avcodec/h264: mmx2, sse2, avx 10-bit h chroma deblock/loop filter
[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 IDCT_ADD_REP_FUNC2(, 8_422, 8, mmx)
82
83 IDCT_ADD_REP_FUNC2(, 8_422, 10, sse2)
84 IDCT_ADD_REP_FUNC2(, 8_422, 10, avx)
85
86 void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
87 void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
88
89 /***********************************/
90 /* deblocking */
91
92 void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40],
93                                          int8_t ref[2][40],
94                                          int16_t mv[2][40][2],
95                                          int bidir, int edges, int step,
96                                          int mask_mv0, int mask_mv1, int field);
97
98 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
99 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
100                                                                int stride,    \
101                                                                int alpha,     \
102                                                                int beta,      \
103                                                                int8_t *tc0);
104 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
105 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
106                                                                int stride,    \
107                                                                int alpha,     \
108                                                                int beta);
109
110 #define LF_FUNCS(type, depth)                   \
111 LF_FUNC(h,  chroma,          depth, mmxext)     \
112 LF_IFUNC(h, chroma_intra,    depth, mmxext)     \
113 LF_FUNC(h,  chroma422,       depth, mmxext)     \
114 LF_IFUNC(h, chroma422_intra, depth, mmxext)     \
115 LF_FUNC(v,  chroma,          depth, mmxext)     \
116 LF_IFUNC(v, chroma_intra,    depth, mmxext)     \
117 LF_FUNC(h,  luma,            depth, mmxext)     \
118 LF_IFUNC(h, luma_intra,      depth, mmxext)     \
119 LF_FUNC(h,  luma,            depth, sse2)       \
120 LF_IFUNC(h, luma_intra,      depth, sse2)       \
121 LF_FUNC(v,  luma,            depth, sse2)       \
122 LF_IFUNC(v, luma_intra,      depth, sse2)       \
123 LF_FUNC(h,  chroma,          depth, sse2)       \
124 LF_IFUNC(h, chroma_intra,    depth, sse2)       \
125 LF_FUNC(h,  chroma422,       depth, sse2)       \
126 LF_IFUNC(h, chroma422_intra, depth, sse2)       \
127 LF_FUNC(v,  chroma,          depth, sse2)       \
128 LF_IFUNC(v, chroma_intra,    depth, sse2)       \
129 LF_FUNC(h,  luma,            depth, avx)        \
130 LF_IFUNC(h, luma_intra,      depth, avx)        \
131 LF_FUNC(v,  luma,            depth, avx)        \
132 LF_IFUNC(v, luma_intra,      depth, avx)        \
133 LF_FUNC(h,  chroma,          depth, avx)        \
134 LF_IFUNC(h, chroma_intra,    depth, avx)        \
135 LF_FUNC(h,  chroma422,       depth, avx)        \
136 LF_IFUNC(h, chroma422_intra, depth, avx)        \
137 LF_FUNC(v,  chroma,          depth, avx)        \
138 LF_IFUNC(v, chroma_intra,    depth, avx)
139
140 LF_FUNCS(uint8_t,   8)
141 LF_FUNCS(uint16_t, 10)
142
143 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
144 LF_FUNC(v8, luma, 8, mmxext)
145 static void deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha,
146                                     int beta, int8_t *tc0)
147 {
148     if ((tc0[0] & tc0[1]) >= 0)
149         ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0);
150     if ((tc0[2] & tc0[3]) >= 0)
151         ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2);
152 }
153 LF_IFUNC(v8, luma_intra, 8, mmxext)
154 static void deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride,
155                                           int alpha, int beta)
156 {
157     ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta);
158     ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta);
159 }
160 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
161
162 LF_FUNC(v,  luma,       10, mmxext)
163 LF_IFUNC(v, luma_intra, 10, mmxext)
164
165 /***********************************/
166 /* weighted prediction */
167
168 #define H264_WEIGHT(W, OPT)                                             \
169 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, ptrdiff_t stride,   \
170                                       int height, int log2_denom,       \
171                                       int weight, int offset);
172
173 #define H264_BIWEIGHT(W, OPT)                                           \
174 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
175                                         ptrdiff_t stride, int height,   \
176                                         int log2_denom, int weightd,    \
177                                         int weights, int offset);
178
179 #define H264_BIWEIGHT_MMX(W)                    \
180     H264_WEIGHT(W, mmxext)                      \
181     H264_BIWEIGHT(W, mmxext)
182
183 #define H264_BIWEIGHT_MMX_SSE(W)                \
184     H264_BIWEIGHT_MMX(W)                        \
185     H264_WEIGHT(W, sse2)                        \
186     H264_BIWEIGHT(W, sse2)                      \
187     H264_BIWEIGHT(W, ssse3)
188
189 H264_BIWEIGHT_MMX_SSE(16)
190 H264_BIWEIGHT_MMX_SSE(8)
191 H264_BIWEIGHT_MMX(4)
192
193 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
194 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
195                                                     ptrdiff_t stride,   \
196                                                     int height,         \
197                                                     int log2_denom,     \
198                                                     int weight,         \
199                                                     int offset);
200
201 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
202 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
203                                                       uint8_t *src,     \
204                                                       ptrdiff_t stride, \
205                                                       int height,       \
206                                                       int log2_denom,   \
207                                                       int weightd,      \
208                                                       int weights,      \
209                                                       int offset);
210
211 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
212     H264_WEIGHT_10(W, DEPTH, sse2)              \
213     H264_WEIGHT_10(W, DEPTH, sse4)              \
214     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
215     H264_BIWEIGHT_10(W, DEPTH, sse4)
216
217 H264_BIWEIGHT_10_SSE(16, 10)
218 H264_BIWEIGHT_10_SSE(8,  10)
219 H264_BIWEIGHT_10_SSE(4,  10)
220
221 av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
222                                  const int chroma_format_idc)
223 {
224 #if HAVE_YASM
225     int cpu_flags = av_get_cpu_flags();
226
227     if (EXTERNAL_MMXEXT(cpu_flags) && chroma_format_idc <= 1)
228         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
229
230     if (bit_depth == 8) {
231         if (EXTERNAL_MMX(cpu_flags)) {
232             c->h264_idct_dc_add   =
233             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
234             c->h264_idct8_dc_add  =
235             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
236
237             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
238             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
239             if (chroma_format_idc <= 1) {
240                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
241             } else {
242                 c->h264_idct_add8 = ff_h264_idct_add8_422_8_mmx;
243             }
244             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
245             if (cpu_flags & AV_CPU_FLAG_CMOV)
246                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
247         }
248         if (EXTERNAL_MMXEXT(cpu_flags)) {
249             c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmxext;
250             c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
251             c->h264_idct_add16   = ff_h264_idct_add16_8_mmxext;
252             c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmxext;
253             if (chroma_format_idc <= 1)
254                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
255             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
256
257             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmxext;
258             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
259             if (chroma_format_idc <= 1) {
260                 c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmxext;
261                 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
262             } else {
263                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma422_8_mmxext;
264                 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma422_intra_8_mmxext;
265             }
266 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
267             c->h264_v_loop_filter_luma       = deblock_v_luma_8_mmxext;
268             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmxext;
269             c->h264_v_loop_filter_luma_intra = deblock_v_luma_intra_8_mmxext;
270             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
271 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
272             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
273             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
274             c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
275
276             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
277             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
278             c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
279         }
280         if (EXTERNAL_SSE2(cpu_flags)) {
281             c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
282
283             c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
284             c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
285             if (chroma_format_idc <= 1)
286                 c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
287             c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
288             c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
289
290             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
291             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
292
293             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
294             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
295
296             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
297             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
298             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
299             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
300         }
301         if (EXTERNAL_SSSE3(cpu_flags)) {
302             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
303             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
304         }
305         if (EXTERNAL_AVX(cpu_flags)) {
306             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
307             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
308             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
309             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
310         }
311     } else if (bit_depth == 10) {
312         if (EXTERNAL_MMXEXT(cpu_flags)) {
313 #if ARCH_X86_32
314             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmxext;
315             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
316             if (chroma_format_idc <= 1) {
317                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_10_mmxext;
318             }
319             c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmxext;
320             c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmxext;
321             c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmxext;
322             c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmxext;
323 #endif /* ARCH_X86_32 */
324             c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
325         }
326         if (EXTERNAL_SSE2(cpu_flags)) {
327             c->h264_idct_add     = ff_h264_idct_add_10_sse2;
328             c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
329
330             c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
331             if (chroma_format_idc <= 1) {
332                 c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
333             } else {
334                 c->h264_idct_add8 = ff_h264_idct_add8_422_10_sse2;
335             }
336             c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
337 #if HAVE_ALIGNED_STACK
338             c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
339             c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
340 #endif /* HAVE_ALIGNED_STACK */
341
342             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
343             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
344             c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
345
346             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
347             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
348             c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
349
350             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
351             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
352             if (chroma_format_idc <= 1) {
353                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_10_sse2;
354             }
355 #if HAVE_ALIGNED_STACK
356             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
357             c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
358             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
359             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
360 #endif /* HAVE_ALIGNED_STACK */
361         }
362         if (EXTERNAL_SSE4(cpu_flags)) {
363             c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
364             c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
365             c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
366
367             c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
368             c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
369             c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
370         }
371         if (EXTERNAL_AVX(cpu_flags)) {
372             c->h264_idct_dc_add  =
373             c->h264_idct_add     = ff_h264_idct_add_10_avx;
374             c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
375
376             c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
377             if (chroma_format_idc <= 1) {
378                 c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
379             } else {
380                 c->h264_idct_add8 = ff_h264_idct_add8_422_10_avx;
381             }
382             c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
383 #if HAVE_ALIGNED_STACK
384             c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
385             c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
386 #endif /* HAVE_ALIGNED_STACK */
387
388             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
389             c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
390             if (chroma_format_idc <= 1) {
391                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_10_avx;
392             }
393 #if HAVE_ALIGNED_STACK
394             c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
395             c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
396             c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
397             c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
398 #endif /* HAVE_ALIGNED_STACK */
399         }
400     }
401 #endif
402 }