNoiseRatio_Q8[ b ] ) - 8 * 128; + + /* Sum-of-squares */ + sumSquared = silk_SMLABB( sumSquared, SNR_Q7, SNR_Q7 ); /* Q14 */ + + /* Tilt measure */ + if( speech_nrg < ( (opus_int32)1 << 20 ) ) { + /* Scale down SNR value for small subband speech energies */ + SNR_Q7 = silk_SMULWB( silk_LSHIFT( silk_SQRT_APPROX( speech_nrg ), 6 ), SNR_Q7 ); + } + input_tilt = silk_SMLAWB( input_tilt, tiltWeights[ b ], SNR_Q7 ); + } else { + NrgToNoiseRatio_Q8[ b ] = 256; + } + } + + /* Mean-of-squares */ + sumSquared = silk_DIV32_16( sumSquared, VAD_N_BANDS ); /* Q14 */ + + /* Root-mean-square approximation, scale to dBs, and write to output pointer */ + pSNR_dB_Q7 = (opus_int16)( 3 * silk_SQRT_APPROX( sumSquared ) ); /* Q7 */ + + /*********************************/ + /* Speech Probability Estimation */ + /*********************************/ + SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - VAD_NEGATIVE_OFFSET_Q5 ); + + /**************************/ + /* Frequency Tilt Measure */ + /**************************/ + psEncC->input_tilt_Q15 = silk_LSHIFT( silk_sigm_Q15( input_tilt ) - 16384, 1 ); + + /**************************************************/ + /* Scale the sigmoid output based on power levels */ + /**************************************************/ + speech_nrg = 0; + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* Accumulate signal-without-noise energies, higher frequency bands have more weight */ + speech_nrg += ( b + 1 ) * silk_RSHIFT( Xnrg[ b ] - psSilk_VAD->NL[ b ], 4 ); + } + + if( psEncC->frame_length == 20 * psEncC->fs_kHz ) { + speech_nrg = silk_RSHIFT32( speech_nrg, 1 ); + } + /* Power scaling */ + if( speech_nrg <= 0 ) { + SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); + } else if( speech_nrg < 16384 ) { + speech_nrg = silk_LSHIFT32( speech_nrg, 16 ); + + /* square-root */ + speech_nrg = silk_SQRT_APPROX( speech_nrg ); + SA_Q15 = silk_SMULWB( 32768 + speech_nrg, SA_Q15 ); + } + + /* Copy the resulting speech activity in Q8 */ + psEncC->speech_activity_Q8 = silk_min_int( silk_RSHIFT( SA_Q15, 7 ), silk_uint8_MAX ); + + /***********************************/ + /* Energy Level and SNR estimation */ + /***********************************/ + /* Smoothing coefficient */ + smooth_coef_Q16 = silk_SMULWB( VAD_SNR_SMOOTH_COEF_Q18, silk_SMULWB( (opus_int32)SA_Q15, SA_Q15 ) ); + + if( psEncC->frame_length == 10 * psEncC->fs_kHz ) { + smooth_coef_Q16 >>= 1; + } + + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* compute smoothed energy-to-noise ratio per band */ + psSilk_VAD->NrgRatioSmth_Q8[ b ] = silk_SMLAWB( psSilk_VAD->NrgRatioSmth_Q8[ b ], + NrgToNoiseRatio_Q8[ b ] - psSilk_VAD->NrgRatioSmth_Q8[ b ], smooth_coef_Q16 ); + + /* signal to noise ratio in dB per band */ + SNR_Q7 = 3 * ( silk_lin2log( psSilk_VAD->NrgRatioSmth_Q8[b] ) - 8 * 128 ); + /* quality = sigmoid( 0.25 * ( SNR_dB - 16 ) ); */ + psEncC->input_quality_bands_Q15[ b ] = silk_sigm_Q15( silk_RSHIFT( SNR_Q7 - 16 * 128, 4 ) ); + } + + RESTORE_STACK; + return( ret ); +} + +/**************************/ +/* Noise level estimation */ +/**************************/ +# if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE +#endif +void silk_VAD_GetNoiseLevels( + const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +) +{ + opus_int k; + opus_int32 nl, nrg, inv_nrg; + opus_int coef, min_coef; + + /* Initially faster smoothing */ + if( psSilk_VAD->counter < 1000 ) { /* 1000 = 20 sec */ + min_coef = silk_DIV32_16( silk_int16_MAX, silk_RSHIFT( psSilk_VAD->counter, 4 ) + 1 ); + /* Increment frame counter */ + psSilk_VAD->counter++; + } else { + min_coef = 0; + } + + for( k = 0; k < VAD_N_BANDS; k++ ) { + /* Get old noise level estimate for current band */ + nl = psSilk_VAD->NL[ k ]; + silk_assert( nl >= 0 ); + + /* Add bias */ + nrg = silk_ADD_POS_SAT32( pX[ k ], psSilk_VAD->NoiseLevelBias[ k ] ); + silk_assert( nrg > 0 ); + + /* Invert energies */ + inv_nrg = silk_DIV32( silk_int32_MAX, nrg ); + silk_assert( inv_nrg >= 0 ); + + /* Less update when subband energy is high */ + if( nrg > silk_LSHIFT( nl, 3 ) ) { + coef = VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 >> 3; + } else if( nrg < nl ) { + coef = VAD_NOISE_LEVEL_SMOOTH_COEF_Q16; + } else { + coef = silk_SMULWB( silk_SMULWW( inv_nrg, nl ), VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 << 1 ); + } + + /* Initially faster smoothing */ + coef = silk_max_int( coef, min_coef ); + + /* Smooth inverse energies */ + psSilk_VAD->inv_NL[ k ] = silk_SMLAWB( psSilk_VAD->inv_NL[ k ], inv_nrg - psSilk_VAD->inv_NL[ k ], coef ); + silk_assert( psSilk_VAD->inv_NL[ k ] >= 0 ); + + /* Compute noise level by inverting again */ + nl = silk_DIV32( silk_int32_MAX, psSilk_VAD->inv_NL[ k ] ); + silk_assert( nl >= 0 ); + + /* Limit noise levels (guarantee 7 bits of head room) */ + nl = silk_min( nl, 0x00FFFFFF ); + + /* Store as part of state */ + psSilk_VAD->NL[ k ] = nl; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/VQ_WMat_EC.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/VQ_WMat_EC.c new file mode 100644 index 000000000..245a7e4b0 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/VQ_WMat_EC.c @@ -0,0 +1,131 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */ +void silk_VQ_WMat_EC_c( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +) +{ + opus_int k, gain_tmp_Q7; + const opus_int8 *cb_row_Q7; + opus_int32 neg_xX_Q24[ 5 ]; + opus_int32 sum1_Q15, sum2_Q24; + opus_int32 bits_res_Q8, bits_tot_Q8; + + /* Negate and convert to new Q domain */ + neg_xX_Q24[ 0 ] = -silk_LSHIFT32( xX_Q17[ 0 ], 7 ); + neg_xX_Q24[ 1 ] = -silk_LSHIFT32( xX_Q17[ 1 ], 7 ); + neg_xX_Q24[ 2 ] = -silk_LSHIFT32( xX_Q17[ 2 ], 7 ); + neg_xX_Q24[ 3 ] = -silk_LSHIFT32( xX_Q17[ 3 ], 7 ); + neg_xX_Q24[ 4 ] = -silk_LSHIFT32( xX_Q17[ 4 ], 7 ); + + /* Loop over codebook */ + *rate_dist_Q8 = silk_int32_MAX; + *res_nrg_Q15 = silk_int32_MAX; + cb_row_Q7 = cb_Q7; + /* If things go really bad, at least *ind is set to something safe. */ + *ind = 0; + for( k = 0; k < L; k++ ) { + opus_int32 penalty; + gain_tmp_Q7 = cb_gain_Q7[k]; + /* Weighted rate */ + /* Quantization error: 1 - 2 * xX * cb + cb' * XX * cb */ + sum1_Q15 = SILK_FIX_CONST( 1.001, 15 ); + + /* Penalty for too large gain */ + penalty = silk_LSHIFT32( silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 11 ); + + /* first row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 0 ], XX_Q17[ 1 ], cb_row_Q7[ 1 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 2 ], cb_row_Q7[ 2 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 3 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 4 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 0 ], cb_row_Q7[ 0 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 0 ] ); + + /* second row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 1 ], XX_Q17[ 7 ], cb_row_Q7[ 2 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 8 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 9 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 6 ], cb_row_Q7[ 1 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 1 ] ); + + /* third row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 2 ], XX_Q17[ 13 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 14 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 12 ], cb_row_Q7[ 2 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 2 ] ); + + /* fourth row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 3 ], XX_Q17[ 19 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 18 ], cb_row_Q7[ 3 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 3 ] ); + + /* last row of XX_Q17 */ + sum2_Q24 = silk_LSHIFT32( neg_xX_Q24[ 4 ], 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 24 ], cb_row_Q7[ 4 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 4 ] ); + + /* find best */ + if( sum1_Q15 >= 0 ) { + /* Translate residual energy to bits using high-rate assumption (6 dB ==> 1 bit/sample) */ + bits_res_Q8 = silk_SMULBB( subfr_len, silk_lin2log( sum1_Q15 + penalty) - (15 << 7) ); + /* In the following line we reduce the codelength component by half ("-1"); seems to slightly improve quality */ + bits_tot_Q8 = silk_ADD_LSHIFT32( bits_res_Q8, cl_Q5[ k ], 3-1 ); + if( bits_tot_Q8 <= *rate_dist_Q8 ) { + *rate_dist_Q8 = bits_tot_Q8; + *res_nrg_Q15 = sum1_Q15 + penalty; + *ind = (opus_int8)k; + *gain_Q7 = gain_tmp_Q7; + } + } + + /* Go to next cbk vector */ + cb_row_Q7 += LTP_ORDER; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/ana_filt_bank_1.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/ana_filt_bank_1.c new file mode 100644 index 000000000..3907f8c1b --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/ana_filt_bank_1.c @@ -0,0 +1,74 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Coefficients for 2-band filter bank based on first-order allpass filters */ +static const opus_int16 A_fb1_20 = 5394 << 1; +static const opus_int16 A_fb1_21 = -24290; /* (opus_int16)(20623 << 1) */ + +/* Split signal into two decimated bands using first-order allpass filters */ +void silk_ana_filt_bank_1( + const opus_int16 *in, /* I Input signal [N] */ + opus_int32 *S, /* I/O State vector [2] */ + opus_int16 *outL, /* O Low band [N/2] */ + opus_int16 *outH, /* O High band [N/2] */ + const opus_int32 N /* I Number of input samples */ +) +{ + opus_int k, N2 = silk_RSHIFT( N, 1 ); + opus_int32 in32, X, Y, out_1, out_2; + + /* Internal variables and state are in Q10 format */ + for( k = 0; k < N2; k++ ) { + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ 2 * k ], 10 ); + + /* All-pass section for even input sample */ + Y = silk_SUB32( in32, S[ 0 ] ); + X = silk_SMLAWB( Y, Y, A_fb1_21 ); + out_1 = silk_ADD32( S[ 0 ], X ); + S[ 0 ] = silk_ADD32( in32, X ); + + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ 2 * k + 1 ], 10 ); + + /* All-pass section for odd input sample, and add to output of previous section */ + Y = silk_SUB32( in32, S[ 1 ] ); + X = silk_SMULWB( Y, A_fb1_20 ); + out_2 = silk_ADD32( S[ 1 ], X ); + S[ 1 ] = silk_ADD32( in32, X ); + + /* Add/subtract, convert back to int16 and store to output */ + outL[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_ADD32( out_2, out_1 ), 11 ) ); + outH[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SUB32( out_2, out_1 ), 11 ) ); + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/check_control_input.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/check_control_input.c new file mode 100644 index 000000000..b4e96e7bf --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/check_control_input.c @@ -0,0 +1,109 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "control.h" +#include "errors.h" + +/* Check encoder control struct */ +opus_int check_control_input( + silk_EncControlStruct *encControl /* I Control structure */ +) +{ + celt_assert( encControl != NULL ); + + if( ( ( encControl->API_sampleRate != 8000 ) && + ( encControl->API_sampleRate != 12000 ) && + ( encControl->API_sampleRate != 16000 ) && + ( encControl->API_sampleRate != 24000 ) && + ( encControl->API_sampleRate != 32000 ) && + ( encControl->API_sampleRate != 44100 ) && +#ifdef ENABLE_QEXT + ( encControl->API_sampleRate != 96000 ) && +#endif + ( encControl->API_sampleRate != 48000 ) ) || + ( ( encControl->desiredInternalSampleRate != 8000 ) && + ( encControl->desiredInternalSampleRate != 12000 ) && + ( encControl->desiredInternalSampleRate != 16000 ) ) || + ( ( encControl->maxInternalSampleRate != 8000 ) && + ( encControl->maxInternalSampleRate != 12000 ) && + ( encControl->maxInternalSampleRate != 16000 ) ) || + ( ( encControl->minInternalSampleRate != 8000 ) && + ( encControl->minInternalSampleRate != 12000 ) && + ( encControl->minInternalSampleRate != 16000 ) ) || + ( encControl->minInternalSampleRate > encControl->desiredInternalSampleRate ) || + ( encControl->maxInternalSampleRate < encControl->desiredInternalSampleRate ) || + ( encControl->minInternalSampleRate > encControl->maxInternalSampleRate ) ) { + celt_assert( 0 ); + return SILK_ENC_FS_NOT_SUPPORTED; + } + if( encControl->payloadSize_ms != 10 && + encControl->payloadSize_ms != 20 && + encControl->payloadSize_ms != 40 && + encControl->payloadSize_ms != 60 ) { + celt_assert( 0 ); + return SILK_ENC_PACKET_SIZE_NOT_SUPPORTED; + } + if( encControl->packetLossPercentage < 0 || encControl->packetLossPercentage > 100 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_LOSS_RATE; + } + if( encControl->useDTX < 0 || encControl->useDTX > 1 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_DTX_SETTING; + } + if( encControl->useCBR < 0 || encControl->useCBR > 1 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_CBR_SETTING; + } + if( encControl->useInBandFEC < 0 || encControl->useInBandFEC > 1 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_INBAND_FEC_SETTING; + } + if( encControl->nChannelsAPI < 1 || encControl->nChannelsAPI > ENCODER_NUM_CHANNELS ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; + } + if( encControl->nChannelsInternal < 1 || encControl->nChannelsInternal > ENCODER_NUM_CHANNELS ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; + } + if( encControl->nChannelsInternal > encControl->nChannelsAPI ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; + } + if( encControl->complexity < 0 || encControl->complexity > 10 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_COMPLEXITY_SETTING; + } + + return SILK_NO_ERROR; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_SNR.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_SNR.c new file mode 100644 index 000000000..455acad6f --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_SNR.c @@ -0,0 +1,113 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +/* These tables hold SNR values divided by 21 (so they fit in 8 bits) + for different target bitrates spaced at 400 bps interval. The first + 10 values are omitted (0-4 kb/s) because they're all zeros. + These tables were obtained by running different SNRs through the + encoder and measuring the active bitrate. */ +static const unsigned char silk_TargetRate_NB_21[117 - 10] = { + 0, 15, 39, 52, 61, 68, + 74, 79, 84, 88, 92, 95, 99,102,105,108,111,114,117,119,122,124, + 126,129,131,133,135,137,139,142,143,145,147,149,151,153,155,157, + 158,160,162,163,165,167,168,170,171,173,174,176,177,179,180,182, + 183,185,186,187,189,190,192,193,194,196,197,199,200,201,203,204, + 205,207,208,209,211,212,213,215,216,217,219,220,221,223,224,225, + 227,228,230,231,232,234,235,236,238,239,241,242,243,245,246,248, + 249,250,252,253,255 +}; + +static const unsigned char silk_TargetRate_MB_21[165 - 10] = { + 0, 0, 28, 43, 52, 59, + 65, 70, 74, 78, 81, 85, 87, 90, 93, 95, 98,100,102,105,107,109, + 111,113,115,116,118,120,122,123,125,127,128,130,131,133,134,136, + 137,138,140,141,143,144,145,147,148,149,151,152,153,154,156,157, + 158,159,160,162,163,164,165,166,167,168,169,171,172,173,174,175, + 176,177,178,179,180,181,182,183,184,185,186,187,188,188,189,190, + 191,192,193,194,195,196,197,198,199,200,201,202,203,203,204,205, + 206,207,208,209,210,211,212,213,214,214,215,216,217,218,219,220, + 221,222,223,224,224,225,226,227,228,229,230,231,232,233,234,235, + 236,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250, + 251,252,253,254,255 +}; + +static const unsigned char silk_TargetRate_WB_21[201 - 10] = { + 0, 0, 0, 8, 29, 41, + 49, 56, 62, 66, 70, 74, 77, 80, 83, 86, 88, 91, 93, 95, 97, 99, + 101,103,105,107,108,110,112,113,115,116,118,119,121,122,123,125, + 126,127,129,130,131,132,134,135,136,137,138,140,141,142,143,144, + 145,146,147,148,149,150,151,152,153,154,156,157,158,159,159,160, + 161,162,163,164,165,166,167,168,169,170,171,171,172,173,174,175, + 176,177,177,178,179,180,181,181,182,183,184,185,185,186,187,188, + 189,189,190,191,192,192,193,194,195,195,196,197,198,198,199,200, + 200,201,202,203,203,204,205,206,206,207,208,209,209,210,211,211, + 212,213,214,214,215,216,216,217,218,219,219,220,221,221,222,223, + 224,224,225,226,226,227,228,229,229,230,231,232,232,233,234,234, + 235,236,237,237,238,239,240,240,241,242,243,243,244,245,246,246, + 247,248,249,249,250,251,252,253,255 +}; + +/* Control SNR of residual quantizer */ +opus_int silk_control_SNR( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + opus_int32 TargetRate_bps /* I Target max bitrate (bps) */ +) +{ + int id; + int bound; + const unsigned char *snr_table; + + psEncC->TargetRate_bps = TargetRate_bps; + if( psEncC->nb_subfr == 2 ) { + TargetRate_bps -= 2000 + psEncC->fs_kHz/16; + } + if( psEncC->fs_kHz == 8 ) { + bound = sizeof(silk_TargetRate_NB_21); + snr_table = silk_TargetRate_NB_21; + } else if( psEncC->fs_kHz == 12 ) { + bound = sizeof(silk_TargetRate_MB_21); + snr_table = silk_TargetRate_MB_21; + } else { + bound = sizeof(silk_TargetRate_WB_21); + snr_table = silk_TargetRate_WB_21; + } + id = (TargetRate_bps+200)/400; + id = silk_min(id - 10, bound-1); + if( id <= 0 ) { + psEncC->SNR_dB_Q7 = 0; + } else { + psEncC->SNR_dB_Q7 = snr_table[id]*21; + } + return SILK_NO_ERROR; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_audio_bandwidth.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_audio_bandwidth.c new file mode 100644 index 000000000..f6d22d839 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_audio_bandwidth.c @@ -0,0 +1,132 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +/* Control internal sampling rate */ +opus_int silk_control_audio_bandwidth( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl /* I Control structure */ +) +{ + opus_int fs_kHz; + opus_int orig_kHz; + opus_int32 fs_Hz; + + orig_kHz = psEncC->fs_kHz; + /* Handle a bandwidth-switching reset where we need to be aware what the last sampling rate was. */ + if( orig_kHz == 0 ) { + orig_kHz = psEncC->sLP.saved_fs_kHz; + } + fs_kHz = orig_kHz; + fs_Hz = silk_SMULBB( fs_kHz, 1000 ); + if( fs_Hz == 0 ) { + /* Encoder has just been initialized */ + fs_Hz = silk_min( psEncC->desiredInternal_fs_Hz, psEncC->API_fs_Hz ); + fs_kHz = silk_DIV32_16( fs_Hz, 1000 ); + } else if( fs_Hz > psEncC->API_fs_Hz || fs_Hz > psEncC->maxInternal_fs_Hz || fs_Hz < psEncC->minInternal_fs_Hz ) { + /* Make sure internal rate is not higher than external rate or maximum allowed, or lower than minimum allowed */ + fs_Hz = psEncC->API_fs_Hz; + fs_Hz = silk_min( fs_Hz, psEncC->maxInternal_fs_Hz ); + fs_Hz = silk_max( fs_Hz, psEncC->minInternal_fs_Hz ); + fs_kHz = silk_DIV32_16( fs_Hz, 1000 ); + } else { + /* State machine for the internal sampling rate switching */ + if( psEncC->sLP.transition_frame_no >= TRANSITION_FRAMES ) { + /* Stop transition phase */ + psEncC->sLP.mode = 0; + } + if( psEncC->allow_bandwidth_switch || encControl->opusCanSwitch ) { + /* Check if we should switch down */ + if( silk_SMULBB( orig_kHz, 1000 ) > psEncC->desiredInternal_fs_Hz ) + { + /* Switch down */ + if( psEncC->sLP.mode == 0 ) { + /* New transition */ + psEncC->sLP.transition_frame_no = TRANSITION_FRAMES; + + /* Reset transition filter state */ + silk_memset( psEncC->sLP.In_LP_State, 0, sizeof( psEncC->sLP.In_LP_State ) ); + } + if( encControl->opusCanSwitch ) { + /* Stop transition phase */ + psEncC->sLP.mode = 0; + + /* Switch to a lower sample frequency */ + fs_kHz = orig_kHz == 16 ? 12 : 8; + } else { + if( psEncC->sLP.transition_frame_no <= 0 ) { + encControl->switchReady = 1; + /* Make room for redundancy */ + encControl->maxBits -= encControl->maxBits * 5 / ( encControl->payloadSize_ms + 5 ); + } else { + /* Direction: down (at double speed) */ + psEncC->sLP.mode = -2; + } + } + } + else + /* Check if we should switch up */ + if( silk_SMULBB( orig_kHz, 1000 ) < psEncC->desiredInternal_fs_Hz ) + { + /* Switch up */ + if( encControl->opusCanSwitch ) { + /* Switch to a higher sample frequency */ + fs_kHz = orig_kHz == 8 ? 12 : 16; + + /* New transition */ + psEncC->sLP.transition_frame_no = 0; + + /* Reset transition filter state */ + silk_memset( psEncC->sLP.In_LP_State, 0, sizeof( psEncC->sLP.In_LP_State ) ); + + /* Direction: up */ + psEncC->sLP.mode = 1; + } else { + if( psEncC->sLP.mode == 0 ) { + encControl->switchReady = 1; + /* Make room for redundancy */ + encControl->maxBits -= encControl->maxBits * 5 / ( encControl->payloadSize_ms + 5 ); + } else { + /* Direction: up */ + psEncC->sLP.mode = 1; + } + } + } else { + if (psEncC->sLP.mode<0) + psEncC->sLP.mode = 1; + } + } + } + + return fs_kHz; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_codec.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_codec.c new file mode 100644 index 000000000..784ffe66d --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/control_codec.c @@ -0,0 +1,423 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#ifdef FIXED_POINT +#include "main_FIX.h" +#define silk_encoder_state_Fxx silk_encoder_state_FIX +#else +#include "main_FLP.h" +#define silk_encoder_state_Fxx silk_encoder_state_FLP +#endif +#include "stack_alloc.h" +#include "tuning_parameters.h" +#include "pitch_est_defines.h" + +static opus_int silk_setup_resamplers( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz /* I */ +); + +static opus_int silk_setup_fs( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz, /* I */ + opus_int PacketSize_ms /* I */ +); + +static opus_int silk_setup_complexity( + silk_encoder_state *psEncC, /* I/O */ + opus_int Complexity /* I */ +); + +static OPUS_INLINE opus_int silk_setup_LBRR( + silk_encoder_state *psEncC, /* I/O */ + const silk_EncControlStruct *encControl /* I */ +); + + +/* Control encoder */ +opus_int silk_control_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl, /* I Control structure */ + const opus_int allow_bw_switch, /* I Flag to allow switching audio bandwidth */ + const opus_int channelNb, /* I Channel number */ + const opus_int force_fs_kHz +) +{ + opus_int fs_kHz, ret = 0; + + psEnc->sCmn.useDTX = encControl->useDTX; + psEnc->sCmn.useCBR = encControl->useCBR; + psEnc->sCmn.API_fs_Hz = encControl->API_sampleRate; + psEnc->sCmn.maxInternal_fs_Hz = encControl->maxInternalSampleRate; + psEnc->sCmn.minInternal_fs_Hz = encControl->minInternalSampleRate; + psEnc->sCmn.desiredInternal_fs_Hz = encControl->desiredInternalSampleRate; + psEnc->sCmn.useInBandFEC = encControl->useInBandFEC; + psEnc->sCmn.nChannelsAPI = encControl->nChannelsAPI; + psEnc->sCmn.nChannelsInternal = encControl->nChannelsInternal; + psEnc->sCmn.allow_bandwidth_switch = allow_bw_switch; + psEnc->sCmn.channelNb = channelNb; + + if( psEnc->sCmn.controlled_since_last_payload != 0 && psEnc->sCmn.prefillFlag == 0 ) { + if( psEnc->sCmn.API_fs_Hz != psEnc->sCmn.prev_API_fs_Hz && psEnc->sCmn.fs_kHz > 0 ) { + /* Change in API sampling rate in the middle of encoding a packet */ + ret += silk_setup_resamplers( psEnc, psEnc->sCmn.fs_kHz ); + } + return ret; + } + + /* Beyond this point we know that there are no previously coded frames in the payload buffer */ + + /********************************************/ + /* Determine internal sampling rate */ + /********************************************/ + fs_kHz = silk_control_audio_bandwidth( &psEnc->sCmn, encControl ); + if( force_fs_kHz ) { + fs_kHz = force_fs_kHz; + } + /********************************************/ + /* Prepare resampler and buffered data */ + /********************************************/ + ret += silk_setup_resamplers( psEnc, fs_kHz ); + + /********************************************/ + /* Set internal sampling frequency */ + /********************************************/ + ret += silk_setup_fs( psEnc, fs_kHz, encControl->payloadSize_ms ); + + /********************************************/ + /* Set encoding complexity */ + /********************************************/ + ret += silk_setup_complexity( &psEnc->sCmn, encControl->complexity ); + + /********************************************/ + /* Set packet loss rate measured by farend */ + /********************************************/ + psEnc->sCmn.PacketLoss_perc = encControl->packetLossPercentage; + + /********************************************/ + /* Set LBRR usage */ + /********************************************/ + ret += silk_setup_LBRR( &psEnc->sCmn, encControl ); + + psEnc->sCmn.controlled_since_last_payload = 1; + + return ret; +} + +static opus_int silk_setup_resamplers( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz /* I */ +) +{ + opus_int ret = SILK_NO_ERROR; + SAVE_STACK; + + if( psEnc->sCmn.fs_kHz != fs_kHz || psEnc->sCmn.prev_API_fs_Hz != psEnc->sCmn.API_fs_Hz ) + { + if( psEnc->sCmn.fs_kHz == 0 ) { + /* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */ + ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, fs_kHz * 1000, 1 ); + } else { + VARDECL( opus_int16, x_buf_API_fs_Hz ); + VARDECL( silk_resampler_state_struct, temp_resampler_state ); +#ifdef FIXED_POINT + opus_int16 *x_bufFIX = psEnc->x_buf; +#else + VARDECL( opus_int16, x_bufFIX ); + opus_int32 new_buf_samples; +#endif + opus_int32 api_buf_samples; + opus_int32 old_buf_samples; + opus_int32 buf_length_ms; + + buf_length_ms = silk_LSHIFT( psEnc->sCmn.nb_subfr * 5, 1 ) + LA_SHAPE_MS; + old_buf_samples = buf_length_ms * psEnc->sCmn.fs_kHz; + +#ifndef FIXED_POINT + new_buf_samples = buf_length_ms * fs_kHz; + ALLOC( x_bufFIX, silk_max( old_buf_samples, new_buf_samples ), + opus_int16 ); + silk_float2short_array( x_bufFIX, psEnc->x_buf, old_buf_samples ); +#endif + + /* Initialize resampler for temporary resampling of x_buf data to API_fs_Hz */ + ALLOC( temp_resampler_state, 1, silk_resampler_state_struct ); + ret += silk_resampler_init( temp_resampler_state, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz, 0 ); + + /* Calculate number of samples to temporarily upsample */ + api_buf_samples = buf_length_ms * silk_DIV32_16( psEnc->sCmn.API_fs_Hz, 1000 ); + + /* Temporary resampling of x_buf data to API_fs_Hz */ + ALLOC( x_buf_API_fs_Hz, api_buf_samples, opus_int16 ); + ret += silk_resampler( temp_resampler_state, x_buf_API_fs_Hz, x_bufFIX, old_buf_samples ); + + /* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */ + ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, silk_SMULBB( fs_kHz, 1000 ), 1 ); + + /* Correct resampler state by resampling buffered data from API_fs_Hz to fs_kHz */ + ret += silk_resampler( &psEnc->sCmn.resampler_state, x_bufFIX, x_buf_API_fs_Hz, api_buf_samples ); + +#ifndef FIXED_POINT + silk_short2float_array( psEnc->x_buf, x_bufFIX, new_buf_samples); +#endif + } + } + + psEnc->sCmn.prev_API_fs_Hz = psEnc->sCmn.API_fs_Hz; + + RESTORE_STACK; + return ret; +} + +static opus_int silk_setup_fs( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz, /* I */ + opus_int PacketSize_ms /* I */ +) +{ + opus_int ret = SILK_NO_ERROR; + + /* Set packet size */ + if( PacketSize_ms != psEnc->sCmn.PacketSize_ms ) { + if( ( PacketSize_ms != 10 ) && + ( PacketSize_ms != 20 ) && + ( PacketSize_ms != 40 ) && + ( PacketSize_ms != 60 ) ) { + ret = SILK_ENC_PACKET_SIZE_NOT_SUPPORTED; + } + if( PacketSize_ms <= 10 ) { + psEnc->sCmn.nFramesPerPacket = 1; + psEnc->sCmn.nb_subfr = PacketSize_ms == 10 ? 2 : 1; + psEnc->sCmn.frame_length = silk_SMULBB( PacketSize_ms, fs_kHz ); + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS_2_SF, fs_kHz ); + if( psEnc->sCmn.fs_kHz == 8 ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_NB_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_iCDF; + } + } else { + psEnc->sCmn.nFramesPerPacket = silk_DIV32_16( PacketSize_ms, MAX_FRAME_LENGTH_MS ); + psEnc->sCmn.nb_subfr = MAX_NB_SUBFR; + psEnc->sCmn.frame_length = silk_SMULBB( 20, fs_kHz ); + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS, fs_kHz ); + if( psEnc->sCmn.fs_kHz == 8 ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_NB_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_iCDF; + } + } + psEnc->sCmn.PacketSize_ms = PacketSize_ms; + psEnc->sCmn.TargetRate_bps = 0; /* trigger new SNR computation */ + } + + /* Set internal sampling frequency */ + celt_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); + celt_assert( psEnc->sCmn.nb_subfr == 2 || psEnc->sCmn.nb_subfr == 4 ); + if( psEnc->sCmn.fs_kHz != fs_kHz ) { + /* reset part of the state */ + silk_memset( &psEnc->sShape, 0, sizeof( psEnc->sShape ) ); + silk_memset( &psEnc->sCmn.sNSQ, 0, sizeof( psEnc->sCmn.sNSQ ) ); + silk_memset( psEnc->sCmn.prev_NLSFq_Q15, 0, sizeof( psEnc->sCmn.prev_NLSFq_Q15 ) ); + silk_memset( &psEnc->sCmn.sLP.In_LP_State, 0, sizeof( psEnc->sCmn.sLP.In_LP_State ) ); + psEnc->sCmn.inputBufIx = 0; + psEnc->sCmn.nFramesEncoded = 0; + psEnc->sCmn.TargetRate_bps = 0; /* trigger new SNR computation */ + + /* Initialize non-zero parameters */ + psEnc->sCmn.prevLag = 100; + psEnc->sCmn.first_frame_after_reset = 1; + psEnc->sShape.LastGainIndex = 10; + psEnc->sCmn.sNSQ.lagPrev = 100; + psEnc->sCmn.sNSQ.prev_gain_Q16 = 65536; + psEnc->sCmn.prevSignalType = TYPE_NO_VOICE_ACTIVITY; + + psEnc->sCmn.fs_kHz = fs_kHz; + if( psEnc->sCmn.fs_kHz == 8 ) { + if( psEnc->sCmn.nb_subfr == MAX_NB_SUBFR ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_NB_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_NB_iCDF; + } + } else { + if( psEnc->sCmn.nb_subfr == MAX_NB_SUBFR ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_iCDF; + } + } + if( psEnc->sCmn.fs_kHz == 8 || psEnc->sCmn.fs_kHz == 12 ) { + psEnc->sCmn.predictLPCOrder = MIN_LPC_ORDER; + psEnc->sCmn.psNLSF_CB = &silk_NLSF_CB_NB_MB; + } else { + psEnc->sCmn.predictLPCOrder = MAX_LPC_ORDER; + psEnc->sCmn.psNLSF_CB = &silk_NLSF_CB_WB; + } + psEnc->sCmn.subfr_length = SUB_FRAME_LENGTH_MS * fs_kHz; + psEnc->sCmn.frame_length = silk_SMULBB( psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr ); + psEnc->sCmn.ltp_mem_length = silk_SMULBB( LTP_MEM_LENGTH_MS, fs_kHz ); + psEnc->sCmn.la_pitch = silk_SMULBB( LA_PITCH_MS, fs_kHz ); + psEnc->sCmn.max_pitch_lag = silk_SMULBB( 18, fs_kHz ); + if( psEnc->sCmn.nb_subfr == MAX_NB_SUBFR ) { + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS, fs_kHz ); + } else { + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS_2_SF, fs_kHz ); + } + if( psEnc->sCmn.fs_kHz == 16 ) { + psEnc->sCmn.pitch_lag_low_bits_iCDF = silk_uniform8_iCDF; + } else if( psEnc->sCmn.fs_kHz == 12 ) { + psEnc->sCmn.pitch_lag_low_bits_iCDF = silk_uniform6_iCDF; + } else { + psEnc->sCmn.pitch_lag_low_bits_iCDF = silk_uniform4_iCDF; + } + } + + /* Check that settings are valid */ + celt_assert( ( psEnc->sCmn.subfr_length * psEnc->sCmn.nb_subfr ) == psEnc->sCmn.frame_length ); + + return ret; +} + +static opus_int silk_setup_complexity( + silk_encoder_state *psEncC, /* I/O */ + opus_int Complexity /* I */ +) +{ + opus_int ret = 0; + + /* Set encoding complexity */ + celt_assert( Complexity >= 0 && Complexity <= 10 ); + if( Complexity < 1 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MIN_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.8, 16 ); + psEncC->pitchEstimationLPCOrder = 6; + psEncC->shapingLPCOrder = 12; + psEncC->la_shape = 3 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 1; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 2; + psEncC->warping_Q16 = 0; + } else if( Complexity < 2 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.76, 16 ); + psEncC->pitchEstimationLPCOrder = 8; + psEncC->shapingLPCOrder = 14; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 1; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 3; + psEncC->warping_Q16 = 0; + } else if( Complexity < 3 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MIN_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.8, 16 ); + psEncC->pitchEstimationLPCOrder = 6; + psEncC->shapingLPCOrder = 12; + psEncC->la_shape = 3 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 2; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 2; + psEncC->warping_Q16 = 0; + } else if( Complexity < 4 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.76, 16 ); + psEncC->pitchEstimationLPCOrder = 8; + psEncC->shapingLPCOrder = 14; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 2; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 4; + psEncC->warping_Q16 = 0; + } else if( Complexity < 6 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.74, 16 ); + psEncC->pitchEstimationLPCOrder = 10; + psEncC->shapingLPCOrder = 16; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 2; + psEncC->useInterpolatedNLSFs = 1; + psEncC->NLSF_MSVQ_Survivors = 6; + psEncC->warping_Q16 = psEncC->fs_kHz * SILK_FIX_CONST( WARPING_MULTIPLIER, 16 ); + } else if( Complexity < 8 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.72, 16 ); + psEncC->pitchEstimationLPCOrder = 12; + psEncC->shapingLPCOrder = 20; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 3; + psEncC->useInterpolatedNLSFs = 1; + psEncC->NLSF_MSVQ_Survivors = 8; + psEncC->warping_Q16 = psEncC->fs_kHz * SILK_FIX_CONST( WARPING_MULTIPLIER, 16 ); + } else { + psEncC->pitchEstimationComplexity = SILK_PE_MAX_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.7, 16 ); + psEncC->pitchEstimationLPCOrder = 16; + psEncC->shapingLPCOrder = 24; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = MAX_DEL_DEC_STATES; + psEncC->useInterpolatedNLSFs = 1; + psEncC->NLSF_MSVQ_Survivors = 16; + psEncC->warping_Q16 = psEncC->fs_kHz * SILK_FIX_CONST( WARPING_MULTIPLIER, 16 ); + } + + /* Do not allow higher pitch estimation LPC order than predict LPC order */ + psEncC->pitchEstimationLPCOrder = silk_min_int( psEncC->pitchEstimationLPCOrder, psEncC->predictLPCOrder ); + psEncC->shapeWinLength = SUB_FRAME_LENGTH_MS * psEncC->fs_kHz + 2 * psEncC->la_shape; + psEncC->Complexity = Complexity; + + celt_assert( psEncC->pitchEstimationLPCOrder <= MAX_FIND_PITCH_LPC_ORDER ); + celt_assert( psEncC->shapingLPCOrder <= MAX_SHAPE_LPC_ORDER ); + celt_assert( psEncC->nStatesDelayedDecision <= MAX_DEL_DEC_STATES ); + celt_assert( psEncC->warping_Q16 <= 32767 ); + celt_assert( psEncC->la_shape <= LA_SHAPE_MAX ); + celt_assert( psEncC->shapeWinLength <= SHAPE_LPC_WIN_MAX ); + + return ret; +} + +static OPUS_INLINE opus_int silk_setup_LBRR( + silk_encoder_state *psEncC, /* I/O */ + const silk_EncControlStruct *encControl /* I */ +) +{ + opus_int LBRR_in_previous_packet, ret = SILK_NO_ERROR; + + LBRR_in_previous_packet = psEncC->LBRR_enabled; + psEncC->LBRR_enabled = encControl->LBRR_coded; + if( psEncC->LBRR_enabled ) { + /* Set gain increase for coding LBRR excitation */ + if( LBRR_in_previous_packet == 0 ) { + /* Previous packet did not have LBRR, and was therefore coded at a higher bitrate */ + psEncC->LBRR_GainIncreases = 7; + } else { + psEncC->LBRR_GainIncreases = silk_max_int( 7 - silk_SMULWB( (opus_int32)psEncC->PacketLoss_perc, SILK_FIX_CONST( 0.2, 16 ) ), 3 ); + } + } + + return ret; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/enc_API.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/enc_API.c new file mode 100644 index 000000000..d0fee4d46 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/enc_API.c @@ -0,0 +1,601 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#include "define.h" +#include "API.h" +#include "control.h" +#include "typedef.h" +#include "stack_alloc.h" +#include "structs.h" +#include "tuning_parameters.h" +#ifdef FIXED_POINT +#include "main_FIX.h" +#else +#include "main_FLP.h" +#endif + +#ifdef ENABLE_DRED +#include "dred_encoder.h" +#endif + +/***************************************/ +/* Read control structure from encoder */ +/***************************************/ +static opus_int silk_QueryEncoder( /* O Returns error code */ + const void *encState, /* I State */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +); + +/****************************************/ +/* Encoder functions */ +/****************************************/ + +opus_int silk_Get_Encoder_Size( /* O Returns error code */ + opus_int *encSizeBytes, /* O Number of bytes in SILK encoder state */ + opus_int channels /* I Number of channels */ +) +{ + opus_int ret = SILK_NO_ERROR; + + *encSizeBytes = sizeof( silk_encoder ); + /* Skip second encoder state for mono. */ + if ( channels == 1 ) { + *encSizeBytes -= sizeof( silk_encoder_state_Fxx ); + } + + return ret; +} + +/*************************/ +/* Init or Reset encoder */ +/*************************/ +opus_int silk_InitEncoder( /* O Returns error code */ + void *encState, /* I/O State */ + int channels, /* I Number of channels */ + int arch, /* I Run-time architecture */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +) +{ + silk_encoder *psEnc; + opus_int n, ret = SILK_NO_ERROR; + + psEnc = (silk_encoder *)encState; + + /* Reset encoder. Skip second encoder state for mono. */ + silk_memset( psEnc, 0, sizeof( silk_encoder ) - (channels==1)*sizeof( silk_encoder_state_Fxx ) ); + for( n = 0; n < channels; n++ ) { + if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ], arch ) ) { + celt_assert( 0 ); + } + } + + psEnc->nChannelsAPI = 1; + psEnc->nChannelsInternal = 1; + + /* Read control structure */ + if( ret += silk_QueryEncoder( encState, encStatus ) ) { + celt_assert( 0 ); + } + + return ret; +} + +/***************************************/ +/* Read control structure from encoder */ +/***************************************/ +static opus_int silk_QueryEncoder( /* O Returns error code */ + const void *encState, /* I State */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +) +{ + opus_int ret = SILK_NO_ERROR; + silk_encoder_state_Fxx *state_Fxx; + silk_encoder *psEnc = (silk_encoder *)encState; + + state_Fxx = psEnc->state_Fxx; + + encStatus->nChannelsAPI = psEnc->nChannelsAPI; + encStatus->nChannelsInternal = psEnc->nChannelsInternal; + encStatus->API_sampleRate = state_Fxx[ 0 ].sCmn.API_fs_Hz; + encStatus->maxInternalSampleRate = state_Fxx[ 0 ].sCmn.maxInternal_fs_Hz; + encStatus->minInternalSampleRate = state_Fxx[ 0 ].sCmn.minInternal_fs_Hz; + encStatus->desiredInternalSampleRate = state_Fxx[ 0 ].sCmn.desiredInternal_fs_Hz; + encStatus->payloadSize_ms = state_Fxx[ 0 ].sCmn.PacketSize_ms; + encStatus->bitRate = state_Fxx[ 0 ].sCmn.TargetRate_bps; + encStatus->packetLossPercentage = state_Fxx[ 0 ].sCmn.PacketLoss_perc; + encStatus->complexity = state_Fxx[ 0 ].sCmn.Complexity; + encStatus->useInBandFEC = state_Fxx[ 0 ].sCmn.useInBandFEC; + encStatus->useDTX = state_Fxx[ 0 ].sCmn.useDTX; + encStatus->useCBR = state_Fxx[ 0 ].sCmn.useCBR; + encStatus->internalSampleRate = silk_SMULBB( state_Fxx[ 0 ].sCmn.fs_kHz, 1000 ); + encStatus->allowBandwidthSwitch = state_Fxx[ 0 ].sCmn.allow_bandwidth_switch; + encStatus->inWBmodeWithoutVariableLP = state_Fxx[ 0 ].sCmn.fs_kHz == 16 && state_Fxx[ 0 ].sCmn.sLP.mode == 0; + + return ret; +} + + +/**************************/ +/* Encode frame with Silk */ +/**************************/ +/* Note: if prefillFlag is set, the input must contain 10 ms of audio, irrespective of what */ +/* encControl->payloadSize_ms is set to */ +opus_int silk_Encode( /* O Returns error code */ + void *encState, /* I/O State */ + silk_EncControlStruct *encControl, /* I Control status */ + const opus_res *samplesIn, /* I Speech sample input vector */ + opus_int nSamplesIn, /* I Number of samples in input vector */ + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */ + const opus_int prefillFlag, /* I Flag to indicate prefilling buffers no coding */ + opus_int activity /* I Decision of Opus voice activity detector */ +) +{ + opus_int n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0; + opus_int nSamplesToBuffer, nSamplesToBufferMax, nBlocksOf10ms; + opus_int nSamplesFromInput = 0, nSamplesFromInputMax; + opus_int speech_act_thr_for_switch_Q8; + opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum; + silk_encoder *psEnc = ( silk_encoder * )encState; + VARDECL( opus_int16, buf ); + opus_int transition, curr_block, tot_blocks; + SAVE_STACK; + + celt_assert( encControl->nChannelsAPI >= encControl->nChannelsInternal && encControl->nChannelsAPI >= psEnc->nChannelsInternal ); + if (encControl->reducedDependency) + { + for( n = 0; n < encControl->nChannelsAPI; n++ ) { + psEnc->state_Fxx[ n ].sCmn.first_frame_after_reset = 1; + } + } + for( n = 0; n < encControl->nChannelsAPI; n++ ) { + psEnc->state_Fxx[ n ].sCmn.nFramesEncoded = 0; + } + /* Check values in encoder control structure */ + if( ( ret = check_control_input( encControl ) ) != 0 ) { + celt_assert( 0 ); + RESTORE_STACK; + return ret; + } + + encControl->switchReady = 0; + + if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) { + /* Mono -> Stereo transition: init state of second channel and stereo state */ + ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ], psEnc->state_Fxx[ 0 ].sCmn.arch ); + silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) ); + silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) ); + psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0; + psEnc->sStereo.mid_side_amp_Q0[ 1 ] = 1; + psEnc->sStereo.mid_side_amp_Q0[ 2 ] = 0; + psEnc->sStereo.mid_side_amp_Q0[ 3 ] = 1; + psEnc->sStereo.width_prev_Q14 = 0; + psEnc->sStereo.smth_width_Q14 = SILK_FIX_CONST( 1, 14 ); + if( psEnc->nChannelsAPI == 2 ) { + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof( silk_resampler_state_struct ) ); + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.In_HP_State, &psEnc->state_Fxx[ 0 ].sCmn.In_HP_State, sizeof( psEnc->state_Fxx[ 1 ].sCmn.In_HP_State ) ); + } + } + + transition = (encControl->payloadSize_ms != psEnc->state_Fxx[ 0 ].sCmn.PacketSize_ms) || (psEnc->nChannelsInternal != encControl->nChannelsInternal); + + psEnc->nChannelsAPI = encControl->nChannelsAPI; + psEnc->nChannelsInternal = encControl->nChannelsInternal; + + nBlocksOf10ms = silk_DIV32( 100 * nSamplesIn, encControl->API_sampleRate ); + tot_blocks = ( nBlocksOf10ms > 1 ) ? nBlocksOf10ms >> 1 : 1; + curr_block = 0; + if( prefillFlag ) { + silk_LP_state save_LP; + /* Only accept input length of 10 ms */ + if( nBlocksOf10ms != 1 ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; + } + if ( prefillFlag == 2 ) { + save_LP = psEnc->state_Fxx[ 0 ].sCmn.sLP; + /* Save the sampling rate so the bandwidth switching code can keep handling transitions. */ + save_LP.saved_fs_kHz = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz; + } + /* Reset Encoder */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch ); + /* Restore the variable LP state. */ + if ( prefillFlag == 2 ) { + psEnc->state_Fxx[ n ].sCmn.sLP = save_LP; + } + celt_assert( !ret ); + } + tmp_payloadSize_ms = encControl->payloadSize_ms; + encControl->payloadSize_ms = 10; + tmp_complexity = encControl->complexity; + encControl->complexity = 0; + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0; + psEnc->state_Fxx[ n ].sCmn.prefillFlag = 1; + } + } else { + /* Only accept input lengths that are a multiple of 10 ms */ + if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; + } + /* Make sure no more than one packet can be produced */ + if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; + } + } + + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + /* Force the side channel to the same rate as the mid */ + opus_int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0; + if( ( ret = silk_control_encoder( &psEnc->state_Fxx[ n ], encControl, psEnc->allowBandwidthSwitch, n, force_fs_kHz ) ) != 0 ) { + silk_assert( 0 ); + RESTORE_STACK; + return ret; + } + if( psEnc->state_Fxx[n].sCmn.first_frame_after_reset || transition ) { + for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) { + psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] = 0; + } + } + psEnc->state_Fxx[ n ].sCmn.inDTX = psEnc->state_Fxx[ n ].sCmn.useDTX; + } + celt_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); + + /* Input buffering/resampling and encoding */ + nSamplesToBufferMax = + 10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz; + nSamplesFromInputMax = + silk_DIV32_16( nSamplesToBufferMax * + psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); + ALLOC( buf, nSamplesFromInputMax, opus_int16 ); + while( 1 ) { + int curr_nBitsUsedLBRR = 0; + nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx; + nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax ); + nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); + /* Resample and write to buffer */ + if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 2 ) { + opus_int id = psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded; + for( n = 0; n < nSamplesFromInput; n++ ) { + buf[ n ] = RES2INT16(samplesIn[ 2 * n ]); + } + /* Making sure to start both resamplers from the same state when switching from mono to stereo */ + if( psEnc->nPrevChannelsInternal == 1 && id==0 ) { + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof(psEnc->state_Fxx[ 1 ].sCmn.resampler_state)); + } + + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; + + nSamplesToBuffer = psEnc->state_Fxx[ 1 ].sCmn.frame_length - psEnc->state_Fxx[ 1 ].sCmn.inputBufIx; + nSamplesToBuffer = silk_min( nSamplesToBuffer, 10 * nBlocksOf10ms * psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); + for( n = 0; n < nSamplesFromInput; n++ ) { + buf[ n ] = RES2INT16(samplesIn[ 2 * n + 1 ]); + } + ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + + psEnc->state_Fxx[ 1 ].sCmn.inputBufIx += nSamplesToBuffer; + } else if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 1 ) { + /* Combine left and right channels before resampling */ + for( n = 0; n < nSamplesFromInput; n++ ) { + sum = RES2INT16(samplesIn[ 2 * n ] + samplesIn[ 2 * n + 1 ]); + buf[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 ); + } + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + /* On the first mono frame, average the results for the two resampler states */ + if( psEnc->nPrevChannelsInternal == 2 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 ) { + ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + for( n = 0; n < psEnc->state_Fxx[ 0 ].sCmn.frame_length; n++ ) { + psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] = + silk_RSHIFT(psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] + + psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx+n+2 ], 1); + } + } + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; + } else { + celt_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 ); + for( n = 0; n < nSamplesFromInput; n++ ) { + buf[n] = RES2INT16(samplesIn[n]); + } + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; + } + + samplesIn += nSamplesFromInput * encControl->nChannelsAPI; + nSamplesIn -= nSamplesFromInput; + + /* Default */ + psEnc->allowBandwidthSwitch = 0; + + /* Silk encoder */ + if( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx >= psEnc->state_Fxx[ 0 ].sCmn.frame_length ) { + /* Enough data in input buffer, so encode */ + celt_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length ); + celt_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length ); + + /* Deal with LBRR data */ + if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 && !prefillFlag ) { + /* Create space at start of payload for VAD and FEC flags */ + opus_uint8 iCDF[ 2 ] = { 0, 0 }; + iCDF[ 0 ] = 256 - silk_RSHIFT( 256, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal ); + ec_enc_icdf( psRangeEnc, 0, iCDF, 8 ); + curr_nBitsUsedLBRR = ec_tell( psRangeEnc ); + + /* Encode any LBRR data from previous packet */ + /* Encode LBRR flags */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + LBRR_symbol = 0; + for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) { + LBRR_symbol |= silk_LSHIFT( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ], i ); + } + psEnc->state_Fxx[ n ].sCmn.LBRR_flag = LBRR_symbol > 0 ? 1 : 0; + if( LBRR_symbol && psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket > 1 ) { + ec_enc_icdf( psRangeEnc, LBRR_symbol - 1, silk_LBRR_flags_iCDF_ptr[ psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket - 2 ], 8 ); + } + } + + /* Code LBRR indices and excitation signals */ + for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) { + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + if( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] ) { + opus_int condCoding; + + if( encControl->nChannelsInternal == 2 && n == 0 ) { + silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ i ] ); + /* For LBRR data there's no need to code the mid-only flag if the side-channel LBRR flag is set */ + if( psEnc->state_Fxx[ 1 ].sCmn.LBRR_flags[ i ] == 0 ) { + silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ i ] ); + } + } + /* Use conditional coding if previous frame available */ + if( i > 0 && psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i - 1 ] ) { + condCoding = CODE_CONDITIONALLY; + } else { + condCoding = CODE_INDEPENDENTLY; + } + silk_encode_indices( &psEnc->state_Fxx[ n ].sCmn, psRangeEnc, i, 1, condCoding ); + silk_encode_pulses( psRangeEnc, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].signalType, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].quantOffsetType, + psEnc->state_Fxx[ n ].sCmn.pulses_LBRR[ i ], psEnc->state_Fxx[ n ].sCmn.frame_length ); + } + } + } + + /* Reset LBRR flags */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + silk_memset( psEnc->state_Fxx[ n ].sCmn.LBRR_flags, 0, sizeof( psEnc->state_Fxx[ n ].sCmn.LBRR_flags ) ); + } + curr_nBitsUsedLBRR = ec_tell( psRangeEnc ) - curr_nBitsUsedLBRR; + } + + silk_HP_variable_cutoff( psEnc->state_Fxx ); + + /* Total target bits for packet */ + nBits = silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 ); + /* Subtract bits used for LBRR */ + if( !prefillFlag ) { + /* psEnc->nBitsUsedLBRR is an exponential moving average of the LBRR usage, + except that for the first LBRR frame it does no averaging and for the first + frame after after LBRR, it goes back to zero immediately. */ + if ( curr_nBitsUsedLBRR < 10 ) { + psEnc->nBitsUsedLBRR = 0; + } else if ( psEnc->nBitsUsedLBRR < 10) { + psEnc->nBitsUsedLBRR = curr_nBitsUsedLBRR; + } else { + psEnc->nBitsUsedLBRR = ( psEnc->nBitsUsedLBRR + curr_nBitsUsedLBRR ) / 2; + } + nBits -= psEnc->nBitsUsedLBRR; + } + /* Divide by number of uncoded frames left in packet */ + nBits = silk_DIV32_16( nBits, psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket ); + /* Convert to bits/second */ + if( encControl->payloadSize_ms == 10 ) { + TargetRate_bps = silk_SMULBB( nBits, 100 ); + } else { + TargetRate_bps = silk_SMULBB( nBits, 50 ); + } + /* Subtract fraction of bits in excess of target in previous frames and packets */ + TargetRate_bps -= silk_DIV32_16( silk_MUL( psEnc->nBitsExceeded, 1000 ), BITRESERVOIR_DECAY_TIME_MS ); + if( !prefillFlag && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded > 0 ) { + /* Compare actual vs target bits so far in this packet */ + opus_int32 bitsBalance = ec_tell( psRangeEnc ) - psEnc->nBitsUsedLBRR - nBits * psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded; + TargetRate_bps -= silk_DIV32_16( silk_MUL( bitsBalance, 1000 ), BITRESERVOIR_DECAY_TIME_MS ); + } + /* Never exceed input bitrate */ + TargetRate_bps = silk_LIMIT( TargetRate_bps, encControl->bitRate, 5000 ); + + /* Convert Left/Right to Mid/Side */ + if( encControl->nChannelsInternal == 2 ) { + silk_stereo_LR_to_MS( &psEnc->sStereo, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ 2 ], &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ 2 ], + psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], &psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], + MStargetRates_bps, TargetRate_bps, psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8, encControl->toMono, + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, psEnc->state_Fxx[ 0 ].sCmn.frame_length ); + if( psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) { + /* Reset side channel encoder memory for first frame with side coding */ + if( psEnc->prev_decode_only_middle == 1 ) { + silk_memset( &psEnc->state_Fxx[ 1 ].sShape, 0, sizeof( psEnc->state_Fxx[ 1 ].sShape ) ); + silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sNSQ, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sNSQ ) ); + silk_memset( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15 ) ); + silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State ) ); + psEnc->state_Fxx[ 1 ].sCmn.prevLag = 100; + psEnc->state_Fxx[ 1 ].sCmn.sNSQ.lagPrev = 100; + psEnc->state_Fxx[ 1 ].sShape.LastGainIndex = 10; + psEnc->state_Fxx[ 1 ].sCmn.prevSignalType = TYPE_NO_VOICE_ACTIVITY; + psEnc->state_Fxx[ 1 ].sCmn.sNSQ.prev_gain_Q16 = 65536; + psEnc->state_Fxx[ 1 ].sCmn.first_frame_after_reset = 1; + } + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ], activity ); + } else { + psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] = 0; + } + if( !prefillFlag ) { + silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] ); + if( psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) { + silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] ); + } + } + } else { + /* Buffering */ + silk_memcpy( psEnc->state_Fxx[ 0 ].sCmn.inputBuf, psEnc->sStereo.sMid, 2 * sizeof( opus_int16 ) ); + silk_memcpy( psEnc->sStereo.sMid, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.frame_length ], 2 * sizeof( opus_int16 ) ); + } + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ], activity ); + + /* Encode */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + opus_int maxBits, useCBR; + + /* Handling rate constraints */ + maxBits = encControl->maxBits; + if( tot_blocks == 2 && curr_block == 0 ) { + maxBits = maxBits * 3 / 5; + } else if( tot_blocks == 3 ) { + if( curr_block == 0 ) { + maxBits = maxBits * 2 / 5; + } else if( curr_block == 1 ) { + maxBits = maxBits * 3 / 4; + } + } + useCBR = encControl->useCBR && curr_block == tot_blocks - 1; + + if( encControl->nChannelsInternal == 1 ) { + channelRate_bps = TargetRate_bps; + } else { + channelRate_bps = MStargetRates_bps[ n ]; + if( n == 0 && MStargetRates_bps[ 1 ] > 0 ) { + useCBR = 0; + /* Give mid up to 1/2 of the max bits for that frame */ + maxBits -= encControl->maxBits / ( tot_blocks * 2 ); + } + } + + if( channelRate_bps > 0 ) { + opus_int condCoding; + + silk_control_SNR( &psEnc->state_Fxx[ n ].sCmn, channelRate_bps ); + + /* Use independent coding if no previous frame available */ + if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - n <= 0 ) { + condCoding = CODE_INDEPENDENTLY; + } else if( n > 0 && psEnc->prev_decode_only_middle ) { + /* If we skipped a side frame in this packet, we don't + need LTP scaling; the LTP state is well-defined. */ + condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING; + } else { + condCoding = CODE_CONDITIONALLY; + } + if( ( ret = silk_encode_frame_Fxx( &psEnc->state_Fxx[ n ], nBytesOut, psRangeEnc, condCoding, maxBits, useCBR ) ) != 0 ) { + silk_assert( 0 ); + } + } + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0; + psEnc->state_Fxx[ n ].sCmn.inputBufIx = 0; + psEnc->state_Fxx[ n ].sCmn.nFramesEncoded++; + } + psEnc->prev_decode_only_middle = psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - 1 ]; + + /* Insert VAD and FEC flags at beginning of bitstream */ + if( *nBytesOut > 0 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket) { + flags = 0; + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) { + flags = silk_LSHIFT( flags, 1 ); + flags |= psEnc->state_Fxx[ n ].sCmn.VAD_flags[ i ]; + } + flags = silk_LSHIFT( flags, 1 ); + flags |= psEnc->state_Fxx[ n ].sCmn.LBRR_flag; + } + if( !prefillFlag ) { + ec_enc_patch_initial_bits( psRangeEnc, flags, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal ); + } + + /* Return zero bytes if all channels DTXed */ + if( psEnc->state_Fxx[ 0 ].sCmn.inDTX && ( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inDTX ) ) { + *nBytesOut = 0; + } + + psEnc->nBitsExceeded += *nBytesOut * 8; + psEnc->nBitsExceeded -= silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 ); + psEnc->nBitsExceeded = silk_LIMIT( psEnc->nBitsExceeded, 0, 10000 ); + + /* Update flag indicating if bandwidth switching is allowed */ + speech_act_thr_for_switch_Q8 = silk_SMLAWB( SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ), + SILK_FIX_CONST( ( 1 - SPEECH_ACTIVITY_DTX_THRES ) / MAX_BANDWIDTH_SWITCH_DELAY_MS, 16 + 8 ), psEnc->timeSinceSwitchAllowed_ms ); + if( psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8 < speech_act_thr_for_switch_Q8 ) { + psEnc->allowBandwidthSwitch = 1; + psEnc->timeSinceSwitchAllowed_ms = 0; + } else { + psEnc->allowBandwidthSwitch = 0; + psEnc->timeSinceSwitchAllowed_ms += encControl->payloadSize_ms; + } + } + + if( nSamplesIn == 0 ) { + break; + } + } else { + break; + } + curr_block++; + } + + psEnc->nPrevChannelsInternal = encControl->nChannelsInternal; + + encControl->allowBandwidthSwitch = psEnc->allowBandwidthSwitch; + encControl->inWBmodeWithoutVariableLP = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == 16 && psEnc->state_Fxx[ 0 ].sCmn.sLP.mode == 0; + encControl->internalSampleRate = silk_SMULBB( psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, 1000 ); + encControl->stereoWidth_Q14 = encControl->toMono ? 0 : psEnc->sStereo.smth_width_Q14; + if( prefillFlag ) { + encControl->payloadSize_ms = tmp_payloadSize_ms; + encControl->complexity = tmp_complexity; + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0; + psEnc->state_Fxx[ n ].sCmn.prefillFlag = 0; + } + } + + encControl->signalType = psEnc->state_Fxx[0].sCmn.indices.signalType; + encControl->offset = silk_Quantization_Offsets_Q10 + [ psEnc->state_Fxx[0].sCmn.indices.signalType >> 1 ] + [ psEnc->state_Fxx[0].sCmn.indices.quantOffsetType ]; + RESTORE_STACK; + return ret; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_indices.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_indices.c new file mode 100644 index 000000000..8a6db2cbe --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_indices.c @@ -0,0 +1,181 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Encode side-information parameters to payload */ +void silk_encode_indices( + silk_encoder_state *psEncC, /* I/O Encoder state */ + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int FrameIndex, /* I Frame number */ + opus_int encode_LBRR, /* I Flag indicating LBRR data is being encoded */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i, k, typeOffset; + opus_int encode_absolute_lagIndex, delta_lagIndex; + opus_int16 ec_ix[ MAX_LPC_ORDER ]; + opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; + const SideInfoIndices *psIndices; + + if( encode_LBRR ) { + psIndices = &psEncC->indices_LBRR[ FrameIndex ]; + } else { + psIndices = &psEncC->indices; + } + + /*******************************************/ + /* Encode signal type and quantizer offset */ + /*******************************************/ + typeOffset = 2 * psIndices->signalType + psIndices->quantOffsetType; + celt_assert( typeOffset >= 0 && typeOffset < 6 ); + celt_assert( encode_LBRR == 0 || typeOffset >= 2 ); + if( encode_LBRR || typeOffset >= 2 ) { + ec_enc_icdf( psRangeEnc, typeOffset - 2, silk_type_offset_VAD_iCDF, 8 ); + } else { + ec_enc_icdf( psRangeEnc, typeOffset, silk_type_offset_no_VAD_iCDF, 8 ); + } + + /****************/ + /* Encode gains */ + /****************/ + /* first subframe */ + if( condCoding == CODE_CONDITIONALLY ) { + /* conditional coding */ + silk_assert( psIndices->GainsIndices[ 0 ] >= 0 && psIndices->GainsIndices[ 0 ] < MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 ); + ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ 0 ], silk_delta_gain_iCDF, 8 ); + } else { + /* independent coding, in two stages: MSB bits followed by 3 LSBs */ + silk_assert( psIndices->GainsIndices[ 0 ] >= 0 && psIndices->GainsIndices[ 0 ] < N_LEVELS_QGAIN ); + ec_enc_icdf( psRangeEnc, silk_RSHIFT( psIndices->GainsIndices[ 0 ], 3 ), silk_gain_iCDF[ psIndices->signalType ], 8 ); + ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ 0 ] & 7, silk_uniform8_iCDF, 8 ); + } + + /* remaining subframes */ + for( i = 1; i < psEncC->nb_subfr; i++ ) { + silk_assert( psIndices->GainsIndices[ i ] >= 0 && psIndices->GainsIndices[ i ] < MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 ); + ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ i ], silk_delta_gain_iCDF, 8 ); + } + + /****************/ + /* Encode NLSFs */ + /****************/ + ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ 0 ], &psEncC->psNLSF_CB->CB1_iCDF[ ( psIndices->signalType >> 1 ) * psEncC->psNLSF_CB->nVectors ], 8 ); + silk_NLSF_unpack( ec_ix, pred_Q8, psEncC->psNLSF_CB, psIndices->NLSFIndices[ 0 ] ); + celt_assert( psEncC->psNLSF_CB->order == psEncC->predictLPCOrder ); + for( i = 0; i < psEncC->psNLSF_CB->order; i++ ) { + if( psIndices->NLSFIndices[ i+1 ] >= NLSF_QUANT_MAX_AMPLITUDE ) { + ec_enc_icdf( psRangeEnc, 2 * NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ i+1 ] - NLSF_QUANT_MAX_AMPLITUDE, silk_NLSF_EXT_iCDF, 8 ); + } else if( psIndices->NLSFIndices[ i+1 ] <= -NLSF_QUANT_MAX_AMPLITUDE ) { + ec_enc_icdf( psRangeEnc, 0, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + ec_enc_icdf( psRangeEnc, -psIndices->NLSFIndices[ i+1 ] - NLSF_QUANT_MAX_AMPLITUDE, silk_NLSF_EXT_iCDF, 8 ); + } else { + ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ i+1 ] + NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + } + } + + /* Encode NLSF interpolation factor */ + if( psEncC->nb_subfr == MAX_NB_SUBFR ) { + silk_assert( psIndices->NLSFInterpCoef_Q2 >= 0 && psIndices->NLSFInterpCoef_Q2 < 5 ); + ec_enc_icdf( psRangeEnc, psIndices->NLSFInterpCoef_Q2, silk_NLSF_interpolation_factor_iCDF, 8 ); + } + + if( psIndices->signalType == TYPE_VOICED ) + { + /*********************/ + /* Encode pitch lags */ + /*********************/ + /* lag index */ + encode_absolute_lagIndex = 1; + if( condCoding == CODE_CONDITIONALLY && psEncC->ec_prevSignalType == TYPE_VOICED ) { + /* Delta Encoding */ + delta_lagIndex = psIndices->lagIndex - psEncC->ec_prevLagIndex; + if( delta_lagIndex < -8 || delta_lagIndex > 11 ) { + delta_lagIndex = 0; + } else { + delta_lagIndex = delta_lagIndex + 9; + encode_absolute_lagIndex = 0; /* Only use delta */ + } + silk_assert( delta_lagIndex >= 0 && delta_lagIndex < 21 ); + ec_enc_icdf( psRangeEnc, delta_lagIndex, silk_pitch_delta_iCDF, 8 ); + } + if( encode_absolute_lagIndex ) { + /* Absolute encoding */ + opus_int32 pitch_high_bits, pitch_low_bits; + pitch_high_bits = silk_DIV32_16( psIndices->lagIndex, silk_RSHIFT( psEncC->fs_kHz, 1 ) ); + pitch_low_bits = psIndices->lagIndex - silk_SMULBB( pitch_high_bits, silk_RSHIFT( psEncC->fs_kHz, 1 ) ); + silk_assert( pitch_low_bits < psEncC->fs_kHz / 2 ); + silk_assert( pitch_high_bits < 32 ); + ec_enc_icdf( psRangeEnc, pitch_high_bits, silk_pitch_lag_iCDF, 8 ); + ec_enc_icdf( psRangeEnc, pitch_low_bits, psEncC->pitch_lag_low_bits_iCDF, 8 ); + } + psEncC->ec_prevLagIndex = psIndices->lagIndex; + + /* Contour index */ + silk_assert( psIndices->contourIndex >= 0 ); + silk_assert( ( psIndices->contourIndex < 34 && psEncC->fs_kHz > 8 && psEncC->nb_subfr == 4 ) || + ( psIndices->contourIndex < 11 && psEncC->fs_kHz == 8 && psEncC->nb_subfr == 4 ) || + ( psIndices->contourIndex < 12 && psEncC->fs_kHz > 8 && psEncC->nb_subfr == 2 ) || + ( psIndices->contourIndex < 3 && psEncC->fs_kHz == 8 && psEncC->nb_subfr == 2 ) ); + ec_enc_icdf( psRangeEnc, psIndices->contourIndex, psEncC->pitch_contour_iCDF, 8 ); + + /********************/ + /* Encode LTP gains */ + /********************/ + /* PERIndex value */ + silk_assert( psIndices->PERIndex >= 0 && psIndices->PERIndex < 3 ); + ec_enc_icdf( psRangeEnc, psIndices->PERIndex, silk_LTP_per_index_iCDF, 8 ); + + /* Codebook Indices */ + for( k = 0; k < psEncC->nb_subfr; k++ ) { + silk_assert( psIndices->LTPIndex[ k ] >= 0 && psIndices->LTPIndex[ k ] < ( 8 << psIndices->PERIndex ) ); + ec_enc_icdf( psRangeEnc, psIndices->LTPIndex[ k ], silk_LTP_gain_iCDF_ptrs[ psIndices->PERIndex ], 8 ); + } + + /**********************/ + /* Encode LTP scaling */ + /**********************/ + if( condCoding == CODE_INDEPENDENTLY ) { + silk_assert( psIndices->LTP_scaleIndex >= 0 && psIndices->LTP_scaleIndex < 3 ); + ec_enc_icdf( psRangeEnc, psIndices->LTP_scaleIndex, silk_LTPscale_iCDF, 8 ); + } + silk_assert( !condCoding || psIndices->LTP_scaleIndex == 0 ); + } + + psEncC->ec_prevSignalType = psIndices->signalType; + + /***************/ + /* Encode seed */ + /***************/ + silk_assert( psIndices->Seed >= 0 && psIndices->Seed < 4 ); + ec_enc_icdf( psRangeEnc, psIndices->Seed, silk_uniform4_iCDF, 8 ); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_pulses.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_pulses.c new file mode 100644 index 000000000..8a1999138 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/encode_pulses.c @@ -0,0 +1,206 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/*********************************************/ +/* Encode quantization indices of excitation */ +/*********************************************/ + +static OPUS_INLINE opus_int combine_and_check( /* return ok */ + opus_int *pulses_comb, /* O */ + const opus_int *pulses_in, /* I */ + opus_int max_pulses, /* I max value for sum of pulses */ + opus_int len /* I number of output values */ +) +{ + opus_int k, sum; + + for( k = 0; k < len; k++ ) { + sum = pulses_in[ 2 * k ] + pulses_in[ 2 * k + 1 ]; + if( sum > max_pulses ) { + return 1; + } + pulses_comb[ k ] = sum; + } + + return 0; +} + +/* Encode quantization indices of excitation */ +void silk_encode_pulses( + ec_enc *psRangeEnc, /* I/O compressor data structure */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I quantOffsetType */ + opus_int8 pulses[], /* I quantization indices */ + const opus_int frame_length /* I Frame length */ +) +{ + opus_int i, k, j, iter, bit, nLS, scale_down, RateLevelIndex = 0; + opus_int32 abs_q, minSumBits_Q5, sumBits_Q5; + VARDECL( opus_int, abs_pulses ); + VARDECL( opus_int, sum_pulses ); + VARDECL( opus_int, nRshifts ); + opus_int pulses_comb[ 8 ]; + opus_int *abs_pulses_ptr; + const opus_int8 *pulses_ptr; + const opus_uint8 *cdf_ptr; + const opus_uint8 *nBits_ptr; + SAVE_STACK; + + silk_memset( pulses_comb, 0, 8 * sizeof( opus_int ) ); /* Fixing Valgrind reported problem*/ + + /****************************/ + /* Prepare for shell coding */ + /****************************/ + /* Calculate number of shell blocks */ + silk_assert( 1 << LOG2_SHELL_CODEC_FRAME_LENGTH == SHELL_CODEC_FRAME_LENGTH ); + iter = silk_RSHIFT( frame_length, LOG2_SHELL_CODEC_FRAME_LENGTH ); + if( iter * SHELL_CODEC_FRAME_LENGTH < frame_length ) { + celt_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ + iter++; + silk_memset( &pulses[ frame_length ], 0, SHELL_CODEC_FRAME_LENGTH * sizeof(opus_int8)); + } + + /* Take the absolute value of the pulses */ + ALLOC( abs_pulses, iter * SHELL_CODEC_FRAME_LENGTH, opus_int ); + silk_assert( !( SHELL_CODEC_FRAME_LENGTH & 3 ) ); + for( i = 0; i < iter * SHELL_CODEC_FRAME_LENGTH; i+=4 ) { + abs_pulses[i+0] = ( opus_int )silk_abs( pulses[ i + 0 ] ); + abs_pulses[i+1] = ( opus_int )silk_abs( pulses[ i + 1 ] ); + abs_pulses[i+2] = ( opus_int )silk_abs( pulses[ i + 2 ] ); + abs_pulses[i+3] = ( opus_int )silk_abs( pulses[ i + 3 ] ); + } + + /* Calc sum pulses per shell code frame */ + ALLOC( sum_pulses, iter, opus_int ); + ALLOC( nRshifts, iter, opus_int ); + abs_pulses_ptr = abs_pulses; + for( i = 0; i < iter; i++ ) { + nRshifts[ i ] = 0; + + while( 1 ) { + /* 1+1 -> 2 */ + scale_down = combine_and_check( pulses_comb, abs_pulses_ptr, silk_max_pulses_table[ 0 ], 8 ); + /* 2+2 -> 4 */ + scale_down += combine_and_check( pulses_comb, pulses_comb, silk_max_pulses_table[ 1 ], 4 ); + /* 4+4 -> 8 */ + scale_down += combine_and_check( pulses_comb, pulses_comb, silk_max_pulses_table[ 2 ], 2 ); + /* 8+8 -> 16 */ + scale_down += combine_and_check( &sum_pulses[ i ], pulses_comb, silk_max_pulses_table[ 3 ], 1 ); + + if( scale_down ) { + /* We need to downscale the quantization signal */ + nRshifts[ i ]++; + for( k = 0; k < SHELL_CODEC_FRAME_LENGTH; k++ ) { + abs_pulses_ptr[ k ] = silk_RSHIFT( abs_pulses_ptr[ k ], 1 ); + } + } else { + /* Jump out of while(1) loop and go to next shell coding frame */ + break; + } + } + abs_pulses_ptr += SHELL_CODEC_FRAME_LENGTH; + } + + /**************/ + /* Rate level */ + /**************/ + /* find rate level that leads to fewest bits for coding of pulses per block info */ + minSumBits_Q5 = silk_int32_MAX; + for( k = 0; k < N_RATE_LEVELS - 1; k++ ) { + nBits_ptr = silk_pulses_per_block_BITS_Q5[ k ]; + sumBits_Q5 = silk_rate_levels_BITS_Q5[ signalType >> 1 ][ k ]; + for( i = 0; i < iter; i++ ) { + if( nRshifts[ i ] > 0 ) { + sumBits_Q5 += nBits_ptr[ SILK_MAX_PULSES + 1 ]; + } else { + sumBits_Q5 += nBits_ptr[ sum_pulses[ i ] ]; + } + } + if( sumBits_Q5 < minSumBits_Q5 ) { + minSumBits_Q5 = sumBits_Q5; + RateLevelIndex = k; + } + } + ec_enc_icdf( psRangeEnc, RateLevelIndex, silk_rate_levels_iCDF[ signalType >> 1 ], 8 ); + + /***************************************************/ + /* Sum-Weighted-Pulses Encoding */ + /***************************************************/ + cdf_ptr = silk_pulses_per_block_iCDF[ RateLevelIndex ]; + for( i = 0; i < iter; i++ ) { + if( nRshifts[ i ] == 0 ) { + ec_enc_icdf( psRangeEnc, sum_pulses[ i ], cdf_ptr, 8 ); + } else { + ec_enc_icdf( psRangeEnc, SILK_MAX_PULSES + 1, cdf_ptr, 8 ); + for( k = 0; k < nRshifts[ i ] - 1; k++ ) { + ec_enc_icdf( psRangeEnc, SILK_MAX_PULSES + 1, silk_pulses_per_block_iCDF[ N_RATE_LEVELS - 1 ], 8 ); + } + ec_enc_icdf( psRangeEnc, sum_pulses[ i ], silk_pulses_per_block_iCDF[ N_RATE_LEVELS - 1 ], 8 ); + } + } + + /******************/ + /* Shell Encoding */ + /******************/ + for( i = 0; i < iter; i++ ) { + if( sum_pulses[ i ] > 0 ) { + silk_shell_encoder( psRangeEnc, &abs_pulses[ i * SHELL_CODEC_FRAME_LENGTH ] ); + } + } + + /****************/ + /* LSB Encoding */ + /****************/ + for( i = 0; i < iter; i++ ) { + if( nRshifts[ i ] > 0 ) { + pulses_ptr = &pulses[ i * SHELL_CODEC_FRAME_LENGTH ]; + nLS = nRshifts[ i ] - 1; + for( k = 0; k < SHELL_CODEC_FRAME_LENGTH; k++ ) { + abs_q = (opus_int8)silk_abs( pulses_ptr[ k ] ); + for( j = nLS; j > 0; j-- ) { + bit = silk_RSHIFT( abs_q, j ) & 1; + ec_enc_icdf( psRangeEnc, bit, silk_lsb_iCDF, 8 ); + } + bit = abs_q & 1; + ec_enc_icdf( psRangeEnc, bit, silk_lsb_iCDF, 8 ); + } + } + } + + /****************/ + /* Encode signs */ + /****************/ + silk_encode_signs( psRangeEnc, pulses, frame_length, signalType, quantOffsetType, sum_pulses ); + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_analysis_filter_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_analysis_filter_FIX.c new file mode 100644 index 000000000..c1a4f0d45 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_analysis_filter_FIX.c @@ -0,0 +1,89 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +void silk_LTP_analysis_filter_FIX( + opus_int16 *LTP_res, /* O LTP residual signal of length MAX_NB_SUBFR * ( pre_length + subfr_length ) */ + const opus_int16 *x, /* I Pointer to input signal with at least max( pitchL ) preceding samples */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],/* I LTP_ORDER LTP coefficients for each MAX_NB_SUBFR subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag, one for each subframe */ + const opus_int32 invGains_Q16[ MAX_NB_SUBFR ], /* I Inverse quantization gains, one for each subframe */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int pre_length /* I Length of the preceding samples starting at &x[0] for each subframe */ +) +{ + const opus_int16 *x_ptr, *x_lag_ptr; + opus_int16 Btmp_Q14[ LTP_ORDER ]; + opus_int16 *LTP_res_ptr; + opus_int k, i; + opus_int32 LTP_est; + + x_ptr = x; + LTP_res_ptr = LTP_res; + for( k = 0; k < nb_subfr; k++ ) { + + x_lag_ptr = x_ptr - pitchL[ k ]; + + Btmp_Q14[ 0 ] = LTPCoef_Q14[ k * LTP_ORDER ]; + Btmp_Q14[ 1 ] = LTPCoef_Q14[ k * LTP_ORDER + 1 ]; + Btmp_Q14[ 2 ] = LTPCoef_Q14[ k * LTP_ORDER + 2 ]; + Btmp_Q14[ 3 ] = LTPCoef_Q14[ k * LTP_ORDER + 3 ]; + Btmp_Q14[ 4 ] = LTPCoef_Q14[ k * LTP_ORDER + 4 ]; + + /* LTP analysis FIR filter */ + for( i = 0; i < subfr_length + pre_length; i++ ) { + LTP_res_ptr[ i ] = x_ptr[ i ]; + + /* Long-term prediction */ + LTP_est = silk_SMULBB( x_lag_ptr[ LTP_ORDER / 2 ], Btmp_Q14[ 0 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ 1 ], Btmp_Q14[ 1 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ 0 ], Btmp_Q14[ 2 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ -1 ], Btmp_Q14[ 3 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ -2 ], Btmp_Q14[ 4 ] ); + + LTP_est = silk_RSHIFT_ROUND( LTP_est, 14 ); /* round and -> Q0*/ + + /* Subtract long-term prediction */ + LTP_res_ptr[ i ] = (opus_int16)silk_SAT16( (opus_int32)x_ptr[ i ] - LTP_est ); + + /* Scale residual */ + LTP_res_ptr[ i ] = silk_SMULWB( invGains_Q16[ k ], LTP_res_ptr[ i ] ); + + x_lag_ptr++; + } + + /* Update pointers */ + LTP_res_ptr += subfr_length + pre_length; + x_ptr += subfr_length; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_scale_ctrl_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_scale_ctrl_FIX.c new file mode 100644 index 000000000..db1016e0b --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/LTP_scale_ctrl_FIX.c @@ -0,0 +1,58 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +/* Calculation of LTP state scaling */ +void silk_LTP_scale_ctrl_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int round_loss; + + if( condCoding == CODE_INDEPENDENTLY ) { + /* Only scale if first frame in packet */ + round_loss = psEnc->sCmn.PacketLoss_perc * psEnc->sCmn.nFramesPerPacket; + if ( psEnc->sCmn.LBRR_flag ) { + /* LBRR reduces the effective loss. In practice, it does not square the loss because + losses aren't independent, but that still seems to work best. We also never go below 2%. */ + round_loss = 2 + silk_SMULBB( round_loss, round_loss ) / 100; + } + psEnc->sCmn.indices.LTP_scaleIndex = silk_SMULBB( psEncCtrl->LTPredCodGain_Q7, round_loss ) > silk_log2lin( 128*7 + 2900-psEnc->sCmn.SNR_dB_Q7 ); + psEnc->sCmn.indices.LTP_scaleIndex += silk_SMULBB( psEncCtrl->LTPredCodGain_Q7, round_loss ) > silk_log2lin( 128*7 + 3900-psEnc->sCmn.SNR_dB_Q7 ); + } else { + /* Default is minimum scaling */ + psEnc->sCmn.indices.LTP_scaleIndex = 0; + } + psEncCtrl->LTP_scale_Q14 = silk_LTPScales_table_Q14[ psEnc->sCmn.indices.LTP_scaleIndex ]; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/apply_sine_window_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/apply_sine_window_FIX.c new file mode 100644 index 000000000..03e088a6d --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/apply_sine_window_FIX.c @@ -0,0 +1,101 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Apply sine window to signal vector. */ +/* Window types: */ +/* 1 -> sine window from 0 to pi/2 */ +/* 2 -> sine window from pi/2 to pi */ +/* Every other sample is linearly interpolated, for speed. */ +/* Window length must be between 16 and 120 (incl) and a multiple of 4. */ + +/* Matlab code for table: + for k=16:9*4:16+2*9*4, fprintf(' %7.d,', -round(65536*pi ./ (k:4:k+8*4))); fprintf('\n'); end +*/ +static const opus_int16 freq_table_Q16[ 27 ] = { + 12111, 9804, 8235, 7100, 6239, 5565, 5022, 4575, 4202, + 3885, 3612, 3375, 3167, 2984, 2820, 2674, 2542, 2422, + 2313, 2214, 2123, 2038, 1961, 1889, 1822, 1760, 1702, +}; + +void silk_apply_sine_window( + opus_int16 px_win[], /* O Pointer to windowed signal */ + const opus_int16 px[], /* I Pointer to input signal */ + const opus_int win_type, /* I Selects a window type */ + const opus_int length /* I Window length, multiple of 4 */ +) +{ + opus_int k, f_Q16, c_Q16; + opus_int32 S0_Q16, S1_Q16; + + celt_assert( win_type == 1 || win_type == 2 ); + + /* Length must be in a range from 16 to 120 and a multiple of 4 */ + celt_assert( length >= 16 && length <= 120 ); + celt_assert( ( length & 3 ) == 0 ); + + /* Frequency */ + k = ( length >> 2 ) - 4; + celt_assert( k >= 0 && k <= 26 ); + f_Q16 = (opus_int)freq_table_Q16[ k ]; + + /* Factor used for cosine approximation */ + c_Q16 = silk_SMULWB( (opus_int32)f_Q16, -f_Q16 ); + silk_assert( c_Q16 >= -32768 ); + + /* initialize state */ + if( win_type == 1 ) { + /* start from 0 */ + S0_Q16 = 0; + /* approximation of sin(f) */ + S1_Q16 = f_Q16 + silk_RSHIFT( length, 3 ); + } else { + /* start from 1 */ + S0_Q16 = ( (opus_int32)1 << 16 ); + /* approximation of cos(f) */ + S1_Q16 = ( (opus_int32)1 << 16 ) + silk_RSHIFT( c_Q16, 1 ) + silk_RSHIFT( length, 4 ); + } + + /* Uses the recursive equation: sin(n*f) = 2 * cos(f) * sin((n-1)*f) - sin((n-2)*f) */ + /* 4 samples at a time */ + for( k = 0; k < length; k += 4 ) { + px_win[ k ] = (opus_int16)silk_SMULWB( silk_RSHIFT( S0_Q16 + S1_Q16, 1 ), px[ k ] ); + px_win[ k + 1 ] = (opus_int16)silk_SMULWB( S1_Q16, px[ k + 1] ); + S0_Q16 = silk_SMULWB( S1_Q16, c_Q16 ) + silk_LSHIFT( S1_Q16, 1 ) - S0_Q16 + 1; + S0_Q16 = silk_min( S0_Q16, ( (opus_int32)1 << 16 ) ); + + px_win[ k + 2 ] = (opus_int16)silk_SMULWB( silk_RSHIFT( S0_Q16 + S1_Q16, 1 ), px[ k + 2] ); + px_win[ k + 3 ] = (opus_int16)silk_SMULWB( S0_Q16, px[ k + 3 ] ); + S1_Q16 = silk_SMULWB( S0_Q16, c_Q16 ) + silk_LSHIFT( S0_Q16, 1 ) - S1_Q16; + S1_Q16 = silk_min( S1_Q16, ( (opus_int32)1 << 16 ) ); + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/autocorr_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/autocorr_FIX.c new file mode 100644 index 000000000..de95c9869 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/autocorr_FIX.c @@ -0,0 +1,48 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "celt_lpc.h" + +/* Compute autocorrelation */ +void silk_autocorr( + opus_int32 *results, /* O Result (length correlationCount) */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *inputData, /* I Input data to correlate */ + const opus_int inputDataSize, /* I Length of input */ + const opus_int correlationCount, /* I Number of correlation taps to compute */ + int arch /* I Run-time architecture */ +) +{ + opus_int corrCount; + corrCount = silk_min_int( inputDataSize, correlationCount ); + *scale = _celt_autocorr(inputData, results, NULL, 0, corrCount-1, inputDataSize, arch); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX.c new file mode 100644 index 000000000..185a12b17 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX.c @@ -0,0 +1,280 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "define.h" +#include "tuning_parameters.h" +#include "pitch.h" + +#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */ + +#define QA 25 +#define N_BITS_HEAD_ROOM 3 +#define MIN_RSHIFTS -16 +#define MAX_RSHIFTS (32 - QA) + +/* Compute reflection coefficients from input signal */ +void silk_burg_modified_c( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +) +{ + opus_int k, n, s, lz, rshifts, reached_max_gain; + opus_int32 C0, num, nrg, rc_Q31, invGain_Q30, Atmp_QA, Atmp1, tmp1, tmp2, x1, x2; + const opus_int16 *x_ptr; + opus_int32 C_first_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 C_last_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 Af_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 CAf[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 CAb[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; + opus_int64 C0_64; + + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + + /* Compute autocorrelations, added over subframes */ + C0_64 = silk_inner_prod16( x, x, subfr_length*nb_subfr, arch ); + lz = silk_CLZ64(C0_64); + rshifts = 32 + 1 + N_BITS_HEAD_ROOM - lz; + if (rshifts > MAX_RSHIFTS) rshifts = MAX_RSHIFTS; + if (rshifts < MIN_RSHIFTS) rshifts = MIN_RSHIFTS; + + if (rshifts > 0) { + C0 = (opus_int32)silk_RSHIFT64(C0_64, rshifts ); + } else { + C0 = silk_LSHIFT32((opus_int32)C0_64, -rshifts ); + } + + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += (opus_int32)silk_RSHIFT64( + silk_inner_prod16( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + int i; + opus_int32 d; + x_ptr = x + s * subfr_length; + celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch ); + for( n = 1; n < D + 1; n++ ) { + for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ ) + d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] ); + xcorr[ n - 1 ] += d; + } + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_LSHIFT32( xcorr[ n - 1 ], -rshifts ); + } + } + } + silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + + /* Initialize */ + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + + invGain_Q30 = (opus_int32)1 << 30; + reached_max_gain = 0; + for( n = 0; n < D; n++ ) { + /* Update first row of correlation matrix (without first element) */ + /* Update last row of correlation matrix (without last element, stored in reversed order) */ + /* Update C * Af */ + /* Update C * flipud(Af) (stored in reversed order) */ + if( rshifts > -2 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], 16 - rshifts ); /* Q(16-rshifts) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 16 - rshifts ); /* Q(16-rshifts) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], QA - 16 ); /* Q(QA-16) */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], QA - 16 ); /* Q(QA-16) */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_SMLAWB( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_SMLAWB( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp_QA = Af_QA[ k ]; + tmp1 = silk_SMLAWB( tmp1, Atmp_QA, x_ptr[ n - k - 1 ] ); /* Q(QA-16) */ + tmp2 = silk_SMLAWB( tmp2, Atmp_QA, x_ptr[ subfr_length - n + k ] ); /* Q(QA-16) */ + } + tmp1 = silk_LSHIFT32( -tmp1, 32 - QA - rshifts ); /* Q(16-rshifts) */ + tmp2 = silk_LSHIFT32( -tmp2, 32 - QA - rshifts ); /* Q(16-rshifts) */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWB( CAf[ k ], tmp1, x_ptr[ n - k ] ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWB( CAb[ k ], tmp2, x_ptr[ subfr_length - n + k - 1 ] ); /* Q( -rshift ) */ + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], -rshifts ); /* Q( -rshifts ) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], -rshifts ); /* Q( -rshifts ) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], 17 ); /* Q17 */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 17 ); /* Q17 */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ + /* We sometimes get overflows in the multiplications (even beyond +/- 2^32), + but they cancel each other and the real result seems to always fit in a 32-bit + signed integer. This was determined experimentally, not theoretically (unfortunately). */ + tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ + tmp2 = silk_MLA_ovflw( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ + } + tmp1 = -tmp1; /* Q17 */ + tmp2 = -tmp2; /* Q17 */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWW( CAf[ k ], tmp1, + silk_LSHIFT32( (opus_int32)x_ptr[ n - k ], -rshifts - 1 ) ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWW( CAb[ k ], tmp2, + silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n + k - 1 ], -rshifts - 1 ) ); /* Q( -rshift ) */ + } + } + } + + /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */ + tmp1 = C_first_row[ n ]; /* Q( -rshifts ) */ + tmp2 = C_last_row[ n ]; /* Q( -rshifts ) */ + num = 0; /* Q( -rshifts ) */ + nrg = silk_ADD32( CAb[ 0 ], CAf[ 0 ] ); /* Q( 1-rshifts ) */ + for( k = 0; k < n; k++ ) { + Atmp_QA = Af_QA[ k ]; + lz = silk_CLZ32( silk_abs( Atmp_QA ) ) - 1; + lz = silk_min( 32 - QA, lz ); + Atmp1 = silk_LSHIFT32( Atmp_QA, lz ); /* Q( QA + lz ) */ + + tmp1 = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( C_last_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + tmp2 = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( C_first_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + num = silk_ADD_LSHIFT32( num, silk_SMMUL( CAb[ n - k ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + nrg = silk_ADD_LSHIFT32( nrg, silk_SMMUL( silk_ADD32( CAb[ k + 1 ], CAf[ k + 1 ] ), + Atmp1 ), 32 - QA - lz ); /* Q( 1-rshifts ) */ + } + CAf[ n + 1 ] = tmp1; /* Q( -rshifts ) */ + CAb[ n + 1 ] = tmp2; /* Q( -rshifts ) */ + num = silk_ADD32( num, tmp2 ); /* Q( -rshifts ) */ + num = silk_LSHIFT32( -num, 1 ); /* Q( 1-rshifts ) */ + + /* Calculate the next order reflection (parcor) coefficient */ + if( silk_abs( num ) < nrg ) { + rc_Q31 = silk_DIV32_varQ( num, nrg, 31 ); + } else { + rc_Q31 = ( num > 0 ) ? silk_int32_MAX : silk_int32_MIN; + } + + /* Update inverse prediction gain */ + tmp1 = ( (opus_int32)1 << 30 ) - silk_SMMUL( rc_Q31, rc_Q31 ); + tmp1 = silk_LSHIFT( silk_SMMUL( invGain_Q30, tmp1 ), 2 ); + if( tmp1 <= minInvGain_Q30 ) { + /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ + tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ + rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } + } + invGain_Q30 = minInvGain_Q30; + reached_max_gain = 1; + } else { + invGain_Q30 = tmp1; + } + + /* Update the AR coefficients */ + for( k = 0; k < (n + 1) >> 1; k++ ) { + tmp1 = Af_QA[ k ]; /* QA */ + tmp2 = Af_QA[ n - k - 1 ]; /* QA */ + Af_QA[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* QA */ + Af_QA[ n - k - 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* QA */ + } + Af_QA[ n ] = silk_RSHIFT32( rc_Q31, 31 - QA ); /* QA */ + + if( reached_max_gain ) { + /* Reached max prediction gain; set remaining coefficients to zero and exit loop */ + for( k = n + 1; k < D; k++ ) { + Af_QA[ k ] = 0; + } + break; + } + + /* Update C * Af and C * Ab */ + for( k = 0; k <= n + 1; k++ ) { + tmp1 = CAf[ k ]; /* Q( -rshifts ) */ + tmp2 = CAb[ n - k + 1 ]; /* Q( -rshifts ) */ + CAf[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + CAb[ n - k + 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + } + } + + if( reached_max_gain ) { + for( k = 0; k < D; k++ ) { + /* Scale coefficients */ + A_Q16[ k ] = -silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); + } + /* Subtract energy of preceding samples from C0 */ + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16( x_ptr, x_ptr, D, arch ), rshifts ); + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= silk_LSHIFT32( silk_inner_prod_aligned( x_ptr, x_ptr, D, arch), -rshifts); + } + } + /* Approximate residual energy */ + *res_nrg = silk_LSHIFT( silk_SMMUL( invGain_Q30, C0 ), 2 ); + *res_nrg_Q = -rshifts; + } else { + /* Return residual energy */ + nrg = CAf[ 0 ]; /* Q( -rshifts ) */ + tmp1 = (opus_int32)1 << 16; /* Q16 */ + for( k = 0; k < D; k++ ) { + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); /* Q16 */ + nrg = silk_SMLAWW( nrg, CAf[ k + 1 ], Atmp1 ); /* Q( -rshifts ) */ + tmp1 = silk_SMLAWW( tmp1, Atmp1, Atmp1 ); /* Q16 */ + A_Q16[ k ] = -Atmp1; + } + *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ), -tmp1 );/* Q( -rshifts ) */ + *res_nrg_Q = -rshifts; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX_sse4_1.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX_sse4_1.c new file mode 100644 index 000000000..c13695ea0 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/burg_modified_FIX_sse4_1.c @@ -0,0 +1,400 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include + +#include "SigProc_FIX.h" +#include "define.h" +#include "tuning_parameters.h" +#include "pitch.h" +#include "celt/x86/x86cpu.h" + +#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */ + +#define QA 25 +#define N_BITS_HEAD_ROOM 3 +#define MIN_RSHIFTS -16 +#define MAX_RSHIFTS (32 - QA) + +/* Compute reflection coefficients from input signal */ +void silk_burg_modified_sse4_1( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +) +{ + opus_int k, n, s, lz, rshifts, reached_max_gain; + opus_int32 C0, num, nrg, rc_Q31, invGain_Q30, Atmp_QA, Atmp1, tmp1, tmp2, x1, x2; + const opus_int16 *x_ptr; + opus_int32 C_first_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 C_last_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 Af_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 CAf[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 CAb[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; + opus_int64 C0_64; + + __m128i FIRST_3210, LAST_3210, ATMP_3210, TMP1_3210, TMP2_3210, T1_3210, T2_3210, PTR_3210, SUBFR_3210, X1_3210, X2_3210; + __m128i CONST1 = _mm_set1_epi32(1); + + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + + /* Compute autocorrelations, added over subframes */ + C0_64 = silk_inner_prod16( x, x, subfr_length*nb_subfr, arch ); + lz = silk_CLZ64(C0_64); + rshifts = 32 + 1 + N_BITS_HEAD_ROOM - lz; + if (rshifts > MAX_RSHIFTS) rshifts = MAX_RSHIFTS; + if (rshifts < MIN_RSHIFTS) rshifts = MIN_RSHIFTS; + + if (rshifts > 0) { + C0 = (opus_int32)silk_RSHIFT64(C0_64, rshifts ); + } else { + C0 = silk_LSHIFT32((opus_int32)C0_64, -rshifts ); + } + + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += (opus_int32)silk_RSHIFT64( + silk_inner_prod16( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + int i; + opus_int32 d; + x_ptr = x + s * subfr_length; + celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch ); + for( n = 1; n < D + 1; n++ ) { + for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ ) + d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] ); + xcorr[ n - 1 ] += d; + } + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_LSHIFT32( xcorr[ n - 1 ], -rshifts ); + } + } + } + silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + + /* Initialize */ + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + + invGain_Q30 = (opus_int32)1 << 30; + reached_max_gain = 0; + for( n = 0; n < D; n++ ) { + /* Update first row of correlation matrix (without first element) */ + /* Update last row of correlation matrix (without last element, stored in reversed order) */ + /* Update C * Af */ + /* Update C * flipud(Af) (stored in reversed order) */ + if( rshifts > -2 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], 16 - rshifts ); /* Q(16-rshifts) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 16 - rshifts ); /* Q(16-rshifts) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], QA - 16 ); /* Q(QA-16) */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], QA - 16 ); /* Q(QA-16) */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_SMLAWB( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_SMLAWB( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp_QA = Af_QA[ k ]; + tmp1 = silk_SMLAWB( tmp1, Atmp_QA, x_ptr[ n - k - 1 ] ); /* Q(QA-16) */ + tmp2 = silk_SMLAWB( tmp2, Atmp_QA, x_ptr[ subfr_length - n + k ] ); /* Q(QA-16) */ + } + tmp1 = silk_LSHIFT32( -tmp1, 32 - QA - rshifts ); /* Q(16-rshifts) */ + tmp2 = silk_LSHIFT32( -tmp2, 32 - QA - rshifts ); /* Q(16-rshifts) */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWB( CAf[ k ], tmp1, x_ptr[ n - k ] ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWB( CAb[ k ], tmp2, x_ptr[ subfr_length - n + k - 1 ] ); /* Q( -rshift ) */ + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], -rshifts ); /* Q( -rshifts ) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], -rshifts ); /* Q( -rshifts ) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], 17 ); /* Q17 */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 17 ); /* Q17 */ + + X1_3210 = _mm_set1_epi32( x1 ); + X2_3210 = _mm_set1_epi32( x2 ); + TMP1_3210 = _mm_setzero_si128(); + TMP2_3210 = _mm_setzero_si128(); + for( k = 0; k < n - 3; k += 4 ) { + PTR_3210 = OP_CVTEPI16_EPI32_M64( &x_ptr[ n - k - 1 - 3 ] ); + SUBFR_3210 = OP_CVTEPI16_EPI32_M64( &x_ptr[ subfr_length - n + k ] ); + FIRST_3210 = _mm_loadu_si128( (__m128i *)(void*)&C_first_row[ k ] ); + PTR_3210 = _mm_shuffle_epi32( PTR_3210, _MM_SHUFFLE( 0, 1, 2, 3 ) ); + LAST_3210 = _mm_loadu_si128( (__m128i *)(void*)&C_last_row[ k ] ); + ATMP_3210 = _mm_loadu_si128( (__m128i *)(void*)&Af_QA[ k ] ); + + T1_3210 = _mm_mullo_epi32( PTR_3210, X1_3210 ); + T2_3210 = _mm_mullo_epi32( SUBFR_3210, X2_3210 ); + + ATMP_3210 = _mm_srai_epi32( ATMP_3210, 7 ); + ATMP_3210 = _mm_add_epi32( ATMP_3210, CONST1 ); + ATMP_3210 = _mm_srai_epi32( ATMP_3210, 1 ); + + FIRST_3210 = _mm_add_epi32( FIRST_3210, T1_3210 ); + LAST_3210 = _mm_add_epi32( LAST_3210, T2_3210 ); + + PTR_3210 = _mm_mullo_epi32( ATMP_3210, PTR_3210 ); + SUBFR_3210 = _mm_mullo_epi32( ATMP_3210, SUBFR_3210 ); + + _mm_storeu_si128( (__m128i *)(void*)&C_first_row[ k ], FIRST_3210 ); + _mm_storeu_si128( (__m128i *)(void*)&C_last_row[ k ], LAST_3210 ); + + TMP1_3210 = _mm_add_epi32( TMP1_3210, PTR_3210 ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, SUBFR_3210 ); + } + + TMP1_3210 = _mm_add_epi32( TMP1_3210, _mm_unpackhi_epi64(TMP1_3210, TMP1_3210 ) ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, _mm_unpackhi_epi64(TMP2_3210, TMP2_3210 ) ); + TMP1_3210 = _mm_add_epi32( TMP1_3210, _mm_shufflelo_epi16(TMP1_3210, 0x0E ) ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, _mm_shufflelo_epi16(TMP2_3210, 0x0E ) ); + + tmp1 += _mm_cvtsi128_si32( TMP1_3210 ); + tmp2 += _mm_cvtsi128_si32( TMP2_3210 ); + + for( ; k < n; k++ ) { + C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ + /* We sometimes get overflows in the multiplications (even beyond +/- 2^32), + but they cancel each other and the real result seems to always fit in a 32-bit + signed integer. This was determined experimentally, not theoretically (unfortunately). */ + tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ + tmp2 = silk_MLA_ovflw( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ + } + + tmp1 = -tmp1; /* Q17 */ + tmp2 = -tmp2; /* Q17 */ + + { + __m128i xmm_tmp1, xmm_tmp2; + __m128i xmm_x_ptr_n_k_x2x0, xmm_x_ptr_n_k_x3x1; + __m128i xmm_x_ptr_sub_x2x0, xmm_x_ptr_sub_x3x1; + + xmm_tmp1 = _mm_set1_epi32( tmp1 ); + xmm_tmp2 = _mm_set1_epi32( tmp2 ); + + for( k = 0; k <= n - 3; k += 4 ) { + xmm_x_ptr_n_k_x2x0 = OP_CVTEPI16_EPI32_M64( &x_ptr[ n - k - 3 ] ); + xmm_x_ptr_sub_x2x0 = OP_CVTEPI16_EPI32_M64( &x_ptr[ subfr_length - n + k - 1 ] ); + + xmm_x_ptr_n_k_x2x0 = _mm_shuffle_epi32( xmm_x_ptr_n_k_x2x0, _MM_SHUFFLE( 0, 1, 2, 3 ) ); + + xmm_x_ptr_n_k_x2x0 = _mm_slli_epi32( xmm_x_ptr_n_k_x2x0, -rshifts - 1 ); + xmm_x_ptr_sub_x2x0 = _mm_slli_epi32( xmm_x_ptr_sub_x2x0, -rshifts - 1 ); + + /* equal shift right 4 bytes, xmm_x_ptr_n_k_x3x1 = _mm_srli_si128(xmm_x_ptr_n_k_x2x0, 4)*/ + xmm_x_ptr_n_k_x3x1 = _mm_shuffle_epi32( xmm_x_ptr_n_k_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_x_ptr_sub_x3x1 = _mm_shuffle_epi32( xmm_x_ptr_sub_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_x_ptr_n_k_x2x0 = _mm_mul_epi32( xmm_x_ptr_n_k_x2x0, xmm_tmp1 ); + xmm_x_ptr_n_k_x3x1 = _mm_mul_epi32( xmm_x_ptr_n_k_x3x1, xmm_tmp1 ); + xmm_x_ptr_sub_x2x0 = _mm_mul_epi32( xmm_x_ptr_sub_x2x0, xmm_tmp2 ); + xmm_x_ptr_sub_x3x1 = _mm_mul_epi32( xmm_x_ptr_sub_x3x1, xmm_tmp2 ); + + xmm_x_ptr_n_k_x2x0 = _mm_srli_epi64( xmm_x_ptr_n_k_x2x0, 16 ); + xmm_x_ptr_n_k_x3x1 = _mm_slli_epi64( xmm_x_ptr_n_k_x3x1, 16 ); + xmm_x_ptr_sub_x2x0 = _mm_srli_epi64( xmm_x_ptr_sub_x2x0, 16 ); + xmm_x_ptr_sub_x3x1 = _mm_slli_epi64( xmm_x_ptr_sub_x3x1, 16 ); + + xmm_x_ptr_n_k_x2x0 = _mm_blend_epi16( xmm_x_ptr_n_k_x2x0, xmm_x_ptr_n_k_x3x1, 0xCC ); + xmm_x_ptr_sub_x2x0 = _mm_blend_epi16( xmm_x_ptr_sub_x2x0, xmm_x_ptr_sub_x3x1, 0xCC ); + + X1_3210 = _mm_loadu_si128( (__m128i *)(void*)&CAf[ k ] ); + PTR_3210 = _mm_loadu_si128( (__m128i *)(void*)&CAb[ k ] ); + + X1_3210 = _mm_add_epi32( X1_3210, xmm_x_ptr_n_k_x2x0 ); + PTR_3210 = _mm_add_epi32( PTR_3210, xmm_x_ptr_sub_x2x0 ); + + _mm_storeu_si128( (__m128i *)(void*)&CAf[ k ], X1_3210 ); + _mm_storeu_si128( (__m128i *)(void*)&CAb[ k ], PTR_3210 ); + } + + for( ; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWW( CAf[ k ], tmp1, + silk_LSHIFT32( (opus_int32)x_ptr[ n - k ], -rshifts - 1 ) ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWW( CAb[ k ], tmp2, + silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n + k - 1 ], -rshifts - 1 ) ); /* Q( -rshift ) */ + } + } + } + } + + /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */ + tmp1 = C_first_row[ n ]; /* Q( -rshifts ) */ + tmp2 = C_last_row[ n ]; /* Q( -rshifts ) */ + num = 0; /* Q( -rshifts ) */ + nrg = silk_ADD32( CAb[ 0 ], CAf[ 0 ] ); /* Q( 1-rshifts ) */ + for( k = 0; k < n; k++ ) { + Atmp_QA = Af_QA[ k ]; + lz = silk_CLZ32( silk_abs( Atmp_QA ) ) - 1; + lz = silk_min( 32 - QA, lz ); + Atmp1 = silk_LSHIFT32( Atmp_QA, lz ); /* Q( QA + lz ) */ + + tmp1 = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( C_last_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + tmp2 = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( C_first_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + num = silk_ADD_LSHIFT32( num, silk_SMMUL( CAb[ n - k ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + nrg = silk_ADD_LSHIFT32( nrg, silk_SMMUL( silk_ADD32( CAb[ k + 1 ], CAf[ k + 1 ] ), + Atmp1 ), 32 - QA - lz ); /* Q( 1-rshifts ) */ + } + CAf[ n + 1 ] = tmp1; /* Q( -rshifts ) */ + CAb[ n + 1 ] = tmp2; /* Q( -rshifts ) */ + num = silk_ADD32( num, tmp2 ); /* Q( -rshifts ) */ + num = silk_LSHIFT32( -num, 1 ); /* Q( 1-rshifts ) */ + + /* Calculate the next order reflection (parcor) coefficient */ + if( silk_abs( num ) < nrg ) { + rc_Q31 = silk_DIV32_varQ( num, nrg, 31 ); + } else { + rc_Q31 = ( num > 0 ) ? silk_int32_MAX : silk_int32_MIN; + } + + /* Update inverse prediction gain */ + tmp1 = ( (opus_int32)1 << 30 ) - silk_SMMUL( rc_Q31, rc_Q31 ); + tmp1 = silk_LSHIFT( silk_SMMUL( invGain_Q30, tmp1 ), 2 ); + if( tmp1 <= minInvGain_Q30 ) { + /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ + tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ + rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } + } + invGain_Q30 = minInvGain_Q30; + reached_max_gain = 1; + } else { + invGain_Q30 = tmp1; + } + + /* Update the AR coefficients */ + for( k = 0; k < (n + 1) >> 1; k++ ) { + tmp1 = Af_QA[ k ]; /* QA */ + tmp2 = Af_QA[ n - k - 1 ]; /* QA */ + Af_QA[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* QA */ + Af_QA[ n - k - 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* QA */ + } + Af_QA[ n ] = silk_RSHIFT32( rc_Q31, 31 - QA ); /* QA */ + + if( reached_max_gain ) { + /* Reached max prediction gain; set remaining coefficients to zero and exit loop */ + for( k = n + 1; k < D; k++ ) { + Af_QA[ k ] = 0; + } + break; + } + + /* Update C * Af and C * Ab */ + for( k = 0; k <= n + 1; k++ ) { + tmp1 = CAf[ k ]; /* Q( -rshifts ) */ + tmp2 = CAb[ n - k + 1 ]; /* Q( -rshifts ) */ + CAf[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + CAb[ n - k + 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + } + } + + if( reached_max_gain ) { + for( k = 0; k < D; k++ ) { + /* Scale coefficients */ + A_Q16[ k ] = -silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); + } + /* Subtract energy of preceding samples from C0 */ + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16( x_ptr, x_ptr, D, arch ), rshifts ); + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= silk_LSHIFT32( silk_inner_prod_aligned( x_ptr, x_ptr, D, arch ), -rshifts ); + } + } + /* Approximate residual energy */ + *res_nrg = silk_LSHIFT( silk_SMMUL( invGain_Q30, C0 ), 2 ); + *res_nrg_Q = -rshifts; + } else { + /* Return residual energy */ + nrg = CAf[ 0 ]; /* Q( -rshifts ) */ + tmp1 = (opus_int32)1 << 16; /* Q16 */ + for( k = 0; k < D; k++ ) { + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); /* Q16 */ + nrg = silk_SMLAWW( nrg, CAf[ k + 1 ], Atmp1 ); /* Q( -rshifts ) */ + tmp1 = silk_SMLAWW( tmp1, Atmp1, Atmp1 ); /* Q16 */ + A_Q16[ k ] = -Atmp1; + } + *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ), -tmp1 );/* Q( -rshifts ) */ + *res_nrg_Q = -rshifts; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int32 res_nrg_c = 0; + opus_int res_nrg_Q_c = 0; + opus_int32 A_Q16_c[ MAX_LPC_ORDER ] = {0}; + + silk_burg_modified_c( + &res_nrg_c, + &res_nrg_Q_c, + A_Q16_c, + x, + minInvGain_Q30, + subfr_length, + nb_subfr, + D, + 0 + ); + + silk_assert( *res_nrg == res_nrg_c ); + silk_assert( *res_nrg_Q == res_nrg_Q_c ); + silk_assert( !memcmp( A_Q16, A_Q16_c, D * sizeof( *A_Q16 ) ) ); + } +#endif +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/corrMatrix_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/corrMatrix_FIX.c new file mode 100644 index 000000000..3b194b107 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/corrMatrix_FIX.c @@ -0,0 +1,149 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/********************************************************************** + * Correlation Matrix Computations for LS estimate. + **********************************************************************/ + +#include "main_FIX.h" + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int16 *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *Xt, /* O Pointer to X'*t correlation vector [order] */ + const opus_int rshifts, /* I Right shifts of correlations */ + int arch /* I Run-time architecture */ +) +{ + opus_int lag, i; + const opus_int16 *ptr1, *ptr2; + opus_int32 inner_prod; + + ptr1 = &x[ order - 1 ]; /* Points to first sample of column 0 of X: X[:,0] */ + ptr2 = t; + /* Calculate X'*t */ + if( rshifts > 0 ) { + /* Right shifting used */ + for( lag = 0; lag < order; lag++ ) { + inner_prod = 0; + for( i = 0; i < L; i++ ) { + inner_prod = silk_ADD_RSHIFT32( inner_prod, silk_SMULBB( ptr1[ i ], ptr2[i] ), rshifts ); + } + Xt[ lag ] = inner_prod; /* X[:,lag]'*t */ + ptr1--; /* Go to next column of X */ + } + } else { + silk_assert( rshifts == 0 ); + for( lag = 0; lag < order; lag++ ) { + Xt[ lag ] = silk_inner_prod_aligned( ptr1, ptr2, L, arch ); /* X[:,lag]'*t */ + ptr1--; /* Go to next column of X */ + } + } +} + +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *XX, /* O Pointer to X'*X correlation matrix [ order x order ] */ + opus_int32 *nrg, /* O Energy of x vector */ + opus_int *rshifts, /* O Right shifts of correlations and energy */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, j, lag; + opus_int32 energy; + const opus_int16 *ptr1, *ptr2; + + /* Calculate energy to find shift used to fit in 32 bits */ + silk_sum_sqr_shift( nrg, rshifts, x, L + order - 1 ); + energy = *nrg; + + /* Calculate energy of first column (0) of X: X[:,0]'*X[:,0] */ + /* Remove contribution of first order - 1 samples */ + for( i = 0; i < order - 1; i++ ) { + energy -= silk_RSHIFT32( silk_SMULBB( x[ i ], x[ i ] ), *rshifts ); + } + + /* Calculate energy of remaining columns of X: X[:,j]'*X[:,j] */ + /* Fill out the diagonal of the correlation matrix */ + matrix_ptr( XX, 0, 0, order ) = energy; + silk_assert( energy >= 0 ); + ptr1 = &x[ order - 1 ]; /* First sample of column 0 of X */ + for( j = 1; j < order; j++ ) { + energy = silk_SUB32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ L - j ], ptr1[ L - j ] ), *rshifts ) ); + energy = silk_ADD32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ -j ], ptr1[ -j ] ), *rshifts ) ); + matrix_ptr( XX, j, j, order ) = energy; + silk_assert( energy >= 0 ); + } + + ptr2 = &x[ order - 2 ]; /* First sample of column 1 of X */ + /* Calculate the remaining elements of the correlation matrix */ + if( *rshifts > 0 ) { + /* Right shifting used */ + for( lag = 1; lag < order; lag++ ) { + /* Inner product of column 0 and column lag: X[:,0]'*X[:,lag] */ + energy = 0; + for( i = 0; i < L; i++ ) { + energy += silk_RSHIFT32( silk_SMULBB( ptr1[ i ], ptr2[i] ), *rshifts ); + } + /* Calculate remaining off diagonal: X[:,j]'*X[:,j + lag] */ + matrix_ptr( XX, lag, 0, order ) = energy; + matrix_ptr( XX, 0, lag, order ) = energy; + for( j = 1; j < ( order - lag ); j++ ) { + energy = silk_SUB32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ L - j ], ptr2[ L - j ] ), *rshifts ) ); + energy = silk_ADD32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ -j ], ptr2[ -j ] ), *rshifts ) ); + matrix_ptr( XX, lag + j, j, order ) = energy; + matrix_ptr( XX, j, lag + j, order ) = energy; + } + ptr2--; /* Update pointer to first sample of next column (lag) in X */ + } + } else { + for( lag = 1; lag < order; lag++ ) { + /* Inner product of column 0 and column lag: X[:,0]'*X[:,lag] */ + energy = silk_inner_prod_aligned( ptr1, ptr2, L, arch ); + matrix_ptr( XX, lag, 0, order ) = energy; + matrix_ptr( XX, 0, lag, order ) = energy; + /* Calculate remaining off diagonal: X[:,j]'*X[:,j + lag] */ + for( j = 1; j < ( order - lag ); j++ ) { + energy = silk_SUB32( energy, silk_SMULBB( ptr1[ L - j ], ptr2[ L - j ] ) ); + energy = silk_SMLABB( energy, ptr1[ -j ], ptr2[ -j ] ); + matrix_ptr( XX, lag + j, j, order ) = energy; + matrix_ptr( XX, j, lag + j, order ) = energy; + } + ptr2--;/* Update pointer to first sample of next column (lag) in X */ + } + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/encode_frame_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/encode_frame_FIX.c new file mode 100644 index 000000000..c2f5281f0 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/encode_frame_FIX.c @@ -0,0 +1,454 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with lower bitrate */ +static OPUS_INLINE void silk_LBRR_encode_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Pointer to Silk FIX encoder control struct */ + const opus_int16 x16[], /* I Input signal */ + opus_int condCoding /* I The type of conditional coding used so far for this frame */ +); + +void silk_encode_do_VAD_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int activity /* I Decision of Opus voice activity detector */ +) +{ + const opus_int activity_threshold = SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ); + + /****************************/ + /* Voice Activity Detection */ + /****************************/ + silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch ); + /* If Opus VAD is inactive and Silk VAD is active: lower Silk VAD to just under the threshold */ + if( activity == VAD_NO_ACTIVITY && psEnc->sCmn.speech_activity_Q8 >= activity_threshold ) { + psEnc->sCmn.speech_activity_Q8 = activity_threshold - 1; + } + + /**************************************************/ + /* Convert speech activity into VAD and DTX flags */ + /**************************************************/ + if( psEnc->sCmn.speech_activity_Q8 < activity_threshold ) { + psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; + psEnc->sCmn.noSpeechCounter++; + if( psEnc->sCmn.noSpeechCounter <= NB_SPEECH_FRAMES_BEFORE_DTX ) { + psEnc->sCmn.inDTX = 0; + } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) { + psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; + psEnc->sCmn.inDTX = 0; + } + psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 0; + } else { + psEnc->sCmn.noSpeechCounter = 0; + psEnc->sCmn.inDTX = 0; + psEnc->sCmn.indices.signalType = TYPE_UNVOICED; + psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 1; + } +} + +/****************/ +/* Encode frame */ +/****************/ +opus_int silk_encode_frame_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int32 *pnBytesOut, /* O Pointer to number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +) +{ + silk_encoder_control_FIX sEncCtrl; + opus_int i, iter, maxIter, found_upper, found_lower, ret = 0; + opus_int16 *x_frame; + ec_enc sRangeEnc_copy, sRangeEnc_copy2; + VARDECL(silk_nsq_state, sNSQ_copy); + opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gainMult_upper; + opus_int32 gainsID, gainsID_lower, gainsID_upper; + opus_int16 gainMult_Q8; + opus_int16 ec_prevLagIndex_copy; + opus_int ec_prevSignalType_copy; + opus_int8 LastGainIndex_copy2; + opus_int gain_lock[ MAX_NB_SUBFR ] = {0}; + opus_int16 best_gain_mult[ MAX_NB_SUBFR ]; + opus_int best_sum[ MAX_NB_SUBFR ]; + opus_int bits_margin; + SAVE_STACK; + + /* Using ALLOC() instead of a regular stack allocation to minimize real stack use when using the pseudostack. + This is useful on some embedded systems. */ + ALLOC(sNSQ_copy, 2, silk_nsq_state); + + /* For CBR, 5 bits below budget is close enough. For VBR, allow up to 25% below the cap if we initially busted the budget. */ + bits_margin = useCBR ? 5 : maxBits/4; + /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ + LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; + + psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3; + + /**************************************************************/ + /* Set up Input Pointers, and insert frame in input buffer */ + /*************************************************************/ + /* start of frame to encode */ + x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; + + /***************************************/ + /* Ensure smooth bandwidth transitions */ + /***************************************/ + silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length ); + + /*******************************************/ + /* Copy new frame to front of input buffer */ + /*******************************************/ + silk_memcpy( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length * sizeof( opus_int16 ) ); + + if( !psEnc->sCmn.prefillFlag ) { + VARDECL( opus_int16, res_pitch ); + VARDECL( opus_uint8, ec_buf_copy ); + opus_int16 *res_pitch_frame; + + ALLOC( res_pitch, + psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + + psEnc->sCmn.ltp_mem_length, opus_int16 ); + /* start of pitch LPC residual frame */ + res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; + + /*****************************************/ + /* Find pitch lags, initial LPC analysis */ + /*****************************************/ + silk_find_pitch_lags_FIX( psEnc, &sEncCtrl, res_pitch, x_frame - psEnc->sCmn.ltp_mem_length, psEnc->sCmn.arch ); + + /************************/ + /* Noise shape analysis */ + /************************/ + silk_noise_shape_analysis_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame, psEnc->sCmn.arch ); + + /***************************************************/ + /* Find linear prediction coefficients (LPC + LTP) */ + /***************************************************/ + silk_find_pred_coefs_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame, condCoding ); + + /****************************************/ + /* Process gains */ + /****************************************/ + silk_process_gains_FIX( psEnc, &sEncCtrl, condCoding ); + + /****************************************/ + /* Low Bitrate Redundant Encoding */ + /****************************************/ + silk_LBRR_encode_FIX( psEnc, &sEncCtrl, x_frame, condCoding ); + + /* Loop over quantizer and entropy coding to control bitrate */ + maxIter = 6; + gainMult_Q8 = SILK_FIX_CONST( 1, 8 ); + found_lower = 0; + found_upper = 0; + gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); + gainsID_lower = -1; + gainsID_upper = -1; + /* Copy part of the input state */ + silk_memcpy( &sRangeEnc_copy, psRangeEnc, sizeof( ec_enc ) ); + silk_memcpy( &sNSQ_copy[0], &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + seed_copy = psEnc->sCmn.indices.Seed; + ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex; + ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType; + ALLOC( ec_buf_copy, 1275, opus_uint8 ); + for( iter = 0; ; iter++ ) { + if( gainsID == gainsID_lower ) { + nBits = nBits_lower; + } else if( gainsID == gainsID_upper ) { + nBits = nBits_upper; + } else { + /* Restore part of the input state */ + if( iter > 0 ) { + silk_memcpy( psRangeEnc, &sRangeEnc_copy, sizeof( ec_enc ) ); + silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy[0], sizeof( silk_nsq_state ) ); + psEnc->sCmn.indices.Seed = seed_copy; + psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; + psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; + } + + /*****************************************/ + /* Noise shaping quantization */ + /*****************************************/ + if( psEnc->sCmn.nStatesDelayedDecision > 1 || psEnc->sCmn.warping_Q16 > 0 ) { + silk_NSQ_del_dec( &psEnc->sCmn, &psEnc->sCmn.sNSQ, &psEnc->sCmn.indices, x_frame, psEnc->sCmn.pulses, + sEncCtrl.PredCoef_Q12[ 0 ], sEncCtrl.LTPCoef_Q14, sEncCtrl.AR_Q13, sEncCtrl.HarmShapeGain_Q14, + sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14, + psEnc->sCmn.arch ); + } else { + silk_NSQ( &psEnc->sCmn, &psEnc->sCmn.sNSQ, &psEnc->sCmn.indices, x_frame, psEnc->sCmn.pulses, + sEncCtrl.PredCoef_Q12[ 0 ], sEncCtrl.LTPCoef_Q14, sEncCtrl.AR_Q13, sEncCtrl.HarmShapeGain_Q14, + sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14, + psEnc->sCmn.arch); + } + + if ( iter == maxIter && !found_lower ) { + silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); + } + + /****************************************/ + /* Encode Parameters */ + /****************************************/ + silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); + + /****************************************/ + /* Encode Excitation Signal */ + /****************************************/ + silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, + psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); + + nBits = ec_tell( psRangeEnc ); + + /* If we still bust after the last iteration, do some damage control. */ + if ( iter == maxIter && !found_lower && nBits > maxBits ) { + silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); + + /* Keep gains the same as the last frame. */ + psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; + for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + psEnc->sCmn.indices.GainsIndices[ i ] = 4; + } + if (condCoding != CODE_CONDITIONALLY) { + psEnc->sCmn.indices.GainsIndices[ 0 ] = sEncCtrl.lastGainIndexPrev; + } + psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; + psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; + /* Clear all pulses. */ + for ( i = 0; i < psEnc->sCmn.frame_length; i++ ) { + psEnc->sCmn.pulses[ i ] = 0; + } + + silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); + + silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, + psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); + + nBits = ec_tell( psRangeEnc ); + } + + if( useCBR == 0 && iter == 0 && nBits <= maxBits ) { + break; + } + } + + if( iter == maxIter ) { + if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) { + /* Restore output state from earlier iteration that did meet the bitrate budget */ + silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); + celt_assert( sRangeEnc_copy2.offs <= 1275 ); + silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs ); + silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy[1], sizeof( silk_nsq_state ) ); + psEnc->sShape.LastGainIndex = LastGainIndex_copy2; + } + break; + } + + if( nBits > maxBits ) { + if( found_lower == 0 && iter >= 2 ) { + /* Adjust the quantizer's rate/distortion tradeoff and discard previous "upper" results */ + sEncCtrl.Lambda_Q10 = silk_ADD_RSHIFT32( sEncCtrl.Lambda_Q10, sEncCtrl.Lambda_Q10, 1 ); + found_upper = 0; + gainsID_upper = -1; + } else { + found_upper = 1; + nBits_upper = nBits; + gainMult_upper = gainMult_Q8; + gainsID_upper = gainsID; + } + } else if( nBits < maxBits - bits_margin ) { + found_lower = 1; + nBits_lower = nBits; + gainMult_lower = gainMult_Q8; + if( gainsID != gainsID_lower ) { + gainsID_lower = gainsID; + /* Copy part of the output state */ + silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); + celt_assert( psRangeEnc->offs <= 1275 ); + silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs ); + silk_memcpy( &sNSQ_copy[1], &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; + } + } else { + /* Close enough */ + break; + } + + if ( !found_lower && nBits > maxBits ) { + int j; + for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + int sum=0; + for ( j = i*psEnc->sCmn.subfr_length; j < (i+1)*psEnc->sCmn.subfr_length; j++ ) { + sum += abs( psEnc->sCmn.pulses[j] ); + } + if ( iter == 0 || (sum < best_sum[i] && !gain_lock[i]) ) { + best_sum[i] = sum; + best_gain_mult[i] = gainMult_Q8; + } else { + gain_lock[i] = 1; + } + } + } + if( ( found_lower & found_upper ) == 0 ) { + /* Adjust gain according to high-rate rate/distortion curve */ + if( nBits > maxBits ) { + gainMult_Q8 = silk_min_32( 1024, gainMult_Q8*3/2 ); + } else { + gainMult_Q8 = silk_max_32( 64, gainMult_Q8*4/5 ); + } + } else { + /* Adjust gain by interpolating */ + gainMult_Q8 = gainMult_lower + silk_DIV32_16( silk_MUL( gainMult_upper - gainMult_lower, maxBits - nBits_lower ), nBits_upper - nBits_lower ); + /* New gain multiplier must be between 25% and 75% of old range (note that gainMult_upper < gainMult_lower) */ + if( gainMult_Q8 > silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ) ) { + gainMult_Q8 = silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ); + } else + if( gainMult_Q8 < silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ) ) { + gainMult_Q8 = silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ); + } + } + + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + opus_int16 tmp; + if ( gain_lock[i] ) { + tmp = best_gain_mult[i]; + } else { + tmp = gainMult_Q8; + } + sEncCtrl.Gains_Q16[ i ] = silk_LSHIFT_SAT32( silk_SMULWB( sEncCtrl.GainsUnq_Q16[ i ], tmp ), 8 ); + } + + /* Quantize gains */ + psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; + silk_gains_quant( psEnc->sCmn.indices.GainsIndices, sEncCtrl.Gains_Q16, + &psEnc->sShape.LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Unique identifier of gains vector */ + gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); + } + } + + /* Update input buffer */ + silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], + ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( opus_int16 ) ); + + /* Exit without entropy coding */ + if( psEnc->sCmn.prefillFlag ) { + /* No payload */ + *pnBytesOut = 0; + RESTORE_STACK; + return ret; + } + + /* Parameters needed for next frame */ + psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; + psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; + + /****************************************/ + /* Finalize payload */ + /****************************************/ + psEnc->sCmn.first_frame_after_reset = 0; + /* Payload size */ + *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 ); + + RESTORE_STACK; + return ret; +} + +/* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate */ +static OPUS_INLINE void silk_LBRR_encode_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Pointer to Silk FIX encoder control struct */ + const opus_int16 x16[], /* I Input signal */ + opus_int condCoding /* I The type of conditional coding used so far for this frame */ +) +{ + opus_int32 TempGains_Q16[ MAX_NB_SUBFR ]; + SideInfoIndices *psIndices_LBRR = &psEnc->sCmn.indices_LBRR[ psEnc->sCmn.nFramesEncoded ]; + VARDECL(silk_nsq_state, sNSQ_LBRR); + SAVE_STACK; + + /*******************************************/ + /* Control use of inband LBRR */ + /*******************************************/ + if( psEnc->sCmn.LBRR_enabled && psEnc->sCmn.speech_activity_Q8 > SILK_FIX_CONST( LBRR_SPEECH_ACTIVITY_THRES, 8 ) ) { + /* Using ALLOC() instead of a regular stack allocation to minimize real stack use when using the pseudostack. + This is useful on some embedded systems. */ + ALLOC(sNSQ_LBRR, 1, silk_nsq_state); + + psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded ] = 1; + + /* Copy noise shaping quantizer state and quantization indices from regular encoding */ + silk_memcpy( &sNSQ_LBRR[0], &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + silk_memcpy( psIndices_LBRR, &psEnc->sCmn.indices, sizeof( SideInfoIndices ) ); + + /* Save original gains */ + silk_memcpy( TempGains_Q16, psEncCtrl->Gains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + + if( psEnc->sCmn.nFramesEncoded == 0 || psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded - 1 ] == 0 ) { + /* First frame in packet or previous frame not LBRR coded */ + psEnc->sCmn.LBRRprevLastGainIndex = psEnc->sShape.LastGainIndex; + + /* Increase Gains to get target LBRR rate */ + psIndices_LBRR->GainsIndices[ 0 ] = psIndices_LBRR->GainsIndices[ 0 ] + psEnc->sCmn.LBRR_GainIncreases; + psIndices_LBRR->GainsIndices[ 0 ] = silk_min_int( psIndices_LBRR->GainsIndices[ 0 ], N_LEVELS_QGAIN - 1 ); + } + + /* Decode to get gains in sync with decoder */ + /* Overwrite unquantized gains with quantized gains */ + silk_gains_dequant( psEncCtrl->Gains_Q16, psIndices_LBRR->GainsIndices, + &psEnc->sCmn.LBRRprevLastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /*****************************************/ + /* Noise shaping quantization */ + /*****************************************/ + if( psEnc->sCmn.nStatesDelayedDecision > 1 || psEnc->sCmn.warping_Q16 > 0 ) { + silk_NSQ_del_dec( &psEnc->sCmn, &sNSQ_LBRR[0], psIndices_LBRR, x16, + psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], psEncCtrl->PredCoef_Q12[ 0 ], psEncCtrl->LTPCoef_Q14, + psEncCtrl->AR_Q13, psEncCtrl->HarmShapeGain_Q14, psEncCtrl->Tilt_Q14, psEncCtrl->LF_shp_Q14, + psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14, psEnc->sCmn.arch ); + } else { + silk_NSQ( &psEnc->sCmn, &sNSQ_LBRR[0], psIndices_LBRR, x16, + psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], psEncCtrl->PredCoef_Q12[ 0 ], psEncCtrl->LTPCoef_Q14, + psEncCtrl->AR_Q13, psEncCtrl->HarmShapeGain_Q14, psEncCtrl->Tilt_Q14, psEncCtrl->LF_shp_Q14, + psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14, psEnc->sCmn.arch ); + } + + /* Restore original gains */ + silk_memcpy( psEncCtrl->Gains_Q16, TempGains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + } + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LPC_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LPC_FIX.c new file mode 100644 index 000000000..c762a0f2a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LPC_FIX.c @@ -0,0 +1,151 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Finds LPC vector from correlations, and converts to NLSF */ +void silk_find_LPC_FIX( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const opus_int16 x[], /* I Input signal */ + const opus_int32 minInvGain_Q30 /* I Inverse of max prediction gain */ +) +{ + opus_int k, subfr_length; + opus_int32 a_Q16[ MAX_LPC_ORDER ]; + opus_int isInterpLower, shift; + opus_int32 res_nrg0, res_nrg1; + opus_int rshift0, rshift1; + + /* Used only for LSF interpolation */ + opus_int32 a_tmp_Q16[ MAX_LPC_ORDER ], res_nrg_interp, res_nrg, res_tmp_nrg; + opus_int res_nrg_interp_Q, res_nrg_Q, res_tmp_nrg_Q; + opus_int16 a_tmp_Q12[ MAX_LPC_ORDER ]; + opus_int16 NLSF0_Q15[ MAX_LPC_ORDER ]; + SAVE_STACK; + + subfr_length = psEncC->subfr_length + psEncC->predictLPCOrder; + + /* Default: no interpolation */ + psEncC->indices.NLSFInterpCoef_Q2 = 4; + + /* Burg AR analysis for the full frame */ + silk_burg_modified( &res_nrg, &res_nrg_Q, a_Q16, x, minInvGain_Q30, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder, psEncC->arch ); + + if( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) { + VARDECL( opus_int16, LPC_res ); + + /* Optimal solution for last 10 ms */ + silk_burg_modified( &res_tmp_nrg, &res_tmp_nrg_Q, a_tmp_Q16, x + 2 * subfr_length, minInvGain_Q30, subfr_length, 2, psEncC->predictLPCOrder, psEncC->arch ); + + /* subtract residual energy here, as that's easier than adding it to the */ + /* residual energy of the first 10 ms in each iteration of the search below */ + shift = res_tmp_nrg_Q - res_nrg_Q; + if( shift >= 0 ) { + if( shift < 32 ) { + res_nrg = res_nrg - silk_RSHIFT( res_tmp_nrg, shift ); + } + } else { + silk_assert( shift > -32 ); + res_nrg = silk_RSHIFT( res_nrg, -shift ) - res_tmp_nrg; + res_nrg_Q = res_tmp_nrg_Q; + } + + /* Convert to NLSFs */ + silk_A2NLSF( NLSF_Q15, a_tmp_Q16, psEncC->predictLPCOrder ); + + ALLOC( LPC_res, 2 * subfr_length, opus_int16 ); + + /* Search over interpolation indices to find the one with lowest residual energy */ + for( k = 3; k >= 0; k-- ) { + /* Interpolate NLSFs for first half */ + silk_interpolate( NLSF0_Q15, psEncC->prev_NLSFq_Q15, NLSF_Q15, k, psEncC->predictLPCOrder ); + + /* Convert to LPC for residual energy evaluation */ + silk_NLSF2A( a_tmp_Q12, NLSF0_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + /* Calculate residual energy with NLSF interpolation */ + silk_LPC_analysis_filter( LPC_res, x, a_tmp_Q12, 2 * subfr_length, psEncC->predictLPCOrder, psEncC->arch ); + + silk_sum_sqr_shift( &res_nrg0, &rshift0, LPC_res + psEncC->predictLPCOrder, subfr_length - psEncC->predictLPCOrder ); + silk_sum_sqr_shift( &res_nrg1, &rshift1, LPC_res + psEncC->predictLPCOrder + subfr_length, subfr_length - psEncC->predictLPCOrder ); + + /* Add subframe energies from first half frame */ + shift = rshift0 - rshift1; + if( shift >= 0 ) { + res_nrg1 = silk_RSHIFT( res_nrg1, shift ); + res_nrg_interp_Q = -rshift0; + } else { + res_nrg0 = silk_RSHIFT( res_nrg0, -shift ); + res_nrg_interp_Q = -rshift1; + } + res_nrg_interp = silk_ADD32( res_nrg0, res_nrg1 ); + + /* Compare with first half energy without NLSF interpolation, or best interpolated value so far */ + shift = res_nrg_interp_Q - res_nrg_Q; + if( shift >= 0 ) { + if( silk_RSHIFT( res_nrg_interp, shift ) < res_nrg ) { + isInterpLower = silk_TRUE; + } else { + isInterpLower = silk_FALSE; + } + } else { + if( -shift < 32 ) { + if( res_nrg_interp < silk_RSHIFT( res_nrg, -shift ) ) { + isInterpLower = silk_TRUE; + } else { + isInterpLower = silk_FALSE; + } + } else { + isInterpLower = silk_FALSE; + } + } + + /* Determine whether current interpolated NLSFs are best so far */ + if( isInterpLower == silk_TRUE ) { + /* Interpolation has lower residual energy */ + res_nrg = res_nrg_interp; + res_nrg_Q = res_nrg_interp_Q; + psEncC->indices.NLSFInterpCoef_Q2 = (opus_int8)k; + } + } + } + + if( psEncC->indices.NLSFInterpCoef_Q2 == 4 ) { + /* NLSF interpolation is currently inactive, calculate NLSFs from full frame AR coefficients */ + silk_A2NLSF( NLSF_Q15, a_Q16, psEncC->predictLPCOrder ); + } + + celt_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LTP_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LTP_FIX.c new file mode 100644 index 000000000..62d4afb25 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_LTP_FIX.c @@ -0,0 +1,99 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "tuning_parameters.h" + +void silk_find_LTP_FIX( + opus_int32 XXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Correlation matrix */ + opus_int32 xXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER ], /* O Correlation vector */ + const opus_int16 r_ptr[], /* I Residual signal after LPC */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, k, extra_shifts; + opus_int xx_shifts, xX_shifts, XX_shifts; + const opus_int16 *lag_ptr; + opus_int32 *XXLTP_Q17_ptr, *xXLTP_Q17_ptr; + opus_int32 xx, nrg, temp; + + xXLTP_Q17_ptr = xXLTP_Q17; + XXLTP_Q17_ptr = XXLTP_Q17; + for( k = 0; k < nb_subfr; k++ ) { + lag_ptr = r_ptr - ( lag[ k ] + LTP_ORDER / 2 ); + + silk_sum_sqr_shift( &xx, &xx_shifts, r_ptr, subfr_length + LTP_ORDER ); /* xx in Q( -xx_shifts ) */ + silk_corrMatrix_FIX( lag_ptr, subfr_length, LTP_ORDER, XXLTP_Q17_ptr, &nrg, &XX_shifts, arch ); /* XXLTP_Q17_ptr and nrg in Q( -XX_shifts ) */ + extra_shifts = xx_shifts - XX_shifts; + if( extra_shifts > 0 ) { + /* Shift XX */ + xX_shifts = xx_shifts; + for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) { + XXLTP_Q17_ptr[ i ] = silk_RSHIFT32( XXLTP_Q17_ptr[ i ], extra_shifts ); /* Q( -xX_shifts ) */ + } + nrg = silk_RSHIFT32( nrg, extra_shifts ); /* Q( -xX_shifts ) */ + } else if( extra_shifts < 0 ) { + /* Shift xx */ + xX_shifts = XX_shifts; + xx = silk_RSHIFT32( xx, -extra_shifts ); /* Q( -xX_shifts ) */ + } else { + xX_shifts = xx_shifts; + } + silk_corrVector_FIX( lag_ptr, r_ptr, subfr_length, LTP_ORDER, xXLTP_Q17_ptr, xX_shifts, arch ); /* xXLTP_Q17_ptr in Q( -xX_shifts ) */ + + /* At this point all correlations are in Q(-xX_shifts) */ + temp = silk_SMLAWB( 1, nrg, SILK_FIX_CONST( LTP_CORR_INV_MAX, 16 ) ); + temp = silk_max( temp, xx ); +TIC(div) +#if 0 + for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) { + XXLTP_Q17_ptr[ i ] = silk_DIV32_varQ( XXLTP_Q17_ptr[ i ], temp, 17 ); + } + for( i = 0; i < LTP_ORDER; i++ ) { + xXLTP_Q17_ptr[ i ] = silk_DIV32_varQ( xXLTP_Q17_ptr[ i ], temp, 17 ); + } +#else + for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) { + XXLTP_Q17_ptr[ i ] = (opus_int32)( silk_LSHIFT64( (opus_int64)XXLTP_Q17_ptr[ i ], 17 ) / temp ); + } + for( i = 0; i < LTP_ORDER; i++ ) { + xXLTP_Q17_ptr[ i ] = (opus_int32)( silk_LSHIFT64( (opus_int64)xXLTP_Q17_ptr[ i ], 17 ) / temp ); + } +#endif +TOC(div) + r_ptr += subfr_length; + XXLTP_Q17_ptr += LTP_ORDER * LTP_ORDER; + xXLTP_Q17_ptr += LTP_ORDER; + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pitch_lags_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pitch_lags_FIX.c new file mode 100644 index 000000000..6c3379f2b --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pitch_lags_FIX.c @@ -0,0 +1,143 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Find pitch lags */ +void silk_find_pitch_lags_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int16 res[], /* O residual */ + const opus_int16 x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +) +{ + opus_int buf_len, i, scale; + opus_int32 thrhld_Q13, res_nrg; + const opus_int16 *x_ptr; + VARDECL( opus_int16, Wsig ); + opus_int16 *Wsig_ptr; + opus_int32 auto_corr[ MAX_FIND_PITCH_LPC_ORDER + 1 ]; + opus_int16 rc_Q15[ MAX_FIND_PITCH_LPC_ORDER ]; + opus_int32 A_Q24[ MAX_FIND_PITCH_LPC_ORDER ]; + opus_int16 A_Q12[ MAX_FIND_PITCH_LPC_ORDER ]; + SAVE_STACK; + + /******************************************/ + /* Set up buffer lengths etc based on Fs */ + /******************************************/ + buf_len = psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + psEnc->sCmn.ltp_mem_length; + + /* Safety check */ + celt_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); + + /*************************************/ + /* Estimate LPC AR coefficients */ + /*************************************/ + + /* Calculate windowed signal */ + + ALLOC( Wsig, psEnc->sCmn.pitch_LPC_win_length, opus_int16 ); + + /* First LA_LTP samples */ + x_ptr = x + buf_len - psEnc->sCmn.pitch_LPC_win_length; + Wsig_ptr = Wsig; + silk_apply_sine_window( Wsig_ptr, x_ptr, 1, psEnc->sCmn.la_pitch ); + + /* Middle un - windowed samples */ + Wsig_ptr += psEnc->sCmn.la_pitch; + x_ptr += psEnc->sCmn.la_pitch; + silk_memcpy( Wsig_ptr, x_ptr, ( psEnc->sCmn.pitch_LPC_win_length - silk_LSHIFT( psEnc->sCmn.la_pitch, 1 ) ) * sizeof( opus_int16 ) ); + + /* Last LA_LTP samples */ + Wsig_ptr += psEnc->sCmn.pitch_LPC_win_length - silk_LSHIFT( psEnc->sCmn.la_pitch, 1 ); + x_ptr += psEnc->sCmn.pitch_LPC_win_length - silk_LSHIFT( psEnc->sCmn.la_pitch, 1 ); + silk_apply_sine_window( Wsig_ptr, x_ptr, 2, psEnc->sCmn.la_pitch ); + + /* Calculate autocorrelation sequence */ + silk_autocorr( auto_corr, &scale, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1, arch ); + + /* Add white noise, as fraction of energy */ + auto_corr[ 0 ] = silk_SMLAWB( auto_corr[ 0 ], auto_corr[ 0 ], SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ) + 1; + + /* Calculate the reflection coefficients using schur */ + res_nrg = silk_schur( rc_Q15, auto_corr, psEnc->sCmn.pitchEstimationLPCOrder ); + + /* Prediction gain */ + psEncCtrl->predGain_Q16 = silk_DIV32_varQ( auto_corr[ 0 ], silk_max_int( res_nrg, 1 ), 16 ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a( A_Q24, rc_Q15, psEnc->sCmn.pitchEstimationLPCOrder ); + + /* Convert From 32 bit Q24 to 16 bit Q12 coefs */ + for( i = 0; i < psEnc->sCmn.pitchEstimationLPCOrder; i++ ) { + A_Q12[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT( A_Q24[ i ], 12 ) ); + } + + /* Do BWE */ + silk_bwexpander( A_Q12, psEnc->sCmn.pitchEstimationLPCOrder, SILK_FIX_CONST( FIND_PITCH_BANDWIDTH_EXPANSION, 16 ) ); + + /*****************************************/ + /* LPC analysis filtering */ + /*****************************************/ + silk_LPC_analysis_filter( res, x, A_Q12, buf_len, psEnc->sCmn.pitchEstimationLPCOrder, psEnc->sCmn.arch ); + + if( psEnc->sCmn.indices.signalType != TYPE_NO_VOICE_ACTIVITY && psEnc->sCmn.first_frame_after_reset == 0 ) { + /* Threshold for pitch estimator */ + thrhld_Q13 = SILK_FIX_CONST( 0.6, 13 ); + thrhld_Q13 = silk_SMLABB( thrhld_Q13, SILK_FIX_CONST( -0.004, 13 ), psEnc->sCmn.pitchEstimationLPCOrder ); + thrhld_Q13 = silk_SMLAWB( thrhld_Q13, SILK_FIX_CONST( -0.1, 21 ), psEnc->sCmn.speech_activity_Q8 ); + thrhld_Q13 = silk_SMLABB( thrhld_Q13, SILK_FIX_CONST( -0.15, 13 ), silk_RSHIFT( psEnc->sCmn.prevSignalType, 1 ) ); + thrhld_Q13 = silk_SMLAWB( thrhld_Q13, SILK_FIX_CONST( -0.1, 14 ), psEnc->sCmn.input_tilt_Q15 ); + thrhld_Q13 = silk_SAT16( thrhld_Q13 ); + + /*****************************************/ + /* Call pitch estimator */ + /*****************************************/ + if( silk_pitch_analysis_core( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, &psEnc->sCmn.indices.contourIndex, + &psEnc->LTPCorr_Q15, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16, + (opus_int)thrhld_Q13, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, + psEnc->sCmn.arch) == 0 ) + { + psEnc->sCmn.indices.signalType = TYPE_VOICED; + } else { + psEnc->sCmn.indices.signalType = TYPE_UNVOICED; + } + } else { + silk_memset( psEncCtrl->pitchL, 0, sizeof( psEncCtrl->pitchL ) ); + psEnc->sCmn.indices.lagIndex = 0; + psEnc->sCmn.indices.contourIndex = 0; + psEnc->LTPCorr_Q15 = 0; + } + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pred_coefs_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pred_coefs_FIX.c new file mode 100644 index 000000000..c49e44f6e --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/find_pred_coefs_FIX.c @@ -0,0 +1,147 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" + +void silk_find_pred_coefs_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + const opus_int16 res_pitch[], /* I Residual from pitch analysis */ + const opus_int16 x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i; + opus_int32 invGains_Q16[ MAX_NB_SUBFR ], local_gains[ MAX_NB_SUBFR ]; + /* Set to NLSF_Q15 to zero so we don't copy junk to the state. */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]={0}; + const opus_int16 *x_ptr; + opus_int16 *x_pre_ptr; + VARDECL( opus_int16, LPC_in_pre ); + opus_int32 min_gain_Q16, minInvGain_Q30; + SAVE_STACK; + + /* weighting for weighted least squares */ + min_gain_Q16 = silk_int32_MAX >> 6; + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + min_gain_Q16 = silk_min( min_gain_Q16, psEncCtrl->Gains_Q16[ i ] ); + } + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + /* Divide to Q16 */ + silk_assert( psEncCtrl->Gains_Q16[ i ] > 0 ); + /* Invert and normalize gains, and ensure that maximum invGains_Q16 is within range of a 16 bit int */ + invGains_Q16[ i ] = silk_DIV32_varQ( min_gain_Q16, psEncCtrl->Gains_Q16[ i ], 16 - 2 ); + + /* Limit inverse */ + invGains_Q16[ i ] = silk_max( invGains_Q16[ i ], 100 ); + + /* Square the inverted gains */ + silk_assert( invGains_Q16[ i ] == silk_SAT16( invGains_Q16[ i ] ) ); + + /* Invert the inverted and normalized gains */ + local_gains[ i ] = silk_DIV32( ( (opus_int32)1 << 16 ), invGains_Q16[ i ] ); + } + + ALLOC( LPC_in_pre, + psEnc->sCmn.nb_subfr * psEnc->sCmn.predictLPCOrder + + psEnc->sCmn.frame_length, opus_int16 ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + VARDECL( opus_int32, xXLTP_Q17 ); + VARDECL( opus_int32, XXLTP_Q17 ); + + /**********/ + /* VOICED */ + /**********/ + celt_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); + + ALLOC( xXLTP_Q17, psEnc->sCmn.nb_subfr * LTP_ORDER, opus_int32 ); + ALLOC( XXLTP_Q17, psEnc->sCmn.nb_subfr * LTP_ORDER * LTP_ORDER, opus_int32 ); + + /* LTP analysis */ + silk_find_LTP_FIX( XXLTP_Q17, xXLTP_Q17, res_pitch, + psEncCtrl->pitchL, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.arch ); + + /* Quantize LTP gain parameters */ + silk_quant_LTP_gains( psEncCtrl->LTPCoef_Q14, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex, + &psEnc->sCmn.sum_log_gain_Q7, &psEncCtrl->LTPredCodGain_Q7, XXLTP_Q17, xXLTP_Q17, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.arch ); + + /* Control LTP scaling */ + silk_LTP_scale_ctrl_FIX( psEnc, psEncCtrl, condCoding ); + + /* Create LTP residual */ + silk_LTP_analysis_filter_FIX( LPC_in_pre, x - psEnc->sCmn.predictLPCOrder, psEncCtrl->LTPCoef_Q14, + psEncCtrl->pitchL, invGains_Q16, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.predictLPCOrder ); + + } else { + /************/ + /* UNVOICED */ + /************/ + /* Create signal with prepended subframes, scaled by inverse gains */ + x_ptr = x - psEnc->sCmn.predictLPCOrder; + x_pre_ptr = LPC_in_pre; + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + silk_scale_copy_vector16( x_pre_ptr, x_ptr, invGains_Q16[ i ], + psEnc->sCmn.subfr_length + psEnc->sCmn.predictLPCOrder ); + x_pre_ptr += psEnc->sCmn.subfr_length + psEnc->sCmn.predictLPCOrder; + x_ptr += psEnc->sCmn.subfr_length; + } + + silk_memset( psEncCtrl->LTPCoef_Q14, 0, psEnc->sCmn.nb_subfr * LTP_ORDER * sizeof( opus_int16 ) ); + psEncCtrl->LTPredCodGain_Q7 = 0; + psEnc->sCmn.sum_log_gain_Q7 = 0; + psEncCtrl->LTP_scale_Q14 = 0; + } + + /* Limit on total predictive coding gain */ + if( psEnc->sCmn.first_frame_after_reset ) { + minInvGain_Q30 = SILK_FIX_CONST( 1.0f / MAX_PREDICTION_POWER_GAIN_AFTER_RESET, 30 ); + } else { + minInvGain_Q30 = silk_log2lin( silk_SMLAWB( 16 << 7, (opus_int32)psEncCtrl->LTPredCodGain_Q7, SILK_FIX_CONST( 1.0 / 3, 16 ) ) ); /* Q16 */ + minInvGain_Q30 = silk_DIV32_varQ( minInvGain_Q30, + silk_SMULWW( SILK_FIX_CONST( MAX_PREDICTION_POWER_GAIN, 0 ), + silk_SMLAWB( SILK_FIX_CONST( 0.25, 18 ), SILK_FIX_CONST( 0.75, 18 ), psEncCtrl->coding_quality_Q14 ) ), 14 ); + } + + /* LPC_in_pre contains the LTP-filtered input for voiced, and the unfiltered input for unvoiced */ + silk_find_LPC_FIX( &psEnc->sCmn, NLSF_Q15, LPC_in_pre, minInvGain_Q30 ); + + /* Quantize LSFs */ + silk_process_NLSFs( &psEnc->sCmn, psEncCtrl->PredCoef_Q12, NLSF_Q15, psEnc->sCmn.prev_NLSFq_Q15 ); + + /* Calculate residual energy using quantized LPC coefficients */ + silk_residual_energy_FIX( psEncCtrl->ResNrg, psEncCtrl->ResNrgQ, LPC_in_pre, psEncCtrl->PredCoef_Q12, local_gains, + psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.predictLPCOrder, psEnc->sCmn.arch ); + + /* Copy to prediction struct for use in next frame for interpolation */ + silk_memcpy( psEnc->sCmn.prev_NLSFq_Q15, NLSF_Q15, sizeof( psEnc->sCmn.prev_NLSFq_Q15 ) ); + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_FIX.c new file mode 100644 index 000000000..549f6eada --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_FIX.c @@ -0,0 +1,54 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a( + opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */ + const opus_int16 *rc_Q15, /* I Reflection coefficients [order] Q15 */ + const opus_int32 order /* I Prediction order */ +) +{ + opus_int k, n; + opus_int32 rc, tmp1, tmp2; + + for( k = 0; k < order; k++ ) { + rc = rc_Q15[ k ]; + for( n = 0; n < (k + 1) >> 1; n++ ) { + tmp1 = A_Q24[ n ]; + tmp2 = A_Q24[ k - n - 1 ]; + A_Q24[ n ] = silk_SMLAWB( tmp1, silk_LSHIFT( tmp2, 1 ), rc ); + A_Q24[ k - n - 1 ] = silk_SMLAWB( tmp2, silk_LSHIFT( tmp1, 1 ), rc ); + } + A_Q24[ k ] = -silk_LSHIFT( (opus_int32)rc_Q15[ k ], 9 ); + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_Q16_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_Q16_FIX.c new file mode 100644 index 000000000..1595aa621 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/k2a_Q16_FIX.c @@ -0,0 +1,54 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a_Q16( + opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */ + const opus_int32 *rc_Q16, /* I Reflection coefficients [order] Q16 */ + const opus_int32 order /* I Prediction order */ +) +{ + opus_int k, n; + opus_int32 rc, tmp1, tmp2; + + for( k = 0; k < order; k++ ) { + rc = rc_Q16[ k ]; + for( n = 0; n < (k + 1) >> 1; n++ ) { + tmp1 = A_Q24[ n ]; + tmp2 = A_Q24[ k - n - 1 ]; + A_Q24[ n ] = silk_SMLAWW( tmp1, tmp2, rc ); + A_Q24[ k - n - 1 ] = silk_SMLAWW( tmp2, tmp1, rc ); + } + A_Q24[ k ] = -silk_LSHIFT( rc, 8 ); + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/main_FIX.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/main_FIX.h new file mode 100644 index 000000000..53a97c5a9 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/main_FIX.h @@ -0,0 +1,233 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MAIN_FIX_H +#define SILK_MAIN_FIX_H + +#include "SigProc_FIX.h" +#include "structs_FIX.h" +#include "control.h" +#include "main.h" +#include "PLC.h" +#include "debug.h" +#include "entenc.h" + +#if ((defined(OPUS_ARM_ASM) && defined(FIXED_POINT)) \ + || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +#include "fixed/arm/warped_autocorrelation_FIX_arm.h" +#endif + + +#define silk_encoder_state_Fxx silk_encoder_state_FIX +#define silk_encode_do_VAD_Fxx silk_encode_do_VAD_FIX +#define silk_encode_frame_Fxx silk_encode_frame_FIX + +#define QC 10 +#define QS 13 + +/*********************/ +/* Encoder Functions */ +/*********************/ + +/* High-pass filter with cutoff frequency adaptation based on pitch lag statistics */ +void silk_HP_variable_cutoff( + silk_encoder_state_Fxx state_Fxx[] /* I/O Encoder states */ +); + +/* Encoder main function */ +void silk_encode_do_VAD_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int activity /* I Decision of Opus voice activity detector */ +); + +/* Encoder main function */ +opus_int silk_encode_frame_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int32 *pnBytesOut, /* O Pointer to number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +); + +/* Initializes the Silk encoder state */ +opus_int silk_init_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk FIX encoder state */ + int arch /* I Run-time architecture */ +); + +/* Control the Silk encoder */ +opus_int silk_control_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl, /* I Control structure */ + const opus_int allow_bw_switch, /* I Flag to allow switching audio bandwidth */ + const opus_int channelNb, /* I Channel number */ + const opus_int force_fs_kHz +); + +/**************************/ +/* Noise shaping analysis */ +/**************************/ +/* Compute noise shaping coefficients and initial gain values */ +void silk_noise_shape_analysis_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ + const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ +); + +/* Autocorrelations for a warped frequency axis */ +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +#if !defined(OVERRIDE_silk_warped_autocorrelation_FIX) +#define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), silk_warped_autocorrelation_FIX_c(corr, scale, input, warping_Q16, length, order)) +#endif + +/* Calculation of LTP state scaling */ +void silk_LTP_scale_ctrl_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/**********************************************/ +/* Prediction Analysis */ +/**********************************************/ +/* Find pitch lags */ +void silk_find_pitch_lags_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int16 res[], /* O residual */ + const opus_int16 x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +); + +/* Find LPC and LTP coefficients */ +void silk_find_pred_coefs_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + const opus_int16 res_pitch[], /* I Residual from pitch analysis */ + const opus_int16 x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/* LPC analysis */ +void silk_find_LPC_FIX( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const opus_int16 x[], /* I Input signal */ + const opus_int32 minInvGain_Q30 /* I Inverse of max prediction gain */ +); + +/* LTP analysis */ +void silk_find_LTP_FIX( + opus_int32 XXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Correlation matrix */ + opus_int32 xXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER ], /* O Correlation vector */ + const opus_int16 r_lpc[], /* I Residual signal after LPC */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +); + +void silk_LTP_analysis_filter_FIX( + opus_int16 *LTP_res, /* O LTP residual signal of length MAX_NB_SUBFR * ( pre_length + subfr_length ) */ + const opus_int16 *x, /* I Pointer to input signal with at least max( pitchL ) preceding samples */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],/* I LTP_ORDER LTP coefficients for each MAX_NB_SUBFR subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag, one for each subframe */ + const opus_int32 invGains_Q16[ MAX_NB_SUBFR ], /* I Inverse quantization gains, one for each subframe */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int pre_length /* I Length of the preceding samples starting at &x[0] for each subframe */ +); + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FIX( + opus_int32 nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + opus_int nrgsQ[ MAX_NB_SUBFR ], /* O Q value per subframe */ + const opus_int16 x[], /* I Input signal */ + opus_int16 a_Q12[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const opus_int32 gains[ MAX_NB_SUBFR ], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +); + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +opus_int32 silk_residual_energy16_covar_FIX( + const opus_int16 *c, /* I Prediction vector */ + const opus_int32 *wXX, /* I Correlation matrix */ + const opus_int32 *wXx, /* I Correlation vector */ + opus_int32 wxx, /* I Signal energy */ + opus_int D, /* I Dimension */ + opus_int cQ /* I Q value for c vector 0 - 15 */ +); + +/* Processing of gains */ +void silk_process_gains_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/******************/ +/* Linear Algebra */ +/******************/ +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *XX, /* O Pointer to X'*X correlation matrix [ order x order ] */ + opus_int32 *nrg, /* O Energy of x vector */ + opus_int *rshifts, /* O Right shifts of correlations */ + int arch /* I Run-time architecture */ +); + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int16 *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *Xt, /* O Pointer to X'*t correlation vector [order] */ + const opus_int rshifts, /* I Right shifts of correlations */ + int arch /* I Run-time architecture */ +); + +#endif /* SILK_MAIN_FIX_H */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/noise_shape_analysis_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/noise_shape_analysis_FIX.c new file mode 100644 index 000000000..49ea59e99 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/noise_shape_analysis_FIX.c @@ -0,0 +1,400 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Compute gain to make warped filter coefficients have a zero mean log frequency response on a */ +/* non-warped frequency scale. (So that it can be implemented with a minimum-phase monic filter.) */ +/* Note: A monic filter is one with the first coefficient equal to 1.0. In Silk we omit the first */ +/* coefficient in an array of coefficients, for monic filters. */ +static OPUS_INLINE opus_int32 warped_gain( /* gain in Q16*/ + const opus_int32 *coefs_Q24, + opus_int lambda_Q16, + opus_int order +) { + opus_int i; + opus_int32 gain_Q24; + + lambda_Q16 = -lambda_Q16; + gain_Q24 = coefs_Q24[ order - 1 ]; + for( i = order - 2; i >= 0; i-- ) { + gain_Q24 = silk_SMLAWB( coefs_Q24[ i ], gain_Q24, lambda_Q16 ); + } + gain_Q24 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 24 ), gain_Q24, -lambda_Q16 ); + return silk_INVERSE32_varQ( gain_Q24, 40 ); +} + +/* Convert warped filter coefficients to monic pseudo-warped coefficients and limit maximum */ +/* amplitude of monic warped coefficients by using bandwidth expansion on the true coefficients */ +static OPUS_INLINE void limit_warped_coefs( + opus_int32 *coefs_Q24, + opus_int lambda_Q16, + opus_int32 limit_Q24, + opus_int order +) { + opus_int i, iter, ind = 0; + opus_int32 tmp, maxabs_Q24, chirp_Q16, gain_Q16; + opus_int32 nom_Q16, den_Q24; + opus_int32 limit_Q20, maxabs_Q20; + + /* Convert to monic coefficients */ + lambda_Q16 = -lambda_Q16; + for( i = order - 1; i > 0; i-- ) { + coefs_Q24[ i - 1 ] = silk_SMLAWB( coefs_Q24[ i - 1 ], coefs_Q24[ i ], lambda_Q16 ); + } + lambda_Q16 = -lambda_Q16; + nom_Q16 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 16 ), -(opus_int32)lambda_Q16, lambda_Q16 ); + den_Q24 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 24 ), coefs_Q24[ 0 ], lambda_Q16 ); + gain_Q16 = silk_DIV32_varQ( nom_Q16, den_Q24, 24 ); + for( i = 0; i < order; i++ ) { + coefs_Q24[ i ] = silk_SMULWW( gain_Q16, coefs_Q24[ i ] ); + } + limit_Q20 = silk_RSHIFT(limit_Q24, 4); + for( iter = 0; iter < 10; iter++ ) { + /* Find maximum absolute value */ + maxabs_Q24 = -1; + for( i = 0; i < order; i++ ) { + tmp = silk_abs_int32( coefs_Q24[ i ] ); + if( tmp > maxabs_Q24 ) { + maxabs_Q24 = tmp; + ind = i; + } + } + /* Use Q20 to avoid any overflow when multiplying by (ind + 1) later. */ + maxabs_Q20 = silk_RSHIFT(maxabs_Q24, 4); + if( maxabs_Q20 <= limit_Q20 ) { + /* Coefficients are within range - done */ + return; + } + + /* Convert back to true warped coefficients */ + for( i = 1; i < order; i++ ) { + coefs_Q24[ i - 1 ] = silk_SMLAWB( coefs_Q24[ i - 1 ], coefs_Q24[ i ], lambda_Q16 ); + } + gain_Q16 = silk_INVERSE32_varQ( gain_Q16, 32 ); + for( i = 0; i < order; i++ ) { + coefs_Q24[ i ] = silk_SMULWW( gain_Q16, coefs_Q24[ i ] ); + } + + /* Apply bandwidth expansion */ + chirp_Q16 = SILK_FIX_CONST( 0.99, 16 ) - silk_DIV32_varQ( + silk_SMULWB( maxabs_Q20 - limit_Q20, silk_SMLABB( SILK_FIX_CONST( 0.8, 10 ), SILK_FIX_CONST( 0.1, 10 ), iter ) ), + silk_MUL( maxabs_Q20, ind + 1 ), 22 ); + silk_bwexpander_32( coefs_Q24, order, chirp_Q16 ); + + /* Convert to monic warped coefficients */ + lambda_Q16 = -lambda_Q16; + for( i = order - 1; i > 0; i-- ) { + coefs_Q24[ i - 1 ] = silk_SMLAWB( coefs_Q24[ i - 1 ], coefs_Q24[ i ], lambda_Q16 ); + } + lambda_Q16 = -lambda_Q16; + nom_Q16 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 16 ), -(opus_int32)lambda_Q16, lambda_Q16 ); + den_Q24 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 24 ), coefs_Q24[ 0 ], lambda_Q16 ); + gain_Q16 = silk_DIV32_varQ( nom_Q16, den_Q24, 24 ); + for( i = 0; i < order; i++ ) { + coefs_Q24[ i ] = silk_SMULWW( gain_Q16, coefs_Q24[ i ] ); + } + } + silk_assert( 0 ); +} + +/**************************************************************/ +/* Compute noise shaping coefficients and initial gain values */ +/**************************************************************/ +void silk_noise_shape_analysis_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ + const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ +) +{ + silk_shape_state_FIX *psShapeSt = &psEnc->sShape; + opus_int k, i, nSamples, nSegs, Qnrg, b_Q14, warping_Q16, scale = 0; + opus_int32 SNR_adj_dB_Q7, HarmShapeGain_Q16, Tilt_Q16, tmp32; + opus_int32 nrg, log_energy_Q7, log_energy_prev_Q7, energy_variation_Q7; + opus_int32 BWExp_Q16, gain_mult_Q16, gain_add_Q16, strength_Q16, b_Q8; + opus_int32 auto_corr[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int32 refl_coef_Q16[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 AR_Q24[ MAX_SHAPE_LPC_ORDER ]; + VARDECL( opus_int16, x_windowed ); + const opus_int16 *x_ptr, *pitch_res_ptr; + SAVE_STACK; + + /* Point to start of first LPC analysis block */ + x_ptr = x - psEnc->sCmn.la_shape; + + /****************/ + /* GAIN CONTROL */ + /****************/ + SNR_adj_dB_Q7 = psEnc->sCmn.SNR_dB_Q7; + + /* Input quality is the average of the quality in the lowest two VAD bands */ + psEncCtrl->input_quality_Q14 = ( opus_int )silk_RSHIFT( (opus_int32)psEnc->sCmn.input_quality_bands_Q15[ 0 ] + + psEnc->sCmn.input_quality_bands_Q15[ 1 ], 2 ); + + /* Coding quality level, between 0.0_Q0 and 1.0_Q0, but in Q14 */ + psEncCtrl->coding_quality_Q14 = silk_RSHIFT( silk_sigm_Q15( silk_RSHIFT_ROUND( SNR_adj_dB_Q7 - + SILK_FIX_CONST( 20.0, 7 ), 4 ) ), 1 ); + + /* Reduce coding SNR during low speech activity */ + if( psEnc->sCmn.useCBR == 0 ) { + b_Q8 = SILK_FIX_CONST( 1.0, 8 ) - psEnc->sCmn.speech_activity_Q8; + b_Q8 = silk_SMULWB( silk_LSHIFT( b_Q8, 8 ), b_Q8 ); + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMULBB( SILK_FIX_CONST( -BG_SNR_DECR_dB, 7 ) >> ( 4 + 1 ), b_Q8 ), /* Q11*/ + silk_SMULWB( SILK_FIX_CONST( 1.0, 14 ) + psEncCtrl->input_quality_Q14, psEncCtrl->coding_quality_Q14 ) ); /* Q12*/ + } + + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce gains for periodic signals */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, SILK_FIX_CONST( HARM_SNR_INCR_dB, 8 ), psEnc->LTPCorr_Q15 ); + } else { + /* For unvoiced signals and low-quality input, adjust the quality slower than SNR_dB setting */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMLAWB( SILK_FIX_CONST( 6.0, 9 ), -SILK_FIX_CONST( 0.4, 18 ), psEnc->sCmn.SNR_dB_Q7 ), + SILK_FIX_CONST( 1.0, 14 ) - psEncCtrl->input_quality_Q14 ); + } + + /*************************/ + /* SPARSENESS PROCESSING */ + /*************************/ + /* Set quantizer offset */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Initially set to 0; may be overruled in process_gains(..) */ + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + /* Sparseness measure, based on relative fluctuations of energy per 2 milliseconds */ + nSamples = silk_LSHIFT( psEnc->sCmn.fs_kHz, 1 ); + energy_variation_Q7 = 0; + log_energy_prev_Q7 = 0; + pitch_res_ptr = pitch_res; + nSegs = silk_SMULBB( SUB_FRAME_LENGTH_MS, psEnc->sCmn.nb_subfr ) / 2; + for( k = 0; k < nSegs; k++ ) { + silk_sum_sqr_shift( &nrg, &scale, pitch_res_ptr, nSamples ); + nrg += silk_RSHIFT( nSamples, scale ); /* Q(-scale)*/ + + log_energy_Q7 = silk_lin2log( nrg ); + if( k > 0 ) { + energy_variation_Q7 += silk_abs( log_energy_Q7 - log_energy_prev_Q7 ); + } + log_energy_prev_Q7 = log_energy_Q7; + pitch_res_ptr += nSamples; + } + + /* Set quantization offset depending on sparseness measure */ + if( energy_variation_Q7 > SILK_FIX_CONST( ENERGY_VARIATION_THRESHOLD_QNT_OFFSET, 7 ) * (nSegs-1) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + } + + /*******************************/ + /* Control bandwidth expansion */ + /*******************************/ + /* More BWE for signals with high prediction gain */ + strength_Q16 = silk_SMULWB( psEncCtrl->predGain_Q16, SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ); + BWExp_Q16 = silk_DIV32_varQ( SILK_FIX_CONST( BANDWIDTH_EXPANSION, 16 ), + silk_SMLAWW( SILK_FIX_CONST( 1.0, 16 ), strength_Q16, strength_Q16 ), 16 ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Slightly more warping in analysis will move quantization noise up in frequency, where it's better masked */ + warping_Q16 = silk_SMLAWB( psEnc->sCmn.warping_Q16, (opus_int32)psEncCtrl->coding_quality_Q14, SILK_FIX_CONST( 0.01, 18 ) ); + } else { + warping_Q16 = 0; + } + + /********************************************/ + /* Compute noise shaping AR coefs and gains */ + /********************************************/ + ALLOC( x_windowed, psEnc->sCmn.shapeWinLength, opus_int16 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Apply window: sine slope followed by flat part followed by cosine slope */ + opus_int shift, slope_part, flat_part; + flat_part = psEnc->sCmn.fs_kHz * 3; + slope_part = silk_RSHIFT( psEnc->sCmn.shapeWinLength - flat_part, 1 ); + + silk_apply_sine_window( x_windowed, x_ptr, 1, slope_part ); + shift = slope_part; + silk_memcpy( x_windowed + shift, x_ptr + shift, flat_part * sizeof(opus_int16) ); + shift += flat_part; + silk_apply_sine_window( x_windowed + shift, x_ptr + shift, 2, slope_part ); + + /* Update pointer: next LPC analysis block */ + x_ptr += psEnc->sCmn.subfr_length; + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Calculate warped auto correlation */ + silk_warped_autocorrelation_FIX( auto_corr, &scale, x_windowed, warping_Q16, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder, arch ); + } else { + /* Calculate regular auto correlation */ + silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, arch ); + } + + /* Add white noise, as a fraction of energy */ + auto_corr[0] = silk_ADD32( auto_corr[0], silk_max_32( silk_SMULWB( silk_RSHIFT( auto_corr[ 0 ], 4 ), + SILK_FIX_CONST( SHAPE_WHITE_NOISE_FRACTION, 20 ) ), 1 ) ); + + /* Calculate the reflection coefficients using schur */ + nrg = silk_schur64( refl_coef_Q16, auto_corr, psEnc->sCmn.shapingLPCOrder ); + silk_assert( nrg >= 0 ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a_Q16( AR_Q24, refl_coef_Q16, psEnc->sCmn.shapingLPCOrder ); + + Qnrg = -scale; /* range: -12...30*/ + silk_assert( Qnrg >= -12 ); + silk_assert( Qnrg <= 30 ); + + /* Make sure that Qnrg is an even number */ + if( Qnrg & 1 ) { + Qnrg -= 1; + nrg >>= 1; + } + + tmp32 = silk_SQRT_APPROX( nrg ); + Qnrg >>= 1; /* range: -6...15*/ + + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT_SAT32( tmp32, 16 - Qnrg ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Adjust gain for warping */ + gain_mult_Q16 = warped_gain( AR_Q24, warping_Q16, psEnc->sCmn.shapingLPCOrder ); + silk_assert( psEncCtrl->Gains_Q16[ k ] > 0 ); + if( psEncCtrl->Gains_Q16[ k ] < SILK_FIX_CONST( 0.25, 16 ) ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + } else { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( silk_RSHIFT_ROUND( psEncCtrl->Gains_Q16[ k ], 1 ), gain_mult_Q16 ); + if ( psEncCtrl->Gains_Q16[ k ] >= ( silk_int32_MAX >> 1 ) ) { + psEncCtrl->Gains_Q16[ k ] = silk_int32_MAX; + } else { + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT32( psEncCtrl->Gains_Q16[ k ], 1 ); + } + } + silk_assert( psEncCtrl->Gains_Q16[ k ] > 0 ); + } + + /* Bandwidth expansion */ + silk_bwexpander_32( AR_Q24, psEnc->sCmn.shapingLPCOrder, BWExp_Q16 ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Convert to monic warped prediction coefficients and limit absolute values */ + limit_warped_coefs( AR_Q24, warping_Q16, SILK_FIX_CONST( 3.999, 24 ), psEnc->sCmn.shapingLPCOrder ); + + /* Convert from Q24 to Q13 and store in int16 */ + for( i = 0; i < psEnc->sCmn.shapingLPCOrder; i++ ) { + psEncCtrl->AR_Q13[ k * MAX_SHAPE_LPC_ORDER + i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( AR_Q24[ i ], 11 ) ); + } + } else { + silk_LPC_fit( &psEncCtrl->AR_Q13[ k * MAX_SHAPE_LPC_ORDER ], AR_Q24, 13, 24, psEnc->sCmn.shapingLPCOrder ); + } + } + + /*****************/ + /* Gain tweaking */ + /*****************/ + /* Increase gains during low speech activity and put lower limit on gains */ + gain_mult_Q16 = silk_log2lin( -silk_SMLAWB( -SILK_FIX_CONST( 16.0, 7 ), SNR_adj_dB_Q7, SILK_FIX_CONST( 0.16, 16 ) ) ); + gain_add_Q16 = silk_log2lin( silk_SMLAWB( SILK_FIX_CONST( 16.0, 7 ), SILK_FIX_CONST( MIN_QGAIN_DB, 7 ), SILK_FIX_CONST( 0.16, 16 ) ) ); + silk_assert( gain_mult_Q16 > 0 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + silk_assert( psEncCtrl->Gains_Q16[ k ] >= 0 ); + psEncCtrl->Gains_Q16[ k ] = silk_ADD_POS_SAT32( psEncCtrl->Gains_Q16[ k ], gain_add_Q16 ); + } + + + /************************************************/ + /* Control low-frequency shaping and noise tilt */ + /************************************************/ + /* Less low frequency shaping for noisy inputs */ + strength_Q16 = silk_MUL( SILK_FIX_CONST( LOW_FREQ_SHAPING, 4 ), silk_SMLAWB( SILK_FIX_CONST( 1.0, 12 ), + SILK_FIX_CONST( LOW_QUALITY_LOW_FREQ_SHAPING_DECR, 13 ), psEnc->sCmn.input_quality_bands_Q15[ 0 ] - SILK_FIX_CONST( 1.0, 15 ) ) ); + strength_Q16 = silk_RSHIFT( silk_MUL( strength_Q16, psEnc->sCmn.speech_activity_Q8 ), 8 ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce low frequencies quantization noise for periodic signals, depending on pitch lag */ + /*f = 400; freqz([1, -0.98 + 2e-4 * f], [1, -0.97 + 7e-4 * f], 2^12, Fs); axis([0, 1000, -10, 1])*/ + opus_int fs_kHz_inv = silk_DIV32_16( SILK_FIX_CONST( 0.2, 14 ), psEnc->sCmn.fs_kHz ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + b_Q14 = fs_kHz_inv + silk_DIV32_16( SILK_FIX_CONST( 3.0, 14 ), psEncCtrl->pitchL[ k ] ); + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ k ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - silk_SMULWB( strength_Q16, b_Q14 ), 16 ); + psEncCtrl->LF_shp_Q14[ k ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + } + silk_assert( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ) < SILK_FIX_CONST( 0.5, 24 ) ); /* Guarantees that second argument to SMULWB() is within range of an opus_int16*/ + Tilt_Q16 = - SILK_FIX_CONST( HP_NOISE_COEF, 16 ) - + silk_SMULWB( SILK_FIX_CONST( 1.0, 16 ) - SILK_FIX_CONST( HP_NOISE_COEF, 16 ), + silk_SMULWB( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ), psEnc->sCmn.speech_activity_Q8 ) ); + } else { + b_Q14 = silk_DIV32_16( 21299, psEnc->sCmn.fs_kHz ); /* 1.3_Q0 = 21299_Q14*/ + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ 0 ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - + silk_SMULWB( strength_Q16, silk_SMULWB( SILK_FIX_CONST( 0.6, 16 ), b_Q14 ) ), 16 ); + psEncCtrl->LF_shp_Q14[ 0 ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + for( k = 1; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->LF_shp_Q14[ k ] = psEncCtrl->LF_shp_Q14[ 0 ]; + } + Tilt_Q16 = -SILK_FIX_CONST( HP_NOISE_COEF, 16 ); + } + + /****************************/ + /* HARMONIC SHAPING CONTROL */ + /****************************/ + if( USE_HARM_SHAPING && psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* More harmonic noise shaping for high bitrates or noisy input */ + HarmShapeGain_Q16 = silk_SMLAWB( SILK_FIX_CONST( HARMONIC_SHAPING, 16 ), + SILK_FIX_CONST( 1.0, 16 ) - silk_SMULWB( SILK_FIX_CONST( 1.0, 18 ) - silk_LSHIFT( psEncCtrl->coding_quality_Q14, 4 ), + psEncCtrl->input_quality_Q14 ), SILK_FIX_CONST( HIGH_RATE_OR_LOW_QUALITY_HARMONIC_SHAPING, 16 ) ); + + /* Less harmonic noise shaping for less periodic signals */ + HarmShapeGain_Q16 = silk_SMULWB( silk_LSHIFT( HarmShapeGain_Q16, 1 ), + silk_SQRT_APPROX( silk_LSHIFT( psEnc->LTPCorr_Q15, 15 ) ) ); + } else { + HarmShapeGain_Q16 = 0; + } + + /*************************/ + /* Smooth over subframes */ + /*************************/ + for( k = 0; k < MAX_NB_SUBFR; k++ ) { + psShapeSt->HarmShapeGain_smth_Q16 = + silk_SMLAWB( psShapeSt->HarmShapeGain_smth_Q16, HarmShapeGain_Q16 - psShapeSt->HarmShapeGain_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + psShapeSt->Tilt_smth_Q16 = + silk_SMLAWB( psShapeSt->Tilt_smth_Q16, Tilt_Q16 - psShapeSt->Tilt_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + + psEncCtrl->HarmShapeGain_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmShapeGain_smth_Q16, 2 ); + psEncCtrl->Tilt_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->Tilt_smth_Q16, 2 ); + } + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/pitch_analysis_core_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/pitch_analysis_core_FIX.c new file mode 100644 index 000000000..14729046d --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/pitch_analysis_core_FIX.c @@ -0,0 +1,721 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/*********************************************************** +* Pitch analyser function +********************************************************** */ +#include "SigProc_FIX.h" +#include "pitch_est_defines.h" +#include "stack_alloc.h" +#include "debug.h" +#include "pitch.h" + +#define SCRATCH_SIZE 22 +#define SF_LENGTH_4KHZ ( PE_SUBFR_LENGTH_MS * 4 ) +#define SF_LENGTH_8KHZ ( PE_SUBFR_LENGTH_MS * 8 ) +#define MIN_LAG_4KHZ ( PE_MIN_LAG_MS * 4 ) +#define MIN_LAG_8KHZ ( PE_MIN_LAG_MS * 8 ) +#define MAX_LAG_4KHZ ( PE_MAX_LAG_MS * 4 ) +#define MAX_LAG_8KHZ ( PE_MAX_LAG_MS * 8 - 1 ) +#define CSTRIDE_4KHZ ( MAX_LAG_4KHZ + 1 - MIN_LAG_4KHZ ) +#define CSTRIDE_8KHZ ( MAX_LAG_8KHZ + 3 - ( MIN_LAG_8KHZ - 2 ) ) +#define D_COMP_MIN ( MIN_LAG_8KHZ - 3 ) +#define D_COMP_MAX ( MAX_LAG_8KHZ + 4 ) +#define D_COMP_STRIDE ( D_COMP_MAX - D_COMP_MIN ) + +typedef opus_int32 silk_pe_stage3_vals[ PE_NB_STAGE3_LAGS ]; + +/************************************************************/ +/* Internally used functions */ +/************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_pe_stage3_vals cross_corr_st3[], /* O 3 DIM correlation array */ + const opus_int16 frame[], /* I vector to correlate */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of a 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +); + +static void silk_P_Ana_calc_energy_st3( + silk_pe_stage3_vals energies_st3[], /* O 3 DIM energy array */ + const opus_int16 frame[], /* I vector to calc energy in */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of one 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +); + +/*************************************************************/ +/* FIXED POINT CORE PITCH ANALYSIS FUNCTION */ +/*************************************************************/ +opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 voiced, 1 unvoiced */ + const opus_int16 *frame_unscaled, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */ + opus_int *pitch_out, /* O 4 pitch lag values */ + opus_int16 *lagIndex, /* O Lag Index */ + opus_int8 *contourIndex, /* O Pitch contour Index */ + opus_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */ + opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ + const opus_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */ + const opus_int search_thres2_Q13, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int Fs_kHz, /* I Sample frequency (kHz) */ + const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ + const opus_int nb_subfr, /* I number of 5 ms subframes */ + int arch /* I Run-time architecture */ +) +{ + VARDECL( opus_int16, frame_8kHz_buf ); + VARDECL( opus_int16, frame_4kHz ); + VARDECL( opus_int16, frame_scaled ); + opus_int32 filt_state[ 6 ]; + const opus_int16 *frame, *frame_8kHz; + opus_int i, k, d, j; + VARDECL( opus_int16, C ); + VARDECL( opus_int32, xcorr32 ); + const opus_int16 *target_ptr, *basis_ptr; + opus_int32 cross_corr, normalizer, energy, energy_basis, energy_target; + opus_int d_srch[ PE_D_SRCH_LENGTH ], Cmax, length_d_srch, length_d_comp, shift; + VARDECL( opus_int16, d_comp ); + opus_int32 sum, threshold, lag_counter; + opus_int CBimax, CBimax_new, CBimax_old, lag, start_lag, end_lag, lag_new; + opus_int32 CC[ PE_NB_CBKS_STAGE2_EXT ], CCmax, CCmax_b, CCmax_new_b, CCmax_new; + VARDECL( silk_pe_stage3_vals, energies_st3 ); + VARDECL( silk_pe_stage3_vals, cross_corr_st3 ); + opus_int frame_length, frame_length_8kHz, frame_length_4kHz; + opus_int sf_length; + opus_int min_lag; + opus_int max_lag; + opus_int32 contour_bias_Q15, diff; + opus_int nb_cbk_search, cbk_size; + opus_int32 delta_lag_log2_sqr_Q7, lag_log2_Q7, prevLag_log2_Q7, prev_lag_bias_Q13; + const opus_int8 *Lag_CB_ptr; + SAVE_STACK; + + /* Check for valid sampling frequency */ + celt_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); + + /* Check for valid complexity setting */ + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + silk_assert( search_thres1_Q16 >= 0 && search_thres1_Q16 <= (1<<16) ); + silk_assert( search_thres2_Q13 >= 0 && search_thres2_Q13 <= (1<<13) ); + + /* Set up frame lengths max / min lag for the sampling frequency */ + frame_length = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * Fs_kHz; + frame_length_4kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 4; + frame_length_8kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 8; + sf_length = PE_SUBFR_LENGTH_MS * Fs_kHz; + min_lag = PE_MIN_LAG_MS * Fs_kHz; + max_lag = PE_MAX_LAG_MS * Fs_kHz - 1; + + /* Downscale input if necessary */ + silk_sum_sqr_shift( &energy, &shift, frame_unscaled, frame_length ); + shift += 3 - silk_CLZ32( energy ); /* at least two bits headroom */ + ALLOC( frame_scaled, frame_length, opus_int16 ); + if( shift > 0 ) { + shift = silk_RSHIFT( shift + 1, 1 ); + for( i = 0; i < frame_length; i++ ) { + frame_scaled[ i ] = silk_RSHIFT( frame_unscaled[ i ], shift ); + } + frame = frame_scaled; + } else { + frame = frame_unscaled; + } + + ALLOC( frame_8kHz_buf, ( Fs_kHz == 8 ) ? 1 : frame_length_8kHz, opus_int16 ); + /* Resample from input sampled at Fs_kHz to 8 kHz */ + if( Fs_kHz == 16 ) { + silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) ); + silk_resampler_down2( filt_state, frame_8kHz_buf, frame, frame_length ); + frame_8kHz = frame_8kHz_buf; + } else if( Fs_kHz == 12 ) { + silk_memset( filt_state, 0, 6 * sizeof( opus_int32 ) ); + silk_resampler_down2_3( filt_state, frame_8kHz_buf, frame, frame_length ); + frame_8kHz = frame_8kHz_buf; + } else { + celt_assert( Fs_kHz == 8 ); + frame_8kHz = frame; + } + + /* Decimate again to 4 kHz */ + silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) );/* Set state to zero */ + ALLOC( frame_4kHz, frame_length_4kHz, opus_int16 ); + silk_resampler_down2( filt_state, frame_4kHz, frame_8kHz, frame_length_8kHz ); + + /* Low-pass filter */ + for( i = frame_length_4kHz - 1; i > 0; i-- ) { + frame_4kHz[ i ] = silk_ADD_SAT16( frame_4kHz[ i ], frame_4kHz[ i - 1 ] ); + } + + + /****************************************************************************** + * FIRST STAGE, operating in 4 khz + ******************************************************************************/ + ALLOC( C, nb_subfr * CSTRIDE_8KHZ, opus_int16 ); + ALLOC( xcorr32, MAX_LAG_4KHZ-MIN_LAG_4KHZ+1, opus_int32 ); + silk_memset( C, 0, (nb_subfr >> 1) * CSTRIDE_4KHZ * sizeof( opus_int16 ) ); + target_ptr = &frame_4kHz[ silk_LSHIFT( SF_LENGTH_4KHZ, 2 ) ]; + for( k = 0; k < nb_subfr >> 1; k++ ) { + /* Check that we are within range of the array */ + celt_assert( target_ptr >= frame_4kHz ); + celt_assert( target_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + basis_ptr = target_ptr - MIN_LAG_4KHZ; + + /* Check that we are within range of the array */ + celt_assert( basis_ptr >= frame_4kHz ); + celt_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + celt_pitch_xcorr( target_ptr, target_ptr - MAX_LAG_4KHZ, xcorr32, SF_LENGTH_8KHZ, MAX_LAG_4KHZ - MIN_LAG_4KHZ + 1, arch ); + + /* Calculate first vector products before loop */ + cross_corr = xcorr32[ MAX_LAG_4KHZ - MIN_LAG_4KHZ ]; + normalizer = silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ, arch ); + normalizer = silk_ADD32( normalizer, silk_inner_prod_aligned( basis_ptr, basis_ptr, SF_LENGTH_8KHZ, arch ) ); + normalizer = silk_ADD32( normalizer, silk_SMULBB( SF_LENGTH_8KHZ, 4000 ) ); + + matrix_ptr( C, k, 0, CSTRIDE_4KHZ ) = + (opus_int16)silk_DIV32_varQ( cross_corr, normalizer, 13 + 1 ); /* Q13 */ + + /* From now on normalizer is computed recursively */ + for( d = MIN_LAG_4KHZ + 1; d <= MAX_LAG_4KHZ; d++ ) { + basis_ptr--; + + /* Check that we are within range of the array */ + silk_assert( basis_ptr >= frame_4kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + cross_corr = xcorr32[ MAX_LAG_4KHZ - d ]; + + /* Add contribution of new sample and remove contribution from oldest sample */ + normalizer = silk_ADD32( normalizer, + silk_SMULBB( basis_ptr[ 0 ], basis_ptr[ 0 ] ) - + silk_SMULBB( basis_ptr[ SF_LENGTH_8KHZ ], basis_ptr[ SF_LENGTH_8KHZ ] ) ); + + matrix_ptr( C, k, d - MIN_LAG_4KHZ, CSTRIDE_4KHZ) = + (opus_int16)silk_DIV32_varQ( cross_corr, normalizer, 13 + 1 ); /* Q13 */ + } + /* Update target pointer */ + target_ptr += SF_LENGTH_8KHZ; + } + + /* Combine two subframes into single correlation measure and apply short-lag bias */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + for( i = MAX_LAG_4KHZ; i >= MIN_LAG_4KHZ; i-- ) { + sum = (opus_int32)matrix_ptr( C, 0, i - MIN_LAG_4KHZ, CSTRIDE_4KHZ ) + + (opus_int32)matrix_ptr( C, 1, i - MIN_LAG_4KHZ, CSTRIDE_4KHZ ); /* Q14 */ + sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q14 */ + C[ i - MIN_LAG_4KHZ ] = (opus_int16)sum; /* Q14 */ + } + } else { + /* Only short-lag bias */ + for( i = MAX_LAG_4KHZ; i >= MIN_LAG_4KHZ; i-- ) { + sum = silk_LSHIFT( (opus_int32)C[ i - MIN_LAG_4KHZ ], 1 ); /* Q14 */ + sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q14 */ + C[ i - MIN_LAG_4KHZ ] = (opus_int16)sum; /* Q14 */ + } + } + + /* Sort */ + length_d_srch = silk_ADD_LSHIFT32( 4, complexity, 1 ); + celt_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); + silk_insertion_sort_decreasing_int16( C, d_srch, CSTRIDE_4KHZ, + length_d_srch ); + + /* Escape if correlation is very low already here */ + Cmax = (opus_int)C[ 0 ]; /* Q14 */ + if( Cmax < SILK_FIX_CONST( 0.2, 14 ) ) { + silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); + *LTPCorr_Q15 = 0; + *lagIndex = 0; + *contourIndex = 0; + RESTORE_STACK; + return 1; + } + + threshold = silk_SMULWB( search_thres1_Q16, Cmax ); + for( i = 0; i < length_d_srch; i++ ) { + /* Convert to 8 kHz indices for the sorted correlation that exceeds the threshold */ + if( C[ i ] > threshold ) { + d_srch[ i ] = silk_LSHIFT( d_srch[ i ] + MIN_LAG_4KHZ, 1 ); + } else { + length_d_srch = i; + break; + } + } + celt_assert( length_d_srch > 0 ); + + ALLOC( d_comp, D_COMP_STRIDE, opus_int16 ); + for( i = D_COMP_MIN; i < D_COMP_MAX; i++ ) { + d_comp[ i - D_COMP_MIN ] = 0; + } + for( i = 0; i < length_d_srch; i++ ) { + d_comp[ d_srch[ i ] - D_COMP_MIN ] = 1; + } + + /* Convolution */ + for( i = D_COMP_MAX - 1; i >= MIN_LAG_8KHZ; i-- ) { + d_comp[ i - D_COMP_MIN ] += + d_comp[ i - 1 - D_COMP_MIN ] + d_comp[ i - 2 - D_COMP_MIN ]; + } + + length_d_srch = 0; + for( i = MIN_LAG_8KHZ; i < MAX_LAG_8KHZ + 1; i++ ) { + if( d_comp[ i + 1 - D_COMP_MIN ] > 0 ) { + d_srch[ length_d_srch ] = i; + length_d_srch++; + } + } + + /* Convolution */ + for( i = D_COMP_MAX - 1; i >= MIN_LAG_8KHZ; i-- ) { + d_comp[ i - D_COMP_MIN ] += d_comp[ i - 1 - D_COMP_MIN ] + + d_comp[ i - 2 - D_COMP_MIN ] + d_comp[ i - 3 - D_COMP_MIN ]; + } + + length_d_comp = 0; + for( i = MIN_LAG_8KHZ; i < D_COMP_MAX; i++ ) { + if( d_comp[ i - D_COMP_MIN ] > 0 ) { + d_comp[ length_d_comp ] = i - 2; + length_d_comp++; + } + } + + /********************************************************************************** + ** SECOND STAGE, operating at 8 kHz, on lag sections with high correlation + *************************************************************************************/ + + /********************************************************************************* + * Find energy of each subframe projected onto its history, for a range of delays + *********************************************************************************/ + silk_memset( C, 0, nb_subfr * CSTRIDE_8KHZ * sizeof( opus_int16 ) ); + + target_ptr = &frame_8kHz[ PE_LTP_MEM_LENGTH_MS * 8 ]; + for( k = 0; k < nb_subfr; k++ ) { + + /* Check that we are within range of the array */ + celt_assert( target_ptr >= frame_8kHz ); + celt_assert( target_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); + + energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ, arch ), 1 ); + for( j = 0; j < length_d_comp; j++ ) { + d = d_comp[ j ]; + basis_ptr = target_ptr - d; + + /* Check that we are within range of the array */ + silk_assert( basis_ptr >= frame_8kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); + + cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, SF_LENGTH_8KHZ, arch ); + if( cross_corr > 0 ) { + energy_basis = silk_inner_prod_aligned( basis_ptr, basis_ptr, SF_LENGTH_8KHZ, arch ); + matrix_ptr( C, k, d - ( MIN_LAG_8KHZ - 2 ), CSTRIDE_8KHZ ) = + (opus_int16)silk_DIV32_varQ( cross_corr, + silk_ADD32( energy_target, + energy_basis ), + 13 + 1 ); /* Q13 */ + } else { + matrix_ptr( C, k, d - ( MIN_LAG_8KHZ - 2 ), CSTRIDE_8KHZ ) = 0; + } + } + target_ptr += SF_LENGTH_8KHZ; + } + + /* search over lag range and lags codebook */ + /* scale factor for lag codebook, as a function of center lag */ + + CCmax = silk_int32_MIN; + CCmax_b = silk_int32_MIN; + + CBimax = 0; /* To avoid returning undefined lag values */ + lag = -1; /* To check if lag with strong enough correlation has been found */ + + if( prevLag > 0 ) { + if( Fs_kHz == 12 ) { + prevLag = silk_DIV32_16( silk_LSHIFT( prevLag, 1 ), 3 ); + } else if( Fs_kHz == 16 ) { + prevLag = silk_RSHIFT( prevLag, 1 ); + } + prevLag_log2_Q7 = silk_lin2log( (opus_int32)prevLag ); + } else { + prevLag_log2_Q7 = 0; + } + silk_assert( search_thres2_Q13 == silk_SAT16( search_thres2_Q13 ) ); + /* Set up stage 2 codebook based on number of subframes */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + cbk_size = PE_NB_CBKS_STAGE2_EXT; + Lag_CB_ptr = &silk_CB_lags_stage2[ 0 ][ 0 ]; + if( Fs_kHz == 8 && complexity > SILK_PE_MIN_COMPLEX ) { + /* If input is 8 khz use a larger codebook here because it is last stage */ + nb_cbk_search = PE_NB_CBKS_STAGE2_EXT; + } else { + nb_cbk_search = PE_NB_CBKS_STAGE2; + } + } else { + cbk_size = PE_NB_CBKS_STAGE2_10MS; + Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE2_10MS; + } + + for( k = 0; k < length_d_srch; k++ ) { + d = d_srch[ k ]; + for( j = 0; j < nb_cbk_search; j++ ) { + CC[ j ] = 0; + for( i = 0; i < nb_subfr; i++ ) { + opus_int d_subfr; + /* Try all codebooks */ + d_subfr = d + matrix_ptr( Lag_CB_ptr, i, j, cbk_size ); + CC[ j ] = CC[ j ] + + (opus_int32)matrix_ptr( C, i, + d_subfr - ( MIN_LAG_8KHZ - 2 ), + CSTRIDE_8KHZ ); + } + } + /* Find best codebook */ + CCmax_new = silk_int32_MIN; + CBimax_new = 0; + for( i = 0; i < nb_cbk_search; i++ ) { + if( CC[ i ] > CCmax_new ) { + CCmax_new = CC[ i ]; + CBimax_new = i; + } + } + + /* Bias towards shorter lags */ + lag_log2_Q7 = silk_lin2log( d ); /* Q7 */ + silk_assert( lag_log2_Q7 == silk_SAT16( lag_log2_Q7 ) ); + silk_assert( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ) ) ); + CCmax_new_b = CCmax_new - silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ), lag_log2_Q7 ), 7 ); /* Q13 */ + + /* Bias towards previous lag */ + silk_assert( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ) ) ); + if( prevLag > 0 ) { + delta_lag_log2_sqr_Q7 = lag_log2_Q7 - prevLag_log2_Q7; + silk_assert( delta_lag_log2_sqr_Q7 == silk_SAT16( delta_lag_log2_sqr_Q7 ) ); + delta_lag_log2_sqr_Q7 = silk_RSHIFT( silk_SMULBB( delta_lag_log2_sqr_Q7, delta_lag_log2_sqr_Q7 ), 7 ); + prev_lag_bias_Q13 = silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ), *LTPCorr_Q15 ), 15 ); /* Q13 */ + prev_lag_bias_Q13 = silk_DIV32( silk_MUL( prev_lag_bias_Q13, delta_lag_log2_sqr_Q7 ), delta_lag_log2_sqr_Q7 + SILK_FIX_CONST( 0.5, 7 ) ); + CCmax_new_b -= prev_lag_bias_Q13; /* Q13 */ + } + + if( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */ + CCmax_new > silk_SMULBB( nb_subfr, search_thres2_Q13 ) && /* Correlation needs to be high enough to be voiced */ + silk_CB_lags_stage2[ 0 ][ CBimax_new ] <= MIN_LAG_8KHZ /* Lag must be in range */ + ) { + CCmax_b = CCmax_new_b; + CCmax = CCmax_new; + lag = d; + CBimax = CBimax_new; + } + } + + if( lag == -1 ) { + /* No suitable candidate found */ + silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); + *LTPCorr_Q15 = 0; + *lagIndex = 0; + *contourIndex = 0; + RESTORE_STACK; + return 1; + } + + /* Output normalized correlation */ + *LTPCorr_Q15 = (opus_int)silk_LSHIFT( silk_DIV32_16( CCmax, nb_subfr ), 2 ); + silk_assert( *LTPCorr_Q15 >= 0 ); + + if( Fs_kHz > 8 ) { + /* Search in original signal */ + + CBimax_old = CBimax; + /* Compensate for decimation */ + silk_assert( lag == silk_SAT16( lag ) ); + if( Fs_kHz == 12 ) { + lag = silk_RSHIFT( silk_SMULBB( lag, 3 ), 1 ); + } else if( Fs_kHz == 16 ) { + lag = silk_LSHIFT( lag, 1 ); + } else { + lag = silk_SMULBB( lag, 3 ); + } + + lag = silk_LIMIT_int( lag, min_lag, max_lag ); + start_lag = silk_max_int( lag - 2, min_lag ); + end_lag = silk_min_int( lag + 2, max_lag ); + lag_new = lag; /* to avoid undefined lag */ + CBimax = 0; /* to avoid undefined lag */ + + CCmax = silk_int32_MIN; + /* pitch lags according to second stage */ + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag + 2 * silk_CB_lags_stage2[ k ][ CBimax_old ]; + } + + /* Set up codebook parameters according to complexity setting and frame length */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + nb_cbk_search = (opus_int)silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + } else { + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + } + + /* Calculate the correlations and energies needed in stage 3 */ + ALLOC( energies_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals ); + ALLOC( cross_corr_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals ); + silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch ); + silk_P_Ana_calc_energy_st3( energies_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch ); + + lag_counter = 0; + silk_assert( lag == silk_SAT16( lag ) ); + contour_bias_Q15 = silk_DIV32_16( SILK_FIX_CONST( PE_FLATCONTOUR_BIAS, 15 ), lag ); + + target_ptr = &frame[ PE_LTP_MEM_LENGTH_MS * Fs_kHz ]; + energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, nb_subfr * sf_length, arch ), 1 ); + for( d = start_lag; d <= end_lag; d++ ) { + for( j = 0; j < nb_cbk_search; j++ ) { + cross_corr = 0; + energy = energy_target; + for( k = 0; k < nb_subfr; k++ ) { + cross_corr = silk_ADD32( cross_corr, + matrix_ptr( cross_corr_st3, k, j, + nb_cbk_search )[ lag_counter ] ); + energy = silk_ADD32( energy, + matrix_ptr( energies_st3, k, j, + nb_cbk_search )[ lag_counter ] ); + silk_assert( energy >= 0 ); + } + if( cross_corr > 0 ) { + CCmax_new = silk_DIV32_varQ( cross_corr, energy, 13 + 1 ); /* Q13 */ + /* Reduce depending on flatness of contour */ + diff = silk_int16_MAX - silk_MUL( contour_bias_Q15, j ); /* Q15 */ + silk_assert( diff == silk_SAT16( diff ) ); + CCmax_new = silk_SMULWB( CCmax_new, diff ); /* Q14 */ + } else { + CCmax_new = 0; + } + + if( CCmax_new > CCmax && ( d + silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag ) { + CCmax = CCmax_new; + lag_new = d; + CBimax = j; + } + } + lag_counter++; + } + + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag_new + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag, PE_MAX_LAG_MS * Fs_kHz ); + } + *lagIndex = (opus_int16)( lag_new - min_lag); + *contourIndex = (opus_int8)CBimax; + } else { /* Fs_kHz == 8 */ + /* Save Lags */ + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], MIN_LAG_8KHZ, PE_MAX_LAG_MS * 8 ); + } + *lagIndex = (opus_int16)( lag - MIN_LAG_8KHZ ); + *contourIndex = (opus_int8)CBimax; + } + celt_assert( *lagIndex >= 0 ); + /* return as voiced */ + RESTORE_STACK; + return 0; +} + +/*********************************************************************** + * Calculates the correlations used in stage 3 search. In order to cover + * the whole lag codebook for all the searched offset lags (lag +- 2), + * the following correlations are needed in each sub frame: + * + * sf1: lag range [-8,...,7] total 16 correlations + * sf2: lag range [-4,...,4] total 9 correlations + * sf3: lag range [-3,....4] total 8 correltions + * sf4: lag range [-6,....8] total 15 correlations + * + * In total 48 correlations. The direct implementation computed in worst + * case 4*12*5 = 240 correlations, but more likely around 120. + ***********************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_pe_stage3_vals cross_corr_st3[], /* O 3 DIM correlation array */ + const opus_int16 frame[], /* I vector to correlate */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of a 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +) +{ + const opus_int16 *target_ptr; + opus_int i, j, k, lag_counter, lag_low, lag_high; + opus_int nb_cbk_search, delta, idx, cbk_size; + VARDECL( opus_int32, scratch_mem ); + VARDECL( opus_int32, xcorr32 ); + const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + SAVE_STACK; + + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + ALLOC( scratch_mem, SCRATCH_SIZE, opus_int32 ); + ALLOC( xcorr32, SCRATCH_SIZE, opus_int32 ); + + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; /* Pointer to middle of frame */ + for( k = 0; k < nb_subfr; k++ ) { + lag_counter = 0; + + /* Calculate the correlations for each subframe */ + lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 ); + celt_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); + celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr32, sf_length, lag_high - lag_low + 1, arch ); + for( j = lag_low; j <= lag_high; j++ ) { + silk_assert( lag_counter < SCRATCH_SIZE ); + scratch_mem[ lag_counter ] = xcorr32[ lag_high - j ]; + lag_counter++; + } + + delta = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + for( i = 0; i < nb_cbk_search; i++ ) { + /* Fill out the 3 dim array that stores the correlations for */ + /* each code_book vector for each start lag */ + idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta; + for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { + silk_assert( idx + j < SCRATCH_SIZE ); + silk_assert( idx + j < lag_counter ); + matrix_ptr( cross_corr_st3, k, i, nb_cbk_search )[ j ] = + scratch_mem[ idx + j ]; + } + } + target_ptr += sf_length; + } + RESTORE_STACK; +} + +/********************************************************************/ +/* Calculate the energies for first two subframes. The energies are */ +/* calculated recursively. */ +/********************************************************************/ +static void silk_P_Ana_calc_energy_st3( + silk_pe_stage3_vals energies_st3[], /* O 3 DIM energy array */ + const opus_int16 frame[], /* I vector to calc energy in */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of one 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +) +{ + const opus_int16 *target_ptr, *basis_ptr; + opus_int32 energy; + opus_int k, i, j, lag_counter; + opus_int nb_cbk_search, delta, idx, cbk_size, lag_diff; + VARDECL( opus_int32, scratch_mem ); + const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + SAVE_STACK; + + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + ALLOC( scratch_mem, SCRATCH_SIZE, opus_int32 ); + + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; + for( k = 0; k < nb_subfr; k++ ) { + lag_counter = 0; + + /* Calculate the energy for first lag */ + basis_ptr = target_ptr - ( start_lag + matrix_ptr( Lag_range_ptr, k, 0, 2 ) ); + energy = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length, arch ); + silk_assert( energy >= 0 ); + scratch_mem[ lag_counter ] = energy; + lag_counter++; + + lag_diff = ( matrix_ptr( Lag_range_ptr, k, 1, 2 ) - matrix_ptr( Lag_range_ptr, k, 0, 2 ) + 1 ); + for( i = 1; i < lag_diff; i++ ) { + /* remove part outside new window */ + energy -= silk_SMULBB( basis_ptr[ sf_length - i ], basis_ptr[ sf_length - i ] ); + silk_assert( energy >= 0 ); + + /* add part that comes into window */ + energy = silk_ADD_SAT32( energy, silk_SMULBB( basis_ptr[ -i ], basis_ptr[ -i ] ) ); + silk_assert( energy >= 0 ); + silk_assert( lag_counter < SCRATCH_SIZE ); + scratch_mem[ lag_counter ] = energy; + lag_counter++; + } + + delta = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + for( i = 0; i < nb_cbk_search; i++ ) { + /* Fill out the 3 dim array that stores the correlations for */ + /* each code_book vector for each start lag */ + idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta; + for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { + silk_assert( idx + j < SCRATCH_SIZE ); + silk_assert( idx + j < lag_counter ); + matrix_ptr( energies_st3, k, i, nb_cbk_search )[ j ] = + scratch_mem[ idx + j ]; + silk_assert( + matrix_ptr( energies_st3, k, i, nb_cbk_search )[ j ] >= 0 ); + } + } + target_ptr += sf_length; + } + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/process_gains_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/process_gains_FIX.c new file mode 100644 index 000000000..05aba3178 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/process_gains_FIX.c @@ -0,0 +1,117 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "tuning_parameters.h" + +/* Processing of gains */ +void silk_process_gains_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + silk_shape_state_FIX *psShapeSt = &psEnc->sShape; + opus_int k; + opus_int32 s_Q16, InvMaxSqrVal_Q16, gain, gain_squared, ResNrg, ResNrgPart, quant_offset_Q10; + + /* Gain reduction when LTP coding gain is high */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /*s = -0.5f * silk_sigmoid( 0.25f * ( psEncCtrl->LTPredCodGain - 12.0f ) ); */ + s_Q16 = -silk_sigm_Q15( silk_RSHIFT_ROUND( psEncCtrl->LTPredCodGain_Q7 - SILK_FIX_CONST( 12.0, 7 ), 4 ) ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMLAWB( psEncCtrl->Gains_Q16[ k ], psEncCtrl->Gains_Q16[ k ], s_Q16 ); + } + } + + /* Limit the quantized signal */ + /* InvMaxSqrVal = pow( 2.0f, 0.33f * ( 21.0f - SNR_dB ) ) / subfr_length; */ + InvMaxSqrVal_Q16 = silk_DIV32_16( silk_log2lin( + silk_SMULWB( SILK_FIX_CONST( 21 + 16 / 0.33, 7 ) - psEnc->sCmn.SNR_dB_Q7, SILK_FIX_CONST( 0.33, 16 ) ) ), psEnc->sCmn.subfr_length ); + + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Soft limit on ratio residual energy and squared gains */ + ResNrg = psEncCtrl->ResNrg[ k ]; + ResNrgPart = silk_SMULWW( ResNrg, InvMaxSqrVal_Q16 ); + if( psEncCtrl->ResNrgQ[ k ] > 0 ) { + ResNrgPart = silk_RSHIFT_ROUND( ResNrgPart, psEncCtrl->ResNrgQ[ k ] ); + } else { + if( ResNrgPart >= silk_RSHIFT( silk_int32_MAX, -psEncCtrl->ResNrgQ[ k ] ) ) { + ResNrgPart = silk_int32_MAX; + } else { + ResNrgPart = silk_LSHIFT( ResNrgPart, -psEncCtrl->ResNrgQ[ k ] ); + } + } + gain = psEncCtrl->Gains_Q16[ k ]; + gain_squared = silk_ADD_SAT32( ResNrgPart, silk_SMMUL( gain, gain ) ); + if( gain_squared < silk_int16_MAX ) { + /* recalculate with higher precision */ + gain_squared = silk_SMLAWW( silk_LSHIFT( ResNrgPart, 16 ), gain, gain ); + silk_assert( gain_squared > 0 ); + gain = silk_SQRT_APPROX( gain_squared ); /* Q8 */ + gain = silk_min( gain, silk_int32_MAX >> 8 ); + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT_SAT32( gain, 8 ); /* Q16 */ + } else { + gain = silk_SQRT_APPROX( gain_squared ); /* Q0 */ + gain = silk_min( gain, silk_int32_MAX >> 16 ); + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT_SAT32( gain, 16 ); /* Q16 */ + } + } + + /* Save unquantized gains and gain Index */ + silk_memcpy( psEncCtrl->GainsUnq_Q16, psEncCtrl->Gains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + psEncCtrl->lastGainIndexPrev = psShapeSt->LastGainIndex; + + /* Quantize gains */ + silk_gains_quant( psEnc->sCmn.indices.GainsIndices, psEncCtrl->Gains_Q16, + &psShapeSt->LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Set quantizer offset for voiced signals. Larger offset when LTP coding gain is low or tilt is high (ie low-pass) */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + if( psEncCtrl->LTPredCodGain_Q7 + silk_RSHIFT( psEnc->sCmn.input_tilt_Q15, 8 ) > SILK_FIX_CONST( 1.0, 7 ) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + } + + /* Quantizer boundary adjustment */ + quant_offset_Q10 = silk_Quantization_Offsets_Q10[ psEnc->sCmn.indices.signalType >> 1 ][ psEnc->sCmn.indices.quantOffsetType ]; + psEncCtrl->Lambda_Q10 = SILK_FIX_CONST( LAMBDA_OFFSET, 10 ) + + silk_SMULBB( SILK_FIX_CONST( LAMBDA_DELAYED_DECISIONS, 10 ), psEnc->sCmn.nStatesDelayedDecision ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_SPEECH_ACT, 18 ), psEnc->sCmn.speech_activity_Q8 ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_INPUT_QUALITY, 12 ), psEncCtrl->input_quality_Q14 ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_CODING_QUALITY, 12 ), psEncCtrl->coding_quality_Q14 ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_QUANT_OFFSET, 16 ), quant_offset_Q10 ); + + silk_assert( psEncCtrl->Lambda_Q10 > 0 ); + silk_assert( psEncCtrl->Lambda_Q10 < SILK_FIX_CONST( 2, 10 ) ); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/regularize_correlations_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/regularize_correlations_FIX.c new file mode 100644 index 000000000..a2836b05f --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/regularize_correlations_FIX.c @@ -0,0 +1,47 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +/* Add noise to matrix diagonal */ +void silk_regularize_correlations_FIX( + opus_int32 *XX, /* I/O Correlation matrices */ + opus_int32 *xx, /* I/O Correlation values */ + opus_int32 noise, /* I Noise to add */ + opus_int D /* I Dimension of XX */ +) +{ + opus_int i; + for( i = 0; i < D; i++ ) { + matrix_ptr( &XX[ 0 ], i, i, D ) = silk_ADD32( matrix_ptr( &XX[ 0 ], i, i, D ), noise ); + } + xx[ 0 ] += noise; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy16_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy16_FIX.c new file mode 100644 index 000000000..7f130f3d3 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy16_FIX.c @@ -0,0 +1,103 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +opus_int32 silk_residual_energy16_covar_FIX( + const opus_int16 *c, /* I Prediction vector */ + const opus_int32 *wXX, /* I Correlation matrix */ + const opus_int32 *wXx, /* I Correlation vector */ + opus_int32 wxx, /* I Signal energy */ + opus_int D, /* I Dimension */ + opus_int cQ /* I Q value for c vector 0 - 15 */ +) +{ + opus_int i, j, lshifts, Qxtra; + opus_int32 c_max, w_max, tmp, tmp2, nrg; + opus_int cn[ MAX_MATRIX_SIZE ]; + const opus_int32 *pRow; + + /* Safety checks */ + celt_assert( D >= 0 ); + celt_assert( D <= 16 ); + celt_assert( cQ > 0 ); + celt_assert( cQ < 16 ); + + lshifts = 16 - cQ; + Qxtra = lshifts; + + c_max = 0; + for( i = 0; i < D; i++ ) { + c_max = silk_max_32( c_max, silk_abs( (opus_int32)c[ i ] ) ); + } + Qxtra = silk_min_int( Qxtra, silk_CLZ32( c_max ) - 17 ); + + w_max = silk_max_32( wXX[ 0 ], wXX[ D * D - 1 ] ); + Qxtra = silk_min_int( Qxtra, silk_CLZ32( silk_MUL( D, silk_RSHIFT( silk_SMULWB( w_max, c_max ), 4 ) ) ) - 5 ); + Qxtra = silk_max_int( Qxtra, 0 ); + for( i = 0; i < D; i++ ) { + cn[ i ] = silk_LSHIFT( ( opus_int )c[ i ], Qxtra ); + silk_assert( silk_abs(cn[i]) <= ( silk_int16_MAX + 1 ) ); /* Check that silk_SMLAWB can be used */ + } + lshifts -= Qxtra; + + /* Compute wxx - 2 * wXx * c */ + tmp = 0; + for( i = 0; i < D; i++ ) { + tmp = silk_SMLAWB( tmp, wXx[ i ], cn[ i ] ); + } + nrg = silk_RSHIFT( wxx, 1 + lshifts ) - tmp; /* Q: -lshifts - 1 */ + + /* Add c' * wXX * c, assuming wXX is symmetric */ + tmp2 = 0; + for( i = 0; i < D; i++ ) { + tmp = 0; + pRow = &wXX[ i * D ]; + for( j = i + 1; j < D; j++ ) { + tmp = silk_SMLAWB( tmp, pRow[ j ], cn[ j ] ); + } + tmp = silk_SMLAWB( tmp, silk_RSHIFT( pRow[ i ], 1 ), cn[ i ] ); + tmp2 = silk_SMLAWB( tmp2, tmp, cn[ i ] ); + } + nrg = silk_ADD_LSHIFT32( nrg, tmp2, lshifts ); /* Q: -lshifts - 1 */ + + /* Keep one bit free always, because we add them for LSF interpolation */ + if( nrg < 1 ) { + nrg = 1; + } else if( nrg > silk_RSHIFT( silk_int32_MAX, lshifts + 2 ) ) { + nrg = silk_int32_MAX >> 1; + } else { + nrg = silk_LSHIFT( nrg, lshifts + 1 ); /* Q0 */ + } + return nrg; + +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy_FIX.c new file mode 100644 index 000000000..6c7cade9a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/residual_energy_FIX.c @@ -0,0 +1,98 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FIX( + opus_int32 nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + opus_int nrgsQ[ MAX_NB_SUBFR ], /* O Q value per subframe */ + const opus_int16 x[], /* I Input signal */ + opus_int16 a_Q12[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const opus_int32 gains[ MAX_NB_SUBFR ], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +) +{ + opus_int offset, i, j, rshift, lz1, lz2; + opus_int16 *LPC_res_ptr; + VARDECL( opus_int16, LPC_res ); + const opus_int16 *x_ptr; + opus_int32 tmp32; + SAVE_STACK; + + x_ptr = x; + offset = LPC_order + subfr_length; + + /* Filter input to create the LPC residual for each frame half, and measure subframe energies */ + ALLOC( LPC_res, ( MAX_NB_SUBFR >> 1 ) * offset, opus_int16 ); + celt_assert( ( nb_subfr >> 1 ) * ( MAX_NB_SUBFR >> 1 ) == nb_subfr ); + for( i = 0; i < nb_subfr >> 1; i++ ) { + /* Calculate half frame LPC residual signal including preceding samples */ + silk_LPC_analysis_filter( LPC_res, x_ptr, a_Q12[ i ], ( MAX_NB_SUBFR >> 1 ) * offset, LPC_order, arch ); + + /* Point to first subframe of the just calculated LPC residual signal */ + LPC_res_ptr = LPC_res + LPC_order; + for( j = 0; j < ( MAX_NB_SUBFR >> 1 ); j++ ) { + /* Measure subframe energy */ + silk_sum_sqr_shift( &nrgs[ i * ( MAX_NB_SUBFR >> 1 ) + j ], &rshift, LPC_res_ptr, subfr_length ); + + /* Set Q values for the measured energy */ + nrgsQ[ i * ( MAX_NB_SUBFR >> 1 ) + j ] = -rshift; + + /* Move to next subframe */ + LPC_res_ptr += offset; + } + /* Move to next frame half */ + x_ptr += ( MAX_NB_SUBFR >> 1 ) * offset; + } + + /* Apply the squared subframe gains */ + for( i = 0; i < nb_subfr; i++ ) { + /* Fully upscale gains and energies */ + lz1 = silk_CLZ32( nrgs[ i ] ) - 1; + lz2 = silk_CLZ32( gains[ i ] ) - 1; + + tmp32 = silk_LSHIFT32( gains[ i ], lz2 ); + + /* Find squared gains */ + tmp32 = silk_SMMUL( tmp32, tmp32 ); /* Q( 2 * lz2 - 32 )*/ + + /* Scale energies */ + nrgs[ i ] = silk_SMMUL( tmp32, silk_LSHIFT32( nrgs[ i ], lz1 ) ); /* Q( nrgsQ[ i ] + lz1 + 2 * lz2 - 32 - 32 )*/ + nrgsQ[ i ] += lz1 + 2 * lz2 - 32 - 32; + } + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur64_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur64_FIX.c new file mode 100644 index 000000000..4b7e19ea5 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur64_FIX.c @@ -0,0 +1,93 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Slower than schur(), but more accurate. */ +/* Uses SMULL(), available on armv4 */ +opus_int32 silk_schur64( /* O returns residual energy */ + opus_int32 rc_Q16[], /* O Reflection coefficients [order] Q16 */ + const opus_int32 c[], /* I Correlations [order+1] */ + opus_int32 order /* I Prediction order */ +) +{ + opus_int k, n; + opus_int32 C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; + opus_int32 Ctmp1_Q30, Ctmp2_Q30, rc_tmp_Q31; + + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + + /* Check for invalid input */ + if( c[ 0 ] <= 0 ) { + silk_memset( rc_Q16, 0, order * sizeof( opus_int32 ) ); + return 0; + } + + k = 0; + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = c[ k ]; + } while( ++k <= order ); + + for( k = 0; k < order; k++ ) { + /* Check that we won't be getting an unstable rc, otherwise stop here. */ + if (silk_abs_int32(C[ k + 1 ][ 0 ]) >= C[ 0 ][ 1 ]) { + if ( C[ k + 1 ][ 0 ] > 0 ) { + rc_Q16[ k ] = -SILK_FIX_CONST( .99f, 16 ); + } else { + rc_Q16[ k ] = SILK_FIX_CONST( .99f, 16 ); + } + k++; + break; + } + + /* Get reflection coefficient: divide two Q30 values and get result in Q31 */ + rc_tmp_Q31 = silk_DIV32_varQ( -C[ k + 1 ][ 0 ], C[ 0 ][ 1 ], 31 ); + + /* Save the output */ + rc_Q16[ k ] = silk_RSHIFT_ROUND( rc_tmp_Q31, 15 ); + + /* Update correlations */ + for( n = 0; n < order - k; n++ ) { + Ctmp1_Q30 = C[ n + k + 1 ][ 0 ]; + Ctmp2_Q30 = C[ n ][ 1 ]; + + /* Multiply and add the highest int32 */ + C[ n + k + 1 ][ 0 ] = Ctmp1_Q30 + silk_SMMUL( silk_LSHIFT( Ctmp2_Q30, 1 ), rc_tmp_Q31 ); + C[ n ][ 1 ] = Ctmp2_Q30 + silk_SMMUL( silk_LSHIFT( Ctmp1_Q30, 1 ), rc_tmp_Q31 ); + } + } + + for(; k < order; k++ ) { + rc_Q16[ k ] = 0; + } + + return silk_max_32( 1, C[ 0 ][ 1 ] ); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur_FIX.c new file mode 100644 index 000000000..2840f6b1a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/schur_FIX.c @@ -0,0 +1,107 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Faster than schur64(), but much less accurate. */ +/* uses SMLAWB(), requiring armv5E and higher. */ +opus_int32 silk_schur( /* O Returns residual energy */ + opus_int16 *rc_Q15, /* O reflection coefficients [order] Q15 */ + const opus_int32 *c, /* I correlations [order+1] */ + const opus_int32 order /* I prediction order */ +) +{ + opus_int k, n, lz; + opus_int32 C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; + opus_int32 Ctmp1, Ctmp2, rc_tmp_Q15; + + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + + /* Get number of leading zeros */ + lz = silk_CLZ32( c[ 0 ] ); + + /* Copy correlations and adjust level to Q30 */ + k = 0; + if( lz < 2 ) { + /* lz must be 1, so shift one to the right */ + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = silk_RSHIFT( c[ k ], 1 ); + } while( ++k <= order ); + } else if( lz > 2 ) { + /* Shift to the left */ + lz -= 2; + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = silk_LSHIFT( c[ k ], lz ); + } while( ++k <= order ); + } else { + /* No need to shift */ + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = c[ k ]; + } while( ++k <= order ); + } + + for( k = 0; k < order; k++ ) { + /* Check that we won't be getting an unstable rc, otherwise stop here. */ + if (silk_abs_int32(C[ k + 1 ][ 0 ]) >= C[ 0 ][ 1 ]) { + if ( C[ k + 1 ][ 0 ] > 0 ) { + rc_Q15[ k ] = -SILK_FIX_CONST( .99f, 15 ); + } else { + rc_Q15[ k ] = SILK_FIX_CONST( .99f, 15 ); + } + k++; + break; + } + + /* Get reflection coefficient */ + rc_tmp_Q15 = -silk_DIV32_16( C[ k + 1 ][ 0 ], silk_max_32( silk_RSHIFT( C[ 0 ][ 1 ], 15 ), 1 ) ); + + /* Clip (shouldn't happen for properly conditioned inputs) */ + rc_tmp_Q15 = silk_SAT16( rc_tmp_Q15 ); + + /* Store */ + rc_Q15[ k ] = (opus_int16)rc_tmp_Q15; + + /* Update correlations */ + for( n = 0; n < order - k; n++ ) { + Ctmp1 = C[ n + k + 1 ][ 0 ]; + Ctmp2 = C[ n ][ 1 ]; + C[ n + k + 1 ][ 0 ] = silk_SMLAWB( Ctmp1, silk_LSHIFT( Ctmp2, 1 ), rc_tmp_Q15 ); + C[ n ][ 1 ] = silk_SMLAWB( Ctmp2, silk_LSHIFT( Ctmp1, 1 ), rc_tmp_Q15 ); + } + } + + for(; k < order; k++ ) { + rc_Q15[ k ] = 0; + } + + /* return residual energy */ + return silk_max_32( 1, C[ 0 ][ 1 ] ); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/structs_FIX.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/structs_FIX.h new file mode 100644 index 000000000..42920d61b --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/structs_FIX.h @@ -0,0 +1,109 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_STRUCTS_FIX_H +#define SILK_STRUCTS_FIX_H + +#include "typedef.h" +#include "main.h" +#include "structs.h" + + +/********************************/ +/* Noise shaping analysis state */ +/********************************/ +typedef struct { + opus_int8 LastGainIndex; + opus_int32 HarmBoost_smth_Q16; + opus_int32 HarmShapeGain_smth_Q16; + opus_int32 Tilt_smth_Q16; +} silk_shape_state_FIX; + +/********************************/ +/* Encoder state FIX */ +/********************************/ +typedef struct { + silk_encoder_state sCmn; /* Common struct, shared with floating-point code */ + silk_shape_state_FIX sShape; /* Shape state */ + + /* Buffer for find pitch and noise shape analysis */ + silk_DWORD_ALIGN opus_int16 x_buf[ 2 * MAX_FRAME_LENGTH + LA_SHAPE_MAX ];/* Buffer for find pitch and noise shape analysis */ + opus_int LTPCorr_Q15; /* Normalized correlation from pitch lag estimator */ + opus_int32 resNrgSmth; +} silk_encoder_state_FIX; + +/************************/ +/* Encoder control FIX */ +/************************/ +typedef struct { + /* Prediction and coding parameters */ + opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; + silk_DWORD_ALIGN opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ]; + opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ]; + opus_int LTP_scale_Q14; + opus_int pitchL[ MAX_NB_SUBFR ]; + + /* Noise shaping parameters */ + /* Testing */ + silk_DWORD_ALIGN opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ]; + opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ]; /* Packs two int16 coefficients per int32 value */ + opus_int Tilt_Q14[ MAX_NB_SUBFR ]; + opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ]; + opus_int Lambda_Q10; + opus_int input_quality_Q14; + opus_int coding_quality_Q14; + + /* measures */ + opus_int32 predGain_Q16; + opus_int LTPredCodGain_Q7; + opus_int32 ResNrg[ MAX_NB_SUBFR ]; /* Residual energy per subframe */ + opus_int ResNrgQ[ MAX_NB_SUBFR ]; /* Q domain for the residual energy > 0 */ + + /* Parameters for CBR mode */ + opus_int32 GainsUnq_Q16[ MAX_NB_SUBFR ]; + opus_int8 lastGainIndexPrev; +} silk_encoder_control_FIX; + +/************************/ +/* Encoder Super Struct */ +/************************/ +typedef struct { + stereo_enc_state sStereo; + opus_int32 nBitsUsedLBRR; + opus_int32 nBitsExceeded; + opus_int nChannelsAPI; + opus_int nChannelsInternal; + opus_int nPrevChannelsInternal; + opus_int timeSinceSwitchAllowed_ms; + opus_int allowBandwidthSwitch; + opus_int prev_decode_only_middle; + /* This needs to be last so we can skip the second state for mono. */ + silk_encoder_state_FIX state_Fxx[ ENCODER_NUM_CHANNELS ]; +} silk_encoder; + + +#endif diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX.c new file mode 100644 index 000000000..dcf84070a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX.c @@ -0,0 +1,102 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "pitch.h" + +/* Copy and multiply a vector by a constant */ +void silk_scale_copy_vector16( + opus_int16 *data_out, + const opus_int16 *data_in, + opus_int32 gain_Q16, /* I Gain in Q16 */ + const opus_int dataSize /* I Length */ +) +{ + opus_int i; + opus_int32 tmp32; + + for( i = 0; i < dataSize; i++ ) { + tmp32 = silk_SMULWB( gain_Q16, data_in[ i ] ); + data_out[ i ] = (opus_int16)silk_CHECK_FIT16( tmp32 ); + } +} + +/* Multiply a vector by a constant */ +void silk_scale_vector32_Q26_lshift_18( + opus_int32 *data1, /* I/O Q0/Q18 */ + opus_int32 gain_Q26, /* I Q26 */ + opus_int dataSize /* I length */ +) +{ + opus_int i; + + for( i = 0; i < dataSize; i++ ) { + data1[ i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( silk_SMULL( data1[ i ], gain_Q26 ), 8 ) ); /* OUTPUT: Q18 */ + } +} + +/* sum = for(i=0;i6, memory access can be reduced by half. */ +opus_int32 silk_inner_prod_aligned( + const opus_int16 *const inVec1, /* I input vector 1 */ + const opus_int16 *const inVec2, /* I input vector 2 */ + const opus_int len, /* I vector lengths */ + int arch /* I Run-time architecture */ +) +{ +#ifdef FIXED_POINT + return celt_inner_prod(inVec1, inVec2, len, arch); +#else + opus_int i; + opus_int32 sum = 0; + for( i = 0; i < len; i++ ) { + sum = silk_SMLABB( sum, inVec1[ i ], inVec2[ i ] ); + } + return sum; +#endif +} + +opus_int64 silk_inner_prod16_c( + const opus_int16 *inVec1, /* I input vector 1 */ + const opus_int16 *inVec2, /* I input vector 2 */ + const opus_int len /* I vector lengths */ +) +{ + opus_int i; + opus_int64 sum = 0; + for( i = 0; i < len; i++ ) { + sum = silk_SMLALBB( sum, inVec1[ i ], inVec2[ i ] ); + } + return sum; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX_sse4_1.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX_sse4_1.c new file mode 100644 index 000000000..deea85791 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/vector_ops_FIX_sse4_1.c @@ -0,0 +1,93 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include +#include "main.h" + +#include "SigProc_FIX.h" +#include "pitch.h" +#include "celt/x86/x86cpu.h" + +opus_int64 silk_inner_prod16_sse4_1( + const opus_int16 *inVec1, /* I input vector 1 */ + const opus_int16 *inVec2, /* I input vector 2 */ + const opus_int len /* I vector lengths */ +) +{ + opus_int i, dataSize4; + opus_int64 sum; + + __m128i xmm_prod_20, xmm_prod_31; + __m128i inVec1_3210, acc1; + __m128i inVec2_3210, acc2; + + sum = 0; + dataSize4 = len & ~3; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for( i = 0; i < dataSize4; i += 4 ) { + inVec1_3210 = OP_CVTEPI16_EPI32_M64( &inVec1[i + 0] ); + inVec2_3210 = OP_CVTEPI16_EPI32_M64( &inVec2[i + 0] ); + xmm_prod_20 = _mm_mul_epi32( inVec1_3210, inVec2_3210 ); + + inVec1_3210 = _mm_shuffle_epi32( inVec1_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + inVec2_3210 = _mm_shuffle_epi32( inVec2_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_prod_31 = _mm_mul_epi32( inVec1_3210, inVec2_3210 ); + + acc1 = _mm_add_epi64( acc1, xmm_prod_20 ); + acc2 = _mm_add_epi64( acc2, xmm_prod_31 ); + } + + acc1 = _mm_add_epi64( acc1, acc2 ); + + /* equal shift right 8 bytes */ + acc2 = _mm_shuffle_epi32( acc1, _MM_SHUFFLE( 0, 0, 3, 2 ) ); + acc1 = _mm_add_epi64( acc1, acc2 ); + + _mm_storel_epi64( (__m128i *)(void *)&sum, acc1 ); + + for( ; i < len; i++ ) { + sum = silk_SMLALBB( sum, inVec1[ i ], inVec2[ i ] ); + } + +#ifdef OPUS_CHECK_ASM + { + opus_int64 sum_c = silk_inner_prod16_c( inVec1, inVec2, len ); + silk_assert( sum == sum_c ); + } +#endif + + return sum; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX.c new file mode 100644 index 000000000..00bd9ff64 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX.c @@ -0,0 +1,92 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +#if defined(__mips) +#include "mips/warped_autocorrelation_FIX_mipsr1.h" +#endif + + +/* Autocorrelations for a warped frequency axis */ +#ifndef OVERRIDE_silk_warped_autocorrelation_FIX_c +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + opus_int n, i, lsh; + opus_int32 tmp1_QS, tmp2_QS; + opus_int32 state_QS[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + + /* Order must be even */ + celt_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* Loop over samples */ + for( n = 0; n < length; n++ ) { + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + state_QS[ i ] = tmp1_QS; + corr_QC[ i ] += silk_RSHIFT64( silk_SMULL( tmp1_QS, state_QS[ 0 ] ), 2 * QS - QC ); + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp2_QS; + corr_QC[ i + 1 ] += silk_RSHIFT64( silk_SMULL( tmp2_QS, state_QS[ 0 ] ), 2 * QS - QC ); + } + state_QS[ order ] = tmp1_QS; + corr_QC[ order ] += silk_RSHIFT64( silk_SMULL( tmp1_QS, state_QS[ 0 ] ), 2 * QS - QC ); + } + + lsh = silk_CLZ64( corr_QC[ 0 ] ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + if( lsh >= 0 ) { + for( i = 0; i < order + 1; i++ ) { + corr[ i ] = (opus_int32)silk_CHECK_FIT32( silk_LSHIFT64( corr_QC[ i ], lsh ) ); + } + } else { + for( i = 0; i < order + 1; i++ ) { + corr[ i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr_QC[ i ], -lsh ) ); + } + } + silk_assert( corr_QC[ 0 ] >= 0 ); /* If breaking, decrease QC*/ +} +#endif /* OVERRIDE_silk_warped_autocorrelation_FIX_c */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_arm.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_arm.h new file mode 100644 index 000000000..1992e4328 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_arm.h @@ -0,0 +1,68 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_WARPED_AUTOCORRELATION_FIX_ARM_H +# define SILK_WARPED_AUTOCORRELATION_FIX_ARM_H + +# include "celt/arm/armcpu.h" + +# if defined(FIXED_POINT) + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void silk_warped_autocorrelation_FIX_neon( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), PRESUME_NEON(silk_warped_autocorrelation_FIX)(corr, scale, input, warping_Q16, length, order)) +# endif +# endif + +# if !defined(OVERRIDE_silk_warped_autocorrelation_FIX) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const SILK_WARPED_AUTOCORRELATION_FIX_IMPL[OPUS_ARCHMASK+1])(opus_int32*, opus_int*, const opus_int16*, const opus_int, const opus_int, const opus_int); +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((*SILK_WARPED_AUTOCORRELATION_FIX_IMPL[(arch)&OPUS_ARCHMASK])(corr, scale, input, warping_Q16, length, order)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), silk_warped_autocorrelation_FIX_neon(corr, scale, input, warping_Q16, length, order)) +# endif +# endif + +# endif /* end FIXED_POINT */ + +#endif /* end SILK_WARPED_AUTOCORRELATION_FIX_ARM_H */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_mipsr1.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_mipsr1.h new file mode 100644 index 000000000..9d9cb0551 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_mipsr1.h @@ -0,0 +1,189 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ +#define __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +#if defined (__mips_dsp) && __mips == 32 + +#define MIPS_MADD __builtin_mips_madd +#define MIPS_SHILO __builtin_mips_shilo + +#define OVERRIDE_silk_warped_autocorrelation_FIX_c + +#elif defined (__mips_isa_rev) && __mips == 32 + +static inline long long MIPS_MADD(long long acc, int a, int b) { + return acc + (long long)a * b; +} + +static inline long long MIPS_SHILO(long long acc, int sh) { + return (sh >= 0) ? (acc >> sh) : (acc << -sh); +} + +#define OVERRIDE_silk_warped_autocorrelation_FIX_c + +#endif + +/* Autocorrelations for a warped frequency axis */ +#if defined (OVERRIDE_silk_warped_autocorrelation_FIX_c) + +#undef QC +#define QC 10 + +#undef QS +#define QS 14 + +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + opus_int n, i, lsh; + opus_int32 tmp1_QS=0, tmp2_QS=0, tmp3_QS=0, tmp4_QS=0, tmp5_QS=0, tmp6_QS=0, tmp7_QS=0, tmp8_QS=0, start_1=0, start_2=0, start_3=0; + opus_int32 state_QS[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 temp64; + + opus_int32 val; + val = 2 * QS - QC; + + /* Order must be even */ + silk_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* Loop over samples */ + for( n = 0; n < length; n=n+4 ) { + + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + start_1 = tmp1_QS; + tmp3_QS = silk_LSHIFT32( (opus_int32)input[ n+1], QS ); + start_2 = tmp3_QS; + tmp5_QS = silk_LSHIFT32( (opus_int32)input[ n+2], QS ); + start_3 = tmp5_QS; + tmp7_QS = silk_LSHIFT32( (opus_int32)input[ n+3], QS ); + + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + corr_QC[ i ] = MIPS_MADD( corr_QC[ i ], tmp1_QS, start_1); + + tmp4_QS = silk_SMLAWB( tmp1_QS, tmp2_QS - tmp3_QS, warping_Q16 ); + corr_QC[ i ] = MIPS_MADD( corr_QC[ i ], tmp3_QS, start_2); + + tmp6_QS = silk_SMLAWB( tmp3_QS, tmp4_QS - tmp5_QS, warping_Q16 ); + corr_QC[ i ] = MIPS_MADD( corr_QC[ i ], tmp5_QS, start_3); + + tmp8_QS = silk_SMLAWB( tmp5_QS, tmp6_QS - tmp7_QS, warping_Q16 ); + state_QS[ i ] = tmp7_QS; + corr_QC[ i ] = MIPS_MADD( corr_QC[ i ], tmp7_QS, state_QS[0]); + + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + corr_QC[ i+1 ] = MIPS_MADD( corr_QC[ i+1 ], tmp2_QS, start_1); + + tmp3_QS = silk_SMLAWB( tmp2_QS, tmp1_QS - tmp4_QS, warping_Q16 ); + corr_QC[ i+1 ] = MIPS_MADD( corr_QC[ i+1 ], tmp4_QS, start_2); + + tmp5_QS = silk_SMLAWB( tmp4_QS, tmp3_QS - tmp6_QS, warping_Q16 ); + corr_QC[ i+1 ] = MIPS_MADD( corr_QC[ i+1 ], tmp6_QS, start_3); + + tmp7_QS = silk_SMLAWB( tmp6_QS, tmp5_QS - tmp8_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp8_QS; + corr_QC[ i+1 ] = MIPS_MADD( corr_QC[ i+1 ], tmp8_QS, state_QS[ 0 ]); + + } + state_QS[ order ] = tmp7_QS; + + corr_QC[ order ] = MIPS_MADD( corr_QC[ order ], tmp1_QS, start_1); + corr_QC[ order ] = MIPS_MADD( corr_QC[ order ], tmp3_QS, start_2); + corr_QC[ order ] = MIPS_MADD( corr_QC[ order ], tmp5_QS, start_3); + corr_QC[ order ] = MIPS_MADD( corr_QC[ order ], tmp7_QS, state_QS[ 0 ]); + } + + for(;n< length; n++ ) { + + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + state_QS[ i ] = tmp1_QS; + corr_QC[ i ] = MIPS_MADD( corr_QC[ i ], tmp1_QS, state_QS[ 0 ]); + + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp2_QS; + corr_QC[ i+1 ] = MIPS_MADD( corr_QC[ i+1 ], tmp2_QS, state_QS[ 0 ]); + } + state_QS[ order ] = tmp1_QS; + corr_QC[ order ] = MIPS_MADD( corr_QC[ order ], tmp1_QS, state_QS[ 0 ]); + } + + temp64 = corr_QC[ 0 ]; + temp64 = MIPS_SHILO(temp64, val); + + lsh = silk_CLZ64( temp64 ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + if( lsh >= 0 ) { + for( i = 0; i < order + 1; i++ ) { + temp64 = corr_QC[ i ]; + //temp64 = MIPS_SHILO(temp64, val); + temp64 = (val >= 0) ? (temp64 >> val) : (temp64 << -val); + corr[ i ] = (opus_int32)silk_CHECK_FIT32( MIPS_SHILO( temp64, -lsh ) ); + } + } else { + for( i = 0; i < order + 1; i++ ) { + temp64 = corr_QC[ i ]; + //temp64 = MIPS_SHILO(temp64, val); + temp64 = (val >= 0) ? (temp64 >> val) : (temp64 << -val); + corr[ i ] = (opus_int32)silk_CHECK_FIT32( MIPS_SHILO( temp64, -lsh ) ); + } + } + + corr_QC[ 0 ] = MIPS_SHILO(corr_QC[ 0 ], val); + + silk_assert( corr_QC[ 0 ] >= 0 ); /* If breaking, decrease QC*/ +} +#endif /* OVERRIDE_silk_warped_autocorrelation_FIX_c */ + +#endif /* __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_neon_intr.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_neon_intr.c new file mode 100644 index 000000000..6f3be025c --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/fixed/warped_autocorrelation_FIX_neon_intr.c @@ -0,0 +1,265 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc., Jean-Marc Valin +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#ifdef OPUS_CHECK_ASM +# include +#endif +#include "stack_alloc.h" +#include "main_FIX.h" + +static OPUS_INLINE void calc_corr( const opus_int32 *const input_QS, opus_int64 *const corr_QC, const opus_int offset, const int32x4_t state_QS_s32x4 ) +{ + int64x2_t corr_QC_s64x2[ 2 ], t_s64x2[ 2 ]; + const int32x4_t input_QS_s32x4 = vld1q_s32( input_QS + offset ); + corr_QC_s64x2[ 0 ] = vld1q_s64( corr_QC + offset + 0 ); + corr_QC_s64x2[ 1 ] = vld1q_s64( corr_QC + offset + 2 ); + t_s64x2[ 0 ] = vmull_s32( vget_low_s32( state_QS_s32x4 ), vget_low_s32( input_QS_s32x4 ) ); + t_s64x2[ 1 ] = vmull_s32( vget_high_s32( state_QS_s32x4 ), vget_high_s32( input_QS_s32x4 ) ); + corr_QC_s64x2[ 0 ] = vsraq_n_s64( corr_QC_s64x2[ 0 ], t_s64x2[ 0 ], 2 * QS - QC ); + corr_QC_s64x2[ 1 ] = vsraq_n_s64( corr_QC_s64x2[ 1 ], t_s64x2[ 1 ], 2 * QS - QC ); + vst1q_s64( corr_QC + offset + 0, corr_QC_s64x2[ 0 ] ); + vst1q_s64( corr_QC + offset + 2, corr_QC_s64x2[ 1 ] ); +} + +static OPUS_INLINE int32x4_t calc_state( const int32x4_t state_QS0_s32x4, const int32x4_t state_QS0_1_s32x4, const int32x4_t state_QS1_1_s32x4, const int32x4_t warping_Q16_s32x4 ) +{ + int32x4_t t_s32x4 = vsubq_s32( state_QS0_s32x4, state_QS0_1_s32x4 ); + t_s32x4 = vqdmulhq_s32( t_s32x4, warping_Q16_s32x4 ); + return vaddq_s32( state_QS1_1_s32x4, t_s32x4 ); +} + +void silk_warped_autocorrelation_FIX_neon( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + if( ( MAX_SHAPE_LPC_ORDER > 24 ) || ( order < 6 ) ) { + silk_warped_autocorrelation_FIX_c( corr, scale, input, warping_Q16, length, order ); + } else { + opus_int n, i, lsh; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; /* In reverse order */ + opus_int64 corr_QC_orderT; + int64x2_t lsh_s64x2; + const opus_int orderT = ( order + 3 ) & ~3; + opus_int64 *corr_QCT; + opus_int32 *input_QS; + VARDECL( opus_int32, input_QST ); + VARDECL( opus_int32, state ); + SAVE_STACK; + + /* Order must be even */ + silk_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* The additional +4 is to ensure a later vld1q_s32 call does not overflow. */ + /* Strictly, only +3 is needed but +4 simplifies initialization using the 4x32 neon load. */ + ALLOC( input_QST, length + 2 * MAX_SHAPE_LPC_ORDER + 4, opus_int32 ); + + input_QS = input_QST; + /* input_QS has zero paddings in the beginning and end. */ + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + + /* Loop over samples */ + for( n = 0; n < length - 7; n += 8, input_QS += 8 ) { + const int16x8_t t0_s16x4 = vld1q_s16( input + n ); + vst1q_s32( input_QS + 0, vshll_n_s16( vget_low_s16( t0_s16x4 ), QS ) ); + vst1q_s32( input_QS + 4, vshll_n_s16( vget_high_s16( t0_s16x4 ), QS ) ); + } + for( ; n < length; n++, input_QS++ ) { + input_QS[ 0 ] = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + } + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS = input_QST + MAX_SHAPE_LPC_ORDER - orderT; + + /* The following loop runs ( length + order ) times, with ( order ) extra epilogues. */ + /* The zero paddings in input_QS guarantee corr_QC's correctness even with the extra epilogues. */ + /* The values of state_QS will be polluted by the extra epilogues, however they are temporary values. */ + + /* Keep the C code here to help understand the intrinsics optimization. */ + /* + { + opus_int32 state_QS[ 2 ][ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int32 *state_QST[ 3 ]; + state_QST[ 0 ] = state_QS[ 0 ]; + state_QST[ 1 ] = state_QS[ 1 ]; + for( n = 0; n < length + order; n++, input_QS++ ) { + state_QST[ 0 ][ orderT ] = input_QS[ orderT ]; + for( i = 0; i < orderT; i++ ) { + corr_QC[ i ] += silk_RSHIFT64( silk_SMULL( state_QST[ 0 ][ i ], input_QS[ i ] ), 2 * QS - QC ); + state_QST[ 1 ][ i ] = silk_SMLAWB( state_QST[ 1 ][ i + 1 ], state_QST[ 0 ][ i ] - state_QST[ 0 ][ i + 1 ], warping_Q16 ); + } + state_QST[ 2 ] = state_QST[ 0 ]; + state_QST[ 0 ] = state_QST[ 1 ]; + state_QST[ 1 ] = state_QST[ 2 ]; + } + } + */ + + { + const int32x4_t warping_Q16_s32x4 = vdupq_n_s32( warping_Q16 << 15 ); + const opus_int32 *in = input_QS + orderT; + opus_int o = orderT; + int32x4_t state_QS_s32x4[ 3 ][ 2 ]; + + /* The additional +4 is to ensure a later vld1q_s32 call does not overflow. */ + ALLOC( state, length + order + 4, opus_int32 ); + state_QS_s32x4[ 2 ][ 1 ] = vdupq_n_s32( 0 ); + + /* Calculate 8 taps of all inputs in each loop. */ + do { + state_QS_s32x4[ 0 ][ 0 ] = state_QS_s32x4[ 0 ][ 1 ] = + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 1 ][ 1 ] = vdupq_n_s32( 0 ); + n = 0; + do { + calc_corr( input_QS + n, corr_QC, o - 8, state_QS_s32x4[ 0 ][ 0 ] ); + calc_corr( input_QS + n, corr_QC, o - 4, state_QS_s32x4[ 0 ][ 1 ] ); + state_QS_s32x4[ 2 ][ 1 ] = vld1q_s32( in + n ); + vst1q_lane_s32( state + n, state_QS_s32x4[ 0 ][ 0 ], 0 ); + state_QS_s32x4[ 2 ][ 0 ] = vextq_s32( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 0 ][ 1 ], 1 ); + state_QS_s32x4[ 2 ][ 1 ] = vextq_s32( state_QS_s32x4[ 0 ][ 1 ], state_QS_s32x4[ 2 ][ 1 ], 1 ); + state_QS_s32x4[ 0 ][ 0 ] = calc_state( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], state_QS_s32x4[ 1 ][ 0 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 0 ][ 1 ] = calc_state( state_QS_s32x4[ 0 ][ 1 ], state_QS_s32x4[ 2 ][ 1 ], state_QS_s32x4[ 1 ][ 1 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 2 ][ 0 ]; + state_QS_s32x4[ 1 ][ 1 ] = state_QS_s32x4[ 2 ][ 1 ]; + } while( ++n < ( length + order ) ); + in = state; + o -= 8; + } while( o > 4 ); + + if( o ) { + /* Calculate the last 4 taps of all inputs. */ + opus_int32 *stateT = state; + silk_assert( o == 4 ); + state_QS_s32x4[ 0 ][ 0 ] = state_QS_s32x4[ 1 ][ 0 ] = vdupq_n_s32( 0 ); + n = length + order; + do { + calc_corr( input_QS, corr_QC, 0, state_QS_s32x4[ 0 ][ 0 ] ); + state_QS_s32x4[ 2 ][ 0 ] = vld1q_s32( stateT ); + vst1q_lane_s32( stateT, state_QS_s32x4[ 0 ][ 0 ], 0 ); + state_QS_s32x4[ 2 ][ 0 ] = vextq_s32( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], 1 ); + state_QS_s32x4[ 0 ][ 0 ] = calc_state( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], state_QS_s32x4[ 1 ][ 0 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 2 ][ 0 ]; + input_QS++; + stateT++; + } while( --n ); + } + } + + { + const opus_int16 *inputT = input; + int32x4_t t_s32x4; + int64x1_t t_s64x1; + int64x2_t t_s64x2 = vdupq_n_s64( 0 ); + for( n = 0; n <= length - 8; n += 8 ) { + int16x8_t input_s16x8 = vld1q_s16( inputT ); + t_s32x4 = vmull_s16( vget_low_s16( input_s16x8 ), vget_low_s16( input_s16x8 ) ); + t_s32x4 = vmlal_s16( t_s32x4, vget_high_s16( input_s16x8 ), vget_high_s16( input_s16x8 ) ); + t_s64x2 = vaddw_s32( t_s64x2, vget_low_s32( t_s32x4 ) ); + t_s64x2 = vaddw_s32( t_s64x2, vget_high_s32( t_s32x4 ) ); + inputT += 8; + } + t_s64x1 = vadd_s64( vget_low_s64( t_s64x2 ), vget_high_s64( t_s64x2 ) ); + corr_QC_orderT = vget_lane_s64( t_s64x1, 0 ); + for( ; n < length; n++ ) { + corr_QC_orderT += silk_SMULL( input[ n ], input[ n ] ); + } + corr_QC_orderT = silk_LSHIFT64( corr_QC_orderT, QC ); + corr_QC[ orderT ] = corr_QC_orderT; + } + + corr_QCT = corr_QC + orderT - order; + lsh = silk_CLZ64( corr_QC_orderT ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + lsh_s64x2 = vdupq_n_s64( lsh ); + for( i = 0; i <= order - 3; i += 4 ) { + int32x4_t corr_s32x4; + int64x2_t corr_QC0_s64x2, corr_QC1_s64x2; + corr_QC0_s64x2 = vld1q_s64( corr_QCT + i ); + corr_QC1_s64x2 = vld1q_s64( corr_QCT + i + 2 ); + corr_QC0_s64x2 = vshlq_s64( corr_QC0_s64x2, lsh_s64x2 ); + corr_QC1_s64x2 = vshlq_s64( corr_QC1_s64x2, lsh_s64x2 ); + corr_s32x4 = vcombine_s32( vmovn_s64( corr_QC1_s64x2 ), vmovn_s64( corr_QC0_s64x2 ) ); + corr_s32x4 = vrev64q_s32( corr_s32x4 ); + vst1q_s32( corr + order - i - 3, corr_s32x4 ); + } + if( lsh >= 0 ) { + for( ; i < order + 1; i++ ) { + corr[ order - i ] = (opus_int32)silk_CHECK_FIT32( silk_LSHIFT64( corr_QCT[ i ], lsh ) ); + } + } else { + for( ; i < order + 1; i++ ) { + corr[ order - i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr_QCT[ i ], -lsh ) ); + } + } + silk_assert( corr_QCT[ order ] >= 0 ); /* If breaking, decrease QC*/ + RESTORE_STACK; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int32 corr_c[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int scale_c; + silk_warped_autocorrelation_FIX_c( corr_c, &scale_c, input, warping_Q16, length, order ); + silk_assert( !memcmp( corr_c, corr, sizeof( corr_c[ 0 ] ) * ( order + 1 ) ) ); + silk_assert( scale_c == *scale ); + } +#endif +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/SigProc_FLP.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/SigProc_FLP.h new file mode 100644 index 000000000..783099beb --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/SigProc_FLP.h @@ -0,0 +1,197 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FLP_H +#define SILK_SIGPROC_FLP_H + +#include "SigProc_FIX.h" +#include "float_cast.h" +#include "main.h" +#include + + +/********************************************************************/ +/* SIGNAL PROCESSING FUNCTIONS */ +/********************************************************************/ + +/* Chirp (bw expand) LP AR filter */ +void silk_bwexpander_FLP( + silk_float *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I length of ar */ + const silk_float chirp /* I chirp factor (typically in range (0..1) ) */ +); + +/* compute inverse of LPC prediction gain, and */ +/* test if LPC coefficients are stable (all poles within unit circle) */ +/* this code is based on silk_FLP_a2k() */ +silk_float silk_LPC_inverse_pred_gain_FLP( /* O return inverse prediction gain, energy domain */ + const silk_float *A, /* I prediction coefficients [order] */ + opus_int32 order /* I prediction order */ +); + +silk_float silk_schur_FLP( /* O returns residual energy */ + silk_float refl_coef[], /* O reflection coefficients (length order) */ + const silk_float auto_corr[], /* I autocorrelation sequence (length order+1) */ + opus_int order /* I order */ +); + +void silk_k2a_FLP( + silk_float *A, /* O prediction coefficients [order] */ + const silk_float *rc, /* I reflection coefficients [order] */ + opus_int32 order /* I prediction order */ +); + +/* compute autocorrelation */ +void silk_autocorrelation_FLP( + silk_float *results, /* O result (length correlationCount) */ + const silk_float *inputData, /* I input data to correlate */ + opus_int inputDataSize, /* I length of input */ + opus_int correlationCount, /* I number of correlation taps to compute */ + int arch +); + +opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, 1 unvoiced */ + const silk_float *frame, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */ + opus_int *pitch_out, /* O Pitch lag values [nb_subfr] */ + opus_int16 *lagIndex, /* O Lag Index */ + opus_int8 *contourIndex, /* O Pitch contour Index */ + silk_float *LTPCorr, /* I/O Normalized correlation; input: value from previous frame */ + opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ + const silk_float search_thres1, /* I First stage threshold for lag candidates 0 - 1 */ + const silk_float search_thres2, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int Fs_kHz, /* I sample frequency (kHz) */ + const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ + const opus_int nb_subfr, /* I Number of 5 ms subframes */ + int arch /* I Run-time architecture */ +); + +void silk_insertion_sort_decreasing_FLP( + silk_float *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +); + +/* Compute reflection coefficients from input signal */ +silk_float silk_burg_modified_FLP( /* O returns residual energy */ + silk_float A[], /* O prediction coefficients (length order) */ + const silk_float x[], /* I input signal, length: nb_subfr*(D+L_sub) */ + const silk_float minInvGain, /* I minimum inverse prediction gain */ + const opus_int subfr_length, /* I input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I number of subframes stacked in x */ + const opus_int D, /* I order */ + int arch +); + +/* multiply a vector by a constant */ +void silk_scale_vector_FLP( + silk_float *data1, + silk_float gain, + opus_int dataSize +); + +/* copy and multiply a vector by a constant */ +void silk_scale_copy_vector_FLP( + silk_float *data_out, + const silk_float *data_in, + silk_float gain, + opus_int dataSize +); + +/* inner product of two silk_float arrays, with result as double */ +double silk_inner_product_FLP_c( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize +); + +#ifndef OVERRIDE_inner_product_FLP +#define silk_inner_product_FLP(data1, data2, dataSize, arch) ((void)arch,silk_inner_product_FLP_c(data1, data2, dataSize)) +#endif + + +/* sum of squares of a silk_float array, with result as double */ +double silk_energy_FLP( + const silk_float *data, + opus_int dataSize +); + +/********************************************************************/ +/* MACROS */ +/********************************************************************/ + +#define PI (3.1415926536f) + +#define silk_min_float( a, b ) (((a) < (b)) ? (a) : (b)) +#define silk_max_float( a, b ) (((a) > (b)) ? (a) : (b)) +#define silk_abs_float( a ) ((silk_float)fabs(a)) + +/* sigmoid function */ +static OPUS_INLINE silk_float silk_sigmoid( silk_float x ) +{ + return (silk_float)(1.0 / (1.0 + exp(-x))); +} + +/* floating-point to integer conversion (rounding) */ +static OPUS_INLINE opus_int32 silk_float2int( silk_float x ) +{ + return (opus_int32)float2int( x ); +} + +/* floating-point to integer conversion (rounding) */ +static OPUS_INLINE void silk_float2short_array( + opus_int16 *out, + const silk_float *in, + opus_int32 length +) +{ + opus_int32 k; + for( k = length - 1; k >= 0; k-- ) { + out[k] = silk_SAT16( (opus_int32)float2int( in[k] ) ); + } +} + +/* integer to floating-point conversion */ +static OPUS_INLINE void silk_short2float_array( + silk_float *out, + const opus_int16 *in, + opus_int32 length +) +{ + opus_int32 k; + for( k = length - 1; k >= 0; k-- ) { + out[k] = (silk_float)in[k]; + } +} + +/* using log2() helps the fixed-point conversion */ +static OPUS_INLINE silk_float silk_log2( double x ) +{ + return ( silk_float )( 3.32192809488736 * log10( x ) ); +} + +#endif /* SILK_SIGPROC_FLP_H */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/main_FLP.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/main_FLP.h new file mode 100644 index 000000000..8173bb32a --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/main_FLP.h @@ -0,0 +1,282 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MAIN_FLP_H +#define SILK_MAIN_FLP_H + +#include "SigProc_FLP.h" +#include "SigProc_FIX.h" +#include "structs_FLP.h" +#include "main.h" +#include "define.h" +#include "debug.h" +#include "entenc.h" + + +#define silk_encoder_state_Fxx silk_encoder_state_FLP +#define silk_encode_do_VAD_Fxx silk_encode_do_VAD_FLP +#define silk_encode_frame_Fxx silk_encode_frame_FLP + +/*********************/ +/* Encoder Functions */ +/*********************/ + +/* High-pass filter with cutoff frequency adaptation based on pitch lag statistics */ +void silk_HP_variable_cutoff( + silk_encoder_state_Fxx state_Fxx[] /* I/O Encoder states */ +); + +/* Encoder main function */ +void silk_encode_do_VAD_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int activity /* I Decision of Opus voice activity detector */ +); + +/* Encoder main function */ +opus_int silk_encode_frame_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int32 *pnBytesOut, /* O Number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +); + +/* Initializes the Silk encoder state */ +opus_int silk_init_encoder( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + int arch /* I Run-tim architecture */ +); + +/* Control the Silk encoder */ +opus_int silk_control_encoder( + silk_encoder_state_FLP *psEnc, /* I/O Pointer to Silk encoder state FLP */ + silk_EncControlStruct *encControl, /* I Control structure */ + const opus_int allow_bw_switch, /* I Flag to allow switching audio bandwidth */ + const opus_int channelNb, /* I Channel number */ + const opus_int force_fs_kHz +); + +/**************************/ +/* Noise shaping analysis */ +/**************************/ +/* Compute noise shaping coefficients and initial gain values */ +void silk_noise_shape_analysis_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float *pitch_res, /* I LPC residual from pitch analysis */ + const silk_float *x /* I Input signal [frame_length + la_shape] */ +); + +/* Autocorrelations for a warped frequency axis */ +void silk_warped_autocorrelation_FLP( + silk_float *corr, /* O Result [order + 1] */ + const silk_float *input, /* I Input data to correlate */ + const silk_float warping, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +/* Calculation of LTP state scaling */ +void silk_LTP_scale_ctrl_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/**********************************************/ +/* Prediction Analysis */ +/**********************************************/ +/* Find pitch lags */ +void silk_find_pitch_lags_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + silk_float res[], /* O Residual */ + const silk_float x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +); + +/* Find LPC and LTP coefficients */ +void silk_find_pred_coefs_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float res_pitch[], /* I Residual from pitch analysis */ + const silk_float x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/* LPC analysis */ +void silk_find_LPC_FLP( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const silk_float x[], /* I Input signal */ + const silk_float minInvGain, /* I Prediction gain from LTP (dB) */ + int arch +); + +/* LTP analysis */ +void silk_find_LTP_FLP( + silk_float XX[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Weight for LTP quantization */ + silk_float xX[ MAX_NB_SUBFR * LTP_ORDER ], /* O Weight for LTP quantization */ + const silk_float r_ptr[], /* I LPC residual */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I number of subframes */ + int arch +); + +void silk_LTP_analysis_filter_FLP( + silk_float *LTP_res, /* O LTP res MAX_NB_SUBFR*(pre_lgth+subfr_lngth) */ + const silk_float *x, /* I Input signal, with preceding samples */ + const silk_float B[ LTP_ORDER * MAX_NB_SUBFR ], /* I LTP coefficients for each subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const silk_float invGains[ MAX_NB_SUBFR ], /* I Inverse quantization gains */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I number of subframes */ + const opus_int pre_length /* I Preceding samples for each subframe */ +); + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FLP( + silk_float nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + const silk_float x[], /* I Input signal */ + silk_float a[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const silk_float gains[], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I number of subframes */ + const opus_int LPC_order /* I LPC order */ +); + +/* 16th order LPC analysis filter */ +void silk_LPC_analysis_filter_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length, /* I Length of input signal */ + const opus_int Order /* I LPC order */ +); + +/* LTP tap quantizer */ +void silk_quant_LTP_gains_FLP( + silk_float B[ MAX_NB_SUBFR * LTP_ORDER ], /* O Quantized LTP gains */ + opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook index */ + opus_int8 *periodicity_index, /* O Periodicity index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ + silk_float *pred_gain_dB, /* O LTP prediction gain */ + const silk_float XX[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I Correlation matrix */ + const silk_float xX[ MAX_NB_SUBFR * LTP_ORDER ], /* I Correlation vector */ + const opus_int subfr_len, /* I Number of samples per subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +); + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +silk_float silk_residual_energy_covar_FLP( /* O Weighted residual energy */ + const silk_float *c, /* I Filter coefficients */ + silk_float *wXX, /* I/O Weighted correlation matrix, reg. out */ + const silk_float *wXx, /* I Weighted correlation vector */ + const silk_float wxx, /* I Weighted correlation value */ + const opus_int D /* I Dimension */ +); + +/* Processing of gains */ +void silk_process_gains_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/******************/ +/* Linear Algebra */ +/******************/ +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FLP( + const silk_float *x, /* I x vector [ L+order-1 ] used to create X */ + const opus_int L, /* I Length of vectors */ + const opus_int Order, /* I Max lag for correlation */ + silk_float *XX, /* O X'*X correlation matrix [order x order] */ + int arch +); + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FLP( + const silk_float *x, /* I x vector [L+order-1] used to create X */ + const silk_float *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vecors */ + const opus_int Order, /* I Max lag for correlation */ + silk_float *Xt, /* O X'*t correlation vector [order] */ + int arch +); + +/* Apply sine window to signal vector. */ +/* Window types: */ +/* 1 -> sine window from 0 to pi/2 */ +/* 2 -> sine window from pi/2 to pi */ +void silk_apply_sine_window_FLP( + silk_float px_win[], /* O Pointer to windowed signal */ + const silk_float px[], /* I Pointer to input signal */ + const opus_int win_type, /* I Selects a window type */ + const opus_int length /* I Window length, multiple of 4 */ +); + +/* Wrapper functions. Call flp / fix code */ + +/* Convert AR filter coefficients to NLSF parameters */ +void silk_A2NLSF_FLP( + opus_int16 *NLSF_Q15, /* O NLSF vector [ LPC_order ] */ + const silk_float *pAR, /* I LPC coefficients [ LPC_order ] */ + const opus_int LPC_order /* I LPC order */ +); + +/* Convert NLSF parameters to AR prediction filter coefficients */ +void silk_NLSF2A_FLP( + silk_float *pAR, /* O LPC coefficients [ LPC_order ] */ + const opus_int16 *NLSF_Q15, /* I NLSF vector [ LPC_order ] */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +); + +/* Limit, stabilize, and quantize NLSFs */ +void silk_process_NLSFs_FLP( + silk_encoder_state *psEncC, /* I/O Encoder state */ + silk_float PredCoef[ 2 ][ MAX_LPC_ORDER ], /* O Prediction coefficients */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ], /* I/O Normalized LSFs (quant out) (0 - (2^15-1)) */ + const opus_int16 prev_NLSF_Q15[ MAX_LPC_ORDER ] /* I Previous Normalized LSFs (0 - (2^15-1)) */ +); + +/* Floating-point Silk NSQ wrapper */ +void silk_NSQ_wrapper_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + SideInfoIndices *psIndices, /* I/O Quantization indices */ + silk_nsq_state *psNSQ, /* I/O Noise Shaping Quantzation state */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const silk_float x[] /* I Prefiltered input signal */ +); + +#endif diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/structs_FLP.h b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/structs_FLP.h new file mode 100644 index 000000000..8eac15c19 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/float/structs_FLP.h @@ -0,0 +1,105 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_STRUCTS_FLP_H +#define SILK_STRUCTS_FLP_H + +#include "typedef.h" +#include "main.h" +#include "structs.h" + + +/********************************/ +/* Noise shaping analysis state */ +/********************************/ +typedef struct { + opus_int8 LastGainIndex; + silk_float HarmShapeGain_smth; + silk_float Tilt_smth; +} silk_shape_state_FLP; + +/********************************/ +/* Encoder state FLP */ +/********************************/ +typedef struct { + silk_encoder_state sCmn; /* Common struct, shared with fixed-point code */ + silk_shape_state_FLP sShape; /* Noise shaping state */ + + /* Buffer for find pitch and noise shape analysis */ + silk_float x_buf[ 2 * MAX_FRAME_LENGTH + LA_SHAPE_MAX ];/* Buffer for find pitch and noise shape analysis */ + silk_float LTPCorr; /* Normalized correlation from pitch lag estimator */ +} silk_encoder_state_FLP; + +/************************/ +/* Encoder control FLP */ +/************************/ +typedef struct { + /* Prediction and coding parameters */ + silk_float Gains[ MAX_NB_SUBFR ]; + silk_float PredCoef[ 2 ][ MAX_LPC_ORDER ]; /* holds interpolated and final coefficients */ + silk_float LTPCoef[LTP_ORDER * MAX_NB_SUBFR]; + silk_float LTP_scale; + opus_int pitchL[ MAX_NB_SUBFR ]; + + /* Noise shaping parameters */ + silk_float AR[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ]; + silk_float LF_MA_shp[ MAX_NB_SUBFR ]; + silk_float LF_AR_shp[ MAX_NB_SUBFR ]; + silk_float Tilt[ MAX_NB_SUBFR ]; + silk_float HarmShapeGain[ MAX_NB_SUBFR ]; + silk_float Lambda; + silk_float input_quality; + silk_float coding_quality; + + /* Measures */ + silk_float predGain; + silk_float LTPredCodGain; + silk_float ResNrg[ MAX_NB_SUBFR ]; /* Residual energy per subframe */ + + /* Parameters for CBR mode */ + opus_int32 GainsUnq_Q16[ MAX_NB_SUBFR ]; + opus_int8 lastGainIndexPrev; +} silk_encoder_control_FLP; + +/************************/ +/* Encoder Super Struct */ +/************************/ +typedef struct { + stereo_enc_state sStereo; + opus_int32 nBitsUsedLBRR; + opus_int32 nBitsExceeded; + opus_int nChannelsAPI; + opus_int nChannelsInternal; + opus_int nPrevChannelsInternal; + opus_int timeSinceSwitchAllowed_ms; + opus_int allowBandwidthSwitch; + opus_int prev_decode_only_middle; + /* This needs to be last so we can skip the second state for mono. */ + silk_encoder_state_FLP state_Fxx[ ENCODER_NUM_CHANNELS ]; +} silk_encoder; + +#endif diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/init_encoder.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/init_encoder.c new file mode 100644 index 000000000..10d41287f --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/init_encoder.c @@ -0,0 +1,68 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#ifdef FIXED_POINT +#include "main_FIX.h" +#else +#include "main_FLP.h" +#endif +#include "tuning_parameters.h" +#include "cpu_support.h" + +#ifdef ENABLE_DRED +#include "dred_encoder.h" +#endif + +/*********************************/ +/* Initialize Silk Encoder state */ +/*********************************/ +opus_int silk_init_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk FIX encoder state */ + int arch /* I Run-time architecture */ +) +{ + opus_int ret = 0; + + /* Clear the entire encoder state */ + silk_memset( psEnc, 0, sizeof( silk_encoder_state_Fxx ) ); + + psEnc->sCmn.arch = arch; + + psEnc->sCmn.variable_HP_smth1_Q15 = silk_LSHIFT( silk_lin2log( SILK_FIX_CONST( VARIABLE_HP_MIN_CUTOFF_HZ, 16 ) ) - ( 16 << 7 ), 8 ); + psEnc->sCmn.variable_HP_smth2_Q15 = psEnc->sCmn.variable_HP_smth1_Q15; + + /* Used to deactivate LSF interpolation, pitch prediction */ + psEnc->sCmn.first_frame_after_reset = 1; + + /* Initialize Silk VAD */ + ret += silk_VAD_Init( &psEnc->sCmn.sVAD ); + + return ret; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/process_NLSFs.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/process_NLSFs.c new file mode 100644 index 000000000..d13080954 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/process_NLSFs.c @@ -0,0 +1,107 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Limit, stabilize, convert and quantize NLSFs */ +void silk_process_NLSFs( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ], /* O Prediction coefficients */ + opus_int16 pNLSF_Q15[ MAX_LPC_ORDER ], /* I/O Normalized LSFs (quant out) (0 - (2^15-1)) */ + const opus_int16 prev_NLSFq_Q15[ MAX_LPC_ORDER ] /* I Previous Normalized LSFs (0 - (2^15-1)) */ +) +{ + opus_int i, doInterpolate; + opus_int NLSF_mu_Q20; + opus_int16 i_sqr_Q15; + opus_int16 pNLSF0_temp_Q15[ MAX_LPC_ORDER ]; + opus_int16 pNLSFW_QW[ MAX_LPC_ORDER ]; + opus_int16 pNLSFW0_temp_QW[ MAX_LPC_ORDER ]; + + silk_assert( psEncC->speech_activity_Q8 >= 0 ); + silk_assert( psEncC->speech_activity_Q8 <= SILK_FIX_CONST( 1.0, 8 ) ); + celt_assert( psEncC->useInterpolatedNLSFs == 1 || psEncC->indices.NLSFInterpCoef_Q2 == ( 1 << 2 ) ); + + /***********************/ + /* Calculate mu values */ + /***********************/ + /* NLSF_mu = 0.003 - 0.0015 * psEnc->speech_activity; */ + NLSF_mu_Q20 = silk_SMLAWB( SILK_FIX_CONST( 0.003, 20 ), SILK_FIX_CONST( -0.001, 28 ), psEncC->speech_activity_Q8 ); + if( psEncC->nb_subfr == 2 ) { + /* Multiply by 1.5 for 10 ms packets */ + NLSF_mu_Q20 = silk_ADD_RSHIFT( NLSF_mu_Q20, NLSF_mu_Q20, 1 ); + } + + celt_assert( NLSF_mu_Q20 > 0 ); + silk_assert( NLSF_mu_Q20 <= SILK_FIX_CONST( 0.005, 20 ) ); + + /* Calculate NLSF weights */ + silk_NLSF_VQ_weights_laroia( pNLSFW_QW, pNLSF_Q15, psEncC->predictLPCOrder ); + + /* Update NLSF weights for interpolated NLSFs */ + doInterpolate = ( psEncC->useInterpolatedNLSFs == 1 ) && ( psEncC->indices.NLSFInterpCoef_Q2 < 4 ); + if( doInterpolate ) { + /* Calculate the interpolated NLSF vector for the first half */ + silk_interpolate( pNLSF0_temp_Q15, prev_NLSFq_Q15, pNLSF_Q15, + psEncC->indices.NLSFInterpCoef_Q2, psEncC->predictLPCOrder ); + + /* Calculate first half NLSF weights for the interpolated NLSFs */ + silk_NLSF_VQ_weights_laroia( pNLSFW0_temp_QW, pNLSF0_temp_Q15, psEncC->predictLPCOrder ); + + /* Update NLSF weights with contribution from first half */ + i_sqr_Q15 = silk_LSHIFT( silk_SMULBB( psEncC->indices.NLSFInterpCoef_Q2, psEncC->indices.NLSFInterpCoef_Q2 ), 11 ); + for( i = 0; i < psEncC->predictLPCOrder; i++ ) { + pNLSFW_QW[ i ] = silk_ADD16( silk_RSHIFT( pNLSFW_QW[ i ], 1 ), silk_RSHIFT( + silk_SMULBB( pNLSFW0_temp_QW[ i ], i_sqr_Q15 ), 16) ); + silk_assert( pNLSFW_QW[ i ] >= 1 ); + } + } + + silk_NLSF_encode( psEncC->indices.NLSFIndices, pNLSF_Q15, psEncC->psNLSF_CB, pNLSFW_QW, + NLSF_mu_Q20, psEncC->NLSF_MSVQ_Survivors, psEncC->indices.signalType ); + + /* Convert quantized NLSFs back to LPC coefficients */ + silk_NLSF2A( PredCoef_Q12[ 1 ], pNLSF_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + if( doInterpolate ) { + /* Calculate the interpolated, quantized LSF vector for the first half */ + silk_interpolate( pNLSF0_temp_Q15, prev_NLSFq_Q15, pNLSF_Q15, + psEncC->indices.NLSFInterpCoef_Q2, psEncC->predictLPCOrder ); + + /* Convert back to LPC coefficients */ + silk_NLSF2A( PredCoef_Q12[ 0 ], pNLSF0_temp_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + } else { + /* Copy LPC coefficients for first half from second half */ + celt_assert( psEncC->predictLPCOrder <= MAX_LPC_ORDER ); + silk_memcpy( PredCoef_Q12[ 0 ], PredCoef_Q12[ 1 ], psEncC->predictLPCOrder * sizeof( opus_int16 ) ); + } +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/quant_LTP_gains.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/quant_LTP_gains.c new file mode 100644 index 000000000..d6b8eff8d --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/quant_LTP_gains.c @@ -0,0 +1,132 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +void silk_quant_LTP_gains( + opus_int16 B_Q14[ MAX_NB_SUBFR * LTP_ORDER ], /* O Quantized LTP gains */ + opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook Index */ + opus_int8 *periodicity_index, /* O Periodicity Index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ + opus_int *pred_gain_dB_Q7, /* O LTP prediction gain */ + const opus_int32 XX_Q17[ MAX_NB_SUBFR*LTP_ORDER*LTP_ORDER ], /* I Correlation matrix in Q18 */ + const opus_int32 xX_Q17[ MAX_NB_SUBFR*LTP_ORDER ], /* I Correlation vector in Q18 */ + const opus_int subfr_len, /* I Number of samples per subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +) +{ + opus_int j, k, cbk_size; + opus_int8 temp_idx[ MAX_NB_SUBFR ]; + const opus_uint8 *cl_ptr_Q5; + const opus_int8 *cbk_ptr_Q7; + const opus_uint8 *cbk_gain_ptr_Q7; + const opus_int32 *XX_Q17_ptr, *xX_Q17_ptr; + opus_int32 res_nrg_Q15_subfr, res_nrg_Q15, rate_dist_Q7_subfr, rate_dist_Q7, min_rate_dist_Q7; + opus_int32 sum_log_gain_tmp_Q7, best_sum_log_gain_Q7, max_gain_Q7; + opus_int gain_Q7; + + /***************************************************/ + /* iterate over different codebooks with different */ + /* rates/distortions, and choose best */ + /***************************************************/ + min_rate_dist_Q7 = silk_int32_MAX; + best_sum_log_gain_Q7 = 0; + for( k = 0; k < 3; k++ ) { + /* Safety margin for pitch gain control, to take into account factors + such as state rescaling/rewhitening. */ + opus_int32 gain_safety = SILK_FIX_CONST( 0.4, 7 ); + + cl_ptr_Q5 = silk_LTP_gain_BITS_Q5_ptrs[ k ]; + cbk_ptr_Q7 = silk_LTP_vq_ptrs_Q7[ k ]; + cbk_gain_ptr_Q7 = silk_LTP_vq_gain_ptrs_Q7[ k ]; + cbk_size = silk_LTP_vq_sizes[ k ]; + + /* Set up pointers to first subframe */ + XX_Q17_ptr = XX_Q17; + xX_Q17_ptr = xX_Q17; + + res_nrg_Q15 = 0; + rate_dist_Q7 = 0; + sum_log_gain_tmp_Q7 = *sum_log_gain_Q7; + for( j = 0; j < nb_subfr; j++ ) { + max_gain_Q7 = silk_log2lin( ( SILK_FIX_CONST( MAX_SUM_LOG_GAIN_DB / 6.0, 7 ) - sum_log_gain_tmp_Q7 ) + + SILK_FIX_CONST( 7, 7 ) ) - gain_safety; + silk_VQ_WMat_EC( + &temp_idx[ j ], /* O index of best codebook vector */ + &res_nrg_Q15_subfr, /* O residual energy */ + &rate_dist_Q7_subfr, /* O best weighted quantization error + mu * rate */ + &gain_Q7, /* O sum of absolute LTP coefficients */ + XX_Q17_ptr, /* I correlation matrix */ + xX_Q17_ptr, /* I correlation vector */ + cbk_ptr_Q7, /* I codebook */ + cbk_gain_ptr_Q7, /* I codebook effective gains */ + cl_ptr_Q5, /* I code length for each codebook vector */ + subfr_len, /* I number of samples per subframe */ + max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + cbk_size, /* I number of vectors in codebook */ + arch /* I Run-time architecture */ + ); + + res_nrg_Q15 = silk_ADD_POS_SAT32( res_nrg_Q15, res_nrg_Q15_subfr ); + rate_dist_Q7 = silk_ADD_POS_SAT32( rate_dist_Q7, rate_dist_Q7_subfr ); + sum_log_gain_tmp_Q7 = silk_max(0, sum_log_gain_tmp_Q7 + + silk_lin2log( gain_safety + gain_Q7 ) - SILK_FIX_CONST( 7, 7 )); + + XX_Q17_ptr += LTP_ORDER * LTP_ORDER; + xX_Q17_ptr += LTP_ORDER; + } + + if( rate_dist_Q7 <= min_rate_dist_Q7 ) { + min_rate_dist_Q7 = rate_dist_Q7; + *periodicity_index = (opus_int8)k; + silk_memcpy( cbk_index, temp_idx, nb_subfr * sizeof( opus_int8 ) ); + best_sum_log_gain_Q7 = sum_log_gain_tmp_Q7; + } + } + + cbk_ptr_Q7 = silk_LTP_vq_ptrs_Q7[ *periodicity_index ]; + for( j = 0; j < nb_subfr; j++ ) { + for( k = 0; k < LTP_ORDER; k++ ) { + B_Q14[ j * LTP_ORDER + k ] = silk_LSHIFT( cbk_ptr_Q7[ cbk_index[ j ] * LTP_ORDER + k ], 7 ); + } + } + + if( nb_subfr == 2 ) { + res_nrg_Q15 = silk_RSHIFT32( res_nrg_Q15, 1 ); + } else { + res_nrg_Q15 = silk_RSHIFT32( res_nrg_Q15, 2 ); + } + + *sum_log_gain_Q7 = best_sum_log_gain_Q7; + *pred_gain_dB_Q7 = (opus_int)silk_SMULBB( -3, silk_lin2log( res_nrg_Q15 ) - ( 15 << 7 ) ); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_LR_to_MS.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_LR_to_MS.c new file mode 100644 index 000000000..751452cb1 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_LR_to_MS.c @@ -0,0 +1,229 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/* Convert Left/Right stereo signal to adaptive Mid/Side representation */ +void silk_stereo_LR_to_MS( + stereo_enc_state *state, /* I/O State */ + opus_int16 x1[], /* I/O Left input signal, becomes mid signal */ + opus_int16 x2[], /* I/O Right input signal, becomes side signal */ + opus_int8 ix[ 2 ][ 3 ], /* O Quantization indices */ + opus_int8 *mid_only_flag, /* O Flag: only mid signal coded */ + opus_int32 mid_side_rates_bps[], /* O Bitrates for mid and side signals */ + opus_int32 total_rate_bps, /* I Total bitrate */ + opus_int prev_speech_act_Q8, /* I Speech activity level in previous frame */ + opus_int toMono, /* I Last frame before a stereo->mono transition */ + opus_int fs_kHz, /* I Sample rate (kHz) */ + opus_int frame_length /* I Number of samples */ +) +{ + opus_int n, is10msFrame, denom_Q16, delta0_Q13, delta1_Q13; + opus_int32 sum, diff, smooth_coef_Q16, pred_Q13[ 2 ], pred0_Q13, pred1_Q13; + opus_int32 LP_ratio_Q14, HP_ratio_Q14, frac_Q16, frac_3_Q16, min_mid_rate_bps, width_Q14, w_Q24, deltaw_Q24; + VARDECL( opus_int16, side ); + VARDECL( opus_int16, LP_mid ); + VARDECL( opus_int16, HP_mid ); + VARDECL( opus_int16, LP_side ); + VARDECL( opus_int16, HP_side ); + opus_int16 *mid = &x1[ -2 ]; + SAVE_STACK; + + ALLOC( side, frame_length + 2, opus_int16 ); + /* Convert to basic mid/side signals */ + for( n = 0; n < frame_length + 2; n++ ) { + sum = x1[ n - 2 ] + (opus_int32)x2[ n - 2 ]; + diff = x1[ n - 2 ] - (opus_int32)x2[ n - 2 ]; + mid[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 ); + side[ n ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( diff, 1 ) ); + } + + /* Buffering */ + silk_memcpy( mid, state->sMid, 2 * sizeof( opus_int16 ) ); + silk_memcpy( side, state->sSide, 2 * sizeof( opus_int16 ) ); + silk_memcpy( state->sMid, &mid[ frame_length ], 2 * sizeof( opus_int16 ) ); + silk_memcpy( state->sSide, &side[ frame_length ], 2 * sizeof( opus_int16 ) ); + + /* LP and HP filter mid signal */ + ALLOC( LP_mid, frame_length, opus_int16 ); + ALLOC( HP_mid, frame_length, opus_int16 ); + for( n = 0; n < frame_length; n++ ) { + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); + LP_mid[ n ] = sum; + HP_mid[ n ] = mid[ n + 1 ] - sum; + } + + /* LP and HP filter side signal */ + ALLOC( LP_side, frame_length, opus_int16 ); + ALLOC( HP_side, frame_length, opus_int16 ); + for( n = 0; n < frame_length; n++ ) { + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT32( side[ n ] + (opus_int32)side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); + LP_side[ n ] = sum; + HP_side[ n ] = side[ n + 1 ] - sum; + } + + /* Find energies and predictors */ + is10msFrame = frame_length == 10 * fs_kHz; + smooth_coef_Q16 = is10msFrame ? + SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF / 2, 16 ) : + SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF, 16 ); + smooth_coef_Q16 = silk_SMULWB( silk_SMULBB( prev_speech_act_Q8, prev_speech_act_Q8 ), smooth_coef_Q16 ); + + pred_Q13[ 0 ] = silk_stereo_find_predictor( &LP_ratio_Q14, LP_mid, LP_side, &state->mid_side_amp_Q0[ 0 ], frame_length, smooth_coef_Q16 ); + pred_Q13[ 1 ] = silk_stereo_find_predictor( &HP_ratio_Q14, HP_mid, HP_side, &state->mid_side_amp_Q0[ 2 ], frame_length, smooth_coef_Q16 ); + /* Ratio of the norms of residual and mid signals */ + frac_Q16 = silk_SMLABB( HP_ratio_Q14, LP_ratio_Q14, 3 ); + frac_Q16 = silk_min( frac_Q16, SILK_FIX_CONST( 1, 16 ) ); + + /* Determine bitrate distribution between mid and side, and possibly reduce stereo width */ + total_rate_bps -= is10msFrame ? 1200 : 600; /* Subtract approximate bitrate for coding stereo parameters */ + if( total_rate_bps < 1 ) { + total_rate_bps = 1; + } + min_mid_rate_bps = silk_SMLABB( 2000, fs_kHz, 600 ); + silk_assert( min_mid_rate_bps < 32767 ); + /* Default bitrate distribution: 8 parts for Mid and (5+3*frac) parts for Side. so: mid_rate = ( 8 / ( 13 + 3 * frac ) ) * total_ rate */ + frac_3_Q16 = silk_MUL( 3, frac_Q16 ); + mid_side_rates_bps[ 0 ] = silk_DIV32_varQ( total_rate_bps, SILK_FIX_CONST( 8 + 5, 16 ) + frac_3_Q16, 16+3 ); + /* If Mid bitrate below minimum, reduce stereo width */ + if( mid_side_rates_bps[ 0 ] < min_mid_rate_bps ) { + mid_side_rates_bps[ 0 ] = min_mid_rate_bps; + mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ]; + /* width = 4 * ( 2 * side_rate - min_rate ) / ( ( 1 + 3 * frac ) * min_rate ) */ + width_Q14 = silk_DIV32_varQ( silk_LSHIFT( mid_side_rates_bps[ 1 ], 1 ) - min_mid_rate_bps, + silk_SMULWB( SILK_FIX_CONST( 1, 16 ) + frac_3_Q16, min_mid_rate_bps ), 14+2 ); + width_Q14 = silk_LIMIT( width_Q14, 0, SILK_FIX_CONST( 1, 14 ) ); + } else { + mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ]; + width_Q14 = SILK_FIX_CONST( 1, 14 ); + } + + /* Smoother */ + state->smth_width_Q14 = (opus_int16)silk_SMLAWB( state->smth_width_Q14, width_Q14 - state->smth_width_Q14, smooth_coef_Q16 ); + + /* At very low bitrates or for inputs that are nearly amplitude panned, switch to panned-mono coding */ + *mid_only_flag = 0; + if( toMono ) { + /* Last frame before stereo->mono transition; collapse stereo width */ + width_Q14 = 0; + pred_Q13[ 0 ] = 0; + pred_Q13[ 1 ] = 0; + silk_stereo_quant_pred( pred_Q13, ix ); + } else if( state->width_prev_Q14 == 0 && + ( 8 * total_rate_bps < 13 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.05, 14 ) ) ) + { + /* Code as panned-mono; previous frame already had zero width */ + /* Scale down and quantize predictors */ + pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 ); + pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 ); + silk_stereo_quant_pred( pred_Q13, ix ); + /* Collapse stereo width */ + width_Q14 = 0; + pred_Q13[ 0 ] = 0; + pred_Q13[ 1 ] = 0; + mid_side_rates_bps[ 0 ] = total_rate_bps; + mid_side_rates_bps[ 1 ] = 0; + *mid_only_flag = 1; + } else if( state->width_prev_Q14 != 0 && + ( 8 * total_rate_bps < 11 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.02, 14 ) ) ) + { + /* Transition to zero-width stereo */ + /* Scale down and quantize predictors */ + pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 ); + pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 ); + silk_stereo_quant_pred( pred_Q13, ix ); + /* Collapse stereo width */ + width_Q14 = 0; + pred_Q13[ 0 ] = 0; + pred_Q13[ 1 ] = 0; + } else if( state->smth_width_Q14 > SILK_FIX_CONST( 0.95, 14 ) ) { + /* Full-width stereo coding */ + silk_stereo_quant_pred( pred_Q13, ix ); + width_Q14 = SILK_FIX_CONST( 1, 14 ); + } else { + /* Reduced-width stereo coding; scale down and quantize predictors */ + pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 ); + pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 ); + silk_stereo_quant_pred( pred_Q13, ix ); + width_Q14 = state->smth_width_Q14; + } + + /* Make sure to keep on encoding until the tapered output has been transmitted */ + if( *mid_only_flag == 1 ) { + state->silent_side_len += frame_length - STEREO_INTERP_LEN_MS * fs_kHz; + if( state->silent_side_len < LA_SHAPE_MS * fs_kHz ) { + *mid_only_flag = 0; + } else { + /* Limit to avoid wrapping around */ + state->silent_side_len = 10000; + } + } else { + state->silent_side_len = 0; + } + + if( *mid_only_flag == 0 && mid_side_rates_bps[ 1 ] < 1 ) { + mid_side_rates_bps[ 1 ] = 1; + mid_side_rates_bps[ 0 ] = silk_max_int( 1, total_rate_bps - mid_side_rates_bps[ 1 ]); + } + + /* Interpolate predictors and subtract prediction from side channel */ + pred0_Q13 = -state->pred_prev_Q13[ 0 ]; + pred1_Q13 = -state->pred_prev_Q13[ 1 ]; + w_Q24 = silk_LSHIFT( state->width_prev_Q14, 10 ); + denom_Q16 = silk_DIV32_16( (opus_int32)1 << 16, STEREO_INTERP_LEN_MS * fs_kHz ); + delta0_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 0 ] - state->pred_prev_Q13[ 0 ], denom_Q16 ), 16 ); + delta1_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 1 ] - state->pred_prev_Q13[ 1 ], denom_Q16 ), 16 ); + deltaw_Q24 = silk_LSHIFT( silk_SMULWB( width_Q14 - state->width_prev_Q14, denom_Q16 ), 10 ); + for( n = 0; n < STEREO_INTERP_LEN_MS * fs_kHz; n++ ) { + pred0_Q13 += delta0_Q13; + pred1_Q13 += delta1_Q13; + w_Q24 += deltaw_Q24; + sum = silk_LSHIFT( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ + sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ + x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); + } + + pred0_Q13 = -pred_Q13[ 0 ]; + pred1_Q13 = -pred_Q13[ 1 ]; + w_Q24 = silk_LSHIFT( width_Q14, 10 ); + for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) { + sum = silk_LSHIFT( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ + sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ + x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); + } + state->pred_prev_Q13[ 0 ] = (opus_int16)pred_Q13[ 0 ]; + state->pred_prev_Q13[ 1 ] = (opus_int16)pred_Q13[ 1 ]; + state->width_prev_Q14 = (opus_int16)width_Q14; + RESTORE_STACK; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_encode_pred.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_encode_pred.c new file mode 100644 index 000000000..03becb673 --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_encode_pred.c @@ -0,0 +1,62 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Entropy code the mid/side quantization indices */ +void silk_stereo_encode_pred( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int8 ix[ 2 ][ 3 ] /* I Quantization indices */ +) +{ + opus_int n; + + /* Entropy coding */ + n = 5 * ix[ 0 ][ 2 ] + ix[ 1 ][ 2 ]; + celt_assert( n < 25 ); + ec_enc_icdf( psRangeEnc, n, silk_stereo_pred_joint_iCDF, 8 ); + for( n = 0; n < 2; n++ ) { + celt_assert( ix[ n ][ 0 ] < 3 ); + celt_assert( ix[ n ][ 1 ] < STEREO_QUANT_SUB_STEPS ); + ec_enc_icdf( psRangeEnc, ix[ n ][ 0 ], silk_uniform3_iCDF, 8 ); + ec_enc_icdf( psRangeEnc, ix[ n ][ 1 ], silk_uniform5_iCDF, 8 ); + } +} + +/* Entropy code the mid-only flag */ +void silk_stereo_encode_mid_only( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int8 mid_only_flag +) +{ + /* Encode flag that only mid channel is coded */ + ec_enc_icdf( psRangeEnc, mid_only_flag, silk_stereo_only_code_mid_iCDF, 8 ); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_quant_pred.c b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_quant_pred.c new file mode 100644 index 000000000..d4ced6c3e --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/silk/stereo_quant_pred.c @@ -0,0 +1,73 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Quantize mid/side predictors */ +void silk_stereo_quant_pred( + opus_int32 pred_Q13[], /* I/O Predictors (out: quantized) */ + opus_int8 ix[ 2 ][ 3 ] /* O Quantization indices */ +) +{ + opus_int i, j, n; + opus_int32 low_Q13, step_Q13, lvl_Q13, err_min_Q13, err_Q13, quant_pred_Q13 = 0; + + /* Quantize */ + for( n = 0; n < 2; n++ ) { + /* Brute-force search over quantization levels */ + err_min_Q13 = silk_int32_MAX; + for( i = 0; i < STEREO_QUANT_TAB_SIZE - 1; i++ ) { + low_Q13 = silk_stereo_pred_quant_Q13[ i ]; + step_Q13 = silk_SMULWB( silk_stereo_pred_quant_Q13[ i + 1 ] - low_Q13, + SILK_FIX_CONST( 0.5 / STEREO_QUANT_SUB_STEPS, 16 ) ); + for( j = 0; j < STEREO_QUANT_SUB_STEPS; j++ ) { + lvl_Q13 = silk_SMLABB( low_Q13, step_Q13, 2 * j + 1 ); + err_Q13 = silk_abs( pred_Q13[ n ] - lvl_Q13 ); + if( err_Q13 < err_min_Q13 ) { + err_min_Q13 = err_Q13; + quant_pred_Q13 = lvl_Q13; + ix[ n ][ 0 ] = i; + ix[ n ][ 1 ] = j; + } else { + /* Error increasing, so we're past the optimum */ + goto done; + } + } + } + done: + ix[ n ][ 2 ] = silk_DIV32_16( ix[ n ][ 0 ], 3 ); + ix[ n ][ 0 ] -= ix[ n ][ 2 ] * 3; + pred_Q13[ n ] = quant_pred_Q13; + } + + /* Subtract second from first predictor (helps when actually applying these) */ + pred_Q13[ 0 ] -= pred_Q13[ 1 ]; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/src/opus_pcm_to_frame.c b/code/sdk/zhihui/test_opus_frame_to_pcm/src/opus_pcm_to_frame.c new file mode 100644 index 000000000..da1b4111e --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/src/opus_pcm_to_frame.c @@ -0,0 +1,228 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include +#include +#include +#include "opus.h" +#include "opus_frame_to_pcm.h" + +#define MAX_PACKET_SIZE 15000 +#define MAX_PCM_SAMPLES 5760 /* 120ms at 48kHz */ + +/* Encode PCM buffer to Opus packet - no file I/O */ +int opus_pcm_to_frame_encode_packet(const short *pcm, int pcm_samples_per_channel, + int sample_rate, int channels, + int bitrate, int application, + unsigned char *out_packet, int out_packet_max_len, + int *out_encoded_len) { + OpusEncoder *encoder = NULL; + int error = OPUS_OK; + int encoded_bytes = 0; + + if (!pcm || !out_packet || out_packet_max_len <= 0 || + pcm_samples_per_channel <= 0 || sample_rate <= 0 || channels <= 0) { + return 1; + } + + encoder = opus_encoder_create(sample_rate, channels, application, &error); + if (error != OPUS_OK || !encoder) { + if (encoder) opus_encoder_destroy(encoder); + printf("[opus_pcm_to_frame] opus_encoder_create failed: %s (code=%d), rate=%d ch=%d\n", + opus_strerror(error), error, sample_rate, channels); + return 1; + } + + error = opus_encoder_ctl(encoder, OPUS_SET_BITRATE(bitrate)); + if (error != OPUS_OK) { + opus_encoder_destroy(encoder); + return 1; + } + error = opus_encoder_ctl(encoder, OPUS_SET_LSB_DEPTH(16)); + if (error != OPUS_OK) { + opus_encoder_destroy(encoder); + return 1; + } + opus_encoder_ctl(encoder, OPUS_RESET_STATE); + + encoded_bytes = opus_encode(encoder, (const opus_int16 *)pcm, pcm_samples_per_channel, + out_packet, out_packet_max_len); + opus_encoder_destroy(encoder); + + if (encoded_bytes <= 0) { + if (encoded_bytes < 0) { + printf("[opus_pcm_to_frame] opus_encode error: %s (code=%d), samples=%d, rate=%d\n", + opus_strerror(encoded_bytes), encoded_bytes, pcm_samples_per_channel, sample_rate); + } + return 1; + } + if (out_encoded_len) { + *out_encoded_len = encoded_bytes; + } + return 0; +} + +/* Encode PCM data to a single Opus frame file - external API */ +int opus_pcm_to_frame_encode_file(const char *input_file, int sample_rate, int channels, + int bitrate, int application, const char *output_file) { + OpusEncoder *encoder = NULL; + int error = OPUS_OK; + opus_int16 *pcm = NULL; + FILE *fin = NULL; + FILE *fout = NULL; + unsigned char *packet_data = NULL; + int encoded_bytes = 0; + int pcm_samples_read = 0; + + printf("=== Encoding PCM to Opus Frame ===\n"); + printf("Input: %s\n", input_file); + printf("Output: %s\n", output_file ? output_file : "(stdout)"); + printf("Sample rate: %d Hz, Channels: %d\n", sample_rate, channels); + printf("Bitrate: %d bps, Application: %d\n", bitrate, application); + + /* Read input file */ + fin = fopen(input_file, "rb"); + if (!fin) { + fprintf(stderr, "Error opening input file: %s\n", input_file); + return 1; + } + + /* Allocate buffer for PCM data */ + pcm = (opus_int16*)malloc(MAX_PCM_SAMPLES * channels * sizeof(opus_int16)); + if (!pcm) { + fprintf(stderr, "Error allocating PCM buffer\n"); + fclose(fin); + return 1; + } + + /* Read PCM data from file */ + pcm_samples_read = fread(pcm, sizeof(opus_int16), MAX_PCM_SAMPLES * channels, fin); + fclose(fin); + + if (pcm_samples_read <= 0) { + fprintf(stderr, "Error reading PCM file or file is empty\n"); + free(pcm); + return 1; + } + + /* Calculate actual samples per channel */ + int samples_per_channel = pcm_samples_read / channels; + if (samples_per_channel <= 0) { + fprintf(stderr, "Error: Invalid PCM data (samples per channel: %d)\n", samples_per_channel); + free(pcm); + return 1; + } + + printf("Read %d samples per channel (%d total samples)\n", samples_per_channel, pcm_samples_read); + + /* Create encoder */ + encoder = opus_encoder_create(sample_rate, channels, application, &error); + if (error != OPUS_OK) { + fprintf(stderr, "Error creating encoder: %s\n", opus_strerror(error)); + free(pcm); + return 1; + } + + /* Set encoder parameters */ + error = opus_encoder_ctl(encoder, OPUS_SET_BITRATE(bitrate)); + if (error != OPUS_OK) { + fprintf(stderr, "Warning: Failed to set bitrate: %s\n", opus_strerror(error)); + } + + /* Set LSB depth to 16 bits for 16-bit PCM input */ + error = opus_encoder_ctl(encoder, OPUS_SET_LSB_DEPTH(16)); + if (error != OPUS_OK) { + fprintf(stderr, "Warning: Failed to set LSB depth: %s\n", opus_strerror(error)); + } + + /* Reset encoder state to ensure clean encoding */ + opus_encoder_ctl(encoder, OPUS_RESET_STATE); + + /* Allocate buffer for encoded packet */ + packet_data = (unsigned char*)malloc(MAX_PACKET_SIZE); + if (!packet_data) { + fprintf(stderr, "Error allocating packet buffer\n"); + opus_encoder_destroy(encoder); + free(pcm); + return 1; + } + + /* Encode PCM data */ + encoded_bytes = opus_encode(encoder, pcm, samples_per_channel, packet_data, MAX_PACKET_SIZE); + if (encoded_bytes <= 0) { + fprintf(stderr, "Error encoding PCM data: %s\n", opus_strerror(encoded_bytes)); + opus_encoder_destroy(encoder); + free(pcm); + free(packet_data); + return 1; + } + + printf("Encoded %d bytes\n", encoded_bytes); + + /* Open output file */ + if (output_file) { + fout = fopen(output_file, "wb"); + if (!fout) { + fprintf(stderr, "Error opening output file: %s\n", output_file); + opus_encoder_destroy(encoder); + free(pcm); + free(packet_data); + return 1; + } + } else { + fout = stdout; + } + + /* Write encoded packet data */ + if (fwrite(packet_data, 1, encoded_bytes, fout) != encoded_bytes) { + fprintf(stderr, "Error writing Opus packet data\n"); + if (fout != stdout) fclose(fout); + opus_encoder_destroy(encoder); + free(pcm); + free(packet_data); + return 1; + } + + if (fout != stdout) { + fclose(fout); + printf("Saved Opus frame to: %s\n", output_file); + printf("Opus file size: %d bytes\n", encoded_bytes); + } + + /* Cleanup */ + opus_encoder_destroy(encoder); + free(pcm); + free(packet_data); + + return 0; +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/analysis.c b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/analysis.c new file mode 100644 index 000000000..c353230ad --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/analysis.c @@ -0,0 +1,982 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define ANALYSIS_C + +#ifdef MLP_TRAINING +#include +#endif + +#include "mathops.h" +#include "kiss_fft.h" +#include "celt.h" +#include "modes.h" +#include "arch.h" +#include "quant_bands.h" +#include "analysis.h" +#include "mlp.h" +#include "stack_alloc.h" +#include "float_cast.h" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +#ifndef DISABLE_FLOAT_API + +#define TRANSITION_PENALTY 10 + +static const float dct_table[128] = { + 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, + 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, + 0.351851f, 0.338330f, 0.311806f, 0.273300f, 0.224292f, 0.166664f, 0.102631f, 0.034654f, + -0.034654f,-0.102631f,-0.166664f,-0.224292f,-0.273300f,-0.311806f,-0.338330f,-0.351851f, + 0.346760f, 0.293969f, 0.196424f, 0.068975f,-0.068975f,-0.196424f,-0.293969f,-0.346760f, + -0.346760f,-0.293969f,-0.196424f,-0.068975f, 0.068975f, 0.196424f, 0.293969f, 0.346760f, + 0.338330f, 0.224292f, 0.034654f,-0.166664f,-0.311806f,-0.351851f,-0.273300f,-0.102631f, + 0.102631f, 0.273300f, 0.351851f, 0.311806f, 0.166664f,-0.034654f,-0.224292f,-0.338330f, + 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f, + 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f, + 0.311806f, 0.034654f,-0.273300f,-0.338330f,-0.102631f, 0.224292f, 0.351851f, 0.166664f, + -0.166664f,-0.351851f,-0.224292f, 0.102631f, 0.338330f, 0.273300f,-0.034654f,-0.311806f, + 0.293969f,-0.068975f,-0.346760f,-0.196424f, 0.196424f, 0.346760f, 0.068975f,-0.293969f, + -0.293969f, 0.068975f, 0.346760f, 0.196424f,-0.196424f,-0.346760f,-0.068975f, 0.293969f, + 0.273300f,-0.166664f,-0.338330f, 0.034654f, 0.351851f, 0.102631f,-0.311806f,-0.224292f, + 0.224292f, 0.311806f,-0.102631f,-0.351851f,-0.034654f, 0.338330f, 0.166664f,-0.273300f, +}; + +static const float analysis_window[240] = { + 0.000043f, 0.000171f, 0.000385f, 0.000685f, 0.001071f, 0.001541f, 0.002098f, 0.002739f, + 0.003466f, 0.004278f, 0.005174f, 0.006156f, 0.007222f, 0.008373f, 0.009607f, 0.010926f, + 0.012329f, 0.013815f, 0.015385f, 0.017037f, 0.018772f, 0.020590f, 0.022490f, 0.024472f, + 0.026535f, 0.028679f, 0.030904f, 0.033210f, 0.035595f, 0.038060f, 0.040604f, 0.043227f, + 0.045928f, 0.048707f, 0.051564f, 0.054497f, 0.057506f, 0.060591f, 0.063752f, 0.066987f, + 0.070297f, 0.073680f, 0.077136f, 0.080665f, 0.084265f, 0.087937f, 0.091679f, 0.095492f, + 0.099373f, 0.103323f, 0.107342f, 0.111427f, 0.115579f, 0.119797f, 0.124080f, 0.128428f, + 0.132839f, 0.137313f, 0.141849f, 0.146447f, 0.151105f, 0.155823f, 0.160600f, 0.165435f, + 0.170327f, 0.175276f, 0.180280f, 0.185340f, 0.190453f, 0.195619f, 0.200838f, 0.206107f, + 0.211427f, 0.216797f, 0.222215f, 0.227680f, 0.233193f, 0.238751f, 0.244353f, 0.250000f, + 0.255689f, 0.261421f, 0.267193f, 0.273005f, 0.278856f, 0.284744f, 0.290670f, 0.296632f, + 0.302628f, 0.308658f, 0.314721f, 0.320816f, 0.326941f, 0.333097f, 0.339280f, 0.345492f, + 0.351729f, 0.357992f, 0.364280f, 0.370590f, 0.376923f, 0.383277f, 0.389651f, 0.396044f, + 0.402455f, 0.408882f, 0.415325f, 0.421783f, 0.428254f, 0.434737f, 0.441231f, 0.447736f, + 0.454249f, 0.460770f, 0.467298f, 0.473832f, 0.480370f, 0.486912f, 0.493455f, 0.500000f, + 0.506545f, 0.513088f, 0.519630f, 0.526168f, 0.532702f, 0.539230f, 0.545751f, 0.552264f, + 0.558769f, 0.565263f, 0.571746f, 0.578217f, 0.584675f, 0.591118f, 0.597545f, 0.603956f, + 0.610349f, 0.616723f, 0.623077f, 0.629410f, 0.635720f, 0.642008f, 0.648271f, 0.654508f, + 0.660720f, 0.666903f, 0.673059f, 0.679184f, 0.685279f, 0.691342f, 0.697372f, 0.703368f, + 0.709330f, 0.715256f, 0.721144f, 0.726995f, 0.732807f, 0.738579f, 0.744311f, 0.750000f, + 0.755647f, 0.761249f, 0.766807f, 0.772320f, 0.777785f, 0.783203f, 0.788573f, 0.793893f, + 0.799162f, 0.804381f, 0.809547f, 0.814660f, 0.819720f, 0.824724f, 0.829673f, 0.834565f, + 0.839400f, 0.844177f, 0.848895f, 0.853553f, 0.858151f, 0.862687f, 0.867161f, 0.871572f, + 0.875920f, 0.880203f, 0.884421f, 0.888573f, 0.892658f, 0.896677f, 0.900627f, 0.904508f, + 0.908321f, 0.912063f, 0.915735f, 0.919335f, 0.922864f, 0.926320f, 0.929703f, 0.933013f, + 0.936248f, 0.939409f, 0.942494f, 0.945503f, 0.948436f, 0.951293f, 0.954072f, 0.956773f, + 0.959396f, 0.961940f, 0.964405f, 0.966790f, 0.969096f, 0.971321f, 0.973465f, 0.975528f, + 0.977510f, 0.979410f, 0.981228f, 0.982963f, 0.984615f, 0.986185f, 0.987671f, 0.989074f, + 0.990393f, 0.991627f, 0.992778f, 0.993844f, 0.994826f, 0.995722f, 0.996534f, 0.997261f, + 0.997902f, 0.998459f, 0.998929f, 0.999315f, 0.999615f, 0.999829f, 0.999957f, 1.000000f, +}; + +static const int tbands[NB_TBANDS+1] = { + 4, 8, 12, 16, 20, 24, 28, 32, 40, 48, 56, 64, 80, 96, 112, 136, 160, 192, 240 +}; + +#define NB_TONAL_SKIP_BANDS 9 + +static opus_val32 silk_resampler_down2_hp( + opus_val32 *S, /* I/O State vector [ 2 ] */ + opus_val32 *out, /* O Output signal [ floor(len/2) ] */ + const opus_val32 *in, /* I Input signal [ len ] */ + int inLen /* I Number of input samples */ +) +{ + int k, len2 = inLen/2; + opus_val32 in32, out32, out32_hp, Y, X; + opus_val64 hp_ener = 0; + /* Internal variables and state are in Q10 format */ + for( k = 0; k < len2; k++ ) { + /* Convert to Q10 */ + in32 = in[ 2 * k ]; + + /* All-pass section for even input sample */ + Y = SUB32( in32, S[ 0 ] ); + X = MULT16_32_Q15(QCONST16(0.6074371f, 15), Y); + out32 = ADD32( S[ 0 ], X ); + S[ 0 ] = ADD32( in32, X ); + out32_hp = out32; + /* Convert to Q10 */ + in32 = in[ 2 * k + 1 ]; + + /* All-pass section for odd input sample, and add to output of previous section */ + Y = SUB32( in32, S[ 1 ] ); + X = MULT16_32_Q15(QCONST16(0.15063f, 15), Y); + out32 = ADD32( out32, S[ 1 ] ); + out32 = ADD32( out32, X ); + S[ 1 ] = ADD32( in32, X ); + + Y = SUB32( -in32, S[ 2 ] ); + X = MULT16_32_Q15(QCONST16(0.15063f, 15), Y); + out32_hp = ADD32( out32_hp, S[ 2 ] ); + out32_hp = ADD32( out32_hp, X ); + S[ 2 ] = ADD32( -in32, X ); + + /* len2 can be up to 480, so we shift by 8 to make it fit. */ + hp_ener += SHR64(out32_hp*(opus_val64)out32_hp, 8); + /* Add, convert back to int16 and store to output */ + out[ k ] = HALF32(out32); + } +#ifdef FIXED_POINT + /* Fitting in 32 bits. */ + hp_ener = hp_ener >> (2*SIG_SHIFT); + if (hp_ener > 2147483647) hp_ener = 2147483647; +#endif + return (opus_val32)hp_ener; +} + +static opus_val32 downmix_and_resample(downmix_func downmix, const void *_x, opus_val32 *y, opus_val32 S[3], int subframe, int offset, int c1, int c2, int C, int Fs) +{ + VARDECL(opus_val32, tmp); + int j; + opus_val32 ret = 0; + SAVE_STACK; + + if (subframe==0) return 0; + if (Fs == 48000) + { + subframe *= 2; + offset *= 2; + } else if (Fs == 16000) { + subframe = subframe*2/3; + offset = offset*2/3; + } + else if (Fs != 24000) celt_assert(0); + ALLOC(tmp, subframe, opus_val32); + + downmix(_x, tmp, subframe, offset, c1, c2, C); + if ((c2==-2 && C==2) || c2>-1) { + for (j=0;jarch = opus_select_arch(); + tonal->Fs = Fs; + /* Clear remaining fields. */ + tonality_analysis_reset(tonal); +} + +void tonality_analysis_reset(TonalityAnalysisState *tonal) +{ + /* Clear non-reusable fields. */ + char *start = (char*)&tonal->TONALITY_ANALYSIS_RESET_START; + OPUS_CLEAR(start, sizeof(TonalityAnalysisState) - (start - (char*)tonal)); +} + +void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len) +{ + int pos; + int curr_lookahead; + float tonality_max; + float tonality_avg; + int tonality_count; + int i; + int pos0; + float prob_avg; + float prob_count; + float prob_min, prob_max; + float vad_prob; + int mpos, vpos; + int bandwidth_span; + + pos = tonal->read_pos; + curr_lookahead = tonal->write_pos-tonal->read_pos; + if (curr_lookahead<0) + curr_lookahead += DETECT_SIZE; + + tonal->read_subframe += len/(tonal->Fs/400); + while (tonal->read_subframe>=8) + { + tonal->read_subframe -= 8; + tonal->read_pos++; + } + if (tonal->read_pos>=DETECT_SIZE) + tonal->read_pos-=DETECT_SIZE; + + /* On long frames, look at the second analysis window rather than the first. */ + if (len > tonal->Fs/50 && pos != tonal->write_pos) + { + pos++; + if (pos==DETECT_SIZE) + pos=0; + } + if (pos == tonal->write_pos) + pos--; + if (pos<0) + pos = DETECT_SIZE-1; + pos0 = pos; + OPUS_COPY(info_out, &tonal->info[pos], 1); + if (!info_out->valid) + return; + tonality_max = tonality_avg = info_out->tonality; + tonality_count = 1; + /* Look at the neighbouring frames and pick largest bandwidth found (to be safe). */ + bandwidth_span = 6; + /* If possible, look ahead for a tone to compensate for the delay in the tone detector. */ + for (i=0;i<3;i++) + { + pos++; + if (pos==DETECT_SIZE) + pos = 0; + if (pos == tonal->write_pos) + break; + tonality_max = MAX32(tonality_max, tonal->info[pos].tonality); + tonality_avg += tonal->info[pos].tonality; + tonality_count++; + info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth); + bandwidth_span--; + } + pos = pos0; + /* Look back in time to see if any has a wider bandwidth than the current frame. */ + for (i=0;iwrite_pos) + break; + info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth); + } + info_out->tonality = MAX32(tonality_avg/tonality_count, tonality_max-.2f); + + mpos = vpos = pos0; + /* If we have enough look-ahead, compensate for the ~5-frame delay in the music prob and + ~1 frame delay in the VAD prob. */ + if (curr_lookahead > 15) + { + mpos += 5; + if (mpos>=DETECT_SIZE) + mpos -= DETECT_SIZE; + vpos += 1; + if (vpos>=DETECT_SIZE) + vpos -= DETECT_SIZE; + } + + /* The following calculations attempt to minimize a "badness function" + for the transition. When switching from speech to music, the badness + of switching at frame k is + b_k = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T) + where + v_i is the activity probability (VAD) at frame i, + p_i is the music probability at frame i + T is the probability threshold for switching + S is the penalty for switching during active audio rather than silence + the current frame has index i=0 + + Rather than apply badness to directly decide when to switch, what we compute + instead is the threshold for which the optimal switching point is now. When + considering whether to switch now (frame 0) or at frame k, we have: + S*v_0 = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T) + which gives us: + T = ( \sum_{i=0}^{k-1} v_i*p_i + S*(v_k-v_0) ) / ( \sum_{i=0}^{k-1} v_i ) + We take the min threshold across all positive values of k (up to the maximum + amount of lookahead we have) to give us the threshold for which the current + frame is the optimal switch point. + + The last step is that we need to consider whether we want to switch at all. + For that we use the average of the music probability over the entire window. + If the threshold is higher than that average we're not going to + switch, so we compute a min with the average as well. The result of all these + min operations is music_prob_min, which gives the threshold for switching to music + if we're currently encoding for speech. + + We do the exact opposite to compute music_prob_max which is used for switching + from music to speech. + */ + prob_min = 1.f; + prob_max = 0.f; + vad_prob = tonal->info[vpos].activity_probability; + prob_count = MAX16(.1f, vad_prob); + prob_avg = MAX16(.1f, vad_prob)*tonal->info[mpos].music_prob; + while (1) + { + float pos_vad; + mpos++; + if (mpos==DETECT_SIZE) + mpos = 0; + if (mpos == tonal->write_pos) + break; + vpos++; + if (vpos==DETECT_SIZE) + vpos = 0; + if (vpos == tonal->write_pos) + break; + pos_vad = tonal->info[vpos].activity_probability; + prob_min = MIN16((prob_avg - TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_min); + prob_max = MAX16((prob_avg + TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_max); + prob_count += MAX16(.1f, pos_vad); + prob_avg += MAX16(.1f, pos_vad)*tonal->info[mpos].music_prob; + } + info_out->music_prob = prob_avg/prob_count; + prob_min = MIN16(prob_avg/prob_count, prob_min); + prob_max = MAX16(prob_avg/prob_count, prob_max); + prob_min = MAX16(prob_min, 0.f); + prob_max = MIN16(prob_max, 1.f); + + /* If we don't have enough look-ahead, do our best to make a decent decision. */ + if (curr_lookahead < 10) + { + float pmin, pmax; + pmin = prob_min; + pmax = prob_max; + pos = pos0; + /* Look for min/max in the past. */ + for (i=0;icount-1, 15);i++) + { + pos--; + if (pos < 0) + pos = DETECT_SIZE-1; + pmin = MIN16(pmin, tonal->info[pos].music_prob); + pmax = MAX16(pmax, tonal->info[pos].music_prob); + } + /* Bias against switching on active audio. */ + pmin = MAX16(0.f, pmin - .1f*vad_prob); + pmax = MIN16(1.f, pmax + .1f*vad_prob); + prob_min += (1.f-.1f*curr_lookahead)*(pmin - prob_min); + prob_max += (1.f-.1f*curr_lookahead)*(pmax - prob_max); + } + info_out->music_prob_min = prob_min; + info_out->music_prob_max = prob_max; + + /* printf("%f %f %f %f %f\n", prob_min, prob_max, prob_avg/prob_count, vad_prob, info_out->music_prob); */ +} + +static const float std_feature_bias[9] = { + 5.684947f, 3.475288f, 1.770634f, 1.599784f, 3.773215f, + 2.163313f, 1.260756f, 1.116868f, 1.918795f +}; + +#define LEAKAGE_OFFSET 2.5f +#define LEAKAGE_SLOPE 2.f + +#ifdef FIXED_POINT +/* For fixed-point, the input is +/-2^15 shifted up by SIG_SHIFT, so we need to + compensate for that in the energy. */ +#define SCALE_COMPENS (1.f/((opus_int32)1<<(15+SIG_SHIFT))) +#define SCALE_ENER(e) ((SCALE_COMPENS*SCALE_COMPENS)*(e)) +#else +#define SCALE_ENER(e) ((1.f/32768/32768)*e) +#endif + +#ifdef FIXED_POINT +static int is_digital_silence32(const opus_val32* pcm, int frame_size, int channels, int lsb_depth) +{ + int silence = 0; + opus_val32 sample_max = 0; +#ifdef MLP_TRAINING + return 0; +#endif + sample_max = celt_maxabs32(pcm, frame_size*channels); + + silence = (sample_max == 0); + (void)lsb_depth; + return silence; +} +#else +#define is_digital_silence32(pcm, frame_size, channels, lsb_depth) is_digital_silence(pcm, frame_size, channels, lsb_depth) +#endif + +static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix) +{ + int i, b; + const kiss_fft_state *kfft; + VARDECL(kiss_fft_cpx, in); + VARDECL(kiss_fft_cpx, out); + int N = 480, N2=240; + float * OPUS_RESTRICT A = tonal->angle; + float * OPUS_RESTRICT dA = tonal->d_angle; + float * OPUS_RESTRICT d2A = tonal->d2_angle; + VARDECL(float, tonality); + VARDECL(float, noisiness); + float band_tonality[NB_TBANDS]; + float logE[NB_TBANDS]; + float BFCC[8]; + float features[25]; + float frame_tonality; + float max_frame_tonality; + /*float tw_sum=0;*/ + float frame_noisiness; + const float pi4 = (float)(M_PI*M_PI*M_PI*M_PI); + float slope=0; + float frame_stationarity; + float relativeE; + float frame_probs[2]; + float alpha, alphaE, alphaE2; + float frame_loudness; + float bandwidth_mask; + int is_masked[NB_TBANDS+1]; + int bandwidth=0; + float maxE = 0; + float noise_floor; + int remaining; + AnalysisInfo *info; + float hp_ener; + float tonality2[240]; + float midE[8]; + float spec_variability=0; + float band_log2[NB_TBANDS+1]; + float leakage_from[NB_TBANDS+1]; + float leakage_to[NB_TBANDS+1]; + float layer_out[MAX_NEURONS]; + float below_max_pitch; + float above_max_pitch; + int is_silence; + SAVE_STACK; + + if (!tonal->initialized) + { + tonal->mem_fill = 240; + tonal->initialized = 1; + } + alpha = 1.f/IMIN(10, 1+tonal->count); + alphaE = 1.f/IMIN(25, 1+tonal->count); + /* Noise floor related decay for bandwidth detection: -2.2 dB/second */ + alphaE2 = 1.f/IMIN(100, 1+tonal->count); + if (tonal->count <= 1) alphaE2 = 1; + + if (tonal->Fs == 48000) + { + /* len and offset are now at 24 kHz. */ + len/= 2; + offset /= 2; + } else if (tonal->Fs == 16000) { + len = 3*len/2; + offset = 3*offset/2; + } + + kfft = celt_mode->mdct.kfft[0]; + tonal->hp_ener_accum += (float)downmix_and_resample(downmix, x, + &tonal->inmem[tonal->mem_fill], tonal->downmix_state, + IMIN(len, ANALYSIS_BUF_SIZE-tonal->mem_fill), offset, c1, c2, C, tonal->Fs); + if (tonal->mem_fill+len < ANALYSIS_BUF_SIZE) + { + tonal->mem_fill += len; + /* Don't have enough to update the analysis */ + RESTORE_STACK; + return; + } + hp_ener = tonal->hp_ener_accum; + info = &tonal->info[tonal->write_pos++]; + if (tonal->write_pos>=DETECT_SIZE) + tonal->write_pos-=DETECT_SIZE; + + is_silence = is_digital_silence32(tonal->inmem, ANALYSIS_BUF_SIZE, 1, lsb_depth); + + ALLOC(in, 480, kiss_fft_cpx); + ALLOC(out, 480, kiss_fft_cpx); + ALLOC(tonality, 240, float); + ALLOC(noisiness, 240, float); + for (i=0;iinmem[i]); + in[i].i = (kiss_fft_scalar)(w*tonal->inmem[N2+i]); + in[N-i-1].r = (kiss_fft_scalar)(w*tonal->inmem[N-i-1]); + in[N-i-1].i = (kiss_fft_scalar)(w*tonal->inmem[N+N2-i-1]); + } + OPUS_MOVE(tonal->inmem, tonal->inmem+ANALYSIS_BUF_SIZE-240, 240); + remaining = len - (ANALYSIS_BUF_SIZE-tonal->mem_fill); + tonal->hp_ener_accum = (float)downmix_and_resample(downmix, x, + &tonal->inmem[240], tonal->downmix_state, remaining, + offset+ANALYSIS_BUF_SIZE-tonal->mem_fill, c1, c2, C, tonal->Fs); + tonal->mem_fill = 240 + remaining; + if (is_silence) + { + /* On silence, copy the previous analysis. */ + int prev_pos = tonal->write_pos-2; + if (prev_pos < 0) + prev_pos += DETECT_SIZE; + OPUS_COPY(info, &tonal->info[prev_pos], 1); + RESTORE_STACK; + return; + } + opus_fft(kfft, in, out, tonal->arch); +#ifndef FIXED_POINT + /* If there's any NaN on the input, the entire output will be NaN, so we only need to check one value. */ + if (celt_isnan(out[0].r)) + { + info->valid = 0; + RESTORE_STACK; + return; + } +#endif + + for (i=1;iactivity = 0; + frame_noisiness = 0; + frame_stationarity = 0; + if (!tonal->count) + { + for (b=0;blowE[b] = 1e10; + tonal->highE[b] = -1e10; + } + } + relativeE = 0; + frame_loudness = 0; + /* The energy of the very first band is special because of DC. */ + { + float E = 0; + float X1r, X2r; + X1r = 2*(float)out[0].r; + X2r = 2*(float)out[0].i; + E = X1r*X1r + X2r*X2r; + for (i=1;i<4;i++) + { + float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r + + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i; + E += binE; + } + E = SCALE_ENER(E); + band_log2[0] = .5f*1.442695f*(float)log(E+1e-10f); + } + for (b=0;bvalid = 0; + RESTORE_STACK; + return; + } +#endif + + tonal->E[tonal->E_count][b] = E; + frame_noisiness += nE/(1e-15f+E); + + frame_loudness += (float)sqrt(E+1e-10f); + logE[b] = (float)log(E+1e-10f); + band_log2[b+1] = .5f*1.442695f*(float)log(E+1e-10f); + tonal->logE[tonal->E_count][b] = logE[b]; + if (tonal->count==0) + tonal->highE[b] = tonal->lowE[b] = logE[b]; + if (tonal->highE[b] > tonal->lowE[b] + 7.5) + { + if (tonal->highE[b] - logE[b] > logE[b] - tonal->lowE[b]) + tonal->highE[b] -= .01f; + else + tonal->lowE[b] += .01f; + } + if (logE[b] > tonal->highE[b]) + { + tonal->highE[b] = logE[b]; + tonal->lowE[b] = MAX32(tonal->highE[b]-15, tonal->lowE[b]); + } else if (logE[b] < tonal->lowE[b]) + { + tonal->lowE[b] = logE[b]; + tonal->highE[b] = MIN32(tonal->lowE[b]+15, tonal->highE[b]); + } + relativeE += (logE[b]-tonal->lowE[b])/(1e-5f + (tonal->highE[b]-tonal->lowE[b])); + + L1=L2=0; + for (i=0;iE[i][b]); + L2 += tonal->E[i][b]; + } + + stationarity = MIN16(0.99f,L1/(float)sqrt(1e-15+NB_FRAMES*L2)); + stationarity *= stationarity; + stationarity *= stationarity; + frame_stationarity += stationarity; + /*band_tonality[b] = tE/(1e-15+E)*/; + band_tonality[b] = MAX16(tE/(1e-15f+E), stationarity*tonal->prev_band_tonality[b]); +#if 0 + if (b>=NB_TONAL_SKIP_BANDS) + { + frame_tonality += tweight[b]*band_tonality[b]; + tw_sum += tweight[b]; + } +#else + frame_tonality += band_tonality[b]; + if (b>=NB_TBANDS-NB_TONAL_SKIP_BANDS) + frame_tonality -= band_tonality[b-NB_TBANDS+NB_TONAL_SKIP_BANDS]; +#endif + max_frame_tonality = MAX16(max_frame_tonality, (1.f+.03f*(b-NB_TBANDS))*frame_tonality); + slope += band_tonality[b]*(b-8); + /*printf("%f %f ", band_tonality[b], stationarity);*/ + tonal->prev_band_tonality[b] = band_tonality[b]; + } + + leakage_from[0] = band_log2[0]; + leakage_to[0] = band_log2[0] - LEAKAGE_OFFSET; + for (b=1;b=0;b--) + { + float leak_slope = LEAKAGE_SLOPE*(tbands[b+1]-tbands[b])/4; + leakage_from[b] = MIN16(leakage_from[b+1]+leak_slope, leakage_from[b]); + leakage_to[b] = MAX16(leakage_to[b+1]-leak_slope, leakage_to[b]); + } + celt_assert(NB_TBANDS+1 <= LEAK_BANDS); + for (b=0;bleak_boost[b] = IMIN(255, (int)floor(.5 + 64.f*boost)); + } + for (;bleak_boost[b] = 0; + + for (i=0;ilogE[i][k] - tonal->logE[j][k]; + dist += tmp*tmp; + } + if (j!=i) + mindist = MIN32(mindist, dist); + } + spec_variability += mindist; + } + spec_variability = (float)sqrt(spec_variability/NB_FRAMES/NB_TBANDS); + bandwidth_mask = 0; + bandwidth = 0; + maxE = 0; + noise_floor = 5.7e-4f/(1<<(IMAX(0,lsb_depth-8))); + noise_floor *= noise_floor; + below_max_pitch=0; + above_max_pitch=0; + for (b=0;bmeanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); + Em = MAX32(E, tonal->meanE[b]); + /* Consider the band "active" only if all these conditions are met: + 1) less than 90 dB below the peak band (maximal masking possible considering + both the ATH and the loudness-dependent slope of the spreading function) + 2) above the PCM quantization noise floor + We use b+1 because the first CELT band isn't included in tbands[] + */ + if (E*1e9f > maxE && (Em > 3*noise_floor*(band_end-band_start) || E > noise_floor*(band_end-band_start))) + bandwidth = b+1; + /* Check if the band is masked (see below). */ + is_masked[b] = E < (tonal->prev_bandwidth >= b+1 ? .01f : .05f)*bandwidth_mask; + /* Use a simple follower with 13 dB/Bark slope for spreading function. */ + bandwidth_mask = MAX32(.05f*bandwidth_mask, E); + } + /* Special case for the last two bands, for which we don't have spectrum but only + the energy above 12 kHz. The difficulty here is that the high-pass we use + leaks some LF energy, so we need to increase the threshold without accidentally cutting + off the band. */ + if (tonal->Fs == 48000) { + float noise_ratio; + float Em; + float E = hp_ener*(1.f/(60*60)); + noise_ratio = tonal->prev_bandwidth==20 ? 10.f : 30.f; + +#ifdef FIXED_POINT + /* silk_resampler_down2_hp() shifted right by an extra 8 bits. */ + E *= 256.f*(1.f/Q15ONE)*(1.f/Q15ONE); +#endif + above_max_pitch += E; + tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); + Em = MAX32(E, tonal->meanE[b]); + if (Em > 3*noise_ratio*noise_floor*160 || E > noise_ratio*noise_floor*160) + bandwidth = 20; + /* Check if the band is masked (see below). */ + is_masked[b] = E < (tonal->prev_bandwidth == 20 ? .01f : .05f)*bandwidth_mask; + } + if (above_max_pitch > below_max_pitch) + info->max_pitch_ratio = below_max_pitch/above_max_pitch; + else + info->max_pitch_ratio = 1; + /* In some cases, resampling aliasing can create a small amount of energy in the first band + being cut. So if the last band is masked, we don't include it. */ + if (bandwidth == 20 && is_masked[NB_TBANDS]) + bandwidth-=2; + else if (bandwidth > 0 && bandwidth <= NB_TBANDS && is_masked[bandwidth-1]) + bandwidth--; + if (tonal->count<=2) + bandwidth = 20; + frame_loudness = 20*(float)log10(frame_loudness); + tonal->Etracker = MAX32(tonal->Etracker-.003f, frame_loudness); + tonal->lowECount *= (1-alphaE); + if (frame_loudness < tonal->Etracker-30) + tonal->lowECount += alphaE; + + for (i=0;i<8;i++) + { + float sum=0; + for (b=0;b<16;b++) + sum += dct_table[i*16+b]*logE[b]; + BFCC[i] = sum; + } + for (i=0;i<8;i++) + { + float sum=0; + for (b=0;b<16;b++) + sum += dct_table[i*16+b]*.5f*(tonal->highE[b]+tonal->lowE[b]); + midE[i] = sum; + } + + frame_stationarity /= NB_TBANDS; + relativeE /= NB_TBANDS; + if (tonal->count<10) + relativeE = .5f; + frame_noisiness /= NB_TBANDS; +#if 1 + info->activity = frame_noisiness + (1-frame_noisiness)*relativeE; +#else + info->activity = .5*(1+frame_noisiness-frame_stationarity); +#endif + frame_tonality = (max_frame_tonality/(NB_TBANDS-NB_TONAL_SKIP_BANDS)); + frame_tonality = MAX16(frame_tonality, tonal->prev_tonality*.8f); + tonal->prev_tonality = frame_tonality; + + slope /= 8*8; + info->tonality_slope = slope; + + tonal->E_count = (tonal->E_count+1)%NB_FRAMES; + tonal->count = IMIN(tonal->count+1, ANALYSIS_COUNT_MAX); + info->tonality = frame_tonality; + + for (i=0;i<4;i++) + features[i] = -0.12299f*(BFCC[i]+tonal->mem[i+24]) + 0.49195f*(tonal->mem[i]+tonal->mem[i+16]) + 0.69693f*tonal->mem[i+8] - 1.4349f*tonal->cmean[i]; + + for (i=0;i<4;i++) + tonal->cmean[i] = (1-alpha)*tonal->cmean[i] + alpha*BFCC[i]; + + for (i=0;i<4;i++) + features[4+i] = 0.63246f*(BFCC[i]-tonal->mem[i+24]) + 0.31623f*(tonal->mem[i]-tonal->mem[i+16]); + for (i=0;i<3;i++) + features[8+i] = 0.53452f*(BFCC[i]+tonal->mem[i+24]) - 0.26726f*(tonal->mem[i]+tonal->mem[i+16]) -0.53452f*tonal->mem[i+8]; + + if (tonal->count > 5) + { + for (i=0;i<9;i++) + tonal->std[i] = (1-alpha)*tonal->std[i] + alpha*features[i]*features[i]; + } + for (i=0;i<4;i++) + features[i] = BFCC[i]-midE[i]; + + for (i=0;i<8;i++) + { + tonal->mem[i+24] = tonal->mem[i+16]; + tonal->mem[i+16] = tonal->mem[i+8]; + tonal->mem[i+8] = tonal->mem[i]; + tonal->mem[i] = BFCC[i]; + } + for (i=0;i<9;i++) + features[11+i] = (float)sqrt(tonal->std[i]) - std_feature_bias[i]; + features[18] = spec_variability - 0.78f; + features[20] = info->tonality - 0.154723f; + features[21] = info->activity - 0.724643f; + features[22] = frame_stationarity - 0.743717f; + features[23] = info->tonality_slope + 0.069216f; + features[24] = tonal->lowECount - 0.067930f; + + analysis_compute_dense(&layer0, layer_out, features); + analysis_compute_gru(&layer1, tonal->rnn_state, layer_out); + analysis_compute_dense(&layer2, frame_probs, tonal->rnn_state); + + /* Probability of speech or music vs noise */ + info->activity_probability = frame_probs[1]; + info->music_prob = frame_probs[0]; + + /*printf("%f %f %f\n", frame_probs[0], frame_probs[1], info->music_prob);*/ +#ifdef MLP_TRAINING + for (i=0;i<25;i++) + printf("%f ", features[i]); + printf("\n"); +#endif + + info->bandwidth = bandwidth; + tonal->prev_bandwidth = bandwidth; + /*printf("%d %d\n", info->bandwidth, info->opus_bandwidth);*/ + info->noisiness = frame_noisiness; + info->valid = 1; + RESTORE_STACK; +} + +void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, const void *analysis_pcm, + int analysis_frame_size, int frame_size, int c1, int c2, int C, opus_int32 Fs, + int lsb_depth, downmix_func downmix, AnalysisInfo *analysis_info) +{ + int offset; + int pcm_len; + + analysis_frame_size -= analysis_frame_size&1; + if (analysis_pcm != NULL) + { + /* Avoid overflow/wrap-around of the analysis buffer */ + analysis_frame_size = IMIN((DETECT_SIZE-5)*Fs/50, analysis_frame_size); + + pcm_len = analysis_frame_size - analysis->analysis_offset; + offset = analysis->analysis_offset; + while (pcm_len>0) { + tonality_analysis(analysis, celt_mode, analysis_pcm, IMIN(Fs/50, pcm_len), offset, c1, c2, C, lsb_depth, downmix); + offset += Fs/50; + pcm_len -= Fs/50; + } + analysis->analysis_offset = analysis_frame_size; + + analysis->analysis_offset -= frame_size; + } + + tonality_get_info(analysis, analysis_info, frame_size); +} + +#endif /* DISABLE_FLOAT_API */ diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/mlp.c b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/mlp.c new file mode 100644 index 000000000..b1dbab5ae --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/mlp.c @@ -0,0 +1,131 @@ +/* Copyright (c) 2008-2011 Octasic Inc. + 2012-2017 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include +#include "opus_types.h" +#include "opus_defines.h" +#include "arch.h" +#include "mlp.h" + +#define fmadd(a, b, c) ((a)*(b)+(c)) +static OPUS_INLINE float tansig_approx(float x) +{ + const float N0 = 952.52801514f; + const float N1 = 96.39235687f; + const float N2 = 0.60863042f; + const float D0 = 952.72399902f; + const float D1 = 413.36801147f; + const float D2 = 11.88600922f; + float X2, num, den; + X2 = x*x; + num = fmadd(fmadd(N2, X2, N1), X2, N0); + den = fmadd(fmadd(D2, X2, D1), X2, D0); + num = num*x/den; + return MAX32(-1.f, MIN32(1.f, num)); +} + +static OPUS_INLINE float sigmoid_approx(float x) +{ + return .5f + .5f*tansig_approx(.5f*x); +} + +static void gemm_accum(float *out, const opus_int8 *weights, int rows, int cols, int col_stride, const float *x) +{ + int i, j; + for (i=0;inb_inputs; + N = layer->nb_neurons; + stride = N; + for (i=0;ibias[i]; + gemm_accum(output, layer->input_weights, N, M, stride, input); + for (i=0;isigmoid) { + for (i=0;inb_inputs; + N = gru->nb_neurons; + stride = 3*N; + /* Compute update gate. */ + for (i=0;ibias[i]; + gemm_accum(z, gru->input_weights, N, M, stride, input); + gemm_accum(z, gru->recurrent_weights, N, N, stride, state); + for (i=0;ibias[N + i]; + gemm_accum(r, &gru->input_weights[N], N, M, stride, input); + gemm_accum(r, &gru->recurrent_weights[N], N, N, stride, state); + for (i=0;ibias[2*N + i]; + for (i=0;iinput_weights[2*N], N, M, stride, input); + gemm_accum(h, &gru->recurrent_weights[2*N], N, N, stride, tmp); + for (i=0;i +#include "celt.h" +#include "entenc.h" +#include "modes.h" +#include "API.h" +#include "stack_alloc.h" +#include "float_cast.h" +#include "opus.h" +#include "arch.h" +#include "pitch.h" +#include "opus_private.h" +#include "os_support.h" +#include "cpu_support.h" +#include "analysis.h" +#include "mathops.h" +#include "tuning_parameters.h" + +#ifdef ENABLE_DRED +#include "dred_coding.h" +#endif + +#ifdef FIXED_POINT +#include "fixed/structs_FIX.h" +#else +#include "float/structs_FLP.h" +#endif +#ifdef ENABLE_OSCE_TRAINING_DATA +#include +#endif + +#ifdef ENABLE_QEXT +#define MAX_ENCODER_BUFFER 960 +#else +#define MAX_ENCODER_BUFFER 480 +#endif + +#define PSEUDO_SNR_THRESHOLD 316.23f /* 10^(25/10) */ + +typedef struct { + opus_val32 XX, XY, YY; + opus_val16 smoothed_width; + opus_val16 max_follower; +} StereoWidthState; + +struct OpusEncoder { + int celt_enc_offset; + int silk_enc_offset; + silk_EncControlStruct silk_mode; +#ifdef ENABLE_DRED + DREDEnc dred_encoder; +#endif + int application; + int channels; + int delay_compensation; + int force_channels; + int signal_type; + int user_bandwidth; + int max_bandwidth; + int user_forced_mode; + int voice_ratio; + opus_int32 Fs; + int use_vbr; + int vbr_constraint; + int variable_duration; + opus_int32 bitrate_bps; + opus_int32 user_bitrate_bps; + int lsb_depth; + int encoder_buffer; + int lfe; + int arch; + int use_dtx; /* general DTX for both SILK and CELT */ + int fec_config; +#ifndef DISABLE_FLOAT_API + TonalityAnalysisState analysis; +#endif +#ifdef ENABLE_QEXT + int enable_qext; +#endif + +#define OPUS_ENCODER_RESET_START stream_channels + int stream_channels; + opus_int16 hybrid_stereo_width_Q14; + opus_int32 variable_HP_smth2_Q15; + opus_val16 prev_HB_gain; + opus_val32 hp_mem[4]; + int mode; + int prev_mode; + int prev_channels; + int prev_framesize; + int bandwidth; + /* Bandwidth determined automatically from the rate (before any other adjustment) */ + int auto_bandwidth; + int silk_bw_switch; + /* Sampling rate (at the API level) */ + int first; + celt_glog * energy_masking; + StereoWidthState width_mem; +#ifndef DISABLE_FLOAT_API + int detected_bandwidth; +#endif + int nb_no_activity_ms_Q1; + opus_val32 peak_signal_energy; +#ifdef ENABLE_DRED + int dred_duration; + int dred_q0; + int dred_dQ; + int dred_qmax; + int dred_target_chunks; + unsigned char activity_mem[DRED_MAX_FRAMES*4]; /* 2.5ms resolution*/ +#endif + int nonfinal_frame; /* current frame is not the final in a packet */ + opus_uint32 rangeFinal; + /* Needs to be the last field because it may be partially or completely omitted. */ + opus_res delay_buffer[MAX_ENCODER_BUFFER*2]; +}; + +/* Transition tables for the voice and music. First column is the + middle (memoriless) threshold. The second column is the hysteresis + (difference with the middle) */ +static const opus_int32 mono_voice_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 13500, 1000, /* WB<->SWB */ + 14000, 2000, /* SWB<->FB */ +}; +static const opus_int32 mono_music_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 11000, 1000, /* WB<->SWB */ + 12000, 2000, /* SWB<->FB */ +}; +static const opus_int32 stereo_voice_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 13500, 1000, /* WB<->SWB */ + 14000, 2000, /* SWB<->FB */ +}; +static const opus_int32 stereo_music_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 11000, 1000, /* WB<->SWB */ + 12000, 2000, /* SWB<->FB */ +}; +/* Threshold bit-rates for switching between mono and stereo */ +static const opus_int32 stereo_voice_threshold = 19000; +static const opus_int32 stereo_music_threshold = 17000; + +/* Threshold bit-rate for switching between SILK/hybrid and CELT-only */ +static const opus_int32 mode_thresholds[2][2] = { + /* voice */ /* music */ + { 64000, 10000}, /* mono */ + { 44000, 10000}, /* stereo */ +}; + +static const opus_int32 fec_thresholds[] = { + 12000, 1000, /* NB */ + 14000, 1000, /* MB */ + 16000, 1000, /* WB */ + 20000, 1000, /* SWB */ + 22000, 1000, /* FB */ +}; + +int opus_encoder_get_size(int channels) +{ + int ret; + ret = opus_encoder_init(NULL, 48000, channels, OPUS_APPLICATION_AUDIO); + if (ret < 0) + return 0; + else + return ret; +} + +int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int application) +{ + void *silk_enc=NULL; + CELTEncoder *celt_enc=NULL; + int err; + int ret, silkEncSizeBytes, celtEncSizeBytes=0; + int tot_size; + int base_size; + + if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000 +#ifdef ENABLE_QEXT + &&Fs!=96000 +#endif + )||(channels!=1&&channels!=2)|| + (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO + && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY + && application != OPUS_APPLICATION_RESTRICTED_SILK + && application != OPUS_APPLICATION_RESTRICTED_CELT)) + return OPUS_BAD_ARG; + + /* Create SILK encoder */ + ret = silk_Get_Encoder_Size( &silkEncSizeBytes, channels ); + if (ret) + return OPUS_BAD_ARG; + silkEncSizeBytes = align(silkEncSizeBytes); + if (application == OPUS_APPLICATION_RESTRICTED_CELT) + silkEncSizeBytes = 0; + if (application != OPUS_APPLICATION_RESTRICTED_SILK) + celtEncSizeBytes = celt_encoder_get_size(channels); + base_size = align(sizeof(OpusEncoder)); + if (application == OPUS_APPLICATION_RESTRICTED_SILK || application == OPUS_APPLICATION_RESTRICTED_CELT) { + base_size = align(base_size - MAX_ENCODER_BUFFER*2*sizeof(opus_res)); + } else if (channels==1) + base_size = align(base_size - MAX_ENCODER_BUFFER*sizeof(opus_res)); + tot_size = base_size+silkEncSizeBytes+celtEncSizeBytes; + if (st == NULL) { + return tot_size; + } + OPUS_CLEAR((char*)st, tot_size); + st->silk_enc_offset = base_size; + st->celt_enc_offset = st->silk_enc_offset+silkEncSizeBytes; + + st->stream_channels = st->channels = channels; + + st->Fs = Fs; + + st->arch = opus_select_arch(); + + if (application != OPUS_APPLICATION_RESTRICTED_CELT) + { + silk_enc = (char*)st+st->silk_enc_offset; + ret = silk_InitEncoder( silk_enc, st->channels, st->arch, &st->silk_mode ); + } + if(ret)return OPUS_INTERNAL_ERROR; + + /* default SILK parameters */ + st->silk_mode.nChannelsAPI = channels; + st->silk_mode.nChannelsInternal = channels; + st->silk_mode.API_sampleRate = st->Fs; + st->silk_mode.maxInternalSampleRate = 16000; + st->silk_mode.minInternalSampleRate = 8000; + st->silk_mode.desiredInternalSampleRate = 16000; + st->silk_mode.payloadSize_ms = 20; + st->silk_mode.bitRate = 25000; + st->silk_mode.packetLossPercentage = 0; + st->silk_mode.complexity = 9; + st->silk_mode.useInBandFEC = 0; + st->silk_mode.useDRED = 0; + st->silk_mode.useDTX = 0; + st->silk_mode.useCBR = 0; + st->silk_mode.reducedDependency = 0; + + /* Create CELT encoder */ + /* Initialize CELT encoder */ + if (application != OPUS_APPLICATION_RESTRICTED_SILK) + { + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + err = celt_encoder_init(celt_enc, Fs, channels, st->arch); + if(err!=OPUS_OK)return OPUS_INTERNAL_ERROR; + celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(st->silk_mode.complexity)); + } + +#ifdef ENABLE_DRED + /* Initialize DRED Encoder */ + dred_encoder_init( &st->dred_encoder, Fs, channels ); +#endif + + st->use_vbr = 1; + /* Makes constrained VBR the default (safer for real-time use) */ + st->vbr_constraint = 1; + st->user_bitrate_bps = OPUS_AUTO; + st->bitrate_bps = 3000+Fs*channels; + st->application = application; + st->signal_type = OPUS_AUTO; + st->user_bandwidth = OPUS_AUTO; + st->max_bandwidth = OPUS_BANDWIDTH_FULLBAND; + st->force_channels = OPUS_AUTO; + st->user_forced_mode = OPUS_AUTO; + st->voice_ratio = -1; + if (application != OPUS_APPLICATION_RESTRICTED_CELT && application != OPUS_APPLICATION_RESTRICTED_SILK) + st->encoder_buffer = st->Fs/100; + else + st->encoder_buffer = 0; + st->lsb_depth = 24; + st->variable_duration = OPUS_FRAMESIZE_ARG; + + /* Delay compensation of 4 ms (2.5 ms for SILK's extra look-ahead + + 1.5 ms for SILK resamplers and stereo prediction) */ + st->delay_compensation = st->Fs/250; + + st->hybrid_stereo_width_Q14 = 1 << 14; + st->prev_HB_gain = Q15ONE; + st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); + st->first = 1; + st->mode = MODE_HYBRID; + st->bandwidth = OPUS_BANDWIDTH_FULLBAND; + +#ifndef DISABLE_FLOAT_API + tonality_analysis_init(&st->analysis, st->Fs); + st->analysis.application = st->application; +#endif + + return OPUS_OK; +} + +static unsigned char gen_toc(int mode, int framerate, int bandwidth, int channels) +{ + int period; + unsigned char toc; + period = 0; + while (framerate < 400) + { + framerate <<= 1; + period++; + } + if (mode == MODE_SILK_ONLY) + { + toc = (bandwidth-OPUS_BANDWIDTH_NARROWBAND)<<5; + toc |= (period-2)<<3; + } else if (mode == MODE_CELT_ONLY) + { + int tmp = bandwidth-OPUS_BANDWIDTH_MEDIUMBAND; + if (tmp < 0) + tmp = 0; + toc = 0x80; + toc |= tmp << 5; + toc |= period<<3; + } else /* Hybrid */ + { + toc = 0x60; + toc |= (bandwidth-OPUS_BANDWIDTH_SUPERWIDEBAND)<<4; + toc |= (period-2)<<3; + } + toc |= (channels==2)<<2; + return toc; +} + +#ifdef FIXED_POINT +/* Second order ARMA filter, alternative implementation */ +void silk_biquad_res( + const opus_res *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [2] */ + opus_res *out, /* O output signal */ + const opus_int32 len, /* I signal length (must be even) */ + int stride +) +{ + /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ + opus_int k; + opus_int32 inval, A0_U_Q28, A0_L_Q28, A1_U_Q28, A1_L_Q28, out32_Q14; + + /* Negate A_Q28 values and split in two parts */ + A0_L_Q28 = ( -A_Q28[ 0 ] ) & 0x00003FFF; /* lower part */ + A0_U_Q28 = silk_RSHIFT( -A_Q28[ 0 ], 14 ); /* upper part */ + A1_L_Q28 = ( -A_Q28[ 1 ] ) & 0x00003FFF; /* lower part */ + A1_U_Q28 = silk_RSHIFT( -A_Q28[ 1 ], 14 ); /* upper part */ + + for( k = 0; k < len; k++ ) { + /* S[ 0 ], S[ 1 ]: Q12 */ + inval = RES2INT16(in[ k*stride ]); + out32_Q14 = silk_LSHIFT( silk_SMLAWB( S[ 0 ], B_Q28[ 0 ], inval ), 2 ); + + S[ 0 ] = S[1] + silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14, A0_L_Q28 ), 14 ); + S[ 0 ] = silk_SMLAWB( S[ 0 ], out32_Q14, A0_U_Q28 ); + S[ 0 ] = silk_SMLAWB( S[ 0 ], B_Q28[ 1 ], inval); + + S[ 1 ] = silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14, A1_L_Q28 ), 14 ); + S[ 1 ] = silk_SMLAWB( S[ 1 ], out32_Q14, A1_U_Q28 ); + S[ 1 ] = silk_SMLAWB( S[ 1 ], B_Q28[ 2 ], inval ); + + /* Scale back to Q0 and saturate */ + out[ k*stride ] = INT16TORES( silk_SAT16( silk_RSHIFT( out32_Q14 + (1<<14) - 1, 14 ) ) ); + } +} +#else +static void silk_biquad_res( + const opus_res *in, /* I: Input signal */ + const opus_int32 *B_Q28, /* I: MA coefficients [3] */ + const opus_int32 *A_Q28, /* I: AR coefficients [2] */ + opus_val32 *S, /* I/O: State vector [2] */ + opus_res *out, /* O: Output signal */ + const opus_int32 len, /* I: Signal length (must be even) */ + int stride +) +{ + /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ + opus_int k; + opus_val32 vout; + opus_val32 inval; + opus_val32 A[2], B[3]; + + A[0] = (opus_val32)(A_Q28[0] * (1.f/((opus_int32)1<<28))); + A[1] = (opus_val32)(A_Q28[1] * (1.f/((opus_int32)1<<28))); + B[0] = (opus_val32)(B_Q28[0] * (1.f/((opus_int32)1<<28))); + B[1] = (opus_val32)(B_Q28[1] * (1.f/((opus_int32)1<<28))); + B[2] = (opus_val32)(B_Q28[2] * (1.f/((opus_int32)1<<28))); + + /* Negate A_Q28 values and split in two parts */ + + for( k = 0; k < len; k++ ) { + /* S[ 0 ], S[ 1 ]: Q12 */ + inval = in[ k*stride ]; + vout = S[ 0 ] + B[0]*inval; + + S[ 0 ] = S[1] - vout*A[0] + B[1]*inval; + + S[ 1 ] = - vout*A[1] + B[2]*inval + VERY_SMALL; + + /* Scale back to Q0 and saturate */ + out[ k*stride ] = vout; + } +} +#endif + +static void hp_cutoff(const opus_res *in, opus_int32 cutoff_Hz, opus_res *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs, int arch) +{ + opus_int32 B_Q28[ 3 ], A_Q28[ 2 ]; + opus_int32 Fc_Q19, r_Q28, r_Q22; + (void)arch; + + silk_assert( cutoff_Hz <= silk_int32_MAX / SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ) ); + Fc_Q19 = silk_DIV32_16( silk_SMULBB( SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ), cutoff_Hz ), Fs/1000 ); + silk_assert( Fc_Q19 > 0 && Fc_Q19 < 32768 ); + + r_Q28 = SILK_FIX_CONST( 1.0, 28 ) - silk_MUL( SILK_FIX_CONST( 0.92, 9 ), Fc_Q19 ); + + /* b = r * [ 1; -2; 1 ]; */ + /* a = [ 1; -2 * r * ( 1 - 0.5 * Fc^2 ); r^2 ]; */ + B_Q28[ 0 ] = r_Q28; + B_Q28[ 1 ] = silk_LSHIFT( -r_Q28, 1 ); + B_Q28[ 2 ] = r_Q28; + + /* -r * ( 2 - Fc * Fc ); */ + r_Q22 = silk_RSHIFT( r_Q28, 6 ); + A_Q28[ 0 ] = silk_SMULWW( r_Q22, silk_SMULWW( Fc_Q19, Fc_Q19 ) - SILK_FIX_CONST( 2.0, 22 ) ); + A_Q28[ 1 ] = silk_SMULWW( r_Q22, r_Q22 ); + +#if defined(FIXED_POINT) && !defined(ENABLE_RES24) + if( channels == 1 ) { + silk_biquad_alt_stride1( in, B_Q28, A_Q28, hp_mem, out, len ); + } else { + silk_biquad_alt_stride2( in, B_Q28, A_Q28, hp_mem, out, len, arch ); + } +#else + silk_biquad_res( in, B_Q28, A_Q28, hp_mem, out, len, channels ); + if( channels == 2 ) { + silk_biquad_res( in+1, B_Q28, A_Q28, hp_mem+2, out+1, len, channels ); + } +#endif +} + +#ifdef FIXED_POINT +static void dc_reject(const opus_res *in, opus_int32 cutoff_Hz, opus_res *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs) +{ + int c, i; + int shift; + + /* Approximates -round(log2(6.3*cutoff_Hz/Fs)) */ + shift=celt_ilog2(Fs/(cutoff_Hz*4)); + for (c=0;csilk_mode.useInBandFEC) { + dred_frac = MIN16(.7f, 3.f*st->silk_mode.packetLossPercentage/100.f); + bitrate_offset = 20000; + } else { + if (st->silk_mode.packetLossPercentage > 5) { + dred_frac = MIN16(.8f, .55f + st->silk_mode.packetLossPercentage/100.f); + } else { + dred_frac = 12*st->silk_mode.packetLossPercentage/100.f; + } + bitrate_offset = 12000; + } + /* Account for the fact that longer packets require less redundancy. */ + dred_frac = dred_frac/(dred_frac + (1-dred_frac)*(frame_size*50.f)/st->Fs); + /* Approximate fit based on a few experiments. Could probably be improved. */ + q0 = IMIN(15, IMAX(4, 51 - 3*EC_ILOG(IMAX(1, bitrate_bps-bitrate_offset)))); + dQ = bitrate_bps-bitrate_offset > 36000 ? 3 : 5; + qmax = 15; + target_dred_bitrate = IMAX(0, (int)(dred_frac*(bitrate_bps-bitrate_offset))); + if (st->dred_duration > 0) { + opus_int32 target_bits = bitrate_to_bits(target_dred_bitrate, st->Fs, frame_size); + max_dred_bits = estimate_dred_bitrate(q0, dQ, qmax, st->dred_duration, target_bits, &target_chunks); + } else { + max_dred_bits = 0; + target_chunks=0; + } + dred_bitrate = IMIN(target_dred_bitrate, bits_to_bitrate(max_dred_bits, st->Fs, frame_size)); + /* If we can't afford enough bits, don't bother with DRED at all. */ + if (target_chunks < 2) + dred_bitrate = 0; + st->dred_q0 = q0; + st->dred_dQ = dQ; + st->dred_qmax = qmax; + st->dred_target_chunks = target_chunks; + return dred_bitrate; +} +#endif + +static opus_int32 user_bitrate_to_bitrate(OpusEncoder *st, int frame_size, int max_data_bytes) +{ + opus_int32 max_bitrate, user_bitrate; + if(!frame_size)frame_size=st->Fs/400; + max_bitrate = bits_to_bitrate(max_data_bytes*8, st->Fs, frame_size); + if (st->user_bitrate_bps==OPUS_AUTO) + user_bitrate = 60*st->Fs/frame_size + st->Fs*st->channels; + else if (st->user_bitrate_bps==OPUS_BITRATE_MAX) + user_bitrate = 1500000; + else + user_bitrate = st->user_bitrate_bps; + return IMIN(user_bitrate, max_bitrate); +} + +#ifndef DISABLE_FLOAT_API +void downmix_float(const void *_x, opus_val32 *y, int subframe, int offset, int c1, int c2, int C) +{ + const float *x; + int j; + + x = (const float *)_x; + for (j=0;j-1) + { + for (j=0;j 65536.f) y[j] = 65536.f; + if (celt_isnan(y[j])) y[j] = 0; + } +#endif +} +#endif + +void downmix_int(const void *_x, opus_val32 *y, int subframe, int offset, int c1, int c2, int C) +{ + const opus_int16 *x; + int j; + + x = (const opus_int16 *)_x; + for (j=0;j-1) + { + for (j=0;j-1) + { + for (j=0;j= OPUS_FRAMESIZE_2_5_MS && variable_duration <= OPUS_FRAMESIZE_120_MS) + { + if (variable_duration <= OPUS_FRAMESIZE_40_MS) + new_size = (Fs/400)<<(variable_duration-OPUS_FRAMESIZE_2_5_MS); + else + new_size = (variable_duration-OPUS_FRAMESIZE_2_5_MS-2)*Fs/50; + } + else + return -1; + if (new_size>frame_size) + return -1; + if (400*new_size!=Fs && 200*new_size!=Fs && 100*new_size!=Fs && + 50*new_size!=Fs && 25*new_size!=Fs && 50*new_size!=3*Fs && + 50*new_size!=4*Fs && 50*new_size!=5*Fs && 50*new_size!=6*Fs) + return -1; + if (application == OPUS_APPLICATION_RESTRICTED_SILK && new_size < Fs/100) + return -1; + return new_size; +} + +opus_val16 compute_stereo_width(const opus_res *pcm, int frame_size, opus_int32 Fs, StereoWidthState *mem) +{ + opus_val32 xx, xy, yy; + opus_val16 sqrt_xx, sqrt_yy; + opus_val16 qrrt_xx, qrrt_yy; + int frame_rate; + int i; + opus_val16 short_alpha; +#ifdef FIXED_POINT + int shift = celt_ilog2(frame_size)-2; +#endif + frame_rate = Fs/frame_size; + short_alpha = MULT16_16(25, Q15ONE)/IMAX(50,frame_rate); + xx=xy=yy=0; + /* Unroll by 4. The frame size is always a multiple of 4 *except* for + 2.5 ms frames at 12 kHz. Since this setting is very rare (and very + stupid), we just discard the last two samples. */ + for (i=0;iXX += MULT16_32_Q15(short_alpha, xx-mem->XX); + /*mem->XY += MULT16_32_Q15(short_alpha, xy-mem->XY);*/ + /* Rewritten to avoid overflows on abrupt sign change. */ + mem->XY = MULT16_32_Q15(Q15ONE - short_alpha, mem->XY) + MULT16_32_Q15(short_alpha, xy); + mem->YY += MULT16_32_Q15(short_alpha, yy-mem->YY); + mem->XX = MAX32(0, mem->XX); + mem->XY = MAX32(0, mem->XY); + mem->YY = MAX32(0, mem->YY); + if (MAX32(mem->XX, mem->YY)>QCONST16(8e-4f, 18)) + { + opus_val16 corr; + opus_val16 ldiff; + opus_val16 width; + sqrt_xx = celt_sqrt(mem->XX); + sqrt_yy = celt_sqrt(mem->YY); + qrrt_xx = celt_sqrt(sqrt_xx); + qrrt_yy = celt_sqrt(sqrt_yy); + /* Inter-channel correlation */ + mem->XY = MIN32(mem->XY, sqrt_xx*sqrt_yy); + corr = SHR32(frac_div32(mem->XY,EPSILON+MULT16_16(sqrt_xx,sqrt_yy)),16); + /* Approximate loudness difference */ + ldiff = MULT16_16(Q15ONE, ABS16(qrrt_xx-qrrt_yy))/(EPSILON+qrrt_xx+qrrt_yy); + width = MULT16_16_Q15(MIN16(Q15ONE, celt_sqrt(QCONST32(1.f,30)-MULT16_16(corr,corr))), ldiff); + /* Smoothing over one second */ + mem->smoothed_width += (width-mem->smoothed_width)/frame_rate; + /* Peak follower */ + mem->max_follower = MAX16(mem->max_follower-QCONST16(.02f,15)/frame_rate, mem->smoothed_width); + } + /*printf("%f %f %f %f %f ", corr/(float)Q15ONE, ldiff/(float)Q15ONE, width/(float)Q15ONE, mem->smoothed_width/(float)Q15ONE, mem->max_follower/(float)Q15ONE);*/ + return EXTRACT16(MIN32(Q15ONE, MULT16_16(20, mem->max_follower))); +} + +static int decide_fec(int useInBandFEC, int PacketLoss_perc, int last_fec, int mode, int *bandwidth, opus_int32 rate) +{ + int orig_bandwidth; + if (!useInBandFEC || PacketLoss_perc == 0 || mode == MODE_CELT_ONLY) + return 0; + orig_bandwidth = *bandwidth; + for (;;) + { + opus_int32 hysteresis; + opus_int32 LBRR_rate_thres_bps; + /* Compute threshold for using FEC at the current bandwidth setting */ + LBRR_rate_thres_bps = fec_thresholds[2*(*bandwidth - OPUS_BANDWIDTH_NARROWBAND)]; + hysteresis = fec_thresholds[2*(*bandwidth - OPUS_BANDWIDTH_NARROWBAND) + 1]; + if (last_fec == 1) LBRR_rate_thres_bps -= hysteresis; + if (last_fec == 0) LBRR_rate_thres_bps += hysteresis; + LBRR_rate_thres_bps = silk_SMULWB( silk_MUL( LBRR_rate_thres_bps, + 125 - silk_min( PacketLoss_perc, 25 ) ), SILK_FIX_CONST( 0.01, 16 ) ); + /* If loss <= 5%, we look at whether we have enough rate to enable FEC. + If loss > 5%, we decrease the bandwidth until we can enable FEC. */ + if (rate > LBRR_rate_thres_bps) + return 1; + else if (PacketLoss_perc <= 5) + return 0; + else if (*bandwidth > OPUS_BANDWIDTH_NARROWBAND) + (*bandwidth)--; + else + break; + } + /* Couldn't find any bandwidth to enable FEC, keep original bandwidth. */ + *bandwidth = orig_bandwidth; + return 0; +} + +static int compute_silk_rate_for_hybrid(int rate, int bandwidth, int frame20ms, int vbr, int fec, int channels) { + int entry; + int i; + int N; + int silk_rate; + static const int rate_table[][5] = { + /* |total| |-------- SILK------------| + |-- No FEC -| |--- FEC ---| + 10ms 20ms 10ms 20ms */ + { 0, 0, 0, 0, 0}, + {12000, 10000, 10000, 11000, 11000}, + {16000, 13500, 13500, 15000, 15000}, + {20000, 16000, 16000, 18000, 18000}, + {24000, 18000, 18000, 21000, 21000}, + {32000, 22000, 22000, 28000, 28000}, + {64000, 38000, 38000, 50000, 50000} + }; + /* Do the allocation per-channel. */ + rate /= channels; + entry = 1 + frame20ms + 2*fec; + N = sizeof(rate_table)/sizeof(rate_table[0]); + for (i=1;i rate) break; + } + if (i == N) + { + silk_rate = rate_table[i-1][entry]; + /* For now, just give 50% of the extra bits to SILK. */ + silk_rate += (rate-rate_table[i-1][0])/2; + } else { + opus_int32 lo, hi, x0, x1; + lo = rate_table[i-1][entry]; + hi = rate_table[i][entry]; + x0 = rate_table[i-1][0]; + x1 = rate_table[i][0]; + silk_rate = (lo*(x1-rate) + hi*(rate-x0))/(x1-x0); + } + if (!vbr) + { + /* Tiny boost to SILK for CBR. We should probably tune this better. */ + silk_rate += 100; + } + if (bandwidth==OPUS_BANDWIDTH_SUPERWIDEBAND) + silk_rate += 300; + silk_rate *= channels; + /* Small adjustment for stereo (calibrated for 32 kb/s, haven't tried other bitrates). */ + if (channels == 2 && rate >= 12000) + silk_rate -= 1000; + return silk_rate; +} + +/* Returns the equivalent bitrate corresponding to 20 ms frames, + complexity 10 VBR operation. */ +static opus_int32 compute_equiv_rate(opus_int32 bitrate, int channels, + int frame_rate, int vbr, int mode, int complexity, int loss) +{ + opus_int32 equiv; + equiv = bitrate; + /* Take into account overhead from smaller frames. */ + if (frame_rate > 50) + equiv -= (40*channels+20)*(frame_rate - 50); + /* CBR is about a 8% penalty for both SILK and CELT. */ + if (!vbr) + equiv -= equiv/12; + /* Complexity makes about 10% difference (from 0 to 10) in general. */ + equiv = equiv * (90+complexity)/100; + if (mode == MODE_SILK_ONLY || mode == MODE_HYBRID) + { + /* SILK complexity 0-1 uses the non-delayed-decision NSQ, which + costs about 20%. */ + if (complexity<2) + equiv = equiv*4/5; + equiv -= equiv*loss/(6*loss + 10); + } else if (mode == MODE_CELT_ONLY) { + /* CELT complexity 0-4 doesn't have the pitch filter, which costs + about 10%. */ + if (complexity<5) + equiv = equiv*9/10; + } else { + /* Mode not known yet */ + /* Half the SILK loss*/ + equiv -= equiv*loss/(12*loss + 20); + } + return equiv; +} + +int is_digital_silence(const opus_res* pcm, int frame_size, int channels, int lsb_depth) +{ + int silence = 0; + opus_val32 sample_max = 0; +#ifdef MLP_TRAINING + return 0; +#endif + sample_max = celt_maxabs_res(pcm, frame_size*channels); + +#ifdef FIXED_POINT + silence = (sample_max == 0); + (void)lsb_depth; +#else + silence = (sample_max <= (opus_val16) 1 / (1 << lsb_depth)); +#endif + + return silence; +} + +#ifdef FIXED_POINT +static opus_val32 compute_frame_energy(const opus_res *pcm, int frame_size, int channels, int arch) +{ + int i; + opus_val32 sample_max; + int max_shift; + int shift; + opus_val32 energy = 0; + int len = frame_size*channels; + (void)arch; + /* Max amplitude in the signal */ + sample_max = RES2INT16(celt_maxabs_res(pcm, len)); + + /* Compute the right shift required in the MAC to avoid an overflow */ + max_shift = celt_ilog2(len); + shift = IMAX(0, (celt_ilog2(1+sample_max) << 1) + max_shift - 28); + + /* Compute the energy */ + for (i=0; i NB_SPEECH_FRAMES_BEFORE_DTX*20*2) + { + if (*nb_no_activity_ms_Q1 <= (NB_SPEECH_FRAMES_BEFORE_DTX + MAX_CONSECUTIVE_DTX)*20*2) + /* Valid frame for DTX! */ + return 1; + else + (*nb_no_activity_ms_Q1) = NB_SPEECH_FRAMES_BEFORE_DTX*20*2; + } + } else + (*nb_no_activity_ms_Q1) = 0; + + return 0; +} + +static int compute_redundancy_bytes(opus_int32 max_data_bytes, opus_int32 bitrate_bps, int frame_rate, int channels) +{ + int redundancy_bytes_cap; + int redundancy_bytes; + opus_int32 redundancy_rate; + int base_bits; + opus_int32 available_bits; + base_bits = (40*channels+20); + + /* Equivalent rate for 5 ms frames. */ + redundancy_rate = bitrate_bps + base_bits*(200 - frame_rate); + /* For VBR, further increase the bitrate if we can afford it. It's pretty short + and we'll avoid artefacts. */ + redundancy_rate = 3*redundancy_rate/2; + redundancy_bytes = redundancy_rate/1600; + + /* Compute the max rate we can use given CBR or VBR with cap. */ + available_bits = max_data_bytes*8 - 2*base_bits; + redundancy_bytes_cap = (available_bits*240/(240+48000/frame_rate) + base_bits)/8; + redundancy_bytes = IMIN(redundancy_bytes, redundancy_bytes_cap); + /* It we can't get enough bits for redundancy to be worth it, rely on the decoder PLC. */ + if (redundancy_bytes > 4 + 8*channels) + redundancy_bytes = IMIN(257, redundancy_bytes); + else + redundancy_bytes = 0; + return redundancy_bytes; +} + +static opus_int32 opus_encode_frame_native(OpusEncoder *st, const opus_res *pcm, int frame_size, + unsigned char *data, opus_int32 max_data_bytes, + int float_api, int first_frame, +#ifdef ENABLE_DRED + opus_int32 dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + AnalysisInfo *analysis_info, +#endif + int is_silence, int redundancy, int celt_to_silk, int prefill, + opus_int32 equiv_rate, int to_celt); + +opus_int32 opus_encode_native(OpusEncoder *st, const opus_res *pcm, int frame_size, + unsigned char *data, opus_int32 out_data_bytes, int lsb_depth, + const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, + int analysis_channels, downmix_func downmix, int float_api) +{ + void *silk_enc=NULL; + CELTEncoder *celt_enc=NULL; + int i; + int ret=0; + int prefill=0; + int redundancy = 0; + int celt_to_silk = 0; + int to_celt = 0; + int voice_est; /* Probability of voice in Q7 */ + opus_int32 equiv_rate; + int frame_rate; + opus_int32 max_rate; /* Max bitrate we're allowed to use */ + int curr_bandwidth; + opus_int32 max_data_bytes; /* Max number of bytes we're allowed to use */ + opus_int32 cbr_bytes=-1; + opus_val16 stereo_width; + const CELTMode *celt_mode=NULL; + int packet_size_cap = 1276; +#ifndef DISABLE_FLOAT_API + AnalysisInfo analysis_info; + int analysis_read_pos_bak=-1; + int analysis_read_subframe_bak=-1; +#endif + int is_silence = 0; +#ifdef ENABLE_DRED + opus_int32 dred_bitrate_bps; +#endif + ALLOC_STACK; + +#ifdef ENABLE_QEXT + if (st->enable_qext) packet_size_cap = QEXT_PACKET_SIZE_CAP; +#endif + + /* Just avoid insane packet sizes here, but the real bounds are applied later on. */ + max_data_bytes = IMIN(packet_size_cap*6, out_data_bytes); + + st->rangeFinal = 0; + if (frame_size <= 0 || max_data_bytes <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + /* Cannot encode 100 ms in 1 byte */ + if (max_data_bytes==1 && st->Fs==(frame_size*10)) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + + if (st->application != OPUS_APPLICATION_RESTRICTED_CELT) + silk_enc = (char*)st+st->silk_enc_offset; + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + + lsb_depth = IMIN(lsb_depth, st->lsb_depth); + + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); + is_silence = is_digital_silence(pcm, frame_size, st->channels, lsb_depth); +#ifndef DISABLE_FLOAT_API + analysis_info.valid = 0; +#ifdef FIXED_POINT + if (st->silk_mode.complexity >= 10 && st->Fs>=16000 && st->Fs<=48000 && st->application != OPUS_APPLICATION_RESTRICTED_SILK) +#else + if (st->silk_mode.complexity >= 7 && st->Fs>=16000 && st->Fs<=48000 && st->application != OPUS_APPLICATION_RESTRICTED_SILK) +#endif + { + analysis_read_pos_bak = st->analysis.read_pos; + analysis_read_subframe_bak = st->analysis.read_subframe; + run_analysis(&st->analysis, celt_mode, analysis_pcm, analysis_size, frame_size, + c1, c2, analysis_channels, st->Fs, + lsb_depth, downmix, &analysis_info); + + } else if (st->analysis.initialized) { + tonality_analysis_reset(&st->analysis); + } +#else + (void)analysis_pcm; + (void)analysis_size; + (void)c1; + (void)c2; + (void)analysis_channels; + (void)downmix; +#endif + + /* Reset voice_ratio if this frame is not silent or if analysis is disabled. + * Otherwise, preserve voice_ratio from the last non-silent frame */ + if (!is_silence) + st->voice_ratio = -1; +#ifndef DISABLE_FLOAT_API + st->detected_bandwidth = 0; + if (analysis_info.valid) + { + int analysis_bandwidth; + if (st->signal_type == OPUS_AUTO) + { + float prob; + if (st->prev_mode == 0) + prob = analysis_info.music_prob; + else if (st->prev_mode == MODE_CELT_ONLY) + prob = analysis_info.music_prob_max; + else + prob = analysis_info.music_prob_min; + st->voice_ratio = (int)floor(.5+100*(1-prob)); + } + + analysis_bandwidth = analysis_info.bandwidth; + if (analysis_bandwidth<=12) + st->detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (analysis_bandwidth<=14) + st->detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (analysis_bandwidth<=16) + st->detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (analysis_bandwidth<=18) + st->detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else + st->detected_bandwidth = OPUS_BANDWIDTH_FULLBAND; + } +#else + st->voice_ratio = -1; +#endif + + /* Track the peak signal energy */ +#ifndef DISABLE_FLOAT_API + if (!analysis_info.valid || analysis_info.activity_probability > DTX_ACTIVITY_THRESHOLD) +#endif + { + if (!is_silence) + { + st->peak_signal_energy = MAX32(MULT16_32_Q15(QCONST16(0.999f, 15), st->peak_signal_energy), + compute_frame_energy(pcm, frame_size, st->channels, st->arch)); + } + } + if (st->channels==2 && st->force_channels!=1) + stereo_width = compute_stereo_width(pcm, frame_size, st->Fs, &st->width_mem); + else + stereo_width = 0; + st->bitrate_bps = user_bitrate_to_bitrate(st, frame_size, max_data_bytes); + + frame_rate = st->Fs/frame_size; + if (!st->use_vbr) + { + cbr_bytes = IMIN((bitrate_to_bits(st->bitrate_bps, st->Fs, frame_size)+4)/8, max_data_bytes); + st->bitrate_bps = bits_to_bitrate(cbr_bytes*8, st->Fs, frame_size); + /* Make sure we provide at least one byte to avoid failing. */ + max_data_bytes = IMAX(1, cbr_bytes); + } +#ifdef ENABLE_DRED + /* Allocate some of the bits to DRED if needed. */ + dred_bitrate_bps = compute_dred_bitrate(st, st->bitrate_bps, frame_size); + st->bitrate_bps -= dred_bitrate_bps; +#endif + if (max_data_bytes<3 || st->bitrate_bps < 3*frame_rate*8 + || (frame_rate<50 && (max_data_bytes*(opus_int32)frame_rate<300 || st->bitrate_bps < 2400))) + { + /*If the space is too low to do something useful, emit 'PLC' frames.*/ + int tocmode = st->mode; + int bw = st->bandwidth == 0 ? OPUS_BANDWIDTH_NARROWBAND : st->bandwidth; + int packet_code = 0; + int num_multiframes = 0; + + if (tocmode==0) + tocmode = MODE_SILK_ONLY; + if (frame_rate>100) + tocmode = MODE_CELT_ONLY; + /* 40 ms -> 2 x 20 ms if in CELT_ONLY or HYBRID mode */ + if (frame_rate==25 && tocmode!=MODE_SILK_ONLY) + { + frame_rate = 50; + packet_code = 1; + } + + /* >= 60 ms frames */ + if (frame_rate<=16) + { + /* 1 x 60 ms, 2 x 40 ms, 2 x 60 ms */ + if (out_data_bytes==1 || (tocmode==MODE_SILK_ONLY && frame_rate!=10)) + { + tocmode = MODE_SILK_ONLY; + + packet_code = frame_rate <= 12; + frame_rate = frame_rate == 12 ? 25 : 16; + } + else + { + num_multiframes = 50/frame_rate; + frame_rate = 50; + packet_code = 3; + } + } + + if(tocmode==MODE_SILK_ONLY&&bw>OPUS_BANDWIDTH_WIDEBAND) + bw=OPUS_BANDWIDTH_WIDEBAND; + else if (tocmode==MODE_CELT_ONLY&&bw==OPUS_BANDWIDTH_MEDIUMBAND) + bw=OPUS_BANDWIDTH_NARROWBAND; + else if (tocmode==MODE_HYBRID&&bw<=OPUS_BANDWIDTH_SUPERWIDEBAND) + bw=OPUS_BANDWIDTH_SUPERWIDEBAND; + + data[0] = gen_toc(tocmode, frame_rate, bw, st->stream_channels); + data[0] |= packet_code; + + ret = packet_code <= 1 ? 1 : 2; + + max_data_bytes = IMAX(max_data_bytes, ret); + + if (packet_code==3) + data[1] = num_multiframes; + + if (!st->use_vbr) + { + ret = opus_packet_pad(data, ret, max_data_bytes); + if (ret == OPUS_OK) + ret = max_data_bytes; + else + ret = OPUS_INTERNAL_ERROR; + } + RESTORE_STACK; + return ret; + } + max_rate = bits_to_bitrate(max_data_bytes*8, st->Fs, frame_size); + + /* Equivalent 20-ms rate for mode/channel/bandwidth decisions */ + equiv_rate = compute_equiv_rate(st->bitrate_bps, st->channels, st->Fs/frame_size, + st->use_vbr, 0, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + + if (st->signal_type == OPUS_SIGNAL_VOICE) + voice_est = 127; + else if (st->signal_type == OPUS_SIGNAL_MUSIC) + voice_est = 0; + else if (st->voice_ratio >= 0) + { + voice_est = st->voice_ratio*327>>8; + /* For AUDIO, never be more than 90% confident of having speech */ + if (st->application == OPUS_APPLICATION_AUDIO) + voice_est = IMIN(voice_est, 115); + } else if (st->application == OPUS_APPLICATION_VOIP) + voice_est = 115; + else + voice_est = 48; + + if (st->force_channels!=OPUS_AUTO && st->channels == 2) + { + st->stream_channels = st->force_channels; + } else { +#ifdef FUZZING + (void)stereo_music_threshold; + (void)stereo_voice_threshold; + /* Random mono/stereo decision */ + if (st->channels == 2 && (rand()&0x1F)==0) + st->stream_channels = 3-st->stream_channels; +#else + /* Rate-dependent mono-stereo decision */ + if (st->channels == 2) + { + opus_int32 stereo_threshold; + stereo_threshold = stereo_music_threshold + ((voice_est*voice_est*(stereo_voice_threshold-stereo_music_threshold))>>14); + if (st->stream_channels == 2) + stereo_threshold -= 1000; + else + stereo_threshold += 1000; + st->stream_channels = (equiv_rate > stereo_threshold) ? 2 : 1; + } else { + st->stream_channels = st->channels; + } +#endif + } + /* Update equivalent rate for channels decision. */ + equiv_rate = compute_equiv_rate(st->bitrate_bps, st->stream_channels, st->Fs/frame_size, + st->use_vbr, 0, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + + /* Allow SILK DTX if DTX is enabled but the generalized DTX cannot be used, + e.g. because of the complexity setting or sample rate. */ +#ifndef DISABLE_FLOAT_API + st->silk_mode.useDTX = st->use_dtx && !(analysis_info.valid || is_silence); +#else + st->silk_mode.useDTX = st->use_dtx && !is_silence; +#endif + + /* Mode selection depending on application and signal type */ + if (st->application == OPUS_APPLICATION_RESTRICTED_SILK) + { + st->mode = MODE_SILK_ONLY; + } else if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY || st->application == OPUS_APPLICATION_RESTRICTED_CELT) + { + st->mode = MODE_CELT_ONLY; + } else if (st->user_forced_mode == OPUS_AUTO) + { +#ifdef FUZZING + (void)stereo_width; + (void)mode_thresholds; + /* Random mode switching */ + if ((rand()&0xF)==0) + { + if ((rand()&0x1)==0) + st->mode = MODE_CELT_ONLY; + else + st->mode = MODE_SILK_ONLY; + } else { + if (st->prev_mode==MODE_CELT_ONLY) + st->mode = MODE_CELT_ONLY; + else + st->mode = MODE_SILK_ONLY; + } +#else + opus_int32 mode_voice, mode_music; + opus_int32 threshold; + + /* Interpolate based on stereo width */ + mode_voice = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[0][0]) + + MULT16_32_Q15(stereo_width,mode_thresholds[1][0])); + mode_music = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[1][1]) + + MULT16_32_Q15(stereo_width,mode_thresholds[1][1])); + /* Interpolate based on speech/music probability */ + threshold = mode_music + ((voice_est*voice_est*(mode_voice-mode_music))>>14); + /* Bias towards SILK for VoIP because of some useful features */ + if (st->application == OPUS_APPLICATION_VOIP) + threshold += 8000; + + /*printf("%f %d\n", stereo_width/(float)Q15ONE, threshold);*/ + /* Hysteresis */ + if (st->prev_mode == MODE_CELT_ONLY) + threshold -= 4000; + else if (st->prev_mode>0) + threshold += 4000; + + st->mode = (equiv_rate >= threshold) ? MODE_CELT_ONLY: MODE_SILK_ONLY; + + /* When FEC is enabled and there's enough packet loss, use SILK. + Unless the FEC is set to 2, in which case we don't switch to SILK if we're confident we have music. */ + if (st->silk_mode.useInBandFEC && st->silk_mode.packetLossPercentage > (128-voice_est)>>4 && (st->fec_config != 2 || voice_est > 25)) + st->mode = MODE_SILK_ONLY; + /* When encoding voice and DTX is enabled but the generalized DTX cannot be used, + use SILK in order to make use of its DTX. */ + if (st->silk_mode.useDTX && voice_est > 100) + st->mode = MODE_SILK_ONLY; +#endif + + /* If max_data_bytes represents less than 6 kb/s, switch to CELT-only mode */ + if (max_data_bytes < bitrate_to_bits(frame_rate > 50 ? 9000 : 6000, st->Fs, frame_size)/8) + st->mode = MODE_CELT_ONLY; + } else { + st->mode = st->user_forced_mode; + } + + /* Override the chosen mode to make sure we meet the requested frame size */ + if (st->mode != MODE_CELT_ONLY && frame_size < st->Fs/100) + { + celt_assert(st->application != OPUS_APPLICATION_RESTRICTED_SILK); + st->mode = MODE_CELT_ONLY; + } + if (st->lfe && st->application != OPUS_APPLICATION_RESTRICTED_SILK) + st->mode = MODE_CELT_ONLY; + + if (st->prev_mode > 0 && + ((st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) || + (st->mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY))) + { + redundancy = 1; + celt_to_silk = (st->mode != MODE_CELT_ONLY); + if (!celt_to_silk) + { + /* Switch to SILK/hybrid if frame size is 10 ms or more*/ + if (frame_size >= st->Fs/100) + { + st->mode = st->prev_mode; + to_celt = 1; + } else { + redundancy=0; + } + } + } + + /* When encoding multiframes, we can ask for a switch to CELT only in the last frame. This switch + * is processed above as the requested mode shouldn't interrupt stereo->mono transition. */ + if (st->stream_channels == 1 && st->prev_channels ==2 && st->silk_mode.toMono==0 + && st->mode != MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY) + { + /* Delay stereo->mono transition by two frames so that SILK can do a smooth downmix */ + st->silk_mode.toMono = 1; + st->stream_channels = 2; + } else { + st->silk_mode.toMono = 0; + } + + /* Update equivalent rate with mode decision. */ + equiv_rate = compute_equiv_rate(st->bitrate_bps, st->stream_channels, st->Fs/frame_size, + st->use_vbr, st->mode, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + + if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) + { + silk_EncControlStruct dummy; + silk_InitEncoder( silk_enc, st->channels, st->arch, &dummy); + prefill=1; + } + + /* Automatic (rate-dependent) bandwidth selection */ + if (st->mode == MODE_CELT_ONLY || st->first || st->silk_mode.allowBandwidthSwitch) + { + const opus_int32 *voice_bandwidth_thresholds, *music_bandwidth_thresholds; + opus_int32 bandwidth_thresholds[8]; + int bandwidth = OPUS_BANDWIDTH_FULLBAND; + + if (st->channels==2 && st->force_channels!=1) + { + voice_bandwidth_thresholds = stereo_voice_bandwidth_thresholds; + music_bandwidth_thresholds = stereo_music_bandwidth_thresholds; + } else { + voice_bandwidth_thresholds = mono_voice_bandwidth_thresholds; + music_bandwidth_thresholds = mono_music_bandwidth_thresholds; + } + /* Interpolate bandwidth thresholds depending on voice estimation */ + for (i=0;i<8;i++) + { + bandwidth_thresholds[i] = music_bandwidth_thresholds[i] + + ((voice_est*voice_est*(voice_bandwidth_thresholds[i]-music_bandwidth_thresholds[i]))>>14); + } + do { + int threshold, hysteresis; + threshold = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)]; + hysteresis = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)+1]; + if (!st->first) + { + if (st->auto_bandwidth >= bandwidth) + threshold -= hysteresis; + else + threshold += hysteresis; + } + if (equiv_rate >= threshold) + break; + } while (--bandwidth>OPUS_BANDWIDTH_NARROWBAND); + /* We don't use mediumband anymore, except when explicitly requested or during + mode transitions. */ + if (bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + bandwidth = OPUS_BANDWIDTH_WIDEBAND; + st->bandwidth = st->auto_bandwidth = bandwidth; + /* Prevents any transition to SWB/FB until the SILK layer has fully + switched to WB mode and turned the variable LP filter off */ + if (!st->first && st->mode != MODE_CELT_ONLY && !st->silk_mode.inWBmodeWithoutVariableLP && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + } + + if (st->bandwidth>st->max_bandwidth) + st->bandwidth = st->max_bandwidth; + + if (st->user_bandwidth != OPUS_AUTO) + st->bandwidth = st->user_bandwidth; + + /* This prevents us from using hybrid at unsafe CBR/max rates */ + if (st->mode != MODE_CELT_ONLY && max_rate < 15000) + { + st->bandwidth = IMIN(st->bandwidth, OPUS_BANDWIDTH_WIDEBAND); + } + + /* Prevents Opus from wasting bits on frequencies that are above + the Nyquist rate of the input signal */ + if (st->Fs <= 24000 && st->bandwidth > OPUS_BANDWIDTH_SUPERWIDEBAND) + st->bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + if (st->Fs <= 16000 && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + if (st->Fs <= 12000 && st->bandwidth > OPUS_BANDWIDTH_MEDIUMBAND) + st->bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + if (st->Fs <= 8000 && st->bandwidth > OPUS_BANDWIDTH_NARROWBAND) + st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; +#ifndef DISABLE_FLOAT_API + /* Use detected bandwidth to reduce the encoded bandwidth. */ + if (st->detected_bandwidth && st->user_bandwidth == OPUS_AUTO) + { + int min_detected_bandwidth; + /* Makes bandwidth detection more conservative just in case the detector + gets it wrong when we could have coded a high bandwidth transparently. + When operating in SILK/hybrid mode, we don't go below wideband to avoid + more complicated switches that require redundancy. */ + if (equiv_rate <= 18000*st->stream_channels && st->mode == MODE_CELT_ONLY) + min_detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (equiv_rate <= 24000*st->stream_channels && st->mode == MODE_CELT_ONLY) + min_detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (equiv_rate <= 30000*st->stream_channels) + min_detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (equiv_rate <= 44000*st->stream_channels) + min_detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else + min_detected_bandwidth = OPUS_BANDWIDTH_FULLBAND; + + st->detected_bandwidth = IMAX(st->detected_bandwidth, min_detected_bandwidth); + st->bandwidth = IMIN(st->bandwidth, st->detected_bandwidth); + } +#endif + st->silk_mode.LBRR_coded = decide_fec(st->silk_mode.useInBandFEC, st->silk_mode.packetLossPercentage, + st->silk_mode.LBRR_coded, st->mode, &st->bandwidth, equiv_rate); + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, OPUS_SET_LSB_DEPTH(lsb_depth)); + + /* CELT mode doesn't support mediumband, use wideband instead */ + if (st->mode == MODE_CELT_ONLY && st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + if (st->lfe) + st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; + + curr_bandwidth = st->bandwidth; + + if (st->application == OPUS_APPLICATION_RESTRICTED_SILK && curr_bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->bandwidth = curr_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + /* Chooses the appropriate mode for speech + *NEVER* switch to/from CELT-only mode here as this will invalidate some assumptions */ + if (st->mode == MODE_SILK_ONLY && curr_bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->mode = MODE_HYBRID; + if (st->mode == MODE_HYBRID && curr_bandwidth <= OPUS_BANDWIDTH_WIDEBAND) + st->mode = MODE_SILK_ONLY; + + /* Can't support higher than >60 ms frames, and >20 ms when in Hybrid or CELT-only modes */ + if ((frame_size > st->Fs/50 && (st->mode != MODE_SILK_ONLY)) || frame_size > 3*st->Fs/50) + { + int enc_frame_size; + int nb_frames; + VARDECL(unsigned char, tmp_data); + VARDECL(OpusRepacketizer, rp); + int max_header_bytes; + opus_int32 repacketize_len; + opus_int32 max_len_sum; + opus_int32 tot_size=0; + unsigned char *curr_data; + int tmp_len; + int dtx_count = 0; + int bak_to_mono; + + if (st->mode == MODE_SILK_ONLY) + { + if (frame_size == 2*st->Fs/25) /* 80 ms -> 2x 40 ms */ + enc_frame_size = st->Fs/25; + else if (frame_size == 3*st->Fs/25) /* 120 ms -> 2x 60 ms */ + enc_frame_size = 3*st->Fs/50; + else /* 100 ms -> 5x 20 ms */ + enc_frame_size = st->Fs/50; + } + else + enc_frame_size = st->Fs/50; + + nb_frames = frame_size/enc_frame_size; + +#ifndef DISABLE_FLOAT_API + if (analysis_read_pos_bak!= -1) + { + /* Reset analysis position to the beginning of the first frame so we + can use it one frame at a time. */ + st->analysis.read_pos = analysis_read_pos_bak; + st->analysis.read_subframe = analysis_read_subframe_bak; + } +#endif + + /* Worst cases: + * 2 frames: Code 2 with different compressed sizes + * >2 frames: Code 3 VBR */ + max_header_bytes = nb_frames == 2 ? 3 : (2+(nb_frames-1)*2); +#ifdef ENABLE_QEXT + /* Cover the use of the separators that are the only thing that can get us over + once we consider that we need to subtract the extension overhead in each + of the individual frames. Also consider that a separator can get our padding + from 254 to 255, which costs an extra length byte (at most once). */ + if (st->enable_qext) max_header_bytes += (nb_frames-1) + 1; +#endif + + if (st->use_vbr || st->user_bitrate_bps==OPUS_BITRATE_MAX) + repacketize_len = out_data_bytes; + else { + celt_assert(cbr_bytes>=0); + repacketize_len = IMIN(cbr_bytes, out_data_bytes); + } + max_len_sum = nb_frames + repacketize_len - max_header_bytes; + + ALLOC(tmp_data, max_len_sum, unsigned char); + curr_data = tmp_data; + ALLOC(rp, 1, OpusRepacketizer); + opus_repacketizer_init(rp); + + + bak_to_mono = st->silk_mode.toMono; + if (bak_to_mono) + st->force_channels = 1; + else + st->prev_channels = st->stream_channels; + + for (i=0;isilk_mode.toMono = 0; + st->nonfinal_frame = i<(nb_frames-1); + + /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */ + frame_to_celt = to_celt && i==nb_frames-1; + frame_redundancy = redundancy && (frame_to_celt || (!to_celt && i==0)); + + curr_max = IMIN(bitrate_to_bits(st->bitrate_bps, st->Fs, enc_frame_size)/8, max_len_sum/nb_frames); +#ifdef ENABLE_DRED + curr_max = IMIN(curr_max, (max_len_sum-bitrate_to_bits(dred_bitrate_bps, st->Fs, frame_size)/8)/nb_frames); + if (first_frame) curr_max += bitrate_to_bits(dred_bitrate_bps, st->Fs, frame_size)/8; +#endif +#ifdef ENABLE_QEXT + /* Leave room for signaling the extension size once we repacketize. */ + if (st->enable_qext) + curr_max -= curr_max/254; +#endif + curr_max = IMIN(max_len_sum-tot_size, curr_max); +#ifndef DISABLE_FLOAT_API + if (analysis_read_pos_bak != -1) { + /* Get analysis for current frame. */ + tonality_get_info(&st->analysis, &analysis_info, enc_frame_size); + } +#endif + is_silence = is_digital_silence(pcm+i*(st->channels*enc_frame_size), enc_frame_size, st->channels, lsb_depth); + + tmp_len = opus_encode_frame_native(st, pcm+i*(st->channels*enc_frame_size), enc_frame_size, curr_data, curr_max, float_api, first_frame, +#ifdef ENABLE_DRED + dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + &analysis_info, +#endif + is_silence, frame_redundancy, celt_to_silk, prefill, + equiv_rate, frame_to_celt + ); + if (tmp_len<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } else if (tmp_len==1) { + dtx_count++; + } + ret = opus_repacketizer_cat(rp, curr_data, tmp_len); + + if (ret<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + tot_size += tmp_len; + curr_data += tmp_len; + } + ret = opus_repacketizer_out_range_impl(rp, 0, nb_frames, data, repacketize_len, 0, !st->use_vbr && (dtx_count != nb_frames), NULL, 0); + if (ret<0) + { + ret = OPUS_INTERNAL_ERROR; + } + st->silk_mode.toMono = bak_to_mono; + RESTORE_STACK; + return ret; + } else { + ret = opus_encode_frame_native(st, pcm, frame_size, data, max_data_bytes, float_api, 1, +#ifdef ENABLE_DRED + dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + &analysis_info, +#endif + is_silence, redundancy, celt_to_silk, prefill, + equiv_rate, to_celt + ); + RESTORE_STACK; + return ret; + } +} + +static opus_int32 opus_encode_frame_native(OpusEncoder *st, const opus_res *pcm, int frame_size, + unsigned char *data, opus_int32 orig_max_data_bytes, + int float_api, int first_frame, +#ifdef ENABLE_DRED + opus_int32 dred_bitrate_bps, +#endif +#ifndef DISABLE_FLOAT_API + AnalysisInfo *analysis_info, +#endif + int is_silence, int redundancy, int celt_to_silk, int prefill, + opus_int32 equiv_rate, int to_celt) +{ + void *silk_enc=NULL; + CELTEncoder *celt_enc=NULL; + const CELTMode *celt_mode=NULL; + int i; + int ret=0; + int max_data_bytes; + opus_int32 nBytes; + ec_enc enc; + int bits_target; + int start_band = 0; + int redundancy_bytes = 0; /* Number of bytes to use for redundancy frame */ + int nb_compr_bytes; + opus_uint32 redundant_rng = 0; + int cutoff_Hz; + int hp_freq_smth1; + opus_val16 HB_gain; + int apply_padding; + int frame_rate; + int curr_bandwidth; + int delay_compensation; + int total_buffer; + opus_int activity = VAD_NO_DECISION; + VARDECL(opus_res, pcm_buf); + VARDECL(opus_res, tmp_prefill); + SAVE_STACK; + + max_data_bytes = IMIN(orig_max_data_bytes, 1276); + st->rangeFinal = 0; + if (st->application != OPUS_APPLICATION_RESTRICTED_CELT) + silk_enc = (char*)st+st->silk_enc_offset; + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); + } + curr_bandwidth = st->bandwidth; + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY || st->application == OPUS_APPLICATION_RESTRICTED_CELT || st->application == OPUS_APPLICATION_RESTRICTED_SILK) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + total_buffer = delay_compensation; + + frame_rate = st->Fs/frame_size; + + if (is_silence) + { + activity = !is_silence; + } +#ifndef DISABLE_FLOAT_API + else if (analysis_info->valid) { + activity = analysis_info->activity_probability >= DTX_ACTIVITY_THRESHOLD; + if (!activity) + { + /* Mark as active if this noise frame is sufficiently loud */ + opus_val32 noise_energy = compute_frame_energy(pcm, frame_size, st->channels, st->arch); + activity = st->peak_signal_energy < (PSEUDO_SNR_THRESHOLD * noise_energy); + } + } +#endif + else if (st->mode == MODE_CELT_ONLY) { + opus_val32 noise_energy = compute_frame_energy(pcm, frame_size, st->channels, st->arch); + /* Boosting peak energy a bit because we didn't just average the active frames. */ + activity = st->peak_signal_energy < (QCONST16(PSEUDO_SNR_THRESHOLD, 0) * (opus_val64)HALF32(noise_energy)); + } + + /* For the first frame at a new SILK bandwidth */ + if (st->silk_bw_switch) + { + redundancy = 1; + celt_to_silk = 1; + st->silk_bw_switch = 0; + /* Do a prefill without resetting the sampling rate control. */ + prefill=2; + } + + /* If we decided to go with CELT, make sure redundancy is off, no matter what + we decided earlier. */ + if (st->mode == MODE_CELT_ONLY) + redundancy = 0; + + if (redundancy) + { + redundancy_bytes = compute_redundancy_bytes(max_data_bytes, st->bitrate_bps, frame_rate, st->stream_channels); + if (redundancy_bytes == 0) + redundancy = 0; + } + if (st->application == OPUS_APPLICATION_RESTRICTED_SILK) + { + redundancy = 0; + redundancy_bytes = 0; + } + + /* printf("%d %d %d %d\n", st->bitrate_bps, st->stream_channels, st->mode, curr_bandwidth); */ + bits_target = IMIN(8*(max_data_bytes-redundancy_bytes), bitrate_to_bits(st->bitrate_bps, st->Fs, frame_size)) - 8; + + data += 1; + + ec_enc_init(&enc, data, orig_max_data_bytes-1); + + ALLOC(pcm_buf, (total_buffer+frame_size)*st->channels, opus_res); + OPUS_COPY(pcm_buf, &st->delay_buffer[(st->encoder_buffer-total_buffer)*st->channels], total_buffer*st->channels); + + if (st->mode == MODE_CELT_ONLY) + hp_freq_smth1 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); + else + hp_freq_smth1 = ((silk_encoder*)silk_enc)->state_Fxx[0].sCmn.variable_HP_smth1_Q15; + + st->variable_HP_smth2_Q15 = silk_SMLAWB( st->variable_HP_smth2_Q15, + hp_freq_smth1 - st->variable_HP_smth2_Q15, SILK_FIX_CONST( VARIABLE_HP_SMTH_COEF2, 16 ) ); + + /* convert from log scale to Hertz */ + cutoff_Hz = silk_log2lin( silk_RSHIFT( st->variable_HP_smth2_Q15, 8 ) ); + + if (st->application == OPUS_APPLICATION_VOIP) + { + hp_cutoff(pcm, cutoff_Hz, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs, st->arch); + +#ifdef ENABLE_OSCE_TRAINING_DATA + /* write out high pass filtered clean signal*/ + static FILE *fout =NULL; + if (fout == NULL) + { + fout = fopen("clean_hp.s16", "wb"); + } + + { + int idx; + opus_int16 tmp; + for (idx = 0; idx < frame_size; idx++) + { + tmp = (opus_int16) (32768 * pcm_buf[total_buffer + idx] + 0.5f); + fwrite(&tmp, sizeof(tmp), 1, fout); + } + } +#endif + } else { +#ifdef ENABLE_QEXT + /* FIXME: Avoid glitching when we switch qext on/off dynamically. */ + if (st->enable_qext) OPUS_COPY(&pcm_buf[total_buffer*st->channels], pcm, frame_size*st->channels); + else +#endif + dc_reject(pcm, 3, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); + } +#ifndef FIXED_POINT + if (float_api) + { + opus_val32 sum; + sum = celt_inner_prod(&pcm_buf[total_buffer*st->channels], &pcm_buf[total_buffer*st->channels], frame_size*st->channels, st->arch); + /* This should filter out both NaNs and ridiculous signals that could + cause NaNs further down. */ + if (!(sum < 1e9f) || celt_isnan(sum)) + { + OPUS_CLEAR(&pcm_buf[total_buffer*st->channels], frame_size*st->channels); + st->hp_mem[0] = st->hp_mem[1] = st->hp_mem[2] = st->hp_mem[3] = 0; + } + } +#else + (void)float_api; +#endif + +#ifdef ENABLE_DRED + /* Compute the DRED features. Needs to be before SILK because of DTX. */ + if ( st->dred_duration > 0 && st->dred_encoder.loaded ) { + int frame_size_400Hz; + /* DRED Encoder */ + dred_compute_latents( &st->dred_encoder, &pcm_buf[total_buffer*st->channels], frame_size, total_buffer, st->arch ); + frame_size_400Hz = frame_size*400/st->Fs; + OPUS_MOVE(&st->activity_mem[frame_size_400Hz], st->activity_mem, 4*DRED_MAX_FRAMES-frame_size_400Hz); + for (i=0;iactivity_mem[i] = activity; + } else { + st->dred_encoder.latents_buffer_fill = 0; + OPUS_CLEAR(st->activity_mem, 4*DRED_MAX_FRAMES); + } +#endif + + /* SILK processing */ + HB_gain = Q15ONE; + if (st->mode != MODE_CELT_ONLY) + { + opus_int32 total_bitRate, celt_rate; + const opus_res *pcm_silk; + + /* Distribute bits between SILK and CELT */ + total_bitRate = bits_to_bitrate(bits_target, st->Fs, frame_size); + if( st->mode == MODE_HYBRID ) { + /* Base rate for SILK */ + st->silk_mode.bitRate = compute_silk_rate_for_hybrid(total_bitRate, + curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded, + st->stream_channels); + if (!st->energy_masking) + { + /* Increasingly attenuate high band when it gets allocated fewer bits */ + celt_rate = total_bitRate - st->silk_mode.bitRate; + HB_gain = Q15ONE - SHR32(celt_exp2(-celt_rate * QCONST16(1.f/1024, 10)), 1); + } + } else { + /* SILK gets all bits */ + st->silk_mode.bitRate = total_bitRate; + } + + /* Surround masking for SILK */ + if (st->energy_masking && st->use_vbr && !st->lfe) + { + opus_val32 mask_sum=0; + celt_glog masking_depth; + opus_int32 rate_offset; + int c; + int end = 17; + opus_int16 srate = 16000; + if (st->bandwidth == OPUS_BANDWIDTH_NARROWBAND) + { + end = 13; + srate = 8000; + } else if (st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + { + end = 15; + srate = 12000; + } + for (c=0;cchannels;c++) + { + for(i=0;ienergy_masking[21*c+i], + GCONST(.5f)), -GCONST(2.0f)); + if (mask > 0) + mask = HALF32(mask); + mask_sum += mask; + } + } + /* Conservative rate reduction, we cut the masking in half */ + masking_depth = mask_sum / end*st->channels; + masking_depth += GCONST(.2f); + rate_offset = (opus_int32)PSHR32(MULT16_16(srate, SHR32(masking_depth, DB_SHIFT-10)), 10); + rate_offset = MAX32(rate_offset, -2*st->silk_mode.bitRate/3); + /* Split the rate change between the SILK and CELT part for hybrid. */ + if (st->bandwidth==OPUS_BANDWIDTH_SUPERWIDEBAND || st->bandwidth==OPUS_BANDWIDTH_FULLBAND) + st->silk_mode.bitRate += 3*rate_offset/5; + else + st->silk_mode.bitRate += rate_offset; + } + + st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs; + st->silk_mode.nChannelsAPI = st->channels; + st->silk_mode.nChannelsInternal = st->stream_channels; + if (curr_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { + st->silk_mode.desiredInternalSampleRate = 8000; + } else if (curr_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { + st->silk_mode.desiredInternalSampleRate = 12000; + } else { + celt_assert( st->mode == MODE_HYBRID || curr_bandwidth == OPUS_BANDWIDTH_WIDEBAND ); + st->silk_mode.desiredInternalSampleRate = 16000; + } + if( st->mode == MODE_HYBRID ) { + /* Don't allow bandwidth reduction at lowest bitrates in hybrid mode */ + st->silk_mode.minInternalSampleRate = 16000; + } else { + st->silk_mode.minInternalSampleRate = 8000; + } + + st->silk_mode.maxInternalSampleRate = 16000; + if (st->mode == MODE_SILK_ONLY) + { + opus_int32 effective_max_rate = bits_to_bitrate(max_data_bytes*8, st->Fs, frame_size); + if (frame_rate > 50) + effective_max_rate = effective_max_rate*2/3; + if (effective_max_rate < 8000) + { + st->silk_mode.maxInternalSampleRate = 12000; + st->silk_mode.desiredInternalSampleRate = IMIN(12000, st->silk_mode.desiredInternalSampleRate); + } + if (effective_max_rate < 7000) + { + st->silk_mode.maxInternalSampleRate = 8000; + st->silk_mode.desiredInternalSampleRate = IMIN(8000, st->silk_mode.desiredInternalSampleRate); + } +#ifdef ENABLE_QEXT + /* At 96 kHz, we don't have the input resampler to do 8 or 12 kHz. */ + if (st->Fs==96000) st->silk_mode.maxInternalSampleRate = st->silk_mode.desiredInternalSampleRate = 16000; +#endif + } + + st->silk_mode.useCBR = !st->use_vbr; + + /* Call SILK encoder for the low band */ + + /* Max bits for SILK, counting ToC, redundancy bytes, and optionally redundancy. */ + st->silk_mode.maxBits = (max_data_bytes-1)*8; + if (redundancy && redundancy_bytes >= 2) + { + /* Counting 1 bit for redundancy position and 20 bits for flag+size (only for hybrid). */ + st->silk_mode.maxBits -= redundancy_bytes*8 + 1; + if (st->mode == MODE_HYBRID) + st->silk_mode.maxBits -= 20; + } + if (st->silk_mode.useCBR) + { + /* When we're in CBR mode, but we have non-SILK data to encode, switch SILK to VBR with cap to + save on complexity. Any variations will be absorbed by CELT and/or DRED and we can still + produce a constant bitrate without wasting bits. */ +#ifdef ENABLE_DRED + if (st->mode == MODE_HYBRID || dred_bitrate_bps > 0) +#else + if (st->mode == MODE_HYBRID) +#endif + { + /* Allow SILK to steal up to 25% of the remaining bits */ + opus_int16 other_bits = IMAX(0, st->silk_mode.maxBits - st->silk_mode.bitRate * frame_size / st->Fs); + st->silk_mode.maxBits = IMAX(0, st->silk_mode.maxBits - other_bits*3/4); + st->silk_mode.useCBR = 0; + } + } else { + /* Constrained VBR. */ + if (st->mode == MODE_HYBRID) + { + /* Compute SILK bitrate corresponding to the max total bits available */ + opus_int32 maxBitRate = compute_silk_rate_for_hybrid(st->silk_mode.maxBits*st->Fs / frame_size, + curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded, + st->stream_channels); + st->silk_mode.maxBits = bitrate_to_bits(maxBitRate, st->Fs, frame_size); + } + } + + if (prefill && st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + opus_int32 zero=0; + int prefill_offset; + /* Use a smooth onset for the SILK prefill to avoid the encoder trying to encode + a discontinuity. The exact location is what we need to avoid leaving any "gap" + in the audio when mixing with the redundant CELT frame. Here we can afford to + overwrite st->delay_buffer because the only thing that uses it before it gets + rewritten is tmp_prefill[] and even then only the part after the ramp really + gets used (rather than sent to the encoder and discarded) */ + prefill_offset = st->channels*(st->encoder_buffer-st->delay_compensation-st->Fs/400); + gain_fade(st->delay_buffer+prefill_offset, st->delay_buffer+prefill_offset, + 0, Q15ONE, celt_mode->overlap, st->Fs/400, st->channels, celt_mode->window, st->Fs); + OPUS_CLEAR(st->delay_buffer, prefill_offset); + pcm_silk = st->delay_buffer; + silk_Encode( silk_enc, &st->silk_mode, pcm_silk, st->encoder_buffer, NULL, &zero, prefill, activity ); + /* Prevent a second switch in the real encode call. */ + st->silk_mode.opusCanSwitch = 0; + } + + pcm_silk = pcm_buf+total_buffer*st->channels; + ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0, activity ); + if( ret ) { + /*fprintf (stderr, "SILK encode error: %d\n", ret);*/ + /* Handle error */ + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + + /* Extract SILK internal bandwidth for signaling in first byte */ + if( st->mode == MODE_SILK_ONLY ) { + if( st->silk_mode.internalSampleRate == 8000 ) { + curr_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + } else if( st->silk_mode.internalSampleRate == 12000 ) { + curr_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + } else if( st->silk_mode.internalSampleRate == 16000 ) { + curr_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + } + } else { + celt_assert( st->silk_mode.internalSampleRate == 16000 ); + } + + st->silk_mode.opusCanSwitch = st->silk_mode.switchReady && !st->nonfinal_frame; + + if (activity == VAD_NO_DECISION) { + activity = (st->silk_mode.signalType != TYPE_NO_VOICE_ACTIVITY); +#ifdef ENABLE_DRED + for (i=0;iFs;i++) + st->activity_mem[i] = activity; +#endif + } + if (nBytes==0) + { + st->rangeFinal = 0; + data[-1] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); + RESTORE_STACK; + return 1; + } + + /* FIXME: How do we allocate the redundancy for CBR? */ + if (st->silk_mode.opusCanSwitch) + { + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + redundancy_bytes = compute_redundancy_bytes(max_data_bytes, st->bitrate_bps, frame_rate, st->stream_channels); + redundancy = (redundancy_bytes != 0); + } + celt_to_silk = 0; + st->silk_bw_switch = 1; + } + } + + /* CELT processing */ + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + int endband=21; + + switch(curr_bandwidth) + { + case OPUS_BANDWIDTH_NARROWBAND: + endband = 13; + break; + case OPUS_BANDWIDTH_MEDIUMBAND: + case OPUS_BANDWIDTH_WIDEBAND: + endband = 17; + break; + case OPUS_BANDWIDTH_SUPERWIDEBAND: + endband = 19; + break; + case OPUS_BANDWIDTH_FULLBAND: + endband = 21; + break; + } + celt_encoder_ctl(celt_enc, CELT_SET_END_BAND(endband)); + celt_encoder_ctl(celt_enc, CELT_SET_CHANNELS(st->stream_channels)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); + } + if (st->mode != MODE_SILK_ONLY) + { + opus_val32 celt_pred=2; + /* We may still decide to disable prediction later */ + if (st->silk_mode.reducedDependency) + celt_pred = 0; + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(celt_pred)); + } + + ALLOC(tmp_prefill, st->channels*st->Fs/400, opus_res); + if (st->mode != MODE_SILK_ONLY && st->mode != st->prev_mode && st->prev_mode > 0 + && st->application != OPUS_APPLICATION_RESTRICTED_CELT) + { + OPUS_COPY(tmp_prefill, &st->delay_buffer[(st->encoder_buffer-total_buffer-st->Fs/400)*st->channels], st->channels*st->Fs/400); + } + + if (st->channels*(st->encoder_buffer-(frame_size+total_buffer)) > 0) + { + OPUS_MOVE(st->delay_buffer, &st->delay_buffer[st->channels*frame_size], st->channels*(st->encoder_buffer-frame_size-total_buffer)); + OPUS_COPY(&st->delay_buffer[st->channels*(st->encoder_buffer-frame_size-total_buffer)], + &pcm_buf[0], + (frame_size+total_buffer)*st->channels); + } else { + OPUS_COPY(st->delay_buffer, &pcm_buf[(frame_size+total_buffer-st->encoder_buffer)*st->channels], st->encoder_buffer*st->channels); + } + /* gain_fade() and stereo_fade() need to be after the buffer copying + because we don't want any of this to affect the SILK part */ + if( ( st->prev_HB_gain < Q15ONE || HB_gain < Q15ONE ) && celt_mode != NULL ) { + gain_fade(pcm_buf, pcm_buf, + st->prev_HB_gain, HB_gain, celt_mode->overlap, frame_size, st->channels, celt_mode->window, st->Fs); + } + st->prev_HB_gain = HB_gain; + if (st->mode != MODE_HYBRID || st->stream_channels==1) + { + if (equiv_rate > 32000) + st->silk_mode.stereoWidth_Q14 = 16384; + else if (equiv_rate < 16000) + st->silk_mode.stereoWidth_Q14 = 0; + else + st->silk_mode.stereoWidth_Q14 = 16384 - 2048*(opus_int32)(32000-equiv_rate)/(equiv_rate-14000); + } + if( !st->energy_masking && st->channels == 2 ) { + /* Apply stereo width reduction (at low bitrates) */ + if( st->hybrid_stereo_width_Q14 < (1 << 14) || st->silk_mode.stereoWidth_Q14 < (1 << 14) ) { + opus_val16 g1, g2; + g1 = st->hybrid_stereo_width_Q14; + g2 = (opus_val16)(st->silk_mode.stereoWidth_Q14); +#ifdef FIXED_POINT + g1 = g1==16384 ? Q15ONE : SHL16(g1,1); + g2 = g2==16384 ? Q15ONE : SHL16(g2,1); +#else + g1 *= (1.f/16384); + g2 *= (1.f/16384); +#endif + if ( celt_mode != NULL ) + { + stereo_fade(pcm_buf, pcm_buf, g1, g2, celt_mode->overlap, + frame_size, st->channels, celt_mode->window, st->Fs); + } + st->hybrid_stereo_width_Q14 = st->silk_mode.stereoWidth_Q14; + } + } + + if ( st->mode != MODE_CELT_ONLY && ec_tell(&enc)+17+20*(st->mode == MODE_HYBRID) <= 8*(max_data_bytes-1)) + { + /* For SILK mode, the redundancy is inferred from the length */ + if (st->mode == MODE_HYBRID) + ec_enc_bit_logp(&enc, redundancy, 12); + if (redundancy) + { + int max_redundancy; + ec_enc_bit_logp(&enc, celt_to_silk, 1); + if (st->mode == MODE_HYBRID) + { + /* Reserve the 8 bits needed for the redundancy length, + and at least a few bits for CELT if possible */ + max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+8+3+7)>>3); + } + else + max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+7)>>3); + /* Target the same bit-rate for redundancy as for the rest, + up to a max of 257 bytes */ + redundancy_bytes = IMIN(max_redundancy, redundancy_bytes); + redundancy_bytes = IMIN(257, IMAX(2, redundancy_bytes)); + if (st->mode == MODE_HYBRID) + ec_enc_uint(&enc, redundancy_bytes-2, 256); + } + } else { + redundancy = 0; + } + + if (!redundancy) + { + st->silk_bw_switch = 0; + redundancy_bytes = 0; + } + if (st->mode != MODE_CELT_ONLY)start_band=17; + + if (st->mode == MODE_SILK_ONLY) + { + ret = (ec_tell(&enc)+7)>>3; + ec_enc_done(&enc); + nb_compr_bytes = ret; + } else { + nb_compr_bytes = (max_data_bytes-1)-redundancy_bytes; +#ifdef ENABLE_QEXT + if (st->mode == MODE_CELT_ONLY && st->enable_qext) { + celt_assert(redundancy_bytes==0); + nb_compr_bytes = orig_max_data_bytes-1; + } +#endif +#ifdef ENABLE_DRED + if (st->dred_duration > 0) + { + int max_celt_bytes; + opus_int32 dred_bytes = bitrate_to_bits(dred_bitrate_bps, st->Fs, frame_size)/8; + /* Allow CELT to steal up to 25% of the remaining bits. */ + max_celt_bytes = nb_compr_bytes - dred_bytes*3/4; + /* But try to give CELT at least 5 bytes to prevent a mismatch with + the redundancy signaling. */ + max_celt_bytes = IMAX((ec_tell(&enc)+7)/8 + 5, max_celt_bytes); + /* Subject to the original max. */ + nb_compr_bytes = IMIN(nb_compr_bytes, max_celt_bytes); + } +#endif + ec_enc_shrink(&enc, nb_compr_bytes); + } + +#ifndef DISABLE_FLOAT_API + if (redundancy || st->mode != MODE_SILK_ONLY) + celt_encoder_ctl(celt_enc, CELT_SET_ANALYSIS(analysis_info)); +#endif + if (st->mode == MODE_HYBRID) { + SILKInfo info; + info.signalType = st->silk_mode.signalType; + info.offset = st->silk_mode.offset; + celt_encoder_ctl(celt_enc, CELT_SET_SILK_INFO(&info)); + } + + /* 5 ms redundant frame for CELT->SILK */ + if (redundancy && celt_to_silk) + { + int err; + celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); + err = celt_encode_with_ec(celt_enc, pcm_buf, st->Fs/200, data+nb_compr_bytes, redundancy_bytes, NULL); + if (err < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng)); + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + } + + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(start_band)); + + data[-1] = 0; + if (st->mode != MODE_SILK_ONLY) + { + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(st->use_vbr)); + if (st->mode == MODE_HYBRID) + { + if( st->use_vbr ) { + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps-st->silk_mode.bitRate)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(0)); + } + } else { + if (st->use_vbr) + { + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps)); + } + } +#ifdef ENABLE_DRED + /* When Using DRED CBR, we can actually make the CELT part VBR and have DRED pick up the slack. */ + if (!st->use_vbr && st->dred_duration > 0) + { + opus_int32 celt_bitrate = st->bitrate_bps; + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(0)); + if (st->mode == MODE_HYBRID) { + celt_bitrate -= st->silk_mode.bitRate; + } + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(celt_bitrate)); + } +#endif + if (st->mode != st->prev_mode && st->prev_mode > 0 && st->application != OPUS_APPLICATION_RESTRICTED_CELT) + { + unsigned char dummy[2]; + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + + /* Prefilling */ + celt_encode_with_ec(celt_enc, tmp_prefill, st->Fs/400, dummy, 2, NULL); + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0)); + } + /* If false, we already busted the budget and we'll end up with a "PLC frame" */ + if (ec_tell(&enc) <= 8*nb_compr_bytes) + { +#ifdef ENABLE_QEXT + if (st->mode == MODE_CELT_ONLY) celt_encoder_ctl(celt_enc, OPUS_SET_QEXT(st->enable_qext)); +#endif + ret = celt_encode_with_ec(celt_enc, pcm_buf, frame_size, NULL, nb_compr_bytes, &enc); +#ifdef ENABLE_QEXT + celt_encoder_ctl(celt_enc, OPUS_SET_QEXT(0)); +#endif + if (ret < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + /* Put CELT->SILK redundancy data in the right place. */ + if (redundancy && celt_to_silk && st->mode==MODE_HYBRID && nb_compr_bytes != ret) + { + OPUS_MOVE(data+ret, data+nb_compr_bytes, redundancy_bytes); + nb_compr_bytes = ret+redundancy_bytes; + } + } + celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&st->rangeFinal)); + } else { + st->rangeFinal = enc.rng; + } + + /* 5 ms redundant frame for SILK->CELT */ + if (redundancy && !celt_to_silk) + { + int err; + unsigned char dummy[2]; + int N2, N4; + N2 = st->Fs/200; + N4 = st->Fs/400; + + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0)); + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); + + if (st->mode == MODE_HYBRID) + { + /* Shrink packet to what the encoder actually used. */ + nb_compr_bytes = ret; + ec_enc_shrink(&enc, nb_compr_bytes); + } + /* NOTE: We could speed this up slightly (at the expense of code size) by just adding a function that prefills the buffer */ + celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2-N4), N4, dummy, 2, NULL); + + err = celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2), N2, data+nb_compr_bytes, redundancy_bytes, NULL); + if (err < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng)); + } + + + + /* Signalling the mode in the first byte */ + data--; + data[0] |= gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); + + st->rangeFinal ^= redundant_rng; + + if (to_celt) + st->prev_mode = MODE_CELT_ONLY; + else + st->prev_mode = st->mode; + st->prev_channels = st->stream_channels; + st->prev_framesize = frame_size; + + st->first = 0; + + /* DTX decision */ + if (st->use_dtx && !st->silk_mode.useDTX) + { + if (decide_dtx_mode(activity, &st->nb_no_activity_ms_Q1, 2*1000*frame_size/st->Fs)) + { + st->rangeFinal = 0; + data[0] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); + RESTORE_STACK; + return 1; + } + } else { + st->nb_no_activity_ms_Q1 = 0; + } + + /* In the unlikely case that the SILK encoder busted its target, tell + the decoder to call the PLC */ + if (ec_tell(&enc) > (max_data_bytes-1)*8) + { + if (max_data_bytes < 2) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + data[1] = 0; + ret = 1; + st->rangeFinal = 0; + } else if (st->mode==MODE_SILK_ONLY&&!redundancy) + { + /*When in LPC only mode it's perfectly + reasonable to strip off trailing zero bytes as + the required range decoder behavior is to + fill these in. This can't be done when the MDCT + modes are used because the decoder needs to know + the actual length for allocation purposes.*/ + while(ret>2&&data[ret]==0)ret--; + } + /* Count ToC and redundancy */ + ret += 1+redundancy_bytes; + apply_padding = !st->use_vbr; +#ifdef ENABLE_DRED + if (st->dred_duration > 0 && st->dred_encoder.loaded && first_frame) { + opus_extension_data extension; + unsigned char buf[DRED_MAX_DATA_SIZE]; + int dred_chunks; + int dred_bytes_left; + dred_chunks = IMIN((st->dred_duration+5)/4, DRED_NUM_REDUNDANCY_FRAMES/2); + if (st->use_vbr) dred_chunks = IMIN(dred_chunks, st->dred_target_chunks); + /* Remaining space for DRED, accounting for cost the 3 extra bytes for code 3, padding length, and extension number. */ + dred_bytes_left = IMIN(DRED_MAX_DATA_SIZE, orig_max_data_bytes-ret-3); + /* Account for the extra bytes required to signal large padding length. */ + dred_bytes_left -= (dred_bytes_left+1+DRED_EXPERIMENTAL_BYTES)/255; + /* Check whether we actually have something to encode. */ + if (dred_chunks >= 1 && dred_bytes_left >= DRED_MIN_BYTES+DRED_EXPERIMENTAL_BYTES) { + int dred_bytes; +#ifdef DRED_EXPERIMENTAL_VERSION + /* Add temporary extension type and version. + These bytes will be removed once extension is finalized. */ + buf[0] = 'D'; + buf[1] = DRED_EXPERIMENTAL_VERSION; +#endif + dred_bytes = dred_encode_silk_frame(&st->dred_encoder, buf+DRED_EXPERIMENTAL_BYTES, dred_chunks, dred_bytes_left-DRED_EXPERIMENTAL_BYTES, + st->dred_q0, st->dred_dQ, st->dred_qmax, st->activity_mem, st->arch); + if (dred_bytes > 0) { + dred_bytes += DRED_EXPERIMENTAL_BYTES; + celt_assert(dred_bytes <= dred_bytes_left); + extension.id = DRED_EXTENSION_ID; + extension.frame = 0; + extension.data = buf; + extension.len = dred_bytes; + ret = opus_packet_pad_impl(data, ret, orig_max_data_bytes, !st->use_vbr, &extension, 1); + if (ret < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + apply_padding = 0; + } + } + } +#else + (void)first_frame; /* Avoids a warning about first_frame being unused. */ +#endif + if (apply_padding) + { + if (opus_packet_pad(data, ret, orig_max_data_bytes) != OPUS_OK) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + ret = orig_max_data_bytes; + } + RESTORE_STACK; + return ret; +} + + + +#if defined(FIXED_POINT) && !defined(ENABLE_RES24) +opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int frame_size; + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, st->Fs); + return opus_encode_native(st, pcm, frame_size, data, max_data_bytes, 16, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_int, 0); +} +#else +opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int i, ret; + int frame_size; + VARDECL(opus_res, in); + ALLOC_STACK; + + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, st->Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + ALLOC(in, frame_size*st->channels, opus_res); + + for (i=0;ichannels;i++) + in[i] = INT16TORES(pcm[i]); + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_int, 1); + RESTORE_STACK; + return ret; +} +#endif + +#if defined(FIXED_POINT) && defined(ENABLE_RES24) +opus_int32 opus_encode24(OpusEncoder *st, const opus_int32 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int frame_size; + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, st->Fs); + return opus_encode_native(st, pcm, frame_size, data, max_data_bytes, MAX_ENCODING_DEPTH, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_int24, 0); +} +#else +opus_int32 opus_encode24(OpusEncoder *st, const opus_int32 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int i, ret; + int frame_size; + VARDECL(opus_res, in); + ALLOC_STACK; + + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, st->Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + ALLOC(in, frame_size*st->channels, opus_res); + + for (i=0;ichannels;i++) + in[i] = INT24TORES(pcm[i]); + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, MAX_ENCODING_DEPTH, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_int24, 1); + RESTORE_STACK; + return ret; +} +#endif + + +#ifndef DISABLE_FLOAT_API + +# if !defined(FIXED_POINT) +opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 out_data_bytes) +{ + int frame_size; + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, st->Fs); + return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, MAX_ENCODING_DEPTH, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_float, 1); +} +# else +opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int i, ret; + int frame_size; + VARDECL(opus_res, in); + ALLOC_STACK; + + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, st->Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + ALLOC(in, frame_size*st->channels, opus_res); + + for (i=0;ichannels;i++) + in[i] = FLOAT2RES(pcm[i]); + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, MAX_ENCODING_DEPTH, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_float, 1); + RESTORE_STACK; + return ret; +} +# endif + +#endif + + +int opus_encoder_ctl(OpusEncoder *st, int request, ...) +{ + int ret; + CELTEncoder *celt_enc=NULL; + va_list ap; + + ret = OPUS_OK; + va_start(ap, request); + + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + + switch (request) + { + case OPUS_SET_APPLICATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (st->application == OPUS_APPLICATION_RESTRICTED_SILK || st->application == OPUS_APPLICATION_RESTRICTED_CELT) + { + ret = OPUS_BAD_ARG; + break; + } + if ( (value != OPUS_APPLICATION_VOIP && value != OPUS_APPLICATION_AUDIO + && value != OPUS_APPLICATION_RESTRICTED_LOWDELAY) + || (!st->first && st->application != value)) + { + ret = OPUS_BAD_ARG; + break; + } + st->application = value; +#ifndef DISABLE_FLOAT_API + st->analysis.application = value; +#endif + } + break; + case OPUS_GET_APPLICATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->application; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_AUTO && value != OPUS_BITRATE_MAX) + { + if (value <= 0) + goto bad_arg; + else if (value <= 500) + value = 500; + else if (value > (opus_int32)750000*st->channels) + value = (opus_int32)750000*st->channels; + } + st->user_bitrate_bps = value; + } + break; + case OPUS_GET_BITRATE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = user_bitrate_to_bitrate(st, st->prev_framesize, 1276); + } + break; + case OPUS_SET_FORCE_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if((value<1 || value>st->channels) && value != OPUS_AUTO) + { + goto bad_arg; + } + st->force_channels = value; + } + break; + case OPUS_GET_FORCE_CHANNELS_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->force_channels; + } + break; + case OPUS_SET_MAX_BANDWIDTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) + { + goto bad_arg; + } + st->max_bandwidth = value; + if (st->max_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { + st->silk_mode.maxInternalSampleRate = 8000; + } else if (st->max_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { + st->silk_mode.maxInternalSampleRate = 12000; + } else { + st->silk_mode.maxInternalSampleRate = 16000; + } + } + break; + case OPUS_GET_MAX_BANDWIDTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->max_bandwidth; + } + break; + case OPUS_SET_BANDWIDTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if ((value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) && value != OPUS_AUTO) + { + goto bad_arg; + } + st->user_bandwidth = value; + if (st->user_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { + st->silk_mode.maxInternalSampleRate = 8000; + } else if (st->user_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { + st->silk_mode.maxInternalSampleRate = 12000; + } else { + st->silk_mode.maxInternalSampleRate = 16000; + } + } + break; + case OPUS_GET_BANDWIDTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->bandwidth; + } + break; + case OPUS_SET_DTX_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->use_dtx = value; + } + break; + case OPUS_GET_DTX_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->use_dtx; + } + break; + case OPUS_SET_COMPLEXITY_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>10) + { + goto bad_arg; + } + st->silk_mode.complexity = value; + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(value)); + } + break; + case OPUS_GET_COMPLEXITY_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->silk_mode.complexity; + } + break; + case OPUS_SET_INBAND_FEC_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>2) + { + goto bad_arg; + } + st->fec_config = value; + st->silk_mode.useInBandFEC = (value != 0); + } + break; + case OPUS_GET_INBAND_FEC_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->fec_config; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value < 0 || value > 100) + { + goto bad_arg; + } + st->silk_mode.packetLossPercentage = value; + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, OPUS_SET_PACKET_LOSS_PERC(value)); + } + break; + case OPUS_GET_PACKET_LOSS_PERC_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->silk_mode.packetLossPercentage; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->use_vbr = value; + st->silk_mode.useCBR = 1-value; + } + break; + case OPUS_GET_VBR_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->use_vbr; + } + break; + case OPUS_SET_VOICE_RATIO_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<-1 || value>100) + { + goto bad_arg; + } + st->voice_ratio = value; + } + break; + case OPUS_GET_VOICE_RATIO_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->voice_ratio; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->vbr_constraint = value; + } + break; + case OPUS_GET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->vbr_constraint; + } + break; + case OPUS_SET_SIGNAL_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value!=OPUS_AUTO && value!=OPUS_SIGNAL_VOICE && value!=OPUS_SIGNAL_MUSIC) + { + goto bad_arg; + } + st->signal_type = value; + } + break; + case OPUS_GET_SIGNAL_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->signal_type; + } + break; + case OPUS_GET_LOOKAHEAD_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->Fs/400; + if (st->application != OPUS_APPLICATION_RESTRICTED_LOWDELAY && st->application != OPUS_APPLICATION_RESTRICTED_CELT) + *value += st->delay_compensation; + } + break; + case OPUS_GET_SAMPLE_RATE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->Fs; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 *value = va_arg(ap, opus_uint32*); + if (!value) + { + goto bad_arg; + } + *value = st->rangeFinal; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + { + goto bad_arg; + } + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->lsb_depth; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_FRAMESIZE_ARG && value != OPUS_FRAMESIZE_2_5_MS && + value != OPUS_FRAMESIZE_5_MS && value != OPUS_FRAMESIZE_10_MS && + value != OPUS_FRAMESIZE_20_MS && value != OPUS_FRAMESIZE_40_MS && + value != OPUS_FRAMESIZE_60_MS && value != OPUS_FRAMESIZE_80_MS && + value != OPUS_FRAMESIZE_100_MS && value != OPUS_FRAMESIZE_120_MS) + { + goto bad_arg; + } + st->variable_duration = value; + } + break; + case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->variable_duration; + } + break; + case OPUS_SET_PREDICTION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value > 1 || value < 0) + goto bad_arg; + st->silk_mode.reducedDependency = value; + } + break; + case OPUS_GET_PREDICTION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + goto bad_arg; + *value = st->silk_mode.reducedDependency; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, OPUS_SET_PHASE_INVERSION_DISABLED(value)); + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, OPUS_GET_PHASE_INVERSION_DISABLED(value)); + else + *value = 0; + } + break; +#ifdef ENABLE_DRED + case OPUS_SET_DRED_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>DRED_MAX_FRAMES) + { + goto bad_arg; + } + st->dred_duration = value; + st->silk_mode.useDRED = !!value; + } + break; + case OPUS_GET_DRED_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->dred_duration; + } + break; +#endif +#ifdef ENABLE_QEXT + case OPUS_SET_QEXT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->enable_qext = value; + } + break; + case OPUS_GET_QEXT_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->enable_qext; + } + break; +#endif + case OPUS_RESET_STATE: + { + void *silk_enc; + silk_EncControlStruct dummy; + char *start; + silk_enc = (char*)st+st->silk_enc_offset; +#ifndef DISABLE_FLOAT_API + tonality_analysis_reset(&st->analysis); +#endif + + start = (char*)&st->OPUS_ENCODER_RESET_START; + OPUS_CLEAR(start, st->silk_enc_offset - (start - (char*)st)); + + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + if (st->application != OPUS_APPLICATION_RESTRICTED_CELT) + silk_InitEncoder( silk_enc, st->channels, st->arch, &dummy ); +#ifdef ENABLE_DRED + /* Initialize DRED Encoder */ + dred_encoder_reset( &st->dred_encoder ); +#endif + st->stream_channels = st->channels; + st->hybrid_stereo_width_Q14 = 1 << 14; + st->prev_HB_gain = Q15ONE; + st->first = 1; + st->mode = MODE_HYBRID; + st->bandwidth = OPUS_BANDWIDTH_FULLBAND; + st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); + } + break; + case OPUS_SET_FORCE_MODE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if ((value < MODE_SILK_ONLY || value > MODE_CELT_ONLY) && value != OPUS_AUTO) + { + goto bad_arg; + } + st->user_forced_mode = value; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + ret = celt_encoder_ctl(celt_enc, OPUS_SET_LFE(value)); + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + celt_glog *value = va_arg(ap, celt_glog*); + st->energy_masking = value; + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + ret = celt_encoder_ctl(celt_enc, OPUS_SET_ENERGY_MASK(value)); + } + break; + case OPUS_GET_IN_DTX_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + if (st->silk_mode.useDTX && (st->prev_mode == MODE_SILK_ONLY || st->prev_mode == MODE_HYBRID)) { + /* DTX determined by Silk. */ + silk_encoder *silk_enc = (silk_encoder*)(void *)((char*)st+st->silk_enc_offset); + *value = silk_enc->state_Fxx[0].sCmn.noSpeechCounter >= NB_SPEECH_FRAMES_BEFORE_DTX; + /* Stereo: check second channel unless only the middle channel was encoded. */ + if(*value == 1 && st->silk_mode.nChannelsInternal == 2 && silk_enc->prev_decode_only_middle == 0) { + *value = silk_enc->state_Fxx[1].sCmn.noSpeechCounter >= NB_SPEECH_FRAMES_BEFORE_DTX; + } + } + else if (st->use_dtx) { + /* DTX determined by Opus. */ + *value = st->nb_no_activity_ms_Q1 >= NB_SPEECH_FRAMES_BEFORE_DTX*20*2; + } + else { + *value = 0; + } + } + break; +#ifdef USE_WEIGHTS_FILE + case OPUS_SET_DNN_BLOB_REQUEST: + { + const unsigned char *data = va_arg(ap, const unsigned char *); + opus_int32 len = va_arg(ap, opus_int32); + if(len<0 || data == NULL) + { + goto bad_arg; + } +#ifdef ENABLE_DRED + ret = dred_encoder_load_model(&st->dred_encoder, data, len); +#endif + } + break; +#endif + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (!value) + { + goto bad_arg; + } + celt_assert(celt_enc != NULL); + ret = celt_encoder_ctl(celt_enc, CELT_GET_MODE(value)); + } + break; + default: + /* fprintf(stderr, "unknown opus_encoder_ctl() request: %d", request);*/ + ret = OPUS_UNIMPLEMENTED; + break; + } + va_end(ap); + return ret; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + +void opus_encoder_destroy(OpusEncoder *st) +{ + opus_free(st); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_multistream_encoder.c b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_multistream_encoder.c new file mode 100644 index 000000000..79c0d3b7d --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_multistream_encoder.c @@ -0,0 +1,1363 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include +#include "float_cast.h" +#include "os_support.h" +#include "mathops.h" +#include "mdct.h" +#include "modes.h" +#include "bands.h" +#include "quant_bands.h" +#include "pitch.h" + +typedef struct { + int nb_streams; + int nb_coupled_streams; + unsigned char mapping[8]; +} VorbisLayout; + +/* Index is nb_channel-1*/ +static const VorbisLayout vorbis_mappings[8] = { + {1, 0, {0}}, /* 1: mono */ + {1, 1, {0, 1}}, /* 2: stereo */ + {2, 1, {0, 2, 1}}, /* 3: 1-d surround */ + {2, 2, {0, 1, 2, 3}}, /* 4: quadraphonic surround */ + {3, 2, {0, 4, 1, 2, 3}}, /* 5: 5-channel surround */ + {4, 2, {0, 4, 1, 2, 3, 5}}, /* 6: 5.1 surround */ + {4, 3, {0, 4, 1, 2, 3, 5, 6}}, /* 7: 6.1 surround */ + {5, 3, {0, 6, 1, 2, 3, 4, 5, 7}}, /* 8: 7.1 surround */ +}; + +#ifdef ENABLE_QEXT +#define MAX_OVERLAP 240 +#else +#define MAX_OVERLAP 120 +#endif + +static opus_val32 *ms_get_preemph_mem(OpusMSEncoder *st) +{ + int s; + char *ptr; + int coupled_size, mono_size; + + coupled_size = opus_encoder_init(NULL, st->Fs, 2, st->application); + mono_size = opus_encoder_init(NULL, st->Fs, 1, st->application); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;slayout.nb_streams;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + /* void* cast avoids clang -Wcast-align warning */ + return (opus_val32*)(void*)(ptr+st->layout.nb_channels*MAX_OVERLAP*sizeof(opus_val32)); +} + +static opus_val32 *ms_get_window_mem(OpusMSEncoder *st) +{ + int s; + char *ptr; + int coupled_size, mono_size; + + coupled_size = opus_encoder_init(NULL, st->Fs, 2, st->application); + mono_size = opus_encoder_init(NULL, st->Fs, 1, st->application); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;slayout.nb_streams;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + /* void* cast avoids clang -Wcast-align warning */ + return (opus_val32*)(void*)ptr; +} + +static int validate_ambisonics(int nb_channels, int *nb_streams, int *nb_coupled_streams) +{ + int order_plus_one; + int acn_channels; + int nondiegetic_channels; + + if (nb_channels < 1 || nb_channels > 227) + return 0; + + order_plus_one = isqrt32(nb_channels); + acn_channels = order_plus_one * order_plus_one; + nondiegetic_channels = nb_channels - acn_channels; + + if (nondiegetic_channels != 0 && nondiegetic_channels != 2) + return 0; + + if (nb_streams) + *nb_streams = acn_channels + (nondiegetic_channels != 0); + if (nb_coupled_streams) + *nb_coupled_streams = nondiegetic_channels != 0; + return 1; +} + +static int validate_encoder_layout(const ChannelLayout *layout) +{ + int s; + for (s=0;snb_streams;s++) + { + if (s < layout->nb_coupled_streams) + { + if (get_left_channel(layout, s, -1)==-1) + return 0; + if (get_right_channel(layout, s, -1)==-1) + return 0; + } else { + if (get_mono_channel(layout, s, -1)==-1) + return 0; + } + } + return 1; +} + +static void channel_pos(int channels, int pos[8]) +{ + /* Position in the mix: 0 don't mix, 1: left, 2: center, 3:right */ + if (channels==4) + { + pos[0]=1; + pos[1]=3; + pos[2]=1; + pos[3]=3; + } else if (channels==3||channels==5||channels==6) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=0; + } else if (channels==7) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=2; + pos[6]=0; + } else if (channels==8) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=1; + pos[6]=3; + pos[7]=0; + } +} + +#if 1 +/* Computes a rough approximation of log2(2^a + 2^b) */ +static opus_val16 logSum(celt_glog a, celt_glog b) +{ + celt_glog max; + celt_glog diff; + celt_glog frac; + static const celt_glog diff_table[17] = { + GCONST(0.5000000f), GCONST(0.2924813f), GCONST(0.1609640f), GCONST(0.0849625f), + GCONST(0.0437314f), GCONST(0.0221971f), GCONST(0.0111839f), GCONST(0.0056136f), + GCONST(0.0028123f) + }; + int low; + if (a>b) + { + max = a; + diff = SUB32(a,b); + } else { + max = b; + diff = SUB32(b,a); + } + if (!(diff < GCONST(8.f))) /* inverted to catch NaNs */ + return max; +#ifdef FIXED_POINT + low = SHR32(diff, DB_SHIFT-1); + frac = VSHR32(diff - SHL32(low, DB_SHIFT-1), DB_SHIFT-16); +#else + low = (int)floor(2*diff); + frac = 2*diff - low; +#endif + return max + diff_table[low] + MULT16_32_Q15(frac, SUB32(diff_table[low+1], diff_table[low])); +} +#else +opus_val16 logSum(opus_val16 a, opus_val16 b) +{ + return log2(pow(4, a)+ pow(4, b))/2; +} +#endif + +void surround_analysis(const CELTMode *celt_mode, const void *pcm, celt_glog *bandLogE, opus_val32 *mem, opus_val32 *preemph_mem, + int len, int overlap, int channels, int rate, opus_copy_channel_in_func copy_channel_in, int arch +) +{ + int c; + int i; + int LM; + int pos[8] = {0}; + int upsample; + int frame_size; + int freq_size; + opus_val16 channel_offset; + opus_val32 bandE[21]; + celt_glog maskLogE[3][21]; + VARDECL(opus_val32, in); + VARDECL(opus_res, x); + VARDECL(opus_val32, freq); + SAVE_STACK; + + upsample = resampling_factor(rate); + frame_size = len*upsample; + + /* LM = log2(frame_size / 120) */ + for (LM=0;LMmaxLM;LM++) + if (celt_mode->shortMdctSize<shortMdctSize<preemph, preemph_mem+c, 0); +#ifndef FIXED_POINT + { + opus_val32 sum; + sum = celt_inner_prod(in, in, frame_size+overlap, 0); + /* This should filter out both NaNs and ridiculous signals that could + cause NaNs further down. */ + if (!(sum < 1e18f) || celt_isnan(sum)) + { + OPUS_CLEAR(in, frame_size+overlap); + preemph_mem[c] = 0; + } + } +#endif + OPUS_CLEAR(bandE, 21); + for (frame=0;framemdct, in+freq_size*frame, freq, celt_mode->window, + overlap, celt_mode->maxLM-LM, 1, arch); + if (upsample != 1) + { + int bound = freq_size/upsample; + for (i=0;i=0;i--) + bandLogE[21*c+i] = MAXG(bandLogE[21*c+i], bandLogE[21*c+i+1]-GCONST(2.f)); + if (pos[c]==1) + { + for (i=0;i<21;i++) + maskLogE[0][i] = logSum(maskLogE[0][i], bandLogE[21*c+i]); + } else if (pos[c]==3) + { + for (i=0;i<21;i++) + maskLogE[2][i] = logSum(maskLogE[2][i], bandLogE[21*c+i]); + } else if (pos[c]==2) + { + for (i=0;i<21;i++) + { + maskLogE[0][i] = logSum(maskLogE[0][i], bandLogE[21*c+i]-GCONST(.5f)); + maskLogE[2][i] = logSum(maskLogE[2][i], bandLogE[21*c+i]-GCONST(.5f)); + } + } +#if 0 + for (i=0;i<21;i++) + printf("%f ", bandLogE[21*c+i]); + float sum=0; + for (i=0;i<21;i++) + sum += bandLogE[21*c+i]; + printf("%f ", sum/21); +#endif + OPUS_COPY(mem+c*overlap, in+frame_size, overlap); + } + for (i=0;i<21;i++) + maskLogE[1][i] = MIN32(maskLogE[0][i],maskLogE[2][i]); + channel_offset = HALF16(celt_log2(QCONST32(2.f,14)/(channels-1))); + for (c=0;c<3;c++) + for (i=0;i<21;i++) + maskLogE[c][i] += channel_offset; +#if 0 + for (c=0;c<3;c++) + { + for (i=0;i<21;i++) + printf("%f ", maskLogE[c][i]); + } +#endif + for (c=0;cnb_streams||nb_coupled_streams<0)return 0; + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + return align(sizeof(OpusMSEncoder)) + + nb_coupled_streams * align(coupled_size) + + (nb_streams-nb_coupled_streams) * align(mono_size); +} + +opus_int32 opus_multistream_surround_encoder_get_size(int channels, int mapping_family) +{ + int nb_streams; + int nb_coupled_streams; + opus_int32 size; + + if (mapping_family==0) + { + if (channels==1) + { + nb_streams=1; + nb_coupled_streams=0; + } else if (channels==2) + { + nb_streams=1; + nb_coupled_streams=1; + } else + return 0; + } else if (mapping_family==1 && channels<=8 && channels>=1) + { + nb_streams=vorbis_mappings[channels-1].nb_streams; + nb_coupled_streams=vorbis_mappings[channels-1].nb_coupled_streams; + } else if (mapping_family==255) + { + nb_streams=channels; + nb_coupled_streams=0; + } else if (mapping_family==2) + { + if (!validate_ambisonics(channels, &nb_streams, &nb_coupled_streams)) + return 0; + } else + return 0; + size = opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); + if (channels>2) + { + size += channels*(MAX_OVERLAP*sizeof(opus_val32) + sizeof(opus_val32)); + } + return size; +} + +static int opus_multistream_encoder_init_impl( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + MappingType mapping_type +) +{ + int coupled_size; + int mono_size; + int i, ret; + char *ptr; + + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams) || + (streams+coupled_streams>channels)) + return OPUS_BAD_ARG; + + coupled_size = opus_encoder_init(NULL, Fs, 2, application); + if (coupled_size < 0) + return coupled_size; + mono_size = opus_encoder_init(NULL, Fs, 1, application); + if (mono_size < 0) + return mono_size; + if (st == NULL) + { + int surround_size=0; + if (mapping_type == MAPPING_TYPE_SURROUND) { + surround_size = channels*(MAX_OVERLAP*sizeof(opus_val32) + sizeof(opus_val32)); + } + + return align(sizeof(OpusMSEncoder)) + coupled_streams*align(coupled_size) + + (streams-coupled_streams)*align(mono_size) + surround_size; + } + + st->arch = opus_select_arch(); + st->layout.nb_channels = channels; + st->layout.nb_streams = streams; + st->layout.nb_coupled_streams = coupled_streams; + if (mapping_type != MAPPING_TYPE_SURROUND) + st->lfe_stream = -1; + st->bitrate_bps = OPUS_AUTO; + st->application = application; + st->Fs = Fs; + st->variable_duration = OPUS_FRAMESIZE_ARG; + for (i=0;ilayout.nb_channels;i++) + st->layout.mapping[i] = mapping[i]; + if (!validate_layout(&st->layout)) + return OPUS_BAD_ARG; + if (!validate_encoder_layout(&st->layout)) + return OPUS_BAD_ARG; + if (mapping_type == MAPPING_TYPE_AMBISONICS && + !validate_ambisonics(st->layout.nb_channels, NULL, NULL)) + return OPUS_BAD_ARG; + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (i=0;ilayout.nb_coupled_streams;i++) + { + ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 2, application); + if(ret!=OPUS_OK)return ret; + if (i==st->lfe_stream) + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_SET_LFE(1)); + ptr += align(coupled_size); + } + for (;ilayout.nb_streams;i++) + { + ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 1, application); + if(ret!=OPUS_OK)return ret; + if (i==st->lfe_stream) + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_SET_LFE(1)); + ptr += align(mono_size); + } + if (mapping_type == MAPPING_TYPE_SURROUND) + { + OPUS_CLEAR(ms_get_preemph_mem(st), channels); + OPUS_CLEAR(ms_get_window_mem(st), channels*MAX_OVERLAP); + } + st->mapping_type = mapping_type; + return OPUS_OK; +} + +int opus_multistream_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application +) +{ + return opus_multistream_encoder_init_impl(st, Fs, channels, streams, + coupled_streams, mapping, + application, MAPPING_TYPE_NONE); +} + +int opus_multistream_surround_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application +) +{ + MappingType mapping_type; + int lfe_stream; + + if ((channels>255) || (channels<1)) + return OPUS_BAD_ARG; + lfe_stream = -1; + if (mapping_family==0) + { + if (channels==1) + { + *streams=1; + *coupled_streams=0; + mapping[0]=0; + } else if (channels==2) + { + *streams=1; + *coupled_streams=1; + mapping[0]=0; + mapping[1]=1; + } else + return OPUS_UNIMPLEMENTED; + } else if (mapping_family==1 && channels<=8 && channels>=1) + { + int i; + *streams=vorbis_mappings[channels-1].nb_streams; + *coupled_streams=vorbis_mappings[channels-1].nb_coupled_streams; + for (i=0;i=6) + lfe_stream = *streams-1; + } else if (mapping_family==255) + { + int i; + *streams=channels; + *coupled_streams=0; + for(i=0;i2 && mapping_family==1) { + mapping_type = MAPPING_TYPE_SURROUND; + } else if (mapping_family==2) + { + mapping_type = MAPPING_TYPE_AMBISONICS; + } else + { + mapping_type = MAPPING_TYPE_NONE; + } + if (st != NULL) { + st->lfe_stream = lfe_stream; + } + return opus_multistream_encoder_init_impl(st, Fs, channels, *streams, + *coupled_streams, mapping, + application, mapping_type); +} + +OpusMSEncoder *opus_multistream_encoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + int *error +) +{ + int ret; + OpusMSEncoder *st; + opus_int32 size; + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams) || + (streams+coupled_streams>channels)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + size = opus_multistream_encoder_init(NULL, Fs, channels, streams, coupled_streams, mapping, application); + if (size < 0) + { + if (error) + *error = size; + return NULL; + } + st = (OpusMSEncoder *)opus_alloc(size); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_encoder_init(st, Fs, channels, streams, coupled_streams, mapping, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +OpusMSEncoder *opus_multistream_surround_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application, + int *error +) +{ + int ret; + opus_int32 size; + OpusMSEncoder *st; + if ((channels>255) || (channels<1)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + size = opus_multistream_surround_encoder_init(NULL, Fs, channels, mapping_family, streams, coupled_streams, mapping, application); + if (size < 0) + { + if (error) + *error = size; + return NULL; + } + st = (OpusMSEncoder *)opus_alloc(size); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_surround_encoder_init(st, Fs, channels, mapping_family, streams, coupled_streams, mapping, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +static void surround_rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size, + opus_int32 Fs + ) +{ + int i; + opus_int32 channel_rate; + int stream_offset; + int lfe_offset; + int coupled_ratio; /* Q8 */ + int lfe_ratio; /* Q8 */ + int nb_lfe; + int nb_uncoupled; + int nb_coupled; + int nb_normal; + opus_int32 channel_offset; + opus_int32 bitrate; + int total; + + nb_lfe = (st->lfe_stream!=-1); + nb_coupled = st->layout.nb_coupled_streams; + nb_uncoupled = st->layout.nb_streams-nb_coupled-nb_lfe; + nb_normal = 2*nb_coupled + nb_uncoupled; + + /* Give each non-LFE channel enough bits per channel for coding band energy. */ + channel_offset = 40*IMAX(50, Fs/frame_size); + + if (st->bitrate_bps==OPUS_AUTO) + { + bitrate = nb_normal*(channel_offset + Fs + 10000) + 8000*nb_lfe; + } else if (st->bitrate_bps==OPUS_BITRATE_MAX) + { + bitrate = nb_normal*750000 + nb_lfe*128000; + } else { + bitrate = st->bitrate_bps; + } + + /* Give LFE some basic stream_channel allocation but never exceed 1/20 of the + total rate for the non-energy part to avoid problems at really low rate. */ + lfe_offset = IMIN(bitrate/20, 3000) + 15*IMAX(50, Fs/frame_size); + + /* We give each stream (coupled or uncoupled) a starting bitrate. + This models the main saving of coupled channels over uncoupled. */ + stream_offset = (bitrate - channel_offset*nb_normal - lfe_offset*nb_lfe)/nb_normal/2; + stream_offset = IMAX(0, IMIN(20000, stream_offset)); + + /* Coupled streams get twice the mono rate after the offset is allocated. */ + coupled_ratio = 512; + /* Should depend on the bitrate, for now we assume LFE gets 1/8 the bits of mono */ + lfe_ratio = 32; + + total = (nb_uncoupled<<8) /* mono */ + + coupled_ratio*nb_coupled /* stereo */ + + nb_lfe*lfe_ratio; + channel_rate = 256*(opus_int64)(bitrate - lfe_offset*nb_lfe - stream_offset*(nb_coupled+nb_uncoupled) - channel_offset*nb_normal)/total; + + for (i=0;ilayout.nb_streams;i++) + { + if (ilayout.nb_coupled_streams) + rate[i] = 2*channel_offset + IMAX(0, stream_offset+(channel_rate*coupled_ratio>>8)); + else if (i!=st->lfe_stream) + rate[i] = channel_offset + IMAX(0, stream_offset + channel_rate); + else + rate[i] = IMAX(0, lfe_offset+(channel_rate*lfe_ratio>>8)); + } +} + +static void ambisonics_rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size, + opus_int32 Fs + ) +{ + int i; + opus_int32 total_rate; + opus_int32 per_stream_rate; + + const int nb_channels = st->layout.nb_streams + st->layout.nb_coupled_streams; + + if (st->bitrate_bps==OPUS_AUTO) + { + total_rate = (st->layout.nb_coupled_streams + st->layout.nb_streams) * + (Fs+60*Fs/frame_size) + st->layout.nb_streams * (opus_int32)15000; + } else if (st->bitrate_bps==OPUS_BITRATE_MAX) + { + total_rate = nb_channels * 750000; + } else + { + total_rate = st->bitrate_bps; + } + + /* Allocate equal number of bits to Ambisonic (uncoupled) and non-diegetic + * (coupled) streams */ + per_stream_rate = total_rate / st->layout.nb_streams; + for (i = 0; i < st->layout.nb_streams; i++) + { + rate[i] = per_stream_rate; + } +} + +static opus_int32 rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size + ) +{ + int i; + opus_int32 rate_sum=0; + opus_int32 Fs; + char *ptr; + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); + + if (st->mapping_type == MAPPING_TYPE_AMBISONICS) { + ambisonics_rate_allocation(st, rate, frame_size, Fs); + } else + { + surround_rate_allocation(st, rate, frame_size, Fs); + } + + for (i=0;ilayout.nb_streams;i++) + { + rate[i] = IMAX(rate[i], 500); + rate_sum += rate[i]; + } + return rate_sum; +} + +/* Max size in case the encoder decides to return six frames (6 x 20 ms = 120 ms) */ +#ifdef ENABLE_QEXT +#define MS_FRAME_TMP (6*QEXT_PACKET_SIZE_CAP+12) +#else +#define MS_FRAME_TMP (6*1275+12) +#endif + +int opus_multistream_encode_native +( + OpusMSEncoder *st, + opus_copy_channel_in_func copy_channel_in, + const void *pcm, + int analysis_frame_size, + unsigned char *data, + opus_int32 max_data_bytes, + int lsb_depth, + downmix_func downmix, + int float_api, + void *user_data +) +{ + opus_int32 Fs; + int coupled_size; + int mono_size; + int s; + char *ptr; + int tot_size; + VARDECL(opus_res, buf); + VARDECL(celt_glog, bandSMR); + unsigned char tmp_data[MS_FRAME_TMP]; + OpusRepacketizer rp; + opus_int32 vbr; + const CELTMode *celt_mode=NULL; + opus_int32 bitrates[256]; + celt_glog bandLogE[42]; + opus_val32 *mem = NULL; + opus_val32 *preemph_mem=NULL; + int frame_size; + opus_int32 rate_sum; + opus_int32 smallest_packet; + ALLOC_STACK; + + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + preemph_mem = ms_get_preemph_mem(st); + mem = ms_get_window_mem(st); + } + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_VBR(&vbr)); + if (st->application != OPUS_APPLICATION_RESTRICTED_SILK) + opus_encoder_ctl((OpusEncoder*)ptr, CELT_GET_MODE(&celt_mode)); + + frame_size = frame_size_select(st->application, analysis_frame_size, st->variable_duration, Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + /* Smallest packet the encoder can produce. */ + smallest_packet = st->layout.nb_streams*2-1; + /* 100 ms needs an extra byte per stream for the ToC. */ + if (Fs/frame_size == 10) + smallest_packet += st->layout.nb_streams; + if (max_data_bytes < smallest_packet) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + ALLOC(buf, 2*frame_size, opus_res); + coupled_size = opus_encoder_init(NULL, st->Fs, 2, st->application); + mono_size = opus_encoder_init(NULL, st->Fs, 1, st->application); + + ALLOC(bandSMR, 21*st->layout.nb_channels, celt_glog); + if (st->mapping_type == MAPPING_TYPE_SURROUND && st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + surround_analysis(celt_mode, pcm, bandSMR, mem, preemph_mem, frame_size, celt_mode->overlap, st->layout.nb_channels, Fs, copy_channel_in, st->arch); + } + + /* Compute bitrate allocation between streams (this could be a lot better) */ + rate_sum = rate_allocation(st, bitrates, frame_size); + + if (!vbr) + { + if (st->bitrate_bps == OPUS_AUTO) + { + max_data_bytes = IMIN(max_data_bytes, (bitrate_to_bits(rate_sum, Fs, frame_size)+4)/8); + } else if (st->bitrate_bps != OPUS_BITRATE_MAX) + { + max_data_bytes = IMIN(max_data_bytes, IMAX(smallest_packet, + (bitrate_to_bits(st->bitrate_bps, Fs, frame_size)+4)/8)); + } + } + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;slayout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrates[s])); + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + opus_int32 equiv_rate; + equiv_rate = st->bitrate_bps; + if (frame_size*50 < Fs) + equiv_rate -= 60*(Fs/frame_size - 50)*st->layout.nb_channels; + if (equiv_rate > 10000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_FULLBAND)); + else if (equiv_rate > 7000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_SUPERWIDEBAND)); + else if (equiv_rate > 5000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_WIDEBAND)); + else + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_NARROWBAND)); + if (s < st->layout.nb_coupled_streams) + { + /* To preserve the spatial image, force stereo CELT on coupled streams */ + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); + opus_encoder_ctl(enc, OPUS_SET_FORCE_CHANNELS(2)); + } + } + else if (st->mapping_type == MAPPING_TYPE_AMBISONICS) { + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); + } + } + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + /* Counting ToC */ + tot_size = 0; + for (s=0;slayout.nb_streams;s++) + { + OpusEncoder *enc; + int len; + int curr_max; + int c1, c2; + int ret; + + opus_repacketizer_init(&rp); + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + { + int i; + int left, right; + left = get_left_channel(&st->layout, s, -1); + right = get_right_channel(&st->layout, s, -1); + (*copy_channel_in)(buf, 2, + pcm, st->layout.nb_channels, left, frame_size, user_data); + (*copy_channel_in)(buf+1, 2, + pcm, st->layout.nb_channels, right, frame_size, user_data); + ptr += align(coupled_size); + if (st->mapping_type == MAPPING_TYPE_SURROUND && st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + for (i=0;i<21;i++) + { + bandLogE[i] = bandSMR[21*left+i]; + bandLogE[21+i] = bandSMR[21*right+i]; + } + } + c1 = left; + c2 = right; + } else { + int i; + int chan = get_mono_channel(&st->layout, s, -1); + (*copy_channel_in)(buf, 1, + pcm, st->layout.nb_channels, chan, frame_size, user_data); + ptr += align(mono_size); + if (st->mapping_type == MAPPING_TYPE_SURROUND && st->application != OPUS_APPLICATION_RESTRICTED_SILK) + { + for (i=0;i<21;i++) + bandLogE[i] = bandSMR[21*chan+i]; + } + c1 = chan; + c2 = -1; + } + if (st->mapping_type == MAPPING_TYPE_SURROUND && st->application != OPUS_APPLICATION_RESTRICTED_SILK) + opus_encoder_ctl(enc, OPUS_SET_ENERGY_MASK(bandLogE)); + /* number of bytes left (+Toc) */ + curr_max = max_data_bytes - tot_size; + /* Reserve one byte for the last stream and two for the others */ + curr_max -= IMAX(0,2*(st->layout.nb_streams-s-1)-1); + /* For 100 ms, reserve an extra byte per stream for the ToC */ + if (Fs/frame_size == 10) + curr_max -= st->layout.nb_streams-s-1; + curr_max = IMIN(curr_max,MS_FRAME_TMP); + /* Repacketizer will add one or two bytes for self-delimited frames */ + if (s != st->layout.nb_streams-1) curr_max -= curr_max>253 ? 2 : 1; + if (!vbr && s == st->layout.nb_streams-1) + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bits_to_bitrate(curr_max*8, Fs, frame_size))); + len = opus_encode_native(enc, buf, frame_size, tmp_data, curr_max, lsb_depth, + pcm, analysis_frame_size, c1, c2, st->layout.nb_channels, downmix, float_api); + if (len<0) + { + RESTORE_STACK; + return len; + } + /* We need to use the repacketizer to add the self-delimiting lengths + while taking into account the fact that the encoder can now return + more than one frame at a time (e.g. 60 ms CELT-only) */ + ret = opus_repacketizer_cat(&rp, tmp_data, len); + /* If the opus_repacketizer_cat() fails, then something's seriously wrong + with the encoder. */ + if (ret != OPUS_OK) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + len = opus_repacketizer_out_range_impl(&rp, 0, opus_repacketizer_get_nb_frames(&rp), + data, max_data_bytes-tot_size, s != st->layout.nb_streams-1, !vbr && s == st->layout.nb_streams-1, NULL, 0); + data += len; + tot_size += len; + } + /*printf("\n");*/ + RESTORE_STACK; + return tot_size; +} + +#if !defined(DISABLE_FLOAT_API) +static void opus_copy_channel_in_float( + opus_res *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + const float *float_src; + opus_int32 i; + (void)user_data; + float_src = (const float *)src; + for (i=0;iFs, 2, st->application); + mono_size = opus_encoder_init(NULL, st->Fs, 1, st->application); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + switch (request) + { + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_AUTO && value != OPUS_BITRATE_MAX) + { + if (value <= 0) + goto bad_arg; + value = IMIN(750000*st->layout.nb_channels, IMAX(500*st->layout.nb_channels, value)); + } + st->bitrate_bps = value; + } + break; + case OPUS_GET_BITRATE_REQUEST: + { + int s; + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = 0; + for (s=0;slayout.nb_streams;s++) + { + opus_int32 rate; + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + opus_encoder_ctl(enc, request, &rate); + *value += rate; + } + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + case OPUS_GET_VBR_REQUEST: + case OPUS_GET_APPLICATION_REQUEST: + case OPUS_GET_BANDWIDTH_REQUEST: + case OPUS_GET_COMPLEXITY_REQUEST: + case OPUS_GET_PACKET_LOSS_PERC_REQUEST: + case OPUS_GET_DTX_REQUEST: + case OPUS_GET_VOICE_RATIO_REQUEST: + case OPUS_GET_VBR_CONSTRAINT_REQUEST: + case OPUS_GET_SIGNAL_REQUEST: + case OPUS_GET_LOOKAHEAD_REQUEST: + case OPUS_GET_SAMPLE_RATE_REQUEST: + case OPUS_GET_INBAND_FEC_REQUEST: + case OPUS_GET_FORCE_CHANNELS_REQUEST: + case OPUS_GET_PREDICTION_DISABLED_REQUEST: + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + case OPUS_GET_QEXT_REQUEST: + { + OpusEncoder *enc; + /* For int32* GET params, just query the first stream */ + opus_int32 *value = va_arg(ap, opus_int32*); + enc = (OpusEncoder*)ptr; + ret = opus_encoder_ctl(enc, request, value); + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + int s; + opus_uint32 *value = va_arg(ap, opus_uint32*); + opus_uint32 tmp; + if (!value) + { + goto bad_arg; + } + *value=0; + for (s=0;slayout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, request, &tmp); + if (ret != OPUS_OK) break; + *value ^= tmp; + } + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + case OPUS_SET_COMPLEXITY_REQUEST: + case OPUS_SET_VBR_REQUEST: + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + case OPUS_SET_MAX_BANDWIDTH_REQUEST: + case OPUS_SET_BANDWIDTH_REQUEST: + case OPUS_SET_SIGNAL_REQUEST: + case OPUS_SET_APPLICATION_REQUEST: + case OPUS_SET_INBAND_FEC_REQUEST: + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + case OPUS_SET_DTX_REQUEST: + case OPUS_SET_FORCE_MODE_REQUEST: + case OPUS_SET_FORCE_CHANNELS_REQUEST: + case OPUS_SET_PREDICTION_DISABLED_REQUEST: + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + case OPUS_SET_QEXT_REQUEST: + { + int s; + /* This works for int32 params */ + opus_int32 value = va_arg(ap, opus_int32); + for (s=0;slayout.nb_streams;s++) + { + OpusEncoder *enc; + + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, request, value); + if (ret != OPUS_OK) + break; + } + } + break; + case OPUS_MULTISTREAM_GET_ENCODER_STATE_REQUEST: + { + int s; + opus_int32 stream_id; + OpusEncoder **value; + stream_id = va_arg(ap, opus_int32); + if (stream_id<0 || stream_id >= st->layout.nb_streams) + goto bad_arg; + value = va_arg(ap, OpusEncoder**); + if (!value) + { + goto bad_arg; + } + for (s=0;slayout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + *value = (OpusEncoder*)ptr; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->variable_duration = value; + } + break; + case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->variable_duration; + } + break; + case OPUS_RESET_STATE: + { + int s; + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + OPUS_CLEAR(ms_get_preemph_mem(st), st->layout.nb_channels); + OPUS_CLEAR(ms_get_window_mem(st), st->layout.nb_channels*MAX_OVERLAP); + } + for (s=0;slayout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, OPUS_RESET_STATE); + if (ret != OPUS_OK) + break; + } + } + break; + default: + ret = OPUS_UNIMPLEMENTED; + break; + } + return ret; +bad_arg: + return OPUS_BAD_ARG; +} + +int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) +{ + int ret; + va_list ap; + va_start(ap, request); + ret = opus_multistream_encoder_ctl_va_list(st, request, ap); + va_end(ap); + return ret; +} + +void opus_multistream_encoder_destroy(OpusMSEncoder *st) +{ + opus_free(st); +} diff --git a/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_projection_encoder.c b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_projection_encoder.c new file mode 100644 index 000000000..9b03900ae --- /dev/null +++ b/code/sdk/zhihui/test_opus_frame_to_pcm/src_lib/opus_projection_encoder.c @@ -0,0 +1,521 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "os_support.h" +#include "opus_private.h" +#include "opus_defines.h" +#include "opus_projection.h" +#include "opus_multistream.h" +#include "stack_alloc.h" +#include "mapping_matrix.h" + +struct OpusProjectionEncoder +{ + opus_int32 mixing_matrix_size_in_bytes; + opus_int32 demixing_matrix_size_in_bytes; + /* Encoder states go here */ +}; + +#if !defined(DISABLE_FLOAT_API) +static void opus_projection_copy_channel_in_float( + opus_res *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_float((const MappingMatrix*)user_data, + (const float*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} +#endif + +static void opus_projection_copy_channel_in_short( + opus_res *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_short((const MappingMatrix*)user_data, + (const opus_int16*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} + +static void opus_projection_copy_channel_in_int24( + opus_res *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_int24((const MappingMatrix*)user_data, + (const opus_int32*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} + +static int get_order_plus_one_from_channels(int channels, int *order_plus_one) +{ + int order_plus_one_; + int acn_channels; + int nondiegetic_channels; + + /* Allowed numbers of channels: + * (1 + n)^2 + 2j, for n = 0...14 and j = 0 or 1. + */ + if (channels < 1 || channels > 227) + return OPUS_BAD_ARG; + + order_plus_one_ = isqrt32(channels); + acn_channels = order_plus_one_ * order_plus_one_; + nondiegetic_channels = channels - acn_channels; + if (nondiegetic_channels != 0 && nondiegetic_channels != 2) + return OPUS_BAD_ARG; + + if (order_plus_one) + *order_plus_one = order_plus_one_; + return OPUS_OK; +} + +static int get_streams_from_channels(int channels, int mapping_family, + int *streams, int *coupled_streams, + int *order_plus_one) +{ + if (mapping_family == 3) + { + if (get_order_plus_one_from_channels(channels, order_plus_one) != OPUS_OK) + return OPUS_BAD_ARG; + if (streams) + *streams = (channels + 1) / 2; + if (coupled_streams) + *coupled_streams = channels / 2; + return OPUS_OK; + } + return OPUS_BAD_ARG; +} + +static MappingMatrix *get_mixing_matrix(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder))); +} + +static MappingMatrix *get_enc_demixing_matrix(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder) + + st->mixing_matrix_size_in_bytes)); +} + +static OpusMSEncoder *get_multistream_encoder(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (OpusMSEncoder *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder) + + st->mixing_matrix_size_in_bytes + + st->demixing_matrix_size_in_bytes)); +} + +opus_int32 opus_projection_ambisonics_encoder_get_size(int channels, + int mapping_family) +{ + int nb_streams; + int nb_coupled_streams; + int order_plus_one; + int mixing_matrix_rows, mixing_matrix_cols; + int demixing_matrix_rows, demixing_matrix_cols; + opus_int32 mixing_matrix_size, demixing_matrix_size; + opus_int32 encoder_size; + int ret; + + ret = get_streams_from_channels(channels, mapping_family, &nb_streams, + &nb_coupled_streams, &order_plus_one); + if (ret != OPUS_OK) + return 0; + + if (order_plus_one == 2) + { + mixing_matrix_rows = mapping_matrix_foa_mixing.rows; + mixing_matrix_cols = mapping_matrix_foa_mixing.cols; + demixing_matrix_rows = mapping_matrix_foa_demixing.rows; + demixing_matrix_cols = mapping_matrix_foa_demixing.cols; + } + else if (order_plus_one == 3) + { + mixing_matrix_rows = mapping_matrix_soa_mixing.rows; + mixing_matrix_cols = mapping_matrix_soa_mixing.cols; + demixing_matrix_rows = mapping_matrix_soa_demixing.rows; + demixing_matrix_cols = mapping_matrix_soa_demixing.cols; + } + else if (order_plus_one == 4) + { + mixing_matrix_rows = mapping_matrix_toa_mixing.rows; + mixing_matrix_cols = mapping_matrix_toa_mixing.cols; + demixing_matrix_rows = mapping_matrix_toa_demixing.rows; + demixing_matrix_cols = mapping_matrix_toa_demixing.cols; + } + else if (order_plus_one == 5) + { + mixing_matrix_rows = mapping_matrix_fourthoa_mixing.rows; + mixing_matrix_cols = mapping_matrix_fourthoa_mixing.cols; + demixing_matrix_rows = mapping_matrix_fourthoa_demixing.rows; + demixing_matrix_cols = mapping_matrix_fourthoa_demixing.cols; + } + else if (order_plus_one == 6) + { + mixing_matrix_rows = mapping_matrix_fifthoa_mixing.rows; + mixing_matrix_cols = mapping_matrix_fifthoa_mixing.cols; + demixing_matrix_rows = mapping_matrix_fifthoa_demixing.rows; + demixing_matrix_cols = mapping_matrix_fifthoa_demixing.cols; + } + else + return 0; + + mixing_matrix_size = + mapping_matrix_get_size(mixing_matrix_rows, mixing_matrix_cols); + if (!mixing_matrix_size) + return 0; + + demixing_matrix_size = + mapping_matrix_get_size(demixing_matrix_rows, demixing_matrix_cols); + if (!demixing_matrix_size) + return 0; + + encoder_size = + opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); + if (!encoder_size) + return 0; + + return align(sizeof(OpusProjectionEncoder)) + + mixing_matrix_size + demixing_matrix_size + encoder_size; +} + +int opus_projection_ambisonics_encoder_init(OpusProjectionEncoder *st, opus_int32 Fs, + int channels, int mapping_family, + int *streams, int *coupled_streams, + int application) +{ + MappingMatrix *mixing_matrix; + MappingMatrix *demixing_matrix; + OpusMSEncoder *ms_encoder; + int i; + int ret; + int order_plus_one; + unsigned char mapping[255]; + + if (streams == NULL || coupled_streams == NULL) { + return OPUS_BAD_ARG; + } + + if (get_streams_from_channels(channels, mapping_family, streams, + coupled_streams, &order_plus_one) != OPUS_OK) + return OPUS_BAD_ARG; + + if (mapping_family == 3) + { + /* Assign mixing matrix based on available pre-computed matrices. */ + mixing_matrix = get_mixing_matrix(st); + if (order_plus_one == 2) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_foa_mixing.rows, + mapping_matrix_foa_mixing.cols, mapping_matrix_foa_mixing.gain, + mapping_matrix_foa_mixing_data, + sizeof(mapping_matrix_foa_mixing_data)); + } + else if (order_plus_one == 3) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_soa_mixing.rows, + mapping_matrix_soa_mixing.cols, mapping_matrix_soa_mixing.gain, + mapping_matrix_soa_mixing_data, + sizeof(mapping_matrix_soa_mixing_data)); + } + else if (order_plus_one == 4) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_toa_mixing.rows, + mapping_matrix_toa_mixing.cols, mapping_matrix_toa_mixing.gain, + mapping_matrix_toa_mixing_data, + sizeof(mapping_matrix_toa_mixing_data)); + } + else if (order_plus_one == 5) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_fourthoa_mixing.rows, + mapping_matrix_fourthoa_mixing.cols, mapping_matrix_fourthoa_mixing.gain, + mapping_matrix_fourthoa_mixing_data, + sizeof(mapping_matrix_fourthoa_mixing_data)); + } + else if (order_plus_one == 6) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_fifthoa_mixing.rows, + mapping_matrix_fifthoa_mixing.cols, mapping_matrix_fifthoa_mixing.gain, + mapping_matrix_fifthoa_mixing_data, + sizeof(mapping_matrix_fifthoa_mixing_data)); + } + else + return OPUS_BAD_ARG; + + st->mixing_matrix_size_in_bytes = mapping_matrix_get_size( + mixing_matrix->rows, mixing_matrix->cols); + if (!st->mixing_matrix_size_in_bytes) + return OPUS_BAD_ARG; + + /* Assign demixing matrix based on available pre-computed matrices. */ + demixing_matrix = get_enc_demixing_matrix(st); + if (order_plus_one == 2) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_foa_demixing.rows, + mapping_matrix_foa_demixing.cols, mapping_matrix_foa_demixing.gain, + mapping_matrix_foa_demixing_data, + sizeof(mapping_matrix_foa_demixing_data)); + } + else if (order_plus_one == 3) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_soa_demixing.rows, + mapping_matrix_soa_demixing.cols, mapping_matrix_soa_demixing.gain, + mapping_matrix_soa_demixing_data, + sizeof(mapping_matrix_soa_demixing_data)); + } + else if (order_plus_one == 4) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_toa_demixing.rows, + mapping_matrix_toa_demixing.cols, mapping_matrix_toa_demixing.gain, + mapping_matrix_toa_demixing_data, + sizeof(mapping_matrix_toa_demixing_data)); + } + else if (order_plus_one == 5) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_fourthoa_demixing.rows, + mapping_matrix_fourthoa_demixing.cols, mapping_matrix_fourthoa_demixing.gain, + mapping_matrix_fourthoa_demixing_data, + sizeof(mapping_matrix_fourthoa_demixing_data)); + } + else if (order_plus_one == 6) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_fifthoa_demixing.rows, + mapping_matrix_fifthoa_demixing.cols, mapping_matrix_fifthoa_demixing.gain, + mapping_matrix_fifthoa_demixing_data, + sizeof(mapping_matrix_fifthoa_demixing_data)); + } + else + return OPUS_BAD_ARG; + + st->demixing_matrix_size_in_bytes = mapping_matrix_get_size( + demixing_matrix->rows, demixing_matrix->cols); + if (!st->demixing_matrix_size_in_bytes) + return OPUS_BAD_ARG; + } + else + return OPUS_UNIMPLEMENTED; + + /* Ensure matrices are large enough for desired coding scheme. */ + if (*streams + *coupled_streams > mixing_matrix->rows || + channels > mixing_matrix->cols || + channels > demixing_matrix->rows || + *streams + *coupled_streams > demixing_matrix->cols) + return OPUS_BAD_ARG; + + /* Set trivial mapping so each input channel pairs with a matrix column. */ + for (i = 0; i < channels; i++) + mapping[i] = i; + + /* Initialize multistream encoder with provided settings. */ + ms_encoder = get_multistream_encoder(st); + ret = opus_multistream_encoder_init(ms_encoder, Fs, channels, *streams, + *coupled_streams, mapping, application); + return ret; +} + +OpusProjectionEncoder *opus_projection_ambisonics_encoder_create( + opus_int32 Fs, int channels, int mapping_family, int *streams, + int *coupled_streams, int application, int *error) +{ + int size; + int ret; + OpusProjectionEncoder *st; + + /* Allocate space for the projection encoder. */ + size = opus_projection_ambisonics_encoder_get_size(channels, mapping_family); + if (!size) { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + st = (OpusProjectionEncoder *)opus_alloc(size); + if (!st) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + + /* Initialize projection encoder with provided settings. */ + ret = opus_projection_ambisonics_encoder_init(st, Fs, channels, + mapping_family, streams, coupled_streams, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +int opus_projection_encode(OpusProjectionEncoder *st, const opus_int16 *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_short, pcm, frame_size, data, + max_data_bytes, 16, downmix_int, 0, get_mixing_matrix(st)); +} + +int opus_projection_encode24(OpusProjectionEncoder *st, const opus_int32 *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_int24, pcm, frame_size, data, + max_data_bytes, MAX_ENCODING_DEPTH, downmix_int, 0, get_mixing_matrix(st)); +} + +#ifndef DISABLE_FLOAT_API +int opus_projection_encode_float(OpusProjectionEncoder *st, const float *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_float, pcm, frame_size, data, + max_data_bytes, MAX_ENCODING_DEPTH, downmix_float, 1, get_mixing_matrix(st)); +} +#endif + +void opus_projection_encoder_destroy(OpusProjectionEncoder *st) +{ + opus_free(st); +} + +int opus_projection_encoder_ctl(OpusProjectionEncoder *st, int request, ...) +{ + va_list ap; + MappingMatrix *demixing_matrix; + OpusMSEncoder *ms_encoder; + int ret = OPUS_OK; + + ms_encoder = get_multistream_encoder(st); + demixing_matrix = get_enc_demixing_matrix(st); + + va_start(ap, request); + switch(request) + { + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = + ms_encoder->layout.nb_channels * (ms_encoder->layout.nb_streams + + ms_encoder->layout.nb_coupled_streams) * sizeof(opus_int16); + } + break; + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = demixing_matrix->gain; + } + break; + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST: + { + int i, j, k, l; + int nb_input_streams; + int nb_output_streams; + unsigned char *external_char; + opus_int16 *internal_short; + opus_int32 external_size; + opus_int32 internal_size; + + /* (I/O is in relation to the decoder's perspective). */ + nb_input_streams = ms_encoder->layout.nb_streams + + ms_encoder->layout.nb_coupled_streams; + nb_output_streams = ms_encoder->layout.nb_channels; + + external_char = va_arg(ap, unsigned char *); + external_size = va_arg(ap, opus_int32); + if (!external_char) + { + goto bad_arg; + } + internal_short = mapping_matrix_get_data(demixing_matrix); + internal_size = nb_input_streams * nb_output_streams * sizeof(opus_int16); + if (external_size != internal_size) + { + goto bad_arg; + } + + /* Copy demixing matrix subset to output destination. */ + l = 0; + for (i = 0; i < nb_input_streams; i++) { + for (j = 0; j < nb_output_streams; j++) { + k = demixing_matrix->rows * i + j; + external_char[2*l] = (unsigned char)internal_short[k]; + external_char[2*l+1] = (unsigned char)(internal_short[k] >> 8); + l++; + } + } + } + break; + default: + { + ret = opus_multistream_encoder_ctl_va_list(ms_encoder, request, ap); + } + break; + } + va_end(ap); + return ret; + +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} -- 2.17.1