7be78a82074a528bccf72aebc9308734c3e733b3
[ffmpeg.git] / libavcodec / x86 / h264dsp_init.c
1 /*
2  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3  *
4  * This file is part of Libav.
5  *
6  * Libav is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * Libav is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with Libav; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 #include "libavutil/cpu.h"
22 #include "libavutil/x86/asm.h"
23 #include "libavcodec/h264dsp.h"
24 #include "dsputil_mmx.h"
25
26 /***********************************/
27 /* IDCT */
28 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT)                                  \
29 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst,    \
30                                                        int16_t *block,  \
31                                                        int stride);
32
33 IDCT_ADD_FUNC(, 8, mmx)
34 IDCT_ADD_FUNC(, 10, sse2)
35 IDCT_ADD_FUNC(_dc, 8, mmx2)
36 IDCT_ADD_FUNC(_dc, 10, mmx2)
37 IDCT_ADD_FUNC(8_dc, 8, mmx2)
38 IDCT_ADD_FUNC(8_dc, 10, sse2)
39 IDCT_ADD_FUNC(8, 8, mmx)
40 IDCT_ADD_FUNC(8, 8, sse2)
41 IDCT_ADD_FUNC(8, 10, sse2)
42 #if HAVE_AVX_EXTERNAL
43 IDCT_ADD_FUNC(, 10, avx)
44 IDCT_ADD_FUNC(8_dc, 10, avx)
45 IDCT_ADD_FUNC(8, 10, avx)
46 #endif
47
48
49 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT)                         \
50 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT       \
51     (uint8_t *dst, const int *block_offset,                             \
52      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
53
54 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
55 IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
56 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
57 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
58 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
59 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
60 IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
61 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
62 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
63 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
64 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
65 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
66 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
67 #if HAVE_AVX_EXTERNAL
68 IDCT_ADD_REP_FUNC(, 16, 10, avx)
69 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
70 #endif
71
72
73 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT)                      \
74 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT     \
75     (uint8_t **dst, const int *block_offset,                          \
76      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
77
78 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
79 IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
80 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
81 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
82 #if HAVE_AVX_EXTERNAL
83 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
84 #endif
85
86 void ff_h264_luma_dc_dequant_idct_mmx(DCTELEM *output, DCTELEM *input, int qmul);
87 void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
88
89 /***********************************/
90 /* deblocking */
91
92 void ff_h264_loop_filter_strength_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
93                                        int8_t ref[2][40], int16_t mv[2][40][2],
94                                        int bidir, int edges, int step,
95                                        int mask_mv0, int mask_mv1, int field);
96
97 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
98 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
99                                                                int stride,    \
100                                                                int alpha,     \
101                                                                int beta,      \
102                                                                int8_t *tc0);
103 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
104 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
105                                                                int stride,    \
106                                                                int alpha,     \
107                                                                int beta);
108
109 #define LF_FUNCS(type, depth)                   \
110 LF_FUNC(h,  chroma,       depth, mmx2)          \
111 LF_IFUNC(h, chroma_intra, depth, mmx2)          \
112 LF_FUNC(v,  chroma,       depth, mmx2)          \
113 LF_IFUNC(v, chroma_intra, depth, mmx2)          \
114 LF_FUNC(h,  luma,         depth, mmx2)          \
115 LF_IFUNC(h, luma_intra,   depth, mmx2)          \
116 LF_FUNC(h,  luma,         depth, sse2)          \
117 LF_IFUNC(h, luma_intra,   depth, sse2)          \
118 LF_FUNC(v,  luma,         depth, sse2)          \
119 LF_IFUNC(v, luma_intra,   depth, sse2)          \
120 LF_FUNC(h,  chroma,       depth, sse2)          \
121 LF_IFUNC(h, chroma_intra, depth, sse2)          \
122 LF_FUNC(v,  chroma,       depth, sse2)          \
123 LF_IFUNC(v, chroma_intra, depth, sse2)          \
124 LF_FUNC(h,  luma,         depth, avx)           \
125 LF_IFUNC(h, luma_intra,   depth, avx)           \
126 LF_FUNC(v,  luma,         depth, avx)           \
127 LF_IFUNC(v, luma_intra,   depth, avx)           \
128 LF_FUNC(h,  chroma,       depth, avx)           \
129 LF_IFUNC(h, chroma_intra, depth, avx)           \
130 LF_FUNC(v,  chroma,       depth, avx)           \
131 LF_IFUNC(v, chroma_intra, depth, avx)
132
133 LF_FUNCS(uint8_t,   8)
134 LF_FUNCS(uint16_t, 10)
135
136 #if ARCH_X86_32
137 LF_FUNC(v8, luma, 8, mmx2)
138 static void ff_deblock_v_luma_8_mmx2(uint8_t *pix, int stride, int alpha,
139                                      int beta, int8_t *tc0)
140 {
141     if ((tc0[0] & tc0[1]) >= 0)
142         ff_deblock_v8_luma_8_mmx2(pix + 0, stride, alpha, beta, tc0);
143     if ((tc0[2] & tc0[3]) >= 0)
144         ff_deblock_v8_luma_8_mmx2(pix + 8, stride, alpha, beta, tc0 + 2);
145 }
146
147 LF_IFUNC(v8, luma_intra, 8, mmx2)
148 static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride,
149                                            int alpha, int beta)
150 {
151     ff_deblock_v8_luma_intra_8_mmx2(pix + 0, stride, alpha, beta);
152     ff_deblock_v8_luma_intra_8_mmx2(pix + 8, stride, alpha, beta);
153 }
154 #endif /* ARCH_X86_32 */
155
156 LF_FUNC(v,  luma,       10, mmx2)
157 LF_IFUNC(v, luma_intra, 10, mmx2)
158
159 /***********************************/
160 /* weighted prediction */
161
162 #define H264_WEIGHT(W, OPT)                                             \
163 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride,         \
164                                       int height, int log2_denom,       \
165                                       int weight, int offset);
166
167 #define H264_BIWEIGHT(W, OPT)                                           \
168 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
169                                         int stride, int height,         \
170                                         int log2_denom, int weightd,    \
171                                         int weights, int offset);
172
173 #define H264_BIWEIGHT_MMX(W)                    \
174     H264_WEIGHT(W, mmx2)                        \
175     H264_BIWEIGHT(W, mmx2)
176
177 #define H264_BIWEIGHT_MMX_SSE(W)                \
178     H264_BIWEIGHT_MMX(W)                        \
179     H264_WEIGHT(W, sse2)                        \
180     H264_BIWEIGHT(W, sse2)                      \
181     H264_BIWEIGHT(W, ssse3)
182
183 H264_BIWEIGHT_MMX_SSE(16)
184 H264_BIWEIGHT_MMX_SSE(8)
185 H264_BIWEIGHT_MMX(4)
186
187 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
188 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
189                                                     int stride,         \
190                                                     int height,         \
191                                                     int log2_denom,     \
192                                                     int weight,         \
193                                                     int offset);
194
195 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
196 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
197                                                       uint8_t *src,     \
198                                                       int stride,       \
199                                                       int height,       \
200                                                       int log2_denom,   \
201                                                       int weightd,      \
202                                                       int weights,      \
203                                                       int offset);
204
205 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
206     H264_WEIGHT_10(W, DEPTH, sse2)              \
207     H264_WEIGHT_10(W, DEPTH, sse4)              \
208     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
209     H264_BIWEIGHT_10(W, DEPTH, sse4)
210
211 H264_BIWEIGHT_10_SSE(16, 10)
212 H264_BIWEIGHT_10_SSE(8,  10)
213 H264_BIWEIGHT_10_SSE(4,  10)
214
215 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
216                          const int chroma_format_idc)
217 {
218 #if HAVE_YASM
219     int mm_flags = av_get_cpu_flags();
220
221     if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMXEXT)
222         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
223
224     if (bit_depth == 8) {
225         if (mm_flags & AV_CPU_FLAG_MMX) {
226             c->h264_idct_dc_add   =
227             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
228             c->h264_idct8_dc_add  =
229             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
230
231             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
232             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
233             if (chroma_format_idc == 1)
234                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
235             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
236             if (mm_flags & AV_CPU_FLAG_CMOV)
237                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
238
239             if (mm_flags & AV_CPU_FLAG_MMXEXT) {
240                 c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmx2;
241                 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmx2;
242                 c->h264_idct_add16   = ff_h264_idct_add16_8_mmx2;
243                 c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmx2;
244                 if (chroma_format_idc == 1)
245                     c->h264_idct_add8 = ff_h264_idct_add8_8_mmx2;
246                 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx2;
247
248                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmx2;
249                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmx2;
250                 if (chroma_format_idc == 1) {
251                     c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmx2;
252                     c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmx2;
253                 }
254 #if ARCH_X86_32
255                 c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_mmx2;
256                 c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmx2;
257                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
258                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
259 #endif /* ARCH_X86_32 */
260                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmx2;
261                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmx2;
262                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmx2;
263
264                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmx2;
265                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmx2;
266                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmx2;
267
268                 if (mm_flags & AV_CPU_FLAG_SSE2) {
269                     c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
270
271                     c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
272                     c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
273                     if (chroma_format_idc == 1)
274                         c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
275                     c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
276                     c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
277
278                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
279                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
280
281                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
282                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
283
284 #if HAVE_ALIGNED_STACK
285                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
286                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
287                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
288                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
289 #endif /* HAVE_ALIGNED_STACK */
290                 }
291                 if (mm_flags & AV_CPU_FLAG_SSSE3) {
292                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
293                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
294                 }
295                 if (mm_flags & AV_CPU_FLAG_AVX) {
296 #if HAVE_ALIGNED_STACK
297                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
298                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
299                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
300                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
301 #endif /* HAVE_ALIGNED_STACK */
302                 }
303             }
304         }
305     } else if (bit_depth == 10) {
306         if (mm_flags & AV_CPU_FLAG_MMX) {
307             if (mm_flags & AV_CPU_FLAG_MMXEXT) {
308 #if ARCH_X86_32
309                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmx2;
310                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmx2;
311                 c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmx2;
312                 c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmx2;
313                 c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmx2;
314                 c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmx2;
315 #endif /* ARCH_X86_32 */
316                 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmx2;
317                 if (mm_flags & AV_CPU_FLAG_SSE2) {
318                     c->h264_idct_add     = ff_h264_idct_add_10_sse2;
319                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
320
321                     c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
322                     if (chroma_format_idc == 1)
323                         c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
324                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
325 #if HAVE_ALIGNED_STACK
326                     c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
327                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
328 #endif /* HAVE_ALIGNED_STACK */
329
330                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
331                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
332                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
333
334                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
335                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
336                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
337
338                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
339                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
340 #if HAVE_ALIGNED_STACK
341                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
342                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
343                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
344                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
345 #endif /* HAVE_ALIGNED_STACK */
346                 }
347                 if (mm_flags & AV_CPU_FLAG_SSE4) {
348                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
349                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
350                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
351
352                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
353                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
354                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
355                 }
356 #if HAVE_AVX_EXTERNAL
357                 if (mm_flags & AV_CPU_FLAG_AVX) {
358                     c->h264_idct_dc_add  =
359                     c->h264_idct_add     = ff_h264_idct_add_10_avx;
360                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
361
362                     c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
363                     if (chroma_format_idc == 1)
364                         c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
365                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
366 #if HAVE_ALIGNED_STACK
367                     c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
368                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
369 #endif /* HAVE_ALIGNED_STACK */
370
371                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
372                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
373 #if HAVE_ALIGNED_STACK
374                     c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
375                     c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
376                     c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
377                     c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
378 #endif /* HAVE_ALIGNED_STACK */
379                 }
380 #endif /* HAVE_AVX_EXTERNAL */
381             }
382         }
383     }
384 #endif /* HAVE_YASM */
385 }