Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / dsputil_mmx.c @ 911e21a3

History | View | Annotate | Download (117 KB)

1
/*
2
 * MMX optimized DSP utils
3
 * Copyright (c) 2000, 2001 Fabrice Bellard.
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 *
22
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
23
 */
24

    
25
#include "libavutil/x86_cpu.h"
26
#include "libavcodec/dsputil.h"
27
#include "libavcodec/h263.h"
28
#include "libavcodec/mpegvideo.h"
29
#include "libavcodec/simple_idct.h"
30
#include "dsputil_mmx.h"
31
#include "mmx.h"
32
#include "vp3dsp_mmx.h"
33
#include "vp3dsp_sse2.h"
34
#include "idct_xvid.h"
35

    
36
//#undef NDEBUG
37
//#include <assert.h>
38

    
39
int mm_flags; /* multimedia extension flags */
40

    
41
/* pixel operations */
42
DECLARE_ALIGNED_8 (const uint64_t, ff_bone) = 0x0101010101010101ULL;
43
DECLARE_ALIGNED_8 (const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
44

    
45
DECLARE_ALIGNED_16(const uint64_t, ff_pdw_80000000[2]) =
46
{0x8000000080000000ULL, 0x8000000080000000ULL};
47

    
48
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_3  ) = 0x0003000300030003ULL;
49
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_4  ) = 0x0004000400040004ULL;
50
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
51
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_8  ) = 0x0008000800080008ULL;
52
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
53
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
54
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
55
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
56
DECLARE_ALIGNED_16(const xmm_t,    ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
57
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
58
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_64 ) = 0x0040004000400040ULL;
59
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
60
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
61
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
62

    
63
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1  ) = 0x0101010101010101ULL;
64
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3  ) = 0x0303030303030303ULL;
65
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
66
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
67
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL;
68
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
69

    
70
DECLARE_ALIGNED_16(const double, ff_pd_1[2]) = { 1.0, 1.0 };
71
DECLARE_ALIGNED_16(const double, ff_pd_2[2]) = { 2.0, 2.0 };
72

    
73
#define JUMPALIGN() asm volatile (ASMALIGN(3)::)
74
#define MOVQ_ZERO(regd)  asm volatile ("pxor %%" #regd ", %%" #regd ::)
75

    
76
#define MOVQ_BFE(regd) \
77
    asm volatile ( \
78
    "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
79
    "paddb %%" #regd ", %%" #regd " \n\t" ::)
80

    
81
#ifndef PIC
82
#define MOVQ_BONE(regd)  asm volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
83
#define MOVQ_WTWO(regd)  asm volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
84
#else
85
// for shared library it's better to use this way for accessing constants
86
// pcmpeqd -> -1
87
#define MOVQ_BONE(regd) \
88
    asm volatile ( \
89
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
90
    "psrlw $15, %%" #regd " \n\t" \
91
    "packuswb %%" #regd ", %%" #regd " \n\t" ::)
92

    
93
#define MOVQ_WTWO(regd) \
94
    asm volatile ( \
95
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
96
    "psrlw $15, %%" #regd " \n\t" \
97
    "psllw $1, %%" #regd " \n\t"::)
98

    
99
#endif
100

    
101
// using regr as temporary and for the output result
102
// first argument is unmodifed and second is trashed
103
// regfe is supposed to contain 0xfefefefefefefefe
104
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
105
    "movq " #rega ", " #regr "  \n\t"\
106
    "pand " #regb ", " #regr "  \n\t"\
107
    "pxor " #rega ", " #regb "  \n\t"\
108
    "pand " #regfe "," #regb "  \n\t"\
109
    "psrlq $1, " #regb "        \n\t"\
110
    "paddb " #regb ", " #regr " \n\t"
111

    
112
#define PAVGB_MMX(rega, regb, regr, regfe) \
113
    "movq " #rega ", " #regr "  \n\t"\
114
    "por  " #regb ", " #regr "  \n\t"\
115
    "pxor " #rega ", " #regb "  \n\t"\
116
    "pand " #regfe "," #regb "  \n\t"\
117
    "psrlq $1, " #regb "        \n\t"\
118
    "psubb " #regb ", " #regr " \n\t"
119

    
120
// mm6 is supposed to contain 0xfefefefefefefefe
121
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \
122
    "movq " #rega ", " #regr "  \n\t"\
123
    "movq " #regc ", " #regp "  \n\t"\
124
    "pand " #regb ", " #regr "  \n\t"\
125
    "pand " #regd ", " #regp "  \n\t"\
126
    "pxor " #rega ", " #regb "  \n\t"\
127
    "pxor " #regc ", " #regd "  \n\t"\
128
    "pand %%mm6, " #regb "      \n\t"\
129
    "pand %%mm6, " #regd "      \n\t"\
130
    "psrlq $1, " #regb "        \n\t"\
131
    "psrlq $1, " #regd "        \n\t"\
132
    "paddb " #regb ", " #regr " \n\t"\
133
    "paddb " #regd ", " #regp " \n\t"
134

    
135
#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
136
    "movq " #rega ", " #regr "  \n\t"\
137
    "movq " #regc ", " #regp "  \n\t"\
138
    "por  " #regb ", " #regr "  \n\t"\
139
    "por  " #regd ", " #regp "  \n\t"\
140
    "pxor " #rega ", " #regb "  \n\t"\
141
    "pxor " #regc ", " #regd "  \n\t"\
142
    "pand %%mm6, " #regb "      \n\t"\
143
    "pand %%mm6, " #regd "      \n\t"\
144
    "psrlq $1, " #regd "        \n\t"\
145
    "psrlq $1, " #regb "        \n\t"\
146
    "psubb " #regb ", " #regr " \n\t"\
147
    "psubb " #regd ", " #regp " \n\t"
148

    
149
/***********************************/
150
/* MMX no rounding */
151
#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
152
#define SET_RND  MOVQ_WONE
153
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
154
#define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
155

    
156
#include "dsputil_mmx_rnd.h"
157

    
158
#undef DEF
159
#undef SET_RND
160
#undef PAVGBP
161
#undef PAVGB
162
/***********************************/
163
/* MMX rounding */
164

    
165
#define DEF(x, y) x ## _ ## y ##_mmx
166
#define SET_RND  MOVQ_WTWO
167
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
168
#define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)
169

    
170
#include "dsputil_mmx_rnd.h"
171

    
172
#undef DEF
173
#undef SET_RND
174
#undef PAVGBP
175
#undef PAVGB
176

    
177
/***********************************/
178
/* 3Dnow specific */
179

    
180
#define DEF(x) x ## _3dnow
181
#define PAVGB "pavgusb"
182

    
183
#include "dsputil_mmx_avg.h"
184

    
185
#undef DEF
186
#undef PAVGB
187

    
188
/***********************************/
189
/* MMX2 specific */
190

    
191
#define DEF(x) x ## _mmx2
192

    
193
/* Introduced only in MMX2 set */
194
#define PAVGB "pavgb"
195

    
196
#include "dsputil_mmx_avg.h"
197

    
198
#undef DEF
199
#undef PAVGB
200

    
201
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
202
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
203
#define put_pixels16_mmx2 put_pixels16_mmx
204
#define put_pixels8_mmx2 put_pixels8_mmx
205
#define put_pixels4_mmx2 put_pixels4_mmx
206
#define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
207
#define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
208
#define put_pixels16_3dnow put_pixels16_mmx
209
#define put_pixels8_3dnow put_pixels8_mmx
210
#define put_pixels4_3dnow put_pixels4_mmx
211
#define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
212
#define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
213

    
214
/***********************************/
215
/* standard MMX */
216

    
217
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
218
{
219
    const DCTELEM *p;
220
    uint8_t *pix;
221

    
222
    /* read the pixels */
223
    p = block;
224
    pix = pixels;
225
    /* unrolled loop */
226
        asm volatile(
227
                "movq   %3, %%mm0               \n\t"
228
                "movq   8%3, %%mm1              \n\t"
229
                "movq   16%3, %%mm2             \n\t"
230
                "movq   24%3, %%mm3             \n\t"
231
                "movq   32%3, %%mm4             \n\t"
232
                "movq   40%3, %%mm5             \n\t"
233
                "movq   48%3, %%mm6             \n\t"
234
                "movq   56%3, %%mm7             \n\t"
235
                "packuswb %%mm1, %%mm0          \n\t"
236
                "packuswb %%mm3, %%mm2          \n\t"
237
                "packuswb %%mm5, %%mm4          \n\t"
238
                "packuswb %%mm7, %%mm6          \n\t"
239
                "movq   %%mm0, (%0)             \n\t"
240
                "movq   %%mm2, (%0, %1)         \n\t"
241
                "movq   %%mm4, (%0, %1, 2)      \n\t"
242
                "movq   %%mm6, (%0, %2)         \n\t"
243
                ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
244
                :"memory");
245
        pix += line_size*4;
246
        p += 32;
247

    
248
    // if here would be an exact copy of the code above
249
    // compiler would generate some very strange code
250
    // thus using "r"
251
    asm volatile(
252
            "movq       (%3), %%mm0             \n\t"
253
            "movq       8(%3), %%mm1            \n\t"
254
            "movq       16(%3), %%mm2           \n\t"
255
            "movq       24(%3), %%mm3           \n\t"
256
            "movq       32(%3), %%mm4           \n\t"
257
            "movq       40(%3), %%mm5           \n\t"
258
            "movq       48(%3), %%mm6           \n\t"
259
            "movq       56(%3), %%mm7           \n\t"
260
            "packuswb %%mm1, %%mm0              \n\t"
261
            "packuswb %%mm3, %%mm2              \n\t"
262
            "packuswb %%mm5, %%mm4              \n\t"
263
            "packuswb %%mm7, %%mm6              \n\t"
264
            "movq       %%mm0, (%0)             \n\t"
265
            "movq       %%mm2, (%0, %1)         \n\t"
266
            "movq       %%mm4, (%0, %1, 2)      \n\t"
267
            "movq       %%mm6, (%0, %2)         \n\t"
268
            ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
269
            :"memory");
270
}
271

    
272
static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) =
273
  { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
274

    
275
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
276
{
277
    int i;
278

    
279
    movq_m2r(*vector128, mm1);
280
    for (i = 0; i < 8; i++) {
281
        movq_m2r(*(block), mm0);
282
        packsswb_m2r(*(block + 4), mm0);
283
        block += 8;
284
        paddb_r2r(mm1, mm0);
285
        movq_r2m(mm0, *pixels);
286
        pixels += line_size;
287
    }
288
}
289

    
290
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
291
{
292
    const DCTELEM *p;
293
    uint8_t *pix;
294
    int i;
295

    
296
    /* read the pixels */
297
    p = block;
298
    pix = pixels;
299
    MOVQ_ZERO(mm7);
300
    i = 4;
301
    do {
302
        asm volatile(
303
                "movq   (%2), %%mm0     \n\t"
304
                "movq   8(%2), %%mm1    \n\t"
305
                "movq   16(%2), %%mm2   \n\t"
306
                "movq   24(%2), %%mm3   \n\t"
307
                "movq   %0, %%mm4       \n\t"
308
                "movq   %1, %%mm6       \n\t"
309
                "movq   %%mm4, %%mm5    \n\t"
310
                "punpcklbw %%mm7, %%mm4 \n\t"
311
                "punpckhbw %%mm7, %%mm5 \n\t"
312
                "paddsw %%mm4, %%mm0    \n\t"
313
                "paddsw %%mm5, %%mm1    \n\t"
314
                "movq   %%mm6, %%mm5    \n\t"
315
                "punpcklbw %%mm7, %%mm6 \n\t"
316
                "punpckhbw %%mm7, %%mm5 \n\t"
317
                "paddsw %%mm6, %%mm2    \n\t"
318
                "paddsw %%mm5, %%mm3    \n\t"
319
                "packuswb %%mm1, %%mm0  \n\t"
320
                "packuswb %%mm3, %%mm2  \n\t"
321
                "movq   %%mm0, %0       \n\t"
322
                "movq   %%mm2, %1       \n\t"
323
                :"+m"(*pix), "+m"(*(pix+line_size))
324
                :"r"(p)
325
                :"memory");
326
        pix += line_size*2;
327
        p += 16;
328
    } while (--i);
329
}
330

    
331
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
332
{
333
    asm volatile(
334
         "lea (%3, %3), %%"REG_a"       \n\t"
335
         ASMALIGN(3)
336
         "1:                            \n\t"
337
         "movd (%1), %%mm0              \n\t"
338
         "movd (%1, %3), %%mm1          \n\t"
339
         "movd %%mm0, (%2)              \n\t"
340
         "movd %%mm1, (%2, %3)          \n\t"
341
         "add %%"REG_a", %1             \n\t"
342
         "add %%"REG_a", %2             \n\t"
343
         "movd (%1), %%mm0              \n\t"
344
         "movd (%1, %3), %%mm1          \n\t"
345
         "movd %%mm0, (%2)              \n\t"
346
         "movd %%mm1, (%2, %3)          \n\t"
347
         "add %%"REG_a", %1             \n\t"
348
         "add %%"REG_a", %2             \n\t"
349
         "subl $4, %0                   \n\t"
350
         "jnz 1b                        \n\t"
351
         : "+g"(h), "+r" (pixels),  "+r" (block)
352
         : "r"((x86_reg)line_size)
353
         : "%"REG_a, "memory"
354
        );
355
}
356

    
357
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
358
{
359
    asm volatile(
360
         "lea (%3, %3), %%"REG_a"       \n\t"
361
         ASMALIGN(3)
362
         "1:                            \n\t"
363
         "movq (%1), %%mm0              \n\t"
364
         "movq (%1, %3), %%mm1          \n\t"
365
         "movq %%mm0, (%2)              \n\t"
366
         "movq %%mm1, (%2, %3)          \n\t"
367
         "add %%"REG_a", %1             \n\t"
368
         "add %%"REG_a", %2             \n\t"
369
         "movq (%1), %%mm0              \n\t"
370
         "movq (%1, %3), %%mm1          \n\t"
371
         "movq %%mm0, (%2)              \n\t"
372
         "movq %%mm1, (%2, %3)          \n\t"
373
         "add %%"REG_a", %1             \n\t"
374
         "add %%"REG_a", %2             \n\t"
375
         "subl $4, %0                   \n\t"
376
         "jnz 1b                        \n\t"
377
         : "+g"(h), "+r" (pixels),  "+r" (block)
378
         : "r"((x86_reg)line_size)
379
         : "%"REG_a, "memory"
380
        );
381
}
382

    
383
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
384
{
385
    asm volatile(
386
         "lea (%3, %3), %%"REG_a"       \n\t"
387
         ASMALIGN(3)
388
         "1:                            \n\t"
389
         "movq (%1), %%mm0              \n\t"
390
         "movq 8(%1), %%mm4             \n\t"
391
         "movq (%1, %3), %%mm1          \n\t"
392
         "movq 8(%1, %3), %%mm5         \n\t"
393
         "movq %%mm0, (%2)              \n\t"
394
         "movq %%mm4, 8(%2)             \n\t"
395
         "movq %%mm1, (%2, %3)          \n\t"
396
         "movq %%mm5, 8(%2, %3)         \n\t"
397
         "add %%"REG_a", %1             \n\t"
398
         "add %%"REG_a", %2             \n\t"
399
         "movq (%1), %%mm0              \n\t"
400
         "movq 8(%1), %%mm4             \n\t"
401
         "movq (%1, %3), %%mm1          \n\t"
402
         "movq 8(%1, %3), %%mm5         \n\t"
403
         "movq %%mm0, (%2)              \n\t"
404
         "movq %%mm4, 8(%2)             \n\t"
405
         "movq %%mm1, (%2, %3)          \n\t"
406
         "movq %%mm5, 8(%2, %3)         \n\t"
407
         "add %%"REG_a", %1             \n\t"
408
         "add %%"REG_a", %2             \n\t"
409
         "subl $4, %0                   \n\t"
410
         "jnz 1b                        \n\t"
411
         : "+g"(h), "+r" (pixels),  "+r" (block)
412
         : "r"((x86_reg)line_size)
413
         : "%"REG_a, "memory"
414
        );
415
}
416

    
417
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
418
{
419
    asm volatile(
420
         "1:                            \n\t"
421
         "movdqu (%1), %%xmm0           \n\t"
422
         "movdqu (%1,%3), %%xmm1        \n\t"
423
         "movdqu (%1,%3,2), %%xmm2      \n\t"
424
         "movdqu (%1,%4), %%xmm3        \n\t"
425
         "movdqa %%xmm0, (%2)           \n\t"
426
         "movdqa %%xmm1, (%2,%3)        \n\t"
427
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
428
         "movdqa %%xmm3, (%2,%4)        \n\t"
429
         "subl $4, %0                   \n\t"
430
         "lea (%1,%3,4), %1             \n\t"
431
         "lea (%2,%3,4), %2             \n\t"
432
         "jnz 1b                        \n\t"
433
         : "+g"(h), "+r" (pixels),  "+r" (block)
434
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
435
         : "memory"
436
        );
437
}
438

    
439
static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
440
{
441
    asm volatile(
442
         "1:                            \n\t"
443
         "movdqu (%1), %%xmm0           \n\t"
444
         "movdqu (%1,%3), %%xmm1        \n\t"
445
         "movdqu (%1,%3,2), %%xmm2      \n\t"
446
         "movdqu (%1,%4), %%xmm3        \n\t"
447
         "pavgb  (%2), %%xmm0           \n\t"
448
         "pavgb  (%2,%3), %%xmm1        \n\t"
449
         "pavgb  (%2,%3,2), %%xmm2      \n\t"
450
         "pavgb  (%2,%4), %%xmm3        \n\t"
451
         "movdqa %%xmm0, (%2)           \n\t"
452
         "movdqa %%xmm1, (%2,%3)        \n\t"
453
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
454
         "movdqa %%xmm3, (%2,%4)        \n\t"
455
         "subl $4, %0                   \n\t"
456
         "lea (%1,%3,4), %1             \n\t"
457
         "lea (%2,%3,4), %2             \n\t"
458
         "jnz 1b                        \n\t"
459
         : "+g"(h), "+r" (pixels),  "+r" (block)
460
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
461
         : "memory"
462
        );
463
}
464

    
465
static void clear_blocks_mmx(DCTELEM *blocks)
466
{
467
    asm volatile(
468
                "pxor %%mm7, %%mm7              \n\t"
469
                "mov $-128*6, %%"REG_a"         \n\t"
470
                "1:                             \n\t"
471
                "movq %%mm7, (%0, %%"REG_a")    \n\t"
472
                "movq %%mm7, 8(%0, %%"REG_a")   \n\t"
473
                "movq %%mm7, 16(%0, %%"REG_a")  \n\t"
474
                "movq %%mm7, 24(%0, %%"REG_a")  \n\t"
475
                "add $32, %%"REG_a"             \n\t"
476
                " js 1b                         \n\t"
477
                : : "r" (((uint8_t *)blocks)+128*6)
478
                : "%"REG_a
479
        );
480
}
481

    
482
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
483
    x86_reg i=0;
484
    asm volatile(
485
        "jmp 2f                         \n\t"
486
        "1:                             \n\t"
487
        "movq  (%1, %0), %%mm0          \n\t"
488
        "movq  (%2, %0), %%mm1          \n\t"
489
        "paddb %%mm0, %%mm1             \n\t"
490
        "movq %%mm1, (%2, %0)           \n\t"
491
        "movq 8(%1, %0), %%mm0          \n\t"
492
        "movq 8(%2, %0), %%mm1          \n\t"
493
        "paddb %%mm0, %%mm1             \n\t"
494
        "movq %%mm1, 8(%2, %0)          \n\t"
495
        "add $16, %0                    \n\t"
496
        "2:                             \n\t"
497
        "cmp %3, %0                     \n\t"
498
        " js 1b                         \n\t"
499
        : "+r" (i)
500
        : "r"(src), "r"(dst), "r"((x86_reg)w-15)
501
    );
502
    for(; i<w; i++)
503
        dst[i+0] += src[i+0];
504
}
505

    
506
static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
507
    x86_reg i=0;
508
    asm volatile(
509
        "jmp 2f                         \n\t"
510
        "1:                             \n\t"
511
        "movq   (%2, %0), %%mm0         \n\t"
512
        "movq  8(%2, %0), %%mm1         \n\t"
513
        "paddb  (%3, %0), %%mm0         \n\t"
514
        "paddb 8(%3, %0), %%mm1         \n\t"
515
        "movq %%mm0,  (%1, %0)          \n\t"
516
        "movq %%mm1, 8(%1, %0)          \n\t"
517
        "add $16, %0                    \n\t"
518
        "2:                             \n\t"
519
        "cmp %4, %0                     \n\t"
520
        " js 1b                         \n\t"
521
        : "+r" (i)
522
        : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
523
    );
524
    for(; i<w; i++)
525
        dst[i] = src1[i] + src2[i];
526
}
527

    
528
#define H263_LOOP_FILTER \
529
        "pxor %%mm7, %%mm7              \n\t"\
530
        "movq  %0, %%mm0                \n\t"\
531
        "movq  %0, %%mm1                \n\t"\
532
        "movq  %3, %%mm2                \n\t"\
533
        "movq  %3, %%mm3                \n\t"\
534
        "punpcklbw %%mm7, %%mm0         \n\t"\
535
        "punpckhbw %%mm7, %%mm1         \n\t"\
536
        "punpcklbw %%mm7, %%mm2         \n\t"\
537
        "punpckhbw %%mm7, %%mm3         \n\t"\
538
        "psubw %%mm2, %%mm0             \n\t"\
539
        "psubw %%mm3, %%mm1             \n\t"\
540
        "movq  %1, %%mm2                \n\t"\
541
        "movq  %1, %%mm3                \n\t"\
542
        "movq  %2, %%mm4                \n\t"\
543
        "movq  %2, %%mm5                \n\t"\
544
        "punpcklbw %%mm7, %%mm2         \n\t"\
545
        "punpckhbw %%mm7, %%mm3         \n\t"\
546
        "punpcklbw %%mm7, %%mm4         \n\t"\
547
        "punpckhbw %%mm7, %%mm5         \n\t"\
548
        "psubw %%mm2, %%mm4             \n\t"\
549
        "psubw %%mm3, %%mm5             \n\t"\
550
        "psllw $2, %%mm4                \n\t"\
551
        "psllw $2, %%mm5                \n\t"\
552
        "paddw %%mm0, %%mm4             \n\t"\
553
        "paddw %%mm1, %%mm5             \n\t"\
554
        "pxor %%mm6, %%mm6              \n\t"\
555
        "pcmpgtw %%mm4, %%mm6           \n\t"\
556
        "pcmpgtw %%mm5, %%mm7           \n\t"\
557
        "pxor %%mm6, %%mm4              \n\t"\
558
        "pxor %%mm7, %%mm5              \n\t"\
559
        "psubw %%mm6, %%mm4             \n\t"\
560
        "psubw %%mm7, %%mm5             \n\t"\
561
        "psrlw $3, %%mm4                \n\t"\
562
        "psrlw $3, %%mm5                \n\t"\
563
        "packuswb %%mm5, %%mm4          \n\t"\
564
        "packsswb %%mm7, %%mm6          \n\t"\
565
        "pxor %%mm7, %%mm7              \n\t"\
566
        "movd %4, %%mm2                 \n\t"\
567
        "punpcklbw %%mm2, %%mm2         \n\t"\
568
        "punpcklbw %%mm2, %%mm2         \n\t"\
569
        "punpcklbw %%mm2, %%mm2         \n\t"\
570
        "psubusb %%mm4, %%mm2           \n\t"\
571
        "movq %%mm2, %%mm3              \n\t"\
572
        "psubusb %%mm4, %%mm3           \n\t"\
573
        "psubb %%mm3, %%mm2             \n\t"\
574
        "movq %1, %%mm3                 \n\t"\
575
        "movq %2, %%mm4                 \n\t"\
576
        "pxor %%mm6, %%mm3              \n\t"\
577
        "pxor %%mm6, %%mm4              \n\t"\
578
        "paddusb %%mm2, %%mm3           \n\t"\
579
        "psubusb %%mm2, %%mm4           \n\t"\
580
        "pxor %%mm6, %%mm3              \n\t"\
581
        "pxor %%mm6, %%mm4              \n\t"\
582
        "paddusb %%mm2, %%mm2           \n\t"\
583
        "packsswb %%mm1, %%mm0          \n\t"\
584
        "pcmpgtb %%mm0, %%mm7           \n\t"\
585
        "pxor %%mm7, %%mm0              \n\t"\
586
        "psubb %%mm7, %%mm0             \n\t"\
587
        "movq %%mm0, %%mm1              \n\t"\
588
        "psubusb %%mm2, %%mm0           \n\t"\
589
        "psubb %%mm0, %%mm1             \n\t"\
590
        "pand %5, %%mm1                 \n\t"\
591
        "psrlw $2, %%mm1                \n\t"\
592
        "pxor %%mm7, %%mm1              \n\t"\
593
        "psubb %%mm7, %%mm1             \n\t"\
594
        "movq %0, %%mm5                 \n\t"\
595
        "movq %3, %%mm6                 \n\t"\
596
        "psubb %%mm1, %%mm5             \n\t"\
597
        "paddb %%mm1, %%mm6             \n\t"
598

    
599
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
600
    if(ENABLE_ANY_H263) {
601
    const int strength= ff_h263_loop_filter_strength[qscale];
602

    
603
    asm volatile(
604

    
605
        H263_LOOP_FILTER
606

    
607
        "movq %%mm3, %1                 \n\t"
608
        "movq %%mm4, %2                 \n\t"
609
        "movq %%mm5, %0                 \n\t"
610
        "movq %%mm6, %3                 \n\t"
611
        : "+m" (*(uint64_t*)(src - 2*stride)),
612
          "+m" (*(uint64_t*)(src - 1*stride)),
613
          "+m" (*(uint64_t*)(src + 0*stride)),
614
          "+m" (*(uint64_t*)(src + 1*stride))
615
        : "g" (2*strength), "m"(ff_pb_FC)
616
    );
617
    }
618
}
619

    
620
static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
621
    asm volatile( //FIXME could save 1 instruction if done as 8x4 ...
622
        "movd  %4, %%mm0                \n\t"
623
        "movd  %5, %%mm1                \n\t"
624
        "movd  %6, %%mm2                \n\t"
625
        "movd  %7, %%mm3                \n\t"
626
        "punpcklbw %%mm1, %%mm0         \n\t"
627
        "punpcklbw %%mm3, %%mm2         \n\t"
628
        "movq %%mm0, %%mm1              \n\t"
629
        "punpcklwd %%mm2, %%mm0         \n\t"
630
        "punpckhwd %%mm2, %%mm1         \n\t"
631
        "movd  %%mm0, %0                \n\t"
632
        "punpckhdq %%mm0, %%mm0         \n\t"
633
        "movd  %%mm0, %1                \n\t"
634
        "movd  %%mm1, %2                \n\t"
635
        "punpckhdq %%mm1, %%mm1         \n\t"
636
        "movd  %%mm1, %3                \n\t"
637

    
638
        : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
639
          "=m" (*(uint32_t*)(dst + 1*dst_stride)),
640
          "=m" (*(uint32_t*)(dst + 2*dst_stride)),
641
          "=m" (*(uint32_t*)(dst + 3*dst_stride))
642
        :  "m" (*(uint32_t*)(src + 0*src_stride)),
643
           "m" (*(uint32_t*)(src + 1*src_stride)),
644
           "m" (*(uint32_t*)(src + 2*src_stride)),
645
           "m" (*(uint32_t*)(src + 3*src_stride))
646
    );
647
}
648

    
649
static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
650
    if(ENABLE_ANY_H263) {
651
    const int strength= ff_h263_loop_filter_strength[qscale];
652
    DECLARE_ALIGNED(8, uint64_t, temp[4]);
653
    uint8_t *btemp= (uint8_t*)temp;
654

    
655
    src -= 2;
656

    
657
    transpose4x4(btemp  , src           , 8, stride);
658
    transpose4x4(btemp+4, src + 4*stride, 8, stride);
659
    asm volatile(
660
        H263_LOOP_FILTER // 5 3 4 6
661

    
662
        : "+m" (temp[0]),
663
          "+m" (temp[1]),
664
          "+m" (temp[2]),
665
          "+m" (temp[3])
666
        : "g" (2*strength), "m"(ff_pb_FC)
667
    );
668

    
669
    asm volatile(
670
        "movq %%mm5, %%mm1              \n\t"
671
        "movq %%mm4, %%mm0              \n\t"
672
        "punpcklbw %%mm3, %%mm5         \n\t"
673
        "punpcklbw %%mm6, %%mm4         \n\t"
674
        "punpckhbw %%mm3, %%mm1         \n\t"
675
        "punpckhbw %%mm6, %%mm0         \n\t"
676
        "movq %%mm5, %%mm3              \n\t"
677
        "movq %%mm1, %%mm6              \n\t"
678
        "punpcklwd %%mm4, %%mm5         \n\t"
679
        "punpcklwd %%mm0, %%mm1         \n\t"
680
        "punpckhwd %%mm4, %%mm3         \n\t"
681
        "punpckhwd %%mm0, %%mm6         \n\t"
682
        "movd %%mm5, (%0)               \n\t"
683
        "punpckhdq %%mm5, %%mm5         \n\t"
684
        "movd %%mm5, (%0,%2)            \n\t"
685
        "movd %%mm3, (%0,%2,2)          \n\t"
686
        "punpckhdq %%mm3, %%mm3         \n\t"
687
        "movd %%mm3, (%0,%3)            \n\t"
688
        "movd %%mm1, (%1)               \n\t"
689
        "punpckhdq %%mm1, %%mm1         \n\t"
690
        "movd %%mm1, (%1,%2)            \n\t"
691
        "movd %%mm6, (%1,%2,2)          \n\t"
692
        "punpckhdq %%mm6, %%mm6         \n\t"
693
        "movd %%mm6, (%1,%3)            \n\t"
694
        :: "r" (src),
695
           "r" (src + 4*stride),
696
           "r" ((x86_reg)   stride ),
697
           "r" ((x86_reg)(3*stride))
698
    );
699
    }
700
}
701

    
702
/* draw the edges of width 'w' of an image of size width, height
703
   this mmx version can only handle w==8 || w==16 */
704
static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
705
{
706
    uint8_t *ptr, *last_line;
707
    int i;
708

    
709
    last_line = buf + (height - 1) * wrap;
710
    /* left and right */
711
    ptr = buf;
712
    if(w==8)
713
    {
714
        asm volatile(
715
                "1:                             \n\t"
716
                "movd (%0), %%mm0               \n\t"
717
                "punpcklbw %%mm0, %%mm0         \n\t"
718
                "punpcklwd %%mm0, %%mm0         \n\t"
719
                "punpckldq %%mm0, %%mm0         \n\t"
720
                "movq %%mm0, -8(%0)             \n\t"
721
                "movq -8(%0, %2), %%mm1         \n\t"
722
                "punpckhbw %%mm1, %%mm1         \n\t"
723
                "punpckhwd %%mm1, %%mm1         \n\t"
724
                "punpckhdq %%mm1, %%mm1         \n\t"
725
                "movq %%mm1, (%0, %2)           \n\t"
726
                "add %1, %0                     \n\t"
727
                "cmp %3, %0                     \n\t"
728
                " jb 1b                         \n\t"
729
                : "+r" (ptr)
730
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
731
        );
732
    }
733
    else
734
    {
735
        asm volatile(
736
                "1:                             \n\t"
737
                "movd (%0), %%mm0               \n\t"
738
                "punpcklbw %%mm0, %%mm0         \n\t"
739
                "punpcklwd %%mm0, %%mm0         \n\t"
740
                "punpckldq %%mm0, %%mm0         \n\t"
741
                "movq %%mm0, -8(%0)             \n\t"
742
                "movq %%mm0, -16(%0)            \n\t"
743
                "movq -8(%0, %2), %%mm1         \n\t"
744
                "punpckhbw %%mm1, %%mm1         \n\t"
745
                "punpckhwd %%mm1, %%mm1         \n\t"
746
                "punpckhdq %%mm1, %%mm1         \n\t"
747
                "movq %%mm1, (%0, %2)           \n\t"
748
                "movq %%mm1, 8(%0, %2)          \n\t"
749
                "add %1, %0                     \n\t"
750
                "cmp %3, %0                     \n\t"
751
                " jb 1b                         \n\t"
752
                : "+r" (ptr)
753
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
754
        );
755
    }
756

    
757
    for(i=0;i<w;i+=4) {
758
        /* top and bottom (and hopefully also the corners) */
759
        ptr= buf - (i + 1) * wrap - w;
760
        asm volatile(
761
                "1:                             \n\t"
762
                "movq (%1, %0), %%mm0           \n\t"
763
                "movq %%mm0, (%0)               \n\t"
764
                "movq %%mm0, (%0, %2)           \n\t"
765
                "movq %%mm0, (%0, %2, 2)        \n\t"
766
                "movq %%mm0, (%0, %3)           \n\t"
767
                "add $8, %0                     \n\t"
768
                "cmp %4, %0                     \n\t"
769
                " jb 1b                         \n\t"
770
                : "+r" (ptr)
771
                : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
772
        );
773
        ptr= last_line + (i + 1) * wrap - w;
774
        asm volatile(
775
                "1:                             \n\t"
776
                "movq (%1, %0), %%mm0           \n\t"
777
                "movq %%mm0, (%0)               \n\t"
778
                "movq %%mm0, (%0, %2)           \n\t"
779
                "movq %%mm0, (%0, %2, 2)        \n\t"
780
                "movq %%mm0, (%0, %3)           \n\t"
781
                "add $8, %0                     \n\t"
782
                "cmp %4, %0                     \n\t"
783
                " jb 1b                         \n\t"
784
                : "+r" (ptr)
785
                : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
786
        );
787
    }
788
}
789

    
790
#define PAETH(cpu, abs3)\
791
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
792
{\
793
    x86_reg i = -bpp;\
794
    x86_reg end = w-3;\
795
    asm volatile(\
796
        "pxor      %%mm7, %%mm7 \n"\
797
        "movd    (%1,%0), %%mm0 \n"\
798
        "movd    (%2,%0), %%mm1 \n"\
799
        "punpcklbw %%mm7, %%mm0 \n"\
800
        "punpcklbw %%mm7, %%mm1 \n"\
801
        "add       %4, %0 \n"\
802
        "1: \n"\
803
        "movq      %%mm1, %%mm2 \n"\
804
        "movd    (%2,%0), %%mm1 \n"\
805
        "movq      %%mm2, %%mm3 \n"\
806
        "punpcklbw %%mm7, %%mm1 \n"\
807
        "movq      %%mm2, %%mm4 \n"\
808
        "psubw     %%mm1, %%mm3 \n"\
809
        "psubw     %%mm0, %%mm4 \n"\
810
        "movq      %%mm3, %%mm5 \n"\
811
        "paddw     %%mm4, %%mm5 \n"\
812
        abs3\
813
        "movq      %%mm4, %%mm6 \n"\
814
        "pminsw    %%mm5, %%mm6 \n"\
815
        "pcmpgtw   %%mm6, %%mm3 \n"\
816
        "pcmpgtw   %%mm5, %%mm4 \n"\
817
        "movq      %%mm4, %%mm6 \n"\
818
        "pand      %%mm3, %%mm4 \n"\
819
        "pandn     %%mm3, %%mm6 \n"\
820
        "pandn     %%mm0, %%mm3 \n"\
821
        "movd    (%3,%0), %%mm0 \n"\
822
        "pand      %%mm1, %%mm6 \n"\
823
        "pand      %%mm4, %%mm2 \n"\
824
        "punpcklbw %%mm7, %%mm0 \n"\
825
        "movq      %6,    %%mm5 \n"\
826
        "paddw     %%mm6, %%mm0 \n"\
827
        "paddw     %%mm2, %%mm3 \n"\
828
        "paddw     %%mm3, %%mm0 \n"\
829
        "pand      %%mm5, %%mm0 \n"\
830
        "movq      %%mm0, %%mm3 \n"\
831
        "packuswb  %%mm3, %%mm3 \n"\
832
        "movd      %%mm3, (%1,%0) \n"\
833
        "add       %4, %0 \n"\
834
        "cmp       %5, %0 \n"\
835
        "jle 1b \n"\
836
        :"+r"(i)\
837
        :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
838
         "m"(ff_pw_255)\
839
        :"memory"\
840
    );\
841
}
842

    
843
#define ABS3_MMX2\
844
        "psubw     %%mm5, %%mm7 \n"\
845
        "pmaxsw    %%mm7, %%mm5 \n"\
846
        "pxor      %%mm6, %%mm6 \n"\
847
        "pxor      %%mm7, %%mm7 \n"\
848
        "psubw     %%mm3, %%mm6 \n"\
849
        "psubw     %%mm4, %%mm7 \n"\
850
        "pmaxsw    %%mm6, %%mm3 \n"\
851
        "pmaxsw    %%mm7, %%mm4 \n"\
852
        "pxor      %%mm7, %%mm7 \n"
853

    
854
#define ABS3_SSSE3\
855
        "pabsw     %%mm3, %%mm3 \n"\
856
        "pabsw     %%mm4, %%mm4 \n"\
857
        "pabsw     %%mm5, %%mm5 \n"
858

    
859
PAETH(mmx2, ABS3_MMX2)
860
#ifdef HAVE_SSSE3
861
PAETH(ssse3, ABS3_SSSE3)
862
#endif
863

    
864
#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
865
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\
866
        "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\
867
        "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\
868
        "movq "#in7", " #m3 "             \n\t" /* d */\
869
        "movq "#in0", %%mm5               \n\t" /* D */\
870
        "paddw " #m3 ", %%mm5             \n\t" /* x4 */\
871
        "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\
872
        "movq "#in1", %%mm5               \n\t" /* C */\
873
        "movq "#in2", %%mm6               \n\t" /* B */\
874
        "paddw " #m6 ", %%mm5             \n\t" /* x3 */\
875
        "paddw " #m5 ", %%mm6             \n\t" /* x2 */\
876
        "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\
877
        "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\
878
        "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\
879
        "paddw " #rnd ", %%mm4            \n\t" /* x2 */\
880
        "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
881
        "psraw $5, %%mm5                  \n\t"\
882
        "packuswb %%mm5, %%mm5            \n\t"\
883
        OP(%%mm5, out, %%mm7, d)
884

    
885
#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
886
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
887
    uint64_t temp;\
888
\
889
    asm volatile(\
890
        "pxor %%mm7, %%mm7                \n\t"\
891
        "1:                               \n\t"\
892
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
893
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
894
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
895
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
896
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
897
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
898
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
899
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
900
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
901
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
902
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
903
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
904
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
905
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
906
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
907
        "paddw %%mm3, %%mm5               \n\t" /* b */\
908
        "paddw %%mm2, %%mm6               \n\t" /* c */\
909
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
910
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
911
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
912
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
913
        "paddw %%mm4, %%mm0               \n\t" /* a */\
914
        "paddw %%mm1, %%mm5               \n\t" /* d */\
915
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
916
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
917
        "paddw %6, %%mm6                  \n\t"\
918
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
919
        "psraw $5, %%mm0                  \n\t"\
920
        "movq %%mm0, %5                   \n\t"\
921
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
922
        \
923
        "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\
924
        "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\
925
        "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\
926
        "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\
927
        "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\
928
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\
929
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\
930
        "paddw %%mm0, %%mm2               \n\t" /* b */\
931
        "paddw %%mm5, %%mm3               \n\t" /* c */\
932
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
933
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
934
        "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\
935
        "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\
936
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\
937
        "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\
938
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
939
        "paddw %%mm2, %%mm1               \n\t" /* a */\
940
        "paddw %%mm6, %%mm4               \n\t" /* d */\
941
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
942
        "psubw %%mm4, %%mm3               \n\t" /* - 6b +3c - d */\
943
        "paddw %6, %%mm1                  \n\t"\
944
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b +3c - d */\
945
        "psraw $5, %%mm3                  \n\t"\
946
        "movq %5, %%mm1                   \n\t"\
947
        "packuswb %%mm3, %%mm1            \n\t"\
948
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
949
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
950
        \
951
        "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\
952
        "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\
953
        "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\
954
        "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\
955
        "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\
956
        "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\
957
        "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\
958
        "paddw %%mm1, %%mm5               \n\t" /* b */\
959
        "paddw %%mm4, %%mm0               \n\t" /* c */\
960
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
961
        "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\
962
        "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\
963
        "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\
964
        "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\
965
        "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\
966
        "paddw %%mm3, %%mm2               \n\t" /* d */\
967
        "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\
968
        "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\
969
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\
970
        "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\
971
        "paddw %%mm2, %%mm6               \n\t" /* a */\
972
        "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
973
        "paddw %6, %%mm0                  \n\t"\
974
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
975
        "psraw $5, %%mm0                  \n\t"\
976
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
977
        \
978
        "paddw %%mm5, %%mm3               \n\t" /* a */\
979
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\
980
        "paddw %%mm4, %%mm6               \n\t" /* b */\
981
        "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\
982
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\
983
        "paddw %%mm1, %%mm4               \n\t" /* c */\
984
        "paddw %%mm2, %%mm5               \n\t" /* d */\
985
        "paddw %%mm6, %%mm6               \n\t" /* 2b */\
986
        "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\
987
        "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
988
        "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\
989
        "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\
990
        "paddw %6, %%mm4                  \n\t"\
991
        "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\
992
        "psraw $5, %%mm4                  \n\t"\
993
        "packuswb %%mm4, %%mm0            \n\t"\
994
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
995
        \
996
        "add %3, %0                       \n\t"\
997
        "add %4, %1                       \n\t"\
998
        "decl %2                          \n\t"\
999
        " jnz 1b                          \n\t"\
1000
        : "+a"(src), "+c"(dst), "+D"(h)\
1001
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1002
        : "memory"\
1003
    );\
1004
}\
1005
\
1006
static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1007
    int i;\
1008
    int16_t temp[16];\
1009
    /* quick HACK, XXX FIXME MUST be optimized */\
1010
    for(i=0; i<h; i++)\
1011
    {\
1012
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1013
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1014
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1015
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1016
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1017
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1018
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1019
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1020
        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1021
        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1022
        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1023
        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1024
        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1025
        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1026
        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1027
        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1028
        asm volatile(\
1029
            "movq (%0), %%mm0               \n\t"\
1030
            "movq 8(%0), %%mm1              \n\t"\
1031
            "paddw %2, %%mm0                \n\t"\
1032
            "paddw %2, %%mm1                \n\t"\
1033
            "psraw $5, %%mm0                \n\t"\
1034
            "psraw $5, %%mm1                \n\t"\
1035
            "packuswb %%mm1, %%mm0          \n\t"\
1036
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1037
            "movq 16(%0), %%mm0             \n\t"\
1038
            "movq 24(%0), %%mm1             \n\t"\
1039
            "paddw %2, %%mm0                \n\t"\
1040
            "paddw %2, %%mm1                \n\t"\
1041
            "psraw $5, %%mm0                \n\t"\
1042
            "psraw $5, %%mm1                \n\t"\
1043
            "packuswb %%mm1, %%mm0          \n\t"\
1044
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1045
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1046
            : "memory"\
1047
        );\
1048
        dst+=dstStride;\
1049
        src+=srcStride;\
1050
    }\
1051
}\
1052
\
1053
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1054
    asm volatile(\
1055
        "pxor %%mm7, %%mm7                \n\t"\
1056
        "1:                               \n\t"\
1057
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
1058
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
1059
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1060
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
1061
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
1062
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
1063
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
1064
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1065
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1066
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
1067
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
1068
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
1069
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
1070
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
1071
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
1072
        "paddw %%mm3, %%mm5               \n\t" /* b */\
1073
        "paddw %%mm2, %%mm6               \n\t" /* c */\
1074
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1075
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
1076
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
1077
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
1078
        "paddw %%mm4, %%mm0               \n\t" /* a */\
1079
        "paddw %%mm1, %%mm5               \n\t" /* d */\
1080
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1081
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
1082
        "paddw %5, %%mm6                  \n\t"\
1083
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1084
        "psraw $5, %%mm0                  \n\t"\
1085
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1086
        \
1087
        "movd 5(%0), %%mm5                \n\t" /* FGHI */\
1088
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\
1089
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\
1090
        "paddw %%mm5, %%mm1               \n\t" /* a */\
1091
        "paddw %%mm6, %%mm2               \n\t" /* b */\
1092
        "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\
1093
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\
1094
        "paddw %%mm6, %%mm3               \n\t" /* c */\
1095
        "paddw %%mm5, %%mm4               \n\t" /* d */\
1096
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1097
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1098
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1099
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1100
        "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\
1101
        "paddw %5, %%mm1                  \n\t"\
1102
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\
1103
        "psraw $5, %%mm3                  \n\t"\
1104
        "packuswb %%mm3, %%mm0            \n\t"\
1105
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
1106
        \
1107
        "add %3, %0                       \n\t"\
1108
        "add %4, %1                       \n\t"\
1109
        "decl %2                          \n\t"\
1110
        " jnz 1b                          \n\t"\
1111
        : "+a"(src), "+c"(dst), "+d"(h)\
1112
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
1113
        : "memory"\
1114
    );\
1115
}\
1116
\
1117
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1118
    int i;\
1119
    int16_t temp[8];\
1120
    /* quick HACK, XXX FIXME MUST be optimized */\
1121
    for(i=0; i<h; i++)\
1122
    {\
1123
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1124
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1125
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1126
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1127
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1128
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1129
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1130
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1131
        asm volatile(\
1132
            "movq (%0), %%mm0           \n\t"\
1133
            "movq 8(%0), %%mm1          \n\t"\
1134
            "paddw %2, %%mm0            \n\t"\
1135
            "paddw %2, %%mm1            \n\t"\
1136
            "psraw $5, %%mm0            \n\t"\
1137
            "psraw $5, %%mm1            \n\t"\
1138
            "packuswb %%mm1, %%mm0      \n\t"\
1139
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1140
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1141
            :"memory"\
1142
        );\
1143
        dst+=dstStride;\
1144
        src+=srcStride;\
1145
    }\
1146
}
1147

    
1148
#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1149
\
1150
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1151
    uint64_t temp[17*4];\
1152
    uint64_t *temp_ptr= temp;\
1153
    int count= 17;\
1154
\
1155
    /*FIXME unroll */\
1156
    asm volatile(\
1157
        "pxor %%mm7, %%mm7              \n\t"\
1158
        "1:                             \n\t"\
1159
        "movq (%0), %%mm0               \n\t"\
1160
        "movq (%0), %%mm1               \n\t"\
1161
        "movq 8(%0), %%mm2              \n\t"\
1162
        "movq 8(%0), %%mm3              \n\t"\
1163
        "punpcklbw %%mm7, %%mm0         \n\t"\
1164
        "punpckhbw %%mm7, %%mm1         \n\t"\
1165
        "punpcklbw %%mm7, %%mm2         \n\t"\
1166
        "punpckhbw %%mm7, %%mm3         \n\t"\
1167
        "movq %%mm0, (%1)               \n\t"\
1168
        "movq %%mm1, 17*8(%1)           \n\t"\
1169
        "movq %%mm2, 2*17*8(%1)         \n\t"\
1170
        "movq %%mm3, 3*17*8(%1)         \n\t"\
1171
        "add $8, %1                     \n\t"\
1172
        "add %3, %0                     \n\t"\
1173
        "decl %2                        \n\t"\
1174
        " jnz 1b                        \n\t"\
1175
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1176
        : "r" ((x86_reg)srcStride)\
1177
        : "memory"\
1178
    );\
1179
    \
1180
    temp_ptr= temp;\
1181
    count=4;\
1182
    \
1183
/*FIXME reorder for speed */\
1184
    asm volatile(\
1185
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1186
        "1:                             \n\t"\
1187
        "movq (%0), %%mm0               \n\t"\
1188
        "movq 8(%0), %%mm1              \n\t"\
1189
        "movq 16(%0), %%mm2             \n\t"\
1190
        "movq 24(%0), %%mm3             \n\t"\
1191
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1192
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1193
        "add %4, %1                     \n\t"\
1194
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1195
        \
1196
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1197
        "add %4, %1                     \n\t"\
1198
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1199
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1200
        "add %4, %1                     \n\t"\
1201
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1202
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1203
        "add %4, %1                     \n\t"\
1204
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1205
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1206
        "add %4, %1                     \n\t"\
1207
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1208
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1209
        "add %4, %1                     \n\t"\
1210
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1211
        \
1212
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1213
        "add %4, %1                     \n\t"  \
1214
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1215
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1216
        \
1217
        "add $136, %0                   \n\t"\
1218
        "add %6, %1                     \n\t"\
1219
        "decl %2                        \n\t"\
1220
        " jnz 1b                        \n\t"\
1221
        \
1222
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1223
        : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(x86_reg)dstStride)\
1224
        :"memory"\
1225
    );\
1226
}\
1227
\
1228
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1229
    uint64_t temp[9*2];\
1230
    uint64_t *temp_ptr= temp;\
1231
    int count= 9;\
1232
\
1233
    /*FIXME unroll */\
1234
    asm volatile(\
1235
        "pxor %%mm7, %%mm7              \n\t"\
1236
        "1:                             \n\t"\
1237
        "movq (%0), %%mm0               \n\t"\
1238
        "movq (%0), %%mm1               \n\t"\
1239
        "punpcklbw %%mm7, %%mm0         \n\t"\
1240
        "punpckhbw %%mm7, %%mm1         \n\t"\
1241
        "movq %%mm0, (%1)               \n\t"\
1242
        "movq %%mm1, 9*8(%1)            \n\t"\
1243
        "add $8, %1                     \n\t"\
1244
        "add %3, %0                     \n\t"\
1245
        "decl %2                        \n\t"\
1246
        " jnz 1b                        \n\t"\
1247
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1248
        : "r" ((x86_reg)srcStride)\
1249
        : "memory"\
1250
    );\
1251
    \
1252
    temp_ptr= temp;\
1253
    count=2;\
1254
    \
1255
/*FIXME reorder for speed */\
1256
    asm volatile(\
1257
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1258
        "1:                             \n\t"\
1259
        "movq (%0), %%mm0               \n\t"\
1260
        "movq 8(%0), %%mm1              \n\t"\
1261
        "movq 16(%0), %%mm2             \n\t"\
1262
        "movq 24(%0), %%mm3             \n\t"\
1263
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1264
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1265
        "add %4, %1                     \n\t"\
1266
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1267
        \
1268
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1269
        "add %4, %1                     \n\t"\
1270
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1271
        \
1272
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1273
        "add %4, %1                     \n\t"\
1274
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1275
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1276
                \
1277
        "add $72, %0                    \n\t"\
1278
        "add %6, %1                     \n\t"\
1279
        "decl %2                        \n\t"\
1280
        " jnz 1b                        \n\t"\
1281
         \
1282
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1283
        : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(x86_reg)dstStride)\
1284
        : "memory"\
1285
   );\
1286
}\
1287
\
1288
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1289
    OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\
1290
}\
1291
\
1292
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1293
    uint64_t temp[8];\
1294
    uint8_t * const half= (uint8_t*)temp;\
1295
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1296
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1297
}\
1298
\
1299
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1300
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1301
}\
1302
\
1303
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1304
    uint64_t temp[8];\
1305
    uint8_t * const half= (uint8_t*)temp;\
1306
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1307
    OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
1308
}\
1309
\
1310
static void OPNAME ## qpel8_mc01_ ## 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_v_lowpass_ ## MMX(half, src, 8, stride);\
1314
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1315
}\
1316
\
1317
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1318
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1319
}\
1320
\
1321
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1322
    uint64_t temp[8];\
1323
    uint8_t * const half= (uint8_t*)temp;\
1324
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1325
    OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
1326
}\
1327
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1328
    uint64_t half[8 + 9];\
1329
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1330
    uint8_t * const halfHV= ((uint8_t*)half);\
1331
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1332
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1333
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1334
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1335
}\
1336
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1337
    uint64_t half[8 + 9];\
1338
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1339
    uint8_t * const halfHV= ((uint8_t*)half);\
1340
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1341
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1342
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1343
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1344
}\
1345
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1346
    uint64_t half[8 + 9];\
1347
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1348
    uint8_t * const halfHV= ((uint8_t*)half);\
1349
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1350
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1351
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1352
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1353
}\
1354
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1355
    uint64_t half[8 + 9];\
1356
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1357
    uint8_t * const halfHV= ((uint8_t*)half);\
1358
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1359
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1360
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1361
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1362
}\
1363
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1364
    uint64_t half[8 + 9];\
1365
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1366
    uint8_t * const halfHV= ((uint8_t*)half);\
1367
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1368
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1369
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1370
}\
1371
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1372
    uint64_t half[8 + 9];\
1373
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1374
    uint8_t * const halfHV= ((uint8_t*)half);\
1375
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1376
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1377
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1378
}\
1379
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1380
    uint64_t half[8 + 9];\
1381
    uint8_t * const halfH= ((uint8_t*)half);\
1382
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1383
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1384
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1385
}\
1386
static void OPNAME ## qpel8_mc32_ ## 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+1, halfH, 8, stride, 9);\
1391
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1392
}\
1393
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1394
    uint64_t half[9];\
1395
    uint8_t * const halfH= ((uint8_t*)half);\
1396
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1397
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1398
}\
1399
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1400
    OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\
1401
}\
1402
\
1403
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1404
    uint64_t temp[32];\
1405
    uint8_t * const half= (uint8_t*)temp;\
1406
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1407
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1408
}\
1409
\
1410
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1411
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1412
}\
1413
\
1414
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1415
    uint64_t temp[32];\
1416
    uint8_t * const half= (uint8_t*)temp;\
1417
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1418
    OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
1419
}\
1420
\
1421
static void OPNAME ## qpel16_mc01_ ## 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_v_lowpass_ ## MMX(half, src, 16, stride);\
1425
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1426
}\
1427
\
1428
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1429
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1430
}\
1431
\
1432
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1433
    uint64_t temp[32];\
1434
    uint8_t * const half= (uint8_t*)temp;\
1435
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1436
    OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
1437
}\
1438
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1439
    uint64_t half[16*2 + 17*2];\
1440
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1441
    uint8_t * const halfHV= ((uint8_t*)half);\
1442
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1443
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1444
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1445
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1446
}\
1447
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1448
    uint64_t half[16*2 + 17*2];\
1449
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1450
    uint8_t * const halfHV= ((uint8_t*)half);\
1451
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1452
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1453
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1454
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1455
}\
1456
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1457
    uint64_t half[16*2 + 17*2];\
1458
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1459
    uint8_t * const halfHV= ((uint8_t*)half);\
1460
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1461
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1462
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1463
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1464
}\
1465
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1466
    uint64_t half[16*2 + 17*2];\
1467
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1468
    uint8_t * const halfHV= ((uint8_t*)half);\
1469
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1470
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1471
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1472
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1473
}\
1474
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1475
    uint64_t half[16*2 + 17*2];\
1476
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1477
    uint8_t * const halfHV= ((uint8_t*)half);\
1478
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1479
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1480
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1481
}\
1482
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1483
    uint64_t half[16*2 + 17*2];\
1484
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1485
    uint8_t * const halfHV= ((uint8_t*)half);\
1486
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1487
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1488
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1489
}\
1490
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1491
    uint64_t half[17*2];\
1492
    uint8_t * const halfH= ((uint8_t*)half);\
1493
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1494
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1495
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1496
}\
1497
static void OPNAME ## qpel16_mc32_ ## 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+1, halfH, 16, stride, 17);\
1502
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1503
}\
1504
static void OPNAME ## qpel16_mc22_ ## 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
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1509
}
1510

    
1511
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"
1512
#define AVG_3DNOW_OP(a,b,temp, size) \
1513
"mov" #size " " #b ", " #temp "   \n\t"\
1514
"pavgusb " #temp ", " #a "        \n\t"\
1515
"mov" #size " " #a ", " #b "      \n\t"
1516
#define AVG_MMX2_OP(a,b,temp, size) \
1517
"mov" #size " " #b ", " #temp "   \n\t"\
1518
"pavgb " #temp ", " #a "          \n\t"\
1519
"mov" #size " " #a ", " #b "      \n\t"
1520

    
1521
QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
1522
QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
1523
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1524
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
1525
QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
1526
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1527
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
1528
QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
1529
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1530

    
1531
/***********************************/
1532
/* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
1533

    
1534
#define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
1535
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1536
    OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
1537
}
1538
#define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
1539
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1540
    OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
1541
}
1542

    
1543
#define QPEL_2TAP(OPNAME, SIZE, MMX)\
1544
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
1545
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
1546
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
1547
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
1548
                          OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
1549
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
1550
                          OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
1551
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
1552
                          OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
1553
static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1554
    OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
1555
}\
1556
static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1557
    OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
1558
}\
1559
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\
1560
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\
1561
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\
1562
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\
1563
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\
1564
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\
1565
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\
1566
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
1567

    
1568
QPEL_2TAP(put_, 16, mmx2)
1569
QPEL_2TAP(avg_, 16, mmx2)
1570
QPEL_2TAP(put_,  8, mmx2)
1571
QPEL_2TAP(avg_,  8, mmx2)
1572
QPEL_2TAP(put_, 16, 3dnow)
1573
QPEL_2TAP(avg_, 16, 3dnow)
1574
QPEL_2TAP(put_,  8, 3dnow)
1575
QPEL_2TAP(avg_,  8, 3dnow)
1576

    
1577

    
1578
#if 0
1579
static void just_return() { return; }
1580
#endif
1581

    
1582
static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1583
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){
1584
    const int w = 8;
1585
    const int ix = ox>>(16+shift);
1586
    const int iy = oy>>(16+shift);
1587
    const int oxs = ox>>4;
1588
    const int oys = oy>>4;
1589
    const int dxxs = dxx>>4;
1590
    const int dxys = dxy>>4;
1591
    const int dyxs = dyx>>4;
1592
    const int dyys = dyy>>4;
1593
    const uint16_t r4[4] = {r,r,r,r};
1594
    const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
1595
    const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
1596
    const uint64_t shift2 = 2*shift;
1597
    uint8_t edge_buf[(h+1)*stride];
1598
    int x, y;
1599

    
1600
    const int dxw = (dxx-(1<<(16+shift)))*(w-1);
1601
    const int dyh = (dyy-(1<<(16+shift)))*(h-1);
1602
    const int dxh = dxy*(h-1);
1603
    const int dyw = dyx*(w-1);
1604
    if( // non-constant fullpel offset (3% of blocks)
1605
        ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |
1606
         (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)
1607
        // uses more than 16 bits of subpel mv (only at huge resolution)
1608
        || (dxx|dxy|dyx|dyy)&15 )
1609
    {
1610
        //FIXME could still use mmx for some of the rows
1611
        ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
1612
        return;
1613
    }
1614

    
1615
    src += ix + iy*stride;
1616
    if( (unsigned)ix >= width-w ||
1617
        (unsigned)iy >= height-h )
1618
    {
1619
        ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
1620
        src = edge_buf;
1621
    }
1622

    
1623
    asm volatile(
1624
        "movd         %0, %%mm6 \n\t"
1625
        "pxor      %%mm7, %%mm7 \n\t"
1626
        "punpcklwd %%mm6, %%mm6 \n\t"
1627
        "punpcklwd %%mm6, %%mm6 \n\t"
1628
        :: "r"(1<<shift)
1629
    );
1630

    
1631
    for(x=0; x<w; x+=4){
1632
        uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
1633
                            oxs - dxys + dxxs*(x+1),
1634
                            oxs - dxys + dxxs*(x+2),
1635
                            oxs - dxys + dxxs*(x+3) };
1636
        uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
1637
                            oys - dyys + dyxs*(x+1),
1638
                            oys - dyys + dyxs*(x+2),
1639
                            oys - dyys + dyxs*(x+3) };
1640

    
1641
        for(y=0; y<h; y++){
1642
            asm volatile(
1643
                "movq   %0,  %%mm4 \n\t"
1644
                "movq   %1,  %%mm5 \n\t"
1645
                "paddw  %2,  %%mm4 \n\t"
1646
                "paddw  %3,  %%mm5 \n\t"
1647
                "movq   %%mm4, %0  \n\t"
1648
                "movq   %%mm5, %1  \n\t"
1649
                "psrlw  $12, %%mm4 \n\t"
1650
                "psrlw  $12, %%mm5 \n\t"
1651
                : "+m"(*dx4), "+m"(*dy4)
1652
                : "m"(*dxy4), "m"(*dyy4)
1653
            );
1654

    
1655
            asm volatile(
1656
                "movq   %%mm6, %%mm2 \n\t"
1657
                "movq   %%mm6, %%mm1 \n\t"
1658
                "psubw  %%mm4, %%mm2 \n\t"
1659
                "psubw  %%mm5, %%mm1 \n\t"
1660
                "movq   %%mm2, %%mm0 \n\t"
1661
                "movq   %%mm4, %%mm3 \n\t"
1662
                "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
1663
                "pmullw %%mm5, %%mm3 \n\t" // dx*dy
1664
                "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
1665
                "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
1666

    
1667
                "movd   %4,    %%mm5 \n\t"
1668
                "movd   %3,    %%mm4 \n\t"
1669
                "punpcklbw %%mm7, %%mm5 \n\t"
1670
                "punpcklbw %%mm7, %%mm4 \n\t"
1671
                "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
1672
                "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
1673

    
1674
                "movd   %2,    %%mm5 \n\t"
1675
                "movd   %1,    %%mm4 \n\t"
1676
                "punpcklbw %%mm7, %%mm5 \n\t"
1677
                "punpcklbw %%mm7, %%mm4 \n\t"
1678
                "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
1679
                "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
1680
                "paddw  %5,    %%mm1 \n\t"
1681
                "paddw  %%mm3, %%mm2 \n\t"
1682
                "paddw  %%mm1, %%mm0 \n\t"
1683
                "paddw  %%mm2, %%mm0 \n\t"
1684

    
1685
                "psrlw    %6,    %%mm0 \n\t"
1686
                "packuswb %%mm0, %%mm0 \n\t"
1687
                "movd     %%mm0, %0    \n\t"
1688

    
1689
                : "=m"(dst[x+y*stride])
1690
                : "m"(src[0]), "m"(src[1]),
1691
                  "m"(src[stride]), "m"(src[stride+1]),
1692
                  "m"(*r4), "m"(shift2)
1693
            );
1694
            src += stride;
1695
        }
1696
        src += 4-h*stride;
1697
    }
1698
}
1699

    
1700
#define PREFETCH(name, op) \
1701
static void name(void *mem, int stride, int h){\
1702
    const uint8_t *p= mem;\
1703
    do{\
1704
        asm volatile(#op" %0" :: "m"(*p));\
1705
        p+= stride;\
1706
    }while(--h);\
1707
}
1708
PREFETCH(prefetch_mmx2,  prefetcht0)
1709
PREFETCH(prefetch_3dnow, prefetch)
1710
#undef PREFETCH
1711

    
1712
#include "h264dsp_mmx.c"
1713

    
1714
/* CAVS specific */
1715
void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);
1716
void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx);
1717

    
1718
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1719
    put_pixels8_mmx(dst, src, stride, 8);
1720
}
1721
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1722
    avg_pixels8_mmx(dst, src, stride, 8);
1723
}
1724
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1725
    put_pixels16_mmx(dst, src, stride, 16);
1726
}
1727
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1728
    avg_pixels16_mmx(dst, src, stride, 16);
1729
}
1730

    
1731
/* VC1 specific */
1732
void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx);
1733

    
1734
void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1735
    put_pixels8_mmx(dst, src, stride, 8);
1736
}
1737

    
1738
/* external functions, from idct_mmx.c */
1739
void ff_mmx_idct(DCTELEM *block);
1740
void ff_mmxext_idct(DCTELEM *block);
1741

    
1742
/* XXX: those functions should be suppressed ASAP when all IDCTs are
1743
   converted */
1744
#ifdef CONFIG_GPL
1745
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1746
{
1747
    ff_mmx_idct (block);
1748
    put_pixels_clamped_mmx(block, dest, line_size);
1749
}
1750
static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1751
{
1752
    ff_mmx_idct (block);
1753
    add_pixels_clamped_mmx(block, dest, line_size);
1754
}
1755
static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1756
{
1757
    ff_mmxext_idct (block);
1758
    put_pixels_clamped_mmx(block, dest, line_size);
1759
}
1760
static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1761
{
1762
    ff_mmxext_idct (block);
1763
    add_pixels_clamped_mmx(block, dest, line_size);
1764
}
1765
#endif
1766
static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
1767
{
1768
    ff_idct_xvid_mmx (block);
1769
    put_pixels_clamped_mmx(block, dest, line_size);
1770
}
1771
static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
1772
{
1773
    ff_idct_xvid_mmx (block);
1774
    add_pixels_clamped_mmx(block, dest, line_size);
1775
}
1776
static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
1777
{
1778
    ff_idct_xvid_mmx2 (block);
1779
    put_pixels_clamped_mmx(block, dest, line_size);
1780
}
1781
static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
1782
{
1783
    ff_idct_xvid_mmx2 (block);
1784
    add_pixels_clamped_mmx(block, dest, line_size);
1785
}
1786

    
1787
static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
1788
{
1789
    int i;
1790
    asm volatile("pxor %%mm7, %%mm7":);
1791
    for(i=0; i<blocksize; i+=2) {
1792
        asm volatile(
1793
            "movq    %0,    %%mm0 \n\t"
1794
            "movq    %1,    %%mm1 \n\t"
1795
            "movq    %%mm0, %%mm2 \n\t"
1796
            "movq    %%mm1, %%mm3 \n\t"
1797
            "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
1798
            "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
1799
            "pslld   $31,   %%mm2 \n\t" // keep only the sign bit
1800
            "pxor    %%mm2, %%mm1 \n\t"
1801
            "movq    %%mm3, %%mm4 \n\t"
1802
            "pand    %%mm1, %%mm3 \n\t"
1803
            "pandn   %%mm1, %%mm4 \n\t"
1804
            "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1805
            "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1806
            "movq    %%mm3, %1    \n\t"
1807
            "movq    %%mm0, %0    \n\t"
1808
            :"+m"(mag[i]), "+m"(ang[i])
1809
            ::"memory"
1810
        );
1811
    }
1812
    asm volatile("femms");
1813
}
1814
static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
1815
{
1816
    int i;
1817

    
1818
    asm volatile(
1819
            "movaps  %0,     %%xmm5 \n\t"
1820
        ::"m"(ff_pdw_80000000[0])
1821
    );
1822
    for(i=0; i<blocksize; i+=4) {
1823
        asm volatile(
1824
            "movaps  %0,     %%xmm0 \n\t"
1825
            "movaps  %1,     %%xmm1 \n\t"
1826
            "xorps   %%xmm2, %%xmm2 \n\t"
1827
            "xorps   %%xmm3, %%xmm3 \n\t"
1828
            "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
1829
            "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
1830
            "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit
1831
            "xorps   %%xmm2, %%xmm1 \n\t"
1832
            "movaps  %%xmm3, %%xmm4 \n\t"
1833
            "andps   %%xmm1, %%xmm3 \n\t"
1834
            "andnps  %%xmm1, %%xmm4 \n\t"
1835
            "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1836
            "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1837
            "movaps  %%xmm3, %1     \n\t"
1838
            "movaps  %%xmm0, %0     \n\t"
1839
            :"+m"(mag[i]), "+m"(ang[i])
1840
            ::"memory"
1841
        );
1842
    }
1843
}
1844

    
1845
#define IF1(x) x
1846
#define IF0(x)
1847

    
1848
#define MIX5(mono,stereo)\
1849
    asm volatile(\
1850
        "movss          0(%2), %%xmm5 \n"\
1851
        "movss          8(%2), %%xmm6 \n"\
1852
        "movss         24(%2), %%xmm7 \n"\
1853
        "shufps    $0, %%xmm5, %%xmm5 \n"\
1854
        "shufps    $0, %%xmm6, %%xmm6 \n"\
1855
        "shufps    $0, %%xmm7, %%xmm7 \n"\
1856
        "1: \n"\
1857
        "movaps       (%0,%1), %%xmm0 \n"\
1858
        "movaps  0x400(%0,%1), %%xmm1 \n"\
1859
        "movaps  0x800(%0,%1), %%xmm2 \n"\
1860
        "movaps  0xc00(%0,%1), %%xmm3 \n"\
1861
        "movaps 0x1000(%0,%1), %%xmm4 \n"\
1862
        "mulps         %%xmm5, %%xmm0 \n"\
1863
        "mulps         %%xmm6, %%xmm1 \n"\
1864
        "mulps         %%xmm5, %%xmm2 \n"\
1865
        "mulps         %%xmm7, %%xmm3 \n"\
1866
        "mulps         %%xmm7, %%xmm4 \n"\
1867
 stereo("addps         %%xmm1, %%xmm0 \n")\
1868
        "addps         %%xmm1, %%xmm2 \n"\
1869
        "addps         %%xmm3, %%xmm0 \n"\
1870
        "addps         %%xmm4, %%xmm2 \n"\
1871
   mono("addps         %%xmm2, %%xmm0 \n")\
1872
        "movaps  %%xmm0,      (%0,%1) \n"\
1873
 stereo("movaps  %%xmm2, 0x400(%0,%1) \n")\
1874
        "add $16, %0 \n"\
1875
        "jl 1b \n"\
1876
        :"+&r"(i)\
1877
        :"r"(samples[0]+len), "r"(matrix)\
1878
        :"memory"\
1879
    );
1880

    
1881
#define MIX_MISC(stereo)\
1882
    asm volatile(\
1883
        "1: \n"\
1884
        "movaps  (%3,%0), %%xmm0 \n"\
1885
 stereo("movaps   %%xmm0, %%xmm1 \n")\
1886
        "mulps    %%xmm6, %%xmm0 \n"\
1887
 stereo("mulps    %%xmm7, %%xmm1 \n")\
1888
        "lea 1024(%3,%0), %1 \n"\
1889
        "mov %5, %2 \n"\
1890
        "2: \n"\
1891
        "movaps   (%1),   %%xmm2 \n"\
1892
 stereo("movaps   %%xmm2, %%xmm3 \n")\
1893
        "mulps   (%4,%2), %%xmm2 \n"\
1894
 stereo("mulps 16(%4,%2), %%xmm3 \n")\
1895
        "addps    %%xmm2, %%xmm0 \n"\
1896
 stereo("addps    %%xmm3, %%xmm1 \n")\
1897
        "add $1024, %1 \n"\
1898
        "add $32, %2 \n"\
1899
        "jl 2b \n"\
1900
        "movaps   %%xmm0,     (%3,%0) \n"\
1901
 stereo("movaps   %%xmm1, 1024(%3,%0) \n")\
1902
        "add $16, %0 \n"\
1903
        "jl 1b \n"\
1904
        :"+&r"(i), "=&r"(j), "=&r"(k)\
1905
        :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\
1906
        :"memory"\
1907
    );
1908

    
1909
static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
1910
{
1911
    int (*matrix_cmp)[2] = (int(*)[2])matrix;
1912
    intptr_t i,j,k;
1913

    
1914
    i = -len*sizeof(float);
1915
    if(in_ch == 5 && out_ch == 2 && !(matrix_cmp[0][1]|matrix_cmp[2][0]|matrix_cmp[3][1]|matrix_cmp[4][0]|(matrix_cmp[1][0]^matrix_cmp[1][1])|(matrix_cmp[0][0]^matrix_cmp[2][1]))) {
1916
        MIX5(IF0,IF1);
1917
    } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {
1918
        MIX5(IF1,IF0);
1919
    } else {
1920
        DECLARE_ALIGNED_16(float, matrix_simd[in_ch][2][4]);
1921
        j = 2*in_ch*sizeof(float);
1922
        asm volatile(
1923
            "1: \n"
1924
            "sub $8, %0 \n"
1925
            "movss     (%2,%0), %%xmm6 \n"
1926
            "movss    4(%2,%0), %%xmm7 \n"
1927
            "shufps $0, %%xmm6, %%xmm6 \n"
1928
            "shufps $0, %%xmm7, %%xmm7 \n"
1929
            "movaps %%xmm6,   (%1,%0,4) \n"
1930
            "movaps %%xmm7, 16(%1,%0,4) \n"
1931
            "jg 1b \n"
1932
            :"+&r"(j)
1933
            :"r"(matrix_simd), "r"(matrix)
1934
            :"memory"
1935
        );
1936
        if(out_ch == 2) {
1937
            MIX_MISC(IF1);
1938
        } else {
1939
            MIX_MISC(IF0);
1940
        }
1941
    }
1942
}
1943

    
1944
static void vector_fmul_3dnow(float *dst, const float *src, int len){
1945
    x86_reg i = (len-4)*4;
1946
    asm volatile(
1947
        "1: \n\t"
1948
        "movq    (%1,%0), %%mm0 \n\t"
1949
        "movq   8(%1,%0), %%mm1 \n\t"
1950
        "pfmul   (%2,%0), %%mm0 \n\t"
1951
        "pfmul  8(%2,%0), %%mm1 \n\t"
1952
        "movq   %%mm0,  (%1,%0) \n\t"
1953
        "movq   %%mm1, 8(%1,%0) \n\t"
1954
        "sub  $16, %0 \n\t"
1955
        "jge 1b \n\t"
1956
        "femms  \n\t"
1957
        :"+r"(i)
1958
        :"r"(dst), "r"(src)
1959
        :"memory"
1960
    );
1961
}
1962
static void vector_fmul_sse(float *dst, const float *src, int len){
1963
    x86_reg i = (len-8)*4;
1964
    asm volatile(
1965
        "1: \n\t"
1966
        "movaps    (%1,%0), %%xmm0 \n\t"
1967
        "movaps  16(%1,%0), %%xmm1 \n\t"
1968
        "mulps     (%2,%0), %%xmm0 \n\t"
1969
        "mulps   16(%2,%0), %%xmm1 \n\t"
1970
        "movaps  %%xmm0,   (%1,%0) \n\t"
1971
        "movaps  %%xmm1, 16(%1,%0) \n\t"
1972
        "sub  $32, %0 \n\t"
1973
        "jge 1b \n\t"
1974
        :"+r"(i)
1975
        :"r"(dst), "r"(src)
1976
        :"memory"
1977
    );
1978
}
1979

    
1980
static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
1981
    x86_reg i = len*4-16;
1982
    asm volatile(
1983
        "1: \n\t"
1984
        "pswapd   8(%1), %%mm0 \n\t"
1985
        "pswapd    (%1), %%mm1 \n\t"
1986
        "pfmul  (%3,%0), %%mm0 \n\t"
1987
        "pfmul 8(%3,%0), %%mm1 \n\t"
1988
        "movq  %%mm0,  (%2,%0) \n\t"
1989
        "movq  %%mm1, 8(%2,%0) \n\t"
1990
        "add   $16, %1 \n\t"
1991
        "sub   $16, %0 \n\t"
1992
        "jge   1b \n\t"
1993
        :"+r"(i), "+r"(src1)
1994
        :"r"(dst), "r"(src0)
1995
    );
1996
    asm volatile("femms");
1997
}
1998
static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
1999
    x86_reg i = len*4-32;
2000
    asm volatile(
2001
        "1: \n\t"
2002
        "movaps        16(%1), %%xmm0 \n\t"
2003
        "movaps          (%1), %%xmm1 \n\t"
2004
        "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
2005
        "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
2006
        "mulps        (%3,%0), %%xmm0 \n\t"
2007
        "mulps      16(%3,%0), %%xmm1 \n\t"
2008
        "movaps     %%xmm0,   (%2,%0) \n\t"
2009
        "movaps     %%xmm1, 16(%2,%0) \n\t"
2010
        "add    $32, %1 \n\t"
2011
        "sub    $32, %0 \n\t"
2012
        "jge    1b \n\t"
2013
        :"+r"(i), "+r"(src1)
2014
        :"r"(dst), "r"(src0)
2015
    );
2016
}
2017

    
2018
static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1,
2019
                                      const float *src2, int src3, int len, int step){
2020
    x86_reg i = (len-4)*4;
2021
    if(step == 2 && src3 == 0){
2022
        dst += (len-4)*2;
2023
        asm volatile(
2024
            "1: \n\t"
2025
            "movq   (%2,%0),  %%mm0 \n\t"
2026
            "movq  8(%2,%0),  %%mm1 \n\t"
2027
            "pfmul  (%3,%0),  %%mm0 \n\t"
2028
            "pfmul 8(%3,%0),  %%mm1 \n\t"
2029
            "pfadd  (%4,%0),  %%mm0 \n\t"
2030
            "pfadd 8(%4,%0),  %%mm1 \n\t"
2031
            "movd     %%mm0,   (%1) \n\t"
2032
            "movd     %%mm1, 16(%1) \n\t"
2033
            "psrlq      $32,  %%mm0 \n\t"
2034
            "psrlq      $32,  %%mm1 \n\t"
2035
            "movd     %%mm0,  8(%1) \n\t"
2036
            "movd     %%mm1, 24(%1) \n\t"
2037
            "sub  $32, %1 \n\t"
2038
            "sub  $16, %0 \n\t"
2039
            "jge  1b \n\t"
2040
            :"+r"(i), "+r"(dst)
2041
            :"r"(src0), "r"(src1), "r"(src2)
2042
            :"memory"
2043
        );
2044
    }
2045
    else if(step == 1 && src3 == 0){
2046
        asm volatile(
2047
            "1: \n\t"
2048
            "movq    (%2,%0), %%mm0 \n\t"
2049
            "movq   8(%2,%0), %%mm1 \n\t"
2050
            "pfmul   (%3,%0), %%mm0 \n\t"
2051
            "pfmul  8(%3,%0), %%mm1 \n\t"
2052
            "pfadd   (%4,%0), %%mm0 \n\t"
2053
            "pfadd  8(%4,%0), %%mm1 \n\t"
2054
            "movq  %%mm0,   (%1,%0) \n\t"
2055
            "movq  %%mm1,  8(%1,%0) \n\t"
2056
            "sub  $16, %0 \n\t"
2057
            "jge  1b \n\t"
2058
            :"+r"(i)
2059
            :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2060
            :"memory"
2061
        );
2062
    }
2063
    else
2064
        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
2065
    asm volatile("femms");
2066
}
2067
static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1,
2068
                                    const float *src2, int src3, int len, int step){
2069
    x86_reg i = (len-8)*4;
2070
    if(step == 2 && src3 == 0){
2071
        dst += (len-8)*2;
2072
        asm volatile(
2073
            "1: \n\t"
2074
            "movaps   (%2,%0), %%xmm0 \n\t"
2075
            "movaps 16(%2,%0), %%xmm1 \n\t"
2076
            "mulps    (%3,%0), %%xmm0 \n\t"
2077
            "mulps  16(%3,%0), %%xmm1 \n\t"
2078
            "addps    (%4,%0), %%xmm0 \n\t"
2079
            "addps  16(%4,%0), %%xmm1 \n\t"
2080
            "movss     %%xmm0,   (%1) \n\t"
2081
            "movss     %%xmm1, 32(%1) \n\t"
2082
            "movhlps   %%xmm0, %%xmm2 \n\t"
2083
            "movhlps   %%xmm1, %%xmm3 \n\t"
2084
            "movss     %%xmm2, 16(%1) \n\t"
2085
            "movss     %%xmm3, 48(%1) \n\t"
2086
            "shufps $0xb1, %%xmm0, %%xmm0 \n\t"
2087
            "shufps $0xb1, %%xmm1, %%xmm1 \n\t"
2088
            "movss     %%xmm0,  8(%1) \n\t"
2089
            "movss     %%xmm1, 40(%1) \n\t"
2090
            "movhlps   %%xmm0, %%xmm2 \n\t"
2091
            "movhlps   %%xmm1, %%xmm3 \n\t"
2092
            "movss     %%xmm2, 24(%1) \n\t"
2093
            "movss     %%xmm3, 56(%1) \n\t"
2094
            "sub  $64, %1 \n\t"
2095
            "sub  $32, %0 \n\t"
2096
            "jge  1b \n\t"
2097
            :"+r"(i), "+r"(dst)
2098
            :"r"(src0), "r"(src1), "r"(src2)
2099
            :"memory"
2100
        );
2101
    }
2102
    else if(step == 1 && src3 == 0){
2103
        asm volatile(
2104
            "1: \n\t"
2105
            "movaps   (%2,%0), %%xmm0 \n\t"
2106
            "movaps 16(%2,%0), %%xmm1 \n\t"
2107
            "mulps    (%3,%0), %%xmm0 \n\t"
2108
            "mulps  16(%3,%0), %%xmm1 \n\t"
2109
            "addps    (%4,%0), %%xmm0 \n\t"
2110
            "addps  16(%4,%0), %%xmm1 \n\t"
2111
            "movaps %%xmm0,   (%1,%0) \n\t"
2112
            "movaps %%xmm1, 16(%1,%0) \n\t"
2113
            "sub  $32, %0 \n\t"
2114
            "jge  1b \n\t"
2115
            :"+r"(i)
2116
            :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2117
            :"memory"
2118
        );
2119
    }
2120
    else
2121
        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
2122
}
2123

    
2124
static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
2125
                                      const float *win, float add_bias, int len){
2126
#ifdef HAVE_6REGS
2127
    if(add_bias == 0){
2128
        x86_reg i = -len*4;
2129
        x86_reg j = len*4-8;
2130
        asm volatile(
2131
            "1: \n"
2132
            "pswapd  (%5,%1), %%mm1 \n"
2133
            "movq    (%5,%0), %%mm0 \n"
2134
            "pswapd  (%4,%1), %%mm5 \n"
2135
            "movq    (%3,%0), %%mm4 \n"
2136
            "movq      %%mm0, %%mm2 \n"
2137
            "movq      %%mm1, %%mm3 \n"
2138
            "pfmul     %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
2139
            "pfmul     %%mm5, %%mm3 \n" // src1[    j]*win[len+j]
2140
            "pfmul     %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
2141
            "pfmul     %%mm5, %%mm0 \n" // src1[    j]*win[len+i]
2142
            "pfadd     %%mm3, %%mm2 \n"
2143
            "pfsub     %%mm0, %%mm1 \n"
2144
            "pswapd    %%mm2, %%mm2 \n"
2145
            "movq      %%mm1, (%2,%0) \n"
2146
            "movq      %%mm2, (%2,%1) \n"
2147
            "sub $8, %1 \n"
2148
            "add $8, %0 \n"
2149
            "jl 1b \n"
2150
            "femms \n"
2151
            :"+r"(i), "+r"(j)
2152
            :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2153
        );
2154
    }else
2155
#endif
2156
        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2157
}
2158

    
2159
static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
2160
                                   const float *win, float add_bias, int len){
2161
#ifdef HAVE_6REGS
2162
    if(add_bias == 0){
2163
        x86_reg i = -len*4;
2164
        x86_reg j = len*4-16;
2165
        asm volatile(
2166
            "1: \n"
2167
            "movaps       (%5,%1), %%xmm1 \n"
2168
            "movaps       (%5,%0), %%xmm0 \n"
2169
            "movaps       (%4,%1), %%xmm5 \n"
2170
            "movaps       (%3,%0), %%xmm4 \n"
2171
            "shufps $0x1b, %%xmm1, %%xmm1 \n"
2172
            "shufps $0x1b, %%xmm5, %%xmm5 \n"
2173
            "movaps        %%xmm0, %%xmm2 \n"
2174
            "movaps        %%xmm1, %%xmm3 \n"
2175
            "mulps         %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
2176
            "mulps         %%xmm5, %%xmm3 \n" // src1[    j]*win[len+j]
2177
            "mulps         %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
2178
            "mulps         %%xmm5, %%xmm0 \n" // src1[    j]*win[len+i]
2179
            "addps         %%xmm3, %%xmm2 \n"
2180
            "subps         %%xmm0, %%xmm1 \n"
2181
            "shufps $0x1b, %%xmm2, %%xmm2 \n"
2182
            "movaps        %%xmm1, (%2,%0) \n"
2183
            "movaps        %%xmm2, (%2,%1) \n"
2184
            "sub $16, %1 \n"
2185
            "add $16, %0 \n"
2186
            "jl 1b \n"
2187
            :"+r"(i), "+r"(j)
2188
            :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2189
        );
2190
    }else
2191
#endif
2192
        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2193
}
2194

    
2195
static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len)
2196
{
2197
    x86_reg i = -4*len;
2198
    asm volatile(
2199
        "movss  %3, %%xmm4 \n"
2200
        "shufps $0, %%xmm4, %%xmm4 \n"
2201
        "1: \n"
2202
        "cvtpi2ps   (%2,%0), %%xmm0 \n"
2203
        "cvtpi2ps  8(%2,%0), %%xmm1 \n"
2204
        "cvtpi2ps 16(%2,%0), %%xmm2 \n"
2205
        "cvtpi2ps 24(%2,%0), %%xmm3 \n"
2206
        "movlhps  %%xmm1,    %%xmm0 \n"
2207
        "movlhps  %%xmm3,    %%xmm2 \n"
2208
        "mulps    %%xmm4,    %%xmm0 \n"
2209
        "mulps    %%xmm4,    %%xmm2 \n"
2210
        "movaps   %%xmm0,   (%1,%0) \n"
2211
        "movaps   %%xmm2, 16(%1,%0) \n"
2212
        "add $32, %0 \n"
2213
        "jl 1b \n"
2214
        :"+r"(i)
2215
        :"r"(dst+len), "r"(src+len), "xm"(mul)
2216
    );
2217
}
2218

    
2219
static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len)
2220
{
2221
    x86_reg i = -4*len;
2222
    asm volatile(
2223
        "movss  %3, %%xmm4 \n"
2224
        "shufps $0, %%xmm4, %%xmm4 \n"
2225
        "1: \n"
2226
        "cvtdq2ps   (%2,%0), %%xmm0 \n"
2227
        "cvtdq2ps 16(%2,%0), %%xmm1 \n"
2228
        "mulps    %%xmm4,    %%xmm0 \n"
2229
        "mulps    %%xmm4,    %%xmm1 \n"
2230
        "movaps   %%xmm0,   (%1,%0) \n"
2231
        "movaps   %%xmm1, 16(%1,%0) \n"
2232
        "add $32, %0 \n"
2233
        "jl 1b \n"
2234
        :"+r"(i)
2235
        :"r"(dst+len), "r"(src+len), "xm"(mul)
2236
    );
2237
}
2238

    
2239
static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){
2240
    // not bit-exact: pf2id uses different rounding than C and SSE
2241
    asm volatile(
2242
        "add        %0          , %0        \n\t"
2243
        "lea         (%2,%0,2)  , %2        \n\t"
2244
        "add        %0          , %1        \n\t"
2245
        "neg        %0                      \n\t"
2246
        "1:                                 \n\t"
2247
        "pf2id       (%2,%0,2)  , %%mm0     \n\t"
2248
        "pf2id      8(%2,%0,2)  , %%mm1     \n\t"
2249
        "pf2id     16(%2,%0,2)  , %%mm2     \n\t"
2250
        "pf2id     24(%2,%0,2)  , %%mm3     \n\t"
2251
        "packssdw   %%mm1       , %%mm0     \n\t"
2252
        "packssdw   %%mm3       , %%mm2     \n\t"
2253
        "movq       %%mm0       ,  (%1,%0)  \n\t"
2254
        "movq       %%mm2       , 8(%1,%0)  \n\t"
2255
        "add        $16         , %0        \n\t"
2256
        " js 1b                             \n\t"
2257
        "femms                              \n\t"
2258
        :"+r"(len), "+r"(dst), "+r"(src)
2259
    );
2260
}
2261
static void float_to_int16_sse(int16_t *dst, const float *src, long len){
2262
    asm volatile(
2263
        "add        %0          , %0        \n\t"
2264
        "lea         (%2,%0,2)  , %2        \n\t"
2265
        "add        %0          , %1        \n\t"
2266
        "neg        %0                      \n\t"
2267
        "1:                                 \n\t"
2268
        "cvtps2pi    (%2,%0,2)  , %%mm0     \n\t"
2269
        "cvtps2pi   8(%2,%0,2)  , %%mm1     \n\t"
2270
        "cvtps2pi  16(%2,%0,2)  , %%mm2     \n\t"
2271
        "cvtps2pi  24(%2,%0,2)  , %%mm3     \n\t"
2272
        "packssdw   %%mm1       , %%mm0     \n\t"
2273
        "packssdw   %%mm3       , %%mm2     \n\t"
2274
        "movq       %%mm0       ,  (%1,%0)  \n\t"
2275
        "movq       %%mm2       , 8(%1,%0)  \n\t"
2276
        "add        $16         , %0        \n\t"
2277
        " js 1b                             \n\t"
2278
        "emms                               \n\t"
2279
        :"+r"(len), "+r"(dst), "+r"(src)
2280
    );
2281
}
2282

    
2283
static void float_to_int16_sse2(int16_t *dst, const float *src, long len){
2284
    asm volatile(
2285
        "add        %0          , %0        \n\t"
2286
        "lea         (%2,%0,2)  , %2        \n\t"
2287
        "add        %0          , %1        \n\t"
2288
        "neg        %0                      \n\t"
2289
        "1:                                 \n\t"
2290
        "cvtps2dq    (%2,%0,2)  , %%xmm0    \n\t"
2291
        "cvtps2dq  16(%2,%0,2)  , %%xmm1    \n\t"
2292
        "packssdw   %%xmm1      , %%xmm0    \n\t"
2293
        "movdqa     %%xmm0      ,  (%1,%0)  \n\t"
2294
        "add        $16         , %0        \n\t"
2295
        " js 1b                             \n\t"
2296
        :"+r"(len), "+r"(dst), "+r"(src)
2297
    );
2298
}
2299

    
2300
#define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \
2301
/* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\
2302
static av_noinline void float_to_int16_interleave2_##cpu(int16_t *dst, const float **src, long len, int channels){\
2303
    DECLARE_ALIGNED_16(int16_t, tmp[len]);\
2304
    int i,j,c;\
2305
    for(c=0; c<channels; c++){\
2306
        float_to_int16_##cpu(tmp, src[c], len);\
2307
        for(i=0, j=c; i<len; i++, j+=channels)\
2308
            dst[j] = tmp[i];\
2309
    }\
2310
}\
2311
\
2312
static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, long len, int channels){\
2313
    if(channels==1)\
2314
        float_to_int16_##cpu(dst, src[0], len);\
2315
    else if(channels>2)\
2316
        float_to_int16_interleave2_##cpu(dst, src, len, channels);\
2317
    else{\
2318
        const float *src0 = src[0];\
2319
        const float *src1 = src[1];\
2320
        asm volatile(\
2321
            "shl $2, %0 \n"\
2322
            "add %0, %1 \n"\
2323
            "add %0, %2 \n"\
2324
            "add %0, %3 \n"\
2325
            "neg %0 \n"\
2326
            body\
2327
            :"+r"(len), "+r"(dst), "+r"(src0), "+r"(src1)\
2328
        );\
2329
    }\
2330
}
2331

    
2332
FLOAT_TO_INT16_INTERLEAVE(3dnow,
2333
    "1:                         \n"
2334
    "pf2id     (%2,%0), %%mm0   \n"
2335
    "pf2id    8(%2,%0), %%mm1   \n"
2336
    "pf2id     (%3,%0), %%mm2   \n"
2337
    "pf2id    8(%3,%0), %%mm3   \n"
2338
    "packssdw    %%mm1, %%mm0   \n"
2339
    "packssdw    %%mm3, %%mm2   \n"
2340
    "movq        %%mm0, %%mm1   \n"
2341
    "punpcklwd   %%mm2, %%mm0   \n"
2342
    "punpckhwd   %%mm2, %%mm1   \n"
2343
    "movq        %%mm0,  (%1,%0)\n"
2344
    "movq        %%mm1, 8(%1,%0)\n"
2345
    "add $16, %0                \n"
2346
    "js 1b                      \n"
2347
    "femms                      \n"
2348
)
2349

    
2350
FLOAT_TO_INT16_INTERLEAVE(sse,
2351
    "1:                         \n"
2352
    "cvtps2pi  (%2,%0), %%mm0   \n"
2353
    "cvtps2pi 8(%2,%0), %%mm1   \n"
2354
    "cvtps2pi  (%3,%0), %%mm2   \n"
2355
    "cvtps2pi 8(%3,%0), %%mm3   \n"
2356
    "packssdw    %%mm1, %%mm0   \n"
2357
    "packssdw    %%mm3, %%mm2   \n"
2358
    "movq        %%mm0, %%mm1   \n"
2359
    "punpcklwd   %%mm2, %%mm0   \n"
2360
    "punpckhwd   %%mm2, %%mm1   \n"
2361
    "movq        %%mm0,  (%1,%0)\n"
2362
    "movq        %%mm1, 8(%1,%0)\n"
2363
    "add $16, %0                \n"
2364
    "js 1b                      \n"
2365
    "emms                       \n"
2366
)
2367

    
2368
FLOAT_TO_INT16_INTERLEAVE(sse2,
2369
    "1:                         \n"
2370
    "cvtps2dq  (%2,%0), %%xmm0  \n"
2371
    "cvtps2dq  (%3,%0), %%xmm1  \n"
2372
    "packssdw   %%xmm1, %%xmm0  \n"
2373
    "movhlps    %%xmm0, %%xmm1  \n"
2374
    "punpcklwd  %%xmm1, %%xmm0  \n"
2375
    "movdqa     %%xmm0, (%1,%0) \n"
2376
    "add $16, %0                \n"
2377
    "js 1b                      \n"
2378
)
2379

    
2380

    
2381
extern void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width);
2382
extern void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width);
2383
extern void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
2384
extern void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
2385
extern void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2386
                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2387
extern void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2388
                          int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2389

    
2390

    
2391
static void add_int16_sse2(int16_t * v1, int16_t * v2, int order)
2392
{
2393
    x86_reg o = -(order << 1);
2394
    v1 += order;
2395
    v2 += order;
2396
    asm volatile(
2397
        "1:                          \n\t"
2398
        "movdqu   (%1,%2),   %%xmm0  \n\t"
2399
        "movdqu 16(%1,%2),   %%xmm1  \n\t"
2400
        "paddw    (%0,%2),   %%xmm0  \n\t"
2401
        "paddw  16(%0,%2),   %%xmm1  \n\t"
2402
        "movdqa   %%xmm0,    (%0,%2) \n\t"
2403
        "movdqa   %%xmm1,  16(%0,%2) \n\t"
2404
        "add      $32,       %2      \n\t"
2405
        "js       1b                 \n\t"
2406
        : "+r"(v1), "+r"(v2), "+r"(o)
2407
    );
2408
}
2409

    
2410
static void sub_int16_sse2(int16_t * v1, int16_t * v2, int order)
2411
{
2412
    x86_reg o = -(order << 1);
2413
    v1 += order;
2414
    v2 += order;
2415
    asm volatile(
2416
        "1:                           \n\t"
2417
        "movdqa    (%0,%2),   %%xmm0  \n\t"
2418
        "movdqa  16(%0,%2),   %%xmm2  \n\t"
2419
        "movdqu    (%1,%2),   %%xmm1  \n\t"
2420
        "movdqu  16(%1,%2),   %%xmm3  \n\t"
2421
        "psubw     %%xmm1,    %%xmm0  \n\t"
2422
        "psubw     %%xmm3,    %%xmm2  \n\t"
2423
        "movdqa    %%xmm0,    (%0,%2) \n\t"
2424
        "movdqa    %%xmm2,  16(%0,%2) \n\t"
2425
        "add       $32,       %2      \n\t"
2426
        "js        1b                 \n\t"
2427
        : "+r"(v1), "+r"(v2), "+r"(o)
2428
    );
2429
}
2430

    
2431
static int32_t scalarproduct_int16_sse2(int16_t * v1, int16_t * v2, int order, int shift)
2432
{
2433
    int res = 0;
2434
    DECLARE_ALIGNED_16(int64_t, sh);
2435
    x86_reg o = -(order << 1);
2436

    
2437
    v1 += order;
2438
    v2 += order;
2439
    sh = shift;
2440
    asm volatile(
2441
        "pxor      %%xmm7,  %%xmm7        \n\t"
2442
        "1:                               \n\t"
2443
        "movdqu    (%0,%3), %%xmm0        \n\t"
2444
        "movdqu  16(%0,%3), %%xmm1        \n\t"
2445
        "pmaddwd   (%1,%3), %%xmm0        \n\t"
2446
        "pmaddwd 16(%1,%3), %%xmm1        \n\t"
2447
        "paddd     %%xmm0,  %%xmm7        \n\t"
2448
        "paddd     %%xmm1,  %%xmm7        \n\t"
2449
        "add       $32,     %3            \n\t"
2450
        "js        1b                     \n\t"
2451
        "movhlps   %%xmm7,  %%xmm2        \n\t"
2452
        "paddd     %%xmm2,  %%xmm7        \n\t"
2453
        "psrad     %4,      %%xmm7        \n\t"
2454
        "pshuflw   $0x4E,   %%xmm7,%%xmm2 \n\t"
2455
        "paddd     %%xmm2,  %%xmm7        \n\t"
2456
        "movd      %%xmm7,  %2            \n\t"
2457
        : "+r"(v1), "+r"(v2), "=r"(res), "+r"(o)
2458
        : "m"(sh)
2459
    );
2460
    return res;
2461
}
2462

    
2463
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2464
{
2465
    mm_flags = mm_support();
2466

    
2467
    if (avctx->dsp_mask) {
2468
        if (avctx->dsp_mask & FF_MM_FORCE)
2469
            mm_flags |= (avctx->dsp_mask & 0xffff);
2470
        else
2471
            mm_flags &= ~(avctx->dsp_mask & 0xffff);
2472
    }
2473

    
2474
#if 0
2475
    av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2476
    if (mm_flags & MM_MMX)
2477
        av_log(avctx, AV_LOG_INFO, " mmx");
2478
    if (mm_flags & MM_MMXEXT)
2479
        av_log(avctx, AV_LOG_INFO, " mmxext");
2480
    if (mm_flags & MM_3DNOW)
2481
        av_log(avctx, AV_LOG_INFO, " 3dnow");
2482
    if (mm_flags & MM_SSE)
2483
        av_log(avctx, AV_LOG_INFO, " sse");
2484
    if (mm_flags & MM_SSE2)
2485
        av_log(avctx, AV_LOG_INFO, " sse2");
2486
    av_log(avctx, AV_LOG_INFO, "\n");
2487
#endif
2488

    
2489
    if (mm_flags & MM_MMX) {
2490
        const int idct_algo= avctx->idct_algo;
2491

    
2492
        if(avctx->lowres==0){
2493
            if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2494
                c->idct_put= ff_simple_idct_put_mmx;
2495
                c->idct_add= ff_simple_idct_add_mmx;
2496
                c->idct    = ff_simple_idct_mmx;
2497
                c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2498
#ifdef CONFIG_GPL
2499
            }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2500
                if(mm_flags & MM_MMXEXT){
2501
                    c->idct_put= ff_libmpeg2mmx2_idct_put;
2502
                    c->idct_add= ff_libmpeg2mmx2_idct_add;
2503
                    c->idct    = ff_mmxext_idct;
2504
                }else{
2505
                    c->idct_put= ff_libmpeg2mmx_idct_put;
2506
                    c->idct_add= ff_libmpeg2mmx_idct_add;
2507
                    c->idct    = ff_mmx_idct;
2508
                }
2509
                c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2510
#endif
2511
            }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER) &&
2512
                     idct_algo==FF_IDCT_VP3 &&
2513
                     avctx->codec->id!=CODEC_ID_THEORA &&
2514
                     !(avctx->flags & CODEC_FLAG_BITEXACT)){
2515
                if(mm_flags & MM_SSE2){
2516
                    c->idct_put= ff_vp3_idct_put_sse2;
2517
                    c->idct_add= ff_vp3_idct_add_sse2;
2518
                    c->idct    = ff_vp3_idct_sse2;
2519
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2520
                }else{
2521
                    ff_vp3_dsp_init_mmx();
2522
                    c->idct_put= ff_vp3_idct_put_mmx;
2523
                    c->idct_add= ff_vp3_idct_add_mmx;
2524
                    c->idct    = ff_vp3_idct_mmx;
2525
                    c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2526
                }
2527
            }else if(idct_algo==FF_IDCT_CAVS){
2528
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2529
            }else if(idct_algo==FF_IDCT_XVIDMMX){
2530
                if(mm_flags & MM_SSE2){
2531
                    c->idct_put= ff_idct_xvid_sse2_put;
2532
                    c->idct_add= ff_idct_xvid_sse2_add;
2533
                    c->idct    = ff_idct_xvid_sse2;
2534
                    c->idct_permutation_type= FF_SSE2_IDCT_PERM;
2535
                }else if(mm_flags & MM_MMXEXT){
2536
                    c->idct_put= ff_idct_xvid_mmx2_put;
2537
                    c->idct_add= ff_idct_xvid_mmx2_add;
2538
                    c->idct    = ff_idct_xvid_mmx2;
2539
                }else{
2540
                    c->idct_put= ff_idct_xvid_mmx_put;
2541
                    c->idct_add= ff_idct_xvid_mmx_add;
2542
                    c->idct    = ff_idct_xvid_mmx;
2543
                }
2544
            }
2545
        }
2546

    
2547
        c->put_pixels_clamped = put_pixels_clamped_mmx;
2548
        c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;
2549
        c->add_pixels_clamped = add_pixels_clamped_mmx;
2550
        c->clear_blocks = clear_blocks_mmx;
2551

    
2552
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2553
        c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
2554
        c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
2555
        c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
2556
        c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU
2557

    
2558
        SET_HPEL_FUNCS(put, 0, 16, mmx);
2559
        SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2560
        SET_HPEL_FUNCS(avg, 0, 16, mmx);
2561
        SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2562
        SET_HPEL_FUNCS(put, 1, 8, mmx);
2563
        SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
2564
        SET_HPEL_FUNCS(avg, 1, 8, mmx);
2565
        SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
2566

    
2567
        c->gmc= gmc_mmx;
2568

    
2569
        c->add_bytes= add_bytes_mmx;
2570
        c->add_bytes_l2= add_bytes_l2_mmx;
2571

    
2572
        c->draw_edges = draw_edges_mmx;
2573

    
2574
        if (ENABLE_ANY_H263) {
2575
            c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2576
            c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2577
        }
2578
        c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_rnd;
2579
        c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;
2580
        c->put_no_rnd_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_nornd;
2581

    
2582
        c->h264_idct_dc_add=
2583
        c->h264_idct_add= ff_h264_idct_add_mmx;
2584
        c->h264_idct8_dc_add=
2585
        c->h264_idct8_add= ff_h264_idct8_add_mmx;
2586
        if (mm_flags & MM_SSE2)
2587
            c->h264_idct8_add= ff_h264_idct8_add_sse2;
2588

    
2589
        if (mm_flags & MM_MMXEXT) {
2590
            c->prefetch = prefetch_mmx2;
2591

    
2592
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2593
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2594

    
2595
            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2596
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2597
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2598

    
2599
            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2600
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2601

    
2602
            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2603
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2604
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2605

    
2606
            c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;
2607
            c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;
2608

    
2609
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2610
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2611
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2612
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2613
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2614
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2615
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2616
            }
2617

    
2618
#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2619
            c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \
2620
            c->PFX ## _pixels_tab[IDX][ 1] = PFX ## SIZE ## _mc10_ ## CPU; \
2621
            c->PFX ## _pixels_tab[IDX][ 2] = PFX ## SIZE ## _mc20_ ## CPU; \
2622
            c->PFX ## _pixels_tab[IDX][ 3] = PFX ## SIZE ## _mc30_ ## CPU; \
2623
            c->PFX ## _pixels_tab[IDX][ 4] = PFX ## SIZE ## _mc01_ ## CPU; \
2624
            c->PFX ## _pixels_tab[IDX][ 5] = PFX ## SIZE ## _mc11_ ## CPU; \
2625
            c->PFX ## _pixels_tab[IDX][ 6] = PFX ## SIZE ## _mc21_ ## CPU; \
2626
            c->PFX ## _pixels_tab[IDX][ 7] = PFX ## SIZE ## _mc31_ ## CPU; \
2627
            c->PFX ## _pixels_tab[IDX][ 8] = PFX ## SIZE ## _mc02_ ## CPU; \
2628
            c->PFX ## _pixels_tab[IDX][ 9] = PFX ## SIZE ## _mc12_ ## CPU; \
2629
            c->PFX ## _pixels_tab[IDX][10] = PFX ## SIZE ## _mc22_ ## CPU; \
2630
            c->PFX ## _pixels_tab[IDX][11] = PFX ## SIZE ## _mc32_ ## CPU; \
2631
            c->PFX ## _pixels_tab[IDX][12] = PFX ## SIZE ## _mc03_ ## CPU; \
2632
            c->PFX ## _pixels_tab[IDX][13] = PFX ## SIZE ## _mc13_ ## CPU; \
2633
            c->PFX ## _pixels_tab[IDX][14] = PFX ## SIZE ## _mc23_ ## CPU; \
2634
            c->PFX ## _pixels_tab[IDX][15] = PFX ## SIZE ## _mc33_ ## CPU
2635

    
2636
            SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2);
2637
            SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2);
2638
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2);
2639
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2);
2640
            SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2);
2641
            SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2);
2642

    
2643
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2);
2644
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2);
2645
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2);
2646
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2);
2647
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2);
2648
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2);
2649

    
2650
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2);
2651
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2);
2652
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2);
2653
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2);
2654

    
2655
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2_rnd;
2656
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;
2657
            c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;
2658
            c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;
2659
            c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
2660
            c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
2661
            c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
2662
            c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;
2663
            c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;
2664
            c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;
2665
            c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
2666

    
2667
            c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
2668
            c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
2669
            c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
2670
            c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
2671
            c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
2672
            c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
2673
            c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
2674
            c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
2675

    
2676
            c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
2677
            c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
2678
            c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
2679
            c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
2680
            c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
2681
            c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
2682
            c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
2683
            c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
2684

    
2685
            if (ENABLE_CAVS_DECODER)
2686
                ff_cavsdsp_init_mmx2(c, avctx);
2687

    
2688
            if (ENABLE_VC1_DECODER || ENABLE_WMV3_DECODER)
2689
                ff_vc1dsp_init_mmx(c, avctx);
2690

    
2691
            c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2;
2692
        } else if (mm_flags & MM_3DNOW) {
2693
            c->prefetch = prefetch_3dnow;
2694

    
2695
            c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
2696
            c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
2697

    
2698
            c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
2699
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
2700
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
2701

    
2702
            c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
2703
            c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
2704

    
2705
            c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
2706
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
2707
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
2708

    
2709
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2710
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
2711
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
2712
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
2713
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
2714
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
2715
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
2716
            }
2717

    
2718
            SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow);
2719
            SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow);
2720
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow);
2721
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow);
2722
            SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow);
2723
            SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow);
2724

    
2725
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow);
2726
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow);
2727
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow);
2728
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow);
2729
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow);
2730
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow);
2731

    
2732
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow);
2733
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow);
2734
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow);
2735
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow);
2736

    
2737
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow_rnd;
2738
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;
2739

    
2740
            if (ENABLE_CAVS_DECODER)
2741
                ff_cavsdsp_init_3dnow(c, avctx);
2742
        }
2743

    
2744

    
2745
#define H264_QPEL_FUNCS(x, y, CPU)\
2746
            c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\
2747
            c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\
2748
            c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\
2749
            c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;
2750
        if((mm_flags & MM_SSE2) && !(mm_flags & MM_3DNOW)){
2751
            // these functions are slower than mmx on AMD, but faster on Intel
2752
/* FIXME works in most codecs, but crashes svq1 due to unaligned chroma
2753
            c->put_pixels_tab[0][0] = put_pixels16_sse2;
2754
            c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
2755
*/
2756
            H264_QPEL_FUNCS(0, 0, sse2);
2757
        }
2758
        if(mm_flags & MM_SSE2){
2759
            H264_QPEL_FUNCS(0, 1, sse2);
2760
            H264_QPEL_FUNCS(0, 2, sse2);
2761
            H264_QPEL_FUNCS(0, 3, sse2);
2762
            H264_QPEL_FUNCS(1, 1, sse2);
2763
            H264_QPEL_FUNCS(1, 2, sse2);
2764
            H264_QPEL_FUNCS(1, 3, sse2);
2765
            H264_QPEL_FUNCS(2, 1, sse2);
2766
            H264_QPEL_FUNCS(2, 2, sse2);
2767
            H264_QPEL_FUNCS(2, 3, sse2);
2768
            H264_QPEL_FUNCS(3, 1, sse2);
2769
            H264_QPEL_FUNCS(3, 2, sse2);
2770
            H264_QPEL_FUNCS(3, 3, sse2);
2771
        }
2772
#ifdef HAVE_SSSE3
2773
        if(mm_flags & MM_SSSE3){
2774
            H264_QPEL_FUNCS(1, 0, ssse3);
2775
            H264_QPEL_FUNCS(1, 1, ssse3);
2776
            H264_QPEL_FUNCS(1, 2, ssse3);
2777
            H264_QPEL_FUNCS(1, 3, ssse3);
2778
            H264_QPEL_FUNCS(2, 0, ssse3);
2779
            H264_QPEL_FUNCS(2, 1, ssse3);
2780
            H264_QPEL_FUNCS(2, 2, ssse3);
2781
            H264_QPEL_FUNCS(2, 3, ssse3);
2782
            H264_QPEL_FUNCS(3, 0, ssse3);
2783
            H264_QPEL_FUNCS(3, 1, ssse3);
2784
            H264_QPEL_FUNCS(3, 2, ssse3);
2785
            H264_QPEL_FUNCS(3, 3, ssse3);
2786
            c->put_no_rnd_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_nornd;
2787
            c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_rnd;
2788
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_ssse3_rnd;
2789
            c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_ssse3;
2790
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3;
2791
            c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
2792
        }
2793
#endif
2794

    
2795
#ifdef CONFIG_SNOW_DECODER
2796
        if(mm_flags & MM_SSE2 & 0){
2797
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
2798
#ifdef HAVE_7REGS
2799
            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
2800
#endif
2801
            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
2802
        }
2803
        else{
2804
            if(mm_flags & MM_MMXEXT){
2805
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
2806
#ifdef HAVE_7REGS
2807
            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
2808
#endif
2809
            }
2810
            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
2811
        }
2812
#endif
2813

    
2814
        if(mm_flags & MM_3DNOW){
2815
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
2816
            c->vector_fmul = vector_fmul_3dnow;
2817
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2818
                c->float_to_int16 = float_to_int16_3dnow;
2819
                c->float_to_int16_interleave = float_to_int16_interleave_3dnow;
2820
            }
2821
        }
2822
        if(mm_flags & MM_3DNOWEXT){
2823
            c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
2824
            c->vector_fmul_window = vector_fmul_window_3dnow2;
2825
        }
2826
        if(mm_flags & MM_SSE){
2827
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
2828
            c->ac3_downmix = ac3_downmix_sse;
2829
            c->vector_fmul = vector_fmul_sse;
2830
            c->vector_fmul_reverse = vector_fmul_reverse_sse;
2831
            c->vector_fmul_add_add = vector_fmul_add_add_sse;
2832
            c->vector_fmul_window = vector_fmul_window_sse;
2833
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse;
2834
            c->float_to_int16 = float_to_int16_sse;
2835
            c->float_to_int16_interleave = float_to_int16_interleave_sse;
2836
        }
2837
        if(mm_flags & MM_3DNOW)
2838
            c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse
2839
        if(mm_flags & MM_SSE2){
2840
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2;
2841
            c->float_to_int16 = float_to_int16_sse2;
2842
            c->float_to_int16_interleave = float_to_int16_interleave_sse2;
2843
            c->add_int16 = add_int16_sse2;
2844
            c->sub_int16 = sub_int16_sse2;
2845
            c->scalarproduct_int16 = scalarproduct_int16_sse2;
2846
        }
2847
    }
2848

    
2849
    if (ENABLE_ENCODERS)
2850
        dsputilenc_init_mmx(c, avctx);
2851

    
2852
#if 0
2853
    // for speed testing
2854
    get_pixels = just_return;
2855
    put_pixels_clamped = just_return;
2856
    add_pixels_clamped = just_return;
2857

2858
    pix_abs16x16 = just_return;
2859
    pix_abs16x16_x2 = just_return;
2860
    pix_abs16x16_y2 = just_return;
2861
    pix_abs16x16_xy2 = just_return;
2862

2863
    put_pixels_tab[0] = just_return;
2864
    put_pixels_tab[1] = just_return;
2865
    put_pixels_tab[2] = just_return;
2866
    put_pixels_tab[3] = just_return;
2867

2868
    put_no_rnd_pixels_tab[0] = just_return;
2869
    put_no_rnd_pixels_tab[1] = just_return;
2870
    put_no_rnd_pixels_tab[2] = just_return;
2871
    put_no_rnd_pixels_tab[3] = just_return;
2872

2873
    avg_pixels_tab[0] = just_return;
2874
    avg_pixels_tab[1] = just_return;
2875
    avg_pixels_tab[2] = just_return;
2876
    avg_pixels_tab[3] = just_return;
2877

2878
    avg_no_rnd_pixels_tab[0] = just_return;
2879
    avg_no_rnd_pixels_tab[1] = just_return;
2880
    avg_no_rnd_pixels_tab[2] = just_return;
2881
    avg_no_rnd_pixels_tab[3] = just_return;
2882

2883
    //av_fdct = just_return;
2884
    //ff_idct = just_return;
2885
#endif
2886
}