FFmpeg: libavcodec/x86/h264_intrapred

00001 /*
00002  * Copyright (c) 2010 Jason Garrett-Glaser
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include "libavutil/cpu.h"
00022 #include "libavcodec/h264pred.h"
00023 
00024 #define PRED4x4(TYPE, DEPTH, OPT) \
00025 void ff_pred4x4_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, const uint8_t *topright, int stride);
00026 
00027 PRED4x4(dc, 10, mmxext)
00028 PRED4x4(down_left, 10, sse2)
00029 PRED4x4(down_left, 10, avx)
00030 PRED4x4(down_right, 10, sse2)
00031 PRED4x4(down_right, 10, ssse3)
00032 PRED4x4(down_right, 10, avx)
00033 PRED4x4(vertical_left, 10, sse2)
00034 PRED4x4(vertical_left, 10, avx)
00035 PRED4x4(vertical_right, 10, sse2)
00036 PRED4x4(vertical_right, 10, ssse3)
00037 PRED4x4(vertical_right, 10, avx)
00038 PRED4x4(horizontal_up, 10, mmxext)
00039 PRED4x4(horizontal_down, 10, sse2)
00040 PRED4x4(horizontal_down, 10, ssse3)
00041 PRED4x4(horizontal_down, 10, avx)
00042 
00043 #define PRED8x8(TYPE, DEPTH, OPT) \
00044 void ff_pred8x8_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
00045 
00046 PRED8x8(vertical, 10, sse2)
00047 PRED8x8(horizontal, 10, sse2)
00048 
00049 void ff_pred16x16_vertical_mmx     (uint8_t *src, int stride);
00050 void ff_pred16x16_vertical_sse     (uint8_t *src, int stride);
00051 void ff_pred16x16_horizontal_mmx   (uint8_t *src, int stride);
00052 void ff_pred16x16_horizontal_mmxext(uint8_t *src, int stride);
00053 void ff_pred16x16_horizontal_ssse3 (uint8_t *src, int stride);
00054 void ff_pred16x16_dc_mmxext        (uint8_t *src, int stride);
00055 void ff_pred16x16_dc_sse2          (uint8_t *src, int stride);
00056 void ff_pred16x16_dc_ssse3         (uint8_t *src, int stride);
00057 void ff_pred16x16_plane_h264_mmx   (uint8_t *src, int stride);
00058 void ff_pred16x16_plane_h264_mmx2  (uint8_t *src, int stride);
00059 void ff_pred16x16_plane_h264_sse2  (uint8_t *src, int stride);
00060 void ff_pred16x16_plane_h264_ssse3 (uint8_t *src, int stride);
00061 void ff_pred16x16_plane_rv40_mmx   (uint8_t *src, int stride);
00062 void ff_pred16x16_plane_rv40_mmx2  (uint8_t *src, int stride);
00063 void ff_pred16x16_plane_rv40_sse2  (uint8_t *src, int stride);
00064 void ff_pred16x16_plane_rv40_ssse3 (uint8_t *src, int stride);
00065 void ff_pred16x16_plane_svq3_mmx   (uint8_t *src, int stride);
00066 void ff_pred16x16_plane_svq3_mmx2  (uint8_t *src, int stride);
00067 void ff_pred16x16_plane_svq3_sse2  (uint8_t *src, int stride);
00068 void ff_pred16x16_plane_svq3_ssse3 (uint8_t *src, int stride);
00069 void ff_pred16x16_tm_vp8_mmx       (uint8_t *src, int stride);
00070 void ff_pred16x16_tm_vp8_mmxext    (uint8_t *src, int stride);
00071 void ff_pred16x16_tm_vp8_sse2      (uint8_t *src, int stride);
00072 void ff_pred8x8_top_dc_mmxext      (uint8_t *src, int stride);
00073 void ff_pred8x8_dc_rv40_mmxext     (uint8_t *src, int stride);
00074 void ff_pred8x8_dc_mmxext          (uint8_t *src, int stride);
00075 void ff_pred8x8_vertical_mmx       (uint8_t *src, int stride);
00076 void ff_pred8x8_horizontal_mmx     (uint8_t *src, int stride);
00077 void ff_pred8x8_horizontal_mmxext  (uint8_t *src, int stride);
00078 void ff_pred8x8_horizontal_ssse3   (uint8_t *src, int stride);
00079 void ff_pred8x8_plane_mmx          (uint8_t *src, int stride);
00080 void ff_pred8x8_plane_mmx2         (uint8_t *src, int stride);
00081 void ff_pred8x8_plane_sse2         (uint8_t *src, int stride);
00082 void ff_pred8x8_plane_ssse3        (uint8_t *src, int stride);
00083 void ff_pred8x8_tm_vp8_mmx         (uint8_t *src, int stride);
00084 void ff_pred8x8_tm_vp8_mmxext      (uint8_t *src, int stride);
00085 void ff_pred8x8_tm_vp8_sse2        (uint8_t *src, int stride);
00086 void ff_pred8x8_tm_vp8_ssse3       (uint8_t *src, int stride);
00087 void ff_pred8x8l_top_dc_mmxext     (uint8_t *src, int has_topleft, int has_topright, int stride);
00088 void ff_pred8x8l_top_dc_ssse3      (uint8_t *src, int has_topleft, int has_topright, int stride);
00089 void ff_pred8x8l_dc_mmxext         (uint8_t *src, int has_topleft, int has_topright, int stride);
00090 void ff_pred8x8l_dc_ssse3          (uint8_t *src, int has_topleft, int has_topright, int stride);
00091 void ff_pred8x8l_horizontal_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
00092 void ff_pred8x8l_horizontal_ssse3  (uint8_t *src, int has_topleft, int has_topright, int stride);
00093 void ff_pred8x8l_vertical_mmxext   (uint8_t *src, int has_topleft, int has_topright, int stride);
00094 void ff_pred8x8l_vertical_ssse3    (uint8_t *src, int has_topleft, int has_topright, int stride);
00095 void ff_pred8x8l_down_left_mmxext  (uint8_t *src, int has_topleft, int has_topright, int stride);
00096 void ff_pred8x8l_down_left_sse2    (uint8_t *src, int has_topleft, int has_topright, int stride);
00097 void ff_pred8x8l_down_left_ssse3   (uint8_t *src, int has_topleft, int has_topright, int stride);
00098 void ff_pred8x8l_down_right_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
00099 void ff_pred8x8l_down_right_sse2   (uint8_t *src, int has_topleft, int has_topright, int stride);
00100 void ff_pred8x8l_down_right_ssse3  (uint8_t *src, int has_topleft, int has_topright, int stride);
00101 void ff_pred8x8l_vertical_right_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
00102 void ff_pred8x8l_vertical_right_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
00103 void ff_pred8x8l_vertical_right_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
00104 void ff_pred8x8l_vertical_left_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
00105 void ff_pred8x8l_vertical_left_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
00106 void ff_pred8x8l_horizontal_up_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
00107 void ff_pred8x8l_horizontal_up_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
00108 void ff_pred8x8l_horizontal_down_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
00109 void ff_pred8x8l_horizontal_down_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
00110 void ff_pred8x8l_horizontal_down_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
00111 void ff_pred4x4_dc_mmxext          (uint8_t *src, const uint8_t *topright, int stride);
00112 void ff_pred4x4_down_left_mmxext   (uint8_t *src, const uint8_t *topright, int stride);
00113 void ff_pred4x4_down_right_mmxext  (uint8_t *src, const uint8_t *topright, int stride);
00114 void ff_pred4x4_vertical_left_mmxext(uint8_t *src, const uint8_t *topright, int stride);
00115 void ff_pred4x4_vertical_right_mmxext(uint8_t *src, const uint8_t *topright, int stride);
00116 void ff_pred4x4_horizontal_up_mmxext(uint8_t *src, const uint8_t *topright, int stride);
00117 void ff_pred4x4_horizontal_down_mmxext(uint8_t *src, const uint8_t *topright, int stride);
00118 void ff_pred4x4_tm_vp8_mmx         (uint8_t *src, const uint8_t *topright, int stride);
00119 void ff_pred4x4_tm_vp8_mmxext      (uint8_t *src, const uint8_t *topright, int stride);
00120 void ff_pred4x4_tm_vp8_ssse3       (uint8_t *src, const uint8_t *topright, int stride);
00121 void ff_pred4x4_vertical_vp8_mmxext(uint8_t *src, const uint8_t *topright, int stride);
00122 
00123 void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth)
00124 {
00125     int mm_flags = av_get_cpu_flags();
00126 
00127 #if HAVE_YASM
00128     if (bit_depth == 8) {
00129         if (mm_flags & AV_CPU_FLAG_MMX) {
00130             h->pred16x16[VERT_PRED8x8         ] = ff_pred16x16_vertical_mmx;
00131             h->pred16x16[HOR_PRED8x8          ] = ff_pred16x16_horizontal_mmx;
00132             h->pred8x8  [VERT_PRED8x8         ] = ff_pred8x8_vertical_mmx;
00133             h->pred8x8  [HOR_PRED8x8          ] = ff_pred8x8_horizontal_mmx;
00134             if (codec_id == CODEC_ID_VP8) {
00135                 h->pred16x16[PLANE_PRED8x8    ] = ff_pred16x16_tm_vp8_mmx;
00136                 h->pred8x8  [PLANE_PRED8x8    ] = ff_pred8x8_tm_vp8_mmx;
00137                 h->pred4x4  [TM_VP8_PRED      ] = ff_pred4x4_tm_vp8_mmx;
00138             } else {
00139                 h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_plane_mmx;
00140                 if (codec_id == CODEC_ID_SVQ3) {
00141                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_mmx;
00142                 } else if (codec_id == CODEC_ID_RV40) {
00143                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_mmx;
00144                 } else {
00145                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_mmx;
00146                 }
00147             }
00148         }
00149 
00150         if (mm_flags & AV_CPU_FLAG_MMX2) {
00151             h->pred16x16[HOR_PRED8x8            ] = ff_pred16x16_horizontal_mmxext;
00152             h->pred16x16[DC_PRED8x8             ] = ff_pred16x16_dc_mmxext;
00153             h->pred8x8  [HOR_PRED8x8            ] = ff_pred8x8_horizontal_mmxext;
00154             h->pred8x8l [TOP_DC_PRED            ] = ff_pred8x8l_top_dc_mmxext;
00155             h->pred8x8l [DC_PRED                ] = ff_pred8x8l_dc_mmxext;
00156             h->pred8x8l [HOR_PRED               ] = ff_pred8x8l_horizontal_mmxext;
00157             h->pred8x8l [VERT_PRED              ] = ff_pred8x8l_vertical_mmxext;
00158             h->pred8x8l [DIAG_DOWN_RIGHT_PRED   ] = ff_pred8x8l_down_right_mmxext;
00159             h->pred8x8l [VERT_RIGHT_PRED        ] = ff_pred8x8l_vertical_right_mmxext;
00160             h->pred8x8l [HOR_UP_PRED            ] = ff_pred8x8l_horizontal_up_mmxext;
00161             h->pred8x8l [DIAG_DOWN_LEFT_PRED    ] = ff_pred8x8l_down_left_mmxext;
00162             h->pred8x8l [HOR_DOWN_PRED          ] = ff_pred8x8l_horizontal_down_mmxext;
00163             h->pred4x4  [DIAG_DOWN_RIGHT_PRED   ] = ff_pred4x4_down_right_mmxext;
00164             h->pred4x4  [VERT_RIGHT_PRED        ] = ff_pred4x4_vertical_right_mmxext;
00165             h->pred4x4  [HOR_DOWN_PRED          ] = ff_pred4x4_horizontal_down_mmxext;
00166             h->pred4x4  [DC_PRED                ] = ff_pred4x4_dc_mmxext;
00167             if (codec_id == CODEC_ID_VP8 || codec_id == CODEC_ID_H264) {
00168                 h->pred4x4  [DIAG_DOWN_LEFT_PRED] = ff_pred4x4_down_left_mmxext;
00169             }
00170             if (codec_id == CODEC_ID_SVQ3 || codec_id == CODEC_ID_H264) {
00171                 h->pred4x4  [VERT_LEFT_PRED     ] = ff_pred4x4_vertical_left_mmxext;
00172             }
00173             if (codec_id != CODEC_ID_RV40) {
00174                 h->pred4x4  [HOR_UP_PRED        ] = ff_pred4x4_horizontal_up_mmxext;
00175             }
00176             if (codec_id == CODEC_ID_SVQ3 || codec_id == CODEC_ID_H264) {
00177                 h->pred8x8  [TOP_DC_PRED8x8     ] = ff_pred8x8_top_dc_mmxext;
00178                 h->pred8x8  [DC_PRED8x8         ] = ff_pred8x8_dc_mmxext;
00179             }
00180             if (codec_id == CODEC_ID_VP8) {
00181                 h->pred16x16[PLANE_PRED8x8      ] = ff_pred16x16_tm_vp8_mmxext;
00182                 h->pred8x8  [DC_PRED8x8         ] = ff_pred8x8_dc_rv40_mmxext;
00183                 h->pred8x8  [PLANE_PRED8x8      ] = ff_pred8x8_tm_vp8_mmxext;
00184                 h->pred4x4  [TM_VP8_PRED        ] = ff_pred4x4_tm_vp8_mmxext;
00185                 h->pred4x4  [VERT_PRED          ] = ff_pred4x4_vertical_vp8_mmxext;
00186             } else {
00187                 h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_plane_mmx2;
00188                 if (codec_id == CODEC_ID_SVQ3) {
00189                     h->pred16x16[PLANE_PRED8x8  ] = ff_pred16x16_plane_svq3_mmx2;
00190                 } else if (codec_id == CODEC_ID_RV40) {
00191                     h->pred16x16[PLANE_PRED8x8  ] = ff_pred16x16_plane_rv40_mmx2;
00192                 } else {
00193                     h->pred16x16[PLANE_PRED8x8  ] = ff_pred16x16_plane_h264_mmx2;
00194                 }
00195             }
00196         }
00197 
00198         if (mm_flags & AV_CPU_FLAG_SSE) {
00199             h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_sse;
00200         }
00201 
00202         if (mm_flags & AV_CPU_FLAG_SSE2) {
00203             h->pred16x16[DC_PRED8x8           ] = ff_pred16x16_dc_sse2;
00204             h->pred8x8l [DIAG_DOWN_LEFT_PRED  ] = ff_pred8x8l_down_left_sse2;
00205             h->pred8x8l [DIAG_DOWN_RIGHT_PRED ] = ff_pred8x8l_down_right_sse2;
00206             h->pred8x8l [VERT_RIGHT_PRED      ] = ff_pred8x8l_vertical_right_sse2;
00207             h->pred8x8l [VERT_LEFT_PRED       ] = ff_pred8x8l_vertical_left_sse2;
00208             h->pred8x8l [HOR_DOWN_PRED        ] = ff_pred8x8l_horizontal_down_sse2;
00209             if (codec_id == CODEC_ID_VP8) {
00210                 h->pred16x16[PLANE_PRED8x8    ] = ff_pred16x16_tm_vp8_sse2;
00211                 h->pred8x8  [PLANE_PRED8x8    ] = ff_pred8x8_tm_vp8_sse2;
00212             } else {
00213                 h->pred8x8  [PLANE_PRED8x8    ] = ff_pred8x8_plane_sse2;
00214                 if (codec_id == CODEC_ID_SVQ3) {
00215                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_sse2;
00216                 } else if (codec_id == CODEC_ID_RV40) {
00217                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_sse2;
00218                 } else {
00219                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_sse2;
00220                 }
00221             }
00222         }
00223 
00224         if (mm_flags & AV_CPU_FLAG_SSSE3) {
00225             h->pred16x16[HOR_PRED8x8          ] = ff_pred16x16_horizontal_ssse3;
00226             h->pred16x16[DC_PRED8x8           ] = ff_pred16x16_dc_ssse3;
00227             h->pred8x8  [HOR_PRED8x8          ] = ff_pred8x8_horizontal_ssse3;
00228             h->pred8x8l [TOP_DC_PRED          ] = ff_pred8x8l_top_dc_ssse3;
00229             h->pred8x8l [DC_PRED              ] = ff_pred8x8l_dc_ssse3;
00230             h->pred8x8l [HOR_PRED             ] = ff_pred8x8l_horizontal_ssse3;
00231             h->pred8x8l [VERT_PRED            ] = ff_pred8x8l_vertical_ssse3;
00232             h->pred8x8l [DIAG_DOWN_LEFT_PRED  ] = ff_pred8x8l_down_left_ssse3;
00233             h->pred8x8l [DIAG_DOWN_RIGHT_PRED ] = ff_pred8x8l_down_right_ssse3;
00234             h->pred8x8l [VERT_RIGHT_PRED      ] = ff_pred8x8l_vertical_right_ssse3;
00235             h->pred8x8l [VERT_LEFT_PRED       ] = ff_pred8x8l_vertical_left_ssse3;
00236             h->pred8x8l [HOR_UP_PRED          ] = ff_pred8x8l_horizontal_up_ssse3;
00237             h->pred8x8l [HOR_DOWN_PRED        ] = ff_pred8x8l_horizontal_down_ssse3;
00238             if (codec_id == CODEC_ID_VP8) {
00239                 h->pred8x8  [PLANE_PRED8x8    ] = ff_pred8x8_tm_vp8_ssse3;
00240                 h->pred4x4  [TM_VP8_PRED      ] = ff_pred4x4_tm_vp8_ssse3;
00241             } else {
00242                 h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_plane_ssse3;
00243                 if (codec_id == CODEC_ID_SVQ3) {
00244                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_ssse3;
00245                 } else if (codec_id == CODEC_ID_RV40) {
00246                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_ssse3;
00247                 } else {
00248                     h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_ssse3;
00249                 }
00250             }
00251         }
00252     } else if (bit_depth == 10) {
00253         if (mm_flags & AV_CPU_FLAG_MMX2) {
00254             h->pred4x4[DC_PRED             ] = ff_pred4x4_dc_10_mmxext;
00255             h->pred4x4[HOR_UP_PRED         ] = ff_pred4x4_horizontal_up_10_mmxext;
00256         }
00257         if (mm_flags & AV_CPU_FLAG_SSE2) {
00258             h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_sse2;
00259             h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_sse2;
00260             h->pred4x4[VERT_LEFT_PRED      ] = ff_pred4x4_vertical_left_10_sse2;
00261             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_sse2;
00262             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_sse2;
00263 
00264             h->pred8x8[VERT_PRED8x8        ] = ff_pred8x8_vertical_10_sse2;
00265             h->pred8x8[HOR_PRED8x8         ] = ff_pred8x8_horizontal_10_sse2;
00266         }
00267         if (mm_flags & AV_CPU_FLAG_SSSE3) {
00268             h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_ssse3;
00269             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_ssse3;
00270             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_ssse3;
00271         }
00272 #if HAVE_AVX
00273         if (mm_flags & AV_CPU_FLAG_AVX) {
00274             h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_avx;
00275             h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_avx;
00276             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_avx;
00277             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_avx;
00278         }
00279 #endif /* HAVE_AVX */
00280     }
00281 #endif /* HAVE_YASM */
00282 }
libavcodec/x86/h264_intrapred_init.c
