Statistics
| Branch: | Revision:

ffmpeg / libavcodec / x86 / cavsdsp_mmx.c @ f079a64a

History | View | Annotate | Download (19.2 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 "libavcodec/cavsdsp.h"
29
#include "dsputil_mmx.h"
30

    
31
/*****************************************************************************
32
 *
33
 * inverse transform
34
 *
35
 ****************************************************************************/
36

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

    
49
        "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */
50
        "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */
51
        "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */
52
        "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */
53
        "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */
54
        "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */
55
        "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */
56
        "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */
57
        "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
58
        "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
59
        "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
60
        "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
61

    
62
        "movq   %%mm5, %%mm4  \n\t"
63
        "movq   %%mm7, %%mm6  \n\t"
64
        "movq   %%mm3, %%mm0  \n\t"
65
        "movq   %%mm1, %%mm2  \n\t"
66
        SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */
67
        "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */
68
        "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */
69
        "paddw  %%mm7, %%mm7  \n\t"
70
        "paddw  %%mm5, %%mm5  \n\t"
71
        "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */
72
        "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */
73

    
74
        SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */
75
        "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */
76
        "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */
77
        "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */
78
        "paddw  %%mm1, %%mm1  \n\t"
79
        "paddw  %%mm3, %%mm3  \n\t"
80
        "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */
81
        "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */
82

    
83
        "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */
84
        "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */
85
        "movq   %%mm2, %%mm4  \n\t"
86
        "movq   %%mm6, %%mm0  \n\t"
87
        "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */
88
        "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */
89
        "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */
90
        "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */
91
        "paddw  %%mm2, %%mm2  \n\t"
92
        "paddw  %%mm0, %%mm0  \n\t"
93
        "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
94
        "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
95

    
96
        "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */
97
        "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */
98
        SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */
99
        "psllw  $3,    %%mm0  \n\t"
100
        "psllw  $3,    %%mm2  \n\t"
101
        "paddw  %1,    %%mm0  \n\t" /* add rounding bias */
102
        "paddw  %1,    %%mm2  \n\t" /* add rounding bias */
103

    
104
        SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */
105
        SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */
106
        SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */
107
        SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */
108
        SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */
109
        SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */
110
        :: "r"(block), "m"(bias)
111
    );
112
}
113

    
114
static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
115
{
116
    int i;
117
    DECLARE_ALIGNED(8, int16_t, b2)[64];
118

    
119
    for(i=0; i<2; i++){
120
        DECLARE_ALIGNED(8, uint64_t, tmp);
121

    
122
        cavs_idct8_1d(block+4*i, ff_pw_4.a);
123

    
124
        __asm__ volatile(
125
            "psraw     $3, %%mm7  \n\t"
126
            "psraw     $3, %%mm6  \n\t"
127
            "psraw     $3, %%mm5  \n\t"
128
            "psraw     $3, %%mm4  \n\t"
129
            "psraw     $3, %%mm3  \n\t"
130
            "psraw     $3, %%mm2  \n\t"
131
            "psraw     $3, %%mm1  \n\t"
132
            "psraw     $3, %%mm0  \n\t"
133
            "movq   %%mm7,    %0   \n\t"
134
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
135
            "movq   %%mm0,  8(%1)  \n\t"
136
            "movq   %%mm6, 24(%1)  \n\t"
137
            "movq   %%mm7, 40(%1)  \n\t"
138
            "movq   %%mm4, 56(%1)  \n\t"
139
            "movq    %0,    %%mm7  \n\t"
140
            TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
141
            "movq   %%mm7,   (%1)  \n\t"
142
            "movq   %%mm1, 16(%1)  \n\t"
143
            "movq   %%mm0, 32(%1)  \n\t"
144
            "movq   %%mm3, 48(%1)  \n\t"
145
            : "=m"(tmp)
146
            : "r"(b2+32*i)
147
            : "memory"
148
        );
149
    }
150

    
151
    for(i=0; i<2; i++){
152
        cavs_idct8_1d(b2+4*i, ff_pw_64.a);
153

    
154
        __asm__ volatile(
155
            "psraw     $7, %%mm7  \n\t"
156
            "psraw     $7, %%mm6  \n\t"
157
            "psraw     $7, %%mm5  \n\t"
158
            "psraw     $7, %%mm4  \n\t"
159
            "psraw     $7, %%mm3  \n\t"
160
            "psraw     $7, %%mm2  \n\t"
161
            "psraw     $7, %%mm1  \n\t"
162
            "psraw     $7, %%mm0  \n\t"
163
            "movq   %%mm7,    (%0)  \n\t"
164
            "movq   %%mm5,  16(%0)  \n\t"
165
            "movq   %%mm3,  32(%0)  \n\t"
166
            "movq   %%mm1,  48(%0)  \n\t"
167
            "movq   %%mm0,  64(%0)  \n\t"
168
            "movq   %%mm2,  80(%0)  \n\t"
169
            "movq   %%mm4,  96(%0)  \n\t"
170
            "movq   %%mm6, 112(%0)  \n\t"
171
            :: "r"(b2+4*i)
172
            : "memory"
173
        );
174
    }
175

    
176
    add_pixels_clamped_mmx(b2, dst, stride);
177
}
178

    
179
/*****************************************************************************
180
 *
181
 * motion compensation
182
 *
183
 ****************************************************************************/
184

    
185
/* vertical filter [-1 -2 96 42 -7  0]  */
186
#define QPEL_CAVSV1(A,B,C,D,E,F,OP,MUL2) \
187
        "movd (%0), "#F"            \n\t"\
188
        "movq "#C", %%mm6           \n\t"\
189
        "pmullw %5, %%mm6           \n\t"\
190
        "movq "#D", %%mm7           \n\t"\
191
        "pmullw "MANGLE(MUL2)", %%mm7\n\t"\
192
        "psllw $3, "#E"             \n\t"\
193
        "psubw "#E", %%mm6          \n\t"\
194
        "psraw $3, "#E"             \n\t"\
195
        "paddw %%mm7, %%mm6         \n\t"\
196
        "paddw "#E", %%mm6          \n\t"\
197
        "paddw "#B", "#B"           \n\t"\
198
        "pxor %%mm7, %%mm7          \n\t"\
199
        "add %2, %0                 \n\t"\
200
        "punpcklbw %%mm7, "#F"      \n\t"\
201
        "psubw "#B", %%mm6          \n\t"\
202
        "psraw $1, "#B"             \n\t"\
203
        "psubw "#A", %%mm6          \n\t"\
204
        "paddw %4, %%mm6            \n\t"\
205
        "psraw $7, %%mm6            \n\t"\
206
        "packuswb %%mm6, %%mm6      \n\t"\
207
        OP(%%mm6, (%1), A, d)            \
208
        "add %3, %1                 \n\t"
209

    
210
/* vertical filter [ 0 -1  5  5 -1  0]  */
211
#define QPEL_CAVSV2(A,B,C,D,E,F,OP,MUL2) \
212
        "movd (%0), "#F"            \n\t"\
213
        "movq "#C", %%mm6           \n\t"\
214
        "paddw "#D", %%mm6          \n\t"\
215
        "pmullw %5, %%mm6           \n\t"\
216
        "add %2, %0                 \n\t"\
217
        "punpcklbw %%mm7, "#F"      \n\t"\
218
        "psubw "#B", %%mm6          \n\t"\
219
        "psubw "#E", %%mm6          \n\t"\
220
        "paddw %4, %%mm6            \n\t"\
221
        "psraw $3, %%mm6            \n\t"\
222
        "packuswb %%mm6, %%mm6      \n\t"\
223
        OP(%%mm6, (%1), A, d)            \
224
        "add %3, %1                 \n\t"
225

    
226
/* vertical filter [ 0 -7 42 96 -2 -1]  */
227
#define QPEL_CAVSV3(A,B,C,D,E,F,OP,MUL2) \
228
        "movd (%0), "#F"            \n\t"\
229
        "movq "#C", %%mm6           \n\t"\
230
        "pmullw "MANGLE(MUL2)", %%mm6\n\t"\
231
        "movq "#D", %%mm7           \n\t"\
232
        "pmullw %5, %%mm7           \n\t"\
233
        "psllw $3, "#B"             \n\t"\
234
        "psubw "#B", %%mm6          \n\t"\
235
        "psraw $3, "#B"             \n\t"\
236
        "paddw %%mm7, %%mm6         \n\t"\
237
        "paddw "#B", %%mm6          \n\t"\
238
        "paddw "#E", "#E"           \n\t"\
239
        "pxor %%mm7, %%mm7          \n\t"\
240
        "add %2, %0                 \n\t"\
241
        "punpcklbw %%mm7, "#F"      \n\t"\
242
        "psubw "#E", %%mm6          \n\t"\
243
        "psraw $1, "#E"             \n\t"\
244
        "psubw "#F", %%mm6          \n\t"\
245
        "paddw %4, %%mm6            \n\t"\
246
        "psraw $7, %%mm6            \n\t"\
247
        "packuswb %%mm6, %%mm6      \n\t"\
248
        OP(%%mm6, (%1), A, d)            \
249
        "add %3, %1                 \n\t"
250

    
251

    
252
#define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
253
    int w= 2;\
254
    src -= 2*srcStride;\
255
    \
256
    while(w--){\
257
      __asm__ volatile(\
258
        "pxor %%mm7, %%mm7          \n\t"\
259
        "movd (%0), %%mm0           \n\t"\
260
        "add %2, %0                 \n\t"\
261
        "movd (%0), %%mm1           \n\t"\
262
        "add %2, %0                 \n\t"\
263
        "movd (%0), %%mm2           \n\t"\
264
        "add %2, %0                 \n\t"\
265
        "movd (%0), %%mm3           \n\t"\
266
        "add %2, %0                 \n\t"\
267
        "movd (%0), %%mm4           \n\t"\
268
        "add %2, %0                 \n\t"\
269
        "punpcklbw %%mm7, %%mm0     \n\t"\
270
        "punpcklbw %%mm7, %%mm1     \n\t"\
271
        "punpcklbw %%mm7, %%mm2     \n\t"\
272
        "punpcklbw %%mm7, %%mm3     \n\t"\
273
        "punpcklbw %%mm7, %%mm4     \n\t"\
274
        VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
275
        VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
276
        VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
277
        VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
278
        VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
279
        VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
280
        VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
281
        VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
282
        \
283
        : "+a"(src), "+c"(dst)\
284
        : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD), "m"(MUL1)\
285
        : "memory"\
286
     );\
287
     if(h==16){\
288
        __asm__ volatile(\
289
            VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
290
            VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
291
            VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
292
            VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
293
            VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
294
            VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
295
            VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
296
            VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
297
            \
298
           : "+a"(src), "+c"(dst)\
299
           : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1)\
300
           : "memory"\
301
        );\
302
     }\
303
     src += 4-(h+5)*srcStride;\
304
     dst += 4-h*dstStride;\
305
   }
306

    
307
#define QPEL_CAVS(OPNAME, OP, MMX)\
308
static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
309
    int h=8;\
310
    __asm__ volatile(\
311
        "pxor %%mm7, %%mm7          \n\t"\
312
        "movq %5, %%mm6             \n\t"\
313
        "1:                         \n\t"\
314
        "movq    (%0), %%mm0        \n\t"\
315
        "movq   1(%0), %%mm2        \n\t"\
316
        "movq %%mm0, %%mm1          \n\t"\
317
        "movq %%mm2, %%mm3          \n\t"\
318
        "punpcklbw %%mm7, %%mm0     \n\t"\
319
        "punpckhbw %%mm7, %%mm1     \n\t"\
320
        "punpcklbw %%mm7, %%mm2     \n\t"\
321
        "punpckhbw %%mm7, %%mm3     \n\t"\
322
        "paddw %%mm2, %%mm0         \n\t"\
323
        "paddw %%mm3, %%mm1         \n\t"\
324
        "pmullw %%mm6, %%mm0        \n\t"\
325
        "pmullw %%mm6, %%mm1        \n\t"\
326
        "movq   -1(%0), %%mm2       \n\t"\
327
        "movq    2(%0), %%mm4       \n\t"\
328
        "movq %%mm2, %%mm3          \n\t"\
329
        "movq %%mm4, %%mm5          \n\t"\
330
        "punpcklbw %%mm7, %%mm2     \n\t"\
331
        "punpckhbw %%mm7, %%mm3     \n\t"\
332
        "punpcklbw %%mm7, %%mm4     \n\t"\
333
        "punpckhbw %%mm7, %%mm5     \n\t"\
334
        "paddw %%mm4, %%mm2         \n\t"\
335
        "paddw %%mm3, %%mm5         \n\t"\
336
        "psubw %%mm2, %%mm0         \n\t"\
337
        "psubw %%mm5, %%mm1         \n\t"\
338
        "movq %6, %%mm5             \n\t"\
339
        "paddw %%mm5, %%mm0         \n\t"\
340
        "paddw %%mm5, %%mm1         \n\t"\
341
        "psraw $3, %%mm0            \n\t"\
342
        "psraw $3, %%mm1            \n\t"\
343
        "packuswb %%mm1, %%mm0      \n\t"\
344
        OP(%%mm0, (%1),%%mm5, q)         \
345
        "add %3, %0                 \n\t"\
346
        "add %4, %1                 \n\t"\
347
        "decl %2                    \n\t"\
348
        " jnz 1b                    \n\t"\
349
        : "+a"(src), "+c"(dst), "+m"(h)\
350
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
351
        : "memory"\
352
    );\
353
}\
354
\
355
static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
356
  QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
357
}\
358
\
359
static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
360
  QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
361
}\
362
\
363
static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
364
  QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
365
}\
366
\
367
static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
368
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
369
}\
370
static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
371
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
372
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
373
}\
374
\
375
static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
376
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
377
}\
378
static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
379
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
380
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
381
}\
382
\
383
static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
384
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
385
}\
386
static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
387
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
388
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
389
}\
390
\
391
static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
392
    OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
393
    OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
394
    src += 8*srcStride;\
395
    dst += 8*dstStride;\
396
    OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
397
    OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
398
}\
399

    
400
#define CAVS_MC(OPNAME, SIZE, MMX) \
401
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
402
    OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
403
}\
404
\
405
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
406
    OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
407
}\
408
\
409
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
410
    OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
411
}\
412
\
413
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
414
    OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
415
}\
416

    
417
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
418
#define AVG_3DNOW_OP(a,b,temp, size) \
419
"mov" #size " " #b ", " #temp "   \n\t"\
420
"pavgusb " #temp ", " #a "        \n\t"\
421
"mov" #size " " #a ", " #b "      \n\t"
422
#define AVG_MMX2_OP(a,b,temp, size) \
423
"mov" #size " " #b ", " #temp "   \n\t"\
424
"pavgb " #temp ", " #a "          \n\t"\
425
"mov" #size " " #a ", " #b "      \n\t"
426

    
427
QPEL_CAVS(put_,       PUT_OP, 3dnow)
428
QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
429
QPEL_CAVS(put_,       PUT_OP, mmx2)
430
QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)
431

    
432
CAVS_MC(put_, 8, 3dnow)
433
CAVS_MC(put_, 16,3dnow)
434
CAVS_MC(avg_, 8, 3dnow)
435
CAVS_MC(avg_, 16,3dnow)
436
CAVS_MC(put_, 8, mmx2)
437
CAVS_MC(put_, 16,mmx2)
438
CAVS_MC(avg_, 8, mmx2)
439
CAVS_MC(avg_, 16,mmx2)
440

    
441
static void ff_cavsdsp_init_mmx2(CAVSDSPContext* c, AVCodecContext *avctx) {
442
#define dspfunc(PFX, IDX, NUM) \
443
    c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
444
    c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
445
    c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
446
    c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
447
    c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \
448

    
449
    dspfunc(put_cavs_qpel, 0, 16);
450
    dspfunc(put_cavs_qpel, 1, 8);
451
    dspfunc(avg_cavs_qpel, 0, 16);
452
    dspfunc(avg_cavs_qpel, 1, 8);
453
#undef dspfunc
454
    c->cavs_idct8_add = cavs_idct8_add_mmx;
455
}
456

    
457
static void ff_cavsdsp_init_3dnow(CAVSDSPContext* c, AVCodecContext *avctx) {
458
#define dspfunc(PFX, IDX, NUM) \
459
    c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
460
    c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
461
    c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
462
    c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
463
    c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
464

    
465
    dspfunc(put_cavs_qpel, 0, 16);
466
    dspfunc(put_cavs_qpel, 1, 8);
467
    dspfunc(avg_cavs_qpel, 0, 16);
468
    dspfunc(avg_cavs_qpel, 1, 8);
469
#undef dspfunc
470
    c->cavs_idct8_add = cavs_idct8_add_mmx;
471
}
472

    
473
void ff_cavsdsp_init_mmx(CAVSDSPContext *c, AVCodecContext *avctx)
474
{
475
    int mm_flags = mm_support();
476

    
477
    if (mm_flags & FF_MM_MMX2)  ff_cavsdsp_init_mmx2 (c, avctx);
478
    if (mm_flags & FF_MM_3DNOW) ff_cavsdsp_init_3dnow(c, avctx);
479
}