h264: convert loop filter strength dsp function to yasm.
[ffmpeg.git] / libavcodec / x86 / h264dsp_mmx.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_cpu.h"
23 #include "libavcodec/h264dsp.h"
24 #include "dsputil_mmx.h"
25
26 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1  ) = 0x0103010301030103ULL;
27
28 /***********************************/
29 /* IDCT */
30 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
31 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT (uint8_t *dst, int16_t *block, 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
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
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 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
78 IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
79 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
80 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
81 #if HAVE_AVX
82 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
83 #endif
84
85 void ff_h264_luma_dc_dequant_idct_mmx (DCTELEM *output, DCTELEM *input, int qmul);
86 void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
87
88 /***********************************/
89 /* deblocking */
90
91 void ff_h264_loop_filter_strength_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
92                                        int8_t ref[2][40], int16_t mv[2][40][2],
93                                        int bidir, int edges, int step,
94                                        int mask_mv0, int mask_mv1, int field);
95
96 #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
97 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
98                                                                 int alpha, int beta, int8_t *tc0);
99 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
100 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
101                                                                 int alpha, 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, int beta, int8_t *tc0)
133 {
134     if((tc0[0] & tc0[1]) >= 0)
135         ff_deblock_v8_luma_8_mmx2(pix+0, stride, alpha, beta, tc0);
136     if((tc0[2] & tc0[3]) >= 0)
137         ff_deblock_v8_luma_8_mmx2(pix+8, stride, alpha, beta, tc0+2);
138 }
139 LF_IFUNC(v8, luma_intra,        8, mmx2)
140 static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride, int alpha, int beta)
141 {
142     ff_deblock_v8_luma_intra_8_mmx2(pix+0, stride, alpha, beta);
143     ff_deblock_v8_luma_intra_8_mmx2(pix+8, stride, alpha, beta);
144 }
145 #endif /* ARCH_X86_32 */
146
147 LF_FUNC (v,  luma,            10, mmx2)
148 LF_IFUNC(v,  luma_intra,      10, mmx2)
149
150 /***********************************/
151 /* weighted prediction */
152
153 #define H264_WEIGHT(W, OPT) \
154 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, \
155     int stride, int height, int log2_denom, int weight, int offset);
156
157 #define H264_BIWEIGHT(W, OPT) \
158 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, \
159     uint8_t *src, int stride, int height, int log2_denom, int weightd, \
160     int weights, int offset);
161
162 #define H264_BIWEIGHT_MMX(W) \
163 H264_WEIGHT  (W, mmx2) \
164 H264_BIWEIGHT(W, mmx2)
165
166 #define H264_BIWEIGHT_MMX_SSE(W) \
167 H264_BIWEIGHT_MMX(W) \
168 H264_WEIGHT      (W, sse2) \
169 H264_BIWEIGHT    (W, sse2) \
170 H264_BIWEIGHT    (W, ssse3)
171
172 H264_BIWEIGHT_MMX_SSE(16)
173 H264_BIWEIGHT_MMX_SSE( 8)
174 H264_BIWEIGHT_MMX    ( 4)
175
176 #define H264_WEIGHT_10(W, DEPTH, OPT) \
177 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
178     int stride, int height, int log2_denom, int weight, int offset);
179
180 #define H264_BIWEIGHT_10(W, DEPTH, OPT) \
181 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT \
182     (uint8_t *dst, uint8_t *src, int stride, int height, int log2_denom, \
183      int weightd, int weights, int offset);
184
185 #define H264_BIWEIGHT_10_SSE(W, DEPTH) \
186 H264_WEIGHT_10  (W, DEPTH, sse2) \
187 H264_WEIGHT_10  (W, DEPTH, sse4) \
188 H264_BIWEIGHT_10(W, DEPTH, sse2) \
189 H264_BIWEIGHT_10(W, DEPTH, sse4)
190
191 H264_BIWEIGHT_10_SSE(16, 10)
192 H264_BIWEIGHT_10_SSE( 8, 10)
193 H264_BIWEIGHT_10_SSE( 4, 10)
194
195 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
196 {
197     int mm_flags = av_get_cpu_flags();
198
199 #if HAVE_YASM
200     if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMX2) {
201         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
202     }
203
204     if (bit_depth == 8) {
205     if (mm_flags & AV_CPU_FLAG_MMX) {
206         c->h264_idct_dc_add         =
207         c->h264_idct_add            = ff_h264_idct_add_8_mmx;
208         c->h264_idct8_dc_add        =
209         c->h264_idct8_add           = ff_h264_idct8_add_8_mmx;
210
211         c->h264_idct_add16          = ff_h264_idct_add16_8_mmx;
212         c->h264_idct8_add4          = ff_h264_idct8_add4_8_mmx;
213         if (chroma_format_idc == 1)
214             c->h264_idct_add8       = ff_h264_idct_add8_8_mmx;
215         c->h264_idct_add16intra     = ff_h264_idct_add16intra_8_mmx;
216         if (mm_flags & AV_CPU_FLAG_CMOV)
217             c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
218
219         if (mm_flags & AV_CPU_FLAG_MMX2) {
220             c->h264_idct_dc_add    = ff_h264_idct_dc_add_8_mmx2;
221             c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_8_mmx2;
222             c->h264_idct_add16     = ff_h264_idct_add16_8_mmx2;
223             c->h264_idct8_add4     = ff_h264_idct8_add4_8_mmx2;
224             if (chroma_format_idc == 1)
225                 c->h264_idct_add8  = ff_h264_idct_add8_8_mmx2;
226             c->h264_idct_add16intra= ff_h264_idct_add16intra_8_mmx2;
227
228             c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_8_mmx2;
229             c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_8_mmx2;
230             if (chroma_format_idc == 1) {
231                 c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_8_mmx2;
232                 c->h264_h_loop_filter_chroma_intra= ff_deblock_h_chroma_intra_8_mmx2;
233             }
234 #if ARCH_X86_32
235             c->h264_v_loop_filter_luma= ff_deblock_v_luma_8_mmx2;
236             c->h264_h_loop_filter_luma= ff_deblock_h_luma_8_mmx2;
237             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
238             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
239 #endif
240             c->weight_h264_pixels_tab[0]= ff_h264_weight_16_mmx2;
241             c->weight_h264_pixels_tab[1]= ff_h264_weight_8_mmx2;
242             c->weight_h264_pixels_tab[2]= ff_h264_weight_4_mmx2;
243
244             c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16_mmx2;
245             c->biweight_h264_pixels_tab[1]= ff_h264_biweight_8_mmx2;
246             c->biweight_h264_pixels_tab[2]= ff_h264_biweight_4_mmx2;
247
248             if (mm_flags&AV_CPU_FLAG_SSE2) {
249                 c->h264_idct8_add           = ff_h264_idct8_add_8_sse2;
250
251                 c->h264_idct_add16          = ff_h264_idct_add16_8_sse2;
252                 c->h264_idct8_add4          = ff_h264_idct8_add4_8_sse2;
253                 if (chroma_format_idc == 1)
254                     c->h264_idct_add8       = ff_h264_idct_add8_8_sse2;
255                 c->h264_idct_add16intra     = ff_h264_idct_add16intra_8_sse2;
256                 c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_sse2;
257
258                 c->weight_h264_pixels_tab[0]= ff_h264_weight_16_sse2;
259                 c->weight_h264_pixels_tab[1]= ff_h264_weight_8_sse2;
260
261                 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16_sse2;
262                 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_8_sse2;
263
264 #if HAVE_ALIGNED_STACK
265                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
266                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
267                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
268                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
269 #endif
270             }
271             if (mm_flags&AV_CPU_FLAG_SSSE3) {
272                 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16_ssse3;
273                 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_8_ssse3;
274             }
275             if (mm_flags&AV_CPU_FLAG_AVX) {
276 #if HAVE_ALIGNED_STACK
277                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
278                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
279                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
280                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
281 #endif
282             }
283         }
284     }
285 #endif
286     } else if (bit_depth == 10) {
287 #if HAVE_YASM
288     if (mm_flags & AV_CPU_FLAG_MMX) {
289         if (mm_flags & AV_CPU_FLAG_MMX2) {
290 #if ARCH_X86_32
291             c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_mmx2;
292             c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_mmx2;
293             c->h264_v_loop_filter_luma= ff_deblock_v_luma_10_mmx2;
294             c->h264_h_loop_filter_luma= ff_deblock_h_luma_10_mmx2;
295             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmx2;
296             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmx2;
297 #endif
298             c->h264_idct_dc_add= ff_h264_idct_dc_add_10_mmx2;
299             if (mm_flags&AV_CPU_FLAG_SSE2) {
300                 c->h264_idct_add       = ff_h264_idct_add_10_sse2;
301                 c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_10_sse2;
302
303                 c->h264_idct_add16     = ff_h264_idct_add16_10_sse2;
304                 if (chroma_format_idc == 1)
305                     c->h264_idct_add8  = ff_h264_idct_add8_10_sse2;
306                 c->h264_idct_add16intra= ff_h264_idct_add16intra_10_sse2;
307 #if HAVE_ALIGNED_STACK
308                 c->h264_idct8_add      = ff_h264_idct8_add_10_sse2;
309                 c->h264_idct8_add4     = ff_h264_idct8_add4_10_sse2;
310 #endif
311
312                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
313                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
314                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
315
316                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
317                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
318                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
319
320                 c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_sse2;
321                 c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_sse2;
322 #if HAVE_ALIGNED_STACK
323                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
324                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
325                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
326                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
327 #endif
328             }
329             if (mm_flags&AV_CPU_FLAG_SSE4) {
330                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
331                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
332                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
333
334                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
335                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
336                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
337             }
338 #if HAVE_AVX
339             if (mm_flags&AV_CPU_FLAG_AVX) {
340                 c->h264_idct_dc_add    =
341                 c->h264_idct_add       = ff_h264_idct_add_10_avx;
342                 c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_10_avx;
343
344                 c->h264_idct_add16     = ff_h264_idct_add16_10_avx;
345                 if (chroma_format_idc == 1)
346                     c->h264_idct_add8  = ff_h264_idct_add8_10_avx;
347                 c->h264_idct_add16intra= ff_h264_idct_add16intra_10_avx;
348 #if HAVE_ALIGNED_STACK
349                 c->h264_idct8_add      = ff_h264_idct8_add_10_avx;
350                 c->h264_idct8_add4     = ff_h264_idct8_add4_10_avx;
351 #endif
352
353                 c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_avx;
354                 c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_avx;
355 #if HAVE_ALIGNED_STACK
356                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
357                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
358                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
359                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
360 #endif
361             }
362 #endif /* HAVE_AVX */
363         }
364     }
365     }
366 #endif
367 }