FFmpeg: libavcodec/x86/h264dsp

00001 /*
00002  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
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 "libavutil/x86/asm.h"
00023 #include "libavcodec/h264dsp.h"
00024 #include "dsputil_mmx.h"
00025 
00026 /***********************************/
00027 /* IDCT */
00028 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT)                                  \
00029 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst,    \
00030                                                        int16_t *block,  \
00031                                                        int stride);
00032 
00033 IDCT_ADD_FUNC(, 8, mmx)
00034 IDCT_ADD_FUNC(, 10, sse2)
00035 IDCT_ADD_FUNC(_dc, 8, mmx2)
00036 IDCT_ADD_FUNC(_dc, 10, mmx2)
00037 IDCT_ADD_FUNC(8_dc, 8, mmx2)
00038 IDCT_ADD_FUNC(8_dc, 10, sse2)
00039 IDCT_ADD_FUNC(8, 8, mmx)
00040 IDCT_ADD_FUNC(8, 8, sse2)
00041 IDCT_ADD_FUNC(8, 10, sse2)
00042 #if HAVE_AVX
00043 IDCT_ADD_FUNC(, 10, avx)
00044 IDCT_ADD_FUNC(8_dc, 10, avx)
00045 IDCT_ADD_FUNC(8, 10, avx)
00046 #endif
00047 
00048 
00049 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT)                         \
00050 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT       \
00051     (uint8_t *dst, const int *block_offset,                             \
00052      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
00053 
00054 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
00055 IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
00056 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
00057 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
00058 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
00059 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
00060 IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
00061 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
00062 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
00063 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
00064 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
00065 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
00066 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
00067 #if HAVE_AVX
00068 IDCT_ADD_REP_FUNC(, 16, 10, avx)
00069 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
00070 #endif
00071 
00072 
00073 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT)                      \
00074 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT     \
00075     (uint8_t **dst, const int *block_offset,                          \
00076      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
00077 
00078 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
00079 IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
00080 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
00081 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
00082 #if HAVE_AVX
00083 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
00084 #endif
00085 
00086 void ff_h264_luma_dc_dequant_idct_mmx(DCTELEM *output, DCTELEM *input, int qmul);
00087 void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
00088 
00089 /***********************************/
00090 /* deblocking */
00091 
00092 void ff_h264_loop_filter_strength_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
00093                                        int8_t ref[2][40], int16_t mv[2][40][2],
00094                                        int bidir, int edges, int step,
00095                                        int mask_mv0, int mask_mv1, int field);
00096 
00097 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
00098 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
00099                                                                int stride,    \
00100                                                                int alpha,     \
00101                                                                int beta,      \
00102                                                                int8_t *tc0);
00103 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
00104 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
00105                                                                int stride,    \
00106                                                                int alpha,     \
00107                                                                int beta);
00108 
00109 #define LF_FUNCS(type, depth)                   \
00110 LF_FUNC(h,  chroma,       depth, mmx2)          \
00111 LF_IFUNC(h, chroma_intra, depth, mmx2)          \
00112 LF_FUNC(v,  chroma,       depth, mmx2)          \
00113 LF_IFUNC(v, chroma_intra, depth, mmx2)          \
00114 LF_FUNC(h,  luma,         depth, mmx2)          \
00115 LF_IFUNC(h, luma_intra,   depth, mmx2)          \
00116 LF_FUNC(h,  luma,         depth, sse2)          \
00117 LF_IFUNC(h, luma_intra,   depth, sse2)          \
00118 LF_FUNC(v,  luma,         depth, sse2)          \
00119 LF_IFUNC(v, luma_intra,   depth, sse2)          \
00120 LF_FUNC(h,  chroma,       depth, sse2)          \
00121 LF_IFUNC(h, chroma_intra, depth, sse2)          \
00122 LF_FUNC(v,  chroma,       depth, sse2)          \
00123 LF_IFUNC(v, chroma_intra, depth, sse2)          \
00124 LF_FUNC(h,  luma,         depth, avx)           \
00125 LF_IFUNC(h, luma_intra,   depth, avx)           \
00126 LF_FUNC(v,  luma,         depth, avx)           \
00127 LF_IFUNC(v, luma_intra,   depth, avx)           \
00128 LF_FUNC(h,  chroma,       depth, avx)           \
00129 LF_IFUNC(h, chroma_intra, depth, avx)           \
00130 LF_FUNC(v,  chroma,       depth, avx)           \
00131 LF_IFUNC(v, chroma_intra, depth, avx)
00132 
00133 LF_FUNCS(uint8_t,   8)
00134 LF_FUNCS(uint16_t, 10)
00135 
00136 #if ARCH_X86_32 && HAVE_YASM
00137 LF_FUNC(v8, luma, 8, mmx2)
00138 static void ff_deblock_v_luma_8_mmx2(uint8_t *pix, int stride, int alpha,
00139                                      int beta, int8_t *tc0)
00140 {
00141     if ((tc0[0] & tc0[1]) >= 0)
00142         ff_deblock_v8_luma_8_mmx2(pix + 0, stride, alpha, beta, tc0);
00143     if ((tc0[2] & tc0[3]) >= 0)
00144         ff_deblock_v8_luma_8_mmx2(pix + 8, stride, alpha, beta, tc0 + 2);
00145 }
00146 
00147 LF_IFUNC(v8, luma_intra, 8, mmx2)
00148 static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride,
00149                                            int alpha, int beta)
00150 {
00151     ff_deblock_v8_luma_intra_8_mmx2(pix + 0, stride, alpha, beta);
00152     ff_deblock_v8_luma_intra_8_mmx2(pix + 8, stride, alpha, beta);
00153 }
00154 #endif /* ARCH_X86_32 */
00155 
00156 LF_FUNC(v,  luma,       10, mmx2)
00157 LF_IFUNC(v, luma_intra, 10, mmx2)
00158 
00159 /***********************************/
00160 /* weighted prediction */
00161 
00162 #define H264_WEIGHT(W, OPT)                                             \
00163 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride,         \
00164                                       int height, int log2_denom,       \
00165                                       int weight, int offset);
00166 
00167 #define H264_BIWEIGHT(W, OPT)                                           \
00168 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
00169                                         int stride, int height,         \
00170                                         int log2_denom, int weightd,    \
00171                                         int weights, int offset);
00172 
00173 #define H264_BIWEIGHT_MMX(W)                    \
00174     H264_WEIGHT(W, mmx2)                        \
00175     H264_BIWEIGHT(W, mmx2)
00176 
00177 #define H264_BIWEIGHT_MMX_SSE(W)                \
00178     H264_BIWEIGHT_MMX(W)                        \
00179     H264_WEIGHT(W, sse2)                        \
00180     H264_BIWEIGHT(W, sse2)                      \
00181     H264_BIWEIGHT(W, ssse3)
00182 
00183 H264_BIWEIGHT_MMX_SSE(16)
00184 H264_BIWEIGHT_MMX_SSE(8)
00185 H264_BIWEIGHT_MMX(4)
00186 
00187 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
00188 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
00189                                                     int stride,         \
00190                                                     int height,         \
00191                                                     int log2_denom,     \
00192                                                     int weight,         \
00193                                                     int offset);
00194 
00195 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
00196 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
00197                                                       uint8_t *src,     \
00198                                                       int stride,       \
00199                                                       int height,       \
00200                                                       int log2_denom,   \
00201                                                       int weightd,      \
00202                                                       int weights,      \
00203                                                       int offset);
00204 
00205 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
00206     H264_WEIGHT_10(W, DEPTH, sse2)              \
00207     H264_WEIGHT_10(W, DEPTH, sse4)              \
00208     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
00209     H264_BIWEIGHT_10(W, DEPTH, sse4)
00210 
00211 H264_BIWEIGHT_10_SSE(16, 10)
00212 H264_BIWEIGHT_10_SSE(8,  10)
00213 H264_BIWEIGHT_10_SSE(4,  10)
00214 
00215 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
00216                          const int chroma_format_idc)
00217 {
00218 #if HAVE_YASM
00219     int mm_flags = av_get_cpu_flags();
00220 
00221     if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMXEXT)
00222         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
00223 
00224     if (bit_depth == 8) {
00225         if (mm_flags & AV_CPU_FLAG_MMX) {
00226             c->h264_idct_dc_add   =
00227             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
00228             c->h264_idct8_dc_add  =
00229             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
00230 
00231             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
00232             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
00233             if (chroma_format_idc == 1)
00234                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
00235             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
00236             if (mm_flags & AV_CPU_FLAG_CMOV)
00237                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
00238 
00239             if (mm_flags & AV_CPU_FLAG_MMXEXT) {
00240                 c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmx2;
00241                 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmx2;
00242                 c->h264_idct_add16   = ff_h264_idct_add16_8_mmx2;
00243                 c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmx2;
00244                 if (chroma_format_idc == 1)
00245                     c->h264_idct_add8 = ff_h264_idct_add8_8_mmx2;
00246                 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx2;
00247 
00248                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmx2;
00249                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmx2;
00250                 if (chroma_format_idc == 1) {
00251                     c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmx2;
00252                     c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmx2;
00253                 }
00254 #if ARCH_X86_32
00255                 c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_mmx2;
00256                 c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmx2;
00257                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
00258                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
00259 #endif /* ARCH_X86_32 */
00260                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmx2;
00261                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmx2;
00262                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmx2;
00263 
00264                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmx2;
00265                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmx2;
00266                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmx2;
00267 
00268                 if (mm_flags & AV_CPU_FLAG_SSE2) {
00269                     c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
00270 
00271                     c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
00272                     c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
00273                     if (chroma_format_idc == 1)
00274                         c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
00275                     c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
00276                     c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
00277 
00278                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
00279                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
00280 
00281                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
00282                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
00283 
00284 #if HAVE_ALIGNED_STACK
00285                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
00286                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
00287                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
00288                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
00289 #endif /* HAVE_ALIGNED_STACK */
00290                 }
00291                 if (mm_flags & AV_CPU_FLAG_SSSE3) {
00292                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
00293                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
00294                 }
00295                 if (HAVE_AVX && mm_flags & AV_CPU_FLAG_AVX) {
00296 #if HAVE_ALIGNED_STACK
00297                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
00298                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
00299                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
00300                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
00301 #endif /* HAVE_ALIGNED_STACK */
00302                 }
00303             }
00304         }
00305     } else if (bit_depth == 10) {
00306         if (mm_flags & AV_CPU_FLAG_MMX) {
00307             if (mm_flags & AV_CPU_FLAG_MMXEXT) {
00308 #if ARCH_X86_32
00309                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmx2;
00310                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmx2;
00311                 c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmx2;
00312                 c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmx2;
00313                 c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmx2;
00314                 c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmx2;
00315 #endif /* ARCH_X86_32 */
00316                 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmx2;
00317                 if (mm_flags & AV_CPU_FLAG_SSE2) {
00318                     c->h264_idct_add     = ff_h264_idct_add_10_sse2;
00319                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
00320 
00321                     c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
00322                     if (chroma_format_idc == 1)
00323                         c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
00324                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
00325 #if HAVE_ALIGNED_STACK
00326                     c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
00327                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
00328 #endif /* HAVE_ALIGNED_STACK */
00329 
00330                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
00331                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
00332                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
00333 
00334                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
00335                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
00336                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
00337 
00338                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
00339                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
00340 #if HAVE_ALIGNED_STACK
00341                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
00342                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
00343                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
00344                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
00345 #endif /* HAVE_ALIGNED_STACK */
00346                 }
00347                 if (mm_flags & AV_CPU_FLAG_SSE4) {
00348                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
00349                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
00350                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
00351 
00352                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
00353                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
00354                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
00355                 }
00356 #if HAVE_AVX
00357                 if (mm_flags & AV_CPU_FLAG_AVX) {
00358                     c->h264_idct_dc_add  =
00359                     c->h264_idct_add     = ff_h264_idct_add_10_avx;
00360                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
00361 
00362                     c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
00363                     if (chroma_format_idc == 1)
00364                         c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
00365                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
00366 #if HAVE_ALIGNED_STACK
00367                     c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
00368                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
00369 #endif /* HAVE_ALIGNED_STACK */
00370 
00371                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
00372                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
00373 #if HAVE_ALIGNED_STACK
00374                     c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
00375                     c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
00376                     c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
00377                     c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
00378 #endif /* HAVE_ALIGNED_STACK */
00379                 }
00380 #endif /* HAVE_AVX */
00381             }
00382         }
00383     }
00384 #endif /* HAVE_YASM */
00385 }
libavcodec/x86/h264dsp_mmx.c
