Merge commit '21d8f4da91673051ad7055cf63d989488a56368e'
[ffmpeg.git] / libavcodec / x86 / h264dsp_init.c
1 /*
2  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg 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  * FFmpeg 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 FFmpeg; 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/attributes.h"
22 #include "libavutil/cpu.h"
23 #include "libavutil/x86/asm.h"
24 #include "libavutil/x86/cpu.h"
25 #include "libavcodec/h264dsp.h"
26
27 /***********************************/
28 /* IDCT */
29 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT)                                  \
30 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst,    \
31                                                        int16_t *block,  \
32                                                        int stride);
33
34 IDCT_ADD_FUNC(, 8, mmx)
35 IDCT_ADD_FUNC(, 10, sse2)
36 IDCT_ADD_FUNC(_dc, 8, mmxext)
37 IDCT_ADD_FUNC(_dc, 10, mmxext)
38 IDCT_ADD_FUNC(8_dc, 8, mmxext)
39 IDCT_ADD_FUNC(8_dc, 10, sse2)
40 IDCT_ADD_FUNC(8, 8, mmx)
41 IDCT_ADD_FUNC(8, 8, sse2)
42 IDCT_ADD_FUNC(8, 10, sse2)
43 IDCT_ADD_FUNC(, 10, avx)
44 IDCT_ADD_FUNC(8_dc, 10, avx)
45 IDCT_ADD_FUNC(8, 10, avx)
46
47
48 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT)                         \
49 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT       \
50     (uint8_t *dst, const int *block_offset,                             \
51      int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
52
53 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
54 IDCT_ADD_REP_FUNC(8, 4, 8, mmxext)
55 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
56 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
57 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
58 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
59 IDCT_ADD_REP_FUNC(, 16, 8, mmxext)
60 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
61 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
62 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
63 IDCT_ADD_REP_FUNC(, 16intra, 8, mmxext)
64 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
65 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
66 IDCT_ADD_REP_FUNC(, 16, 10, avx)
67 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
68
69
70 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT)                      \
71 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT     \
72     (uint8_t **dst, const int *block_offset,                          \
73      int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
74
75 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
76 IDCT_ADD_REP_FUNC2(, 8, 8, mmxext)
77 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
78 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
79 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
80
81 void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
82 void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
83
84 /***********************************/
85 /* deblocking */
86
87 void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40],
88                                          int8_t ref[2][40],
89                                          int16_t mv[2][40][2],
90                                          int bidir, int edges, int step,
91                                          int mask_mv0, int mask_mv1, int field);
92
93 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
94 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
95                                                                int stride,    \
96                                                                int alpha,     \
97                                                                int beta,      \
98                                                                int8_t *tc0);
99 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
100 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
101                                                                int stride,    \
102                                                                int alpha,     \
103                                                                int beta);
104
105 #define LF_FUNCS(type, depth)                   \
106 LF_FUNC(h,  chroma,       depth, mmxext)        \
107 LF_IFUNC(h, chroma_intra, depth, mmxext)        \
108 LF_FUNC(v,  chroma,       depth, mmxext)        \
109 LF_IFUNC(v, chroma_intra, depth, mmxext)        \
110 LF_FUNC(h,  luma,         depth, mmxext)        \
111 LF_IFUNC(h, luma_intra,   depth, mmxext)        \
112 LF_FUNC(h,  luma,         depth, sse2)          \
113 LF_IFUNC(h, luma_intra,   depth, sse2)          \
114 LF_FUNC(v,  luma,         depth, sse2)          \
115 LF_IFUNC(v, luma_intra,   depth, sse2)          \
116 LF_FUNC(h,  chroma,       depth, sse2)          \
117 LF_IFUNC(h, chroma_intra, depth, sse2)          \
118 LF_FUNC(v,  chroma,       depth, sse2)          \
119 LF_IFUNC(v, chroma_intra, depth, sse2)          \
120 LF_FUNC(h,  luma,         depth, avx)           \
121 LF_IFUNC(h, luma_intra,   depth, avx)           \
122 LF_FUNC(v,  luma,         depth, avx)           \
123 LF_IFUNC(v, luma_intra,   depth, avx)           \
124 LF_FUNC(h,  chroma,       depth, avx)           \
125 LF_IFUNC(h, chroma_intra, depth, avx)           \
126 LF_FUNC(v,  chroma,       depth, avx)           \
127 LF_IFUNC(v, chroma_intra, depth, avx)
128
129 LF_FUNCS(uint8_t,   8)
130 LF_FUNCS(uint16_t, 10)
131
132 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
133 LF_FUNC(v8, luma, 8, mmxext)
134 static void deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha,
135                                     int beta, int8_t *tc0)
136 {
137     if ((tc0[0] & tc0[1]) >= 0)
138         ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0);
139     if ((tc0[2] & tc0[3]) >= 0)
140         ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2);
141 }
142 LF_IFUNC(v8, luma_intra, 8, mmxext)
143 static void deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride,
144                                           int alpha, int beta)
145 {
146     ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta);
147     ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta);
148 }
149 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
150
151 LF_FUNC(v,  luma,       10, mmxext)
152 LF_IFUNC(v, luma_intra, 10, mmxext)
153
154 /***********************************/
155 /* weighted prediction */
156
157 #define H264_WEIGHT(W, OPT)                                             \
158 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride,         \
159                                       int height, int log2_denom,       \
160                                       int weight, int offset);
161
162 #define H264_BIWEIGHT(W, OPT)                                           \
163 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
164                                         int stride, int height,         \
165                                         int log2_denom, int weightd,    \
166                                         int weights, int offset);
167
168 #define H264_BIWEIGHT_MMX(W)                    \
169     H264_WEIGHT(W, mmxext)                      \
170     H264_BIWEIGHT(W, mmxext)
171
172 #define H264_BIWEIGHT_MMX_SSE(W)                \
173     H264_BIWEIGHT_MMX(W)                        \
174     H264_WEIGHT(W, sse2)                        \
175     H264_BIWEIGHT(W, sse2)                      \
176     H264_BIWEIGHT(W, ssse3)
177
178 H264_BIWEIGHT_MMX_SSE(16)
179 H264_BIWEIGHT_MMX_SSE(8)
180 H264_BIWEIGHT_MMX(4)
181
182 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
183 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
184                                                     int stride,         \
185                                                     int height,         \
186                                                     int log2_denom,     \
187                                                     int weight,         \
188                                                     int offset);
189
190 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
191 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
192                                                       uint8_t *src,     \
193                                                       int stride,       \
194                                                       int height,       \
195                                                       int log2_denom,   \
196                                                       int weightd,      \
197                                                       int weights,      \
198                                                       int offset);
199
200 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
201     H264_WEIGHT_10(W, DEPTH, sse2)              \
202     H264_WEIGHT_10(W, DEPTH, sse4)              \
203     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
204     H264_BIWEIGHT_10(W, DEPTH, sse4)
205
206 H264_BIWEIGHT_10_SSE(16, 10)
207 H264_BIWEIGHT_10_SSE(8,  10)
208 H264_BIWEIGHT_10_SSE(4,  10)
209
210 av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
211                                  const int chroma_format_idc)
212 {
213 #if HAVE_YASM
214     int cpu_flags = av_get_cpu_flags();
215
216     if (chroma_format_idc <= 1 && EXTERNAL_MMXEXT(cpu_flags))
217         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
218
219     if (bit_depth == 8) {
220         if (EXTERNAL_MMX(cpu_flags)) {
221             c->h264_idct_dc_add   =
222             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
223             c->h264_idct8_dc_add  =
224             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
225
226             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
227             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
228             if (chroma_format_idc <= 1)
229                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
230             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
231             if (cpu_flags & AV_CPU_FLAG_CMOV)
232                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
233         }
234         if (EXTERNAL_MMXEXT(cpu_flags)) {
235             c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmxext;
236             c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
237             c->h264_idct_add16   = ff_h264_idct_add16_8_mmxext;
238             c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmxext;
239             if (chroma_format_idc <= 1)
240                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
241             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
242
243             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmxext;
244             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
245             if (chroma_format_idc <= 1) {
246                 c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmxext;
247                 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
248             }
249 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
250             c->h264_v_loop_filter_luma       = deblock_v_luma_8_mmxext;
251             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmxext;
252             c->h264_v_loop_filter_luma_intra = deblock_v_luma_intra_8_mmxext;
253             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
254 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
255             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
256             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
257             c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
258
259             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
260             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
261             c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
262         }
263         if (EXTERNAL_SSE2(cpu_flags)) {
264             c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
265
266             c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
267             c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
268             if (chroma_format_idc <= 1)
269                 c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
270             c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
271             c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
272
273             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
274             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
275
276             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
277             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
278
279             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
280             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
281             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
282             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
283         }
284         if (EXTERNAL_SSSE3(cpu_flags)) {
285             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
286             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
287         }
288         if (EXTERNAL_AVX(cpu_flags)) {
289             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
290             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
291             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
292             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
293         }
294     } else if (bit_depth == 10) {
295         if (EXTERNAL_MMXEXT(cpu_flags)) {
296 #if ARCH_X86_32
297             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmxext;
298             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
299             c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmxext;
300             c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmxext;
301             c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmxext;
302             c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmxext;
303 #endif /* ARCH_X86_32 */
304             c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
305         }
306         if (EXTERNAL_SSE2(cpu_flags)) {
307             c->h264_idct_add     = ff_h264_idct_add_10_sse2;
308             c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
309
310             c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
311             if (chroma_format_idc <= 1)
312                 c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
313             c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
314 #if HAVE_ALIGNED_STACK
315             c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
316             c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
317 #endif /* HAVE_ALIGNED_STACK */
318
319             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
320             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
321             c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
322
323             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
324             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
325             c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
326
327             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
328             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
329 #if HAVE_ALIGNED_STACK
330             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
331             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
332             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
333             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
334 #endif /* HAVE_ALIGNED_STACK */
335         }
336         if (EXTERNAL_SSE4(cpu_flags)) {
337             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
338             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
339             c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
340
341             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
342             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
343             c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
344         }
345         if (EXTERNAL_AVX(cpu_flags)) {
346             c->h264_idct_dc_add  =
347             c->h264_idct_add     = ff_h264_idct_add_10_avx;
348             c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
349
350             c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
351             if (chroma_format_idc <= 1)
352                 c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
353             c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
354 #if HAVE_ALIGNED_STACK
355             c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
356             c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
357 #endif /* HAVE_ALIGNED_STACK */
358
359             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
360             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
361 #if HAVE_ALIGNED_STACK
362             c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
363             c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
364             c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
365             c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
366 #endif /* HAVE_ALIGNED_STACK */
367         }
368     }
369 #endif
370 }