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 "libavutil/x86/cpu.h"
00024 #include "libavcodec/h264dsp.h"
00025 #include "dsputil_mmx.h"
00026 
00027 /***********************************/
00028 /* IDCT */
00029 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT)                                  \
00030 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst,    \
00031                                                        int16_t *block,  \
00032                                                        int stride);
00033 
00034 IDCT_ADD_FUNC(, 8, mmx)
00035 IDCT_ADD_FUNC(, 10, sse2)
00036 IDCT_ADD_FUNC(_dc, 8, mmx2)
00037 IDCT_ADD_FUNC(_dc, 10, mmx2)
00038 IDCT_ADD_FUNC(8_dc, 8, mmx2)
00039 IDCT_ADD_FUNC(8_dc, 10, sse2)
00040 IDCT_ADD_FUNC(8, 8, mmx)
00041 IDCT_ADD_FUNC(8, 8, sse2)
00042 IDCT_ADD_FUNC(8, 10, sse2)
00043 IDCT_ADD_FUNC(, 10, avx)
00044 IDCT_ADD_FUNC(8_dc, 10, avx)
00045 IDCT_ADD_FUNC(8, 10, avx)
00046 
00047 
00048 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT)                         \
00049 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT       \
00050     (uint8_t *dst, const int *block_offset,                             \
00051      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
00052 
00053 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
00054 IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
00055 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
00056 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
00057 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
00058 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
00059 IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
00060 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
00061 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
00062 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
00063 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
00064 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
00065 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
00066 IDCT_ADD_REP_FUNC(, 16, 10, avx)
00067 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
00068 
00069 
00070 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT)                      \
00071 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT     \
00072     (uint8_t **dst, const int *block_offset,                          \
00073      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
00074 
00075 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
00076 IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
00077 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
00078 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
00079 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
00080 
00081 void ff_h264_luma_dc_dequant_idct_mmx(DCTELEM *output, DCTELEM *input, int qmul);
00082 void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
00083 
00084 /***********************************/
00085 /* deblocking */
00086 
00087 void ff_h264_loop_filter_strength_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
00088                                        int8_t ref[2][40], int16_t mv[2][40][2],
00089                                        int bidir, int edges, int step,
00090                                        int mask_mv0, int mask_mv1, int field);
00091 
00092 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
00093 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
00094                                                                int stride,    \
00095                                                                int alpha,     \
00096                                                                int beta,      \
00097                                                                int8_t *tc0);
00098 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
00099 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
00100                                                                int stride,    \
00101                                                                int alpha,     \
00102                                                                int beta);
00103 
00104 #define LF_FUNCS(type, depth)                   \
00105 LF_FUNC(h,  chroma,       depth, mmx2)          \
00106 LF_IFUNC(h, chroma_intra, depth, mmx2)          \
00107 LF_FUNC(v,  chroma,       depth, mmx2)          \
00108 LF_IFUNC(v, chroma_intra, depth, mmx2)          \
00109 LF_FUNC(h,  luma,         depth, mmx2)          \
00110 LF_IFUNC(h, luma_intra,   depth, mmx2)          \
00111 LF_FUNC(h,  luma,         depth, sse2)          \
00112 LF_IFUNC(h, luma_intra,   depth, sse2)          \
00113 LF_FUNC(v,  luma,         depth, sse2)          \
00114 LF_IFUNC(v, luma_intra,   depth, sse2)          \
00115 LF_FUNC(h,  chroma,       depth, sse2)          \
00116 LF_IFUNC(h, chroma_intra, depth, sse2)          \
00117 LF_FUNC(v,  chroma,       depth, sse2)          \
00118 LF_IFUNC(v, chroma_intra, depth, sse2)          \
00119 LF_FUNC(h,  luma,         depth, avx)           \
00120 LF_IFUNC(h, luma_intra,   depth, avx)           \
00121 LF_FUNC(v,  luma,         depth, avx)           \
00122 LF_IFUNC(v, luma_intra,   depth, avx)           \
00123 LF_FUNC(h,  chroma,       depth, avx)           \
00124 LF_IFUNC(h, chroma_intra, depth, avx)           \
00125 LF_FUNC(v,  chroma,       depth, avx)           \
00126 LF_IFUNC(v, chroma_intra, depth, avx)
00127 
00128 LF_FUNCS(uint8_t,   8)
00129 LF_FUNCS(uint16_t, 10)
00130 
00131 #if ARCH_X86_32 && HAVE_YASM
00132 LF_FUNC(v8, luma, 8, mmx2)
00133 static void ff_deblock_v_luma_8_mmx2(uint8_t *pix, int stride, int alpha,
00134                                      int beta, int8_t *tc0)
00135 {
00136     if ((tc0[0] & tc0[1]) >= 0)
00137         ff_deblock_v8_luma_8_mmx2(pix + 0, stride, alpha, beta, tc0);
00138     if ((tc0[2] & tc0[3]) >= 0)
00139         ff_deblock_v8_luma_8_mmx2(pix + 8, stride, alpha, beta, tc0 + 2);
00140 }
00141 
00142 LF_IFUNC(v8, luma_intra, 8, mmx2)
00143 static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride,
00144                                            int alpha, int beta)
00145 {
00146     ff_deblock_v8_luma_intra_8_mmx2(pix + 0, stride, alpha, beta);
00147     ff_deblock_v8_luma_intra_8_mmx2(pix + 8, stride, alpha, beta);
00148 }
00149 #endif /* ARCH_X86_32 */
00150 
00151 LF_FUNC(v,  luma,       10, mmx2)
00152 LF_IFUNC(v, luma_intra, 10, mmx2)
00153 
00154 /***********************************/
00155 /* weighted prediction */
00156 
00157 #define H264_WEIGHT(W, OPT)                                             \
00158 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride,         \
00159                                       int height, int log2_denom,       \
00160                                       int weight, int offset);
00161 
00162 #define H264_BIWEIGHT(W, OPT)                                           \
00163 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
00164                                         int stride, int height,         \
00165                                         int log2_denom, int weightd,    \
00166                                         int weights, int offset);
00167 
00168 #define H264_BIWEIGHT_MMX(W)                    \
00169     H264_WEIGHT(W, mmx2)                        \
00170     H264_BIWEIGHT(W, mmx2)
00171 
00172 #define H264_BIWEIGHT_MMX_SSE(W)                \
00173     H264_BIWEIGHT_MMX(W)                        \
00174     H264_WEIGHT(W, sse2)                        \
00175     H264_BIWEIGHT(W, sse2)                      \
00176     H264_BIWEIGHT(W, ssse3)
00177 
00178 H264_BIWEIGHT_MMX_SSE(16)
00179 H264_BIWEIGHT_MMX_SSE(8)
00180 H264_BIWEIGHT_MMX(4)
00181 
00182 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
00183 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
00184                                                     int stride,         \
00185                                                     int height,         \
00186                                                     int log2_denom,     \
00187                                                     int weight,         \
00188                                                     int offset);
00189 
00190 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
00191 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
00192                                                       uint8_t *src,     \
00193                                                       int stride,       \
00194                                                       int height,       \
00195                                                       int log2_denom,   \
00196                                                       int weightd,      \
00197                                                       int weights,      \
00198                                                       int offset);
00199 
00200 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
00201     H264_WEIGHT_10(W, DEPTH, sse2)              \
00202     H264_WEIGHT_10(W, DEPTH, sse4)              \
00203     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
00204     H264_BIWEIGHT_10(W, DEPTH, sse4)
00205 
00206 H264_BIWEIGHT_10_SSE(16, 10)
00207 H264_BIWEIGHT_10_SSE(8,  10)
00208 H264_BIWEIGHT_10_SSE(4,  10)
00209 
00210 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
00211                          const int chroma_format_idc)
00212 {
00213 #if HAVE_YASM
00214     int mm_flags = av_get_cpu_flags();
00215 
00216     if (chroma_format_idc == 1 && EXTERNAL_MMXEXT(mm_flags))
00217         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
00218 
00219     if (bit_depth == 8) {
00220         if (EXTERNAL_MMX(mm_flags)) {
00221             c->h264_idct_dc_add   =
00222             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
00223             c->h264_idct8_dc_add  =
00224             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
00225 
00226             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
00227             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
00228             if (chroma_format_idc == 1)
00229                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
00230             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
00231             if (mm_flags & AV_CPU_FLAG_CMOV)
00232                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
00233 
00234             if (EXTERNAL_MMXEXT(mm_flags)) {
00235                 c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmx2;
00236                 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmx2;
00237                 c->h264_idct_add16   = ff_h264_idct_add16_8_mmx2;
00238                 c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmx2;
00239                 if (chroma_format_idc == 1)
00240                     c->h264_idct_add8 = ff_h264_idct_add8_8_mmx2;
00241                 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx2;
00242 
00243                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmx2;
00244                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmx2;
00245                 if (chroma_format_idc == 1) {
00246                     c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmx2;
00247                     c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmx2;
00248                 }
00249 #if ARCH_X86_32
00250                 c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_mmx2;
00251                 c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmx2;
00252                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
00253                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
00254 #endif /* ARCH_X86_32 */
00255                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmx2;
00256                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmx2;
00257                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmx2;
00258 
00259                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmx2;
00260                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmx2;
00261                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmx2;
00262 
00263                 if (EXTERNAL_SSE2(mm_flags)) {
00264                     c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
00265 
00266                     c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
00267                     c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
00268                     if (chroma_format_idc == 1)
00269                         c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
00270                     c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
00271                     c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
00272 
00273                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
00274                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
00275 
00276                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
00277                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
00278 
00279 #if HAVE_ALIGNED_STACK
00280                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
00281                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
00282                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
00283                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
00284 #endif /* HAVE_ALIGNED_STACK */
00285                 }
00286                 if (EXTERNAL_SSSE3(mm_flags)) {
00287                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
00288                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
00289                 }
00290                 if (EXTERNAL_AVX(mm_flags) && HAVE_ALIGNED_STACK) {
00291                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
00292                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
00293                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
00294                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
00295                 }
00296             }
00297         }
00298     } else if (bit_depth == 10) {
00299         if (EXTERNAL_MMX(mm_flags)) {
00300             if (EXTERNAL_MMXEXT(mm_flags)) {
00301 #if ARCH_X86_32
00302                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmx2;
00303                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmx2;
00304                 c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmx2;
00305                 c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmx2;
00306                 c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmx2;
00307                 c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmx2;
00308 #endif /* ARCH_X86_32 */
00309                 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmx2;
00310                 if (EXTERNAL_SSE2(mm_flags)) {
00311                     c->h264_idct_add     = ff_h264_idct_add_10_sse2;
00312                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
00313 
00314                     c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
00315                     if (chroma_format_idc == 1)
00316                         c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
00317                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
00318 #if HAVE_ALIGNED_STACK
00319                     c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
00320                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
00321 #endif /* HAVE_ALIGNED_STACK */
00322 
00323                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
00324                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
00325                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
00326 
00327                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
00328                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
00329                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
00330 
00331                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
00332                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
00333 #if HAVE_ALIGNED_STACK
00334                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
00335                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
00336                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
00337                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
00338 #endif /* HAVE_ALIGNED_STACK */
00339                 }
00340                 if (EXTERNAL_SSE4(mm_flags)) {
00341                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
00342                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
00343                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
00344 
00345                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
00346                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
00347                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
00348                 }
00349                 if (EXTERNAL_AVX(mm_flags)) {
00350                     c->h264_idct_dc_add  =
00351                     c->h264_idct_add     = ff_h264_idct_add_10_avx;
00352                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
00353 
00354                     c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
00355                     if (chroma_format_idc == 1)
00356                         c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
00357                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
00358 #if HAVE_ALIGNED_STACK
00359                     c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
00360                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
00361 #endif /* HAVE_ALIGNED_STACK */
00362 
00363                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
00364                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
00365 #if HAVE_ALIGNED_STACK
00366                     c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
00367                     c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
00368                     c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
00369                     c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
00370 #endif /* HAVE_ALIGNED_STACK */
00371                 }
00372             }
00373         }
00374     }
00375 #endif
00376 }
libavcodec/x86/h264dsp_init.c
