x86: h264dsp: drop some unnecessary ifdefs around prototype declarations
[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 IDCT_ADD_FUNC(, 10, avx)
43 IDCT_ADD_FUNC(8_dc, 10, avx)
44 IDCT_ADD_FUNC(8, 10, avx)
45
46
47 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT)                         \
48 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT       \
49     (uint8_t *dst, const int *block_offset,                             \
50      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
51
52 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
53 IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
54 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
55 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
56 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
57 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
58 IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
59 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
60 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
61 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
62 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
63 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
64 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
65 IDCT_ADD_REP_FUNC(, 16, 10, avx)
66 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
67
68
69 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT)                      \
70 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT     \
71     (uint8_t **dst, const int *block_offset,                          \
72      DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
73
74 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
75 IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
76 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
77 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
78 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
79
80 void ff_h264_luma_dc_dequant_idct_mmx(DCTELEM *output, DCTELEM *input, int qmul);
81 void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
82
83 /***********************************/
84 /* deblocking */
85
86 void ff_h264_loop_filter_strength_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
87                                        int8_t ref[2][40], int16_t mv[2][40][2],
88                                        int bidir, int edges, int step,
89                                        int mask_mv0, int mask_mv1, int field);
90
91 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
92 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
93                                                                int stride,    \
94                                                                int alpha,     \
95                                                                int beta,      \
96                                                                int8_t *tc0);
97 #define LF_IFUNC(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
103 #define LF_FUNCS(type, depth)                   \
104 LF_FUNC(h,  chroma,       depth, mmx2)          \
105 LF_IFUNC(h, chroma_intra, depth, mmx2)          \
106 LF_FUNC(v,  chroma,       depth, mmx2)          \
107 LF_IFUNC(v, chroma_intra, depth, mmx2)          \
108 LF_FUNC(h,  luma,         depth, mmx2)          \
109 LF_IFUNC(h, luma_intra,   depth, mmx2)          \
110 LF_FUNC(h,  luma,         depth, sse2)          \
111 LF_IFUNC(h, luma_intra,   depth, sse2)          \
112 LF_FUNC(v,  luma,         depth, sse2)          \
113 LF_IFUNC(v, luma_intra,   depth, sse2)          \
114 LF_FUNC(h,  chroma,       depth, sse2)          \
115 LF_IFUNC(h, chroma_intra, depth, sse2)          \
116 LF_FUNC(v,  chroma,       depth, sse2)          \
117 LF_IFUNC(v, chroma_intra, depth, sse2)          \
118 LF_FUNC(h,  luma,         depth, avx)           \
119 LF_IFUNC(h, luma_intra,   depth, avx)           \
120 LF_FUNC(v,  luma,         depth, avx)           \
121 LF_IFUNC(v, luma_intra,   depth, avx)           \
122 LF_FUNC(h,  chroma,       depth, avx)           \
123 LF_IFUNC(h, chroma_intra, depth, avx)           \
124 LF_FUNC(v,  chroma,       depth, avx)           \
125 LF_IFUNC(v, chroma_intra, depth, avx)
126
127 LF_FUNCS(uint8_t,   8)
128 LF_FUNCS(uint16_t, 10)
129
130 #if ARCH_X86_32
131 LF_FUNC(v8, luma, 8, mmx2)
132 static void ff_deblock_v_luma_8_mmx2(uint8_t *pix, int stride, int alpha,
133                                      int beta, int8_t *tc0)
134 {
135     if ((tc0[0] & tc0[1]) >= 0)
136         ff_deblock_v8_luma_8_mmx2(pix + 0, stride, alpha, beta, tc0);
137     if ((tc0[2] & tc0[3]) >= 0)
138         ff_deblock_v8_luma_8_mmx2(pix + 8, stride, alpha, beta, tc0 + 2);
139 }
140
141 LF_IFUNC(v8, luma_intra, 8, mmx2)
142 static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride,
143                                            int alpha, int beta)
144 {
145     ff_deblock_v8_luma_intra_8_mmx2(pix + 0, stride, alpha, beta);
146     ff_deblock_v8_luma_intra_8_mmx2(pix + 8, stride, alpha, beta);
147 }
148 #endif /* ARCH_X86_32 */
149
150 LF_FUNC(v,  luma,       10, mmx2)
151 LF_IFUNC(v, luma_intra, 10, mmx2)
152
153 /***********************************/
154 /* weighted prediction */
155
156 #define H264_WEIGHT(W, OPT)                                             \
157 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride,         \
158                                       int height, int log2_denom,       \
159                                       int weight, int offset);
160
161 #define H264_BIWEIGHT(W, OPT)                                           \
162 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
163                                         int stride, int height,         \
164                                         int log2_denom, int weightd,    \
165                                         int weights, int offset);
166
167 #define H264_BIWEIGHT_MMX(W)                    \
168     H264_WEIGHT(W, mmx2)                        \
169     H264_BIWEIGHT(W, mmx2)
170
171 #define H264_BIWEIGHT_MMX_SSE(W)                \
172     H264_BIWEIGHT_MMX(W)                        \
173     H264_WEIGHT(W, sse2)                        \
174     H264_BIWEIGHT(W, sse2)                      \
175     H264_BIWEIGHT(W, ssse3)
176
177 H264_BIWEIGHT_MMX_SSE(16)
178 H264_BIWEIGHT_MMX_SSE(8)
179 H264_BIWEIGHT_MMX(4)
180
181 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
182 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
183                                                     int stride,         \
184                                                     int height,         \
185                                                     int log2_denom,     \
186                                                     int weight,         \
187                                                     int offset);
188
189 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
190 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
191                                                       uint8_t *src,     \
192                                                       int stride,       \
193                                                       int height,       \
194                                                       int log2_denom,   \
195                                                       int weightd,      \
196                                                       int weights,      \
197                                                       int offset);
198
199 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
200     H264_WEIGHT_10(W, DEPTH, sse2)              \
201     H264_WEIGHT_10(W, DEPTH, sse4)              \
202     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
203     H264_BIWEIGHT_10(W, DEPTH, sse4)
204
205 H264_BIWEIGHT_10_SSE(16, 10)
206 H264_BIWEIGHT_10_SSE(8,  10)
207 H264_BIWEIGHT_10_SSE(4,  10)
208
209 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
210                          const int chroma_format_idc)
211 {
212 #if HAVE_YASM
213     int mm_flags = av_get_cpu_flags();
214
215     if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMXEXT)
216         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
217
218     if (bit_depth == 8) {
219         if (mm_flags & AV_CPU_FLAG_MMX) {
220             c->h264_idct_dc_add   =
221             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
222             c->h264_idct8_dc_add  =
223             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
224
225             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
226             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
227             if (chroma_format_idc == 1)
228                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
229             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
230             if (mm_flags & AV_CPU_FLAG_CMOV)
231                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
232
233             if (mm_flags & AV_CPU_FLAG_MMXEXT) {
234                 c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmx2;
235                 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmx2;
236                 c->h264_idct_add16   = ff_h264_idct_add16_8_mmx2;
237                 c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmx2;
238                 if (chroma_format_idc == 1)
239                     c->h264_idct_add8 = ff_h264_idct_add8_8_mmx2;
240                 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx2;
241
242                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmx2;
243                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmx2;
244                 if (chroma_format_idc == 1) {
245                     c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmx2;
246                     c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmx2;
247                 }
248 #if ARCH_X86_32
249                 c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_mmx2;
250                 c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmx2;
251                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
252                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
253 #endif /* ARCH_X86_32 */
254                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmx2;
255                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmx2;
256                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmx2;
257
258                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmx2;
259                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmx2;
260                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmx2;
261
262                 if (mm_flags & AV_CPU_FLAG_SSE2) {
263                     c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
264
265                     c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
266                     c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
267                     if (chroma_format_idc == 1)
268                         c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
269                     c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
270                     c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
271
272                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
273                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
274
275                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
276                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
277
278 #if HAVE_ALIGNED_STACK
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 #endif /* HAVE_ALIGNED_STACK */
284                 }
285                 if (mm_flags & AV_CPU_FLAG_SSSE3) {
286                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
287                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
288                 }
289                 if (mm_flags & AV_CPU_FLAG_AVX) {
290 #if HAVE_ALIGNED_STACK
291                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
292                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
293                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
294                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
295 #endif /* HAVE_ALIGNED_STACK */
296                 }
297             }
298         }
299     } else if (bit_depth == 10) {
300         if (mm_flags & AV_CPU_FLAG_MMX) {
301             if (mm_flags & AV_CPU_FLAG_MMXEXT) {
302 #if ARCH_X86_32
303                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmx2;
304                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmx2;
305                 c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmx2;
306                 c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmx2;
307                 c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmx2;
308                 c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmx2;
309 #endif /* ARCH_X86_32 */
310                 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmx2;
311                 if (mm_flags & AV_CPU_FLAG_SSE2) {
312                     c->h264_idct_add     = ff_h264_idct_add_10_sse2;
313                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
314
315                     c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
316                     if (chroma_format_idc == 1)
317                         c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
318                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
319 #if HAVE_ALIGNED_STACK
320                     c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
321                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
322 #endif /* HAVE_ALIGNED_STACK */
323
324                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
325                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
326                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
327
328                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
329                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
330                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
331
332                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
333                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
334 #if HAVE_ALIGNED_STACK
335                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
336                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
337                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
338                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
339 #endif /* HAVE_ALIGNED_STACK */
340                 }
341                 if (mm_flags & AV_CPU_FLAG_SSE4) {
342                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
343                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
344                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
345
346                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
347                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
348                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
349                 }
350 #if HAVE_AVX_EXTERNAL
351                 if (mm_flags & AV_CPU_FLAG_AVX) {
352                     c->h264_idct_dc_add  =
353                     c->h264_idct_add     = ff_h264_idct_add_10_avx;
354                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
355
356                     c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
357                     if (chroma_format_idc == 1)
358                         c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
359                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
360 #if HAVE_ALIGNED_STACK
361                     c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
362                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
363 #endif /* HAVE_ALIGNED_STACK */
364
365                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
366                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
367 #if HAVE_ALIGNED_STACK
368                     c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
369                     c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
370                     c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
371                     c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
372 #endif /* HAVE_ALIGNED_STACK */
373                 }
374 #endif /* HAVE_AVX_EXTERNAL */
375             }
376         }
377     }
378 #endif /* HAVE_YASM */
379 }