Unified · Split
| @@ -67,7 +67,8 @@ OBJS-$(CONFIG_MPEGAUDIODSP) += mpegaudiodsp.o \ · 66 unmodified lines | ||
| 67 | 67 | OBJS-$(CONFIG_MPEGVIDEO) += mpegvideo.o mpegvideodsp.o \ |
| 68 | 68 | mpegvideo_motion.o mpegutils.o |
| 69 | 69 | OBJS-$(CONFIG_MPEGVIDEOENC) += mpegvideo_enc.o mpeg12data.o \ |
| 70 | − motion_est.o ratecontrol.o | |
| 70 | + motion_est.o ratecontrol.o \ | |
| 71 | + mpegvideoencdsp.o | |
| 71 | 72 | OBJS-$(CONFIG_QPELDSP) += qpeldsp.o |
| 72 | 73 | OBJS-$(CONFIG_RANGECODER) += rangecoder.o |
| 73 | 74 | RDFT-OBJS-$(CONFIG_HARDCODED_TABLES) += sin_tables.o |
| @@ -464,35 +464,6 @@ static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int · 463 unmodified lines | ||
| 464 | 464 | return score1 + FFABS(score2) * 8; |
| 465 | 465 | } |
| 466 | 466 | |
| 467 | −static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], | |
| 468 | − int16_t basis[64], int scale) | |
| 469 | −{ | |
| 470 | − int i; | |
| 471 | − unsigned int sum = 0; | |
| 472 | − | |
| 473 | − for (i = 0; i < 8 * 8; i++) { | |
| 474 | − int b = rem[i] + ((basis[i] * scale + | |
| 475 | − (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >> | |
| 476 | − (BASIS_SHIFT - RECON_SHIFT)); | |
| 477 | − int w = weight[i]; | |
| 478 | − b >>= RECON_SHIFT; | |
| 479 | − assert(-512 < b && b < 512); | |
| 480 | − | |
| 481 | − sum += (w * b) * (w * b) >> 4; | |
| 482 | − } | |
| 483 | − return sum >> 2; | |
| 484 | −} | |
| 485 | − | |
| 486 | −static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale) | |
| 487 | −{ | |
| 488 | − int i; | |
| 489 | − | |
| 490 | − for (i = 0; i < 8 * 8; i++) | |
| 491 | − rem[i] += (basis[i] * scale + | |
| 492 | − (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >> | |
| 493 | − (BASIS_SHIFT - RECON_SHIFT); | |
| 494 | −} | |
| 495 | − | |
| 496 | 467 | static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b, |
| 497 | 468 | int stride, int h) |
| 498 | 469 | { |
| @@ -1126,9 +1097,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx) · 627 unmodified lines | ||
| 1126 | 1097 | c->nsse[0] = nsse16_c; |
| 1127 | 1098 | c->nsse[1] = nsse8_c; |
| 1128 | 1099 | |
| 1129 | − c->try_8x8basis = try_8x8basis_c; | |
| 1130 | − c->add_8x8basis = add_8x8basis_c; | |
| 1131 | − | |
| 1132 | 1100 | c->shrink[0] = av_image_copy_plane; |
| 1133 | 1101 | c->shrink[1] = ff_shrink22; |
| 1134 | 1102 | c->shrink[2] = ff_shrink44; |
| @@ -87,12 +87,6 @@ typedef struct DSPContext { · 86 unmodified lines | ||
| 87 | 87 | void (*fdct)(int16_t *block /* align 16 */); |
| 88 | 88 | void (*fdct248)(int16_t *block /* align 16 */); |
| 89 | 89 | |
| 90 | − int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], | |
| 91 | − int16_t basis[64], int scale); | |
| 92 | − void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
| 93 | −#define BASIS_SHIFT 16 | |
| 94 | −#define RECON_SHIFT 6 | |
| 95 | − | |
| 96 | 90 | void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, |
| 97 | 91 | int w, int h, int sides); |
| 98 | 92 | #define EDGE_WIDTH 16 |
| @@ -37,6 +37,7 @@ · 36 unmodified lines | ||
| 37 | 37 | #include "hpeldsp.h" |
| 38 | 38 | #include "idctdsp.h" |
| 39 | 39 | #include "mpegvideodsp.h" |
| 40 | +#include "mpegvideoencdsp.h" | |
| 40 | 41 | #include "put_bits.h" |
| 41 | 42 | #include "ratecontrol.h" |
| 42 | 43 | #include "parser.h" |
| @@ -355,6 +356,7 @@ typedef struct MpegEncContext { · 312 unmodified lines | ||
| 355 | 356 | HpelDSPContext hdsp; |
| 356 | 357 | IDCTDSPContext idsp; |
| 357 | 358 | MpegVideoDSPContext mdsp; |
| 359 | + MpegvideoEncDSPContext mpvencdsp; | |
| 358 | 360 | QpelDSPContext qdsp; |
| 359 | 361 | VideoDSPContext vdsp; |
| 360 | 362 | H263DSPContext h263dsp; |
| @@ -701,6 +701,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx) · 700 unmodified lines | ||
| 701 | 701 | if (ARCH_X86) |
| 702 | 702 | ff_MPV_encode_init_x86(s); |
| 703 | 703 | |
| 704 | + ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx); | |
| 704 | 705 | ff_qpeldsp_init(&s->qdsp); |
| 705 | 706 | |
| 706 | 707 | s->avctx->coded_frame = s->current_picture.f; |
| @@ -3871,7 +3872,7 @@ STOP_TIMER("memset rem[]")} · 3164 unmodified lines | ||
| 3871 | 3872 | run_tab[rle_index++]=run; |
| 3872 | 3873 | run=0; |
| 3873 | 3874 | |
| 3874 | − s->dsp.add_8x8basis(rem, basis[j], coeff); | |
| 3875 | + s->mpvencdsp.add_8x8basis(rem, basis[j], coeff); | |
| 3875 | 3876 | }else{ |
| 3876 | 3877 | run++; |
| 3877 | 3878 | } |
| @@ -3885,7 +3886,7 @@ STOP_TIMER("init rem[]") · 7 unmodified lines | ||
| 3885 | 3886 | {START_TIMER |
| 3886 | 3887 | #endif |
| 3887 | 3888 | for(;;){ |
| 3888 | − int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0); | |
| 3889 | + int best_score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0], 0); | |
| 3889 | 3890 | int best_coeff=0; |
| 3890 | 3891 | int best_change=0; |
| 3891 | 3892 | int run2, best_unquant_change=0, analyze_gradient; |
| @@ -3929,7 +3930,8 @@ STOP_TIMER("dct")} · 37 unmodified lines | ||
| 3929 | 3930 | if(new_coeff >= 2048 || new_coeff < 0) |
| 3930 | 3931 | continue; |
| 3931 | 3932 | |
| 3932 | − score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff); | |
| 3933 | + score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0], | |
| 3934 | + new_coeff - old_coeff); | |
| 3933 | 3935 | if(score<best_score){ |
| 3934 | 3936 | best_score= score; |
| 3935 | 3937 | best_coeff= 0; |
| @@ -4052,7 +4054,8 @@ STOP_TIMER("dct")} · 116 unmodified lines | ||
| 4052 | 4054 | unquant_change= new_coeff - old_coeff; |
| 4053 | 4055 | assert((score < 100*lambda && score > -100*lambda) || lambda==0); |
| 4054 | 4056 | |
| 4055 | − score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change); | |
| 4057 | + score += s->mpvencdsp.try_8x8basis(rem, weight, basis[j], | |
| 4058 | + unquant_change); | |
| 4056 | 4059 | if(score<best_score){ |
| 4057 | 4060 | best_score= score; |
| 4058 | 4061 | best_coeff= i; |
| @@ -4126,7 +4129,7 @@ if(256*256*256*64 % count == 0){ · 67 unmodified lines | ||
| 4126 | 4129 | } |
| 4127 | 4130 | } |
| 4128 | 4131 | |
| 4129 | − s->dsp.add_8x8basis(rem, basis[j], best_unquant_change); | |
| 4132 | + s->mpvencdsp.add_8x8basis(rem, basis[j], best_unquant_change); | |
| 4130 | 4133 | }else{ |
| 4131 | 4134 | break; |
| 4132 | 4135 | } |
| @@ -0,0 +1,64 @@ | ||
| 1 | +/* | |
| 2 | + * This file is part of Libav. | |
| 3 | + * | |
| 4 | + * Libav is free software; you can redistribute it and/or | |
| 5 | + * modify it under the terms of the GNU Lesser General Public | |
| 6 | + * License as published by the Free Software Foundation; either | |
| 7 | + * version 2.1 of the License, or (at your option) any later version. | |
| 8 | + * | |
| 9 | + * Libav is distributed in the hope that it will be useful, | |
| 10 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 11 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 12 | + * Lesser General Public License for more details. | |
| 13 | + * | |
| 14 | + * You should have received a copy of the GNU Lesser General Public | |
| 15 | + * License along with Libav; if not, write to the Free Software | |
| 16 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 17 | + */ | |
| 18 | + | |
| 19 | +#include <assert.h> | |
| 20 | +#include <stdint.h> | |
| 21 | + | |
| 22 | +#include "config.h" | |
| 23 | +#include "libavutil/attributes.h" | |
| 24 | +#include "avcodec.h" | |
| 25 | +#include "mpegvideoencdsp.h" | |
| 26 | + | |
| 27 | +static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], | |
| 28 | + int16_t basis[64], int scale) | |
| 29 | +{ | |
| 30 | + int i; | |
| 31 | + unsigned int sum = 0; | |
| 32 | + | |
| 33 | + for (i = 0; i < 8 * 8; i++) { | |
| 34 | + int b = rem[i] + ((basis[i] * scale + | |
| 35 | + (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >> | |
| 36 | + (BASIS_SHIFT - RECON_SHIFT)); | |
| 37 | + int w = weight[i]; | |
| 38 | + b >>= RECON_SHIFT; | |
| 39 | + assert(-512 < b && b < 512); | |
| 40 | + | |
| 41 | + sum += (w * b) * (w * b) >> 4; | |
| 42 | + } | |
| 43 | + return sum >> 2; | |
| 44 | +} | |
| 45 | + | |
| 46 | +static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale) | |
| 47 | +{ | |
| 48 | + int i; | |
| 49 | + | |
| 50 | + for (i = 0; i < 8 * 8; i++) | |
| 51 | + rem[i] += (basis[i] * scale + | |
| 52 | + (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >> | |
| 53 | + (BASIS_SHIFT - RECON_SHIFT); | |
| 54 | +} | |
| 55 | + | |
| 56 | +av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c, | |
| 57 | + AVCodecContext *avctx) | |
| 58 | +{ | |
| 59 | + c->try_8x8basis = try_8x8basis_c; | |
| 60 | + c->add_8x8basis = add_8x8basis_c; | |
| 61 | + | |
| 62 | + if (ARCH_X86) | |
| 63 | + ff_mpegvideoencdsp_init_x86(c, avctx); | |
| 64 | +} | |
| @@ -0,0 +1,41 @@ | ||
| 1 | +/* | |
| 2 | + * This file is part of Libav. | |
| 3 | + * | |
| 4 | + * Libav is free software; you can redistribute it and/or | |
| 5 | + * modify it under the terms of the GNU Lesser General Public | |
| 6 | + * License as published by the Free Software Foundation; either | |
| 7 | + * version 2.1 of the License, or (at your option) any later version. | |
| 8 | + * | |
| 9 | + * Libav is distributed in the hope that it will be useful, | |
| 10 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 11 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 12 | + * Lesser General Public License for more details. | |
| 13 | + * | |
| 14 | + * You should have received a copy of the GNU Lesser General Public | |
| 15 | + * License along with Libav; if not, write to the Free Software | |
| 16 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 17 | + */ | |
| 18 | + | |
| 19 | +#ifndef AVCODEC_MPEGVIDEOENCDSP_H | |
| 20 | +#define AVCODEC_MPEGVIDEOENCDSP_H | |
| 21 | + | |
| 22 | +#include <stdint.h> | |
| 23 | + | |
| 24 | +#include "avcodec.h" | |
| 25 | + | |
| 26 | +#define BASIS_SHIFT 16 | |
| 27 | +#define RECON_SHIFT 6 | |
| 28 | + | |
| 29 | +typedef struct MpegvideoEncDSPContext { | |
| 30 | + int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], | |
| 31 | + int16_t basis[64], int scale); | |
| 32 | + void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); | |
| 33 | + | |
| 34 | +} MpegvideoEncDSPContext; | |
| 35 | + | |
| 36 | +void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c, | |
| 37 | + AVCodecContext *avctx); | |
| 38 | +void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, | |
| 39 | + AVCodecContext *avctx); | |
| 40 | + | |
| 41 | +#endif /* AVCODEC_MPEGVIDEOENCDSP_H */ | |
| @@ -23,7 +23,8 @@ OBJS-$(CONFIG_LPC) += x86/lpc.o · 22 unmodified lines | ||
| 23 | 23 | OBJS-$(CONFIG_MPEGAUDIODSP) += x86/mpegaudiodsp.o |
| 24 | 24 | OBJS-$(CONFIG_MPEGVIDEO) += x86/mpegvideo.o \ |
| 25 | 25 | x86/mpegvideodsp.o |
| 26 | −OBJS-$(CONFIG_MPEGVIDEOENC) += x86/mpegvideoenc.o | |
| 26 | +OBJS-$(CONFIG_MPEGVIDEOENC) += x86/mpegvideoenc.o \ | |
| 27 | + x86/mpegvideoencdsp_init.o | |
| 27 | 28 | OBJS-$(CONFIG_QPELDSP) += x86/qpeldsp_init.o |
| 28 | 29 | OBJS-$(CONFIG_VIDEODSP) += x86/videodsp_init.o |
| 29 | 30 | OBJS-$(CONFIG_VP3DSP) += x86/vp3dsp_init.o |
| @@ -805,72 +805,6 @@ DCT_SAD_FUNC(ssse3) · 804 unmodified lines | ||
| 805 | 805 | #undef HSUM |
| 806 | 806 | #undef DCT_SAD |
| 807 | 807 | |
| 808 | −#define PHADDD(a, t) \ | |
| 809 | − "movq " #a ", " #t " \n\t" \ | |
| 810 | − "psrlq $32, " #a " \n\t" \ | |
| 811 | − "paddd " #t ", " #a " \n\t" | |
| 812 | − | |
| 813 | −/* | |
| 814 | − * pmulhw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15])[16 - 31] | |
| 815 | − * pmulhrw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x8000)[16 - 31] | |
| 816 | − * pmulhrsw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x4000)[15 - 30] | |
| 817 | − */ | |
| 818 | −#define PMULHRW(x, y, s, o) \ | |
| 819 | − "pmulhw " #s ", " #x " \n\t" \ | |
| 820 | − "pmulhw " #s ", " #y " \n\t" \ | |
| 821 | − "paddw " #o ", " #x " \n\t" \ | |
| 822 | − "paddw " #o ", " #y " \n\t" \ | |
| 823 | − "psraw $1, " #x " \n\t" \ | |
| 824 | − "psraw $1, " #y " \n\t" | |
| 825 | −#define DEF(x) x ## _mmx | |
| 826 | −#define SET_RND MOVQ_WONE | |
| 827 | −#define SCALE_OFFSET 1 | |
| 828 | − | |
| 829 | −#include "dsputil_qns_template.c" | |
| 830 | − | |
| 831 | −#undef DEF | |
| 832 | −#undef SET_RND | |
| 833 | −#undef SCALE_OFFSET | |
| 834 | −#undef PMULHRW | |
| 835 | − | |
| 836 | −#define DEF(x) x ## _3dnow | |
| 837 | −#define SET_RND(x) | |
| 838 | −#define SCALE_OFFSET 0 | |
| 839 | −#define PMULHRW(x, y, s, o) \ | |
| 840 | − "pmulhrw " #s ", " #x " \n\t" \ | |
| 841 | − "pmulhrw " #s ", " #y " \n\t" | |
| 842 | − | |
| 843 | −#include "dsputil_qns_template.c" | |
| 844 | − | |
| 845 | −#undef DEF | |
| 846 | −#undef SET_RND | |
| 847 | −#undef SCALE_OFFSET | |
| 848 | −#undef PMULHRW | |
| 849 | − | |
| 850 | −#if HAVE_SSSE3_INLINE | |
| 851 | −#undef PHADDD | |
| 852 | −#define DEF(x) x ## _ssse3 | |
| 853 | −#define SET_RND(x) | |
| 854 | −#define SCALE_OFFSET -1 | |
| 855 | − | |
| 856 | −#define PHADDD(a, t) \ | |
| 857 | − "pshufw $0x0E, " #a ", " #t " \n\t" \ | |
| 858 | − /* faster than phaddd on core2 */ \ | |
| 859 | − "paddd " #t ", " #a " \n\t" | |
| 860 | − | |
| 861 | −#define PMULHRW(x, y, s, o) \ | |
| 862 | − "pmulhrsw " #s ", " #x " \n\t" \ | |
| 863 | − "pmulhrsw " #s ", " #y " \n\t" | |
| 864 | − | |
| 865 | −#include "dsputil_qns_template.c" | |
| 866 | − | |
| 867 | −#undef DEF | |
| 868 | −#undef SET_RND | |
| 869 | −#undef SCALE_OFFSET | |
| 870 | −#undef PMULHRW | |
| 871 | −#undef PHADDD | |
| 872 | −#endif /* HAVE_SSSE3_INLINE */ | |
| 873 | − | |
| 874 | 808 | #endif /* HAVE_INLINE_ASM */ |
| 875 | 809 | |
| 876 | 810 | int ff_sse16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, |
| @@ -921,16 +855,7 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, · 44 unmodified lines | ||
| 921 | 855 | c->nsse[1] = nsse8_mmx; |
| 922 | 856 | if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { |
| 923 | 857 | c->vsad[0] = vsad16_mmx; |
| 924 | − c->try_8x8basis = try_8x8basis_mmx; | |
| 925 | − } | |
| 926 | − c->add_8x8basis = add_8x8basis_mmx; | |
| 927 | − } | |
| 928 | − | |
| 929 | − if (INLINE_AMD3DNOW(cpu_flags)) { | |
| 930 | − if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { | |
| 931 | − c->try_8x8basis = try_8x8basis_3dnow; | |
| 932 | 858 | } |
| 933 | − c->add_8x8basis = add_8x8basis_3dnow; | |
| 934 | 859 | } |
| 935 | 860 | |
| 936 | 861 | if (INLINE_MMXEXT(cpu_flags)) { |
| @@ -956,10 +881,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, · 19 unmodified lines | ||
| 956 | 881 | |
| 957 | 882 | #if HAVE_SSSE3_INLINE |
| 958 | 883 | if (INLINE_SSSE3(cpu_flags)) { |
| 959 | − if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { | |
| 960 | − c->try_8x8basis = try_8x8basis_ssse3; | |
| 961 | − } | |
| 962 | − c->add_8x8basis = add_8x8basis_ssse3; | |
| 963 | 884 | c->sum_abs_dctelem = sum_abs_dctelem_ssse3; |
| 964 | 885 | } |
| 965 | 886 | #endif |
| @@ -1,5 +1,5 @@ | ||
| 1 | 1 | /* |
| 2 | − * DSP utils : QNS functions are compiled 3 times for mmx/3dnow/ssse3 | |
| 2 | + * QNS functions are compiled 3 times for MMX/3DNOW/SSSE3 | |
| 3 | 3 | * Copyright (c) 2004 Michael Niedermayer |
| 4 | 4 | * |
| 5 | 5 | * MMX optimization by Michael Niedermayer <michaelni@gmx.at> |
| @@ -0,0 +1,125 @@ | ||
| 1 | +/* | |
| 2 | + * This file is part of Libav. | |
| 3 | + * | |
| 4 | + * Libav is free software; you can redistribute it and/or | |
| 5 | + * modify it under the terms of the GNU Lesser General Public | |
| 6 | + * License as published by the Free Software Foundation; either | |
| 7 | + * version 2.1 of the License, or (at your option) any later version. | |
| 8 | + * | |
| 9 | + * Libav is distributed in the hope that it will be useful, | |
| 10 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 11 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 12 | + * Lesser General Public License for more details. | |
| 13 | + * | |
| 14 | + * You should have received a copy of the GNU Lesser General Public | |
| 15 | + * License along with Libav; if not, write to the Free Software | |
| 16 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 17 | + */ | |
| 18 | + | |
| 19 | +#include "libavutil/attributes.h" | |
| 20 | +#include "libavutil/cpu.h" | |
| 21 | +#include "libavutil/x86/cpu.h" | |
| 22 | +#include "libavcodec/avcodec.h" | |
| 23 | +#include "libavcodec/mpegvideoencdsp.h" | |
| 24 | + | |
| 25 | +#if HAVE_INLINE_ASM | |
| 26 | + | |
| 27 | +#define PHADDD(a, t) \ | |
| 28 | + "movq " #a ", " #t " \n\t" \ | |
| 29 | + "psrlq $32, " #a " \n\t" \ | |
| 30 | + "paddd " #t ", " #a " \n\t" | |
| 31 | + | |
| 32 | +/* | |
| 33 | + * pmulhw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15])[16 - 31] | |
| 34 | + * pmulhrw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x8000)[16 - 31] | |
| 35 | + * pmulhrsw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x4000)[15 - 30] | |
| 36 | + */ | |
| 37 | +#define PMULHRW(x, y, s, o) \ | |
| 38 | + "pmulhw " #s ", " #x " \n\t" \ | |
| 39 | + "pmulhw " #s ", " #y " \n\t" \ | |
| 40 | + "paddw " #o ", " #x " \n\t" \ | |
| 41 | + "paddw " #o ", " #y " \n\t" \ | |
| 42 | + "psraw $1, " #x " \n\t" \ | |
| 43 | + "psraw $1, " #y " \n\t" | |
| 44 | +#define DEF(x) x ## _mmx | |
| 45 | +#define SET_RND MOVQ_WONE | |
| 46 | +#define SCALE_OFFSET 1 | |
| 47 | + | |
| 48 | +#include "mpegvideoenc_qns_template.c" | |
| 49 | + | |
| 50 | +#undef DEF | |
| 51 | +#undef SET_RND | |
| 52 | +#undef SCALE_OFFSET | |
| 53 | +#undef PMULHRW | |
| 54 | + | |
| 55 | +#define DEF(x) x ## _3dnow | |
| 56 | +#define SET_RND(x) | |
| 57 | +#define SCALE_OFFSET 0 | |
| 58 | +#define PMULHRW(x, y, s, o) \ | |
| 59 | + "pmulhrw " #s ", " #x " \n\t" \ | |
| 60 | + "pmulhrw " #s ", " #y " \n\t" | |
| 61 | + | |
| 62 | +#include "mpegvideoenc_qns_template.c" | |
| 63 | + | |
| 64 | +#undef DEF | |
| 65 | +#undef SET_RND | |
| 66 | +#undef SCALE_OFFSET | |
| 67 | +#undef PMULHRW | |
| 68 | + | |
| 69 | +#if HAVE_SSSE3_INLINE | |
| 70 | +#undef PHADDD | |
| 71 | +#define DEF(x) x ## _ssse3 | |
| 72 | +#define SET_RND(x) | |
| 73 | +#define SCALE_OFFSET -1 | |
| 74 | + | |
| 75 | +#define PHADDD(a, t) \ | |
| 76 | + "pshufw $0x0E, " #a ", " #t " \n\t" \ | |
| 77 | + /* faster than phaddd on core2 */ \ | |
| 78 | + "paddd " #t ", " #a " \n\t" | |
| 79 | + | |
| 80 | +#define PMULHRW(x, y, s, o) \ | |
| 81 | + "pmulhrsw " #s ", " #x " \n\t" \ | |
| 82 | + "pmulhrsw " #s ", " #y " \n\t" | |
| 83 | + | |
| 84 | +#include "mpegvideoenc_qns_template.c" | |
| 85 | + | |
| 86 | +#undef DEF | |
| 87 | +#undef SET_RND | |
| 88 | +#undef SCALE_OFFSET | |
| 89 | +#undef PMULHRW | |
| 90 | +#undef PHADDD | |
| 91 | +#endif /* HAVE_SSSE3_INLINE */ | |
| 92 | + | |
| 93 | +#endif /* HAVE_INLINE_ASM */ | |
| 94 | + | |
| 95 | +av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, | |
| 96 | + AVCodecContext *avctx) | |
| 97 | +{ | |
| 98 | +#if HAVE_INLINE_ASM | |
| 99 | + int cpu_flags = av_get_cpu_flags(); | |
| 100 | + | |
| 101 | + if (INLINE_MMX(cpu_flags)) { | |
| 102 | + if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { | |
| 103 | + c->try_8x8basis = try_8x8basis_mmx; | |
| 104 | + } | |
| 105 | + c->add_8x8basis = add_8x8basis_mmx; | |
| 106 | + } | |
| 107 | + | |
| 108 | + if (INLINE_AMD3DNOW(cpu_flags)) { | |
| 109 | + if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { | |
| 110 | + c->try_8x8basis = try_8x8basis_3dnow; | |
| 111 | + } | |
| 112 | + c->add_8x8basis = add_8x8basis_3dnow; | |
| 113 | + } | |
| 114 | + | |
| 115 | +#if HAVE_SSSE3_INLINE | |
| 116 | + if (INLINE_SSSE3(cpu_flags)) { | |
| 117 | + if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { | |
| 118 | + c->try_8x8basis = try_8x8basis_ssse3; | |
| 119 | + } | |
| 120 | + c->add_8x8basis = add_8x8basis_ssse3; | |
| 121 | + } | |
| 122 | +#endif /* HAVE_SSSE3_INLINE */ | |
| 123 | + | |
| 124 | +#endif /* HAVE_INLINE_ASM */ | |
| 125 | +} | |