diff options
Diffstat (limited to 'vp8/encoder')
-rw-r--r-- | vp8/encoder/firstpass.c | 188 | ||||
-rw-r--r-- | vp8/encoder/onyx_int.h | 1 | ||||
-rw-r--r-- | vp8/encoder/rdopt.c | 2 | ||||
-rw-r--r-- | vp8/encoder/x86/variance_impl_sse2.asm | 33 | ||||
-rw-r--r-- | vp8/encoder/x86/variance_impl_ssse3.asm | 348 | ||||
-rw-r--r-- | vp8/encoder/x86/variance_ssse3.c | 140 | ||||
-rw-r--r-- | vp8/encoder/x86/variance_x86.h | 4 | ||||
-rw-r--r-- | vp8/encoder/x86/x86_csystemdependent.c | 2 |
8 files changed, 642 insertions, 76 deletions
diff --git a/vp8/encoder/firstpass.c b/vp8/encoder/firstpass.c index d489413f6..804b80bd5 100644 --- a/vp8/encoder/firstpass.c +++ b/vp8/encoder/firstpass.c @@ -312,7 +312,9 @@ void vp8_output_stats(const VP8_COMP *cpi, FILE *fpfile; fpfile = fopen("firstpass.stt", "a"); - fprintf(fpfile, "%12.0f %12.0f %12.0f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.0f\n", + fprintf(fpfile, "%12.0f %12.0f %12.0f %12.4f %12.4f %12.4f %12.4f" + " %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.4f %12.0f" + " %12.4f\n", stats->frame, stats->intra_error, stats->coded_error, @@ -320,6 +322,7 @@ void vp8_output_stats(const VP8_COMP *cpi, stats->pcnt_inter, stats->pcnt_motion, stats->pcnt_second_ref, + stats->pcnt_neutral, stats->MVr, stats->mvr_abs, stats->MVc, @@ -327,7 +330,8 @@ void vp8_output_stats(const VP8_COMP *cpi, stats->MVrv, stats->MVcv, stats->mv_in_out_count, - stats->count); + stats->count, + stats->duration); fclose(fpfile); @@ -359,6 +363,7 @@ void vp8_zero_stats(FIRSTPASS_STATS *section) section->pcnt_inter = 0.0; section->pcnt_motion = 0.0; section->pcnt_second_ref = 0.0; + section->pcnt_neutral = 0.0; section->MVr = 0.0; section->mvr_abs = 0.0; section->MVc = 0.0; @@ -378,6 +383,7 @@ void vp8_accumulate_stats(FIRSTPASS_STATS *section, FIRSTPASS_STATS *frame) section->pcnt_inter += frame->pcnt_inter; section->pcnt_motion += frame->pcnt_motion; section->pcnt_second_ref += frame->pcnt_second_ref; + section->pcnt_neutral += frame->pcnt_neutral; section->MVr += frame->MVr; section->mvr_abs += frame->mvr_abs; section->MVc += frame->MVc; @@ -398,6 +404,7 @@ void vp8_avg_stats(FIRSTPASS_STATS *section) section->ssim_weighted_pred_err /= section->count; section->pcnt_inter /= section->count; section->pcnt_second_ref /= section->count; + section->pcnt_neutral /= section->count; section->pcnt_motion /= section->count; section->MVr /= section->count; section->mvr_abs /= section->count; @@ -570,6 +577,7 @@ void vp8_first_pass(VP8_COMP *cpi) int intercount = 0; int second_ref_count = 0; int intrapenalty = 256; + int neutral_count = 0; int sum_in_vectors = 0; @@ -726,6 +734,17 @@ void vp8_first_pass(VP8_COMP *cpi) if (motion_error <= this_error) { + // Keep a count of cases where the inter and intra were + // very close and very low. This helps with scene cut + // detection for example in cropped clips with black bars + // at the sides or top and bottom. + if( (((this_error-intrapenalty) * 9) <= + (motion_error*10)) && + (this_error < (2*intrapenalty)) ) + { + neutral_count++; + } + d->bmi.mv.as_mv.row <<= 3; d->bmi.mv.as_mv.col <<= 3; this_error = motion_error; @@ -854,6 +873,7 @@ void vp8_first_pass(VP8_COMP *cpi) fps.pcnt_inter = 1.0 * (double)intercount / cm->MBs; fps.pcnt_second_ref = 1.0 * (double)second_ref_count / cm->MBs; + fps.pcnt_neutral = 1.0 * (double)neutral_count / cm->MBs; if (mvcount > 0) { @@ -1341,7 +1361,7 @@ void vp8_end_second_pass(VP8_COMP *cpi) // This function gives and estimate of how badly we believe // the predicition quality is decaying from frame to frame. -double gf_prediction_decay_rate(VP8_COMP *cpi, FIRSTPASS_STATS *next_frame) +double get_prediction_decay_rate(VP8_COMP *cpi, FIRSTPASS_STATS *next_frame) { double prediction_decay_rate; double motion_decay; @@ -1376,6 +1396,52 @@ double gf_prediction_decay_rate(VP8_COMP *cpi, FIRSTPASS_STATS *next_frame) return prediction_decay_rate; } +// Function to test for a condition where a complex transition is followed +// by a static section. For example in slide shows where there is a fade +// between slides. This is to help with more optimal kf and gf positioning. +BOOL detect_transition_to_still( + VP8_COMP *cpi, + int frame_interval, + int still_interval, + double loop_decay_rate, + double decay_accumulator ) +{ + BOOL trans_to_still = FALSE; + + // Break clause to detect very still sections after motion + // For example a static image after a fade or other transition + // instead of a clean scene cut. + if ( (frame_interval > MIN_GF_INTERVAL) && + (loop_decay_rate >= 0.999) && + (decay_accumulator < 0.9) ) + { + int j; + FIRSTPASS_STATS * position = cpi->stats_in; + FIRSTPASS_STATS tmp_next_frame; + double decay_rate; + + // Look ahead a few frames to see if static condition + // persists... + for ( j = 0; j < still_interval; j++ ) + { + if (EOF == vp8_input_stats(cpi, &tmp_next_frame)) + break; + + decay_rate = get_prediction_decay_rate(cpi, &tmp_next_frame); + if ( decay_rate < 0.999 ) + break; + } + // Reset file position + reset_fpf_position(cpi, position); + + // Only if it does do we signal a transition to still + if ( j == still_interval ) + trans_to_still = TRUE; + } + + return trans_to_still; +} + // Analyse and define a gf/arf group . static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) { @@ -1528,7 +1594,7 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) if (r > GF_RMAX) r = GF_RMAX; - loop_decay_rate = gf_prediction_decay_rate(cpi, &next_frame); + loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame); // Cumulative effect of decay decay_accumulator = decay_accumulator * loop_decay_rate; @@ -1537,48 +1603,13 @@ static void define_gf_group(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) boost_score += (decay_accumulator * r); // Break clause to detect very still sections after motion - // For example a staic image after a fade or other transition - // instead of a clean key frame. - if ( (i > MIN_GF_INTERVAL) && - (loop_decay_rate >= 0.999) && - (decay_accumulator < 0.9) ) + // For example a staic image after a fade or other transition. + if ( detect_transition_to_still( cpi, i, 5, + loop_decay_rate, decay_accumulator ) ) { - int j; - FIRSTPASS_STATS * position = cpi->stats_in; - FIRSTPASS_STATS tmp_next_frame; - double decay_rate; - - // Look ahead a few frames to see if static condition - // persists... - for ( j = 0; j < 4; j++ ) - { - if (EOF == vp8_input_stats(cpi, &tmp_next_frame)) - break; - - decay_rate = gf_prediction_decay_rate(cpi, &tmp_next_frame); - if ( decay_rate < 0.999 ) - break; - } - reset_fpf_position(cpi, position); // Reset file position - - // Force GF not alt ref - if ( j == 4 ) - { - if (0) - { - FILE *f = fopen("fadegf.stt", "a"); - fprintf(f, " %8d %8d %10.4f %10.4f %10.4f\n", - cpi->common.current_video_frame+i, i, - loop_decay_rate, decay_accumulator, - boost_score ); - fclose(f); - } - - allow_alt_ref = FALSE; - - boost_score = old_boost_score; - break; - } + allow_alt_ref = FALSE; + boost_score = old_boost_score; + break; } // Break out conditions. @@ -2285,7 +2316,7 @@ static BOOL test_candidate_kf(VP8_COMP *cpi, FIRSTPASS_STATS *last_frame, FIRST (next_frame->pcnt_second_ref < 0.10) && ((this_frame->pcnt_inter < 0.05) || ( - (this_frame->pcnt_inter < .25) && + ((this_frame->pcnt_inter - this_frame->pcnt_neutral) < .25) && ((this_frame->intra_error / DOUBLE_DIVIDE_CHECK(this_frame->coded_error)) < 2.5) && ((fabs(last_frame->coded_error - this_frame->coded_error) / DOUBLE_DIVIDE_CHECK(this_frame->coded_error) > .40) || (fabs(last_frame->intra_error - this_frame->intra_error) / DOUBLE_DIVIDE_CHECK(this_frame->intra_error) > .40) || @@ -2332,7 +2363,9 @@ static BOOL test_candidate_kf(VP8_COMP *cpi, FIRSTPASS_STATS *last_frame, FIRST // Test various breakout clauses if ((local_next_frame.pcnt_inter < 0.05) || (next_iiratio < 1.5) || - ((local_next_frame.pcnt_inter < 0.20) && (next_iiratio < 3.0)) || + (((local_next_frame.pcnt_inter - + local_next_frame.pcnt_neutral) < 0.20) && + (next_iiratio < 3.0)) || ((boost_score - old_boost_score) < 0.5) || (local_next_frame.intra_error < 200) ) @@ -2363,13 +2396,13 @@ static BOOL test_candidate_kf(VP8_COMP *cpi, FIRSTPASS_STATS *last_frame, FIRST } void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) { - int i; + int i,j; FIRSTPASS_STATS last_frame; FIRSTPASS_STATS first_frame; FIRSTPASS_STATS next_frame; FIRSTPASS_STATS *start_position; - double decay_accumulator = 0; + double decay_accumulator = 1.0; double boost_score = 0; double old_boost_score = 0.0; double loop_decay_rate; @@ -2379,6 +2412,7 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) double kf_group_intra_err = 0.0; double kf_group_coded_err = 0.0; double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth * cpi->oxcf.two_pass_vbrmin_section / 100); + double recent_loop_decay[8] = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}; vpx_memset(&next_frame, 0, sizeof(next_frame)); // assure clean @@ -2407,6 +2441,7 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) kf_mod_err = calculate_modified_err(cpi, this_frame); // find the next keyframe + i = 0; while (cpi->stats_in < cpi->stats_in_end) { // Accumulate kf group error @@ -2425,9 +2460,34 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) if (cpi->oxcf.auto_key && lookup_next_frame_stats(cpi, &next_frame) != EOF) { + // Normal scene cut check if (test_candidate_kf(cpi, &last_frame, this_frame, &next_frame)) break; + // How fast is prediction quality decaying + loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame); + + // We want to know something about the recent past... rather than + // as used elsewhere where we are concened with decay in prediction + // quality since the last GF or KF. + recent_loop_decay[i%8] = loop_decay_rate; + decay_accumulator = 1.0; + for (j = 0; j < 8; j++) + { + decay_accumulator = decay_accumulator * recent_loop_decay[j]; + } + + // Special check for transition or high motion followed by a + // to a static scene. + if ( detect_transition_to_still( cpi, i, + (cpi->key_frame_frequency-i), + loop_decay_rate, + decay_accumulator ) ) + { + break; + } + + // Step on to the next frame cpi->frames_to_key ++; @@ -2437,6 +2497,8 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) break; } else cpi->frames_to_key ++; + + i++; } // If there is a max kf interval set by the user we must obey it. @@ -2588,32 +2650,8 @@ void vp8_find_next_key_frame(VP8_COMP *cpi, FIRSTPASS_STATS *this_frame) if (r > RMAX) r = RMAX; - // Adjust loop decay rate - //if ( next_frame.pcnt_inter < loop_decay_rate ) - loop_decay_rate = next_frame.pcnt_inter; - - // High % motion -> somewhat higher decay rate - motion_pct = next_frame.pcnt_motion; - motion_decay = (1.0 - (motion_pct / 20.0)); - if (motion_decay < loop_decay_rate) - loop_decay_rate = motion_decay; - - // Adjustment to decay rate based on speed of motion - { - double this_mv_rabs; - double this_mv_cabs; - double distance_factor; - - this_mv_rabs = fabs(next_frame.mvr_abs * motion_pct); - this_mv_cabs = fabs(next_frame.mvc_abs * motion_pct); - - distance_factor = sqrt((this_mv_rabs * this_mv_rabs) + - (this_mv_cabs * this_mv_cabs)) / 250.0; - distance_factor = ((distance_factor > 1.0) - ? 0.0 : (1.0 - distance_factor)); - if (distance_factor < loop_decay_rate) - loop_decay_rate = distance_factor; - } + // How fast is prediction quality decaying + loop_decay_rate = get_prediction_decay_rate(cpi, &next_frame); decay_accumulator = decay_accumulator * loop_decay_rate; decay_accumulator = decay_accumulator < 0.1 ? 0.1 : decay_accumulator; diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h index 5a4b3c185..0d353c31f 100644 --- a/vp8/encoder/onyx_int.h +++ b/vp8/encoder/onyx_int.h @@ -99,6 +99,7 @@ typedef struct double pcnt_inter; double pcnt_motion; double pcnt_second_ref; + double pcnt_neutral; double MVr; double mvr_abs; double MVc; diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c index 3db05155c..867ff6a9c 100644 --- a/vp8/encoder/rdopt.c +++ b/vp8/encoder/rdopt.c @@ -2033,7 +2033,7 @@ int vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int else cpi->zbin_mode_boost = LF_ZEROMV_ZBIN_BOOST; } - else if (vp8_ref_frame_order[mode_index] == SPLITMV) + else if (vp8_mode_order[mode_index] == SPLITMV) cpi->zbin_mode_boost = 0; else cpi->zbin_mode_boost = MV_ZBIN_BOOST; diff --git a/vp8/encoder/x86/variance_impl_sse2.asm b/vp8/encoder/x86/variance_impl_sse2.asm index 6cdc47bc9..5d1a17d44 100644 --- a/vp8/encoder/x86/variance_impl_sse2.asm +++ b/vp8/encoder/x86/variance_impl_sse2.asm @@ -627,6 +627,10 @@ filter_block2d_bil_var_sse2_loop: filter_block2d_bil_var_sse2_sp_only: movsxd rdx, dword ptr arg(6) ; yoffset + + cmp rdx, 0 ; skip all if both xoffset=0 and yoffset=0 + je filter_block2d_bil_var_sse2_full_pixel + shl rdx, 5 lea rdx, [rdx + rcx] ; VFilter @@ -671,6 +675,35 @@ filter_block2d_bil_sp_only_loop: jmp filter_block2d_bil_variance +filter_block2d_bil_var_sse2_full_pixel: + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + pxor xmm0, xmm0 ; + +filter_block2d_bil_full_pixel_loop: + movq xmm1, QWORD PTR [rsi] ; + punpcklbw xmm1, xmm0 ; + + movq xmm2, QWORD PTR [rdi] ; + punpcklbw xmm2, xmm0 ; + + psubw xmm1, xmm2 ; + paddw xmm6, xmm1 ; + + pmaddwd xmm1, xmm1 ; + paddd xmm7, xmm1 ; + + lea rsi, [rsi + rax] ;ref_pixels_per_line + lea rdi, [rdi + rbx] ;src_pixels_per_line + + sub rcx, 1 ; + jnz filter_block2d_bil_full_pixel_loop ; + + jmp filter_block2d_bil_variance + filter_block2d_bil_var_sse2_fp_only: mov rsi, arg(0) ;ref_ptr mov rdi, arg(2) ;src_ptr diff --git a/vp8/encoder/x86/variance_impl_ssse3.asm b/vp8/encoder/x86/variance_impl_ssse3.asm new file mode 100644 index 000000000..b1976328d --- /dev/null +++ b/vp8/encoder/x86/variance_impl_ssse3.asm @@ -0,0 +1,348 @@ +; +; Copyright (c) 2010 The WebM project authors. All Rights Reserved. +; +; Use of this source code is governed by a BSD-style license +; that can be found in the LICENSE file in the root of the source +; tree. An additional intellectual property rights grant can be found +; in the file PATENTS. All contributing project authors may +; be found in the AUTHORS file in the root of the source tree. +; + + +%include "vpx_ports/x86_abi_support.asm" + +%define xmm_filter_shift 7 + + +;void vp8_filter_block2d_bil_var_ssse3 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int xoffset, +; int yoffset, +; int *sum, +; unsigned int *sumsquared;; +; +;) +;Note: The filter coefficient at offset=0 is 128. Since the second register +;for Pmaddubsw is signed bytes, we must calculate zero offset seperately. +global sym(vp8_filter_block2d_bil_var_ssse3) +sym(vp8_filter_block2d_bil_var_ssse3): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 9 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + push rbx + ; end prolog + + pxor xmm6, xmm6 + pxor xmm7, xmm7 + + lea rcx, [GLOBAL(vp8_bilinear_filters_ssse3)] + movsxd rax, dword ptr arg(5) ; xoffset + + cmp rax, 0 ; skip first_pass filter if xoffset=0 + je filter_block2d_bil_var_ssse3_sp_only + + shl rax, 4 ; point to filter coeff with xoffset + lea rax, [rax + rcx] ; HFilter + + movsxd rdx, dword ptr arg(6) ; yoffset + + cmp rdx, 0 ; skip second_pass filter if yoffset=0 + je filter_block2d_bil_var_ssse3_fp_only + + shl rdx, 4 + lea rdx, [rdx + rcx] ; VFilter + + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + + movdqu xmm0, XMMWORD PTR [rsi] + movdqu xmm1, XMMWORD PTR [rsi+1] + movdqa xmm2, xmm0 + + punpcklbw xmm0, xmm1 + punpckhbw xmm2, xmm1 + pmaddubsw xmm0, [rax] + pmaddubsw xmm2, [rax] + + paddw xmm0, [GLOBAL(xmm_bi_rd)] + paddw xmm2, [GLOBAL(xmm_bi_rd)] + psraw xmm0, xmm_filter_shift + psraw xmm2, xmm_filter_shift + + packuswb xmm0, xmm2 + + movsxd rbx, dword ptr arg(1) ;ref_pixels_per_line + lea rsi, [rsi + rbx] +%if ABI_IS_32BIT=0 + movsxd r9, dword ptr arg(3) ;src_pixels_per_line +%endif + +filter_block2d_bil_var_ssse3_loop: + movdqu xmm1, XMMWORD PTR [rsi] + movdqu xmm2, XMMWORD PTR [rsi+1] + movdqa xmm3, xmm1 + + punpcklbw xmm1, xmm2 + punpckhbw xmm3, xmm2 + pmaddubsw xmm1, [rax] + pmaddubsw xmm3, [rax] + + paddw xmm1, [GLOBAL(xmm_bi_rd)] + paddw xmm3, [GLOBAL(xmm_bi_rd)] + psraw xmm1, xmm_filter_shift + psraw xmm3, xmm_filter_shift + packuswb xmm1, xmm3 + + movdqa xmm2, xmm0 + movdqa xmm0, xmm1 + movdqa xmm3, xmm2 + + punpcklbw xmm2, xmm1 + punpckhbw xmm3, xmm1 + pmaddubsw xmm2, [rdx] + pmaddubsw xmm3, [rdx] + + paddw xmm2, [GLOBAL(xmm_bi_rd)] + paddw xmm3, [GLOBAL(xmm_bi_rd)] + psraw xmm2, xmm_filter_shift + psraw xmm3, xmm_filter_shift + + movq xmm1, QWORD PTR [rdi] + pxor xmm4, xmm4 + punpcklbw xmm1, xmm4 + movq xmm5, QWORD PTR [rdi+8] + punpcklbw xmm5, xmm4 + + psubw xmm2, xmm1 + psubw xmm3, xmm5 + paddw xmm6, xmm2 + paddw xmm6, xmm3 + pmaddwd xmm2, xmm2 + pmaddwd xmm3, xmm3 + paddd xmm7, xmm2 + paddd xmm7, xmm3 + + lea rsi, [rsi + rbx] ;ref_pixels_per_line +%if ABI_IS_32BIT + add rdi, dword ptr arg(3) ;src_pixels_per_line +%else + lea rdi, [rdi + r9] +%endif + + sub rcx, 1 + jnz filter_block2d_bil_var_ssse3_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_var_ssse3_sp_only: + movsxd rdx, dword ptr arg(6) ; yoffset + + cmp rdx, 0 ; Both xoffset =0 and yoffset=0 + je filter_block2d_bil_var_ssse3_full_pixel + + shl rdx, 4 + lea rdx, [rdx + rcx] ; VFilter + + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + + movdqu xmm1, XMMWORD PTR [rsi] + movdqa xmm0, xmm1 + + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + lea rsi, [rsi + rax] + +filter_block2d_bil_sp_only_loop: + movdqu xmm3, XMMWORD PTR [rsi] + movdqa xmm2, xmm1 + movdqa xmm0, xmm3 + + punpcklbw xmm1, xmm3 + punpckhbw xmm2, xmm3 + pmaddubsw xmm1, [rdx] + pmaddubsw xmm2, [rdx] + + paddw xmm1, [GLOBAL(xmm_bi_rd)] + paddw xmm2, [GLOBAL(xmm_bi_rd)] + psraw xmm1, xmm_filter_shift + psraw xmm2, xmm_filter_shift + + movq xmm3, QWORD PTR [rdi] + pxor xmm4, xmm4 + punpcklbw xmm3, xmm4 + movq xmm5, QWORD PTR [rdi+8] + punpcklbw xmm5, xmm4 + + psubw xmm1, xmm3 + psubw xmm2, xmm5 + paddw xmm6, xmm1 + paddw xmm6, xmm2 + pmaddwd xmm1, xmm1 + pmaddwd xmm2, xmm2 + paddd xmm7, xmm1 + paddd xmm7, xmm2 + + movdqa xmm1, xmm0 + lea rsi, [rsi + rax] ;ref_pixels_per_line + lea rdi, [rdi + rbx] ;src_pixels_per_line + + sub rcx, 1 + jnz filter_block2d_bil_sp_only_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_var_ssse3_full_pixel: + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + pxor xmm0, xmm0 + +filter_block2d_bil_full_pixel_loop: + movq xmm1, QWORD PTR [rsi] + punpcklbw xmm1, xmm0 + movq xmm2, QWORD PTR [rsi+8] + punpcklbw xmm2, xmm0 + + movq xmm3, QWORD PTR [rdi] + punpcklbw xmm3, xmm0 + movq xmm4, QWORD PTR [rdi+8] + punpcklbw xmm4, xmm0 + + psubw xmm1, xmm3 + psubw xmm2, xmm4 + paddw xmm6, xmm1 + paddw xmm6, xmm2 + pmaddwd xmm1, xmm1 + pmaddwd xmm2, xmm2 + paddd xmm7, xmm1 + paddd xmm7, xmm2 + + lea rsi, [rsi + rax] ;ref_pixels_per_line + lea rdi, [rdi + rbx] ;src_pixels_per_line + sub rcx, 1 + jnz filter_block2d_bil_full_pixel_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_var_ssse3_fp_only: + mov rsi, arg(0) ;ref_ptr + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rdx, dword ptr arg(1) ;ref_pixels_per_line + + pxor xmm0, xmm0 + movsxd rbx, dword ptr arg(3) ;src_pixels_per_line + +filter_block2d_bil_fp_only_loop: + movdqu xmm1, XMMWORD PTR [rsi] + movdqu xmm2, XMMWORD PTR [rsi+1] + movdqa xmm3, xmm1 + + punpcklbw xmm1, xmm2 + punpckhbw xmm3, xmm2 + pmaddubsw xmm1, [rax] + pmaddubsw xmm3, [rax] + + paddw xmm1, [GLOBAL(xmm_bi_rd)] + paddw xmm3, [GLOBAL(xmm_bi_rd)] + psraw xmm1, xmm_filter_shift + psraw xmm3, xmm_filter_shift + + movq xmm2, XMMWORD PTR [rdi] + pxor xmm4, xmm4 + punpcklbw xmm2, xmm4 + movq xmm5, QWORD PTR [rdi+8] + punpcklbw xmm5, xmm4 + + psubw xmm1, xmm2 + psubw xmm3, xmm5 + paddw xmm6, xmm1 + paddw xmm6, xmm3 + pmaddwd xmm1, xmm1 + pmaddwd xmm3, xmm3 + paddd xmm7, xmm1 + paddd xmm7, xmm3 + + lea rsi, [rsi + rdx] + lea rdi, [rdi + rbx] ;src_pixels_per_line + + sub rcx, 1 + jnz filter_block2d_bil_fp_only_loop + + jmp filter_block2d_bil_variance + +filter_block2d_bil_variance: + pxor xmm0, xmm0 + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(7) ;[Sum] + mov rdi, arg(8) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rbx + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +SECTION_RODATA +align 16 +xmm_bi_rd: + times 8 dw 64 +align 16 +vp8_bilinear_filters_ssse3: + times 8 db 128, 0 + times 8 db 112, 16 + times 8 db 96, 32 + times 8 db 80, 48 + times 8 db 64, 64 + times 8 db 48, 80 + times 8 db 32, 96 + times 8 db 16, 112 diff --git a/vp8/encoder/x86/variance_ssse3.c b/vp8/encoder/x86/variance_ssse3.c new file mode 100644 index 000000000..750ae8b86 --- /dev/null +++ b/vp8/encoder/x86/variance_ssse3.c @@ -0,0 +1,140 @@ +/* + * Copyright (c) 2010 The WebM project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + + +#include "vp8/encoder/variance.h" +#include "vp8/common/pragmas.h" +#include "vpx_ports/mem.h" + +extern unsigned int vp8_get16x16var_sse2 +( + const unsigned char *src_ptr, + int source_stride, + const unsigned char *ref_ptr, + int recon_stride, + unsigned int *SSE, + int *Sum +); +extern void vp8_half_horiz_vert_variance16x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); +extern void vp8_half_horiz_variance16x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); +extern void vp8_half_vert_variance16x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); +extern void vp8_filter_block2d_bil_var_ssse3 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int xoffset, + int yoffset, + int *sum, + unsigned int *sumsquared +); + +unsigned int vp8_sub_pixel_variance16x16_ssse3 +( + const unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + const unsigned char *dst_ptr, + int dst_pixels_per_line, + unsigned int *sse +) +{ + int xsum0, xsum1; + unsigned int xxsum0, xxsum1; + + // note we could avoid these if statements if the calling function + // just called the appropriate functions inside. + if (xoffset == 4 && yoffset == 0) + { + vp8_half_horiz_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + &xsum0, &xxsum0); + + vp8_half_horiz_variance16x_h_sse2( + src_ptr + 8, src_pixels_per_line, + dst_ptr + 8, dst_pixels_per_line, 16, + &xsum1, &xxsum1); + + xsum0 += xsum1; + xxsum0 += xxsum1; + } + else if (xoffset == 0 && yoffset == 4) + { + vp8_half_vert_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + &xsum0, &xxsum0); + + vp8_half_vert_variance16x_h_sse2( + src_ptr + 8, src_pixels_per_line, + dst_ptr + 8, dst_pixels_per_line, 16, + &xsum1, &xxsum1); + + xsum0 += xsum1; + xxsum0 += xxsum1; + } + else if (xoffset == 4 && yoffset == 4) + { + vp8_half_horiz_vert_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + &xsum0, &xxsum0); + + vp8_half_horiz_vert_variance16x_h_sse2( + src_ptr + 8, src_pixels_per_line, + dst_ptr + 8, dst_pixels_per_line, 16, + &xsum1, &xxsum1); + + xsum0 += xsum1; + xxsum0 += xxsum1; + } + else + { + vp8_filter_block2d_bil_var_ssse3( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + xoffset, yoffset, + &xsum0, &xxsum0); + } + + *sse = xxsum0; + return (xxsum0 - ((xsum0 * xsum0) >> 8)); +} diff --git a/vp8/encoder/x86/variance_x86.h b/vp8/encoder/x86/variance_x86.h index 6bea15ebc..1e2fb3490 100644 --- a/vp8/encoder/x86/variance_x86.h +++ b/vp8/encoder/x86/variance_x86.h @@ -286,6 +286,7 @@ extern prototype_sad_multi_dif_address(vp8_sad4x4x4d_sse3); #if HAVE_SSSE3 extern prototype_sad_multi_same_address(vp8_sad16x16x3_ssse3); extern prototype_sad_multi_same_address(vp8_sad16x8x3_ssse3); +extern prototype_subpixvariance(vp8_sub_pixel_variance16x16_ssse3); #if !CONFIG_RUNTIME_CPU_DETECT #undef vp8_variance_sad16x16x3 @@ -294,6 +295,9 @@ extern prototype_sad_multi_same_address(vp8_sad16x8x3_ssse3); #undef vp8_variance_sad16x8x3 #define vp8_variance_sad16x8x3 vp8_sad16x8x3_ssse3 +#undef vp8_variance_subpixvar16x16 +#define vp8_variance_subpixvar16x16 vp8_sub_pixel_variance16x16_ssse3 + #endif #endif diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c index 61c603229..c7639a7e4 100644 --- a/vp8/encoder/x86/x86_csystemdependent.c +++ b/vp8/encoder/x86/x86_csystemdependent.c @@ -334,6 +334,8 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi) cpi->rtcd.variance.sad16x16x3 = vp8_sad16x16x3_ssse3; cpi->rtcd.variance.sad16x8x3 = vp8_sad16x8x3_ssse3; + cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_ssse3; + cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3; } |