Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / dsputil_mmx.c @ 764ef400

History | View | Annotate | Download (69.7 KB)

1
/*
2
 * MMX optimized DSP utils
3
 * Copyright (c) 2000, 2001 Fabrice Bellard.
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18
 *
19
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
20
 */
21

    
22
#include "../dsputil.h"
23
#include "../simple_idct.h"
24

    
25
int mm_flags; /* multimedia extension flags */
26

    
27
/* pixel operations */
28
static const uint64_t mm_bone __attribute__ ((aligned(8))) = 0x0101010101010101ULL;
29
static const uint64_t mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
30
static const uint64_t mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002ULL;
31

    
32
static const uint64_t ff_pw_20 __attribute__ ((aligned(8))) = 0x0014001400140014ULL;
33
static const uint64_t ff_pw_3  __attribute__ ((aligned(8))) = 0x0003000300030003ULL;
34
static const uint64_t ff_pw_16 __attribute__ ((aligned(8))) = 0x0010001000100010ULL;
35
static const uint64_t ff_pw_15 __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;
36

    
37
#define JUMPALIGN() __asm __volatile (".balign 8"::)
38
#define MOVQ_ZERO(regd)  __asm __volatile ("pxor %%" #regd ", %%" #regd ::)
39

    
40
#define MOVQ_WONE(regd) \
41
    __asm __volatile ( \
42
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
43
    "psrlw $15, %%" #regd ::)
44

    
45
#define MOVQ_BFE(regd) \
46
    __asm __volatile ( \
47
    "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
48
    "paddb %%" #regd ", %%" #regd " \n\t" ::)
49

    
50
#ifndef PIC
51
#define MOVQ_BONE(regd)  __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_bone))
52
#define MOVQ_WTWO(regd)  __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_wtwo))
53
#else
54
// for shared library it's better to use this way for accessing constants
55
// pcmpeqd -> -1
56
#define MOVQ_BONE(regd) \
57
    __asm __volatile ( \
58
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
59
    "psrlw $15, %%" #regd " \n\t" \
60
    "packuswb %%" #regd ", %%" #regd " \n\t" ::)
61

    
62
#define MOVQ_WTWO(regd) \
63
    __asm __volatile ( \
64
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
65
    "psrlw $15, %%" #regd " \n\t" \
66
    "psllw $1, %%" #regd " \n\t"::)
67

    
68
#endif
69

    
70
// using regr as temporary and for the output result
71
// first argument is unmodifed and second is trashed
72
// regfe is supposed to contain 0xfefefefefefefefe
73
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
74
    "movq " #rega ", " #regr "        \n\t"\
75
    "pand " #regb ", " #regr "        \n\t"\
76
    "pxor " #rega ", " #regb "        \n\t"\
77
    "pand " #regfe "," #regb "        \n\t"\
78
    "psrlq $1, " #regb "         \n\t"\
79
    "paddb " #regb ", " #regr "        \n\t"
80

    
81
#define PAVGB_MMX(rega, regb, regr, regfe) \
82
    "movq " #rega ", " #regr "        \n\t"\
83
    "por  " #regb ", " #regr "        \n\t"\
84
    "pxor " #rega ", " #regb "        \n\t"\
85
    "pand " #regfe "," #regb "        \n\t"\
86
    "psrlq $1, " #regb "        \n\t"\
87
    "psubb " #regb ", " #regr "        \n\t"
88

    
89
// mm6 is supposed to contain 0xfefefefefefefefe
90
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \
91
    "movq " #rega ", " #regr "        \n\t"\
92
    "movq " #regc ", " #regp "        \n\t"\
93
    "pand " #regb ", " #regr "        \n\t"\
94
    "pand " #regd ", " #regp "        \n\t"\
95
    "pxor " #rega ", " #regb "        \n\t"\
96
    "pxor " #regc ", " #regd "        \n\t"\
97
    "pand %%mm6, " #regb "        \n\t"\
98
    "pand %%mm6, " #regd "        \n\t"\
99
    "psrlq $1, " #regb "         \n\t"\
100
    "psrlq $1, " #regd "         \n\t"\
101
    "paddb " #regb ", " #regr "        \n\t"\
102
    "paddb " #regd ", " #regp "        \n\t"
103

    
104
#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
105
    "movq " #rega ", " #regr "        \n\t"\
106
    "movq " #regc ", " #regp "        \n\t"\
107
    "por  " #regb ", " #regr "        \n\t"\
108
    "por  " #regd ", " #regp "        \n\t"\
109
    "pxor " #rega ", " #regb "        \n\t"\
110
    "pxor " #regc ", " #regd "        \n\t"\
111
    "pand %%mm6, " #regb "             \n\t"\
112
    "pand %%mm6, " #regd "             \n\t"\
113
    "psrlq $1, " #regd "        \n\t"\
114
    "psrlq $1, " #regb "        \n\t"\
115
    "psubb " #regb ", " #regr "        \n\t"\
116
    "psubb " #regd ", " #regp "        \n\t"
117

    
118
/***********************************/
119
/* MMX no rounding */
120
#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
121
#define SET_RND  MOVQ_WONE
122
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
123
#define PAVGB(a, b, c, e)                PAVGB_MMX_NO_RND(a, b, c, e)
124

    
125
#include "dsputil_mmx_rnd.h"
126

    
127
#undef DEF
128
#undef SET_RND
129
#undef PAVGBP
130
#undef PAVGB
131
/***********************************/
132
/* MMX rounding */
133

    
134
#define DEF(x, y) x ## _ ## y ##_mmx
135
#define SET_RND  MOVQ_WTWO
136
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
137
#define PAVGB(a, b, c, e)                PAVGB_MMX(a, b, c, e)
138

    
139
#include "dsputil_mmx_rnd.h"
140

    
141
#undef DEF
142
#undef SET_RND
143
#undef PAVGBP
144
#undef PAVGB
145

    
146
/***********************************/
147
/* 3Dnow specific */
148

    
149
#define DEF(x) x ## _3dnow
150
/* for Athlons PAVGUSB is prefered */
151
#define PAVGB "pavgusb"
152

    
153
#include "dsputil_mmx_avg.h"
154

    
155
#undef DEF
156
#undef PAVGB
157

    
158
/***********************************/
159
/* MMX2 specific */
160

    
161
#define DEF(x) x ## _mmx2
162

    
163
/* Introduced only in MMX2 set */
164
#define PAVGB "pavgb"
165

    
166
#include "dsputil_mmx_avg.h"
167

    
168
#undef DEF
169
#undef PAVGB
170

    
171
/***********************************/
172
/* standard MMX */
173

    
174
#ifdef CONFIG_ENCODERS
175
static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
176
{
177
    asm volatile(
178
        "movl $-128, %%eax        \n\t"
179
        "pxor %%mm7, %%mm7        \n\t"
180
        ".balign 16                \n\t"
181
        "1:                        \n\t"
182
        "movq (%0), %%mm0        \n\t"
183
        "movq (%0, %2), %%mm2        \n\t"
184
        "movq %%mm0, %%mm1        \n\t"
185
        "movq %%mm2, %%mm3        \n\t"
186
        "punpcklbw %%mm7, %%mm0        \n\t"
187
        "punpckhbw %%mm7, %%mm1        \n\t"
188
        "punpcklbw %%mm7, %%mm2        \n\t"
189
        "punpckhbw %%mm7, %%mm3        \n\t"
190
        "movq %%mm0, (%1, %%eax)\n\t"
191
        "movq %%mm1, 8(%1, %%eax)\n\t"
192
        "movq %%mm2, 16(%1, %%eax)\n\t"
193
        "movq %%mm3, 24(%1, %%eax)\n\t"
194
        "addl %3, %0                \n\t"
195
        "addl $32, %%eax        \n\t"
196
        "js 1b                        \n\t"
197
        : "+r" (pixels)
198
        : "r" (block+64), "r" (line_size), "r" (line_size*2)
199
        : "%eax"
200
    );
201
}
202

    
203
static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
204
{
205
    asm volatile(
206
        "pxor %%mm7, %%mm7        \n\t"
207
        "movl $-128, %%eax        \n\t"
208
        ".balign 16                \n\t"
209
        "1:                        \n\t"
210
        "movq (%0), %%mm0        \n\t"
211
        "movq (%1), %%mm2        \n\t"
212
        "movq %%mm0, %%mm1        \n\t"
213
        "movq %%mm2, %%mm3        \n\t"
214
        "punpcklbw %%mm7, %%mm0        \n\t"
215
        "punpckhbw %%mm7, %%mm1        \n\t"
216
        "punpcklbw %%mm7, %%mm2        \n\t"
217
        "punpckhbw %%mm7, %%mm3        \n\t"
218
        "psubw %%mm2, %%mm0        \n\t"
219
        "psubw %%mm3, %%mm1        \n\t"
220
        "movq %%mm0, (%2, %%eax)\n\t"
221
        "movq %%mm1, 8(%2, %%eax)\n\t"
222
        "addl %3, %0                \n\t"
223
        "addl %3, %1                \n\t"
224
        "addl $16, %%eax        \n\t"
225
        "jnz 1b                        \n\t"
226
        : "+r" (s1), "+r" (s2)
227
        : "r" (block+64), "r" (stride)
228
        : "%eax"
229
    );
230
}
231
#endif //CONFIG_ENCODERS
232

    
233
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
234
{
235
    const DCTELEM *p;
236
    uint8_t *pix;
237

    
238
    /* read the pixels */
239
    p = block;
240
    pix = pixels;
241
    /* unrolled loop */
242
        __asm __volatile(
243
                "movq        %3, %%mm0\n\t"
244
                "movq        8%3, %%mm1\n\t"
245
                "movq        16%3, %%mm2\n\t"
246
                "movq        24%3, %%mm3\n\t"
247
                "movq        32%3, %%mm4\n\t"
248
                "movq        40%3, %%mm5\n\t"
249
                "movq        48%3, %%mm6\n\t"
250
                "movq        56%3, %%mm7\n\t"
251
                "packuswb %%mm1, %%mm0\n\t"
252
                "packuswb %%mm3, %%mm2\n\t"
253
                "packuswb %%mm5, %%mm4\n\t"
254
                "packuswb %%mm7, %%mm6\n\t"
255
                "movq        %%mm0, (%0)\n\t"
256
                "movq        %%mm2, (%0, %1)\n\t"
257
                "movq        %%mm4, (%0, %1, 2)\n\t"
258
                "movq        %%mm6, (%0, %2)\n\t"
259
                ::"r" (pix), "r" (line_size), "r" (line_size*3), "m"(*p)
260
                :"memory");
261
        pix += line_size*4;
262
        p += 32;
263

    
264
    // if here would be an exact copy of the code above
265
    // compiler would generate some very strange code
266
    // thus using "r"
267
    __asm __volatile(
268
            "movq        (%3), %%mm0\n\t"
269
            "movq        8(%3), %%mm1\n\t"
270
            "movq        16(%3), %%mm2\n\t"
271
            "movq        24(%3), %%mm3\n\t"
272
            "movq        32(%3), %%mm4\n\t"
273
            "movq        40(%3), %%mm5\n\t"
274
            "movq        48(%3), %%mm6\n\t"
275
            "movq        56(%3), %%mm7\n\t"
276
            "packuswb %%mm1, %%mm0\n\t"
277
            "packuswb %%mm3, %%mm2\n\t"
278
            "packuswb %%mm5, %%mm4\n\t"
279
            "packuswb %%mm7, %%mm6\n\t"
280
            "movq        %%mm0, (%0)\n\t"
281
            "movq        %%mm2, (%0, %1)\n\t"
282
            "movq        %%mm4, (%0, %1, 2)\n\t"
283
            "movq        %%mm6, (%0, %2)\n\t"
284
            ::"r" (pix), "r" (line_size), "r" (line_size*3), "r"(p)
285
            :"memory");
286
}
287

    
288
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
289
{
290
    const DCTELEM *p;
291
    uint8_t *pix;
292
    int i;
293

    
294
    /* read the pixels */
295
    p = block;
296
    pix = pixels;
297
    MOVQ_ZERO(mm7);
298
    i = 4;
299
    do {
300
        __asm __volatile(
301
                "movq        (%2), %%mm0\n\t"
302
                "movq        8(%2), %%mm1\n\t"
303
                "movq        16(%2), %%mm2\n\t"
304
                "movq        24(%2), %%mm3\n\t"
305
                "movq        %0, %%mm4\n\t"
306
                "movq        %1, %%mm6\n\t"
307
                "movq        %%mm4, %%mm5\n\t"
308
                "punpcklbw %%mm7, %%mm4\n\t"
309
                "punpckhbw %%mm7, %%mm5\n\t"
310
                "paddsw        %%mm4, %%mm0\n\t"
311
                "paddsw        %%mm5, %%mm1\n\t"
312
                "movq        %%mm6, %%mm5\n\t"
313
                "punpcklbw %%mm7, %%mm6\n\t"
314
                "punpckhbw %%mm7, %%mm5\n\t"
315
                "paddsw        %%mm6, %%mm2\n\t"
316
                "paddsw        %%mm5, %%mm3\n\t"
317
                "packuswb %%mm1, %%mm0\n\t"
318
                "packuswb %%mm3, %%mm2\n\t"
319
                "movq        %%mm0, %0\n\t"
320
                "movq        %%mm2, %1\n\t"
321
                :"+m"(*pix), "+m"(*(pix+line_size))
322
                :"r"(p)
323
                :"memory");
324
        pix += line_size*2;
325
        p += 16;
326
    } while (--i);
327
}
328

    
329
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
330
{
331
    __asm __volatile(
332
         "lea (%3, %3), %%eax                \n\t"
333
         ".balign 8                        \n\t"
334
         "1:                                \n\t"
335
         "movq (%1), %%mm0                \n\t"
336
         "movq (%1, %3), %%mm1                \n\t"
337
              "movq %%mm0, (%2)                \n\t"
338
         "movq %%mm1, (%2, %3)                \n\t"
339
         "addl %%eax, %1                \n\t"
340
         "addl %%eax, %2                       \n\t"
341
         "movq (%1), %%mm0                \n\t"
342
         "movq (%1, %3), %%mm1                \n\t"
343
         "movq %%mm0, (%2)                \n\t"
344
         "movq %%mm1, (%2, %3)                \n\t"
345
         "addl %%eax, %1                \n\t"
346
         "addl %%eax, %2                       \n\t"
347
         "subl $4, %0                        \n\t"
348
         "jnz 1b                        \n\t"
349
         : "+g"(h), "+r" (pixels),  "+r" (block)
350
         : "r"(line_size)
351
         : "%eax", "memory"
352
        );
353
}
354

    
355
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
356
{
357
    __asm __volatile(
358
         "lea (%3, %3), %%eax                \n\t"
359
         ".balign 8                        \n\t"
360
         "1:                                \n\t"
361
         "movq (%1), %%mm0                \n\t"
362
         "movq 8(%1), %%mm4                \n\t"
363
         "movq (%1, %3), %%mm1                \n\t"
364
         "movq 8(%1, %3), %%mm5                \n\t"
365
              "movq %%mm0, (%2)                \n\t"
366
              "movq %%mm4, 8(%2)                \n\t"
367
         "movq %%mm1, (%2, %3)                \n\t"
368
         "movq %%mm5, 8(%2, %3)                \n\t"
369
         "addl %%eax, %1                \n\t"
370
         "addl %%eax, %2                       \n\t"
371
         "movq (%1), %%mm0                \n\t"
372
         "movq 8(%1), %%mm4                \n\t"
373
         "movq (%1, %3), %%mm1                \n\t"
374
         "movq 8(%1, %3), %%mm5                \n\t"
375
         "movq %%mm0, (%2)                \n\t"
376
         "movq %%mm4, 8(%2)                \n\t"
377
         "movq %%mm1, (%2, %3)                \n\t"
378
         "movq %%mm5, 8(%2, %3)                \n\t"
379
         "addl %%eax, %1                \n\t"
380
         "addl %%eax, %2                       \n\t"
381
         "subl $4, %0                        \n\t"
382
         "jnz 1b                        \n\t"
383
         : "+g"(h), "+r" (pixels),  "+r" (block)
384
         : "r"(line_size)
385
         : "%eax", "memory"
386
        );
387
}
388

    
389
static void clear_blocks_mmx(DCTELEM *blocks)
390
{
391
    __asm __volatile(
392
                "pxor %%mm7, %%mm7                \n\t"
393
                "movl $-128*6, %%eax                \n\t"
394
                "1:                                \n\t"
395
                "movq %%mm7, (%0, %%eax)        \n\t"
396
                "movq %%mm7, 8(%0, %%eax)        \n\t"
397
                "movq %%mm7, 16(%0, %%eax)        \n\t"
398
                "movq %%mm7, 24(%0, %%eax)        \n\t"
399
                "addl $32, %%eax                \n\t"
400
                " js 1b                                \n\t"
401
                : : "r" (((int)blocks)+128*6)
402
                : "%eax"
403
        );
404
}
405

    
406
#ifdef CONFIG_ENCODERS
407
static int pix_sum16_mmx(uint8_t * pix, int line_size){
408
    const int h=16;
409
    int sum;
410
    int index= -line_size*h;
411

    
412
    __asm __volatile(
413
                "pxor %%mm7, %%mm7                \n\t"
414
                "pxor %%mm6, %%mm6                \n\t"
415
                "1:                                \n\t"
416
                "movq (%2, %1), %%mm0                \n\t"
417
                "movq (%2, %1), %%mm1                \n\t"
418
                "movq 8(%2, %1), %%mm2                \n\t"
419
                "movq 8(%2, %1), %%mm3                \n\t"
420
                "punpcklbw %%mm7, %%mm0                \n\t"
421
                "punpckhbw %%mm7, %%mm1                \n\t"
422
                "punpcklbw %%mm7, %%mm2                \n\t"
423
                "punpckhbw %%mm7, %%mm3                \n\t"
424
                "paddw %%mm0, %%mm1                \n\t"
425
                "paddw %%mm2, %%mm3                \n\t"
426
                "paddw %%mm1, %%mm3                \n\t"
427
                "paddw %%mm3, %%mm6                \n\t"
428
                "addl %3, %1                        \n\t"
429
                " js 1b                                \n\t"
430
                "movq %%mm6, %%mm5                \n\t"
431
                "psrlq $32, %%mm6                \n\t"
432
                "paddw %%mm5, %%mm6                \n\t"
433
                "movq %%mm6, %%mm5                \n\t"
434
                "psrlq $16, %%mm6                \n\t"
435
                "paddw %%mm5, %%mm6                \n\t"
436
                "movd %%mm6, %0                        \n\t"
437
                "andl $0xFFFF, %0                \n\t"
438
                : "=&r" (sum), "+r" (index)
439
                : "r" (pix - index), "r" (line_size)
440
        );
441

    
442
        return sum;
443
}
444
#endif //CONFIG_ENCODERS
445

    
446
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
447
    int i=0;
448
    asm volatile(
449
        "1:                                \n\t"
450
        "movq  (%1, %0), %%mm0                \n\t"
451
        "movq  (%2, %0), %%mm1                \n\t"
452
        "paddb %%mm0, %%mm1                \n\t"
453
        "movq %%mm1, (%2, %0)                \n\t"
454
        "movq 8(%1, %0), %%mm0                \n\t"
455
        "movq 8(%2, %0), %%mm1                \n\t"
456
        "paddb %%mm0, %%mm1                \n\t"
457
        "movq %%mm1, 8(%2, %0)                \n\t"
458
        "addl $16, %0                        \n\t"
459
        "cmpl %3, %0                        \n\t"
460
        " jb 1b                                \n\t"
461
        : "+r" (i)
462
        : "r"(src), "r"(dst), "r"(w-15)
463
    );
464
    for(; i<w; i++)
465
        dst[i+0] += src[i+0];
466
}
467

    
468
#ifdef CONFIG_ENCODERS
469
static int pix_norm1_mmx(uint8_t *pix, int line_size) {
470
    int tmp;
471
  asm volatile (
472
      "movl $16,%%ecx\n"
473
      "pxor %%mm0,%%mm0\n"
474
      "pxor %%mm7,%%mm7\n"
475
      "1:\n"
476
      "movq (%0),%%mm2\n"        /* mm2 = pix[0-7] */
477
      "movq 8(%0),%%mm3\n"        /* mm3 = pix[8-15] */
478

    
479
      "movq %%mm2,%%mm1\n"        /* mm1 = mm2 = pix[0-7] */
480

    
481
      "punpckhbw %%mm0,%%mm1\n"        /* mm1 = [pix4-7] */
482
      "punpcklbw %%mm0,%%mm2\n"        /* mm2 = [pix0-3] */
483

    
484
      "movq %%mm3,%%mm4\n"        /* mm4 = mm3 = pix[8-15] */
485
      "punpckhbw %%mm0,%%mm3\n"        /* mm3 = [pix12-15] */
486
      "punpcklbw %%mm0,%%mm4\n"        /* mm4 = [pix8-11] */
487

    
488
      "pmaddwd %%mm1,%%mm1\n"        /* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
489
      "pmaddwd %%mm2,%%mm2\n"        /* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */
490

    
491
      "pmaddwd %%mm3,%%mm3\n"
492
      "pmaddwd %%mm4,%%mm4\n"
493

    
494
      "paddd %%mm1,%%mm2\n"        /* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
495
                                          pix2^2+pix3^2+pix6^2+pix7^2) */
496
      "paddd %%mm3,%%mm4\n"
497
      "paddd %%mm2,%%mm7\n"
498

    
499
      "addl %2, %0\n"
500
      "paddd %%mm4,%%mm7\n"
501
      "dec %%ecx\n"
502
      "jnz 1b\n"
503

    
504
      "movq %%mm7,%%mm1\n"
505
      "psrlq $32, %%mm7\n"        /* shift hi dword to lo */
506
      "paddd %%mm7,%%mm1\n"
507
      "movd %%mm1,%1\n"
508
      : "+r" (pix), "=r"(tmp) : "r" (line_size) : "%ecx" );
509
    return tmp;
510
}
511

    
512
static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size) {
513
    int tmp;
514
  asm volatile (
515
      "movl $16,%%ecx\n"
516
      "pxor %%mm0,%%mm0\n"        /* mm0 = 0 */
517
      "pxor %%mm7,%%mm7\n"        /* mm7 holds the sum */
518
      "1:\n"
519
      "movq (%0),%%mm1\n"        /* mm1 = pix1[0-7] */
520
      "movq (%1),%%mm2\n"        /* mm2 = pix2[0-7] */
521
      "movq 8(%0),%%mm3\n"        /* mm3 = pix1[8-15] */
522
      "movq 8(%1),%%mm4\n"        /* mm4 = pix2[8-15] */
523

    
524
      /* todo: mm1-mm2, mm3-mm4 */
525
      /* algo: substract mm1 from mm2 with saturation and vice versa */
526
      /*       OR the results to get absolute difference */
527
      "movq %%mm1,%%mm5\n"
528
      "movq %%mm3,%%mm6\n"
529
      "psubusb %%mm2,%%mm1\n"
530
      "psubusb %%mm4,%%mm3\n"
531
      "psubusb %%mm5,%%mm2\n"
532
      "psubusb %%mm6,%%mm4\n"
533

    
534
      "por %%mm1,%%mm2\n"
535
      "por %%mm3,%%mm4\n"
536

    
537
      /* now convert to 16-bit vectors so we can square them */
538
      "movq %%mm2,%%mm1\n"
539
      "movq %%mm4,%%mm3\n"
540

    
541
      "punpckhbw %%mm0,%%mm2\n"
542
      "punpckhbw %%mm0,%%mm4\n"
543
      "punpcklbw %%mm0,%%mm1\n"        /* mm1 now spread over (mm1,mm2) */
544
      "punpcklbw %%mm0,%%mm3\n"        /* mm4 now spread over (mm3,mm4) */
545

    
546
      "pmaddwd %%mm2,%%mm2\n"
547
      "pmaddwd %%mm4,%%mm4\n"
548
      "pmaddwd %%mm1,%%mm1\n"
549
      "pmaddwd %%mm3,%%mm3\n"
550

    
551
      "addl %3,%0\n"
552
      "addl %3,%1\n"
553

    
554
      "paddd %%mm2,%%mm1\n"
555
      "paddd %%mm4,%%mm3\n"
556
      "paddd %%mm1,%%mm7\n"
557
      "paddd %%mm3,%%mm7\n"
558

    
559
      "decl %%ecx\n"
560
      "jnz 1b\n"
561

    
562
      "movq %%mm7,%%mm1\n"
563
      "psrlq $32, %%mm7\n"        /* shift hi dword to lo */
564
      "paddd %%mm7,%%mm1\n"
565
      "movd %%mm1,%2\n"
566
      : "+r" (pix1), "+r" (pix2), "=r"(tmp) : "r" (line_size) : "ecx");
567
    return tmp;
568
}
569

    
570
static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
571
    int i=0;
572
    asm volatile(
573
        "1:                                \n\t"
574
        "movq  (%2, %0), %%mm0                \n\t"
575
        "movq  (%1, %0), %%mm1                \n\t"
576
        "psubb %%mm0, %%mm1                \n\t"
577
        "movq %%mm1, (%3, %0)                \n\t"
578
        "movq 8(%2, %0), %%mm0                \n\t"
579
        "movq 8(%1, %0), %%mm1                \n\t"
580
        "psubb %%mm0, %%mm1                \n\t"
581
        "movq %%mm1, 8(%3, %0)                \n\t"
582
        "addl $16, %0                        \n\t"
583
        "cmpl %4, %0                        \n\t"
584
        " jb 1b                                \n\t"
585
        : "+r" (i)
586
        : "r"(src1), "r"(src2), "r"(dst), "r"(w-15)
587
    );
588
    for(; i<w; i++)
589
        dst[i+0] = src1[i+0]-src2[i+0];
590
}
591

    
592
static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){
593
    int i=0;
594
    uint8_t l, lt;
595

    
596
    asm volatile(
597
        "1:                                \n\t"
598
        "movq  -1(%1, %0), %%mm0        \n\t" // LT
599
        "movq  (%1, %0), %%mm1                \n\t" // T
600
        "movq  -1(%2, %0), %%mm2        \n\t" // L
601
        "movq  (%2, %0), %%mm3                \n\t" // X
602
        "movq %%mm2, %%mm4                \n\t" // L
603
        "psubb %%mm0, %%mm2                \n\t"
604
        "paddb %%mm1, %%mm2                \n\t" // L + T - LT
605
        "movq %%mm4, %%mm5                \n\t" // L
606
        "pmaxub %%mm1, %%mm4                \n\t" // max(T, L)
607
        "pminub %%mm5, %%mm1                \n\t" // min(T, L)
608
        "pminub %%mm2, %%mm4                \n\t" 
609
        "pmaxub %%mm1, %%mm4                \n\t"
610
        "psubb %%mm4, %%mm3                \n\t" // dst - pred
611
        "movq %%mm3, (%3, %0)                \n\t"
612
        "addl $8, %0                        \n\t"
613
        "cmpl %4, %0                        \n\t"
614
        " jb 1b                                \n\t"
615
        : "+r" (i)
616
        : "r"(src1), "r"(src2), "r"(dst), "r"(w)
617
    );
618

    
619
    l= *left;
620
    lt= *left_top;
621
    
622
    dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
623
    
624
    *left_top= src1[w-1];
625
    *left    = src2[w-1];
626
}
627

    
628
#define LBUTTERFLY2(a1,b1,a2,b2)\
629
    "paddw " #b1 ", " #a1 "                \n\t"\
630
    "paddw " #b2 ", " #a2 "                \n\t"\
631
    "paddw " #b1 ", " #b1 "                \n\t"\
632
    "paddw " #b2 ", " #b2 "                \n\t"\
633
    "psubw " #a1 ", " #b1 "                \n\t"\
634
    "psubw " #a2 ", " #b2 "                \n\t"
635

    
636
#define HADAMARD48\
637
        LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\
638
        LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\
639
        LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\
640
        LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\
641
        LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\
642
        LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\
643

    
644
#define MMABS(a,z)\
645
    "pxor " #z ", " #z "                \n\t"\
646
    "pcmpgtw " #a ", " #z "                \n\t"\
647
    "pxor " #z ", " #a "                \n\t"\
648
    "psubw " #z ", " #a "                \n\t"
649

    
650
#define MMABS_SUM(a,z, sum)\
651
    "pxor " #z ", " #z "                \n\t"\
652
    "pcmpgtw " #a ", " #z "                \n\t"\
653
    "pxor " #z ", " #a "                \n\t"\
654
    "psubw " #z ", " #a "                \n\t"\
655
    "paddusw " #a ", " #sum "                \n\t"
656

    
657
#define MMABS_MMX2(a,z)\
658
    "pxor " #z ", " #z "                \n\t"\
659
    "psubw " #a ", " #z "                \n\t"\
660
    "pmaxsw " #z ", " #a "                \n\t"
661

    
662
#define MMABS_SUM_MMX2(a,z, sum)\
663
    "pxor " #z ", " #z "                \n\t"\
664
    "psubw " #a ", " #z "                \n\t"\
665
    "pmaxsw " #z ", " #a "                \n\t"\
666
    "paddusw " #a ", " #sum "                \n\t"
667
        
668
#define SBUTTERFLY(a,b,t,n)\
669
    "movq " #a ", " #t "                \n\t" /* abcd */\
670
    "punpckl" #n " " #b ", " #a "        \n\t" /* aebf */\
671
    "punpckh" #n " " #b ", " #t "        \n\t" /* cgdh */\
672

    
673
#define TRANSPOSE4(a,b,c,d,t)\
674
    SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
675
    SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
676
    SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
677
    SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
678

    
679
#define LOAD4(o, a, b, c, d)\
680
        "movq "#o"(%1), " #a "                \n\t"\
681
        "movq "#o"+16(%1), " #b "        \n\t"\
682
        "movq "#o"+32(%1), " #c "        \n\t"\
683
        "movq "#o"+48(%1), " #d "        \n\t"
684

    
685
#define STORE4(o, a, b, c, d)\
686
        "movq "#a", "#o"(%1)                \n\t"\
687
        "movq "#b", "#o"+16(%1)                \n\t"\
688
        "movq "#c", "#o"+32(%1)                \n\t"\
689
        "movq "#d", "#o"+48(%1)                \n\t"\
690

    
691
static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride){
692
    uint64_t temp[16] __align8;
693
    int sum=0;
694

    
695
    diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);
696

    
697
    asm volatile(
698
        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
699
        LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)
700
        
701
        HADAMARD48
702
        
703
        "movq %%mm7, 112(%1)                \n\t"
704
        
705
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
706
        STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)
707
        
708
        "movq 112(%1), %%mm7                 \n\t"
709
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
710
        STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)
711

    
712
        LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
713
        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
714
        
715
        HADAMARD48
716
        
717
        "movq %%mm7, 120(%1)                \n\t"
718
        
719
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
720
        STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)
721
        
722
        "movq 120(%1), %%mm7                 \n\t"
723
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
724
        "movq %%mm7, %%mm5                \n\t"//FIXME remove
725
        "movq %%mm6, %%mm7                \n\t"
726
        "movq %%mm0, %%mm6                \n\t"
727
//        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove
728
        
729
        LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
730
//        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
731
        
732
        HADAMARD48
733
        "movq %%mm7, 64(%1)                \n\t"
734
        MMABS(%%mm0, %%mm7)
735
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
736
        MMABS_SUM(%%mm2, %%mm7, %%mm0)
737
        MMABS_SUM(%%mm3, %%mm7, %%mm0)
738
        MMABS_SUM(%%mm4, %%mm7, %%mm0)
739
        MMABS_SUM(%%mm5, %%mm7, %%mm0)
740
        MMABS_SUM(%%mm6, %%mm7, %%mm0)
741
        "movq 64(%1), %%mm1                \n\t"
742
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
743
        "movq %%mm0, 64(%1)                \n\t"
744
        
745
        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
746
        LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)
747
        
748
        HADAMARD48
749
        "movq %%mm7, (%1)                \n\t"
750
        MMABS(%%mm0, %%mm7)
751
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
752
        MMABS_SUM(%%mm2, %%mm7, %%mm0)
753
        MMABS_SUM(%%mm3, %%mm7, %%mm0)
754
        MMABS_SUM(%%mm4, %%mm7, %%mm0)
755
        MMABS_SUM(%%mm5, %%mm7, %%mm0)
756
        MMABS_SUM(%%mm6, %%mm7, %%mm0)
757
        "movq (%1), %%mm1                \n\t"
758
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
759
        "movq 64(%1), %%mm1                \n\t"
760
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
761
        
762
        "movq %%mm0, %%mm1                \n\t"
763
        "psrlq $32, %%mm0                \n\t"
764
        "paddusw %%mm1, %%mm0                \n\t"
765
        "movq %%mm0, %%mm1                \n\t"
766
        "psrlq $16, %%mm0                \n\t"
767
        "paddusw %%mm1, %%mm0                \n\t"
768
        "movd %%mm0, %0                        \n\t"
769
                
770
        : "=r" (sum)
771
        : "r"(temp)
772
    );
773
    return sum&0xFFFF;
774
}
775

    
776
static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride){
777
    uint64_t temp[16] __align8;
778
    int sum=0;
779

    
780
    diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);
781

    
782
    asm volatile(
783
        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
784
        LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)
785
        
786
        HADAMARD48
787
        
788
        "movq %%mm7, 112(%1)                \n\t"
789
        
790
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
791
        STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)
792
        
793
        "movq 112(%1), %%mm7                 \n\t"
794
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
795
        STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)
796

    
797
        LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
798
        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
799
        
800
        HADAMARD48
801
        
802
        "movq %%mm7, 120(%1)                \n\t"
803
        
804
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
805
        STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)
806
        
807
        "movq 120(%1), %%mm7                 \n\t"
808
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
809
        "movq %%mm7, %%mm5                \n\t"//FIXME remove
810
        "movq %%mm6, %%mm7                \n\t"
811
        "movq %%mm0, %%mm6                \n\t"
812
//        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove
813
        
814
        LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
815
//        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
816
        
817
        HADAMARD48
818
        "movq %%mm7, 64(%1)                \n\t"
819
        MMABS_MMX2(%%mm0, %%mm7)
820
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
821
        MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)
822
        MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)
823
        MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)
824
        MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)
825
        MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)
826
        "movq 64(%1), %%mm1                \n\t"
827
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
828
        "movq %%mm0, 64(%1)                \n\t"
829
        
830
        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
831
        LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)
832
        
833
        HADAMARD48
834
        "movq %%mm7, (%1)                \n\t"
835
        MMABS_MMX2(%%mm0, %%mm7)
836
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
837
        MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)
838
        MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)
839
        MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)
840
        MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)
841
        MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)
842
        "movq (%1), %%mm1                \n\t"
843
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
844
        "movq 64(%1), %%mm1                \n\t"
845
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
846
        
847
        "movq %%mm0, %%mm1                \n\t"
848
        "psrlq $32, %%mm0                \n\t"
849
        "paddusw %%mm1, %%mm0                \n\t"
850
        "movq %%mm0, %%mm1                \n\t"
851
        "psrlq $16, %%mm0                \n\t"
852
        "paddusw %%mm1, %%mm0                \n\t"
853
        "movd %%mm0, %0                        \n\t"
854
                
855
        : "=r" (sum)
856
        : "r"(temp)
857
    );
858
    return sum&0xFFFF;
859
}
860

    
861

    
862
WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)
863
WARPER88_1616(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)
864
#endif //CONFIG_ENCODERS
865

    
866
#define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
867
#define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)
868

    
869
#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
870
        "paddw " #m4 ", " #m3 "                \n\t" /* x1 */\
871
        "movq "MANGLE(ff_pw_20)", %%mm4                \n\t" /* 20 */\
872
        "pmullw " #m3 ", %%mm4                \n\t" /* 20x1 */\
873
        "movq "#in7", " #m3 "                \n\t" /* d */\
874
        "movq "#in0", %%mm5                \n\t" /* D */\
875
        "paddw " #m3 ", %%mm5                \n\t" /* x4 */\
876
        "psubw %%mm5, %%mm4                \n\t" /* 20x1 - x4 */\
877
        "movq "#in1", %%mm5                \n\t" /* C */\
878
        "movq "#in2", %%mm6                \n\t" /* B */\
879
        "paddw " #m6 ", %%mm5                \n\t" /* x3 */\
880
        "paddw " #m5 ", %%mm6                \n\t" /* x2 */\
881
        "paddw %%mm6, %%mm6                \n\t" /* 2x2 */\
882
        "psubw %%mm6, %%mm5                \n\t" /* -2x2 + x3 */\
883
        "pmullw "MANGLE(ff_pw_3)", %%mm5        \n\t" /* -6x2 + 3x3 */\
884
        "paddw " #rnd ", %%mm4                \n\t" /* x2 */\
885
        "paddw %%mm4, %%mm5                \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
886
        "psraw $5, %%mm5                \n\t"\
887
        "packuswb %%mm5, %%mm5                \n\t"\
888
        OP(%%mm5, out, %%mm7, d)
889

    
890
#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
891
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
892
    uint64_t temp;\
893
\
894
    asm volatile(\
895
        "pxor %%mm7, %%mm7                \n\t"\
896
        "1:                                \n\t"\
897
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
898
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
899
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
900
        "punpcklbw %%mm7, %%mm0                \n\t" /* 0A0B0C0D */\
901
        "punpckhbw %%mm7, %%mm1                \n\t" /* 0E0F0G0H */\
902
        "pshufw $0x90, %%mm0, %%mm5        \n\t" /* 0A0A0B0C */\
903
        "pshufw $0x41, %%mm0, %%mm6        \n\t" /* 0B0A0A0B */\
904
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
905
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
906
        "psllq $8, %%mm2                \n\t" /* 0ABCDEFG */\
907
        "psllq $16, %%mm3                \n\t" /* 00ABCDEF */\
908
        "psllq $24, %%mm4                \n\t" /* 000ABCDE */\
909
        "punpckhbw %%mm7, %%mm2                \n\t" /* 0D0E0F0G */\
910
        "punpckhbw %%mm7, %%mm3                \n\t" /* 0C0D0E0F */\
911
        "punpckhbw %%mm7, %%mm4                \n\t" /* 0B0C0D0E */\
912
        "paddw %%mm3, %%mm5                \n\t" /* b */\
913
        "paddw %%mm2, %%mm6                \n\t" /* c */\
914
        "paddw %%mm5, %%mm5                \n\t" /* 2b */\
915
        "psubw %%mm5, %%mm6                \n\t" /* c - 2b */\
916
        "pshufw $0x06, %%mm0, %%mm5        \n\t" /* 0C0B0A0A */\
917
        "pmullw "MANGLE(ff_pw_3)", %%mm6                \n\t" /* 3c - 6b */\
918
        "paddw %%mm4, %%mm0                \n\t" /* a */\
919
        "paddw %%mm1, %%mm5                \n\t" /* d */\
920
        "pmullw "MANGLE(ff_pw_20)", %%mm0                \n\t" /* 20a */\
921
        "psubw %%mm5, %%mm0                \n\t" /* 20a - d */\
922
        "paddw %6, %%mm6                \n\t"\
923
        "paddw %%mm6, %%mm0                \n\t" /* 20a - 6b + 3c - d */\
924
        "psraw $5, %%mm0                \n\t"\
925
        "movq %%mm0, %5                        \n\t"\
926
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
927
        \
928
        "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\
929
        "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\
930
        "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\
931
        "psrlq $8, %%mm0                \n\t" /* GHIJKLM0 */\
932
        "psrlq $16, %%mm5                \n\t" /* HIJKLM00 */\
933
        "punpcklbw %%mm7, %%mm0                \n\t" /* 0G0H0I0J */\
934
        "punpcklbw %%mm7, %%mm5                \n\t" /* 0H0I0J0K */\
935
        "paddw %%mm0, %%mm2                \n\t" /* b */\
936
        "paddw %%mm5, %%mm3                \n\t" /* c */\
937
        "paddw %%mm2, %%mm2                \n\t" /* 2b */\
938
        "psubw %%mm2, %%mm3                \n\t" /* c - 2b */\
939
        "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\
940
        "psrlq $24, %%mm6                \n\t" /* IJKLM000 */\
941
        "punpcklbw %%mm7, %%mm2                \n\t" /* 0F0G0H0I */\
942
        "punpcklbw %%mm7, %%mm6                \n\t" /* 0I0J0K0L */\
943
        "pmullw "MANGLE(ff_pw_3)", %%mm3                \n\t" /* 3c - 6b */\
944
        "paddw %%mm2, %%mm1                \n\t" /* a */\
945
        "paddw %%mm6, %%mm4                \n\t" /* d */\
946
        "pmullw "MANGLE(ff_pw_20)", %%mm1                \n\t" /* 20a */\
947
        "psubw %%mm4, %%mm3                \n\t" /* - 6b +3c - d */\
948
        "paddw %6, %%mm1                \n\t"\
949
        "paddw %%mm1, %%mm3                \n\t" /* 20a - 6b +3c - d */\
950
        "psraw $5, %%mm3                \n\t"\
951
        "movq %5, %%mm1                        \n\t"\
952
        "packuswb %%mm3, %%mm1                \n\t"\
953
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
954
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
955
        \
956
        "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\
957
        "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\
958
        "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\
959
        "psrlq $8, %%mm1                \n\t" /* KLMNOPQ0 */\
960
        "psrlq $16, %%mm4                \n\t" /* LMNOPQ00 */\
961
        "punpcklbw %%mm7, %%mm1                \n\t" /* 0K0L0M0N */\
962
        "punpcklbw %%mm7, %%mm4                \n\t" /* 0L0M0N0O */\
963
        "paddw %%mm1, %%mm5                \n\t" /* b */\
964
        "paddw %%mm4, %%mm0                \n\t" /* c */\
965
        "paddw %%mm5, %%mm5                \n\t" /* 2b */\
966
        "psubw %%mm5, %%mm0                \n\t" /* c - 2b */\
967
        "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\
968
        "psrlq $24, %%mm3                \n\t" /* MNOPQ000 */\
969
        "pmullw "MANGLE(ff_pw_3)", %%mm0                \n\t" /* 3c - 6b */\
970
        "punpcklbw %%mm7, %%mm3                \n\t" /* 0M0N0O0P */\
971
        "paddw %%mm3, %%mm2                \n\t" /* d */\
972
        "psubw %%mm2, %%mm0                \n\t" /* -6b + 3c - d */\
973
        "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\
974
        "punpcklbw %%mm7, %%mm2                \n\t" /* 0J0K0L0M */\
975
        "punpckhbw %%mm7, %%mm5                \n\t" /* 0N0O0P0Q */\
976
        "paddw %%mm2, %%mm6                \n\t" /* a */\
977
        "pmullw "MANGLE(ff_pw_20)", %%mm6                \n\t" /* 20a */\
978
        "paddw %6, %%mm0                \n\t"\
979
        "paddw %%mm6, %%mm0                \n\t" /* 20a - 6b + 3c - d */\
980
        "psraw $5, %%mm0                \n\t"\
981
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
982
        \
983
        "paddw %%mm5, %%mm3                \n\t" /* a */\
984
        "pshufw $0xF9, %%mm5, %%mm6        \n\t" /* 0O0P0Q0Q */\
985
        "paddw %%mm4, %%mm6                \n\t" /* b */\
986
        "pshufw $0xBE, %%mm5, %%mm4        \n\t" /* 0P0Q0Q0P */\
987
        "pshufw $0x6F, %%mm5, %%mm5        \n\t" /* 0Q0Q0P0O */\
988
        "paddw %%mm1, %%mm4                \n\t" /* c */\
989
        "paddw %%mm2, %%mm5                \n\t" /* d */\
990
        "paddw %%mm6, %%mm6                \n\t" /* 2b */\
991
        "psubw %%mm6, %%mm4                \n\t" /* c - 2b */\
992
        "pmullw "MANGLE(ff_pw_20)", %%mm3                \n\t" /* 20a */\
993
        "pmullw "MANGLE(ff_pw_3)", %%mm4                \n\t" /* 3c - 6b */\
994
        "psubw %%mm5, %%mm3                \n\t" /* -6b + 3c - d */\
995
        "paddw %6, %%mm4                \n\t"\
996
        "paddw %%mm3, %%mm4                \n\t" /* 20a - 6b + 3c - d */\
997
        "psraw $5, %%mm4                \n\t"\
998
        "packuswb %%mm4, %%mm0                \n\t"\
999
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
1000
        \
1001
        "addl %3, %0                        \n\t"\
1002
        "addl %4, %1                        \n\t"\
1003
        "decl %2                        \n\t"\
1004
        " jnz 1b                                \n\t"\
1005
        : "+a"(src), "+c"(dst), "+m"(h)\
1006
        : "d"(srcStride), "S"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1007
        : "memory"\
1008
    );\
1009
}\
1010
\
1011
static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1012
    int i;\
1013
    int16_t temp[16];\
1014
    /* quick HACK, XXX FIXME MUST be optimized */\
1015
    for(i=0; i<h; i++)\
1016
    {\
1017
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1018
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1019
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1020
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1021
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1022
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1023
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1024
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1025
        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1026
        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1027
        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1028
        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1029
        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1030
        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1031
        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1032
        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1033
        asm volatile(\
1034
            "movq (%0), %%mm0                \n\t"\
1035
            "movq 8(%0), %%mm1                \n\t"\
1036
            "paddw %2, %%mm0                \n\t"\
1037
            "paddw %2, %%mm1                \n\t"\
1038
            "psraw $5, %%mm0                \n\t"\
1039
            "psraw $5, %%mm1                \n\t"\
1040
            "packuswb %%mm1, %%mm0        \n\t"\
1041
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1042
            "movq 16(%0), %%mm0                \n\t"\
1043
            "movq 24(%0), %%mm1                \n\t"\
1044
            "paddw %2, %%mm0                \n\t"\
1045
            "paddw %2, %%mm1                \n\t"\
1046
            "psraw $5, %%mm0                \n\t"\
1047
            "psraw $5, %%mm1                \n\t"\
1048
            "packuswb %%mm1, %%mm0        \n\t"\
1049
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1050
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1051
            : "memory"\
1052
        );\
1053
        dst+=dstStride;\
1054
        src+=srcStride;\
1055
    }\
1056
}\
1057
\
1058
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1059
    uint64_t temp;\
1060
\
1061
    asm volatile(\
1062
        "pxor %%mm7, %%mm7                \n\t"\
1063
        "1:                                \n\t"\
1064
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
1065
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
1066
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1067
        "punpcklbw %%mm7, %%mm0                \n\t" /* 0A0B0C0D */\
1068
        "punpckhbw %%mm7, %%mm1                \n\t" /* 0E0F0G0H */\
1069
        "pshufw $0x90, %%mm0, %%mm5        \n\t" /* 0A0A0B0C */\
1070
        "pshufw $0x41, %%mm0, %%mm6        \n\t" /* 0B0A0A0B */\
1071
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1072
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1073
        "psllq $8, %%mm2                \n\t" /* 0ABCDEFG */\
1074
        "psllq $16, %%mm3                \n\t" /* 00ABCDEF */\
1075
        "psllq $24, %%mm4                \n\t" /* 000ABCDE */\
1076
        "punpckhbw %%mm7, %%mm2                \n\t" /* 0D0E0F0G */\
1077
        "punpckhbw %%mm7, %%mm3                \n\t" /* 0C0D0E0F */\
1078
        "punpckhbw %%mm7, %%mm4                \n\t" /* 0B0C0D0E */\
1079
        "paddw %%mm3, %%mm5                \n\t" /* b */\
1080
        "paddw %%mm2, %%mm6                \n\t" /* c */\
1081
        "paddw %%mm5, %%mm5                \n\t" /* 2b */\
1082
        "psubw %%mm5, %%mm6                \n\t" /* c - 2b */\
1083
        "pshufw $0x06, %%mm0, %%mm5        \n\t" /* 0C0B0A0A */\
1084
        "pmullw "MANGLE(ff_pw_3)", %%mm6                \n\t" /* 3c - 6b */\
1085
        "paddw %%mm4, %%mm0                \n\t" /* a */\
1086
        "paddw %%mm1, %%mm5                \n\t" /* d */\
1087
        "pmullw "MANGLE(ff_pw_20)", %%mm0                \n\t" /* 20a */\
1088
        "psubw %%mm5, %%mm0                \n\t" /* 20a - d */\
1089
        "paddw %6, %%mm6                \n\t"\
1090
        "paddw %%mm6, %%mm0                \n\t" /* 20a - 6b + 3c - d */\
1091
        "psraw $5, %%mm0                \n\t"\
1092
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1093
        \
1094
        "movd 5(%0), %%mm5                \n\t" /* FGHI */\
1095
        "punpcklbw %%mm7, %%mm5                \n\t" /* 0F0G0H0I */\
1096
        "pshufw $0xF9, %%mm5, %%mm6        \n\t" /* 0G0H0I0I */\
1097
        "paddw %%mm5, %%mm1                \n\t" /* a */\
1098
        "paddw %%mm6, %%mm2                \n\t" /* b */\
1099
        "pshufw $0xBE, %%mm5, %%mm6        \n\t" /* 0H0I0I0H */\
1100
        "pshufw $0x6F, %%mm5, %%mm5        \n\t" /* 0I0I0H0G */\
1101
        "paddw %%mm6, %%mm3                \n\t" /* c */\
1102
        "paddw %%mm5, %%mm4                \n\t" /* d */\
1103
        "paddw %%mm2, %%mm2                \n\t" /* 2b */\
1104
        "psubw %%mm2, %%mm3                \n\t" /* c - 2b */\
1105
        "pmullw "MANGLE(ff_pw_20)", %%mm1                \n\t" /* 20a */\
1106
        "pmullw "MANGLE(ff_pw_3)", %%mm3                \n\t" /* 3c - 6b */\
1107
        "psubw %%mm4, %%mm3                \n\t" /* -6b + 3c - d */\
1108
        "paddw %6, %%mm1                \n\t"\
1109
        "paddw %%mm1, %%mm3                \n\t" /* 20a - 6b + 3c - d */\
1110
        "psraw $5, %%mm3                \n\t"\
1111
        "packuswb %%mm3, %%mm0                \n\t"\
1112
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
1113
        \
1114
        "addl %3, %0                        \n\t"\
1115
        "addl %4, %1                        \n\t"\
1116
        "decl %2                        \n\t"\
1117
        " jnz 1b                        \n\t"\
1118
        : "+a"(src), "+c"(dst), "+m"(h)\
1119
        : "S"(srcStride), "D"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1120
        : "memory"\
1121
    );\
1122
}\
1123
\
1124
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1125
    int i;\
1126
    int16_t temp[8];\
1127
    /* quick HACK, XXX FIXME MUST be optimized */\
1128
    for(i=0; i<h; i++)\
1129
    {\
1130
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1131
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1132
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1133
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1134
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1135
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1136
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1137
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1138
        asm volatile(\
1139
            "movq (%0), %%mm0                \n\t"\
1140
            "movq 8(%0), %%mm1                \n\t"\
1141
            "paddw %2, %%mm0                \n\t"\
1142
            "paddw %2, %%mm1                \n\t"\
1143
            "psraw $5, %%mm0                \n\t"\
1144
            "psraw $5, %%mm1                \n\t"\
1145
            "packuswb %%mm1, %%mm0        \n\t"\
1146
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1147
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1148
            :"memory"\
1149
        );\
1150
        dst+=dstStride;\
1151
        src+=srcStride;\
1152
    }\
1153
}
1154

    
1155
#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1156
\
1157
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1158
    uint64_t temp[17*4];\
1159
    uint64_t *temp_ptr= temp;\
1160
    int count= 17;\
1161
\
1162
    /*FIXME unroll */\
1163
    asm volatile(\
1164
        "pxor %%mm7, %%mm7                \n\t"\
1165
        "1:                                \n\t"\
1166
        "movq (%0), %%mm0                \n\t"\
1167
        "movq (%0), %%mm1                \n\t"\
1168
        "movq 8(%0), %%mm2                \n\t"\
1169
        "movq 8(%0), %%mm3                \n\t"\
1170
        "punpcklbw %%mm7, %%mm0                \n\t"\
1171
        "punpckhbw %%mm7, %%mm1                \n\t"\
1172
        "punpcklbw %%mm7, %%mm2                \n\t"\
1173
        "punpckhbw %%mm7, %%mm3                \n\t"\
1174
        "movq %%mm0, (%1)                \n\t"\
1175
        "movq %%mm1, 17*8(%1)                \n\t"\
1176
        "movq %%mm2, 2*17*8(%1)                \n\t"\
1177
        "movq %%mm3, 3*17*8(%1)                \n\t"\
1178
        "addl $8, %1                        \n\t"\
1179
        "addl %3, %0                        \n\t"\
1180
        "decl %2                        \n\t"\
1181
        " jnz 1b                        \n\t"\
1182
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1183
        : "r" (srcStride)\
1184
        : "memory"\
1185
    );\
1186
    \
1187
    temp_ptr= temp;\
1188
    count=4;\
1189
    \
1190
/*FIXME reorder for speed */\
1191
    asm volatile(\
1192
        /*"pxor %%mm7, %%mm7                \n\t"*/\
1193
        "1:                                \n\t"\
1194
        "movq (%0), %%mm0                \n\t"\
1195
        "movq 8(%0), %%mm1                \n\t"\
1196
        "movq 16(%0), %%mm2                \n\t"\
1197
        "movq 24(%0), %%mm3                \n\t"\
1198
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1199
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1200
        "addl %4, %1                        \n\t"\
1201
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1202
        \
1203
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1204
        "addl %4, %1                        \n\t"\
1205
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1206
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1207
        "addl %4, %1                        \n\t"\
1208
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1209
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1210
        "addl %4, %1                        \n\t"\
1211
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1212
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1213
        "addl %4, %1                        \n\t"\
1214
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1215
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1216
        "addl %4, %1                        \n\t"\
1217
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1218
        \
1219
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1220
        "addl %4, %1                        \n\t"  \
1221
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1222
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1223
        \
1224
        "addl $136, %0                        \n\t"\
1225
        "addl %6, %1                        \n\t"\
1226
        "decl %2                        \n\t"\
1227
        " jnz 1b                        \n\t"\
1228
        \
1229
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1230
        : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*dstStride)\
1231
        :"memory"\
1232
    );\
1233
}\
1234
\
1235
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1236
    uint64_t temp[9*4];\
1237
    uint64_t *temp_ptr= temp;\
1238
    int count= 9;\
1239
\
1240
    /*FIXME unroll */\
1241
    asm volatile(\
1242
        "pxor %%mm7, %%mm7                \n\t"\
1243
        "1:                                \n\t"\
1244
        "movq (%0), %%mm0                \n\t"\
1245
        "movq (%0), %%mm1                \n\t"\
1246
        "punpcklbw %%mm7, %%mm0                \n\t"\
1247
        "punpckhbw %%mm7, %%mm1                \n\t"\
1248
        "movq %%mm0, (%1)                \n\t"\
1249
        "movq %%mm1, 9*8(%1)                \n\t"\
1250
        "addl $8, %1                        \n\t"\
1251
        "addl %3, %0                        \n\t"\
1252
        "decl %2                        \n\t"\
1253
        " jnz 1b                        \n\t"\
1254
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1255
        : "r" (srcStride)\
1256
        : "memory"\
1257
    );\
1258
    \
1259
    temp_ptr= temp;\
1260
    count=2;\
1261
    \
1262
/*FIXME reorder for speed */\
1263
    asm volatile(\
1264
        /*"pxor %%mm7, %%mm7                \n\t"*/\
1265
        "1:                                \n\t"\
1266
        "movq (%0), %%mm0                \n\t"\
1267
        "movq 8(%0), %%mm1                \n\t"\
1268
        "movq 16(%0), %%mm2                \n\t"\
1269
        "movq 24(%0), %%mm3                \n\t"\
1270
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1271
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1272
        "addl %4, %1                        \n\t"\
1273
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1274
        \
1275
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1276
        "addl %4, %1                        \n\t"\
1277
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1278
        \
1279
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1280
        "addl %4, %1                        \n\t"\
1281
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1282
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1283
                \
1284
        "addl $72, %0                        \n\t"\
1285
        "addl %6, %1                        \n\t"\
1286
        "decl %2                        \n\t"\
1287
        " jnz 1b                        \n\t"\
1288
         \
1289
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1290
        : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*dstStride)\
1291
        : "memory"\
1292
   );\
1293
}\
1294
\
1295
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1296
    OPNAME ## pixels8_mmx(dst, src, stride, 8);\
1297
}\
1298
\
1299
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1300
    uint64_t temp[8];\
1301
    uint8_t * const half= (uint8_t*)temp;\
1302
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1303
    OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
1304
}\
1305
\
1306
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1307
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1308
}\
1309
\
1310
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1311
    uint64_t temp[8];\
1312
    uint8_t * const half= (uint8_t*)temp;\
1313
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1314
    OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\
1315
}\
1316
\
1317
static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1318
    uint64_t temp[8];\
1319
    uint8_t * const half= (uint8_t*)temp;\
1320
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1321
    OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
1322
}\
1323
\
1324
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1325
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1326
}\
1327
\
1328
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1329
    uint64_t temp[8];\
1330
    uint8_t * const half= (uint8_t*)temp;\
1331
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1332
    OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\
1333
}\
1334
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1335
    uint64_t half[8 + 9];\
1336
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1337
    uint8_t * const halfHV= ((uint8_t*)half);\
1338
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1339
    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
1340
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1341
    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
1342
}\
1343
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1344
    uint64_t half[8 + 9];\
1345
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1346
    uint8_t * const halfHV= ((uint8_t*)half);\
1347
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1348
    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
1349
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1350
    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
1351
}\
1352
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1353
    uint64_t half[8 + 9];\
1354
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1355
    uint8_t * const halfHV= ((uint8_t*)half);\
1356
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1357
    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
1358
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1359
    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
1360
}\
1361
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1362
    uint64_t half[8 + 9];\
1363
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1364
    uint8_t * const halfHV= ((uint8_t*)half);\
1365
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1366
    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
1367
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1368
    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
1369
}\
1370
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1371
    uint64_t half[8 + 9];\
1372
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1373
    uint8_t * const halfHV= ((uint8_t*)half);\
1374
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1375
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1376
    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
1377
}\
1378
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1379
    uint64_t half[8 + 9];\
1380
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1381
    uint8_t * const halfHV= ((uint8_t*)half);\
1382
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1383
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1384
    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
1385
}\
1386
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1387
    uint64_t half[8 + 9];\
1388
    uint8_t * const halfH= ((uint8_t*)half);\
1389
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1390
    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
1391
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1392
}\
1393
static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1394
    uint64_t half[8 + 9];\
1395
    uint8_t * const halfH= ((uint8_t*)half);\
1396
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1397
    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
1398
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1399
}\
1400
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1401
    uint64_t half[9];\
1402
    uint8_t * const halfH= ((uint8_t*)half);\
1403
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1404
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1405
}\
1406
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1407
    OPNAME ## pixels16_mmx(dst, src, stride, 16);\
1408
}\
1409
\
1410
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1411
    uint64_t temp[32];\
1412
    uint8_t * const half= (uint8_t*)temp;\
1413
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1414
    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
1415
}\
1416
\
1417
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1418
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1419
}\
1420
\
1421
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1422
    uint64_t temp[32];\
1423
    uint8_t * const half= (uint8_t*)temp;\
1424
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1425
    OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\
1426
}\
1427
\
1428
static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1429
    uint64_t temp[32];\
1430
    uint8_t * const half= (uint8_t*)temp;\
1431
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1432
    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
1433
}\
1434
\
1435
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1436
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1437
}\
1438
\
1439
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1440
    uint64_t temp[32];\
1441
    uint8_t * const half= (uint8_t*)temp;\
1442
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1443
    OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\
1444
}\
1445
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1446
    uint64_t half[16*2 + 17*2];\
1447
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1448
    uint8_t * const halfHV= ((uint8_t*)half);\
1449
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1450
    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
1451
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1452
    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
1453
}\
1454
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1455
    uint64_t half[16*2 + 17*2];\
1456
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1457
    uint8_t * const halfHV= ((uint8_t*)half);\
1458
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1459
    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
1460
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1461
    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
1462
}\
1463
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1464
    uint64_t half[16*2 + 17*2];\
1465
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1466
    uint8_t * const halfHV= ((uint8_t*)half);\
1467
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1468
    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
1469
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1470
    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
1471
}\
1472
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1473
    uint64_t half[16*2 + 17*2];\
1474
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1475
    uint8_t * const halfHV= ((uint8_t*)half);\
1476
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1477
    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
1478
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1479
    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
1480
}\
1481
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1482
    uint64_t half[16*2 + 17*2];\
1483
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1484
    uint8_t * const halfHV= ((uint8_t*)half);\
1485
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1486
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1487
    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
1488
}\
1489
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1490
    uint64_t half[16*2 + 17*2];\
1491
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1492
    uint8_t * const halfHV= ((uint8_t*)half);\
1493
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1494
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1495
    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
1496
}\
1497
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1498
    uint64_t half[17*2];\
1499
    uint8_t * const halfH= ((uint8_t*)half);\
1500
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1501
    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
1502
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1503
}\
1504
static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1505
    uint64_t half[17*2];\
1506
    uint8_t * const halfH= ((uint8_t*)half);\
1507
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1508
    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
1509
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1510
}\
1511
static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1512
    uint64_t half[17*2];\
1513
    uint8_t * const halfH= ((uint8_t*)half);\
1514
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1515
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1516
}
1517

    
1518

    
1519
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"
1520
#define AVG_3DNOW_OP(a,b,temp, size) \
1521
"mov" #size " " #b ", " #temp "        \n\t"\
1522
"pavgusb " #temp ", " #a "        \n\t"\
1523
"mov" #size " " #a ", " #b "        \n\t"
1524
#define AVG_MMX2_OP(a,b,temp, size) \
1525
"mov" #size " " #b ", " #temp "        \n\t"\
1526
"pavgb " #temp ", " #a "        \n\t"\
1527
"mov" #size " " #a ", " #b "        \n\t"
1528

    
1529
QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
1530
QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
1531
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1532
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
1533
QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
1534
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1535
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
1536
QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
1537
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1538

    
1539
#if 0
1540
static void just_return() { return; }
1541
#endif
1542

    
1543
#define SET_QPEL_FUNC(postfix1, postfix2) \
1544
    c->put_ ## postfix1 = put_ ## postfix2;\
1545
    c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
1546
    c->avg_ ## postfix1 = avg_ ## postfix2;
1547

    
1548
/* external functions, from idct_mmx.c */
1549
void ff_mmx_idct(DCTELEM *block);
1550
void ff_mmxext_idct(DCTELEM *block);
1551

    
1552
/* XXX: those functions should be suppressed ASAP when all IDCTs are
1553
   converted */
1554
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1555
{
1556
    ff_mmx_idct (block);
1557
    put_pixels_clamped_mmx(block, dest, line_size);
1558
}
1559
static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1560
{
1561
    ff_mmx_idct (block);
1562
    add_pixels_clamped_mmx(block, dest, line_size);
1563
}
1564
static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1565
{
1566
    ff_mmxext_idct (block);
1567
    put_pixels_clamped_mmx(block, dest, line_size);
1568
}
1569
static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1570
{
1571
    ff_mmxext_idct (block);
1572
    add_pixels_clamped_mmx(block, dest, line_size);
1573
}
1574
    
1575
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
1576
{
1577
    mm_flags = mm_support();
1578

    
1579
    if (avctx->dsp_mask) {
1580
        if (avctx->dsp_mask & FF_MM_FORCE)
1581
            mm_flags |= (avctx->dsp_mask & 0xffff);
1582
        else
1583
            mm_flags &= ~(avctx->dsp_mask & 0xffff);
1584
    }
1585

    
1586
#if 0
1587
    fprintf(stderr, "libavcodec: CPU flags:");
1588
    if (mm_flags & MM_MMX)
1589
        fprintf(stderr, " mmx");
1590
    if (mm_flags & MM_MMXEXT)
1591
        fprintf(stderr, " mmxext");
1592
    if (mm_flags & MM_3DNOW)
1593
        fprintf(stderr, " 3dnow");
1594
    if (mm_flags & MM_SSE)
1595
        fprintf(stderr, " sse");
1596
    if (mm_flags & MM_SSE2)
1597
        fprintf(stderr, " sse2");
1598
    fprintf(stderr, "\n");
1599
#endif
1600

    
1601
    if (mm_flags & MM_MMX) {
1602
        const int dct_algo = avctx->dct_algo;
1603
        const int idct_algo= avctx->idct_algo;
1604

    
1605
#ifdef CONFIG_ENCODERS
1606
        if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX)
1607
            c->fdct = ff_fdct_mmx;
1608
#endif //CONFIG_ENCODERS
1609

    
1610
        if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
1611
            c->idct_put= ff_simple_idct_put_mmx;
1612
            c->idct_add= ff_simple_idct_add_mmx;
1613
            c->idct    = ff_simple_idct_mmx;
1614
            c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
1615
        }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
1616
            if(mm_flags & MM_MMXEXT){
1617
                c->idct_put= ff_libmpeg2mmx2_idct_put;
1618
                c->idct_add= ff_libmpeg2mmx2_idct_add;
1619
                c->idct    = ff_mmxext_idct;
1620
            }else{
1621
                c->idct_put= ff_libmpeg2mmx_idct_put;
1622
                c->idct_add= ff_libmpeg2mmx_idct_add;
1623
                c->idct    = ff_mmx_idct;
1624
            }
1625
            c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
1626
        }
1627
        
1628
#ifdef CONFIG_ENCODERS
1629
        c->get_pixels = get_pixels_mmx;
1630
        c->diff_pixels = diff_pixels_mmx;
1631
#endif //CONFIG_ENCODERS
1632
        c->put_pixels_clamped = put_pixels_clamped_mmx;
1633
        c->add_pixels_clamped = add_pixels_clamped_mmx;
1634
        c->clear_blocks = clear_blocks_mmx;
1635
#ifdef CONFIG_ENCODERS
1636
        c->pix_sum = pix_sum16_mmx;
1637
#endif //CONFIG_ENCODERS
1638

    
1639
        c->put_pixels_tab[0][0] = put_pixels16_mmx;
1640
        c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;
1641
        c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;
1642
        c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;
1643

    
1644
        c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;
1645
        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
1646
        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
1647
        c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;
1648

    
1649
        c->avg_pixels_tab[0][0] = avg_pixels16_mmx;
1650
        c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;
1651
        c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;
1652
        c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;
1653

    
1654
        c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;
1655
        c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;
1656
        c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;
1657
        c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;
1658

    
1659
        c->put_pixels_tab[1][0] = put_pixels8_mmx;
1660
        c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;
1661
        c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;
1662
        c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;
1663

    
1664
        c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;
1665
        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
1666
        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
1667
        c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;
1668

    
1669
        c->avg_pixels_tab[1][0] = avg_pixels8_mmx;
1670
        c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;
1671
        c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;
1672
        c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;
1673

    
1674
        c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;
1675
        c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
1676
        c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
1677
        c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
1678
                
1679
        c->add_bytes= add_bytes_mmx;
1680
#ifdef CONFIG_ENCODERS
1681
        c->diff_bytes= diff_bytes_mmx;
1682
        
1683
        c->hadamard8_diff[0]= hadamard8_diff16_mmx;
1684
        c->hadamard8_diff[1]= hadamard8_diff_mmx;
1685
        
1686
        c->pix_norm1 = pix_norm1_mmx;
1687
        c->sse[0] = sse16_mmx;
1688
#endif //CONFIG_ENCODERS
1689
        
1690
        if (mm_flags & MM_MMXEXT) {
1691
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
1692
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
1693

    
1694
            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
1695
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
1696
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
1697

    
1698
            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
1699
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
1700

    
1701
            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
1702
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
1703
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
1704

    
1705
#ifdef CONFIG_ENCODERS
1706
            c->hadamard8_diff[0]= hadamard8_diff16_mmx2;
1707
            c->hadamard8_diff[1]= hadamard8_diff_mmx2;
1708
#endif //CONFIG_ENCODERS
1709

    
1710
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1711
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
1712
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
1713
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
1714
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
1715
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
1716
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
1717
            }
1718

    
1719
#if 1
1720
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
1721
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
1722
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
1723
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
1724
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
1725
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
1726
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
1727
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
1728
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
1729
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
1730
            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
1731
            SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)
1732
            SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)
1733
            SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)
1734
            SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)
1735
            SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)
1736
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)
1737
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)
1738
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)
1739
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)
1740
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)
1741
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)
1742
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)
1743
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)
1744
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)
1745
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)
1746
            SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)
1747
            SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)
1748
            SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)
1749
            SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)
1750
            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)
1751
            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)
1752
#endif
1753

    
1754
            c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;
1755
        } else if (mm_flags & MM_3DNOW) {
1756
            c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
1757
            c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
1758

    
1759
            c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
1760
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
1761
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
1762

    
1763
            c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
1764
            c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
1765

    
1766
            c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
1767
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
1768
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
1769

    
1770
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
1771
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
1772
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
1773
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
1774
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
1775
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
1776
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
1777
            }
1778

    
1779
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
1780
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
1781
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)
1782
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)
1783
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)
1784
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)
1785
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)
1786
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)
1787
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)
1788
            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)
1789
            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)
1790
            SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)
1791
            SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)
1792
            SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)
1793
            SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)
1794
            SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)
1795
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)
1796
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)
1797
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)
1798
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)
1799
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)
1800
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)
1801
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)
1802
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)
1803
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)
1804
            SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)
1805
            SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)
1806
            SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)
1807
            SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)
1808
            SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)
1809
            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
1810
            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
1811
        }
1812
    }
1813
        
1814
#ifdef CONFIG_ENCODERS
1815
    dsputil_init_pix_mmx(c, avctx);
1816
#endif //CONFIG_ENCODERS
1817
#if 0
1818
    // for speed testing
1819
    get_pixels = just_return;
1820
    put_pixels_clamped = just_return;
1821
    add_pixels_clamped = just_return;
1822

1823
    pix_abs16x16 = just_return;
1824
    pix_abs16x16_x2 = just_return;
1825
    pix_abs16x16_y2 = just_return;
1826
    pix_abs16x16_xy2 = just_return;
1827

1828
    put_pixels_tab[0] = just_return;
1829
    put_pixels_tab[1] = just_return;
1830
    put_pixels_tab[2] = just_return;
1831
    put_pixels_tab[3] = just_return;
1832

1833
    put_no_rnd_pixels_tab[0] = just_return;
1834
    put_no_rnd_pixels_tab[1] = just_return;
1835
    put_no_rnd_pixels_tab[2] = just_return;
1836
    put_no_rnd_pixels_tab[3] = just_return;
1837

1838
    avg_pixels_tab[0] = just_return;
1839
    avg_pixels_tab[1] = just_return;
1840
    avg_pixels_tab[2] = just_return;
1841
    avg_pixels_tab[3] = just_return;
1842

1843
    avg_no_rnd_pixels_tab[0] = just_return;
1844
    avg_no_rnd_pixels_tab[1] = just_return;
1845
    avg_no_rnd_pixels_tab[2] = just_return;
1846
    avg_no_rnd_pixels_tab[3] = just_return;
1847

1848
    //av_fdct = just_return;
1849
    //ff_idct = just_return;
1850
#endif
1851
}