diff options
Diffstat (limited to 'vp8/encoder')
-rw-r--r-- | vp8/encoder/arm/arm_csystemdependent.c | 4 | ||||
-rw-r--r-- | vp8/encoder/arm/variance_arm.h | 8 | ||||
-rw-r--r-- | vp8/encoder/encodeframe.c | 446 | ||||
-rw-r--r-- | vp8/encoder/ethreading.c | 3 | ||||
-rw-r--r-- | vp8/encoder/firstpass.c | 67 | ||||
-rw-r--r-- | vp8/encoder/generic/csystemdependent.c | 2 | ||||
-rw-r--r-- | vp8/encoder/onyx_if.c | 61 | ||||
-rw-r--r-- | vp8/encoder/onyx_int.h | 8 | ||||
-rw-r--r-- | vp8/encoder/pickinter.c | 48 | ||||
-rw-r--r-- | vp8/encoder/pickinter.h | 3 | ||||
-rw-r--r-- | vp8/encoder/ppc/csystemdependent.c | 6 | ||||
-rw-r--r-- | vp8/encoder/ratectrl.c | 43 | ||||
-rw-r--r-- | vp8/encoder/rdopt.c | 84 | ||||
-rw-r--r-- | vp8/encoder/variance.h | 12 | ||||
-rw-r--r-- | vp8/encoder/variance_c.c | 34 | ||||
-rw-r--r-- | vp8/encoder/x86/variance_mmx.c | 30 | ||||
-rw-r--r-- | vp8/encoder/x86/variance_x86.h | 13 | ||||
-rw-r--r-- | vp8/encoder/x86/x86_csystemdependent.c | 5 |
18 files changed, 455 insertions, 422 deletions
diff --git a/vp8/encoder/arm/arm_csystemdependent.c b/vp8/encoder/arm/arm_csystemdependent.c index db079d5ed..75e3a53d2 100644 --- a/vp8/encoder/arm/arm_csystemdependent.c +++ b/vp8/encoder/arm/arm_csystemdependent.c @@ -54,8 +54,6 @@ void vp8_arch_arm_encoder_init(VP8_COMP *cpi) /*cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c;*/ /*cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_c; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;; cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_c;*/ /*cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_c; @@ -104,8 +102,6 @@ void vp8_arch_arm_encoder_init(VP8_COMP *cpi) /*cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c;*/ cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_neon; - /*cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;*/ cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_neon; cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_neon; diff --git a/vp8/encoder/arm/variance_arm.h b/vp8/encoder/arm/variance_arm.h index ad0d37193..cbfc753b3 100644 --- a/vp8/encoder/arm/variance_arm.h +++ b/vp8/encoder/arm/variance_arm.h @@ -84,8 +84,6 @@ extern prototype_variance(vp8_variance_halfpixvar16x16_hv_neon); //extern prototype_getmbss(vp8_get_mb_ss_c); extern prototype_variance(vp8_mse16x16_neon); extern prototype_get16x16prederror(vp8_get16x16pred_error_neon); -//extern prototype_variance2(vp8_get8x8var_c); -//extern prototype_variance2(vp8_get16x16var_c); extern prototype_get16x16prederror(vp8_get4x4sse_cs_neon); #if !CONFIG_RUNTIME_CPU_DETECT @@ -152,12 +150,6 @@ extern prototype_get16x16prederror(vp8_get4x4sse_cs_neon); #undef vp8_variance_get16x16prederror #define vp8_variance_get16x16prederror vp8_get16x16pred_error_neon -//#undef vp8_variance_get8x8var -//#define vp8_variance_get8x8var vp8_get8x8var_c - -//#undef vp8_variance_get16x16var -//#define vp8_variance_get16x16var vp8_get16x16var_c - #undef vp8_variance_get4x4sse_cs #define vp8_variance_get4x4sse_cs vp8_get4x4sse_cs_neon #endif diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c index ecbf0265f..5cae99f41 100644 --- a/vp8/encoder/encodeframe.c +++ b/vp8/encoder/encodeframe.c @@ -50,6 +50,7 @@ void vp8_build_block_offsets(MACROBLOCK *x); void vp8_setup_block_ptrs(MACROBLOCK *x); int vp8cx_encode_inter_macroblock(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t, int recon_yoffset, int recon_uvoffset); int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t); +static void adjust_act_zbin( VP8_COMP *cpi, MACROBLOCK *x ); #ifdef MODE_STATS unsigned int inter_y_modes[10] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0}; @@ -84,8 +85,6 @@ static unsigned int tt_activity_measure( VP8_COMP *cpi, MACROBLOCK *x ) { unsigned int act; unsigned int sse; - int sum; - /* TODO: This could also be done over smaller areas (8x8), but that would * require extensive changes elsewhere, as lambda is assumed to be fixed * over an entire MB in most of the code. @@ -93,14 +92,9 @@ static unsigned int tt_activity_measure( VP8_COMP *cpi, MACROBLOCK *x ) * lambda using a non-linear combination (e.g., the smallest, or second * smallest, etc.). */ - VARIANCE_INVOKE(&cpi->rtcd.variance, get16x16var)(x->src.y_buffer, - x->src.y_stride, VP8_VAR_OFFS, 0, &sse, &sum); - - /* This requires a full 32 bits of precision. */ - act = (sse<<8) - sum*sum; - - /* Drop 4 to give us some headroom to work with. */ - act = (act + 8) >> 4; + act = VARIANCE_INVOKE(&cpi->rtcd.variance, var16x16)(x->src.y_buffer, + x->src.y_stride, VP8_VAR_OFFS, 0, &sse); + act = act<<4; /* If the region is flat, lower the activity some more. */ if (act < 8<<12) @@ -110,70 +104,122 @@ static unsigned int tt_activity_measure( VP8_COMP *cpi, MACROBLOCK *x ) } // Stub for alternative experimental activity measures. -static unsigned int alt_activity_measure( VP8_COMP *cpi, MACROBLOCK *x ) +extern int encode_intra(VP8_COMP *cpi, MACROBLOCK *x, int use_dc_pred); +static unsigned int alt_activity_measure( VP8_COMP *cpi, + MACROBLOCK *x, int use_dc_pred ) { - unsigned int mb_activity = VP8_ACTIVITY_AVG_MIN; - - x->e_mbd.mode_info_context->mbmi.mode = DC_PRED; - x->e_mbd.mode_info_context->mbmi.uv_mode = DC_PRED; - x->e_mbd.mode_info_context->mbmi.ref_frame = INTRA_FRAME; - - vp8_encode_intra16x16mby(IF_RTCD(&cpi->rtcd), x); - - mb_activity = VARIANCE_INVOKE(&cpi->rtcd.variance, getmbss)(x->src_diff); - - return mb_activity; + return encode_intra(cpi,x, use_dc_pred); } // Measure the activity of the current macroblock // What we measure here is TBD so abstracted to this function -static unsigned int mb_activity_measure( VP8_COMP *cpi, MACROBLOCK *x ) +#define ALT_ACT_MEASURE 1 +static unsigned int mb_activity_measure( VP8_COMP *cpi, MACROBLOCK *x, + int mb_row, int mb_col) { unsigned int mb_activity; - if ( 1 ) + if ( ALT_ACT_MEASURE ) { - // Original activity measure from Tim T's code. - mb_activity = tt_activity_measure( cpi, x ); + int use_dc_pred = (mb_col || mb_row) && (!mb_col || !mb_row); + + // Or use and alternative. + mb_activity = alt_activity_measure( cpi, x, use_dc_pred ); } else { - // Or use and alternative. - mb_activity = alt_activity_measure( cpi, x ); + // Original activity measure from Tim T's code. + mb_activity = tt_activity_measure( cpi, x ); } + if ( mb_activity < VP8_ACTIVITY_AVG_MIN ) + mb_activity = VP8_ACTIVITY_AVG_MIN; + return mb_activity; } // Calculate an "average" mb activity value for the frame +#define ACT_MEDIAN 0 static void calc_av_activity( VP8_COMP *cpi, INT64 activity_sum ) { +#if ACT_MEDIAN + // Find median: Simple n^2 algorithm for experimentation + { + unsigned int median; + unsigned int i,j; + unsigned int * sortlist; + unsigned int tmp; + + // Create a list to sort to + CHECK_MEM_ERROR(sortlist, + vpx_calloc(sizeof(unsigned int), + cpi->common.MBs)); + + // Copy map to sort list + vpx_memcpy( sortlist, cpi->mb_activity_map, + sizeof(unsigned int) * cpi->common.MBs ); + + + // Ripple each value down to its correct position + for ( i = 1; i < cpi->common.MBs; i ++ ) + { + for ( j = i; j > 0; j -- ) + { + if ( sortlist[j] < sortlist[j-1] ) + { + // Swap values + tmp = sortlist[j-1]; + sortlist[j-1] = sortlist[j]; + sortlist[j] = tmp; + } + else + break; + } + } + + // Even number MBs so estimate median as mean of two either side. + median = ( 1 + sortlist[cpi->common.MBs >> 1] + + sortlist[(cpi->common.MBs >> 1) + 1] ) >> 1; + + cpi->activity_avg = median; + + vpx_free(sortlist); + } +#else // Simple mean for now cpi->activity_avg = (unsigned int)(activity_sum/cpi->common.MBs); +#endif + if (cpi->activity_avg < VP8_ACTIVITY_AVG_MIN) cpi->activity_avg = VP8_ACTIVITY_AVG_MIN; + + // Experimental code: return fixed value normalized for several clips + if ( ALT_ACT_MEASURE ) + cpi->activity_avg = 100000; } +#define USE_ACT_INDEX 0 #define OUTPUT_NORM_ACT_STATS 0 -// Calculate a normalized activity value for each mb -static void calc_norm_activity( VP8_COMP *cpi, MACROBLOCK *x ) + +#if USE_ACT_INDEX +// Calculate and activity index for each mb +static void calc_activity_index( VP8_COMP *cpi, MACROBLOCK *x ) { VP8_COMMON *const cm = & cpi->common; int mb_row, mb_col; - unsigned int act; - unsigned int a; - unsigned int b; + INT64 act; + INT64 a; + INT64 b; #if OUTPUT_NORM_ACT_STATS FILE *f = fopen("norm_act.stt", "a"); - fprintf(f, "\n"); + fprintf(f, "\n%12d\n", cpi->activity_avg ); #endif // Reset pointers to start of activity map x->mb_activity_ptr = cpi->mb_activity_map; - x->mb_norm_activity_ptr = cpi->mb_norm_activity_map; // Calculate normalized mb activity number. for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) @@ -185,25 +231,19 @@ static void calc_norm_activity( VP8_COMP *cpi, MACROBLOCK *x ) act = *(x->mb_activity_ptr); // Calculate a normalized activity number - a = act + 2*cpi->activity_avg; - b = 2*act + cpi->activity_avg; + a = act + 4*cpi->activity_avg; + b = 4*act + cpi->activity_avg; if ( b >= a ) - *(x->mb_norm_activity_ptr) = (int)((b + (a>>1))/a); + *(x->activity_ptr) = (int)((b + (a>>1))/a) - 1; else - *(x->mb_norm_activity_ptr) = -(int)((a + (b>>1))/b); - - if ( *(x->mb_norm_activity_ptr) == 0 ) - { - *(x->mb_norm_activity_ptr) = 1; - } + *(x->activity_ptr) = 1 - (int)((a + (b>>1))/b); #if OUTPUT_NORM_ACT_STATS - fprintf(f, " %6d", *(x->mb_norm_activity_ptr)); + fprintf(f, " %6d", *(x->mb_activity_ptr)); #endif // Increment activity map pointers x->mb_activity_ptr++; - x->mb_norm_activity_ptr++; } #if OUTPUT_NORM_ACT_STATS @@ -217,33 +257,44 @@ static void calc_norm_activity( VP8_COMP *cpi, MACROBLOCK *x ) #endif } - +#endif // Loop through all MBs. Note activity of each, average activity and // calculate a normalized activity for each static void build_activity_map( VP8_COMP *cpi ) { MACROBLOCK *const x = & cpi->mb; + MACROBLOCKD *xd = &x->e_mbd; VP8_COMMON *const cm = & cpi->common; +#if ALT_ACT_MEASURE + YV12_BUFFER_CONFIG *new_yv12 = &cm->yv12_fb[cm->new_fb_idx]; + int recon_yoffset; + int recon_y_stride = new_yv12->y_stride; +#endif + int mb_row, mb_col; unsigned int mb_activity; INT64 activity_sum = 0; - // Initialise source buffer pointer - x->src = *cpi->Source; - - // Set pointer to start of activity map - x->mb_activity_ptr = cpi->mb_activity_map; - // for each macroblock row in image for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) { +#if ALT_ACT_MEASURE + // reset above block coeffs + xd->up_available = (mb_row != 0); + recon_yoffset = (mb_row * recon_y_stride * 16); +#endif // for each macroblock col in image for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) { +#if ALT_ACT_MEASURE + xd->dst.y_buffer = new_yv12->y_buffer + recon_yoffset; + xd->left_available = (mb_col != 0); + recon_yoffset += 16; +#endif // measure activity - mb_activity = mb_activity_measure( cpi, x ); + mb_activity = mb_activity_measure( cpi, x, mb_row, mb_col ); // Keep frame sum activity_sum += mb_activity; @@ -258,49 +309,50 @@ static void build_activity_map( VP8_COMP *cpi ) x->src.y_buffer += 16; } + // adjust to the next row of mbs x->src.y_buffer += 16 * x->src.y_stride - 16 * cm->mb_cols; + +#if ALT_ACT_MEASURE + //extend the recon for intra prediction + vp8_extend_mb_row(new_yv12, xd->dst.y_buffer + 16, + xd->dst.u_buffer + 8, xd->dst.v_buffer + 8); +#endif + } // Calculate an "average" MB activity calc_av_activity(cpi, activity_sum); - // Calculate a normalized activity number of each mb - calc_norm_activity( cpi, x ); +#if USE_ACT_INDEX + // Calculate an activity index number of each mb + calc_activity_index( cpi, x ); +#endif + } -// Activity masking based on Tim T's original code +// Macroblock activity masking void vp8_activity_masking(VP8_COMP *cpi, MACROBLOCK *x) { - - unsigned int a; - unsigned int b; - unsigned int act = *(x->mb_activity_ptr); +#if USE_ACT_INDEX + x->rdmult += *(x->mb_activity_ptr) * (x->rdmult >> 2); + x->errorperbit = x->rdmult/x->rddiv; +#else + INT64 a; + INT64 b; + INT64 act = *(x->mb_activity_ptr); // Apply the masking to the RD multiplier. - a = act + 2*cpi->activity_avg; - b = 2*act + cpi->activity_avg; + a = act + (2*cpi->activity_avg); + b = (2*act) + cpi->activity_avg; - //tmp = (unsigned int)(((INT64)tmp*b + (a>>1))/a); x->rdmult = (unsigned int)(((INT64)x->rdmult*b + (a>>1))/a); + x->errorperbit = x->rdmult/x->rddiv; - // For now now zbin adjustment on mode choice - x->act_zbin_adj = 0; -} - -// Stub function to use a normalized activity measure stored at mb level. -void vp8_norm_activity_masking(VP8_COMP *cpi, MACROBLOCK *x) -{ - int norm_act; - - norm_act = *(x->mb_norm_activity_ptr); - if (norm_act > 0) - x->rdmult = norm_act * (x->rdmult); - else - x->rdmult = -(x->rdmult / norm_act); +#endif - // For now now zbin adjustment on mode choice - x->act_zbin_adj = 0; + // Activity based Zbin adjustment + adjust_act_zbin(cpi, x); } static @@ -356,7 +408,6 @@ void encode_mb_row(VP8_COMP *cpi, // Set the mb activity pointer to the start of the row. x->mb_activity_ptr = &cpi->mb_activity_map[map_index]; - x->mb_norm_activity_ptr = &cpi->mb_norm_activity_map[map_index]; // for each macroblock col in image for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) @@ -476,7 +527,6 @@ void encode_mb_row(VP8_COMP *cpi, // Increment the activity mask pointers. x->mb_activity_ptr++; - x->mb_norm_activity_ptr++; /* save the block info */ for (i = 0; i < 16; i++) @@ -525,6 +575,65 @@ void encode_mb_row(VP8_COMP *cpi, #endif } +void init_encode_frame_mb_context(VP8_COMP *cpi) +{ + MACROBLOCK *const x = & cpi->mb; + VP8_COMMON *const cm = & cpi->common; + MACROBLOCKD *const xd = & x->e_mbd; + + // GF active flags data structure + x->gf_active_ptr = (signed char *)cpi->gf_active_flags; + + // Activity map pointer + x->mb_activity_ptr = cpi->mb_activity_map; + + x->vector_range = 32; + + x->act_zbin_adj = 0; + + x->partition_info = x->pi; + + xd->mode_info_context = cm->mi; + xd->mode_info_stride = cm->mode_info_stride; + + xd->frame_type = cm->frame_type; + + xd->frames_since_golden = cm->frames_since_golden; + xd->frames_till_alt_ref_frame = cm->frames_till_alt_ref_frame; + + // reset intra mode contexts + if (cm->frame_type == KEY_FRAME) + vp8_init_mbmode_probs(cm); + + // Copy data over into macro block data sturctures. + x->src = * cpi->Source; + xd->pre = cm->yv12_fb[cm->lst_fb_idx]; + xd->dst = cm->yv12_fb[cm->new_fb_idx]; + + // set up frame for intra coded blocks + vp8_setup_intra_recon(&cm->yv12_fb[cm->new_fb_idx]); + + vp8_build_block_offsets(x); + + vp8_setup_block_dptrs(&x->e_mbd); + + vp8_setup_block_ptrs(x); + + xd->mode_info_context->mbmi.mode = DC_PRED; + xd->mode_info_context->mbmi.uv_mode = DC_PRED; + + xd->left_context = &cm->left_context; + + vp8_zero(cpi->count_mb_ref_frame_usage) + vp8_zero(cpi->ymode_count) + vp8_zero(cpi->uv_mode_count) + + x->mvc = cm->fc.mvc; + + vpx_memset(cm->above_context, 0, + sizeof(ENTROPY_CONTEXT_PLANES) * cm->mb_cols); +} + void vp8_encode_frame(VP8_COMP *cpi) { int mb_row; @@ -536,6 +645,17 @@ void vp8_encode_frame(VP8_COMP *cpi) int segment_counts[MAX_MB_SEGMENTS]; int totalrate; + vpx_memset(segment_counts, 0, sizeof(segment_counts)); + totalrate = 0; + + if (cpi->compressor_speed == 2) + { + if (cpi->oxcf.cpu_used < 0) + cpi->Speed = -(cpi->oxcf.cpu_used); + else + vp8_auto_select_speed(cpi); + } + // Functions setup for all frame types so we can use MC in AltRef if (cm->mcomp_filter_type == SIXTAP) { @@ -560,10 +680,6 @@ void vp8_encode_frame(VP8_COMP *cpi) &cpi->common.rtcd.subpix, bilinear16x16); } - x->gf_active_ptr = (signed char *)cpi->gf_active_flags; // Point to base of GF active flags data structure - - x->vector_range = 32; - // Reset frame count of inter 0,0 motion vector useage. cpi->inter_zz_count = 0; @@ -574,89 +690,34 @@ void vp8_encode_frame(VP8_COMP *cpi) cpi->skip_true_count = 0; cpi->skip_false_count = 0; - x->act_zbin_adj = 0; - #if 0 // Experimental code cpi->frame_distortion = 0; cpi->last_mb_distortion = 0; #endif - totalrate = 0; - - x->partition_info = x->pi; - - xd->mode_info_context = cm->mi; - xd->mode_info_stride = cm->mode_info_stride; - - xd->frame_type = cm->frame_type; - - xd->frames_since_golden = cm->frames_since_golden; - xd->frames_till_alt_ref_frame = cm->frames_till_alt_ref_frame; vp8_zero(cpi->MVcount); - // vp8_zero( Contexts) vp8_zero(cpi->coef_counts); - // reset intra mode contexts - if (cm->frame_type == KEY_FRAME) - vp8_init_mbmode_probs(cm); - - vp8cx_frame_init_quantizer(cpi); - if (cpi->compressor_speed == 2) - { - if (cpi->oxcf.cpu_used < 0) - cpi->Speed = -(cpi->oxcf.cpu_used); - else - vp8_auto_select_speed(cpi); - } + vp8_initialize_rd_consts(cpi, + vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q)); - vp8_initialize_rd_consts(cpi, vp8_dc_quant(cm->base_qindex, cm->y1dc_delta_q)); vp8cx_initialize_me_consts(cpi, cm->base_qindex); - // Copy data over into macro block data sturctures. - x->src = * cpi->Source; - xd->pre = cm->yv12_fb[cm->lst_fb_idx]; - xd->dst = cm->yv12_fb[cm->new_fb_idx]; - - // set up frame new frame for intra coded blocks - - vp8_setup_intra_recon(&cm->yv12_fb[cm->new_fb_idx]); - - vp8_build_block_offsets(x); - - vp8_setup_block_dptrs(&x->e_mbd); - - vp8_setup_block_ptrs(x); - - xd->mode_info_context->mbmi.mode = DC_PRED; - xd->mode_info_context->mbmi.uv_mode = DC_PRED; - - xd->left_context = &cm->left_context; - - vp8_zero(cpi->count_mb_ref_frame_usage) - vp8_zero(cpi->ymode_count) - vp8_zero(cpi->uv_mode_count) - - x->mvc = cm->fc.mvc; - - vpx_memset(cm->above_context, 0, sizeof(ENTROPY_CONTEXT_PLANES) * cm->mb_cols); - if(cpi->oxcf.tuning == VP8_TUNE_SSIM) { - if(1) - { - // Build a frame level activity map - build_activity_map(cpi); - } + // Initialize encode frame context. + init_encode_frame_mb_context(cpi); - // Reset various MB pointers. - x->src = *cpi->Source; - x->mb_activity_ptr = cpi->mb_activity_map; - x->mb_norm_activity_ptr = cpi->mb_norm_activity_map; + // Build a frame level activity map + build_activity_map(cpi); } + // re-initencode frame context. + init_encode_frame_mb_context(cpi); + { struct vpx_usec_timer emr_timer; vpx_usec_timer_start(&emr_timer); @@ -997,27 +1058,24 @@ static void sum_intra_stats(VP8_COMP *cpi, MACROBLOCK *x) // Experimental stub function to create a per MB zbin adjustment based on // some previously calculated measure of MB activity. -void adjust_act_zbin( VP8_COMP *cpi, int rate, MACROBLOCK *x ) +static void adjust_act_zbin( VP8_COMP *cpi, MACROBLOCK *x ) { - INT64 act; +#if USE_ACT_INDEX + x->act_zbin_adj = *(x->mb_activity_ptr); +#else INT64 a; INT64 b; + INT64 act = *(x->mb_activity_ptr); - // Read activity from the map - act = (INT64)(*(x->mb_activity_ptr)); - - // Calculate a zbin adjustment for this mb + // Apply the masking to the RD multiplier. a = act + 4*cpi->activity_avg; b = 4*act + cpi->activity_avg; - if ( b > a ) - //x->act_zbin_adj = (char)((b * 8) / a) - 8; - x->act_zbin_adj = 8; - else - x->act_zbin_adj = 0; - - // Tmp force to 0 to disable. - x->act_zbin_adj = 0; + if ( act > cpi->activity_avg ) + x->act_zbin_adj = (int)(((INT64)b + (a>>1))/a) - 1; + else + x->act_zbin_adj = 1 - (int)(((INT64)a + (b>>1))/b); +#endif } int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t) @@ -1041,55 +1099,29 @@ int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t) Error4x4 = vp8_rd_pick_intra4x4mby_modes(cpi, x, &rate4x4, &rate4x4_tokenonly, &dist4x4, Error16x16); - rate += (Error4x4 < Error16x16) ? rate4x4 : rate16x16; + if (Error4x4 < Error16x16) + { + x->e_mbd.mode_info_context->mbmi.mode = B_PRED; + rate += rate4x4; + } + else + { + rate += rate16x16; + } if(cpi->oxcf.tuning == VP8_TUNE_SSIM) { - adjust_act_zbin( cpi, rate, x ); + adjust_act_zbin( cpi, x ); vp8_update_zbin_extra(cpi, x); } } else - { - int rate2, best_distortion; - MB_PREDICTION_MODE mode, best_mode = DC_PRED; - int this_rd; - Error16x16 = INT_MAX; - - vp8_pick_intra_mbuv_mode(x); - - for (mode = DC_PRED; mode <= TM_PRED; mode ++) - { - int distortion2; + vp8_pick_intra_mode(cpi, x, &rate); - x->e_mbd.mode_info_context->mbmi.mode = mode; - RECON_INVOKE(&cpi->common.rtcd.recon, build_intra_predictors_mby) - (&x->e_mbd); - distortion2 = VARIANCE_INVOKE(&cpi->rtcd.variance, get16x16prederror)(x->src.y_buffer, x->src.y_stride, x->e_mbd.predictor, 16); - rate2 = x->mbmode_cost[x->e_mbd.frame_type][mode]; - this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2); - - if (Error16x16 > this_rd) - { - Error16x16 = this_rd; - best_mode = mode; - best_distortion = distortion2; - } - } - x->e_mbd.mode_info_context->mbmi.mode = best_mode; - - Error4x4 = vp8_pick_intra4x4mby_modes(IF_RTCD(&cpi->rtcd), x, &rate2, &best_distortion); - } - - if (Error4x4 < Error16x16) - { - x->e_mbd.mode_info_context->mbmi.mode = B_PRED; + if (x->e_mbd.mode_info_context->mbmi.mode == B_PRED) vp8_encode_intra4x4mby(IF_RTCD(&cpi->rtcd), x); - } else - { vp8_encode_intra16x16mby(IF_RTCD(&cpi->rtcd), x); - } vp8_encode_intra16x16mbuv(IF_RTCD(&cpi->rtcd), x); sum_intra_stats(cpi, x); @@ -1163,7 +1195,7 @@ int vp8cx_encode_inter_macroblock if(cpi->oxcf.tuning == VP8_TUNE_SSIM) { // Adjust the zbin based on this MB rate. - adjust_act_zbin( cpi, rate, x ); + adjust_act_zbin( cpi, x ); } #if 0 @@ -1193,11 +1225,10 @@ int vp8cx_encode_inter_macroblock { // Experimental code. Special case for gf and arf zeromv modes. // Increase zbin size to supress noise + cpi->zbin_mode_boost = 0; if (cpi->zbin_mode_boost_enabled) { - if ( xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME ) - cpi->zbin_mode_boost = 0; - else + if ( xd->mode_info_context->mbmi.ref_frame != INTRA_FRAME ) { if (xd->mode_info_context->mbmi.mode == ZEROMV) { @@ -1212,9 +1243,6 @@ int vp8cx_encode_inter_macroblock cpi->zbin_mode_boost = MV_ZBIN_BOOST; } } - else - cpi->zbin_mode_boost = 0; - vp8_update_zbin_extra(cpi, x); } diff --git a/vp8/encoder/ethreading.c b/vp8/encoder/ethreading.c index 665b2d5dc..420ed8eff 100644 --- a/vp8/encoder/ethreading.c +++ b/vp8/encoder/ethreading.c @@ -114,8 +114,6 @@ THREAD_FUNCTION thread_encoding_proc(void *p_data) // Set the mb activity pointer to the start of the row. x->mb_activity_ptr = &cpi->mb_activity_map[map_index]; - x->mb_norm_activity_ptr = - &cpi->mb_norm_activity_map[map_index]; // for each macroblock col in image for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) @@ -230,7 +228,6 @@ THREAD_FUNCTION thread_encoding_proc(void *p_data) // Increment the activity mask pointers. x->mb_activity_ptr++; - x->mb_norm_activity_ptr++; /* save the block info */ for (i = 0; i < 16; i++) diff --git a/vp8/encoder/firstpass.c b/vp8/encoder/firstpass.c index 7b4869534..bc1863471 100644 --- a/vp8/encoder/firstpass.c +++ b/vp8/encoder/firstpass.c @@ -81,7 +81,7 @@ static const int cq_level[QINDEX_RANGE] = static void find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame); -static int encode_intra(VP8_COMP *cpi, MACROBLOCK *x, int use_dc_pred) +int encode_intra(VP8_COMP *cpi, MACROBLOCK *x, int use_dc_pred) { int i; @@ -243,33 +243,58 @@ static int frame_max_bits(VP8_COMP *cpi) int max_bits; // For CBR we need to also consider buffer fullness. - // If we are running below the optimal level then we need to gradually tighten up on max_bits. if (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) { - double buffer_fullness_ratio = (double)cpi->buffer_level / DOUBLE_DIVIDE_CHECK((double)cpi->oxcf.optimal_buffer_level); + max_bits = 2 * cpi->av_per_frame_bandwidth; + max_bits -= cpi->buffered_av_per_frame_bandwidth; + max_bits *= ((double)cpi->oxcf.two_pass_vbrmax_section / 100.0); + } + // VBR + else + { + // For VBR base this on the bits and frames left plus the two_pass_vbrmax_section rate passed in by the user + max_bits = (int)(((double)cpi->twopass.bits_left / (cpi->twopass.total_stats->count - (double)cpi->common.current_video_frame)) * ((double)cpi->oxcf.two_pass_vbrmax_section / 100.0)); + } + + // Trap case where we are out of bits + if (max_bits < 0) + max_bits = 0; - // For CBR base this on the target average bits per frame plus the maximum sedction rate passed in by the user - max_bits = (int)(cpi->av_per_frame_bandwidth * ((double)cpi->oxcf.two_pass_vbrmax_section / 100.0)); + return max_bits; +} - // If our buffer is below the optimum level - if (buffer_fullness_ratio < 1.0) - { - // The lower of max_bits / 4 or cpi->av_per_frame_bandwidth / 4. - int min_max_bits = ((cpi->av_per_frame_bandwidth >> 2) < (max_bits >> 2)) ? cpi->av_per_frame_bandwidth >> 2 : max_bits >> 2; - max_bits = (int)(max_bits * buffer_fullness_ratio); +static int gf_group_max_bits(VP8_COMP *cpi) +{ + // Max allocation for a golden frame group + int max_bits; - if (max_bits < min_max_bits) - max_bits = min_max_bits; // Lowest value we will set ... which should allow the buffer to refil. + // For CBR we need to also consider buffer fullness. + if (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) + { + max_bits = cpi->av_per_frame_bandwidth * cpi->baseline_gf_interval; + if (max_bits > cpi->oxcf.optimal_buffer_level) + { + max_bits -= cpi->oxcf.optimal_buffer_level; + max_bits += cpi->buffer_level; } + else + { + max_bits -= (cpi->buffered_av_per_frame_bandwidth + - cpi->av_per_frame_bandwidth) + * cpi->baseline_gf_interval; + } + + max_bits *= ((double)cpi->oxcf.two_pass_vbrmax_section / 100.0); } - // VBR else { // For VBR base this on the bits and frames left plus the two_pass_vbrmax_section rate passed in by the user max_bits = (int)(((double)cpi->twopass.bits_left / (cpi->twopass.total_stats->count - (double)cpi->common.current_video_frame)) * ((double)cpi->oxcf.two_pass_vbrmax_section / 100.0)); + max_bits *= cpi->baseline_gf_interval; } + // Trap case where we are out of bits if (max_bits < 0) max_bits = 0; @@ -1362,7 +1387,7 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) double abs_mv_in_out_accumulator = 0.0; double mod_err_per_mb_accumulator = 0.0; - int max_bits = frame_max_bits(cpi); // Max for a single frame + int max_group_bits; unsigned int allow_alt_ref = cpi->oxcf.play_alternate && cpi->oxcf.lag_in_frames; @@ -1715,8 +1740,9 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) cpi->twopass.gf_group_bits = (cpi->twopass.gf_group_bits < 0) ? 0 : (cpi->twopass.gf_group_bits > cpi->twopass.kf_group_bits) ? cpi->twopass.kf_group_bits : cpi->twopass.gf_group_bits; // Clip cpi->twopass.gf_group_bits based on user supplied data rate variability limit (cpi->oxcf.two_pass_vbrmax_section) - if (cpi->twopass.gf_group_bits > max_bits * cpi->baseline_gf_interval) - cpi->twopass.gf_group_bits = max_bits * cpi->baseline_gf_interval; + max_group_bits = gf_group_max_bits(cpi); + if (cpi->twopass.gf_group_bits > max_group_bits) + cpi->twopass.gf_group_bits = max_group_bits; // Reset the file position reset_fpf_position(cpi, start_pos); @@ -1810,13 +1836,6 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) } } - // Apply an additional limit for CBR - if (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) - { - if (cpi->twopass.gf_bits > (cpi->buffer_level >> 1)) - cpi->twopass.gf_bits = cpi->buffer_level >> 1; - } - // Dont allow a negative value for gf_bits if (cpi->twopass.gf_bits < 0) cpi->twopass.gf_bits = 0; diff --git a/vp8/encoder/generic/csystemdependent.c b/vp8/encoder/generic/csystemdependent.c index 9af3f183a..37885dadf 100644 --- a/vp8/encoder/generic/csystemdependent.c +++ b/vp8/encoder/generic/csystemdependent.c @@ -68,8 +68,6 @@ void vp8_cmachine_specific_config(VP8_COMP *cpi) cpi->rtcd.variance.getmbss = vp8_get_mb_ss_c; cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_c; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_c; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_c;; cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_c; cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_c; diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c index 0296d9290..bebc9417d 100644 --- a/vp8/encoder/onyx_if.c +++ b/vp8/encoder/onyx_if.c @@ -1477,6 +1477,7 @@ static void init_config(VP8_PTR ptr, VP8_CONFIG *oxcf) cpi->rolling_actual_bits = cpi->av_per_frame_bandwidth; cpi->long_rolling_target_bits = cpi->av_per_frame_bandwidth; cpi->long_rolling_actual_bits = cpi->av_per_frame_bandwidth; + cpi->buffered_av_per_frame_bandwidth = cpi->av_per_frame_bandwidth; cpi->total_actual_bits = 0; cpi->total_target_vs_actual = 0; @@ -1572,7 +1573,7 @@ void vp8_change_config(VP8_PTR ptr, VP8_CONFIG *oxcf) break; } - if (cpi->pass == 0) + if (cpi->pass == 0 && cpi->oxcf.end_usage != USAGE_STREAM_FROM_SERVER) cpi->auto_worst_q = 1; cpi->oxcf.worst_allowed_q = q_trans[oxcf->worst_allowed_q]; @@ -3453,7 +3454,8 @@ static void encode_frame_to_data_rate // For CBR if the buffer reaches its maximum level then we can no longer // save up bits for later frames so we might as well use them up // on the current frame. - if ((cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) && + if (cpi->pass == 2 + && (cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER) && (cpi->buffer_level >= cpi->oxcf.optimal_buffer_level) && cpi->buffered_mode) { int Adjustment = cpi->active_worst_quality / 4; // Max adjustment is 1/4 @@ -3544,6 +3546,9 @@ static void encode_frame_to_data_rate } else { + if(cpi->pass != 2) + Q = cpi->avg_frame_qindex; + cpi->active_best_quality = inter_minq[Q]; // For the constant/constrained quality mode we dont want @@ -3845,15 +3850,16 @@ static void encode_frame_to_data_rate (cpi->active_worst_quality < cpi->worst_quality) && (cpi->projected_frame_size > frame_over_shoot_limit)) { - int over_size_percent = ((cpi->projected_frame_size - frame_over_shoot_limit) * 100) / frame_over_shoot_limit; + /* step down active_worst_quality such that the corresponding + * active_best_quality will be equal to the current + * active_worst_quality + 1 + */ + int i; - // If so is there any scope for relaxing it - while ((cpi->active_worst_quality < cpi->worst_quality) && (over_size_percent > 0)) - { - cpi->active_worst_quality++; - top_index = cpi->active_worst_quality; - over_size_percent = (int)(over_size_percent * 0.96); // Assume 1 qstep = about 4% on frame size. - } + for(i=cpi->active_worst_quality; i<cpi->worst_quality; i++) + if(inter_minq[i] >= cpi->active_worst_quality + 1) + break; + cpi->active_worst_quality = i; // If we have updated the active max Q do not call vp8_update_rate_correction_factors() this loop. active_worst_qchanged = TRUE; @@ -4241,10 +4247,9 @@ static void encode_frame_to_data_rate // Update the buffer level variable. // Non-viewable frames are a special case and are treated as pure overhead. - if ( !cm->show_frame ) - cpi->bits_off_target -= cpi->projected_frame_size; - else - cpi->bits_off_target += cpi->av_per_frame_bandwidth - cpi->projected_frame_size; + if ( cm->show_frame ) + cpi->bits_off_target += cpi->av_per_frame_bandwidth; + cpi->bits_off_target -= cpi->projected_frame_size; // Rolling monitors of whether we are over or underspending used to help regulate min and Max Q in two pass. cpi->rolling_target_bits = ((cpi->rolling_target_bits * 3) + cpi->this_frame_target + 2) / 4; @@ -4258,7 +4263,33 @@ static void encode_frame_to_data_rate // Debug stats cpi->total_target_vs_actual += (cpi->this_frame_target - cpi->projected_frame_size); - cpi->buffer_level = cpi->bits_off_target; + // Update the buffered average bitrate + { + long long numerator; + + numerator = cpi->oxcf.maximum_buffer_size + - cpi->buffered_av_per_frame_bandwidth + + cpi->projected_frame_size; + numerator *= cpi->buffered_av_per_frame_bandwidth; + cpi->buffered_av_per_frame_bandwidth = numerator + / cpi->oxcf.maximum_buffer_size; + } + + { + long long tmp = (long long)cpi->buffered_av_per_frame_bandwidth + * cpi->oxcf.maximum_buffer_size + / cpi->av_per_frame_bandwidth; + cpi->buffer_level = cpi->oxcf.maximum_buffer_size + - tmp + + cpi->oxcf.optimal_buffer_level; + } + + // Accumulate overshoot error. + cpi->accumulated_overshoot += + (cpi->projected_frame_size > cpi->av_per_frame_bandwidth) + ? cpi->projected_frame_size - cpi->av_per_frame_bandwidth + : 0; + // Update bits left to the kf and gf groups to account for overshoot or undershoot on these frames if (cm->frame_type == KEY_FRAME) diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h index 663786004..c460b9da9 100644 --- a/vp8/encoder/onyx_int.h +++ b/vp8/encoder/onyx_int.h @@ -47,8 +47,8 @@ #define MIN_THRESHMULT 32 #define MAX_THRESHMULT 512 -#define GF_ZEROMV_ZBIN_BOOST 24 -#define LF_ZEROMV_ZBIN_BOOST 12 +#define GF_ZEROMV_ZBIN_BOOST 12 +#define LF_ZEROMV_ZBIN_BOOST 6 #define MV_ZBIN_BOOST 4 #define ZBIN_OQ_MAX 192 @@ -351,6 +351,10 @@ typedef struct VP8_COMP int per_frame_bandwidth; // Current section per frame bandwidth target int av_per_frame_bandwidth; // Average frame size target for clip int min_frame_bandwidth; // Minimum allocation that should be used for any frame + int buffered_av_per_frame_bandwidth; // Average bitrate over the last buffer + int buffered_av_per_frame_bandwidth_rem; // Average bitrate remainder + int accumulated_overshoot; // Accumulated # of bits spent > target + int inter_frame_target; double output_frame_rate; long long last_time_stamp_seen; diff --git a/vp8/encoder/pickinter.c b/vp8/encoder/pickinter.c index 456059cf8..661ecb79a 100644 --- a/vp8/encoder/pickinter.c +++ b/vp8/encoder/pickinter.c @@ -221,7 +221,7 @@ static int pick_intra4x4block( } -int vp8_pick_intra4x4mby_modes +static int pick_intra4x4mby_modes ( const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *mb, @@ -275,7 +275,7 @@ int vp8_pick_intra4x4mby_modes return error; } -void vp8_pick_intra_mbuv_mode(MACROBLOCK *mb) +static void pick_intra_mbuv_mode(MACROBLOCK *mb) { MACROBLOCKD *x = &mb->e_mbd; @@ -659,10 +659,9 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, switch (this_mode) { case B_PRED: - // Pass best so far to vp8_pick_intra4x4mby_modes to use as breakout + // Pass best so far to pick_intra4x4mby_modes to use as breakout distortion2 = *returndistortion; - vp8_pick_intra4x4mby_modes(IF_RTCD(&cpi->rtcd), x, - &rate, &distortion2); + pick_intra4x4mby_modes(IF_RTCD(&cpi->rtcd), x, &rate, &distortion2); if (distortion2 == INT_MAX) { @@ -956,7 +955,7 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, if (best_mbmode.mode <= B_PRED) { /* set mode_info_context->mbmi.uv_mode */ - vp8_pick_intra_mbuv_mode(x); + pick_intra_mbuv_mode(x); } if (x->e_mbd.mode_info_context->mbmi.mode == B_PRED) @@ -968,3 +967,40 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, } update_mvcount(cpi, &x->e_mbd, &frame_best_ref_mv[xd->mode_info_context->mbmi.ref_frame]); } + + +void vp8_pick_intra_mode(VP8_COMP *cpi, MACROBLOCK *x, int *rate_) +{ + int error4x4, error16x16 = INT_MAX; + int rate, distortion, best_distortion; + MB_PREDICTION_MODE mode, best_mode = DC_PRED; + int this_rd; + + pick_intra_mbuv_mode(x); + + for (mode = DC_PRED; mode <= TM_PRED; mode ++) + { + x->e_mbd.mode_info_context->mbmi.mode = mode; + RECON_INVOKE(&cpi->common.rtcd.recon, build_intra_predictors_mby) + (&x->e_mbd); + distortion = VARIANCE_INVOKE(&cpi->rtcd.variance, get16x16prederror) + (x->src.y_buffer, x->src.y_stride, x->e_mbd.predictor, 16); + rate = x->mbmode_cost[x->e_mbd.frame_type][mode]; + this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion); + + if (error16x16 > this_rd) + { + error16x16 = this_rd; + best_mode = mode; + best_distortion = distortion; + } + } + x->e_mbd.mode_info_context->mbmi.mode = best_mode; + + error4x4 = pick_intra4x4mby_modes(IF_RTCD(&cpi->rtcd), x, &rate, + &best_distortion); + if (error4x4 < error16x16) + x->e_mbd.mode_info_context->mbmi.mode = B_PRED; + + *rate_ = rate; +} diff --git a/vp8/encoder/pickinter.h b/vp8/encoder/pickinter.h index f96fc5376..a0103d165 100644 --- a/vp8/encoder/pickinter.h +++ b/vp8/encoder/pickinter.h @@ -14,7 +14,6 @@ #include "vpx_ports/config.h" #include "vp8/common/onyxc_int.h" -extern int vp8_pick_intra4x4mby_modes(const VP8_ENCODER_RTCD *, MACROBLOCK *mb, int *Rate, int *Distortion); -extern void vp8_pick_intra_mbuv_mode(MACROBLOCK *mb); extern void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int recon_uvoffset, int *returnrate, int *returndistortion, int *returnintra); +extern void vp8_pick_intra_mode(VP8_COMP *cpi, MACROBLOCK *x, int *rate); #endif diff --git a/vp8/encoder/ppc/csystemdependent.c b/vp8/encoder/ppc/csystemdependent.c index 8dfd2a543..0dd097f84 100644 --- a/vp8/encoder/ppc/csystemdependent.c +++ b/vp8/encoder/ppc/csystemdependent.c @@ -49,8 +49,6 @@ void (*vp8_subtract_mbuv)(short *diff, unsigned char *usrc, unsigned char *vsrc, void (*vp8_fast_quantize_b)(BLOCK *b, BLOCKD *d); unsigned int (*vp8_get16x16pred_error)(unsigned char *src_ptr, int src_stride, unsigned char *ref_ptr, int ref_stride); -unsigned int (*vp8_get8x8var)(unsigned char *src_ptr, int source_stride, unsigned char *ref_ptr, int recon_stride, unsigned int *SSE, int *Sum); -unsigned int (*vp8_get16x16var)(unsigned char *src_ptr, int source_stride, unsigned char *ref_ptr, int recon_stride, unsigned int *SSE, int *Sum); unsigned int (*vp8_get4x4sse_cs)(unsigned char *src_ptr, int source_stride, unsigned char *ref_ptr, int recon_stride); // c imports @@ -89,8 +87,6 @@ extern sub_pixel_variance_function sub_pixel_variance16x16_c; extern unsigned int vp8_get_mb_ss_c(short *); extern unsigned int vp8_get16x16pred_error_c(unsigned char *src_ptr, int src_stride, unsigned char *ref_ptr, int ref_stride); -extern unsigned int vp8_get8x8var_c(unsigned char *src_ptr, int source_stride, unsigned char *ref_ptr, int recon_stride, unsigned int *SSE, int *Sum); -extern unsigned int vp8_get16x16var_c(unsigned char *src_ptr, int source_stride, unsigned char *ref_ptr, int recon_stride, unsigned int *SSE, int *Sum); extern unsigned int vp8_get4x4sse_cs_c(unsigned char *src_ptr, int source_stride, unsigned char *ref_ptr, int recon_stride); // ppc @@ -150,8 +146,6 @@ void vp8_cmachine_specific_config(void) vp8_get_mb_ss = vp8_get_mb_ss_c; vp8_get16x16pred_error = vp8_get16x16pred_error_c; - vp8_get8x8var = vp8_get8x8var_ppc; - vp8_get16x16var = vp8_get16x16var_ppc; vp8_get4x4sse_cs = vp8_get4x4sse_cs_c; vp8_sad16x16 = vp8_sad16x16_ppc; diff --git a/vp8/encoder/ratectrl.c b/vp8/encoder/ratectrl.c index 54c394dfc..73e1437b5 100644 --- a/vp8/encoder/ratectrl.c +++ b/vp8/encoder/ratectrl.c @@ -605,10 +605,10 @@ static void calc_gf_params(VP8_COMP *cpi) static void calc_pframe_target_size(VP8_COMP *cpi) { - int min_frame_target; + int min_frame_target, max_frame_target; int Adjustment; - min_frame_target = 0; + min_frame_target = 1; if (cpi->pass == 2) { @@ -616,10 +616,19 @@ static void calc_pframe_target_size(VP8_COMP *cpi) if (min_frame_target < (cpi->av_per_frame_bandwidth >> 5)) min_frame_target = cpi->av_per_frame_bandwidth >> 5; + + max_frame_target = INT_MAX; } - else if (min_frame_target < cpi->per_frame_bandwidth / 4) - min_frame_target = cpi->per_frame_bandwidth / 4; + else + { + if (min_frame_target < cpi->per_frame_bandwidth / 4) + min_frame_target = cpi->per_frame_bandwidth / 4; + /* Don't allow the target to completely deplete the buffer. */ + max_frame_target = cpi->buffer_level + cpi->av_per_frame_bandwidth; + if(max_frame_target < min_frame_target) + max_frame_target = min_frame_target; + } // Special alt reference frame case if (cpi->common.refresh_alt_ref_frame) @@ -1112,6 +1121,32 @@ static void calc_pframe_target_size(VP8_COMP *cpi) } } + + if (cpi->pass==0 && cpi->oxcf.end_usage == USAGE_STREAM_FROM_SERVER){ + /* determine the accumulated error to apply to this frame. Apply + * more of the error when we've been undershooting, less when + * we've been overshooting + */ + long long adjust; + int bitrate_error; + + bitrate_error = cpi->av_per_frame_bandwidth + - cpi->buffered_av_per_frame_bandwidth; + + adjust = cpi->accumulated_overshoot; + adjust *= cpi->av_per_frame_bandwidth + bitrate_error; + adjust /= cpi->oxcf.maximum_buffer_size; + if (adjust > (cpi->this_frame_target - min_frame_target)) + adjust = (cpi->this_frame_target - min_frame_target); + else if (adjust < 0) + adjust = 0; + + cpi->this_frame_target -= adjust; + cpi->accumulated_overshoot -= adjust; + } + + if(cpi->this_frame_target > max_frame_target) + cpi->this_frame_target = max_frame_target; } diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c index d4d6cd7c7..ebfc438f7 100644 --- a/vp8/encoder/rdopt.c +++ b/vp8/encoder/rdopt.c @@ -155,51 +155,50 @@ static int rd_iifactor [ 32 ] = { 4, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }; - /* values are now correlated to quantizer */ static int sad_per_bit16lut[QINDEX_RANGE] = { + 2, 2, 2, 2, 2, 2, 2, 2, + 2, 2, 2, 2, 2, 2, 2, 2, + 3, 3, 3, 3, 3, 3, 3, 3, + 3, 3, 3, 3, 3, 3, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, - 6, 6, 6, 6, 6, 7, 7, 7, - 7, 7, 7, 7, 8, 8, 8, 8, - 8, 8, 8, 8, 8, 8, 9, 9, - 9, 9, 9, 9, 10, 10, 10, 10, - 10, 10, 11, 11, 11, 11, 11, 11, - 12, 12, 12, 12, 12, 12, 12, 13, - 13, 13, 13, 13, 13, 14, 14, 14, - 14, 14, 15, 15, 15, 15, 15, 15, - 16, 16, 16, 16, 16, 16, 17, 17, - 17, 17, 17, 17, 17, 18, 18, 18, - 18, 18, 19, 19, 19, 19, 19, 19, - 20, 20, 20, 21, 21, 21, 21, 22, - 22, 22, 23, 23, 23, 24, 24, 24, - 25, 25, 26, 26, 27, 27, 27, 28, - 28, 28, 29, 29, 30, 30, 31, 31 + 6, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 10, 10, + 10, 10, 10, 10, 10, 10, 11, 11, + 11, 11, 11, 11, 12, 12, 12, 12, + 12, 12, 13, 13, 13, 13, 14, 14 }; static int sad_per_bit4lut[QINDEX_RANGE] = { - 5, 5, 5, 5, 5, 5, 7, 7, + 2, 2, 2, 2, 2, 2, 3, 3, + 3, 3, 3, 3, 3, 3, 3, 3, + 3, 3, 3, 3, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 5, 5, + 5, 5, 5, 5, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, - 8, 8, 8, 8, 10, 10, 10, 10, - 10, 10, 10, 10, 10, 10, 11, 11, - 11, 11, 11, 11, 13, 13, 13, 13, - 13, 13, 14, 14, 14, 14, 14, 14, - 16, 16, 16, 16, 16, 16, 16, 17, - 17, 17, 17, 17, 17, 19, 19, 19, - 19, 19, 20, 20, 20, 20, 20, 20, - 22, 22, 22, 22, 22, 22, 23, 23, - 23, 23, 23, 23, 23, 25, 25, 25, - 25, 25, 26, 26, 26, 26, 26, 26, - 28, 28, 28, 29, 29, 29, 29, 31, - 31, 31, 32, 32, 32, 34, 34, 34, - 35, 35, 37, 37, 38, 38, 38, 40, - 40, 40, 41, 41, 43, 43, 44, 44, + 8, 8, 9, 9, 9, 9, 9, 9, + 10, 10, 10, 10, 10, 10, 10, 10, + 11, 11, 11, 11, 11, 11, 11, 11, + 12, 12, 12, 12, 12, 12, 12, 12, + 13, 13, 13, 13, 13, 13, 13, 14, + 14, 14, 14, 14, 15, 15, 15, 15, + 16, 16, 16, 16, 17, 17, 17, 18, + 18, 18, 19, 19, 19, 20, 20, 20, }; void vp8cx_initialize_me_consts(VP8_COMP *cpi, int QIndex) { - cpi->mb.sadperbit16 = sad_per_bit16lut[QIndex]/2; - cpi->mb.sadperbit4 = sad_per_bit4lut[QIndex]/2; + cpi->mb.sadperbit16 = sad_per_bit16lut[QIndex]; + cpi->mb.sadperbit4 = sad_per_bit4lut[QIndex]; } void vp8_initialize_rd_consts(VP8_COMP *cpi, int Qvalue) @@ -2182,29 +2181,28 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int } else if (x->encode_breakout) { - int sum; unsigned int sse; + unsigned int var; int threshold = (xd->block[0].dequant[1] * xd->block[0].dequant[1] >>4); if(threshold < x->encode_breakout) threshold = x->encode_breakout; - VARIANCE_INVOKE(&cpi->rtcd.variance, get16x16var) - (x->src.y_buffer, x->src.y_stride, - x->e_mbd.predictor, 16, &sse, &sum); + var = VARIANCE_INVOKE(&cpi->rtcd.variance, var16x16) + (x->src.y_buffer, x->src.y_stride, + x->e_mbd.predictor, 16, &sse); if (sse < threshold) { - // Check u and v to make sure skip is ok - int sse2 = 0; + unsigned int q2dc = xd->block[24].dequant[0]; /* If theres is no codeable 2nd order dc or a very small uniform pixel change change */ - if (abs(sum) < (xd->block[24].dequant[0]<<2)|| - ((sum * sum>>8) > sse && abs(sum) <128)) + if ((sse - var < q2dc * q2dc >>4) || + (sse /2 > var && sse-var < 64)) { - sse2 = VP8_UVSSE(x, IF_RTCD(&cpi->rtcd.variance)); - + // Check u and v to make sure skip is ok + int sse2= VP8_UVSSE(x, IF_RTCD(&cpi->rtcd.variance)); if (sse2 * 2 < threshold) { x->skip = 1; diff --git a/vp8/encoder/variance.h b/vp8/encoder/variance.h index 0d7d977d7..d52aa1b1d 100644 --- a/vp8/encoder/variance.h +++ b/vp8/encoder/variance.h @@ -313,16 +313,6 @@ extern prototype_variance(vp8_variance_mse16x16); #endif extern prototype_get16x16prederror(vp8_variance_get16x16prederror); -#ifndef vp8_variance_get8x8var -#define vp8_variance_get8x8var vp8_get8x8var_c -#endif -extern prototype_variance2(vp8_variance_get8x8var); - -#ifndef vp8_variance_get16x16var -#define vp8_variance_get16x16var vp8_get16x16var_c -#endif -extern prototype_variance2(vp8_variance_get16x16var); - #ifndef vp8_variance_get4x4sse_cs #define vp8_variance_get4x4sse_cs vp8_get4x4sse_cs_c #endif @@ -377,8 +367,6 @@ typedef struct vp8_variance_fn_t mse16x16; vp8_get16x16prederror_fn_t get16x16prederror; - vp8_variance2_fn_t get8x8var; - vp8_variance2_fn_t get16x16var; vp8_get16x16prederror_fn_t get4x4sse_cs; vp8_sad_multi_fn_t sad16x16x3; diff --git a/vp8/encoder/variance_c.c b/vp8/encoder/variance_c.c index ede07c8db..c7b9c2209 100644 --- a/vp8/encoder/variance_c.c +++ b/vp8/encoder/variance_c.c @@ -61,40 +61,6 @@ static void variance( } } -unsigned int -vp8_get8x8var_c -( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *SSE, - int *Sum -) -{ - - variance(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8, SSE, Sum); - return (*SSE - (((*Sum) * (*Sum)) >> 6)); -} - -unsigned int -vp8_get16x16var_c -( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *SSE, - int *Sum -) -{ - - variance(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16, SSE, Sum); - return (*SSE - (((*Sum) * (*Sum)) >> 8)); - -} - - unsigned int vp8_variance16x16_c( const unsigned char *src_ptr, diff --git a/vp8/encoder/x86/variance_mmx.c b/vp8/encoder/x86/variance_mmx.c index 4a89868c2..1b05571f1 100644 --- a/vp8/encoder/x86/variance_mmx.c +++ b/vp8/encoder/x86/variance_mmx.c @@ -84,36 +84,6 @@ extern unsigned int vp8_get16x16pred_error_mmx int ref_stride ); -unsigned int vp8_get16x16var_mmx( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *SSE, - int *SUM -) -{ - unsigned int sse0, sse1, sse2, sse3, var; - int sum0, sum1, sum2, sum3, avg; - - - vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1); - vp8_get8x8var_mmx(src_ptr + 8 * source_stride, source_stride, ref_ptr + 8 * recon_stride, recon_stride, &sse2, &sum2) ; - vp8_get8x8var_mmx(src_ptr + 8 * source_stride + 8, source_stride, ref_ptr + 8 * recon_stride + 8, recon_stride, &sse3, &sum3); - - var = sse0 + sse1 + sse2 + sse3; - avg = sum0 + sum1 + sum2 + sum3; - - *SSE = var; - *SUM = avg; - return (var - ((avg * avg) >> 8)); - -} - - - - unsigned int vp8_variance4x4_mmx( const unsigned char *src_ptr, diff --git a/vp8/encoder/x86/variance_x86.h b/vp8/encoder/x86/variance_x86.h index 77e05e1e8..4a640d7aa 100644 --- a/vp8/encoder/x86/variance_x86.h +++ b/vp8/encoder/x86/variance_x86.h @@ -43,7 +43,6 @@ extern prototype_getmbss(vp8_get_mb_ss_mmx); extern prototype_variance(vp8_mse16x16_mmx); extern prototype_get16x16prederror(vp8_get16x16pred_error_mmx); extern prototype_variance2(vp8_get8x8var_mmx); -extern prototype_variance2(vp8_get16x16var_mmx); extern prototype_get16x16prederror(vp8_get4x4sse_cs_mmx); #if !CONFIG_RUNTIME_CPU_DETECT @@ -113,12 +112,6 @@ extern prototype_get16x16prederror(vp8_get4x4sse_cs_mmx); #undef vp8_variance_get16x16prederror #define vp8_variance_get16x16prederror vp8_get16x16pred_error_mmx -#undef vp8_variance_get8x8var -#define vp8_variance_get8x8var vp8_get8x8var_mmx - -#undef vp8_variance_get16x16var -#define vp8_variance_get16x16var vp8_get16x16var_mmx - #undef vp8_variance_get4x4sse_cs #define vp8_variance_get4x4sse_cs vp8_get4x4sse_cs_mmx @@ -219,12 +212,6 @@ extern prototype_variance2(vp8_get16x16var_sse2); #undef vp8_variance_get16x16prederror #define vp8_variance_get16x16prederror vp8_get16x16pred_error_sse2 -#undef vp8_variance_get8x8var -#define vp8_variance_get8x8var vp8_get8x8var_sse2 - -#undef vp8_variance_get16x16var -#define vp8_variance_get16x16var vp8_get16x16var_sse2 - #endif #endif diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c index 378b14066..f33c74a1c 100644 --- a/vp8/encoder/x86/x86_csystemdependent.c +++ b/vp8/encoder/x86/x86_csystemdependent.c @@ -176,8 +176,6 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi) cpi->rtcd.variance.getmbss = vp8_get_mb_ss_mmx; cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_mmx; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_mmx; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_mmx; cpi->rtcd.variance.get4x4sse_cs = vp8_get4x4sse_cs_mmx; cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_mmx; @@ -227,9 +225,6 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi) cpi->rtcd.variance.getmbss = vp8_get_mb_ss_sse2; cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_sse2; - cpi->rtcd.variance.get8x8var = vp8_get8x8var_sse2; - cpi->rtcd.variance.get16x16var = vp8_get16x16var_sse2; - /* cpi->rtcd.variance.get4x4sse_cs not implemented for wmt */; |