Performance improvements: quant_bands: with: 681 decicycles in quant_bands, 8388453 runs, 155 skips without: 1190 decicycles in quant_bands, 8388386 runs, 222 skips Around 42% for the function Twoloop coder: abs_pow34: with/without: 7.82s/8.17s Around 4% for the entire encoder Both: with/without: 7.15s/8.17s Around 12% for the entire encoder Fast coder: abs_pow34: with/without: 3.40s/3.77s Around 10% for the entire encoder Both: with/without: 3.02s/3.77s Around 20% faster for the entire encoder
Showing the changes to libavcodec/aaccoder.c (1 of 13 files). Show all
Unified · Split
| @@ -88,7 +88,7 @@ static void encode_window_bands_info(AACEncContext *s, SingleChannelElement *sce · 87 unmodified lines | |||
| 88 | float next_minrd = INFINITY; | 88 | float next_minrd = INFINITY; |
| 89 | int next_mincb = 0; | 89 | int next_mincb = 0; |
| 90 | 90 | ||
| 91 | abs_pow34_v(s->scoefs, sce->coeffs, 1024); | 91 | s->abs_pow34(s->scoefs, sce->coeffs, 1024); |
| 92 | start = win*128; | 92 | start = win*128; |
| 93 | for (cb = 0; cb < CB_TOT_ALL; cb++) { | 93 | for (cb = 0; cb < CB_TOT_ALL; cb++) { |
| 94 | path[0][cb].cost = 0.0f; | 94 | path[0][cb].cost = 0.0f; |
| @@ -299,7 +299,7 @@ static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s, · 204 unmodified lines | |||
| 299 | } | 299 | } |
| 300 | } | 300 | } |
| 301 | idx = 1; | 301 | idx = 1; |
| 302 | abs_pow34_v(s->scoefs, sce->coeffs, 1024); | 302 | s->abs_pow34(s->scoefs, sce->coeffs, 1024); |
| 303 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { | 303 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { |
| 304 | start = w*128; | 304 | start = w*128; |
| 305 | for (g = 0; g < sce->ics.num_swb; g++) { | 305 | for (g = 0; g < sce->ics.num_swb; g++) { |
| @@ -446,7 +446,7 @@ static void search_for_quantizers_fast(AVCodecContext *avctx, AACEncContext *s, · 140 unmodified lines | |||
| 446 | 446 | ||
| 447 | if (!allz) | 447 | if (!allz) |
| 448 | return; | 448 | return; |
| 449 | abs_pow34_v(s->scoefs, sce->coeffs, 1024); | 449 | s->abs_pow34(s->scoefs, sce->coeffs, 1024); |
| 450 | ff_quantize_band_cost_cache_init(s); | 450 | ff_quantize_band_cost_cache_init(s); |
| 451 | 451 | ||
| 452 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { | 452 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { |
| @@ -652,8 +652,8 @@ static void search_for_pns(AACEncContext *s, AVCodecContext *avctx, SingleChanne · 199 unmodified lines | |||
| 652 | s->fdsp->vector_fmul_scalar(PNS, PNS, scale, sce->ics.swb_sizes[g]); | 652 | s->fdsp->vector_fmul_scalar(PNS, PNS, scale, sce->ics.swb_sizes[g]); |
| 653 | pns_senergy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]); | 653 | pns_senergy = s->fdsp->scalarproduct_float(PNS, PNS, sce->ics.swb_sizes[g]); |
| 654 | pns_energy += pns_senergy; | 654 | pns_energy += pns_senergy; |
| 655 | abs_pow34_v(NOR34, &sce->coeffs[start_c], sce->ics.swb_sizes[g]); | 655 | s->abs_pow34(NOR34, &sce->coeffs[start_c], sce->ics.swb_sizes[g]); |
| 656 | abs_pow34_v(PNS34, PNS, sce->ics.swb_sizes[g]); | 656 | s->abs_pow34(PNS34, PNS, sce->ics.swb_sizes[g]); |
| 657 | dist1 += quantize_band_cost(s, &sce->coeffs[start_c], | 657 | dist1 += quantize_band_cost(s, &sce->coeffs[start_c], |
| 658 | NOR34, | 658 | NOR34, |
| 659 | sce->ics.swb_sizes[g], | 659 | sce->ics.swb_sizes[g], |
| @@ -757,8 +757,9 @@ static void search_for_ms(AACEncContext *s, ChannelElement *cpe) · 97 unmodified lines | |||
| 757 | { | 757 | { |
| 758 | int start = 0, i, w, w2, g, sid_sf_boost, prev_mid, prev_side; | 758 | int start = 0, i, w, w2, g, sid_sf_boost, prev_mid, prev_side; |
| 759 | uint8_t nextband0[128], nextband1[128]; | 759 | uint8_t nextband0[128], nextband1[128]; |
| 760 | float M[128], S[128]; | 760 | float *M = s->scoefs + 128*0, *S = s->scoefs + 128*1; |
| 761 | float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3; | 761 | float *L34 = s->scoefs + 128*2, *R34 = s->scoefs + 128*3; |
| 762 | float *M34 = s->scoefs + 128*4, *S34 = s->scoefs + 128*5; | ||
| 762 | const float lambda = s->lambda; | 763 | const float lambda = s->lambda; |
| 763 | const float mslambda = FFMIN(1.0f, lambda / 120.f); | 764 | const float mslambda = FFMIN(1.0f, lambda / 120.f); |
| 764 | SingleChannelElement *sce0 = &cpe->ch[0]; | 765 | SingleChannelElement *sce0 = &cpe->ch[0]; |
| @@ -789,8 +790,8 @@ static void search_for_ms(AACEncContext *s, ChannelElement *cpe) · 24 unmodified lines | |||
| 789 | S[i] = M[i] | 790 | S[i] = M[i] |
| 790 | - sce1->coeffs[start+(w+w2)*128+i]; | 791 | - sce1->coeffs[start+(w+w2)*128+i]; |
| 791 | } | 792 | } |
| 792 | abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]); | 793 | s->abs_pow34(M34, M, sce0->ics.swb_sizes[g]); |
| 793 | abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]); | 794 | s->abs_pow34(S34, S, sce0->ics.swb_sizes[g]); |
| 794 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++ ) { | 795 | for (i = 0; i < sce0->ics.swb_sizes[g]; i++ ) { |
| 795 | Mmax = FFMAX(Mmax, M34[i]); | 796 | Mmax = FFMAX(Mmax, M34[i]); |
| 796 | Smax = FFMAX(Smax, S34[i]); | 797 | Smax = FFMAX(Smax, S34[i]); |
| @@ -833,10 +834,10 @@ static void search_for_ms(AACEncContext *s, ChannelElement *cpe) · 36 unmodified lines | |||
| 833 | - sce1->coeffs[start+(w+w2)*128+i]; | 834 | - sce1->coeffs[start+(w+w2)*128+i]; |
| 834 | } | 835 | } |
| 835 | 836 | ||
| 836 | abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]); | 837 | s->abs_pow34(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]); |
| 837 | abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]); | 838 | s->abs_pow34(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]); |
| 838 | abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]); | 839 | s->abs_pow34(M34, M, sce0->ics.swb_sizes[g]); |
| 839 | abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]); | 840 | s->abs_pow34(S34, S, sce0->ics.swb_sizes[g]); |
| 840 | dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128], | 841 | dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128], |
| 841 | L34, | 842 | L34, |
| 842 | sce0->ics.swb_sizes[g], | 843 | sce0->ics.swb_sizes[g], |