Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / cavsdsp_mmx.c @ be449fca

History | View | Annotate | Download (19.8 KB)

1
/*
2
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3
 * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>
4
 *
5
 * MMX-optimized DSP functions, based on H.264 optimizations by
6
 * Michael Niedermayer and Loren Merritt
7
 *
8
 * This file is part of FFmpeg.
9
 *
10
 * FFmpeg is free software; you can redistribute it and/or
11
 * modify it under the terms of the GNU Lesser General Public
12
 * License as published by the Free Software Foundation; either
13
 * version 2.1 of the License, or (at your option) any later version.
14
 *
15
 * FFmpeg is distributed in the hope that it will be useful,
16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18
 * Lesser General Public License for more details.
19
 *
20
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with FFmpeg; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23
 */
24

    
25
#include "libavutil/common.h"
26
#include "libavutil/x86_cpu.h"
27
#include "libavcodec/dsputil.h"
28
#include "dsputil_mmx.h"
29

    
30
/*****************************************************************************
31
 *
32
 * inverse transform
33
 *
34
 ****************************************************************************/
35

    
36
static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)
37
{
38
    __asm__ volatile(
39
        "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */
40
        "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */
41
        "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */
42
        "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */
43
        "movq   %%mm4, %%mm0  \n\t"
44
        "movq   %%mm5, %%mm3  \n\t"
45
        "movq   %%mm2, %%mm6  \n\t"
46
        "movq   %%mm7, %%mm1  \n\t"
47

    
48
        "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */
49
        "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */
50
        "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */
51
        "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */
52
        "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */
53
        "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */
54
        "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */
55
        "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */
56
        "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
57
        "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
58
        "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
59
        "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
60

    
61
        "movq   %%mm5, %%mm4  \n\t"
62
        "movq   %%mm7, %%mm6  \n\t"
63
        "movq   %%mm3, %%mm0  \n\t"
64
        "movq   %%mm1, %%mm2  \n\t"
65
        SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */
66
        "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */
67
        "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */
68
        "paddw  %%mm7, %%mm7  \n\t"
69
        "paddw  %%mm5, %%mm5  \n\t"
70
        "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */
71
        "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */
72

    
73
        SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */
74
        "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */
75
        "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */
76
        "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */
77
        "paddw  %%mm1, %%mm1  \n\t"
78
        "paddw  %%mm3, %%mm3  \n\t"
79
        "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */
80
        "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */
81

    
82
        "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */
83
        "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */
84
        "movq   %%mm2, %%mm4  \n\t"
85
        "movq   %%mm6, %%mm0  \n\t"
86
        "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */
87
        "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */
88
        "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */
89
        "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */
90
        "paddw  %%mm2, %%mm2  \n\t"
91
        "paddw  %%mm0, %%mm0  \n\t"
92
        "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
93
        "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
94

    
95
        "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */
96
        "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */
97
        SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */
98
        "psllw  $3,    %%mm0  \n\t"
99
        "psllw  $3,    %%mm2  \n\t"
100
        "paddw  %1,    %%mm0  \n\t" /* add rounding bias */
101
        "paddw  %1,    %%mm2  \n\t" /* add rounding bias */
102

    
103
        SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */
104
        SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */
105
        SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */
106
        SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */
107
        SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */
108
        SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */
109
        :: "r"(block), "m"(bias)
110
    );
111
}
112

    
113
static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
114
{
115
    int i;
116
    DECLARE_ALIGNED_8(int16_t, b2[64]);
117

    
118
    for(i=0; i<2; i++){
119
        DECLARE_ALIGNED_8(uint64_t, tmp);
120

    
121
        cavs_idct8_1d(block+4*i, ff_pw_4);
122

    
123
        __asm__ volatile(
124
            "psraw     $3, %%mm7  \n\t"
125
            "psraw     $3, %%mm6  \n\t"
126
            "psraw     $3, %%mm5  \n\t"
127
            "psraw     $3, %%mm4  \n\t"
128
            "psraw     $3, %%mm3  \n\t"
129
            "psraw     $3, %%mm2  \n\t"
130
            "psraw     $3, %%mm1  \n\t"
131
            "psraw     $3, %%mm0  \n\t"
132
            "movq   %%mm7,    %0   \n\t"
133
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
134
            "movq   %%mm0,  8(%1)  \n\t"
135
            "movq   %%mm6, 24(%1)  \n\t"
136
            "movq   %%mm7, 40(%1)  \n\t"
137
            "movq   %%mm4, 56(%1)  \n\t"
138
            "movq    %0,    %%mm7  \n\t"
139
            TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
140
            "movq   %%mm7,   (%1)  \n\t"
141
            "movq   %%mm1, 16(%1)  \n\t"
142
            "movq   %%mm0, 32(%1)  \n\t"
143
            "movq   %%mm3, 48(%1)  \n\t"
144
            : "=m"(tmp)
145
            : "r"(b2+32*i)
146
            : "memory"
147
        );
148
    }
149

    
150
    for(i=0; i<2; i++){
151
        cavs_idct8_1d(b2+4*i, ff_pw_64);
152

    
153
        __asm__ volatile(
154
            "psraw     $7, %%mm7  \n\t"
155
            "psraw     $7, %%mm6  \n\t"
156
            "psraw     $7, %%mm5  \n\t"
157
            "psraw     $7, %%mm4  \n\t"
158
            "psraw     $7, %%mm3  \n\t"
159
            "psraw     $7, %%mm2  \n\t"
160
            "psraw     $7, %%mm1  \n\t"
161
            "psraw     $7, %%mm0  \n\t"
162
            "movq   %%mm7,    (%0)  \n\t"
163
            "movq   %%mm5,  16(%0)  \n\t"
164
            "movq   %%mm3,  32(%0)  \n\t"
165
            "movq   %%mm1,  48(%0)  \n\t"
166
            "movq   %%mm0,  64(%0)  \n\t"
167
            "movq   %%mm2,  80(%0)  \n\t"
168
            "movq   %%mm4,  96(%0)  \n\t"
169
            "movq   %%mm6, 112(%0)  \n\t"
170
            :: "r"(b2+4*i)
171
            : "memory"
172
        );
173
    }
174

    
175
    add_pixels_clamped_mmx(b2, dst, stride);
176

    
177
    /* clear block */
178
    __asm__ volatile(
179
            "pxor %%mm7, %%mm7   \n\t"
180
            "movq %%mm7, (%0)    \n\t"
181
            "movq %%mm7, 8(%0)   \n\t"
182
            "movq %%mm7, 16(%0)  \n\t"
183
            "movq %%mm7, 24(%0)  \n\t"
184
            "movq %%mm7, 32(%0)  \n\t"
185
            "movq %%mm7, 40(%0)  \n\t"
186
            "movq %%mm7, 48(%0)  \n\t"
187
            "movq %%mm7, 56(%0)  \n\t"
188
            "movq %%mm7, 64(%0)  \n\t"
189
            "movq %%mm7, 72(%0)  \n\t"
190
            "movq %%mm7, 80(%0)  \n\t"
191
            "movq %%mm7, 88(%0)  \n\t"
192
            "movq %%mm7, 96(%0)  \n\t"
193
            "movq %%mm7, 104(%0) \n\t"
194
            "movq %%mm7, 112(%0) \n\t"
195
            "movq %%mm7, 120(%0) \n\t"
196
            :: "r" (block)
197
    );
198
}
199

    
200
/*****************************************************************************
201
 *
202
 * motion compensation
203
 *
204
 ****************************************************************************/
205

    
206
/* vertical filter [-1 -2 96 42 -7  0]  */
207
#define QPEL_CAVSV1(A,B,C,D,E,F,OP)      \
208
        "movd (%0), "#F"            \n\t"\
209
        "movq "#C", %%mm6           \n\t"\
210
        "pmullw %5, %%mm6           \n\t"\
211
        "movq "#D", %%mm7           \n\t"\
212
        "pmullw %6, %%mm7           \n\t"\
213
        "psllw $3, "#E"             \n\t"\
214
        "psubw "#E", %%mm6          \n\t"\
215
        "psraw $3, "#E"             \n\t"\
216
        "paddw %%mm7, %%mm6         \n\t"\
217
        "paddw "#E", %%mm6          \n\t"\
218
        "paddw "#B", "#B"           \n\t"\
219
        "pxor %%mm7, %%mm7          \n\t"\
220
        "add %2, %0                 \n\t"\
221
        "punpcklbw %%mm7, "#F"      \n\t"\
222
        "psubw "#B", %%mm6          \n\t"\
223
        "psraw $1, "#B"             \n\t"\
224
        "psubw "#A", %%mm6          \n\t"\
225
        "paddw %4, %%mm6            \n\t"\
226
        "psraw $7, %%mm6            \n\t"\
227
        "packuswb %%mm6, %%mm6      \n\t"\
228
        OP(%%mm6, (%1), A, d)            \
229
        "add %3, %1                 \n\t"
230

    
231
/* vertical filter [ 0 -1  5  5 -1  0]  */
232
#define QPEL_CAVSV2(A,B,C,D,E,F,OP)      \
233
        "movd (%0), "#F"            \n\t"\
234
        "movq "#C", %%mm6           \n\t"\
235
        "paddw "#D", %%mm6          \n\t"\
236
        "pmullw %5, %%mm6           \n\t"\
237
        "add %2, %0                 \n\t"\
238
        "punpcklbw %%mm7, "#F"      \n\t"\
239
        "psubw "#B", %%mm6          \n\t"\
240
        "psubw "#E", %%mm6          \n\t"\
241
        "paddw %4, %%mm6            \n\t"\
242
        "psraw $3, %%mm6            \n\t"\
243
        "packuswb %%mm6, %%mm6      \n\t"\
244
        OP(%%mm6, (%1), A, d)            \
245
        "add %3, %1                 \n\t"
246

    
247
/* vertical filter [ 0 -7 42 96 -2 -1]  */
248
#define QPEL_CAVSV3(A,B,C,D,E,F,OP)      \
249
        "movd (%0), "#F"            \n\t"\
250
        "movq "#C", %%mm6           \n\t"\
251
        "pmullw %6, %%mm6           \n\t"\
252
        "movq "#D", %%mm7           \n\t"\
253
        "pmullw %5, %%mm7           \n\t"\
254
        "psllw $3, "#B"             \n\t"\
255
        "psubw "#B", %%mm6          \n\t"\
256
        "psraw $3, "#B"             \n\t"\
257
        "paddw %%mm7, %%mm6         \n\t"\
258
        "paddw "#B", %%mm6          \n\t"\
259
        "paddw "#E", "#E"           \n\t"\
260
        "pxor %%mm7, %%mm7          \n\t"\
261
        "add %2, %0                 \n\t"\
262
        "punpcklbw %%mm7, "#F"      \n\t"\
263
        "psubw "#E", %%mm6          \n\t"\
264
        "psraw $1, "#E"             \n\t"\
265
        "psubw "#F", %%mm6          \n\t"\
266
        "paddw %4, %%mm6            \n\t"\
267
        "psraw $7, %%mm6            \n\t"\
268
        "packuswb %%mm6, %%mm6      \n\t"\
269
        OP(%%mm6, (%1), A, d)            \
270
        "add %3, %1                 \n\t"
271

    
272

    
273
#define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
274
    int w= 2;\
275
    src -= 2*srcStride;\
276
    \
277
    while(w--){\
278
      __asm__ volatile(\
279
        "pxor %%mm7, %%mm7          \n\t"\
280
        "movd (%0), %%mm0           \n\t"\
281
        "add %2, %0                 \n\t"\
282
        "movd (%0), %%mm1           \n\t"\
283
        "add %2, %0                 \n\t"\
284
        "movd (%0), %%mm2           \n\t"\
285
        "add %2, %0                 \n\t"\
286
        "movd (%0), %%mm3           \n\t"\
287
        "add %2, %0                 \n\t"\
288
        "movd (%0), %%mm4           \n\t"\
289
        "add %2, %0                 \n\t"\
290
        "punpcklbw %%mm7, %%mm0     \n\t"\
291
        "punpcklbw %%mm7, %%mm1     \n\t"\
292
        "punpcklbw %%mm7, %%mm2     \n\t"\
293
        "punpcklbw %%mm7, %%mm3     \n\t"\
294
        "punpcklbw %%mm7, %%mm4     \n\t"\
295
        VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
296
        VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
297
        VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
298
        VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
299
        VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
300
        VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
301
        VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
302
        VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
303
        \
304
        : "+a"(src), "+c"(dst)\
305
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\
306
        : "memory"\
307
     );\
308
     if(h==16){\
309
        __asm__ volatile(\
310
            VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
311
            VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
312
            VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
313
            VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
314
            VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
315
            VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
316
            VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
317
            VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
318
            \
319
           : "+a"(src), "+c"(dst)\
320
           : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1), "m"(MUL2)\
321
           : "memory"\
322
        );\
323
     }\
324
     src += 4-(h+5)*srcStride;\
325
     dst += 4-h*dstStride;\
326
   }
327

    
328
#define QPEL_CAVS(OPNAME, OP, MMX)\
329
static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
330
    int h=8;\
331
    __asm__ volatile(\
332
        "pxor %%mm7, %%mm7          \n\t"\
333
        "movq %5, %%mm6             \n\t"\
334
        "1:                         \n\t"\
335
        "movq    (%0), %%mm0        \n\t"\
336
        "movq   1(%0), %%mm2        \n\t"\
337
        "movq %%mm0, %%mm1          \n\t"\
338
        "movq %%mm2, %%mm3          \n\t"\
339
        "punpcklbw %%mm7, %%mm0     \n\t"\
340
        "punpckhbw %%mm7, %%mm1     \n\t"\
341
        "punpcklbw %%mm7, %%mm2     \n\t"\
342
        "punpckhbw %%mm7, %%mm3     \n\t"\
343
        "paddw %%mm2, %%mm0         \n\t"\
344
        "paddw %%mm3, %%mm1         \n\t"\
345
        "pmullw %%mm6, %%mm0        \n\t"\
346
        "pmullw %%mm6, %%mm1        \n\t"\
347
        "movq   -1(%0), %%mm2       \n\t"\
348
        "movq    2(%0), %%mm4       \n\t"\
349
        "movq %%mm2, %%mm3          \n\t"\
350
        "movq %%mm4, %%mm5          \n\t"\
351
        "punpcklbw %%mm7, %%mm2     \n\t"\
352
        "punpckhbw %%mm7, %%mm3     \n\t"\
353
        "punpcklbw %%mm7, %%mm4     \n\t"\
354
        "punpckhbw %%mm7, %%mm5     \n\t"\
355
        "paddw %%mm4, %%mm2         \n\t"\
356
        "paddw %%mm3, %%mm5         \n\t"\
357
        "psubw %%mm2, %%mm0         \n\t"\
358
        "psubw %%mm5, %%mm1         \n\t"\
359
        "movq %6, %%mm5             \n\t"\
360
        "paddw %%mm5, %%mm0         \n\t"\
361
        "paddw %%mm5, %%mm1         \n\t"\
362
        "psraw $3, %%mm0            \n\t"\
363
        "psraw $3, %%mm1            \n\t"\
364
        "packuswb %%mm1, %%mm0      \n\t"\
365
        OP(%%mm0, (%1),%%mm5, q)         \
366
        "add %3, %0                 \n\t"\
367
        "add %4, %1                 \n\t"\
368
        "decl %2                    \n\t"\
369
        " jnz 1b                    \n\t"\
370
        : "+a"(src), "+c"(dst), "+m"(h)\
371
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
372
        : "memory"\
373
    );\
374
}\
375
\
376
static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
377
  QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
378
}\
379
\
380
static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
381
  QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
382
}\
383
\
384
static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
385
  QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
386
}\
387
\
388
static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
389
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
390
}\
391
static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
392
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
393
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
394
}\
395
\
396
static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
397
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
398
}\
399
static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
400
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
401
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
402
}\
403
\
404
static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
405
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
406
}\
407
static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
408
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
409
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
410
}\
411
\
412
static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
413
    OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
414
    OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
415
    src += 8*srcStride;\
416
    dst += 8*dstStride;\
417
    OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
418
    OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
419
}\
420

    
421
#define CAVS_MC(OPNAME, SIZE, MMX) \
422
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
423
    OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
424
}\
425
\
426
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
427
    OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
428
}\
429
\
430
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
431
    OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
432
}\
433
\
434
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
435
    OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
436
}\
437

    
438
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
439
#define AVG_3DNOW_OP(a,b,temp, size) \
440
"mov" #size " " #b ", " #temp "   \n\t"\
441
"pavgusb " #temp ", " #a "        \n\t"\
442
"mov" #size " " #a ", " #b "      \n\t"
443
#define AVG_MMX2_OP(a,b,temp, size) \
444
"mov" #size " " #b ", " #temp "   \n\t"\
445
"pavgb " #temp ", " #a "          \n\t"\
446
"mov" #size " " #a ", " #b "      \n\t"
447

    
448
QPEL_CAVS(put_,       PUT_OP, 3dnow)
449
QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
450
QPEL_CAVS(put_,       PUT_OP, mmx2)
451
QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)
452

    
453
CAVS_MC(put_, 8, 3dnow)
454
CAVS_MC(put_, 16,3dnow)
455
CAVS_MC(avg_, 8, 3dnow)
456
CAVS_MC(avg_, 16,3dnow)
457
CAVS_MC(put_, 8, mmx2)
458
CAVS_MC(put_, 16,mmx2)
459
CAVS_MC(avg_, 8, mmx2)
460
CAVS_MC(avg_, 16,mmx2)
461

    
462
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
463
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
464
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
465
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
466

    
467
void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx) {
468
#define dspfunc(PFX, IDX, NUM) \
469
    c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
470
    c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
471
    c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
472
    c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
473
    c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \
474

    
475
    dspfunc(put_cavs_qpel, 0, 16);
476
    dspfunc(put_cavs_qpel, 1, 8);
477
    dspfunc(avg_cavs_qpel, 0, 16);
478
    dspfunc(avg_cavs_qpel, 1, 8);
479
#undef dspfunc
480
    c->cavs_idct8_add = cavs_idct8_add_mmx;
481
}
482

    
483
void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx) {
484
#define dspfunc(PFX, IDX, NUM) \
485
    c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
486
    c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
487
    c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
488
    c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
489
    c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
490

    
491
    dspfunc(put_cavs_qpel, 0, 16);
492
    dspfunc(put_cavs_qpel, 1, 8);
493
    dspfunc(avg_cavs_qpel, 0, 16);
494
    dspfunc(avg_cavs_qpel, 1, 8);
495
#undef dspfunc
496
    c->cavs_idct8_add = cavs_idct8_add_mmx;
497
}