diff options
| -rw-r--r-- | test/android/scrape_gtest_log.py | 31 | ||||
| -rw-r--r-- | vp9/encoder/vp9_encodeframe.c | 96 | ||||
| -rw-r--r-- | vp9/encoder/vp9_onyx_if.c | 109 | ||||
| -rw-r--r-- | vp9/encoder/vp9_onyx_int.h | 3 | ||||
| -rw-r--r-- | vp9/encoder/vp9_picklpf.c | 2 | ||||
| -rw-r--r-- | vp9/encoder/vp9_speed_features.c | 90 |
6 files changed, 152 insertions, 179 deletions
diff --git a/test/android/scrape_gtest_log.py b/test/android/scrape_gtest_log.py index c159c8938..487845c27 100644 --- a/test/android/scrape_gtest_log.py +++ b/test/android/scrape_gtest_log.py @@ -13,16 +13,45 @@ waterfall to gather json results mixed in with gtest logs. This is dubious software engineering. """ +import getopt import json +import os import re import sys def main(): + if len(sys.argv) != 3: + print "Expects a file to write json to!" + exit(1) + + try: + opts, _ = \ + getopt.getopt(sys.argv[1:], \ + 'o:', ['output-json=']) + except getopt.GetOptError: + print 'scrape_gtest_log.py -o <output_json>' + sys.exit(2) + + output_json = '' + for opt, arg in opts: + if opt in ('-o', '--output-json'): + output_json = os.path.join(arg) + blob = sys.stdin.read() json_string = '[' + ','.join('{' + x + '}' for x in re.findall(r'{([^}]*.?)}', blob)) + ']' - print json.dumps(json.loads(json_string), indent=4, sort_keys=True) + print blob + + output = json.dumps(json.loads(json_string), indent=4, sort_keys=True) + print output + + path = os.path.dirname(output_json) + if path and not os.path.exists(path): + os.makedirs(path) + + outfile = open(output_json, 'w') + outfile.write(output) if __name__ == '__main__': sys.exit(main()) diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c index 1ca3c2881..929722563 100644 --- a/vp9/encoder/vp9_encodeframe.c +++ b/vp9/encoder/vp9_encodeframe.c @@ -1048,9 +1048,9 @@ static void rd_pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile, (mi_col <= 1) || (mi_col >= (cm->mi_cols - 2)); if (!is_edge && (complexity > 128)) x->rdmult += ((x->rdmult * (complexity - 128)) / 256); - } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) { + } else if (aq_mode == CYCLIC_REFRESH_AQ) { const uint8_t *const map = cm->seg.update_map ? cpi->segmentation_map - : cm->last_frame_seg_map; + : cm->last_frame_seg_map; // If segment 1, use rdmult for that segment. if (vp9_get_segment_id(cm, map, bsize, mi_row, mi_col)) x->rdmult = vp9_cyclic_refresh_get_rdmult(cpi->cyclic_refresh); @@ -1076,8 +1076,7 @@ static void rd_pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile, vp9_clear_system_state(); *totalrate = (int)round(*totalrate * rdmult_ratio); } - } else if ((cpi->oxcf.aq_mode == COMPLEXITY_AQ) || - (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ)) { + } else if (aq_mode == COMPLEXITY_AQ || aq_mode == CYCLIC_REFRESH_AQ) { x->rdmult = orig_rdmult; } } @@ -2297,6 +2296,7 @@ static void encode_rd_sb_row(VP9_COMP *cpi, const TileInfo *const tile, int mi_row, TOKENEXTRA **tp) { VP9_COMMON *const cm = &cpi->common; MACROBLOCKD *const xd = &cpi->mb.e_mbd; + SPEED_FEATURES *const sf = &cpi->sf; int mi_col; // Initialize the left context for the new SB row @@ -2312,7 +2312,7 @@ static void encode_rd_sb_row(VP9_COMP *cpi, const TileInfo *const tile, BLOCK_SIZE i; MACROBLOCK *x = &cpi->mb; - if (cpi->sf.adaptive_pred_interp_filter) { + if (sf->adaptive_pred_interp_filter) { for (i = BLOCK_4X4; i < BLOCK_8X8; ++i) { const int num_4x4_w = num_4x4_blocks_wide_lookup[i]; const int num_4x4_h = num_4x4_blocks_high_lookup[i]; @@ -2326,48 +2326,48 @@ static void encode_rd_sb_row(VP9_COMP *cpi, const TileInfo *const tile, vp9_zero(cpi->mb.pred_mv); - if ((cpi->sf.partition_search_type == SEARCH_PARTITION && - cpi->sf.use_lastframe_partitioning) || - cpi->sf.partition_search_type == FIXED_PARTITION || - cpi->sf.partition_search_type == VAR_BASED_PARTITION || - cpi->sf.partition_search_type == VAR_BASED_FIXED_PARTITION) { + if ((sf->partition_search_type == SEARCH_PARTITION && + sf->use_lastframe_partitioning) || + sf->partition_search_type == FIXED_PARTITION || + sf->partition_search_type == VAR_BASED_PARTITION || + sf->partition_search_type == VAR_BASED_FIXED_PARTITION) { const int idx_str = cm->mi_stride * mi_row + mi_col; MODE_INFO **mi_8x8 = cm->mi_grid_visible + idx_str; MODE_INFO **prev_mi_8x8 = cm->prev_mi_grid_visible + idx_str; cpi->mb.source_variance = UINT_MAX; - if (cpi->sf.partition_search_type == FIXED_PARTITION) { + if (sf->partition_search_type == FIXED_PARTITION) { set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64); set_fixed_partitioning(cpi, tile, mi_8x8, mi_row, mi_col, - cpi->sf.always_this_block_size); + sf->always_this_block_size); rd_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64, &dummy_rate, &dummy_dist, 1); - } else if (cpi->sf.partition_search_type == VAR_BASED_FIXED_PARTITION) { + } else if (sf->partition_search_type == VAR_BASED_FIXED_PARTITION) { BLOCK_SIZE bsize; set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64); bsize = get_rd_var_based_fixed_partition(cpi, mi_row, mi_col); set_fixed_partitioning(cpi, tile, mi_8x8, mi_row, mi_col, bsize); rd_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64, &dummy_rate, &dummy_dist, 1); - } else if (cpi->sf.partition_search_type == VAR_BASED_PARTITION) { + } else if (sf->partition_search_type == VAR_BASED_PARTITION) { choose_partitioning(cpi, tile, mi_row, mi_col); rd_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64, &dummy_rate, &dummy_dist, 1); } else { if ((cm->current_video_frame - % cpi->sf.last_partitioning_redo_frequency) == 0 + % sf->last_partitioning_redo_frequency) == 0 || cm->prev_mi == 0 || cm->show_frame == 0 || cm->frame_type == KEY_FRAME || cpi->rc.is_src_frame_alt_ref - || ((cpi->sf.use_lastframe_partitioning == + || ((sf->use_lastframe_partitioning == LAST_FRAME_PARTITION_LOW_MOTION) && sb_has_motion(cm, prev_mi_8x8))) { // If required set upper and lower partition size limits - if (cpi->sf.auto_min_max_partition_size) { + if (sf->auto_min_max_partition_size) { set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64); rd_auto_partition_range(cpi, tile, mi_row, mi_col, - &cpi->sf.min_partition_size, - &cpi->sf.max_partition_size); + &sf->min_partition_size, + &sf->max_partition_size); } rd_pick_partition(cpi, tile, tp, mi_row, mi_col, BLOCK_64X64, &dummy_rate, &dummy_dist, 1, INT64_MAX); @@ -2379,11 +2379,11 @@ static void encode_rd_sb_row(VP9_COMP *cpi, const TileInfo *const tile, } } else { // If required set upper and lower partition size limits - if (cpi->sf.auto_min_max_partition_size) { + if (sf->auto_min_max_partition_size) { set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64); rd_auto_partition_range(cpi, tile, mi_row, mi_col, - &cpi->sf.min_partition_size, - &cpi->sf.max_partition_size); + &sf->min_partition_size, + &sf->max_partition_size); } rd_pick_partition(cpi, tile, tp, mi_row, mi_col, BLOCK_64X64, &dummy_rate, &dummy_dist, 1, INT64_MAX); @@ -2448,19 +2448,6 @@ static int check_dual_ref_flags(VP9_COMP *cpi) { } } -static int get_skip_flag(MODE_INFO **mi_8x8, int mis, int ymbs, int xmbs) { - int x, y; - - for (y = 0; y < ymbs; y++) { - for (x = 0; x < xmbs; x++) { - if (!mi_8x8[y * mis + x]->mbmi.skip) - return 0; - } - } - - return 1; -} - static void reset_skip_txfm_size(VP9_COMMON *cm, TX_SIZE txfm_max) { int mi_row, mi_col; const int mis = cm->mi_stride; @@ -2911,12 +2898,7 @@ static void nonrd_use_partition(VP9_COMP *cpi, if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return; - if (bsize >= BLOCK_8X8) { - subsize = mi_8x8[0]->mbmi.sb_type; - } else { - subsize = BLOCK_4X4; - } - + subsize = (bsize >= BLOCK_8X8) ? mi_8x8[0]->mbmi.sb_type : BLOCK_4X4; partition = partition_lookup[bsl][subsize]; switch (partition) { @@ -3201,33 +3183,31 @@ void vp9_encode_frame(VP9_COMP *cpi) { if (cpi->sf.frame_parameter_update) { int i; - REFERENCE_MODE reference_mode; - /* - * This code does a single RD pass over the whole frame assuming - * either compound, single or hybrid prediction as per whatever has - * worked best for that type of frame in the past. - * It also predicts whether another coding mode would have worked - * better that this coding mode. If that is the case, it remembers - * that for subsequent frames. - * It does the same analysis for transform size selection also. - */ + + // This code does a single RD pass over the whole frame assuming + // either compound, single or hybrid prediction as per whatever has + // worked best for that type of frame in the past. + // It also predicts whether another coding mode would have worked + // better that this coding mode. If that is the case, it remembers + // that for subsequent frames. + // It does the same analysis for transform size selection also. const MV_REFERENCE_FRAME frame_type = get_frame_type(cpi); const int64_t *mode_thresh = cpi->rd_prediction_type_threshes[frame_type]; const int64_t *filter_thresh = cpi->rd_filter_threshes[frame_type]; /* prediction (compound, single or hybrid) mode selection */ - if (frame_type == 3 || !cm->allow_comp_inter_inter) - reference_mode = SINGLE_REFERENCE; + if (frame_type == ALTREF_FRAME || !cm->allow_comp_inter_inter) + cm->reference_mode = SINGLE_REFERENCE; else if (mode_thresh[COMPOUND_REFERENCE] > mode_thresh[SINGLE_REFERENCE] && mode_thresh[COMPOUND_REFERENCE] > mode_thresh[REFERENCE_MODE_SELECT] && check_dual_ref_flags(cpi) && cpi->static_mb_pct == 100) - reference_mode = COMPOUND_REFERENCE; + cm->reference_mode = COMPOUND_REFERENCE; else if (mode_thresh[SINGLE_REFERENCE] > mode_thresh[REFERENCE_MODE_SELECT]) - reference_mode = SINGLE_REFERENCE; + cm->reference_mode = SINGLE_REFERENCE; else - reference_mode = REFERENCE_MODE_SELECT; + cm->reference_mode = REFERENCE_MODE_SELECT; if (cm->interp_filter == SWITCHABLE) { if (frame_type != ALTREF_FRAME && @@ -3243,9 +3223,6 @@ void vp9_encode_frame(VP9_COMP *cpi) { } } - cpi->mb.e_mbd.lossless = cpi->oxcf.lossless; - cm->reference_mode = reference_mode; - encode_frame_internal(cpi); for (i = 0; i < REFERENCE_MODES; ++i) { @@ -3324,7 +3301,6 @@ void vp9_encode_frame(VP9_COMP *cpi) { } } } else { - cpi->mb.e_mbd.lossless = cpi->oxcf.lossless; cm->reference_mode = SINGLE_REFERENCE; // Force the usage of the BILINEAR interp_filter. cm->interp_filter = BILINEAR; diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c index d036347bb..07f527875 100644 --- a/vp9/encoder/vp9_onyx_if.c +++ b/vp9/encoder/vp9_onyx_if.c @@ -1677,53 +1677,42 @@ void vp9_remove_compressor(VP9_COMP *cpi) { #endif } +static int64_t get_sse(const uint8_t *a, int a_stride, + const uint8_t *b, int b_stride, + int width, int height) { + const int dw = width % 16; + const int dh = height % 16; + int64_t total_sse = 0; + unsigned int sse = 0; + int sum = 0; + int x, y; + + if (dw > 0) { + variance(&a[width - dw], a_stride, &b[width - dw], b_stride, + dw, height, &sse, &sum); + total_sse += sse; + } + if (dh > 0) { + variance(&a[(height - dh) * a_stride], a_stride, + &b[(height - dh) * b_stride], b_stride, + width - dw, dh, &sse, &sum); + total_sse += sse; + } -static uint64_t calc_plane_error(const uint8_t *orig, int orig_stride, - const uint8_t *recon, int recon_stride, - unsigned int cols, unsigned int rows) { - unsigned int row, col; - uint64_t total_sse = 0; - int diff; - - for (row = 0; row + 16 <= rows; row += 16) { - for (col = 0; col + 16 <= cols; col += 16) { - unsigned int sse; - - vp9_mse16x16(orig + col, orig_stride, recon + col, recon_stride, &sse); + for (y = 0; y < height / 16; ++y) { + const uint8_t *pa = a; + const uint8_t *pb = b; + for (x = 0; x < width / 16; ++x) { + vp9_mse16x16(pa, a_stride, pb, b_stride, &sse); total_sse += sse; - } - - /* Handle odd-sized width */ - if (col < cols) { - unsigned int border_row, border_col; - const uint8_t *border_orig = orig; - const uint8_t *border_recon = recon; - - for (border_row = 0; border_row < 16; border_row++) { - for (border_col = col; border_col < cols; border_col++) { - diff = border_orig[border_col] - border_recon[border_col]; - total_sse += diff * diff; - } - - border_orig += orig_stride; - border_recon += recon_stride; - } - } - - orig += orig_stride * 16; - recon += recon_stride * 16; - } - /* Handle odd-sized height */ - for (; row < rows; row++) { - for (col = 0; col < cols; col++) { - diff = orig[col] - recon[col]; - total_sse += diff * diff; + pa += 16; + pb += 16; } - orig += orig_stride; - recon += recon_stride; + a += 16 * a_stride; + b += 16 * b_stride; } return total_sse; @@ -1751,9 +1740,9 @@ static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b, const int w = widths[i]; const int h = heights[i]; const uint32_t samples = w * h; - const uint64_t sse = calc_plane_error(a_planes[i], a_strides[i], - b_planes[i], b_strides[i], - w, h); + const uint64_t sse = get_sse(a_planes[i], a_strides[i], + b_planes[i], b_strides[i], + w, h); psnr->sse[1 + i] = sse; psnr->samples[1 + i] = samples; psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, 255.0, (double)sse); @@ -2227,7 +2216,7 @@ static void output_frame_level_debug_stats(VP9_COMP *cpi) { vp9_clear_system_state(); - recon_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm)); + recon_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm)); if (cpi->twopass.total_left_stats.coded_error != 0.0) fprintf(f, "%10u %10d %10d %10d %10d %10d " @@ -2398,7 +2387,7 @@ static void encode_with_recode_loop(VP9_COMP *cpi, rc->this_key_frame_forced && (rc->projected_frame_size < rc->max_frame_bandwidth)) { int last_q = q; - int kf_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm)); + int kf_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm)); int high_err_target = cpi->ambient_err; int low_err_target = cpi->ambient_err >> 1; @@ -2755,7 +2744,7 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi, // fixed interval. Note the reconstruction error if it is the frame before // the force key frame if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) { - cpi->ambient_err = vp9_calc_ss_err(cpi->Source, get_frame_new_buffer(cm)); + cpi->ambient_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm)); } // If the encoder forced a KEY_FRAME decision @@ -3517,28 +3506,12 @@ void vp9_set_svc(VP9_COMP *cpi, int use_svc) { return; } -int vp9_calc_ss_err(const YV12_BUFFER_CONFIG *source, - const YV12_BUFFER_CONFIG *reference) { - int i, j; - int total = 0; - - const uint8_t *src = source->y_buffer; - const uint8_t *ref = reference->y_buffer; - - // Loop through the Y plane raw and reconstruction data summing - // (square differences) - for (i = 0; i < source->y_height; i += 16) { - for (j = 0; j < source->y_width; j += 16) { - unsigned int sse; - total += vp9_mse16x16(src + j, source->y_stride, - ref + j, reference->y_stride, &sse); - } - - src += 16 * source->y_stride; - ref += 16 * reference->y_stride; - } +int vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b) { + assert(a->y_crop_width == b->y_crop_width); + assert(a->y_crop_height == b->y_crop_height); - return total; + return (int)get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride, + a->y_crop_width, a->y_crop_height); } diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h index dd980ef9b..de13336c9 100644 --- a/vp9/encoder/vp9_onyx_int.h +++ b/vp9/encoder/vp9_onyx_int.h @@ -605,8 +605,7 @@ static INLINE int get_token_alloc(int mb_rows, int mb_cols) { return mb_rows * mb_cols * (16 * 16 * 3 + 4); } -int vp9_calc_ss_err(const YV12_BUFFER_CONFIG *source, - const YV12_BUFFER_CONFIG *reference); +int vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b); void vp9_alloc_compressor_data(VP9_COMP *cpi); diff --git a/vp9/encoder/vp9_picklpf.c b/vp9/encoder/vp9_picklpf.c index 92ad1e745..3ac85228b 100644 --- a/vp9/encoder/vp9_picklpf.c +++ b/vp9/encoder/vp9_picklpf.c @@ -35,7 +35,7 @@ static int try_filter_frame(const YV12_BUFFER_CONFIG *sd, VP9_COMP *const cpi, int filt_err; vp9_loop_filter_frame(cm, &cpi->mb.e_mbd, filt_level, 1, partial_frame); - filt_err = vp9_calc_ss_err(sd, cm->frame_to_show); + filt_err = vp9_get_y_sse(sd, cm->frame_to_show); // Re-instate the unfiltered frame vpx_yv12_copy_y(&cpi->last_frame_uf, cm->frame_to_show); diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c index 9ca1b5eda..b79e15979 100644 --- a/vp9/encoder/vp9_speed_features.c +++ b/vp9/encoder/vp9_speed_features.c @@ -40,24 +40,21 @@ (1 << THR_ALTR) | \ (1 << THR_GOLD)) -static void set_good_speed_feature(VP9_COMP *cpi, - VP9_COMMON *cm, - SPEED_FEATURES *sf, - int speed) { - int i; +static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm, + SPEED_FEATURES *sf, int speed) { sf->adaptive_rd_thresh = 1; - sf->recode_loop = ((speed < 1) ? ALLOW_RECODE : ALLOW_RECODE_KFMAXBW); + sf->recode_loop = (speed < 1) ? ALLOW_RECODE : ALLOW_RECODE_KFMAXBW; sf->allow_skip_recode = 1; if (speed >= 1) { sf->use_square_partition_only = !frame_is_intra_only(cm); sf->less_rectangular_check = 1; - sf->tx_size_search_method = vp9_frame_is_boosted(cpi) - ? USE_FULL_RD : USE_LARGESTALL; + sf->tx_size_search_method = vp9_frame_is_boosted(cpi) ? USE_FULL_RD + : USE_LARGESTALL; if (MIN(cm->width, cm->height) >= 720) - sf->disable_split_mask = cm->show_frame ? - DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT; + sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT + : DISABLE_ALL_INTER_SPLIT; else sf->disable_split_mask = DISABLE_COMPOUND_SPLIT; sf->use_rd_breakout = 1; @@ -74,19 +71,18 @@ static void set_good_speed_feature(VP9_COMP *cpi, sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V; sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V; } - // Additions or changes from speed 1 for speed >= 2. + if (speed >= 2) { - sf->tx_size_search_method = vp9_frame_is_boosted(cpi) - ? USE_FULL_RD : USE_LARGESTALL; + sf->tx_size_search_method = vp9_frame_is_boosted(cpi) ? USE_FULL_RD + : USE_LARGESTALL; if (MIN(cm->width, cm->height) >= 720) - sf->disable_split_mask = cm->show_frame ? - DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT; + sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT + : DISABLE_ALL_INTER_SPLIT; else sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY; sf->adaptive_pred_interp_filter = 2; - sf->reference_masking = 1; sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH | FLAG_SKIP_INTRA_BESTINTER | @@ -94,13 +90,12 @@ static void set_good_speed_feature(VP9_COMP *cpi, FLAG_SKIP_INTRA_LOWVAR; sf->disable_filter_search_var_thresh = 100; sf->comp_inter_joint_search_thresh = BLOCK_SIZES; - sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX; sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION; sf->adjust_partitioning_from_last_frame = 1; sf->last_partitioning_redo_frequency = 3; } - // Additions or changes for speed 3 and above + if (speed >= 3) { if (MIN(cm->width, cm->height) >= 720) sf->disable_split_mask = DISABLE_ALL_SPLIT; @@ -108,35 +103,32 @@ static void set_good_speed_feature(VP9_COMP *cpi, sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT; sf->recode_loop = ALLOW_RECODE_KFMAXBW; - sf->adaptive_rd_thresh = 3; sf->mode_skip_start = 6; sf->use_fast_coef_updates = ONE_LOOP_REDUCED; sf->use_fast_coef_costing = 1; } - // Additions or changes for speed 3 and above + if (speed >= 4) { sf->use_square_partition_only = 1; sf->tx_size_search_method = USE_LARGESTALL; sf->disable_split_mask = DISABLE_ALL_SPLIT; - sf->adaptive_rd_thresh = 4; - - // Add a couple more skip flags sf->mode_search_skip_flags |= FLAG_SKIP_COMP_REFMISMATCH | FLAG_EARLY_TERMINATE; - sf->disable_filter_search_var_thresh = 200; - sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL; sf->use_lp32x32fdct = 1; } + if (speed >= 5) { + int i; + sf->partition_search_type = FIXED_PARTITION; sf->optimize_coefficients = 0; sf->search_method = HEX; sf->disable_filter_search_var_thresh = 500; - for (i = 0; i < TX_SIZES; i++) { + for (i = 0; i < TX_SIZES; ++i) { sf->intra_y_mode_mask[i] = INTRA_DC_ONLY; sf->intra_uv_mode_mask[i] = INTRA_DC_ONLY; } @@ -144,8 +136,7 @@ static void set_good_speed_feature(VP9_COMP *cpi, } } -static void set_rt_speed_feature(VP9_COMMON *cm, - SPEED_FEATURES *sf, +static void set_rt_speed_feature(VP9_COMMON *cm, SPEED_FEATURES *sf, int speed) { sf->static_segmentation = 0; sf->adaptive_rd_thresh = 1; @@ -155,12 +146,12 @@ static void set_rt_speed_feature(VP9_COMMON *cm, if (speed == 1) { sf->use_square_partition_only = !frame_is_intra_only(cm); sf->less_rectangular_check = 1; - sf->tx_size_search_method = - frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL; + sf->tx_size_search_method = frame_is_intra_only(cm) ? USE_FULL_RD + : USE_LARGESTALL; if (MIN(cm->width, cm->height) >= 720) - sf->disable_split_mask = cm->show_frame ? - DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT; + sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT + : DISABLE_ALL_INTER_SPLIT; else sf->disable_split_mask = DISABLE_COMPOUND_SPLIT; @@ -174,22 +165,22 @@ static void set_rt_speed_feature(VP9_COMMON *cm, sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V; sf->encode_breakout_thresh = 8; } + if (speed >= 2) { sf->use_square_partition_only = !frame_is_intra_only(cm); sf->less_rectangular_check = 1; - sf->tx_size_search_method = - frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL; - + sf->tx_size_search_method = frame_is_intra_only(cm) ? USE_FULL_RD + : USE_LARGESTALL; if (MIN(cm->width, cm->height) >= 720) sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT; else sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY; - sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH - | FLAG_SKIP_INTRA_BESTINTER | FLAG_SKIP_COMP_BESTINTRA - | FLAG_SKIP_INTRA_LOWVAR; - + sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH | + FLAG_SKIP_INTRA_BESTINTER | + FLAG_SKIP_COMP_BESTINTRA | + FLAG_SKIP_INTRA_LOWVAR; sf->use_rd_breakout = 1; sf->adaptive_motion_search = 1; sf->adaptive_pred_interp_filter = 2; @@ -213,6 +204,7 @@ static void set_rt_speed_feature(VP9_COMMON *cm, sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V; sf->encode_breakout_thresh = 200; } + if (speed >= 3) { sf->use_square_partition_only = 1; sf->disable_filter_search_var_thresh = 100; @@ -229,6 +221,7 @@ static void set_rt_speed_feature(VP9_COMMON *cm, sf->lpf_pick = LPF_PICK_FROM_Q; sf->encode_breakout_thresh = 700; } + if (speed >= 4) { int i; sf->last_partitioning_redo_frequency = 4; @@ -254,6 +247,7 @@ static void set_rt_speed_feature(VP9_COMMON *cm, sf->max_intra_bsize = BLOCK_32X32; sf->allow_skip_recode = 1; } + if (speed >= 5) { sf->max_partition_size = BLOCK_32X32; sf->min_partition_size = BLOCK_8X8; @@ -268,21 +262,24 @@ static void set_rt_speed_feature(VP9_COMMON *cm, sf->search_method = FAST_DIAMOND; sf->allow_skip_recode = 0; } + if (speed >= 6) { sf->partition_search_type = VAR_BASED_FIXED_PARTITION; sf->use_nonrd_pick_mode = 1; sf->search_method = FAST_DIAMOND; } + if (speed >= 7) { int i; for (i = 0; i < BLOCK_SIZES; ++i) - sf->disable_inter_mode_mask[i] = 14; // only search NEARESTMV (0) + sf->disable_inter_mode_mask[i] = ~(1 << INTER_OFFSET(NEARESTMV)); } } void vp9_set_speed_features(VP9_COMP *cpi) { SPEED_FEATURES *const sf = &cpi->sf; VP9_COMMON *const cm = &cpi->common; + const VP9_CONFIG *const oxcf = &cpi->oxcf; const int speed = cpi->speed < 0 ? -cpi->speed : cpi->speed; int i; @@ -293,7 +290,7 @@ void vp9_set_speed_features(VP9_COMP *cpi) { sf->subpel_search_method = SUBPEL_TREE; sf->subpel_iters_per_step = 2; sf->subpel_force_stop = 0; - sf->optimize_coefficients = !cpi->oxcf.lossless; + sf->optimize_coefficients = !oxcf->lossless; sf->reduce_first_step_size = 0; sf->auto_mv_step_size = 0; sf->max_step_search_steps = MAX_MVSEARCH_STEPS; @@ -343,7 +340,7 @@ void vp9_set_speed_features(VP9_COMP *cpi) { // Recode loop tolerence %. sf->recode_tolerance = 25; - switch (cpi->oxcf.mode) { + switch (oxcf->mode) { case MODE_BESTQUALITY: case MODE_SECONDPASS_BEST: // This is the best quality mode. cpi->diamond_search_sad = vp9_full_range_search; @@ -360,9 +357,8 @@ void vp9_set_speed_features(VP9_COMP *cpi) { // Slow quant, dct and trellis not worthwhile for first pass // so make sure they are always turned off. - if (cpi->pass == 1) { + if (cpi->pass == 1) sf->optimize_coefficients = 0; - } // No recode for 1 pass. if (cpi->pass == 0) { @@ -370,14 +366,14 @@ void vp9_set_speed_features(VP9_COMP *cpi) { sf->optimize_coefficients = 0; } - if (cpi->sf.subpel_search_method == SUBPEL_TREE) { + if (sf->subpel_search_method == SUBPEL_TREE) { cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree; cpi->find_fractional_mv_step_comp = vp9_find_best_sub_pixel_comp_tree; } - cpi->mb.optimize = cpi->sf.optimize_coefficients == 1 && cpi->pass != 1; + cpi->mb.optimize = sf->optimize_coefficients == 1 && cpi->pass != 1; - if (cpi->encode_breakout && cpi->oxcf.mode == MODE_REALTIME && + if (cpi->encode_breakout && oxcf->mode == MODE_REALTIME && sf->encode_breakout_thresh > cpi->encode_breakout) cpi->encode_breakout = sf->encode_breakout_thresh; |