diff options
Diffstat (limited to 'vp9/simple_encode.cc')
-rw-r--r-- | vp9/simple_encode.cc | 323 |
1 files changed, 295 insertions, 28 deletions
diff --git a/vp9/simple_encode.cc b/vp9/simple_encode.cc index c417a2589..d4eb0c669 100644 --- a/vp9/simple_encode.cc +++ b/vp9/simple_encode.cc @@ -90,12 +90,20 @@ static int img_read(vpx_image_t *img, FILE *file) { return 1; } +// Assume every config in VP9EncoderConfig is less than 100 characters. +#define ENCODE_CONFIG_BUF_SIZE 100 +struct EncodeConfig { + char name[ENCODE_CONFIG_BUF_SIZE]; + char value[ENCODE_CONFIG_BUF_SIZE]; +}; + class SimpleEncode::EncodeImpl { public: VP9_COMP *cpi; vpx_img_fmt_t img_fmt; vpx_image_t tmp_img; std::vector<FIRSTPASS_STATS> first_pass_stats; + std::vector<EncodeConfig> encode_config_list; }; static VP9_COMP *init_encoder(const VP9EncoderConfig *oxcf, @@ -167,7 +175,8 @@ static RefFrameType mv_ref_frame_to_ref_frame_type( static void update_motion_vector_info( const MOTION_VECTOR_INFO *input_motion_vector_info, const int num_rows_4x4, - const int num_cols_4x4, MotionVectorInfo *output_motion_vector_info) { + const int num_cols_4x4, MotionVectorInfo *output_motion_vector_info, + int motion_vector_scale) { const int num_units_4x4 = num_rows_4x4 * num_cols_4x4; for (int i = 0; i < num_units_4x4; ++i) { const MV_REFERENCE_FRAME *in_ref_frame = @@ -185,16 +194,34 @@ static void update_motion_vector_info( mv_ref_frame_to_ref_frame_type(in_ref_frame[1]); output_motion_vector_info[i].mv_row[0] = (double)input_motion_vector_info[i].mv[0].as_mv.row / - kMotionVectorPrecision; + motion_vector_scale; output_motion_vector_info[i].mv_column[0] = (double)input_motion_vector_info[i].mv[0].as_mv.col / - kMotionVectorPrecision; + motion_vector_scale; output_motion_vector_info[i].mv_row[1] = (double)input_motion_vector_info[i].mv[1].as_mv.row / - kMotionVectorPrecision; + motion_vector_scale; output_motion_vector_info[i].mv_column[1] = (double)input_motion_vector_info[i].mv[1].as_mv.col / - kMotionVectorPrecision; + motion_vector_scale; + } +} + +static void update_tpl_stats_info(const TplDepStats *input_tpl_stats_info, + const int show_frame_count, + TplStatsInfo *output_tpl_stats_info) { + int frame_idx; + for (frame_idx = 0; frame_idx < show_frame_count; ++frame_idx) { + output_tpl_stats_info[frame_idx].intra_cost = + input_tpl_stats_info[frame_idx].intra_cost; + output_tpl_stats_info[frame_idx].inter_cost = + input_tpl_stats_info[frame_idx].inter_cost; + output_tpl_stats_info[frame_idx].mc_flow = + input_tpl_stats_info[frame_idx].mc_flow; + output_tpl_stats_info[frame_idx].mc_dep_cost = + input_tpl_stats_info[frame_idx].mc_dep_cost; + output_tpl_stats_info[frame_idx].mc_ref_cost = + input_tpl_stats_info[frame_idx].mc_ref_cost; } } @@ -471,12 +498,13 @@ static bool init_encode_frame_result(EncodeFrameResult *encode_frame_result, encode_frame_result->coding_data.reset( new (std::nothrow) uint8_t[max_coding_data_byte_size]); - encode_frame_result->num_rows_4x4 = get_num_unit_4x4(frame_width); - encode_frame_result->num_cols_4x4 = get_num_unit_4x4(frame_height); + encode_frame_result->num_rows_4x4 = get_num_unit_4x4(frame_height); + encode_frame_result->num_cols_4x4 = get_num_unit_4x4(frame_width); encode_frame_result->partition_info.resize(encode_frame_result->num_rows_4x4 * encode_frame_result->num_cols_4x4); encode_frame_result->motion_vector_info.resize( encode_frame_result->num_rows_4x4 * encode_frame_result->num_cols_4x4); + encode_frame_result->tpl_stats_info.resize(MAX_LAG_BUFFERS); if (encode_frame_result->coding_data.get() == nullptr) { return false; @@ -485,8 +513,20 @@ static bool init_encode_frame_result(EncodeFrameResult *encode_frame_result, frame_height, img_fmt); } +static void encode_frame_result_update_rq_history( + const RATE_QINDEX_HISTORY *rq_history, + EncodeFrameResult *encode_frame_result) { + encode_frame_result->recode_count = rq_history->recode_count; + for (int i = 0; i < encode_frame_result->recode_count; ++i) { + const int q_index = rq_history->q_index_history[i]; + const int rate = rq_history->rate_history[i]; + encode_frame_result->q_index_history.push_back(q_index); + encode_frame_result->rate_history.push_back(rate); + } +} + static void update_encode_frame_result( - EncodeFrameResult *encode_frame_result, + EncodeFrameResult *encode_frame_result, const int show_frame_count, const ENCODE_FRAME_RESULT *encode_frame_info) { encode_frame_result->coding_data_bit_size = encode_frame_result->coding_data_byte_size * 8; @@ -511,9 +551,16 @@ static void update_encode_frame_result( update_motion_vector_info(encode_frame_info->motion_vector_info, encode_frame_result->num_rows_4x4, encode_frame_result->num_cols_4x4, - &encode_frame_result->motion_vector_info[0]); + &encode_frame_result->motion_vector_info[0], + kMotionVectorSubPixelPrecision); update_frame_counts(&encode_frame_info->frame_counts, &encode_frame_result->frame_counts); + if (encode_frame_result->frame_type == kFrameTypeAltRef) { + update_tpl_stats_info(encode_frame_info->tpl_stats_info, show_frame_count, + &encode_frame_result->tpl_stats_info[0]); + } + encode_frame_result_update_rq_history(&encode_frame_info->rq_history, + encode_frame_result); } static void IncreaseGroupOfPictureIndex(GroupOfPicture *group_of_picture) { @@ -612,6 +659,9 @@ static void SetGroupOfPicture(int first_is_key_frame, int use_alt_ref, group_of_picture->show_frame_count = coding_frame_count - use_alt_ref; group_of_picture->start_show_index = first_show_idx; group_of_picture->start_coding_index = start_coding_index; + group_of_picture->first_is_key_frame = first_is_key_frame; + group_of_picture->use_alt_ref = use_alt_ref; + group_of_picture->last_gop_use_alt_ref = last_gop_use_alt_ref; // We need to make a copy of start reference frame info because we // use it to simulate the ref frame update. @@ -692,6 +742,60 @@ static void UpdateGroupOfPicture(const VP9_COMP *cpi, int start_coding_index, start_ref_frame_info, group_of_picture); } +#define SET_STRUCT_VALUE(config, structure, ret, field) \ + if (strcmp(config.name, #field) == 0) { \ + structure->field = atoi(config.value); \ + ret = 1; \ + } + +static void UpdateEncodeConfig(const EncodeConfig &config, + VP9EncoderConfig *oxcf) { + int ret = 0; + SET_STRUCT_VALUE(config, oxcf, ret, key_freq); + SET_STRUCT_VALUE(config, oxcf, ret, two_pass_vbrmin_section); + SET_STRUCT_VALUE(config, oxcf, ret, two_pass_vbrmax_section); + SET_STRUCT_VALUE(config, oxcf, ret, under_shoot_pct); + SET_STRUCT_VALUE(config, oxcf, ret, over_shoot_pct); + SET_STRUCT_VALUE(config, oxcf, ret, max_threads); + SET_STRUCT_VALUE(config, oxcf, ret, frame_parallel_decoding_mode); + SET_STRUCT_VALUE(config, oxcf, ret, tile_columns); + SET_STRUCT_VALUE(config, oxcf, ret, arnr_max_frames); + SET_STRUCT_VALUE(config, oxcf, ret, arnr_strength); + SET_STRUCT_VALUE(config, oxcf, ret, lag_in_frames); + SET_STRUCT_VALUE(config, oxcf, ret, encode_breakout); + SET_STRUCT_VALUE(config, oxcf, ret, enable_tpl_model); + SET_STRUCT_VALUE(config, oxcf, ret, enable_auto_arf); + if (strcmp(config.name, "rc_mode") == 0) { + int rc_mode = atoi(config.value); + if (rc_mode >= VPX_VBR && rc_mode <= VPX_Q) { + oxcf->rc_mode = (enum vpx_rc_mode)rc_mode; + ret = 1; + } else { + fprintf(stderr, "Invalid rc_mode value: %d\n", rc_mode); + } + } + SET_STRUCT_VALUE(config, oxcf, ret, cq_level); + if (ret == 0) { + fprintf(stderr, "Ignored unsupported encode_config %s\n", config.name); + } +} + +static VP9EncoderConfig GetEncodeConfig( + int frame_width, int frame_height, vpx_rational_t frame_rate, + int target_bitrate, int encode_speed, vpx_enc_pass enc_pass, + const std::vector<EncodeConfig> &encode_config_list) { + VP9EncoderConfig oxcf = + vp9_get_encoder_config(frame_width, frame_height, frame_rate, + target_bitrate, encode_speed, enc_pass); + for (const auto &config : encode_config_list) { + UpdateEncodeConfig(config, &oxcf); + } + if (enc_pass == VPX_RC_FIRST_PASS) { + oxcf.lag_in_frames = 0; + } + return oxcf; +} + SimpleEncode::SimpleEncode(int frame_width, int frame_height, int frame_rate_num, int frame_rate_den, int target_bitrate, int num_frames, @@ -703,6 +807,7 @@ SimpleEncode::SimpleEncode(int frame_width, int frame_height, frame_rate_den_ = frame_rate_den; target_bitrate_ = target_bitrate; num_frames_ = num_frames; + encode_speed_ = 0; frame_coding_index_ = 0; show_frame_count_ = 0; @@ -724,16 +829,55 @@ SimpleEncode::SimpleEncode(int frame_width, int frame_height, InitRefFrameInfo(&ref_frame_info_); } +void SimpleEncode::SetEncodeSpeed(int encode_speed) { + encode_speed_ = encode_speed; +} + +StatusCode SimpleEncode::SetEncodeConfig(const char *name, const char *value) { + if (name == nullptr || value == nullptr) { + fprintf(stderr, "SetEncodeConfig: null pointer, name %p value %p\n", name, + value); + return StatusError; + } + EncodeConfig config; + snprintf(config.name, ENCODE_CONFIG_BUF_SIZE, "%s", name); + snprintf(config.value, ENCODE_CONFIG_BUF_SIZE, "%s", value); + impl_ptr_->encode_config_list.push_back(config); + return StatusOk; +} + +StatusCode SimpleEncode::DumpEncodeConfigs(int pass, FILE *fp) { + if (fp == nullptr) { + fprintf(stderr, "DumpEncodeConfigs: null pointer, fp %p\n", fp); + return StatusError; + } + vpx_enc_pass enc_pass; + if (pass == 1) { + enc_pass = VPX_RC_FIRST_PASS; + } else { + enc_pass = VPX_RC_LAST_PASS; + } + const vpx_rational_t frame_rate = + make_vpx_rational(frame_rate_num_, frame_rate_den_); + const VP9EncoderConfig oxcf = + GetEncodeConfig(frame_width_, frame_height_, frame_rate, target_bitrate_, + encode_speed_, enc_pass, impl_ptr_->encode_config_list); + vp9_dump_encoder_config(&oxcf, fp); + return StatusOk; +} + void SimpleEncode::ComputeFirstPassStats() { vpx_rational_t frame_rate = make_vpx_rational(frame_rate_num_, frame_rate_den_); - const VP9EncoderConfig oxcf = - vp9_get_encoder_config(frame_width_, frame_height_, frame_rate, - target_bitrate_, VPX_RC_FIRST_PASS); + const VP9EncoderConfig oxcf = GetEncodeConfig( + frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_, + VPX_RC_FIRST_PASS, impl_ptr_->encode_config_list); VP9_COMP *cpi = init_encoder(&oxcf, impl_ptr_->img_fmt); struct lookahead_ctx *lookahead = cpi->lookahead; int i; int use_highbitdepth = 0; + const int num_rows_16x16 = get_num_unit_16x16(frame_height_); + const int num_cols_16x16 = get_num_unit_16x16(frame_width_); #if CONFIG_VP9_HIGHBITDEPTH use_highbitdepth = cpi->common.use_highbitdepth; #endif @@ -766,6 +910,12 @@ void SimpleEncode::ComputeFirstPassStats() { // vp9_get_compressed_data only generates first pass stats not // compresses data assert(size == 0); + // Get vp9 first pass motion vector info. + std::vector<MotionVectorInfo> mv_info(num_rows_16x16 * num_cols_16x16); + update_motion_vector_info(cpi->fp_motion_vector_info, num_rows_16x16, + num_cols_16x16, mv_info.data(), + kMotionVectorFullPixelPrecision); + fp_motion_vector_info_.push_back(mv_info); } impl_ptr_->first_pass_stats.push_back(vp9_get_frame_stats(&cpi->twopass)); } @@ -776,6 +926,9 @@ void SimpleEncode::ComputeFirstPassStats() { free_encoder(cpi); rewind(in_file_); vpx_img_free(&img); + + // Generate key_frame_map based on impl_ptr_->first_pass_stats. + key_frame_map_ = ComputeKeyFrameMap(); } std::vector<std::vector<double>> SimpleEncode::ObserveFirstPassStats() { @@ -800,9 +953,44 @@ std::vector<std::vector<double>> SimpleEncode::ObserveFirstPassStats() { return output_stats; } -void SimpleEncode::SetExternalGroupOfPicture( - std::vector<int> external_arf_indexes) { - external_arf_indexes_ = external_arf_indexes; +std::vector<std::vector<MotionVectorInfo>> +SimpleEncode::ObserveFirstPassMotionVectors() { + return fp_motion_vector_info_; +} + +void SimpleEncode::SetExternalGroupOfPicturesMap(int *gop_map, + int gop_map_size) { + for (int i = 0; i < gop_map_size; ++i) { + gop_map_.push_back(gop_map[i]); + } + // The following will check and modify gop_map_ to make sure the + // gop_map_ satisfies the constraints. + // 1) Each key frame position should be at the start of a gop. + // 2) The last gop should not use an alt ref. + assert(gop_map_.size() == key_frame_map_.size()); + int last_gop_start = 0; + for (int i = 0; static_cast<size_t>(i) < gop_map_.size(); ++i) { + if (key_frame_map_[i] == 1 && gop_map_[i] == 0) { + fprintf(stderr, "Add an extra gop start at show_idx %d\n", i); + // Insert a gop start at key frame location. + gop_map_[i] |= kGopMapFlagStart; + gop_map_[i] |= kGopMapFlagUseAltRef; + } + if (gop_map_[i] & kGopMapFlagStart) { + last_gop_start = i; + } + } + if (gop_map_[last_gop_start] & kGopMapFlagUseAltRef) { + fprintf(stderr, + "Last group of pictures starting at show_idx %d shouldn't use alt " + "ref\n", + last_gop_start); + gop_map_[last_gop_start] &= ~kGopMapFlagUseAltRef; + } +} + +std::vector<int> SimpleEncode::ObserveExternalGroupOfPicturesMap() { + return gop_map_; } template <typename T> @@ -813,13 +1001,40 @@ T *GetVectorData(const std::vector<T> &v) { return const_cast<T *>(v.data()); } +static GOP_COMMAND GetGopCommand(const std::vector<int> &gop_map, + int start_show_index) { + GOP_COMMAND gop_command; + if (gop_map.size() > 0) { + assert(static_cast<size_t>(start_show_index) < gop_map.size()); + assert((gop_map[start_show_index] & kGopMapFlagStart) != 0); + int end_show_index = start_show_index + 1; + // gop_map[end_show_index] & kGopMapFlagStart == 0 means this is + // the start of a gop. + while (static_cast<size_t>(end_show_index) < gop_map.size() && + (gop_map[end_show_index] & kGopMapFlagStart) == 0) { + ++end_show_index; + } + const int show_frame_count = end_show_index - start_show_index; + int use_alt_ref = (gop_map[start_show_index] & kGopMapFlagUseAltRef) != 0; + if (static_cast<size_t>(end_show_index) == gop_map.size()) { + // This is the last gop group, there must be no altref. + use_alt_ref = 0; + } + gop_command_on(&gop_command, show_frame_count, use_alt_ref); + } else { + gop_command_off(&gop_command); + } + return gop_command; +} + void SimpleEncode::StartEncode() { assert(impl_ptr_->first_pass_stats.size() > 0); vpx_rational_t frame_rate = make_vpx_rational(frame_rate_num_, frame_rate_den_); - VP9EncoderConfig oxcf = - vp9_get_encoder_config(frame_width_, frame_height_, frame_rate, - target_bitrate_, VPX_RC_LAST_PASS); + VP9EncoderConfig oxcf = GetEncodeConfig( + frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_, + VPX_RC_LAST_PASS, impl_ptr_->encode_config_list); + vpx_fixed_buf_t stats; stats.buf = GetVectorData(impl_ptr_->first_pass_stats); stats.sz = sizeof(impl_ptr_->first_pass_stats[0]) * @@ -834,11 +1049,10 @@ void SimpleEncode::StartEncode() { frame_coding_index_ = 0; show_frame_count_ = 0; - encode_command_set_external_arf_indexes(&impl_ptr_->cpi->encode_command, - GetVectorData(external_arf_indexes_)); - UpdateKeyFrameGroup(show_frame_count_); + const GOP_COMMAND gop_command = GetGopCommand(gop_map_, show_frame_count_); + encode_command_set_gop_command(&impl_ptr_->cpi->encode_command, gop_command); UpdateGroupOfPicture(impl_ptr_->cpi, frame_coding_index_, ref_frame_info_, &group_of_picture_); rewind(in_file_); @@ -914,6 +1128,9 @@ void SimpleEncode::PostUpdateState( IncreaseGroupOfPictureIndex(&group_of_picture_); if (IsGroupOfPictureFinished(group_of_picture_)) { + const GOP_COMMAND gop_command = GetGopCommand(gop_map_, show_frame_count_); + encode_command_set_gop_command(&impl_ptr_->cpi->encode_command, + gop_command); // This function needs to be called after ref_frame_info_ is updated // properly in PostUpdateRefFrameInfo() and UpdateKeyFrameGroup(). UpdateGroupOfPicture(impl_ptr_->cpi, frame_coding_index_, ref_frame_info_, @@ -985,7 +1202,10 @@ void SimpleEncode::EncodeFrame(EncodeFrameResult *encode_frame_result) { abort(); } - update_encode_frame_result(encode_frame_result, &encode_frame_info); + const GroupOfPicture group_of_picture = this->ObserveGroupOfPicture(); + const int show_frame_count = group_of_picture.show_frame_count; + update_encode_frame_result(encode_frame_result, show_frame_count, + &encode_frame_info); PostUpdateState(*encode_frame_result); } else { // TODO(angiebird): Clean up encode_frame_result. @@ -1002,26 +1222,73 @@ void SimpleEncode::EncodeFrameWithQuantizeIndex( encode_command_reset_external_quantize_index(&impl_ptr_->cpi->encode_command); } +void SimpleEncode::EncodeFrameWithTargetFrameBits( + EncodeFrameResult *encode_frame_result, int target_frame_bits, + double percent_diff) { + encode_command_set_target_frame_bits(&impl_ptr_->cpi->encode_command, + target_frame_bits, percent_diff); + EncodeFrame(encode_frame_result); + encode_command_reset_target_frame_bits(&impl_ptr_->cpi->encode_command); +} + +static int GetCodingFrameNumFromGopMap(const std::vector<int> &gop_map) { + int start_show_index = 0; + int coding_frame_count = 0; + while (static_cast<size_t>(start_show_index) < gop_map.size()) { + const GOP_COMMAND gop_command = GetGopCommand(gop_map, start_show_index); + start_show_index += gop_command.show_frame_count; + coding_frame_count += gop_command_coding_frame_count(&gop_command); + } + assert(start_show_index == gop_map.size()); + return coding_frame_count; +} + int SimpleEncode::GetCodingFrameNum() const { - assert(impl_ptr_->first_pass_stats.size() - 1 > 0); + assert(impl_ptr_->first_pass_stats.size() > 0); + if (gop_map_.size() > 0) { + return GetCodingFrameNumFromGopMap(gop_map_); + } + // These are the default settings for now. const int multi_layer_arf = 0; const int allow_alt_ref = 1; vpx_rational_t frame_rate = make_vpx_rational(frame_rate_num_, frame_rate_den_); - const VP9EncoderConfig oxcf = - vp9_get_encoder_config(frame_width_, frame_height_, frame_rate, - target_bitrate_, VPX_RC_LAST_PASS); + const VP9EncoderConfig oxcf = GetEncodeConfig( + frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_, + VPX_RC_LAST_PASS, impl_ptr_->encode_config_list); FRAME_INFO frame_info = vp9_get_frame_info(&oxcf); FIRST_PASS_INFO first_pass_info; fps_init_first_pass_info(&first_pass_info, GetVectorData(impl_ptr_->first_pass_stats), num_frames_); - return vp9_get_coding_frame_num(external_arf_indexes_.data(), &oxcf, - &frame_info, &first_pass_info, + return vp9_get_coding_frame_num(&oxcf, &frame_info, &first_pass_info, multi_layer_arf, allow_alt_ref); } +std::vector<int> SimpleEncode::ComputeKeyFrameMap() const { + // The last entry of first_pass_stats is the overall stats. + assert(impl_ptr_->first_pass_stats.size() == num_frames_ + 1); + vpx_rational_t frame_rate = + make_vpx_rational(frame_rate_num_, frame_rate_den_); + const VP9EncoderConfig oxcf = GetEncodeConfig( + frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_, + VPX_RC_LAST_PASS, impl_ptr_->encode_config_list); + FRAME_INFO frame_info = vp9_get_frame_info(&oxcf); + FIRST_PASS_INFO first_pass_info; + fps_init_first_pass_info(&first_pass_info, + GetVectorData(impl_ptr_->first_pass_stats), + num_frames_); + std::vector<int> key_frame_map(num_frames_, 0); + vp9_get_key_frame_map(&oxcf, &frame_info, &first_pass_info, + GetVectorData(key_frame_map)); + return key_frame_map; +} + +std::vector<int> SimpleEncode::ObserveKeyFrameMap() const { + return key_frame_map_; +} + uint64_t SimpleEncode::GetFramePixelCount() const { assert(frame_width_ % 2 == 0); assert(frame_height_ % 2 == 0); |