Statistics
| Branch: | Revision:

ffmpeg / libavcodec / x86 / dsputil_mmx.c @ 84dc2d8a

History | View | Annotate | Download (124 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/mpegvideo.h"
28
#include "libavcodec/simple_idct.h"
29
#include "dsputil_mmx.h"
30
#include "vp3dsp_mmx.h"
31
#include "vp3dsp_sse2.h"
32
#include "vp6dsp_mmx.h"
33
#include "vp6dsp_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_reg,  ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
51
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_8  ) = {0x0008000800080008ULL, 0x0008000800080008ULL};
52
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
53
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
54
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
55
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
56
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
57
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
58
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_64 ) = {0x0040004000400040ULL, 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_1F ) = 0x1F1F1F1F1F1F1F1FULL;
67
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
68
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
69
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL;
70
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
71

    
72
DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
73
DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
74

    
75
#define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::)
76
#define MOVQ_ZERO(regd)  __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
77

    
78
#define MOVQ_BFE(regd) \
79
    __asm__ volatile ( \
80
    "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
81
    "paddb %%" #regd ", %%" #regd " \n\t" ::)
82

    
83
#ifndef PIC
84
#define MOVQ_BONE(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
85
#define MOVQ_WTWO(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
86
#else
87
// for shared library it's better to use this way for accessing constants
88
// pcmpeqd -> -1
89
#define MOVQ_BONE(regd) \
90
    __asm__ volatile ( \
91
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
92
    "psrlw $15, %%" #regd " \n\t" \
93
    "packuswb %%" #regd ", %%" #regd " \n\t" ::)
94

    
95
#define MOVQ_WTWO(regd) \
96
    __asm__ volatile ( \
97
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
98
    "psrlw $15, %%" #regd " \n\t" \
99
    "psllw $1, %%" #regd " \n\t"::)
100

    
101
#endif
102

    
103
// using regr as temporary and for the output result
104
// first argument is unmodifed and second is trashed
105
// regfe is supposed to contain 0xfefefefefefefefe
106
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
107
    "movq " #rega ", " #regr "  \n\t"\
108
    "pand " #regb ", " #regr "  \n\t"\
109
    "pxor " #rega ", " #regb "  \n\t"\
110
    "pand " #regfe "," #regb "  \n\t"\
111
    "psrlq $1, " #regb "        \n\t"\
112
    "paddb " #regb ", " #regr " \n\t"
113

    
114
#define PAVGB_MMX(rega, regb, regr, regfe) \
115
    "movq " #rega ", " #regr "  \n\t"\
116
    "por  " #regb ", " #regr "  \n\t"\
117
    "pxor " #rega ", " #regb "  \n\t"\
118
    "pand " #regfe "," #regb "  \n\t"\
119
    "psrlq $1, " #regb "        \n\t"\
120
    "psubb " #regb ", " #regr " \n\t"
121

    
122
// mm6 is supposed to contain 0xfefefefefefefefe
123
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \
124
    "movq " #rega ", " #regr "  \n\t"\
125
    "movq " #regc ", " #regp "  \n\t"\
126
    "pand " #regb ", " #regr "  \n\t"\
127
    "pand " #regd ", " #regp "  \n\t"\
128
    "pxor " #rega ", " #regb "  \n\t"\
129
    "pxor " #regc ", " #regd "  \n\t"\
130
    "pand %%mm6, " #regb "      \n\t"\
131
    "pand %%mm6, " #regd "      \n\t"\
132
    "psrlq $1, " #regb "        \n\t"\
133
    "psrlq $1, " #regd "        \n\t"\
134
    "paddb " #regb ", " #regr " \n\t"\
135
    "paddb " #regd ", " #regp " \n\t"
136

    
137
#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
138
    "movq " #rega ", " #regr "  \n\t"\
139
    "movq " #regc ", " #regp "  \n\t"\
140
    "por  " #regb ", " #regr "  \n\t"\
141
    "por  " #regd ", " #regp "  \n\t"\
142
    "pxor " #rega ", " #regb "  \n\t"\
143
    "pxor " #regc ", " #regd "  \n\t"\
144
    "pand %%mm6, " #regb "      \n\t"\
145
    "pand %%mm6, " #regd "      \n\t"\
146
    "psrlq $1, " #regd "        \n\t"\
147
    "psrlq $1, " #regb "        \n\t"\
148
    "psubb " #regb ", " #regr " \n\t"\
149
    "psubb " #regd ", " #regp " \n\t"
150

    
151
/***********************************/
152
/* MMX no rounding */
153
#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
154
#define SET_RND  MOVQ_WONE
155
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
156
#define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
157
#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
158

    
159
#include "dsputil_mmx_rnd_template.c"
160

    
161
#undef DEF
162
#undef SET_RND
163
#undef PAVGBP
164
#undef PAVGB
165
/***********************************/
166
/* MMX rounding */
167

    
168
#define DEF(x, y) x ## _ ## y ##_mmx
169
#define SET_RND  MOVQ_WTWO
170
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
171
#define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)
172

    
173
#include "dsputil_mmx_rnd_template.c"
174

    
175
#undef DEF
176
#undef SET_RND
177
#undef PAVGBP
178
#undef PAVGB
179
#undef OP_AVG
180

    
181
/***********************************/
182
/* 3Dnow specific */
183

    
184
#define DEF(x) x ## _3dnow
185
#define PAVGB "pavgusb"
186
#define OP_AVG PAVGB
187

    
188
#include "dsputil_mmx_avg_template.c"
189

    
190
#undef DEF
191
#undef PAVGB
192
#undef OP_AVG
193

    
194
/***********************************/
195
/* MMX2 specific */
196

    
197
#define DEF(x) x ## _mmx2
198

    
199
/* Introduced only in MMX2 set */
200
#define PAVGB "pavgb"
201
#define OP_AVG PAVGB
202

    
203
#include "dsputil_mmx_avg_template.c"
204

    
205
#undef DEF
206
#undef PAVGB
207
#undef OP_AVG
208

    
209
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
210
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
211
#define put_pixels16_mmx2 put_pixels16_mmx
212
#define put_pixels8_mmx2 put_pixels8_mmx
213
#define put_pixels4_mmx2 put_pixels4_mmx
214
#define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
215
#define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
216
#define put_pixels16_3dnow put_pixels16_mmx
217
#define put_pixels8_3dnow put_pixels8_mmx
218
#define put_pixels4_3dnow put_pixels4_mmx
219
#define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
220
#define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
221

    
222
/***********************************/
223
/* standard MMX */
224

    
225
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
226
{
227
    const DCTELEM *p;
228
    uint8_t *pix;
229

    
230
    /* read the pixels */
231
    p = block;
232
    pix = pixels;
233
    /* unrolled loop */
234
        __asm__ volatile(
235
                "movq   %3, %%mm0               \n\t"
236
                "movq   8%3, %%mm1              \n\t"
237
                "movq   16%3, %%mm2             \n\t"
238
                "movq   24%3, %%mm3             \n\t"
239
                "movq   32%3, %%mm4             \n\t"
240
                "movq   40%3, %%mm5             \n\t"
241
                "movq   48%3, %%mm6             \n\t"
242
                "movq   56%3, %%mm7             \n\t"
243
                "packuswb %%mm1, %%mm0          \n\t"
244
                "packuswb %%mm3, %%mm2          \n\t"
245
                "packuswb %%mm5, %%mm4          \n\t"
246
                "packuswb %%mm7, %%mm6          \n\t"
247
                "movq   %%mm0, (%0)             \n\t"
248
                "movq   %%mm2, (%0, %1)         \n\t"
249
                "movq   %%mm4, (%0, %1, 2)      \n\t"
250
                "movq   %%mm6, (%0, %2)         \n\t"
251
                ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
252
                :"memory");
253
        pix += line_size*4;
254
        p += 32;
255

    
256
    // if here would be an exact copy of the code above
257
    // compiler would generate some very strange code
258
    // thus using "r"
259
    __asm__ volatile(
260
            "movq       (%3), %%mm0             \n\t"
261
            "movq       8(%3), %%mm1            \n\t"
262
            "movq       16(%3), %%mm2           \n\t"
263
            "movq       24(%3), %%mm3           \n\t"
264
            "movq       32(%3), %%mm4           \n\t"
265
            "movq       40(%3), %%mm5           \n\t"
266
            "movq       48(%3), %%mm6           \n\t"
267
            "movq       56(%3), %%mm7           \n\t"
268
            "packuswb %%mm1, %%mm0              \n\t"
269
            "packuswb %%mm3, %%mm2              \n\t"
270
            "packuswb %%mm5, %%mm4              \n\t"
271
            "packuswb %%mm7, %%mm6              \n\t"
272
            "movq       %%mm0, (%0)             \n\t"
273
            "movq       %%mm2, (%0, %1)         \n\t"
274
            "movq       %%mm4, (%0, %1, 2)      \n\t"
275
            "movq       %%mm6, (%0, %2)         \n\t"
276
            ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
277
            :"memory");
278
}
279

    
280
DECLARE_ASM_CONST(8, uint8_t, ff_vector128)[8] =
281
  { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
282

    
283
#define put_signed_pixels_clamped_mmx_half(off) \
284
            "movq    "#off"(%2), %%mm1          \n\t"\
285
            "movq 16+"#off"(%2), %%mm2          \n\t"\
286
            "movq 32+"#off"(%2), %%mm3          \n\t"\
287
            "movq 48+"#off"(%2), %%mm4          \n\t"\
288
            "packsswb  8+"#off"(%2), %%mm1      \n\t"\
289
            "packsswb 24+"#off"(%2), %%mm2      \n\t"\
290
            "packsswb 40+"#off"(%2), %%mm3      \n\t"\
291
            "packsswb 56+"#off"(%2), %%mm4      \n\t"\
292
            "paddb %%mm0, %%mm1                 \n\t"\
293
            "paddb %%mm0, %%mm2                 \n\t"\
294
            "paddb %%mm0, %%mm3                 \n\t"\
295
            "paddb %%mm0, %%mm4                 \n\t"\
296
            "movq %%mm1, (%0)                   \n\t"\
297
            "movq %%mm2, (%0, %3)               \n\t"\
298
            "movq %%mm3, (%0, %3, 2)            \n\t"\
299
            "movq %%mm4, (%0, %1)               \n\t"
300

    
301
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
302
{
303
    x86_reg line_skip = line_size;
304
    x86_reg line_skip3;
305

    
306
    __asm__ volatile (
307
            "movq "MANGLE(ff_vector128)", %%mm0 \n\t"
308
            "lea (%3, %3, 2), %1                \n\t"
309
            put_signed_pixels_clamped_mmx_half(0)
310
            "lea (%0, %3, 4), %0                \n\t"
311
            put_signed_pixels_clamped_mmx_half(64)
312
            :"+&r" (pixels), "=&r" (line_skip3)
313
            :"r" (block), "r"(line_skip)
314
            :"memory");
315
}
316

    
317
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
318
{
319
    const DCTELEM *p;
320
    uint8_t *pix;
321
    int i;
322

    
323
    /* read the pixels */
324
    p = block;
325
    pix = pixels;
326
    MOVQ_ZERO(mm7);
327
    i = 4;
328
    do {
329
        __asm__ volatile(
330
                "movq   (%2), %%mm0     \n\t"
331
                "movq   8(%2), %%mm1    \n\t"
332
                "movq   16(%2), %%mm2   \n\t"
333
                "movq   24(%2), %%mm3   \n\t"
334
                "movq   %0, %%mm4       \n\t"
335
                "movq   %1, %%mm6       \n\t"
336
                "movq   %%mm4, %%mm5    \n\t"
337
                "punpcklbw %%mm7, %%mm4 \n\t"
338
                "punpckhbw %%mm7, %%mm5 \n\t"
339
                "paddsw %%mm4, %%mm0    \n\t"
340
                "paddsw %%mm5, %%mm1    \n\t"
341
                "movq   %%mm6, %%mm5    \n\t"
342
                "punpcklbw %%mm7, %%mm6 \n\t"
343
                "punpckhbw %%mm7, %%mm5 \n\t"
344
                "paddsw %%mm6, %%mm2    \n\t"
345
                "paddsw %%mm5, %%mm3    \n\t"
346
                "packuswb %%mm1, %%mm0  \n\t"
347
                "packuswb %%mm3, %%mm2  \n\t"
348
                "movq   %%mm0, %0       \n\t"
349
                "movq   %%mm2, %1       \n\t"
350
                :"+m"(*pix), "+m"(*(pix+line_size))
351
                :"r"(p)
352
                :"memory");
353
        pix += line_size*2;
354
        p += 16;
355
    } while (--i);
356
}
357

    
358
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
359
{
360
    __asm__ volatile(
361
         "lea (%3, %3), %%"REG_a"       \n\t"
362
         ASMALIGN(3)
363
         "1:                            \n\t"
364
         "movd (%1), %%mm0              \n\t"
365
         "movd (%1, %3), %%mm1          \n\t"
366
         "movd %%mm0, (%2)              \n\t"
367
         "movd %%mm1, (%2, %3)          \n\t"
368
         "add %%"REG_a", %1             \n\t"
369
         "add %%"REG_a", %2             \n\t"
370
         "movd (%1), %%mm0              \n\t"
371
         "movd (%1, %3), %%mm1          \n\t"
372
         "movd %%mm0, (%2)              \n\t"
373
         "movd %%mm1, (%2, %3)          \n\t"
374
         "add %%"REG_a", %1             \n\t"
375
         "add %%"REG_a", %2             \n\t"
376
         "subl $4, %0                   \n\t"
377
         "jnz 1b                        \n\t"
378
         : "+g"(h), "+r" (pixels),  "+r" (block)
379
         : "r"((x86_reg)line_size)
380
         : "%"REG_a, "memory"
381
        );
382
}
383

    
384
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
385
{
386
    __asm__ volatile(
387
         "lea (%3, %3), %%"REG_a"       \n\t"
388
         ASMALIGN(3)
389
         "1:                            \n\t"
390
         "movq (%1), %%mm0              \n\t"
391
         "movq (%1, %3), %%mm1          \n\t"
392
         "movq %%mm0, (%2)              \n\t"
393
         "movq %%mm1, (%2, %3)          \n\t"
394
         "add %%"REG_a", %1             \n\t"
395
         "add %%"REG_a", %2             \n\t"
396
         "movq (%1), %%mm0              \n\t"
397
         "movq (%1, %3), %%mm1          \n\t"
398
         "movq %%mm0, (%2)              \n\t"
399
         "movq %%mm1, (%2, %3)          \n\t"
400
         "add %%"REG_a", %1             \n\t"
401
         "add %%"REG_a", %2             \n\t"
402
         "subl $4, %0                   \n\t"
403
         "jnz 1b                        \n\t"
404
         : "+g"(h), "+r" (pixels),  "+r" (block)
405
         : "r"((x86_reg)line_size)
406
         : "%"REG_a, "memory"
407
        );
408
}
409

    
410
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
411
{
412
    __asm__ volatile(
413
         "lea (%3, %3), %%"REG_a"       \n\t"
414
         ASMALIGN(3)
415
         "1:                            \n\t"
416
         "movq (%1), %%mm0              \n\t"
417
         "movq 8(%1), %%mm4             \n\t"
418
         "movq (%1, %3), %%mm1          \n\t"
419
         "movq 8(%1, %3), %%mm5         \n\t"
420
         "movq %%mm0, (%2)              \n\t"
421
         "movq %%mm4, 8(%2)             \n\t"
422
         "movq %%mm1, (%2, %3)          \n\t"
423
         "movq %%mm5, 8(%2, %3)         \n\t"
424
         "add %%"REG_a", %1             \n\t"
425
         "add %%"REG_a", %2             \n\t"
426
         "movq (%1), %%mm0              \n\t"
427
         "movq 8(%1), %%mm4             \n\t"
428
         "movq (%1, %3), %%mm1          \n\t"
429
         "movq 8(%1, %3), %%mm5         \n\t"
430
         "movq %%mm0, (%2)              \n\t"
431
         "movq %%mm4, 8(%2)             \n\t"
432
         "movq %%mm1, (%2, %3)          \n\t"
433
         "movq %%mm5, 8(%2, %3)         \n\t"
434
         "add %%"REG_a", %1             \n\t"
435
         "add %%"REG_a", %2             \n\t"
436
         "subl $4, %0                   \n\t"
437
         "jnz 1b                        \n\t"
438
         : "+g"(h), "+r" (pixels),  "+r" (block)
439
         : "r"((x86_reg)line_size)
440
         : "%"REG_a, "memory"
441
        );
442
}
443

    
444
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
445
{
446
    __asm__ volatile(
447
         "1:                            \n\t"
448
         "movdqu (%1), %%xmm0           \n\t"
449
         "movdqu (%1,%3), %%xmm1        \n\t"
450
         "movdqu (%1,%3,2), %%xmm2      \n\t"
451
         "movdqu (%1,%4), %%xmm3        \n\t"
452
         "movdqa %%xmm0, (%2)           \n\t"
453
         "movdqa %%xmm1, (%2,%3)        \n\t"
454
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
455
         "movdqa %%xmm3, (%2,%4)        \n\t"
456
         "subl $4, %0                   \n\t"
457
         "lea (%1,%3,4), %1             \n\t"
458
         "lea (%2,%3,4), %2             \n\t"
459
         "jnz 1b                        \n\t"
460
         : "+g"(h), "+r" (pixels),  "+r" (block)
461
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
462
         : "memory"
463
        );
464
}
465

    
466
static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
467
{
468
    __asm__ volatile(
469
         "1:                            \n\t"
470
         "movdqu (%1), %%xmm0           \n\t"
471
         "movdqu (%1,%3), %%xmm1        \n\t"
472
         "movdqu (%1,%3,2), %%xmm2      \n\t"
473
         "movdqu (%1,%4), %%xmm3        \n\t"
474
         "pavgb  (%2), %%xmm0           \n\t"
475
         "pavgb  (%2,%3), %%xmm1        \n\t"
476
         "pavgb  (%2,%3,2), %%xmm2      \n\t"
477
         "pavgb  (%2,%4), %%xmm3        \n\t"
478
         "movdqa %%xmm0, (%2)           \n\t"
479
         "movdqa %%xmm1, (%2,%3)        \n\t"
480
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
481
         "movdqa %%xmm3, (%2,%4)        \n\t"
482
         "subl $4, %0                   \n\t"
483
         "lea (%1,%3,4), %1             \n\t"
484
         "lea (%2,%3,4), %2             \n\t"
485
         "jnz 1b                        \n\t"
486
         : "+g"(h), "+r" (pixels),  "+r" (block)
487
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
488
         : "memory"
489
        );
490
}
491

    
492
#define CLEAR_BLOCKS(name,n) \
493
static void name(DCTELEM *blocks)\
494
{\
495
    __asm__ volatile(\
496
                "pxor %%mm7, %%mm7              \n\t"\
497
                "mov     %1, %%"REG_a"          \n\t"\
498
                "1:                             \n\t"\
499
                "movq %%mm7, (%0, %%"REG_a")    \n\t"\
500
                "movq %%mm7, 8(%0, %%"REG_a")   \n\t"\
501
                "movq %%mm7, 16(%0, %%"REG_a")  \n\t"\
502
                "movq %%mm7, 24(%0, %%"REG_a")  \n\t"\
503
                "add $32, %%"REG_a"             \n\t"\
504
                " js 1b                         \n\t"\
505
                : : "r" (((uint8_t *)blocks)+128*n),\
506
                    "i" (-128*n)\
507
                : "%"REG_a\
508
        );\
509
}
510
CLEAR_BLOCKS(clear_blocks_mmx, 6)
511
CLEAR_BLOCKS(clear_block_mmx, 1)
512

    
513
static void clear_block_sse(DCTELEM *block)
514
{
515
    __asm__ volatile(
516
        "xorps  %%xmm0, %%xmm0  \n"
517
        "movaps %%xmm0,    (%0) \n"
518
        "movaps %%xmm0,  16(%0) \n"
519
        "movaps %%xmm0,  32(%0) \n"
520
        "movaps %%xmm0,  48(%0) \n"
521
        "movaps %%xmm0,  64(%0) \n"
522
        "movaps %%xmm0,  80(%0) \n"
523
        "movaps %%xmm0,  96(%0) \n"
524
        "movaps %%xmm0, 112(%0) \n"
525
        :: "r"(block)
526
        : "memory"
527
    );
528
}
529

    
530
static void clear_blocks_sse(DCTELEM *blocks)
531
{\
532
    __asm__ volatile(
533
        "xorps  %%xmm0, %%xmm0  \n"
534
        "mov     %1, %%"REG_a"  \n"
535
        "1:                     \n"
536
        "movaps %%xmm0,    (%0, %%"REG_a") \n"
537
        "movaps %%xmm0,  16(%0, %%"REG_a") \n"
538
        "movaps %%xmm0,  32(%0, %%"REG_a") \n"
539
        "movaps %%xmm0,  48(%0, %%"REG_a") \n"
540
        "movaps %%xmm0,  64(%0, %%"REG_a") \n"
541
        "movaps %%xmm0,  80(%0, %%"REG_a") \n"
542
        "movaps %%xmm0,  96(%0, %%"REG_a") \n"
543
        "movaps %%xmm0, 112(%0, %%"REG_a") \n"
544
        "add $128, %%"REG_a"    \n"
545
        " js 1b                 \n"
546
        : : "r" (((uint8_t *)blocks)+128*6),
547
            "i" (-128*6)
548
        : "%"REG_a
549
    );
550
}
551

    
552
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
553
    x86_reg i=0;
554
    __asm__ volatile(
555
        "jmp 2f                         \n\t"
556
        "1:                             \n\t"
557
        "movq  (%1, %0), %%mm0          \n\t"
558
        "movq  (%2, %0), %%mm1          \n\t"
559
        "paddb %%mm0, %%mm1             \n\t"
560
        "movq %%mm1, (%2, %0)           \n\t"
561
        "movq 8(%1, %0), %%mm0          \n\t"
562
        "movq 8(%2, %0), %%mm1          \n\t"
563
        "paddb %%mm0, %%mm1             \n\t"
564
        "movq %%mm1, 8(%2, %0)          \n\t"
565
        "add $16, %0                    \n\t"
566
        "2:                             \n\t"
567
        "cmp %3, %0                     \n\t"
568
        " js 1b                         \n\t"
569
        : "+r" (i)
570
        : "r"(src), "r"(dst), "r"((x86_reg)w-15)
571
    );
572
    for(; i<w; i++)
573
        dst[i+0] += src[i+0];
574
}
575

    
576
static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
577
    x86_reg i=0;
578
    __asm__ volatile(
579
        "jmp 2f                         \n\t"
580
        "1:                             \n\t"
581
        "movq   (%2, %0), %%mm0         \n\t"
582
        "movq  8(%2, %0), %%mm1         \n\t"
583
        "paddb  (%3, %0), %%mm0         \n\t"
584
        "paddb 8(%3, %0), %%mm1         \n\t"
585
        "movq %%mm0,  (%1, %0)          \n\t"
586
        "movq %%mm1, 8(%1, %0)          \n\t"
587
        "add $16, %0                    \n\t"
588
        "2:                             \n\t"
589
        "cmp %4, %0                     \n\t"
590
        " js 1b                         \n\t"
591
        : "+r" (i)
592
        : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
593
    );
594
    for(; i<w; i++)
595
        dst[i] = src1[i] + src2[i];
596
}
597

    
598
#if HAVE_7REGS && HAVE_TEN_OPERANDS
599
static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top) {
600
    x86_reg w2 = -w;
601
    x86_reg x;
602
    int l = *left & 0xff;
603
    int tl = *left_top & 0xff;
604
    int t;
605
    __asm__ volatile(
606
        "mov    %7, %3 \n"
607
        "1: \n"
608
        "movzx (%3,%4), %2 \n"
609
        "mov    %2, %k3 \n"
610
        "sub   %b1, %b3 \n"
611
        "add   %b0, %b3 \n"
612
        "mov    %2, %1 \n"
613
        "cmp    %0, %2 \n"
614
        "cmovg  %0, %2 \n"
615
        "cmovg  %1, %0 \n"
616
        "cmp   %k3, %0 \n"
617
        "cmovg %k3, %0 \n"
618
        "mov    %7, %3 \n"
619
        "cmp    %2, %0 \n"
620
        "cmovl  %2, %0 \n"
621
        "add (%6,%4), %b0 \n"
622
        "mov   %b0, (%5,%4) \n"
623
        "inc    %4 \n"
624
        "jl 1b \n"
625
        :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
626
        :"r"(dst+w), "r"(diff+w), "rm"(top+w)
627
    );
628
    *left = l;
629
    *left_top = tl;
630
}
631
#endif
632

    
633
#define H263_LOOP_FILTER \
634
        "pxor %%mm7, %%mm7              \n\t"\
635
        "movq  %0, %%mm0                \n\t"\
636
        "movq  %0, %%mm1                \n\t"\
637
        "movq  %3, %%mm2                \n\t"\
638
        "movq  %3, %%mm3                \n\t"\
639
        "punpcklbw %%mm7, %%mm0         \n\t"\
640
        "punpckhbw %%mm7, %%mm1         \n\t"\
641
        "punpcklbw %%mm7, %%mm2         \n\t"\
642
        "punpckhbw %%mm7, %%mm3         \n\t"\
643
        "psubw %%mm2, %%mm0             \n\t"\
644
        "psubw %%mm3, %%mm1             \n\t"\
645
        "movq  %1, %%mm2                \n\t"\
646
        "movq  %1, %%mm3                \n\t"\
647
        "movq  %2, %%mm4                \n\t"\
648
        "movq  %2, %%mm5                \n\t"\
649
        "punpcklbw %%mm7, %%mm2         \n\t"\
650
        "punpckhbw %%mm7, %%mm3         \n\t"\
651
        "punpcklbw %%mm7, %%mm4         \n\t"\
652
        "punpckhbw %%mm7, %%mm5         \n\t"\
653
        "psubw %%mm2, %%mm4             \n\t"\
654
        "psubw %%mm3, %%mm5             \n\t"\
655
        "psllw $2, %%mm4                \n\t"\
656
        "psllw $2, %%mm5                \n\t"\
657
        "paddw %%mm0, %%mm4             \n\t"\
658
        "paddw %%mm1, %%mm5             \n\t"\
659
        "pxor %%mm6, %%mm6              \n\t"\
660
        "pcmpgtw %%mm4, %%mm6           \n\t"\
661
        "pcmpgtw %%mm5, %%mm7           \n\t"\
662
        "pxor %%mm6, %%mm4              \n\t"\
663
        "pxor %%mm7, %%mm5              \n\t"\
664
        "psubw %%mm6, %%mm4             \n\t"\
665
        "psubw %%mm7, %%mm5             \n\t"\
666
        "psrlw $3, %%mm4                \n\t"\
667
        "psrlw $3, %%mm5                \n\t"\
668
        "packuswb %%mm5, %%mm4          \n\t"\
669
        "packsswb %%mm7, %%mm6          \n\t"\
670
        "pxor %%mm7, %%mm7              \n\t"\
671
        "movd %4, %%mm2                 \n\t"\
672
        "punpcklbw %%mm2, %%mm2         \n\t"\
673
        "punpcklbw %%mm2, %%mm2         \n\t"\
674
        "punpcklbw %%mm2, %%mm2         \n\t"\
675
        "psubusb %%mm4, %%mm2           \n\t"\
676
        "movq %%mm2, %%mm3              \n\t"\
677
        "psubusb %%mm4, %%mm3           \n\t"\
678
        "psubb %%mm3, %%mm2             \n\t"\
679
        "movq %1, %%mm3                 \n\t"\
680
        "movq %2, %%mm4                 \n\t"\
681
        "pxor %%mm6, %%mm3              \n\t"\
682
        "pxor %%mm6, %%mm4              \n\t"\
683
        "paddusb %%mm2, %%mm3           \n\t"\
684
        "psubusb %%mm2, %%mm4           \n\t"\
685
        "pxor %%mm6, %%mm3              \n\t"\
686
        "pxor %%mm6, %%mm4              \n\t"\
687
        "paddusb %%mm2, %%mm2           \n\t"\
688
        "packsswb %%mm1, %%mm0          \n\t"\
689
        "pcmpgtb %%mm0, %%mm7           \n\t"\
690
        "pxor %%mm7, %%mm0              \n\t"\
691
        "psubb %%mm7, %%mm0             \n\t"\
692
        "movq %%mm0, %%mm1              \n\t"\
693
        "psubusb %%mm2, %%mm0           \n\t"\
694
        "psubb %%mm0, %%mm1             \n\t"\
695
        "pand %5, %%mm1                 \n\t"\
696
        "psrlw $2, %%mm1                \n\t"\
697
        "pxor %%mm7, %%mm1              \n\t"\
698
        "psubb %%mm7, %%mm1             \n\t"\
699
        "movq %0, %%mm5                 \n\t"\
700
        "movq %3, %%mm6                 \n\t"\
701
        "psubb %%mm1, %%mm5             \n\t"\
702
        "paddb %%mm1, %%mm6             \n\t"
703

    
704
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
705
    if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
706
    const int strength= ff_h263_loop_filter_strength[qscale];
707

    
708
    __asm__ volatile(
709

    
710
        H263_LOOP_FILTER
711

    
712
        "movq %%mm3, %1                 \n\t"
713
        "movq %%mm4, %2                 \n\t"
714
        "movq %%mm5, %0                 \n\t"
715
        "movq %%mm6, %3                 \n\t"
716
        : "+m" (*(uint64_t*)(src - 2*stride)),
717
          "+m" (*(uint64_t*)(src - 1*stride)),
718
          "+m" (*(uint64_t*)(src + 0*stride)),
719
          "+m" (*(uint64_t*)(src + 1*stride))
720
        : "g" (2*strength), "m"(ff_pb_FC)
721
    );
722
    }
723
}
724

    
725
static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
726
    __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
727
        "movd  %4, %%mm0                \n\t"
728
        "movd  %5, %%mm1                \n\t"
729
        "movd  %6, %%mm2                \n\t"
730
        "movd  %7, %%mm3                \n\t"
731
        "punpcklbw %%mm1, %%mm0         \n\t"
732
        "punpcklbw %%mm3, %%mm2         \n\t"
733
        "movq %%mm0, %%mm1              \n\t"
734
        "punpcklwd %%mm2, %%mm0         \n\t"
735
        "punpckhwd %%mm2, %%mm1         \n\t"
736
        "movd  %%mm0, %0                \n\t"
737
        "punpckhdq %%mm0, %%mm0         \n\t"
738
        "movd  %%mm0, %1                \n\t"
739
        "movd  %%mm1, %2                \n\t"
740
        "punpckhdq %%mm1, %%mm1         \n\t"
741
        "movd  %%mm1, %3                \n\t"
742

    
743
        : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
744
          "=m" (*(uint32_t*)(dst + 1*dst_stride)),
745
          "=m" (*(uint32_t*)(dst + 2*dst_stride)),
746
          "=m" (*(uint32_t*)(dst + 3*dst_stride))
747
        :  "m" (*(uint32_t*)(src + 0*src_stride)),
748
           "m" (*(uint32_t*)(src + 1*src_stride)),
749
           "m" (*(uint32_t*)(src + 2*src_stride)),
750
           "m" (*(uint32_t*)(src + 3*src_stride))
751
    );
752
}
753

    
754
static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
755
    if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
756
    const int strength= ff_h263_loop_filter_strength[qscale];
757
    DECLARE_ALIGNED(8, uint64_t, temp)[4];
758
    uint8_t *btemp= (uint8_t*)temp;
759

    
760
    src -= 2;
761

    
762
    transpose4x4(btemp  , src           , 8, stride);
763
    transpose4x4(btemp+4, src + 4*stride, 8, stride);
764
    __asm__ volatile(
765
        H263_LOOP_FILTER // 5 3 4 6
766

    
767
        : "+m" (temp[0]),
768
          "+m" (temp[1]),
769
          "+m" (temp[2]),
770
          "+m" (temp[3])
771
        : "g" (2*strength), "m"(ff_pb_FC)
772
    );
773

    
774
    __asm__ volatile(
775
        "movq %%mm5, %%mm1              \n\t"
776
        "movq %%mm4, %%mm0              \n\t"
777
        "punpcklbw %%mm3, %%mm5         \n\t"
778
        "punpcklbw %%mm6, %%mm4         \n\t"
779
        "punpckhbw %%mm3, %%mm1         \n\t"
780
        "punpckhbw %%mm6, %%mm0         \n\t"
781
        "movq %%mm5, %%mm3              \n\t"
782
        "movq %%mm1, %%mm6              \n\t"
783
        "punpcklwd %%mm4, %%mm5         \n\t"
784
        "punpcklwd %%mm0, %%mm1         \n\t"
785
        "punpckhwd %%mm4, %%mm3         \n\t"
786
        "punpckhwd %%mm0, %%mm6         \n\t"
787
        "movd %%mm5, (%0)               \n\t"
788
        "punpckhdq %%mm5, %%mm5         \n\t"
789
        "movd %%mm5, (%0,%2)            \n\t"
790
        "movd %%mm3, (%0,%2,2)          \n\t"
791
        "punpckhdq %%mm3, %%mm3         \n\t"
792
        "movd %%mm3, (%0,%3)            \n\t"
793
        "movd %%mm1, (%1)               \n\t"
794
        "punpckhdq %%mm1, %%mm1         \n\t"
795
        "movd %%mm1, (%1,%2)            \n\t"
796
        "movd %%mm6, (%1,%2,2)          \n\t"
797
        "punpckhdq %%mm6, %%mm6         \n\t"
798
        "movd %%mm6, (%1,%3)            \n\t"
799
        :: "r" (src),
800
           "r" (src + 4*stride),
801
           "r" ((x86_reg)   stride ),
802
           "r" ((x86_reg)(3*stride))
803
    );
804
    }
805
}
806

    
807
/* draw the edges of width 'w' of an image of size width, height
808
   this mmx version can only handle w==8 || w==16 */
809
static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
810
{
811
    uint8_t *ptr, *last_line;
812
    int i;
813

    
814
    last_line = buf + (height - 1) * wrap;
815
    /* left and right */
816
    ptr = buf;
817
    if(w==8)
818
    {
819
        __asm__ volatile(
820
                "1:                             \n\t"
821
                "movd (%0), %%mm0               \n\t"
822
                "punpcklbw %%mm0, %%mm0         \n\t"
823
                "punpcklwd %%mm0, %%mm0         \n\t"
824
                "punpckldq %%mm0, %%mm0         \n\t"
825
                "movq %%mm0, -8(%0)             \n\t"
826
                "movq -8(%0, %2), %%mm1         \n\t"
827
                "punpckhbw %%mm1, %%mm1         \n\t"
828
                "punpckhwd %%mm1, %%mm1         \n\t"
829
                "punpckhdq %%mm1, %%mm1         \n\t"
830
                "movq %%mm1, (%0, %2)           \n\t"
831
                "add %1, %0                     \n\t"
832
                "cmp %3, %0                     \n\t"
833
                " jb 1b                         \n\t"
834
                : "+r" (ptr)
835
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
836
        );
837
    }
838
    else
839
    {
840
        __asm__ volatile(
841
                "1:                             \n\t"
842
                "movd (%0), %%mm0               \n\t"
843
                "punpcklbw %%mm0, %%mm0         \n\t"
844
                "punpcklwd %%mm0, %%mm0         \n\t"
845
                "punpckldq %%mm0, %%mm0         \n\t"
846
                "movq %%mm0, -8(%0)             \n\t"
847
                "movq %%mm0, -16(%0)            \n\t"
848
                "movq -8(%0, %2), %%mm1         \n\t"
849
                "punpckhbw %%mm1, %%mm1         \n\t"
850
                "punpckhwd %%mm1, %%mm1         \n\t"
851
                "punpckhdq %%mm1, %%mm1         \n\t"
852
                "movq %%mm1, (%0, %2)           \n\t"
853
                "movq %%mm1, 8(%0, %2)          \n\t"
854
                "add %1, %0                     \n\t"
855
                "cmp %3, %0                     \n\t"
856
                " jb 1b                         \n\t"
857
                : "+r" (ptr)
858
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
859
        );
860
    }
861

    
862
    for(i=0;i<w;i+=4) {
863
        /* top and bottom (and hopefully also the corners) */
864
        ptr= buf - (i + 1) * wrap - w;
865
        __asm__ volatile(
866
                "1:                             \n\t"
867
                "movq (%1, %0), %%mm0           \n\t"
868
                "movq %%mm0, (%0)               \n\t"
869
                "movq %%mm0, (%0, %2)           \n\t"
870
                "movq %%mm0, (%0, %2, 2)        \n\t"
871
                "movq %%mm0, (%0, %3)           \n\t"
872
                "add $8, %0                     \n\t"
873
                "cmp %4, %0                     \n\t"
874
                " jb 1b                         \n\t"
875
                : "+r" (ptr)
876
                : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
877
        );
878
        ptr= last_line + (i + 1) * wrap - w;
879
        __asm__ volatile(
880
                "1:                             \n\t"
881
                "movq (%1, %0), %%mm0           \n\t"
882
                "movq %%mm0, (%0)               \n\t"
883
                "movq %%mm0, (%0, %2)           \n\t"
884
                "movq %%mm0, (%0, %2, 2)        \n\t"
885
                "movq %%mm0, (%0, %3)           \n\t"
886
                "add $8, %0                     \n\t"
887
                "cmp %4, %0                     \n\t"
888
                " jb 1b                         \n\t"
889
                : "+r" (ptr)
890
                : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
891
        );
892
    }
893
}
894

    
895
#define PAETH(cpu, abs3)\
896
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
897
{\
898
    x86_reg i = -bpp;\
899
    x86_reg end = w-3;\
900
    __asm__ volatile(\
901
        "pxor      %%mm7, %%mm7 \n"\
902
        "movd    (%1,%0), %%mm0 \n"\
903
        "movd    (%2,%0), %%mm1 \n"\
904
        "punpcklbw %%mm7, %%mm0 \n"\
905
        "punpcklbw %%mm7, %%mm1 \n"\
906
        "add       %4, %0 \n"\
907
        "1: \n"\
908
        "movq      %%mm1, %%mm2 \n"\
909
        "movd    (%2,%0), %%mm1 \n"\
910
        "movq      %%mm2, %%mm3 \n"\
911
        "punpcklbw %%mm7, %%mm1 \n"\
912
        "movq      %%mm2, %%mm4 \n"\
913
        "psubw     %%mm1, %%mm3 \n"\
914
        "psubw     %%mm0, %%mm4 \n"\
915
        "movq      %%mm3, %%mm5 \n"\
916
        "paddw     %%mm4, %%mm5 \n"\
917
        abs3\
918
        "movq      %%mm4, %%mm6 \n"\
919
        "pminsw    %%mm5, %%mm6 \n"\
920
        "pcmpgtw   %%mm6, %%mm3 \n"\
921
        "pcmpgtw   %%mm5, %%mm4 \n"\
922
        "movq      %%mm4, %%mm6 \n"\
923
        "pand      %%mm3, %%mm4 \n"\
924
        "pandn     %%mm3, %%mm6 \n"\
925
        "pandn     %%mm0, %%mm3 \n"\
926
        "movd    (%3,%0), %%mm0 \n"\
927
        "pand      %%mm1, %%mm6 \n"\
928
        "pand      %%mm4, %%mm2 \n"\
929
        "punpcklbw %%mm7, %%mm0 \n"\
930
        "movq      %6,    %%mm5 \n"\
931
        "paddw     %%mm6, %%mm0 \n"\
932
        "paddw     %%mm2, %%mm3 \n"\
933
        "paddw     %%mm3, %%mm0 \n"\
934
        "pand      %%mm5, %%mm0 \n"\
935
        "movq      %%mm0, %%mm3 \n"\
936
        "packuswb  %%mm3, %%mm3 \n"\
937
        "movd      %%mm3, (%1,%0) \n"\
938
        "add       %4, %0 \n"\
939
        "cmp       %5, %0 \n"\
940
        "jle 1b \n"\
941
        :"+r"(i)\
942
        :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
943
         "m"(ff_pw_255)\
944
        :"memory"\
945
    );\
946
}
947

    
948
#define ABS3_MMX2\
949
        "psubw     %%mm5, %%mm7 \n"\
950
        "pmaxsw    %%mm7, %%mm5 \n"\
951
        "pxor      %%mm6, %%mm6 \n"\
952
        "pxor      %%mm7, %%mm7 \n"\
953
        "psubw     %%mm3, %%mm6 \n"\
954
        "psubw     %%mm4, %%mm7 \n"\
955
        "pmaxsw    %%mm6, %%mm3 \n"\
956
        "pmaxsw    %%mm7, %%mm4 \n"\
957
        "pxor      %%mm7, %%mm7 \n"
958

    
959
#define ABS3_SSSE3\
960
        "pabsw     %%mm3, %%mm3 \n"\
961
        "pabsw     %%mm4, %%mm4 \n"\
962
        "pabsw     %%mm5, %%mm5 \n"
963

    
964
PAETH(mmx2, ABS3_MMX2)
965
#if HAVE_SSSE3
966
PAETH(ssse3, ABS3_SSSE3)
967
#endif
968

    
969
#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
970
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\
971
        "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\
972
        "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\
973
        "movq "#in7", " #m3 "             \n\t" /* d */\
974
        "movq "#in0", %%mm5               \n\t" /* D */\
975
        "paddw " #m3 ", %%mm5             \n\t" /* x4 */\
976
        "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\
977
        "movq "#in1", %%mm5               \n\t" /* C */\
978
        "movq "#in2", %%mm6               \n\t" /* B */\
979
        "paddw " #m6 ", %%mm5             \n\t" /* x3 */\
980
        "paddw " #m5 ", %%mm6             \n\t" /* x2 */\
981
        "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\
982
        "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\
983
        "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\
984
        "paddw " #rnd ", %%mm4            \n\t" /* x2 */\
985
        "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
986
        "psraw $5, %%mm5                  \n\t"\
987
        "packuswb %%mm5, %%mm5            \n\t"\
988
        OP(%%mm5, out, %%mm7, d)
989

    
990
#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
991
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
992
    uint64_t temp;\
993
\
994
    __asm__ volatile(\
995
        "pxor %%mm7, %%mm7                \n\t"\
996
        "1:                               \n\t"\
997
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
998
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
999
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1000
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
1001
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
1002
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
1003
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
1004
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1005
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1006
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
1007
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
1008
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
1009
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
1010
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
1011
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
1012
        "paddw %%mm3, %%mm5               \n\t" /* b */\
1013
        "paddw %%mm2, %%mm6               \n\t" /* c */\
1014
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1015
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
1016
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
1017
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
1018
        "paddw %%mm4, %%mm0               \n\t" /* a */\
1019
        "paddw %%mm1, %%mm5               \n\t" /* d */\
1020
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1021
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
1022
        "paddw %6, %%mm6                  \n\t"\
1023
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1024
        "psraw $5, %%mm0                  \n\t"\
1025
        "movq %%mm0, %5                   \n\t"\
1026
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1027
        \
1028
        "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\
1029
        "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\
1030
        "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\
1031
        "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\
1032
        "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\
1033
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\
1034
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\
1035
        "paddw %%mm0, %%mm2               \n\t" /* b */\
1036
        "paddw %%mm5, %%mm3               \n\t" /* c */\
1037
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1038
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1039
        "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\
1040
        "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\
1041
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\
1042
        "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\
1043
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1044
        "paddw %%mm2, %%mm1               \n\t" /* a */\
1045
        "paddw %%mm6, %%mm4               \n\t" /* d */\
1046
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1047
        "psubw %%mm4, %%mm3               \n\t" /* - 6b +3c - d */\
1048
        "paddw %6, %%mm1                  \n\t"\
1049
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b +3c - d */\
1050
        "psraw $5, %%mm3                  \n\t"\
1051
        "movq %5, %%mm1                   \n\t"\
1052
        "packuswb %%mm3, %%mm1            \n\t"\
1053
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
1054
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
1055
        \
1056
        "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\
1057
        "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\
1058
        "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\
1059
        "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\
1060
        "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\
1061
        "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\
1062
        "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\
1063
        "paddw %%mm1, %%mm5               \n\t" /* b */\
1064
        "paddw %%mm4, %%mm0               \n\t" /* c */\
1065
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1066
        "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\
1067
        "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\
1068
        "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\
1069
        "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\
1070
        "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\
1071
        "paddw %%mm3, %%mm2               \n\t" /* d */\
1072
        "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\
1073
        "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\
1074
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\
1075
        "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\
1076
        "paddw %%mm2, %%mm6               \n\t" /* a */\
1077
        "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
1078
        "paddw %6, %%mm0                  \n\t"\
1079
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1080
        "psraw $5, %%mm0                  \n\t"\
1081
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
1082
        \
1083
        "paddw %%mm5, %%mm3               \n\t" /* a */\
1084
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\
1085
        "paddw %%mm4, %%mm6               \n\t" /* b */\
1086
        "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\
1087
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\
1088
        "paddw %%mm1, %%mm4               \n\t" /* c */\
1089
        "paddw %%mm2, %%mm5               \n\t" /* d */\
1090
        "paddw %%mm6, %%mm6               \n\t" /* 2b */\
1091
        "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\
1092
        "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
1093
        "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\
1094
        "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\
1095
        "paddw %6, %%mm4                  \n\t"\
1096
        "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\
1097
        "psraw $5, %%mm4                  \n\t"\
1098
        "packuswb %%mm4, %%mm0            \n\t"\
1099
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
1100
        \
1101
        "add %3, %0                       \n\t"\
1102
        "add %4, %1                       \n\t"\
1103
        "decl %2                          \n\t"\
1104
        " jnz 1b                          \n\t"\
1105
        : "+a"(src), "+c"(dst), "+D"(h)\
1106
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1107
        : "memory"\
1108
    );\
1109
}\
1110
\
1111
static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1112
    int i;\
1113
    int16_t temp[16];\
1114
    /* quick HACK, XXX FIXME MUST be optimized */\
1115
    for(i=0; i<h; i++)\
1116
    {\
1117
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1118
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1119
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1120
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1121
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1122
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1123
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1124
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1125
        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1126
        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1127
        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1128
        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1129
        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1130
        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1131
        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1132
        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1133
        __asm__ volatile(\
1134
            "movq (%0), %%mm0               \n\t"\
1135
            "movq 8(%0), %%mm1              \n\t"\
1136
            "paddw %2, %%mm0                \n\t"\
1137
            "paddw %2, %%mm1                \n\t"\
1138
            "psraw $5, %%mm0                \n\t"\
1139
            "psraw $5, %%mm1                \n\t"\
1140
            "packuswb %%mm1, %%mm0          \n\t"\
1141
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1142
            "movq 16(%0), %%mm0             \n\t"\
1143
            "movq 24(%0), %%mm1             \n\t"\
1144
            "paddw %2, %%mm0                \n\t"\
1145
            "paddw %2, %%mm1                \n\t"\
1146
            "psraw $5, %%mm0                \n\t"\
1147
            "psraw $5, %%mm1                \n\t"\
1148
            "packuswb %%mm1, %%mm0          \n\t"\
1149
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1150
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1151
            : "memory"\
1152
        );\
1153
        dst+=dstStride;\
1154
        src+=srcStride;\
1155
    }\
1156
}\
1157
\
1158
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1159
    __asm__ volatile(\
1160
        "pxor %%mm7, %%mm7                \n\t"\
1161
        "1:                               \n\t"\
1162
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
1163
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
1164
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1165
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
1166
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
1167
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
1168
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
1169
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1170
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1171
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
1172
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
1173
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
1174
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
1175
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
1176
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
1177
        "paddw %%mm3, %%mm5               \n\t" /* b */\
1178
        "paddw %%mm2, %%mm6               \n\t" /* c */\
1179
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1180
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
1181
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
1182
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
1183
        "paddw %%mm4, %%mm0               \n\t" /* a */\
1184
        "paddw %%mm1, %%mm5               \n\t" /* d */\
1185
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1186
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
1187
        "paddw %5, %%mm6                  \n\t"\
1188
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1189
        "psraw $5, %%mm0                  \n\t"\
1190
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1191
        \
1192
        "movd 5(%0), %%mm5                \n\t" /* FGHI */\
1193
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\
1194
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\
1195
        "paddw %%mm5, %%mm1               \n\t" /* a */\
1196
        "paddw %%mm6, %%mm2               \n\t" /* b */\
1197
        "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\
1198
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\
1199
        "paddw %%mm6, %%mm3               \n\t" /* c */\
1200
        "paddw %%mm5, %%mm4               \n\t" /* d */\
1201
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1202
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1203
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1204
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1205
        "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\
1206
        "paddw %5, %%mm1                  \n\t"\
1207
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\
1208
        "psraw $5, %%mm3                  \n\t"\
1209
        "packuswb %%mm3, %%mm0            \n\t"\
1210
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
1211
        \
1212
        "add %3, %0                       \n\t"\
1213
        "add %4, %1                       \n\t"\
1214
        "decl %2                          \n\t"\
1215
        " jnz 1b                          \n\t"\
1216
        : "+a"(src), "+c"(dst), "+d"(h)\
1217
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
1218
        : "memory"\
1219
    );\
1220
}\
1221
\
1222
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1223
    int i;\
1224
    int16_t temp[8];\
1225
    /* quick HACK, XXX FIXME MUST be optimized */\
1226
    for(i=0; i<h; i++)\
1227
    {\
1228
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1229
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1230
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1231
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1232
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1233
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1234
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1235
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1236
        __asm__ volatile(\
1237
            "movq (%0), %%mm0           \n\t"\
1238
            "movq 8(%0), %%mm1          \n\t"\
1239
            "paddw %2, %%mm0            \n\t"\
1240
            "paddw %2, %%mm1            \n\t"\
1241
            "psraw $5, %%mm0            \n\t"\
1242
            "psraw $5, %%mm1            \n\t"\
1243
            "packuswb %%mm1, %%mm0      \n\t"\
1244
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1245
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1246
            :"memory"\
1247
        );\
1248
        dst+=dstStride;\
1249
        src+=srcStride;\
1250
    }\
1251
}
1252

    
1253
#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1254
\
1255
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1256
    uint64_t temp[17*4];\
1257
    uint64_t *temp_ptr= temp;\
1258
    int count= 17;\
1259
\
1260
    /*FIXME unroll */\
1261
    __asm__ volatile(\
1262
        "pxor %%mm7, %%mm7              \n\t"\
1263
        "1:                             \n\t"\
1264
        "movq (%0), %%mm0               \n\t"\
1265
        "movq (%0), %%mm1               \n\t"\
1266
        "movq 8(%0), %%mm2              \n\t"\
1267
        "movq 8(%0), %%mm3              \n\t"\
1268
        "punpcklbw %%mm7, %%mm0         \n\t"\
1269
        "punpckhbw %%mm7, %%mm1         \n\t"\
1270
        "punpcklbw %%mm7, %%mm2         \n\t"\
1271
        "punpckhbw %%mm7, %%mm3         \n\t"\
1272
        "movq %%mm0, (%1)               \n\t"\
1273
        "movq %%mm1, 17*8(%1)           \n\t"\
1274
        "movq %%mm2, 2*17*8(%1)         \n\t"\
1275
        "movq %%mm3, 3*17*8(%1)         \n\t"\
1276
        "add $8, %1                     \n\t"\
1277
        "add %3, %0                     \n\t"\
1278
        "decl %2                        \n\t"\
1279
        " jnz 1b                        \n\t"\
1280
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1281
        : "r" ((x86_reg)srcStride)\
1282
        : "memory"\
1283
    );\
1284
    \
1285
    temp_ptr= temp;\
1286
    count=4;\
1287
    \
1288
/*FIXME reorder for speed */\
1289
    __asm__ volatile(\
1290
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1291
        "1:                             \n\t"\
1292
        "movq (%0), %%mm0               \n\t"\
1293
        "movq 8(%0), %%mm1              \n\t"\
1294
        "movq 16(%0), %%mm2             \n\t"\
1295
        "movq 24(%0), %%mm3             \n\t"\
1296
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1297
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1298
        "add %4, %1                     \n\t"\
1299
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1300
        \
1301
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1302
        "add %4, %1                     \n\t"\
1303
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1304
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1305
        "add %4, %1                     \n\t"\
1306
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1307
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1308
        "add %4, %1                     \n\t"\
1309
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1310
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1311
        "add %4, %1                     \n\t"\
1312
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1313
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1314
        "add %4, %1                     \n\t"\
1315
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1316
        \
1317
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1318
        "add %4, %1                     \n\t"  \
1319
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1320
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1321
        \
1322
        "add $136, %0                   \n\t"\
1323
        "add %6, %1                     \n\t"\
1324
        "decl %2                        \n\t"\
1325
        " jnz 1b                        \n\t"\
1326
        \
1327
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1328
        : "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)\
1329
        :"memory"\
1330
    );\
1331
}\
1332
\
1333
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1334
    uint64_t temp[9*2];\
1335
    uint64_t *temp_ptr= temp;\
1336
    int count= 9;\
1337
\
1338
    /*FIXME unroll */\
1339
    __asm__ volatile(\
1340
        "pxor %%mm7, %%mm7              \n\t"\
1341
        "1:                             \n\t"\
1342
        "movq (%0), %%mm0               \n\t"\
1343
        "movq (%0), %%mm1               \n\t"\
1344
        "punpcklbw %%mm7, %%mm0         \n\t"\
1345
        "punpckhbw %%mm7, %%mm1         \n\t"\
1346
        "movq %%mm0, (%1)               \n\t"\
1347
        "movq %%mm1, 9*8(%1)            \n\t"\
1348
        "add $8, %1                     \n\t"\
1349
        "add %3, %0                     \n\t"\
1350
        "decl %2                        \n\t"\
1351
        " jnz 1b                        \n\t"\
1352
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1353
        : "r" ((x86_reg)srcStride)\
1354
        : "memory"\
1355
    );\
1356
    \
1357
    temp_ptr= temp;\
1358
    count=2;\
1359
    \
1360
/*FIXME reorder for speed */\
1361
    __asm__ volatile(\
1362
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1363
        "1:                             \n\t"\
1364
        "movq (%0), %%mm0               \n\t"\
1365
        "movq 8(%0), %%mm1              \n\t"\
1366
        "movq 16(%0), %%mm2             \n\t"\
1367
        "movq 24(%0), %%mm3             \n\t"\
1368
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1369
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1370
        "add %4, %1                     \n\t"\
1371
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1372
        \
1373
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1374
        "add %4, %1                     \n\t"\
1375
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1376
        \
1377
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1378
        "add %4, %1                     \n\t"\
1379
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1380
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1381
                \
1382
        "add $72, %0                    \n\t"\
1383
        "add %6, %1                     \n\t"\
1384
        "decl %2                        \n\t"\
1385
        " jnz 1b                        \n\t"\
1386
         \
1387
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1388
        : "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)\
1389
        : "memory"\
1390
   );\
1391
}\
1392
\
1393
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1394
    OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\
1395
}\
1396
\
1397
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1398
    uint64_t temp[8];\
1399
    uint8_t * const half= (uint8_t*)temp;\
1400
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1401
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1402
}\
1403
\
1404
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1405
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1406
}\
1407
\
1408
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1409
    uint64_t temp[8];\
1410
    uint8_t * const half= (uint8_t*)temp;\
1411
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1412
    OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
1413
}\
1414
\
1415
static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1416
    uint64_t temp[8];\
1417
    uint8_t * const half= (uint8_t*)temp;\
1418
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1419
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1420
}\
1421
\
1422
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1423
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1424
}\
1425
\
1426
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1427
    uint64_t temp[8];\
1428
    uint8_t * const half= (uint8_t*)temp;\
1429
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1430
    OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
1431
}\
1432
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1433
    uint64_t half[8 + 9];\
1434
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1435
    uint8_t * const halfHV= ((uint8_t*)half);\
1436
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1437
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1438
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1439
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1440
}\
1441
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1442
    uint64_t half[8 + 9];\
1443
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1444
    uint8_t * const halfHV= ((uint8_t*)half);\
1445
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1446
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1447
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1448
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1449
}\
1450
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1451
    uint64_t half[8 + 9];\
1452
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1453
    uint8_t * const halfHV= ((uint8_t*)half);\
1454
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1455
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1456
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1457
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1458
}\
1459
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1460
    uint64_t half[8 + 9];\
1461
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1462
    uint8_t * const halfHV= ((uint8_t*)half);\
1463
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1464
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1465
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1466
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1467
}\
1468
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1469
    uint64_t half[8 + 9];\
1470
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1471
    uint8_t * const halfHV= ((uint8_t*)half);\
1472
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1473
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1474
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1475
}\
1476
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1477
    uint64_t half[8 + 9];\
1478
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1479
    uint8_t * const halfHV= ((uint8_t*)half);\
1480
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1481
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1482
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1483
}\
1484
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1485
    uint64_t half[8 + 9];\
1486
    uint8_t * const halfH= ((uint8_t*)half);\
1487
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1488
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1489
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1490
}\
1491
static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1492
    uint64_t half[8 + 9];\
1493
    uint8_t * const halfH= ((uint8_t*)half);\
1494
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1495
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1496
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1497
}\
1498
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1499
    uint64_t half[9];\
1500
    uint8_t * const halfH= ((uint8_t*)half);\
1501
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1502
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1503
}\
1504
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1505
    OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\
1506
}\
1507
\
1508
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1509
    uint64_t temp[32];\
1510
    uint8_t * const half= (uint8_t*)temp;\
1511
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1512
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1513
}\
1514
\
1515
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1516
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1517
}\
1518
\
1519
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1520
    uint64_t temp[32];\
1521
    uint8_t * const half= (uint8_t*)temp;\
1522
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1523
    OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
1524
}\
1525
\
1526
static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1527
    uint64_t temp[32];\
1528
    uint8_t * const half= (uint8_t*)temp;\
1529
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1530
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1531
}\
1532
\
1533
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1534
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1535
}\
1536
\
1537
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1538
    uint64_t temp[32];\
1539
    uint8_t * const half= (uint8_t*)temp;\
1540
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1541
    OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
1542
}\
1543
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1544
    uint64_t half[16*2 + 17*2];\
1545
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1546
    uint8_t * const halfHV= ((uint8_t*)half);\
1547
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1548
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1549
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1550
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1551
}\
1552
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1553
    uint64_t half[16*2 + 17*2];\
1554
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1555
    uint8_t * const halfHV= ((uint8_t*)half);\
1556
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1557
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1558
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1559
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1560
}\
1561
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1562
    uint64_t half[16*2 + 17*2];\
1563
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1564
    uint8_t * const halfHV= ((uint8_t*)half);\
1565
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1566
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1567
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1568
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1569
}\
1570
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1571
    uint64_t half[16*2 + 17*2];\
1572
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1573
    uint8_t * const halfHV= ((uint8_t*)half);\
1574
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1575
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1576
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1577
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1578
}\
1579
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1580
    uint64_t half[16*2 + 17*2];\
1581
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1582
    uint8_t * const halfHV= ((uint8_t*)half);\
1583
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1584
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1585
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1586
}\
1587
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1588
    uint64_t half[16*2 + 17*2];\
1589
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1590
    uint8_t * const halfHV= ((uint8_t*)half);\
1591
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1592
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1593
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1594
}\
1595
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1596
    uint64_t half[17*2];\
1597
    uint8_t * const halfH= ((uint8_t*)half);\
1598
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1599
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1600
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1601
}\
1602
static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1603
    uint64_t half[17*2];\
1604
    uint8_t * const halfH= ((uint8_t*)half);\
1605
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1606
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1607
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1608
}\
1609
static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1610
    uint64_t half[17*2];\
1611
    uint8_t * const halfH= ((uint8_t*)half);\
1612
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1613
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1614
}
1615

    
1616
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"
1617
#define AVG_3DNOW_OP(a,b,temp, size) \
1618
"mov" #size " " #b ", " #temp "   \n\t"\
1619
"pavgusb " #temp ", " #a "        \n\t"\
1620
"mov" #size " " #a ", " #b "      \n\t"
1621
#define AVG_MMX2_OP(a,b,temp, size) \
1622
"mov" #size " " #b ", " #temp "   \n\t"\
1623
"pavgb " #temp ", " #a "          \n\t"\
1624
"mov" #size " " #a ", " #b "      \n\t"
1625

    
1626
QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
1627
QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
1628
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1629
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
1630
QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
1631
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1632
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
1633
QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
1634
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1635

    
1636
/***********************************/
1637
/* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
1638

    
1639
#define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
1640
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1641
    OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
1642
}
1643
#define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
1644
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1645
    OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
1646
}
1647

    
1648
#define QPEL_2TAP(OPNAME, SIZE, MMX)\
1649
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
1650
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
1651
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
1652
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
1653
                          OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
1654
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
1655
                          OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
1656
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
1657
                          OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
1658
static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1659
    OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
1660
}\
1661
static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1662
    OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
1663
}\
1664
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\
1665
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\
1666
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\
1667
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\
1668
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\
1669
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\
1670
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\
1671
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
1672

    
1673
QPEL_2TAP(put_, 16, mmx2)
1674
QPEL_2TAP(avg_, 16, mmx2)
1675
QPEL_2TAP(put_,  8, mmx2)
1676
QPEL_2TAP(avg_,  8, mmx2)
1677
QPEL_2TAP(put_, 16, 3dnow)
1678
QPEL_2TAP(avg_, 16, 3dnow)
1679
QPEL_2TAP(put_,  8, 3dnow)
1680
QPEL_2TAP(avg_,  8, 3dnow)
1681

    
1682

    
1683
#if 0
1684
static void just_return(void) { return; }
1685
#endif
1686

    
1687
static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1688
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){
1689
    const int w = 8;
1690
    const int ix = ox>>(16+shift);
1691
    const int iy = oy>>(16+shift);
1692
    const int oxs = ox>>4;
1693
    const int oys = oy>>4;
1694
    const int dxxs = dxx>>4;
1695
    const int dxys = dxy>>4;
1696
    const int dyxs = dyx>>4;
1697
    const int dyys = dyy>>4;
1698
    const uint16_t r4[4] = {r,r,r,r};
1699
    const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
1700
    const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
1701
    const uint64_t shift2 = 2*shift;
1702
    uint8_t edge_buf[(h+1)*stride];
1703
    int x, y;
1704

    
1705
    const int dxw = (dxx-(1<<(16+shift)))*(w-1);
1706
    const int dyh = (dyy-(1<<(16+shift)))*(h-1);
1707
    const int dxh = dxy*(h-1);
1708
    const int dyw = dyx*(w-1);
1709
    if( // non-constant fullpel offset (3% of blocks)
1710
        ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |
1711
         (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)
1712
        // uses more than 16 bits of subpel mv (only at huge resolution)
1713
        || (dxx|dxy|dyx|dyy)&15 )
1714
    {
1715
        //FIXME could still use mmx for some of the rows
1716
        ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
1717
        return;
1718
    }
1719

    
1720
    src += ix + iy*stride;
1721
    if( (unsigned)ix >= width-w ||
1722
        (unsigned)iy >= height-h )
1723
    {
1724
        ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
1725
        src = edge_buf;
1726
    }
1727

    
1728
    __asm__ volatile(
1729
        "movd         %0, %%mm6 \n\t"
1730
        "pxor      %%mm7, %%mm7 \n\t"
1731
        "punpcklwd %%mm6, %%mm6 \n\t"
1732
        "punpcklwd %%mm6, %%mm6 \n\t"
1733
        :: "r"(1<<shift)
1734
    );
1735

    
1736
    for(x=0; x<w; x+=4){
1737
        uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
1738
                            oxs - dxys + dxxs*(x+1),
1739
                            oxs - dxys + dxxs*(x+2),
1740
                            oxs - dxys + dxxs*(x+3) };
1741
        uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
1742
                            oys - dyys + dyxs*(x+1),
1743
                            oys - dyys + dyxs*(x+2),
1744
                            oys - dyys + dyxs*(x+3) };
1745

    
1746
        for(y=0; y<h; y++){
1747
            __asm__ volatile(
1748
                "movq   %0,  %%mm4 \n\t"
1749
                "movq   %1,  %%mm5 \n\t"
1750
                "paddw  %2,  %%mm4 \n\t"
1751
                "paddw  %3,  %%mm5 \n\t"
1752
                "movq   %%mm4, %0  \n\t"
1753
                "movq   %%mm5, %1  \n\t"
1754
                "psrlw  $12, %%mm4 \n\t"
1755
                "psrlw  $12, %%mm5 \n\t"
1756
                : "+m"(*dx4), "+m"(*dy4)
1757
                : "m"(*dxy4), "m"(*dyy4)
1758
            );
1759

    
1760
            __asm__ volatile(
1761
                "movq   %%mm6, %%mm2 \n\t"
1762
                "movq   %%mm6, %%mm1 \n\t"
1763
                "psubw  %%mm4, %%mm2 \n\t"
1764
                "psubw  %%mm5, %%mm1 \n\t"
1765
                "movq   %%mm2, %%mm0 \n\t"
1766
                "movq   %%mm4, %%mm3 \n\t"
1767
                "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
1768
                "pmullw %%mm5, %%mm3 \n\t" // dx*dy
1769
                "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
1770
                "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
1771

    
1772
                "movd   %4,    %%mm5 \n\t"
1773
                "movd   %3,    %%mm4 \n\t"
1774
                "punpcklbw %%mm7, %%mm5 \n\t"
1775
                "punpcklbw %%mm7, %%mm4 \n\t"
1776
                "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
1777
                "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
1778

    
1779
                "movd   %2,    %%mm5 \n\t"
1780
                "movd   %1,    %%mm4 \n\t"
1781
                "punpcklbw %%mm7, %%mm5 \n\t"
1782
                "punpcklbw %%mm7, %%mm4 \n\t"
1783
                "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
1784
                "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
1785
                "paddw  %5,    %%mm1 \n\t"
1786
                "paddw  %%mm3, %%mm2 \n\t"
1787
                "paddw  %%mm1, %%mm0 \n\t"
1788
                "paddw  %%mm2, %%mm0 \n\t"
1789

    
1790
                "psrlw    %6,    %%mm0 \n\t"
1791
                "packuswb %%mm0, %%mm0 \n\t"
1792
                "movd     %%mm0, %0    \n\t"
1793

    
1794
                : "=m"(dst[x+y*stride])
1795
                : "m"(src[0]), "m"(src[1]),
1796
                  "m"(src[stride]), "m"(src[stride+1]),
1797
                  "m"(*r4), "m"(shift2)
1798
            );
1799
            src += stride;
1800
        }
1801
        src += 4-h*stride;
1802
    }
1803
}
1804

    
1805
#define PREFETCH(name, op) \
1806
static void name(void *mem, int stride, int h){\
1807
    const uint8_t *p= mem;\
1808
    do{\
1809
        __asm__ volatile(#op" %0" :: "m"(*p));\
1810
        p+= stride;\
1811
    }while(--h);\
1812
}
1813
PREFETCH(prefetch_mmx2,  prefetcht0)
1814
PREFETCH(prefetch_3dnow, prefetch)
1815
#undef PREFETCH
1816

    
1817
#include "h264dsp_mmx.c"
1818
#include "rv40dsp_mmx.c"
1819

    
1820
/* CAVS specific */
1821
void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);
1822
void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx);
1823

    
1824
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1825
    put_pixels8_mmx(dst, src, stride, 8);
1826
}
1827
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1828
    avg_pixels8_mmx(dst, src, stride, 8);
1829
}
1830
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1831
    put_pixels16_mmx(dst, src, stride, 16);
1832
}
1833
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1834
    avg_pixels16_mmx(dst, src, stride, 16);
1835
}
1836

    
1837
/* VC1 specific */
1838
void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx);
1839

    
1840
void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1841
    put_pixels8_mmx(dst, src, stride, 8);
1842
}
1843
void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1844
    avg_pixels8_mmx2(dst, src, stride, 8);
1845
}
1846

    
1847
/* external functions, from idct_mmx.c */
1848
void ff_mmx_idct(DCTELEM *block);
1849
void ff_mmxext_idct(DCTELEM *block);
1850

    
1851
/* XXX: those functions should be suppressed ASAP when all IDCTs are
1852
   converted */
1853
#if CONFIG_GPL
1854
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1855
{
1856
    ff_mmx_idct (block);
1857
    put_pixels_clamped_mmx(block, dest, line_size);
1858
}
1859
static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1860
{
1861
    ff_mmx_idct (block);
1862
    add_pixels_clamped_mmx(block, dest, line_size);
1863
}
1864
static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1865
{
1866
    ff_mmxext_idct (block);
1867
    put_pixels_clamped_mmx(block, dest, line_size);
1868
}
1869
static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1870
{
1871
    ff_mmxext_idct (block);
1872
    add_pixels_clamped_mmx(block, dest, line_size);
1873
}
1874
#endif
1875
static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
1876
{
1877
    ff_idct_xvid_mmx (block);
1878
    put_pixels_clamped_mmx(block, dest, line_size);
1879
}
1880
static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
1881
{
1882
    ff_idct_xvid_mmx (block);
1883
    add_pixels_clamped_mmx(block, dest, line_size);
1884
}
1885
static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
1886
{
1887
    ff_idct_xvid_mmx2 (block);
1888
    put_pixels_clamped_mmx(block, dest, line_size);
1889
}
1890
static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
1891
{
1892
    ff_idct_xvid_mmx2 (block);
1893
    add_pixels_clamped_mmx(block, dest, line_size);
1894
}
1895

    
1896
static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
1897
{
1898
    int i;
1899
    __asm__ volatile("pxor %%mm7, %%mm7":);
1900
    for(i=0; i<blocksize; i+=2) {
1901
        __asm__ volatile(
1902
            "movq    %0,    %%mm0 \n\t"
1903
            "movq    %1,    %%mm1 \n\t"
1904
            "movq    %%mm0, %%mm2 \n\t"
1905
            "movq    %%mm1, %%mm3 \n\t"
1906
            "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
1907
            "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
1908
            "pslld   $31,   %%mm2 \n\t" // keep only the sign bit
1909
            "pxor    %%mm2, %%mm1 \n\t"
1910
            "movq    %%mm3, %%mm4 \n\t"
1911
            "pand    %%mm1, %%mm3 \n\t"
1912
            "pandn   %%mm1, %%mm4 \n\t"
1913
            "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1914
            "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1915
            "movq    %%mm3, %1    \n\t"
1916
            "movq    %%mm0, %0    \n\t"
1917
            :"+m"(mag[i]), "+m"(ang[i])
1918
            ::"memory"
1919
        );
1920
    }
1921
    __asm__ volatile("femms");
1922
}
1923
static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
1924
{
1925
    int i;
1926

    
1927
    __asm__ volatile(
1928
            "movaps  %0,     %%xmm5 \n\t"
1929
        ::"m"(ff_pdw_80000000[0])
1930
    );
1931
    for(i=0; i<blocksize; i+=4) {
1932
        __asm__ volatile(
1933
            "movaps  %0,     %%xmm0 \n\t"
1934
            "movaps  %1,     %%xmm1 \n\t"
1935
            "xorps   %%xmm2, %%xmm2 \n\t"
1936
            "xorps   %%xmm3, %%xmm3 \n\t"
1937
            "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
1938
            "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
1939
            "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit
1940
            "xorps   %%xmm2, %%xmm1 \n\t"
1941
            "movaps  %%xmm3, %%xmm4 \n\t"
1942
            "andps   %%xmm1, %%xmm3 \n\t"
1943
            "andnps  %%xmm1, %%xmm4 \n\t"
1944
            "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1945
            "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1946
            "movaps  %%xmm3, %1     \n\t"
1947
            "movaps  %%xmm0, %0     \n\t"
1948
            :"+m"(mag[i]), "+m"(ang[i])
1949
            ::"memory"
1950
        );
1951
    }
1952
}
1953

    
1954
#define IF1(x) x
1955
#define IF0(x)
1956

    
1957
#define MIX5(mono,stereo)\
1958
    __asm__ volatile(\
1959
        "movss          0(%2), %%xmm5 \n"\
1960
        "movss          8(%2), %%xmm6 \n"\
1961
        "movss         24(%2), %%xmm7 \n"\
1962
        "shufps    $0, %%xmm5, %%xmm5 \n"\
1963
        "shufps    $0, %%xmm6, %%xmm6 \n"\
1964
        "shufps    $0, %%xmm7, %%xmm7 \n"\
1965
        "1: \n"\
1966
        "movaps       (%0,%1), %%xmm0 \n"\
1967
        "movaps  0x400(%0,%1), %%xmm1 \n"\
1968
        "movaps  0x800(%0,%1), %%xmm2 \n"\
1969
        "movaps  0xc00(%0,%1), %%xmm3 \n"\
1970
        "movaps 0x1000(%0,%1), %%xmm4 \n"\
1971
        "mulps         %%xmm5, %%xmm0 \n"\
1972
        "mulps         %%xmm6, %%xmm1 \n"\
1973
        "mulps         %%xmm5, %%xmm2 \n"\
1974
        "mulps         %%xmm7, %%xmm3 \n"\
1975
        "mulps         %%xmm7, %%xmm4 \n"\
1976
 stereo("addps         %%xmm1, %%xmm0 \n")\
1977
        "addps         %%xmm1, %%xmm2 \n"\
1978
        "addps         %%xmm3, %%xmm0 \n"\
1979
        "addps         %%xmm4, %%xmm2 \n"\
1980
   mono("addps         %%xmm2, %%xmm0 \n")\
1981
        "movaps  %%xmm0,      (%0,%1) \n"\
1982
 stereo("movaps  %%xmm2, 0x400(%0,%1) \n")\
1983
        "add $16, %0 \n"\
1984
        "jl 1b \n"\
1985
        :"+&r"(i)\
1986
        :"r"(samples[0]+len), "r"(matrix)\
1987
        :"memory"\
1988
    );
1989

    
1990
#define MIX_MISC(stereo)\
1991
    __asm__ volatile(\
1992
        "1: \n"\
1993
        "movaps  (%3,%0), %%xmm0 \n"\
1994
 stereo("movaps   %%xmm0, %%xmm1 \n")\
1995
        "mulps    %%xmm6, %%xmm0 \n"\
1996
 stereo("mulps    %%xmm7, %%xmm1 \n")\
1997
        "lea 1024(%3,%0), %1 \n"\
1998
        "mov %5, %2 \n"\
1999
        "2: \n"\
2000
        "movaps   (%1),   %%xmm2 \n"\
2001
 stereo("movaps   %%xmm2, %%xmm3 \n")\
2002
        "mulps   (%4,%2), %%xmm2 \n"\
2003
 stereo("mulps 16(%4,%2), %%xmm3 \n")\
2004
        "addps    %%xmm2, %%xmm0 \n"\
2005
 stereo("addps    %%xmm3, %%xmm1 \n")\
2006
        "add $1024, %1 \n"\
2007
        "add $32, %2 \n"\
2008
        "jl 2b \n"\
2009
        "movaps   %%xmm0,     (%3,%0) \n"\
2010
 stereo("movaps   %%xmm1, 1024(%3,%0) \n")\
2011
        "add $16, %0 \n"\
2012
        "jl 1b \n"\
2013
        :"+&r"(i), "=&r"(j), "=&r"(k)\
2014
        :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\
2015
        :"memory"\
2016
    );
2017

    
2018
static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
2019
{
2020
    int (*matrix_cmp)[2] = (int(*)[2])matrix;
2021
    intptr_t i,j,k;
2022

    
2023
    i = -len*sizeof(float);
2024
    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]))) {
2025
        MIX5(IF0,IF1);
2026
    } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {
2027
        MIX5(IF1,IF0);
2028
    } else {
2029
        DECLARE_ALIGNED(16, float, matrix_simd)[in_ch][2][4];
2030
        j = 2*in_ch*sizeof(float);
2031
        __asm__ volatile(
2032
            "1: \n"
2033
            "sub $8, %0 \n"
2034
            "movss     (%2,%0), %%xmm6 \n"
2035
            "movss    4(%2,%0), %%xmm7 \n"
2036
            "shufps $0, %%xmm6, %%xmm6 \n"
2037
            "shufps $0, %%xmm7, %%xmm7 \n"
2038
            "movaps %%xmm6,   (%1,%0,4) \n"
2039
            "movaps %%xmm7, 16(%1,%0,4) \n"
2040
            "jg 1b \n"
2041
            :"+&r"(j)
2042
            :"r"(matrix_simd), "r"(matrix)
2043
            :"memory"
2044
        );
2045
        if(out_ch == 2) {
2046
            MIX_MISC(IF1);
2047
        } else {
2048
            MIX_MISC(IF0);
2049
        }
2050
    }
2051
}
2052

    
2053
static void vector_fmul_3dnow(float *dst, const float *src, int len){
2054
    x86_reg i = (len-4)*4;
2055
    __asm__ volatile(
2056
        "1: \n\t"
2057
        "movq    (%1,%0), %%mm0 \n\t"
2058
        "movq   8(%1,%0), %%mm1 \n\t"
2059
        "pfmul   (%2,%0), %%mm0 \n\t"
2060
        "pfmul  8(%2,%0), %%mm1 \n\t"
2061
        "movq   %%mm0,  (%1,%0) \n\t"
2062
        "movq   %%mm1, 8(%1,%0) \n\t"
2063
        "sub  $16, %0 \n\t"
2064
        "jge 1b \n\t"
2065
        "femms  \n\t"
2066
        :"+r"(i)
2067
        :"r"(dst), "r"(src)
2068
        :"memory"
2069
    );
2070
}
2071
static void vector_fmul_sse(float *dst, const float *src, int len){
2072
    x86_reg i = (len-8)*4;
2073
    __asm__ volatile(
2074
        "1: \n\t"
2075
        "movaps    (%1,%0), %%xmm0 \n\t"
2076
        "movaps  16(%1,%0), %%xmm1 \n\t"
2077
        "mulps     (%2,%0), %%xmm0 \n\t"
2078
        "mulps   16(%2,%0), %%xmm1 \n\t"
2079
        "movaps  %%xmm0,   (%1,%0) \n\t"
2080
        "movaps  %%xmm1, 16(%1,%0) \n\t"
2081
        "sub  $32, %0 \n\t"
2082
        "jge 1b \n\t"
2083
        :"+r"(i)
2084
        :"r"(dst), "r"(src)
2085
        :"memory"
2086
    );
2087
}
2088

    
2089
static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
2090
    x86_reg i = len*4-16;
2091
    __asm__ volatile(
2092
        "1: \n\t"
2093
        "pswapd   8(%1), %%mm0 \n\t"
2094
        "pswapd    (%1), %%mm1 \n\t"
2095
        "pfmul  (%3,%0), %%mm0 \n\t"
2096
        "pfmul 8(%3,%0), %%mm1 \n\t"
2097
        "movq  %%mm0,  (%2,%0) \n\t"
2098
        "movq  %%mm1, 8(%2,%0) \n\t"
2099
        "add   $16, %1 \n\t"
2100
        "sub   $16, %0 \n\t"
2101
        "jge   1b \n\t"
2102
        :"+r"(i), "+r"(src1)
2103
        :"r"(dst), "r"(src0)
2104
    );
2105
    __asm__ volatile("femms");
2106
}
2107
static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
2108
    x86_reg i = len*4-32;
2109
    __asm__ volatile(
2110
        "1: \n\t"
2111
        "movaps        16(%1), %%xmm0 \n\t"
2112
        "movaps          (%1), %%xmm1 \n\t"
2113
        "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
2114
        "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
2115
        "mulps        (%3,%0), %%xmm0 \n\t"
2116
        "mulps      16(%3,%0), %%xmm1 \n\t"
2117
        "movaps     %%xmm0,   (%2,%0) \n\t"
2118
        "movaps     %%xmm1, 16(%2,%0) \n\t"
2119
        "add    $32, %1 \n\t"
2120
        "sub    $32, %0 \n\t"
2121
        "jge    1b \n\t"
2122
        :"+r"(i), "+r"(src1)
2123
        :"r"(dst), "r"(src0)
2124
    );
2125
}
2126

    
2127
static void vector_fmul_add_3dnow(float *dst, const float *src0, const float *src1,
2128
                                  const float *src2, int len){
2129
    x86_reg i = (len-4)*4;
2130
    __asm__ volatile(
2131
        "1: \n\t"
2132
        "movq    (%2,%0), %%mm0 \n\t"
2133
        "movq   8(%2,%0), %%mm1 \n\t"
2134
        "pfmul   (%3,%0), %%mm0 \n\t"
2135
        "pfmul  8(%3,%0), %%mm1 \n\t"
2136
        "pfadd   (%4,%0), %%mm0 \n\t"
2137
        "pfadd  8(%4,%0), %%mm1 \n\t"
2138
        "movq  %%mm0,   (%1,%0) \n\t"
2139
        "movq  %%mm1,  8(%1,%0) \n\t"
2140
        "sub  $16, %0 \n\t"
2141
        "jge  1b \n\t"
2142
        :"+r"(i)
2143
        :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2144
        :"memory"
2145
    );
2146
    __asm__ volatile("femms");
2147
}
2148
static void vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
2149
                                const float *src2, int len){
2150
    x86_reg i = (len-8)*4;
2151
    __asm__ volatile(
2152
        "1: \n\t"
2153
        "movaps   (%2,%0), %%xmm0 \n\t"
2154
        "movaps 16(%2,%0), %%xmm1 \n\t"
2155
        "mulps    (%3,%0), %%xmm0 \n\t"
2156
        "mulps  16(%3,%0), %%xmm1 \n\t"
2157
        "addps    (%4,%0), %%xmm0 \n\t"
2158
        "addps  16(%4,%0), %%xmm1 \n\t"
2159
        "movaps %%xmm0,   (%1,%0) \n\t"
2160
        "movaps %%xmm1, 16(%1,%0) \n\t"
2161
        "sub  $32, %0 \n\t"
2162
        "jge  1b \n\t"
2163
        :"+r"(i)
2164
        :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2165
        :"memory"
2166
    );
2167
}
2168

    
2169
static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
2170
                                      const float *win, float add_bias, int len){
2171
#if HAVE_6REGS
2172
    if(add_bias == 0){
2173
        x86_reg i = -len*4;
2174
        x86_reg j = len*4-8;
2175
        __asm__ volatile(
2176
            "1: \n"
2177
            "pswapd  (%5,%1), %%mm1 \n"
2178
            "movq    (%5,%0), %%mm0 \n"
2179
            "pswapd  (%4,%1), %%mm5 \n"
2180
            "movq    (%3,%0), %%mm4 \n"
2181
            "movq      %%mm0, %%mm2 \n"
2182
            "movq      %%mm1, %%mm3 \n"
2183
            "pfmul     %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
2184
            "pfmul     %%mm5, %%mm3 \n" // src1[    j]*win[len+j]
2185
            "pfmul     %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
2186
            "pfmul     %%mm5, %%mm0 \n" // src1[    j]*win[len+i]
2187
            "pfadd     %%mm3, %%mm2 \n"
2188
            "pfsub     %%mm0, %%mm1 \n"
2189
            "pswapd    %%mm2, %%mm2 \n"
2190
            "movq      %%mm1, (%2,%0) \n"
2191
            "movq      %%mm2, (%2,%1) \n"
2192
            "sub $8, %1 \n"
2193
            "add $8, %0 \n"
2194
            "jl 1b \n"
2195
            "femms \n"
2196
            :"+r"(i), "+r"(j)
2197
            :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2198
        );
2199
    }else
2200
#endif
2201
        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2202
}
2203

    
2204
static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
2205
                                   const float *win, float add_bias, int len){
2206
#if HAVE_6REGS
2207
    if(add_bias == 0){
2208
        x86_reg i = -len*4;
2209
        x86_reg j = len*4-16;
2210
        __asm__ volatile(
2211
            "1: \n"
2212
            "movaps       (%5,%1), %%xmm1 \n"
2213
            "movaps       (%5,%0), %%xmm0 \n"
2214
            "movaps       (%4,%1), %%xmm5 \n"
2215
            "movaps       (%3,%0), %%xmm4 \n"
2216
            "shufps $0x1b, %%xmm1, %%xmm1 \n"
2217
            "shufps $0x1b, %%xmm5, %%xmm5 \n"
2218
            "movaps        %%xmm0, %%xmm2 \n"
2219
            "movaps        %%xmm1, %%xmm3 \n"
2220
            "mulps         %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
2221
            "mulps         %%xmm5, %%xmm3 \n" // src1[    j]*win[len+j]
2222
            "mulps         %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
2223
            "mulps         %%xmm5, %%xmm0 \n" // src1[    j]*win[len+i]
2224
            "addps         %%xmm3, %%xmm2 \n"
2225
            "subps         %%xmm0, %%xmm1 \n"
2226
            "shufps $0x1b, %%xmm2, %%xmm2 \n"
2227
            "movaps        %%xmm1, (%2,%0) \n"
2228
            "movaps        %%xmm2, (%2,%1) \n"
2229
            "sub $16, %1 \n"
2230
            "add $16, %0 \n"
2231
            "jl 1b \n"
2232
            :"+r"(i), "+r"(j)
2233
            :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2234
        );
2235
    }else
2236
#endif
2237
        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2238
}
2239

    
2240
static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len)
2241
{
2242
    x86_reg i = -4*len;
2243
    __asm__ volatile(
2244
        "movss  %3, %%xmm4 \n"
2245
        "shufps $0, %%xmm4, %%xmm4 \n"
2246
        "1: \n"
2247
        "cvtpi2ps   (%2,%0), %%xmm0 \n"
2248
        "cvtpi2ps  8(%2,%0), %%xmm1 \n"
2249
        "cvtpi2ps 16(%2,%0), %%xmm2 \n"
2250
        "cvtpi2ps 24(%2,%0), %%xmm3 \n"
2251
        "movlhps  %%xmm1,    %%xmm0 \n"
2252
        "movlhps  %%xmm3,    %%xmm2 \n"
2253
        "mulps    %%xmm4,    %%xmm0 \n"
2254
        "mulps    %%xmm4,    %%xmm2 \n"
2255
        "movaps   %%xmm0,   (%1,%0) \n"
2256
        "movaps   %%xmm2, 16(%1,%0) \n"
2257
        "add $32, %0 \n"
2258
        "jl 1b \n"
2259
        :"+r"(i)
2260
        :"r"(dst+len), "r"(src+len), "m"(mul)
2261
    );
2262
}
2263

    
2264
static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len)
2265
{
2266
    x86_reg i = -4*len;
2267
    __asm__ volatile(
2268
        "movss  %3, %%xmm4 \n"
2269
        "shufps $0, %%xmm4, %%xmm4 \n"
2270
        "1: \n"
2271
        "cvtdq2ps   (%2,%0), %%xmm0 \n"
2272
        "cvtdq2ps 16(%2,%0), %%xmm1 \n"
2273
        "mulps    %%xmm4,    %%xmm0 \n"
2274
        "mulps    %%xmm4,    %%xmm1 \n"
2275
        "movaps   %%xmm0,   (%1,%0) \n"
2276
        "movaps   %%xmm1, 16(%1,%0) \n"
2277
        "add $32, %0 \n"
2278
        "jl 1b \n"
2279
        :"+r"(i)
2280
        :"r"(dst+len), "r"(src+len), "m"(mul)
2281
    );
2282
}
2283

    
2284
static void vector_clipf_sse(float *dst, const float *src, float min, float max,
2285
                             int len)
2286
{
2287
    x86_reg i = (len-16)*4;
2288
    __asm__ volatile(
2289
        "movss  %3, %%xmm4 \n"
2290
        "movss  %4, %%xmm5 \n"
2291
        "shufps $0, %%xmm4, %%xmm4 \n"
2292
        "shufps $0, %%xmm5, %%xmm5 \n"
2293
        "1: \n\t"
2294
        "movaps    (%2,%0), %%xmm0 \n\t" // 3/1 on intel
2295
        "movaps  16(%2,%0), %%xmm1 \n\t"
2296
        "movaps  32(%2,%0), %%xmm2 \n\t"
2297
        "movaps  48(%2,%0), %%xmm3 \n\t"
2298
        "maxps      %%xmm4, %%xmm0 \n\t"
2299
        "maxps      %%xmm4, %%xmm1 \n\t"
2300
        "maxps      %%xmm4, %%xmm2 \n\t"
2301
        "maxps      %%xmm4, %%xmm3 \n\t"
2302
        "minps      %%xmm5, %%xmm0 \n\t"
2303
        "minps      %%xmm5, %%xmm1 \n\t"
2304
        "minps      %%xmm5, %%xmm2 \n\t"
2305
        "minps      %%xmm5, %%xmm3 \n\t"
2306
        "movaps  %%xmm0,   (%1,%0) \n\t"
2307
        "movaps  %%xmm1, 16(%1,%0) \n\t"
2308
        "movaps  %%xmm2, 32(%1,%0) \n\t"
2309
        "movaps  %%xmm3, 48(%1,%0) \n\t"
2310
        "sub  $64, %0 \n\t"
2311
        "jge 1b \n\t"
2312
        :"+&r"(i)
2313
        :"r"(dst), "r"(src), "m"(min), "m"(max)
2314
        :"memory"
2315
    );
2316
}
2317

    
2318
static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){
2319
    x86_reg reglen = len;
2320
    // not bit-exact: pf2id uses different rounding than C and SSE
2321
    __asm__ volatile(
2322
        "add        %0          , %0        \n\t"
2323
        "lea         (%2,%0,2)  , %2        \n\t"
2324
        "add        %0          , %1        \n\t"
2325
        "neg        %0                      \n\t"
2326
        "1:                                 \n\t"
2327
        "pf2id       (%2,%0,2)  , %%mm0     \n\t"
2328
        "pf2id      8(%2,%0,2)  , %%mm1     \n\t"
2329
        "pf2id     16(%2,%0,2)  , %%mm2     \n\t"
2330
        "pf2id     24(%2,%0,2)  , %%mm3     \n\t"
2331
        "packssdw   %%mm1       , %%mm0     \n\t"
2332
        "packssdw   %%mm3       , %%mm2     \n\t"
2333
        "movq       %%mm0       ,  (%1,%0)  \n\t"
2334
        "movq       %%mm2       , 8(%1,%0)  \n\t"
2335
        "add        $16         , %0        \n\t"
2336
        " js 1b                             \n\t"
2337
        "femms                              \n\t"
2338
        :"+r"(reglen), "+r"(dst), "+r"(src)
2339
    );
2340
}
2341
static void float_to_int16_sse(int16_t *dst, const float *src, long len){
2342
    x86_reg reglen = len;
2343
    __asm__ volatile(
2344
        "add        %0          , %0        \n\t"
2345
        "lea         (%2,%0,2)  , %2        \n\t"
2346
        "add        %0          , %1        \n\t"
2347
        "neg        %0                      \n\t"
2348
        "1:                                 \n\t"
2349
        "cvtps2pi    (%2,%0,2)  , %%mm0     \n\t"
2350
        "cvtps2pi   8(%2,%0,2)  , %%mm1     \n\t"
2351
        "cvtps2pi  16(%2,%0,2)  , %%mm2     \n\t"
2352
        "cvtps2pi  24(%2,%0,2)  , %%mm3     \n\t"
2353
        "packssdw   %%mm1       , %%mm0     \n\t"
2354
        "packssdw   %%mm3       , %%mm2     \n\t"
2355
        "movq       %%mm0       ,  (%1,%0)  \n\t"
2356
        "movq       %%mm2       , 8(%1,%0)  \n\t"
2357
        "add        $16         , %0        \n\t"
2358
        " js 1b                             \n\t"
2359
        "emms                               \n\t"
2360
        :"+r"(reglen), "+r"(dst), "+r"(src)
2361
    );
2362
}
2363

    
2364
static void float_to_int16_sse2(int16_t *dst, const float *src, long len){
2365
    x86_reg reglen = len;
2366
    __asm__ volatile(
2367
        "add        %0          , %0        \n\t"
2368
        "lea         (%2,%0,2)  , %2        \n\t"
2369
        "add        %0          , %1        \n\t"
2370
        "neg        %0                      \n\t"
2371
        "1:                                 \n\t"
2372
        "cvtps2dq    (%2,%0,2)  , %%xmm0    \n\t"
2373
        "cvtps2dq  16(%2,%0,2)  , %%xmm1    \n\t"
2374
        "packssdw   %%xmm1      , %%xmm0    \n\t"
2375
        "movdqa     %%xmm0      ,  (%1,%0)  \n\t"
2376
        "add        $16         , %0        \n\t"
2377
        " js 1b                             \n\t"
2378
        :"+r"(reglen), "+r"(dst), "+r"(src)
2379
    );
2380
}
2381

    
2382
void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len);
2383
void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len);
2384
void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len);
2385
int32_t ff_scalarproduct_int16_mmx2(int16_t *v1, int16_t *v2, int order, int shift);
2386
int32_t ff_scalarproduct_int16_sse2(int16_t *v1, int16_t *v2, int order, int shift);
2387
int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, int16_t *v2, int16_t *v3, int order, int mul);
2388
int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, int16_t *v2, int16_t *v3, int order, int mul);
2389
int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, int16_t *v2, int16_t *v3, int order, int mul);
2390
void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
2391
int  ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, int w, int left);
2392
int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, int left);
2393
void ff_x264_deblock_v_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
2394
void ff_x264_deblock_h_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
2395
void ff_x264_deblock_h_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
2396
void ff_x264_deblock_v_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta);
2397
void ff_x264_deblock_h_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta);
2398

    
2399
#if HAVE_YASM && ARCH_X86_32
2400
void ff_x264_deblock_v8_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
2401
static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
2402
{
2403
    ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
2404
    ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
2405
}
2406
#elif !HAVE_YASM
2407
#define ff_float_to_int16_interleave6_sse(a,b,c)   float_to_int16_interleave_misc_sse(a,b,c,6)
2408
#define ff_float_to_int16_interleave6_3dnow(a,b,c) float_to_int16_interleave_misc_3dnow(a,b,c,6)
2409
#define ff_float_to_int16_interleave6_3dn2(a,b,c)  float_to_int16_interleave_misc_3dnow(a,b,c,6)
2410
#endif
2411
#define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse
2412

    
2413
#define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \
2414
/* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\
2415
static av_noinline void float_to_int16_interleave_misc_##cpu(int16_t *dst, const float **src, long len, int channels){\
2416
    DECLARE_ALIGNED(16, int16_t, tmp)[len];\
2417
    int i,j,c;\
2418
    for(c=0; c<channels; c++){\
2419
        float_to_int16_##cpu(tmp, src[c], len);\
2420
        for(i=0, j=c; i<len; i++, j+=channels)\
2421
            dst[j] = tmp[i];\
2422
    }\
2423
}\
2424
\
2425
static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, long len, int channels){\
2426
    if(channels==1)\
2427
        float_to_int16_##cpu(dst, src[0], len);\
2428
    else if(channels==2){\
2429
        x86_reg reglen = len; \
2430
        const float *src0 = src[0];\
2431
        const float *src1 = src[1];\
2432
        __asm__ volatile(\
2433
            "shl $2, %0 \n"\
2434
            "add %0, %1 \n"\
2435
            "add %0, %2 \n"\
2436
            "add %0, %3 \n"\
2437
            "neg %0 \n"\
2438
            body\
2439
            :"+r"(reglen), "+r"(dst), "+r"(src0), "+r"(src1)\
2440
        );\
2441
    }else if(channels==6){\
2442
        ff_float_to_int16_interleave6_##cpu(dst, src, len);\
2443
    }else\
2444
        float_to_int16_interleave_misc_##cpu(dst, src, len, channels);\
2445
}
2446

    
2447
FLOAT_TO_INT16_INTERLEAVE(3dnow,
2448
    "1:                         \n"
2449
    "pf2id     (%2,%0), %%mm0   \n"
2450
    "pf2id    8(%2,%0), %%mm1   \n"
2451
    "pf2id     (%3,%0), %%mm2   \n"
2452
    "pf2id    8(%3,%0), %%mm3   \n"
2453
    "packssdw    %%mm1, %%mm0   \n"
2454
    "packssdw    %%mm3, %%mm2   \n"
2455
    "movq        %%mm0, %%mm1   \n"
2456
    "punpcklwd   %%mm2, %%mm0   \n"
2457
    "punpckhwd   %%mm2, %%mm1   \n"
2458
    "movq        %%mm0,  (%1,%0)\n"
2459
    "movq        %%mm1, 8(%1,%0)\n"
2460
    "add $16, %0                \n"
2461
    "js 1b                      \n"
2462
    "femms                      \n"
2463
)
2464

    
2465
FLOAT_TO_INT16_INTERLEAVE(sse,
2466
    "1:                         \n"
2467
    "cvtps2pi  (%2,%0), %%mm0   \n"
2468
    "cvtps2pi 8(%2,%0), %%mm1   \n"
2469
    "cvtps2pi  (%3,%0), %%mm2   \n"
2470
    "cvtps2pi 8(%3,%0), %%mm3   \n"
2471
    "packssdw    %%mm1, %%mm0   \n"
2472
    "packssdw    %%mm3, %%mm2   \n"
2473
    "movq        %%mm0, %%mm1   \n"
2474
    "punpcklwd   %%mm2, %%mm0   \n"
2475
    "punpckhwd   %%mm2, %%mm1   \n"
2476
    "movq        %%mm0,  (%1,%0)\n"
2477
    "movq        %%mm1, 8(%1,%0)\n"
2478
    "add $16, %0                \n"
2479
    "js 1b                      \n"
2480
    "emms                       \n"
2481
)
2482

    
2483
FLOAT_TO_INT16_INTERLEAVE(sse2,
2484
    "1:                         \n"
2485
    "cvtps2dq  (%2,%0), %%xmm0  \n"
2486
    "cvtps2dq  (%3,%0), %%xmm1  \n"
2487
    "packssdw   %%xmm1, %%xmm0  \n"
2488
    "movhlps    %%xmm0, %%xmm1  \n"
2489
    "punpcklwd  %%xmm1, %%xmm0  \n"
2490
    "movdqa     %%xmm0, (%1,%0) \n"
2491
    "add $16, %0                \n"
2492
    "js 1b                      \n"
2493
)
2494

    
2495
static void float_to_int16_interleave_3dn2(int16_t *dst, const float **src, long len, int channels){
2496
    if(channels==6)
2497
        ff_float_to_int16_interleave6_3dn2(dst, src, len);
2498
    else
2499
        float_to_int16_interleave_3dnow(dst, src, len, channels);
2500
}
2501

    
2502

    
2503
void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width);
2504
void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width);
2505
void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
2506
void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
2507
void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2508
                                   int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2509
void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2510
                                  int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2511

    
2512

    
2513
float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
2514

    
2515
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2516
{
2517
    mm_flags = mm_support();
2518

    
2519
    if (avctx->dsp_mask) {
2520
        if (avctx->dsp_mask & FF_MM_FORCE)
2521
            mm_flags |= (avctx->dsp_mask & 0xffff);
2522
        else
2523
            mm_flags &= ~(avctx->dsp_mask & 0xffff);
2524
    }
2525

    
2526
#if 0
2527
    av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2528
    if (mm_flags & FF_MM_MMX)
2529
        av_log(avctx, AV_LOG_INFO, " mmx");
2530
    if (mm_flags & FF_MM_MMX2)
2531
        av_log(avctx, AV_LOG_INFO, " mmx2");
2532
    if (mm_flags & FF_MM_3DNOW)
2533
        av_log(avctx, AV_LOG_INFO, " 3dnow");
2534
    if (mm_flags & FF_MM_SSE)
2535
        av_log(avctx, AV_LOG_INFO, " sse");
2536
    if (mm_flags & FF_MM_SSE2)
2537
        av_log(avctx, AV_LOG_INFO, " sse2");
2538
    av_log(avctx, AV_LOG_INFO, "\n");
2539
#endif
2540

    
2541
    if (mm_flags & FF_MM_MMX) {
2542
        const int idct_algo= avctx->idct_algo;
2543

    
2544
        if(avctx->lowres==0){
2545
            if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2546
                c->idct_put= ff_simple_idct_put_mmx;
2547
                c->idct_add= ff_simple_idct_add_mmx;
2548
                c->idct    = ff_simple_idct_mmx;
2549
                c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2550
#if CONFIG_GPL
2551
            }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2552
                if(mm_flags & FF_MM_MMX2){
2553
                    c->idct_put= ff_libmpeg2mmx2_idct_put;
2554
                    c->idct_add= ff_libmpeg2mmx2_idct_add;
2555
                    c->idct    = ff_mmxext_idct;
2556
                }else{
2557
                    c->idct_put= ff_libmpeg2mmx_idct_put;
2558
                    c->idct_add= ff_libmpeg2mmx_idct_add;
2559
                    c->idct    = ff_mmx_idct;
2560
                }
2561
                c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2562
#endif
2563
            }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&
2564
                     idct_algo==FF_IDCT_VP3){
2565
                if(mm_flags & FF_MM_SSE2){
2566
                    c->idct_put= ff_vp3_idct_put_sse2;
2567
                    c->idct_add= ff_vp3_idct_add_sse2;
2568
                    c->idct    = ff_vp3_idct_sse2;
2569
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2570
                }else{
2571
                    c->idct_put= ff_vp3_idct_put_mmx;
2572
                    c->idct_add= ff_vp3_idct_add_mmx;
2573
                    c->idct    = ff_vp3_idct_mmx;
2574
                    c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2575
                }
2576
            }else if(idct_algo==FF_IDCT_CAVS){
2577
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2578
            }else if(idct_algo==FF_IDCT_XVIDMMX){
2579
                if(mm_flags & FF_MM_SSE2){
2580
                    c->idct_put= ff_idct_xvid_sse2_put;
2581
                    c->idct_add= ff_idct_xvid_sse2_add;
2582
                    c->idct    = ff_idct_xvid_sse2;
2583
                    c->idct_permutation_type= FF_SSE2_IDCT_PERM;
2584
                }else if(mm_flags & FF_MM_MMX2){
2585
                    c->idct_put= ff_idct_xvid_mmx2_put;
2586
                    c->idct_add= ff_idct_xvid_mmx2_add;
2587
                    c->idct    = ff_idct_xvid_mmx2;
2588
                }else{
2589
                    c->idct_put= ff_idct_xvid_mmx_put;
2590
                    c->idct_add= ff_idct_xvid_mmx_add;
2591
                    c->idct    = ff_idct_xvid_mmx;
2592
                }
2593
            }
2594
        }
2595

    
2596
        c->put_pixels_clamped = put_pixels_clamped_mmx;
2597
        c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;
2598
        c->add_pixels_clamped = add_pixels_clamped_mmx;
2599
        c->clear_block  = clear_block_mmx;
2600
        c->clear_blocks = clear_blocks_mmx;
2601
        if ((mm_flags & FF_MM_SSE) &&
2602
            !(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)){
2603
            /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
2604
            c->clear_block  = clear_block_sse;
2605
            c->clear_blocks = clear_blocks_sse;
2606
        }
2607

    
2608
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2609
        c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
2610
        c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
2611
        c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
2612
        c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU
2613

    
2614
        SET_HPEL_FUNCS(put, 0, 16, mmx);
2615
        SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2616
        SET_HPEL_FUNCS(avg, 0, 16, mmx);
2617
        SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2618
        SET_HPEL_FUNCS(put, 1, 8, mmx);
2619
        SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
2620
        SET_HPEL_FUNCS(avg, 1, 8, mmx);
2621
        SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
2622

    
2623
        c->gmc= gmc_mmx;
2624

    
2625
        c->add_bytes= add_bytes_mmx;
2626
        c->add_bytes_l2= add_bytes_l2_mmx;
2627

    
2628
        c->draw_edges = draw_edges_mmx;
2629

    
2630
        if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
2631
            c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2632
            c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2633
        }
2634
        c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_rnd;
2635
        c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;
2636
        c->put_no_rnd_vc1_chroma_pixels_tab[0]= put_vc1_chroma_mc8_mmx_nornd;
2637

    
2638
        c->put_rv40_chroma_pixels_tab[0]= put_rv40_chroma_mc8_mmx;
2639
        c->put_rv40_chroma_pixels_tab[1]= put_rv40_chroma_mc4_mmx;
2640

    
2641
        c->h264_idct_dc_add=
2642
        c->h264_idct_add= ff_h264_idct_add_mmx;
2643
        c->h264_idct8_dc_add=
2644
        c->h264_idct8_add= ff_h264_idct8_add_mmx;
2645

    
2646
        c->h264_idct_add16     = ff_h264_idct_add16_mmx;
2647
        c->h264_idct8_add4     = ff_h264_idct8_add4_mmx;
2648
        c->h264_idct_add8      = ff_h264_idct_add8_mmx;
2649
        c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx;
2650

    
2651
        if (CONFIG_VP6_DECODER) {
2652
            c->vp6_filter_diag4 = ff_vp6_filter_diag4_mmx;
2653
        }
2654

    
2655
        if (mm_flags & FF_MM_MMX2) {
2656
            c->prefetch = prefetch_mmx2;
2657

    
2658
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2659
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2660

    
2661
            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2662
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2663
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2664

    
2665
            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2666
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2667

    
2668
            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2669
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2670
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2671

    
2672
            c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;
2673
            c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;
2674
            c->h264_idct_add16     = ff_h264_idct_add16_mmx2;
2675
            c->h264_idct8_add4     = ff_h264_idct8_add4_mmx2;
2676
            c->h264_idct_add8      = ff_h264_idct_add8_mmx2;
2677
            c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2;
2678

    
2679
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2680
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2681
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2682
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2683
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2684
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2685
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2686

    
2687
                if (CONFIG_VP3_DECODER) {
2688
                    c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2;
2689
                    c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;
2690
                }
2691
            }
2692

    
2693
#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2694
            c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \
2695
            c->PFX ## _pixels_tab[IDX][ 1] = PFX ## SIZE ## _mc10_ ## CPU; \
2696
            c->PFX ## _pixels_tab[IDX][ 2] = PFX ## SIZE ## _mc20_ ## CPU; \
2697
            c->PFX ## _pixels_tab[IDX][ 3] = PFX ## SIZE ## _mc30_ ## CPU; \
2698
            c->PFX ## _pixels_tab[IDX][ 4] = PFX ## SIZE ## _mc01_ ## CPU; \
2699
            c->PFX ## _pixels_tab[IDX][ 5] = PFX ## SIZE ## _mc11_ ## CPU; \
2700
            c->PFX ## _pixels_tab[IDX][ 6] = PFX ## SIZE ## _mc21_ ## CPU; \
2701
            c->PFX ## _pixels_tab[IDX][ 7] = PFX ## SIZE ## _mc31_ ## CPU; \
2702
            c->PFX ## _pixels_tab[IDX][ 8] = PFX ## SIZE ## _mc02_ ## CPU; \
2703
            c->PFX ## _pixels_tab[IDX][ 9] = PFX ## SIZE ## _mc12_ ## CPU; \
2704
            c->PFX ## _pixels_tab[IDX][10] = PFX ## SIZE ## _mc22_ ## CPU; \
2705
            c->PFX ## _pixels_tab[IDX][11] = PFX ## SIZE ## _mc32_ ## CPU; \
2706
            c->PFX ## _pixels_tab[IDX][12] = PFX ## SIZE ## _mc03_ ## CPU; \
2707
            c->PFX ## _pixels_tab[IDX][13] = PFX ## SIZE ## _mc13_ ## CPU; \
2708
            c->PFX ## _pixels_tab[IDX][14] = PFX ## SIZE ## _mc23_ ## CPU; \
2709
            c->PFX ## _pixels_tab[IDX][15] = PFX ## SIZE ## _mc33_ ## CPU
2710

    
2711
            SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2);
2712
            SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2);
2713
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2);
2714
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2);
2715
            SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2);
2716
            SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2);
2717

    
2718
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2);
2719
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2);
2720
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2);
2721
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2);
2722
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2);
2723
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2);
2724

    
2725
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2);
2726
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2);
2727
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2);
2728
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2);
2729

    
2730
            c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_mmx2;
2731
            c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_mmx2;
2732

    
2733
            c->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_vc1_chroma_mc8_mmx2_nornd;
2734

    
2735
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2_rnd;
2736
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;
2737
            c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;
2738
            c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;
2739
            c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
2740
            c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
2741
            c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
2742
            c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;
2743
            c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;
2744
            c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;
2745
            c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
2746

    
2747
            c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
2748
            c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
2749
            c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
2750
            c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
2751
            c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
2752
            c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
2753
            c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
2754
            c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
2755

    
2756
            c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
2757
            c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
2758
            c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
2759
            c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
2760
            c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
2761
            c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
2762
            c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
2763
            c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
2764

    
2765
#if HAVE_YASM
2766
            c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
2767
#endif
2768
#if HAVE_7REGS && HAVE_TEN_OPERANDS
2769
            if( mm_flags&FF_MM_3DNOW )
2770
                c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
2771
#endif
2772

    
2773
            if (CONFIG_CAVS_DECODER)
2774
                ff_cavsdsp_init_mmx2(c, avctx);
2775

    
2776
            if (CONFIG_VC1_DECODER)
2777
                ff_vc1dsp_init_mmx(c, avctx);
2778

    
2779
            c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2;
2780
        } else if (mm_flags & FF_MM_3DNOW) {
2781
            c->prefetch = prefetch_3dnow;
2782

    
2783
            c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
2784
            c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
2785

    
2786
            c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
2787
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
2788
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
2789

    
2790
            c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
2791
            c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
2792

    
2793
            c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
2794
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
2795
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
2796

    
2797
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2798
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
2799
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
2800
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
2801
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
2802
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
2803
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
2804
            }
2805

    
2806
            SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow);
2807
            SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow);
2808
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow);
2809
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow);
2810
            SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow);
2811
            SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow);
2812

    
2813
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow);
2814
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow);
2815
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow);
2816
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow);
2817
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow);
2818
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow);
2819

    
2820
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow);
2821
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow);
2822
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow);
2823
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow);
2824

    
2825
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow_rnd;
2826
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;
2827

    
2828
            c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_3dnow;
2829
            c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_3dnow;
2830

    
2831
            if (CONFIG_CAVS_DECODER)
2832
                ff_cavsdsp_init_3dnow(c, avctx);
2833
        }
2834

    
2835

    
2836
#define H264_QPEL_FUNCS(x, y, CPU)\
2837
            c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\
2838
            c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\
2839
            c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\
2840
            c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;
2841
        if((mm_flags & FF_MM_SSE2) && !(mm_flags & FF_MM_3DNOW)){
2842
            // these functions are slower than mmx on AMD, but faster on Intel
2843
            c->put_pixels_tab[0][0] = put_pixels16_sse2;
2844
            c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
2845
            H264_QPEL_FUNCS(0, 0, sse2);
2846
        }
2847
        if(mm_flags & FF_MM_SSE2){
2848
            c->h264_idct8_add = ff_h264_idct8_add_sse2;
2849
            c->h264_idct8_add4= ff_h264_idct8_add4_sse2;
2850

    
2851
            H264_QPEL_FUNCS(0, 1, sse2);
2852
            H264_QPEL_FUNCS(0, 2, sse2);
2853
            H264_QPEL_FUNCS(0, 3, sse2);
2854
            H264_QPEL_FUNCS(1, 1, sse2);
2855
            H264_QPEL_FUNCS(1, 2, sse2);
2856
            H264_QPEL_FUNCS(1, 3, sse2);
2857
            H264_QPEL_FUNCS(2, 1, sse2);
2858
            H264_QPEL_FUNCS(2, 2, sse2);
2859
            H264_QPEL_FUNCS(2, 3, sse2);
2860
            H264_QPEL_FUNCS(3, 1, sse2);
2861
            H264_QPEL_FUNCS(3, 2, sse2);
2862
            H264_QPEL_FUNCS(3, 3, sse2);
2863

    
2864
            if (CONFIG_VP6_DECODER) {
2865
                c->vp6_filter_diag4 = ff_vp6_filter_diag4_sse2;
2866
            }
2867
        }
2868
#if HAVE_SSSE3
2869
        if(mm_flags & FF_MM_SSSE3){
2870
            H264_QPEL_FUNCS(1, 0, ssse3);
2871
            H264_QPEL_FUNCS(1, 1, ssse3);
2872
            H264_QPEL_FUNCS(1, 2, ssse3);
2873
            H264_QPEL_FUNCS(1, 3, ssse3);
2874
            H264_QPEL_FUNCS(2, 0, ssse3);
2875
            H264_QPEL_FUNCS(2, 1, ssse3);
2876
            H264_QPEL_FUNCS(2, 2, ssse3);
2877
            H264_QPEL_FUNCS(2, 3, ssse3);
2878
            H264_QPEL_FUNCS(3, 0, ssse3);
2879
            H264_QPEL_FUNCS(3, 1, ssse3);
2880
            H264_QPEL_FUNCS(3, 2, ssse3);
2881
            H264_QPEL_FUNCS(3, 3, ssse3);
2882
            c->put_no_rnd_vc1_chroma_pixels_tab[0]= put_vc1_chroma_mc8_ssse3_nornd;
2883
            c->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_vc1_chroma_mc8_ssse3_nornd;
2884
            c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_rnd;
2885
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_ssse3_rnd;
2886
            c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_ssse3;
2887
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3;
2888
            c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
2889
#if HAVE_YASM
2890
            c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
2891
            if (mm_flags & FF_MM_SSE4) // not really sse4, just slow on Conroe
2892
                c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
2893
#endif
2894
        }
2895
#endif
2896

    
2897
#if CONFIG_GPL && HAVE_YASM
2898
        if (mm_flags & FF_MM_MMX2){
2899
#if ARCH_X86_32
2900
            c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext;
2901
            c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext;
2902
#endif
2903
            if( mm_flags&FF_MM_SSE2 ){
2904
#if ARCH_X86_64 || !defined(__ICC) || __ICC > 1110
2905
                c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2;
2906
                c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2;
2907
                c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2;
2908
                c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2;
2909
#endif
2910
                c->h264_idct_add16 = ff_h264_idct_add16_sse2;
2911
                c->h264_idct_add8  = ff_h264_idct_add8_sse2;
2912
                c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2;
2913
            }
2914
        }
2915
#endif
2916

    
2917
#if CONFIG_SNOW_DECODER
2918
        if(mm_flags & FF_MM_SSE2 & 0){
2919
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
2920
#if HAVE_7REGS
2921
            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
2922
#endif
2923
            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
2924
        }
2925
        else{
2926
            if(mm_flags & FF_MM_MMX2){
2927
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
2928
#if HAVE_7REGS
2929
            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
2930
#endif
2931
            }
2932
            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
2933
        }
2934
#endif
2935

    
2936
        if(mm_flags & FF_MM_3DNOW){
2937
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
2938
            c->vector_fmul = vector_fmul_3dnow;
2939
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2940
                c->float_to_int16 = float_to_int16_3dnow;
2941
                c->float_to_int16_interleave = float_to_int16_interleave_3dnow;
2942
            }
2943
        }
2944
        if(mm_flags & FF_MM_3DNOWEXT){
2945
            c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
2946
            c->vector_fmul_window = vector_fmul_window_3dnow2;
2947
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2948
                c->float_to_int16_interleave = float_to_int16_interleave_3dn2;
2949
            }
2950
        }
2951
        if(mm_flags & FF_MM_MMX2){
2952
#if HAVE_YASM
2953
            c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;
2954
            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;
2955
#endif
2956
        }
2957
        if(mm_flags & FF_MM_SSE){
2958
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
2959
            c->ac3_downmix = ac3_downmix_sse;
2960
            c->vector_fmul = vector_fmul_sse;
2961
            c->vector_fmul_reverse = vector_fmul_reverse_sse;
2962
            c->vector_fmul_add = vector_fmul_add_sse;
2963
            c->vector_fmul_window = vector_fmul_window_sse;
2964
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse;
2965
            c->vector_clipf = vector_clipf_sse;
2966
            c->float_to_int16 = float_to_int16_sse;
2967
            c->float_to_int16_interleave = float_to_int16_interleave_sse;
2968
#if HAVE_YASM
2969
            c->scalarproduct_float = ff_scalarproduct_float_sse;
2970
#endif
2971
        }
2972
        if(mm_flags & FF_MM_3DNOW)
2973
            c->vector_fmul_add = vector_fmul_add_3dnow; // faster than sse
2974
        if(mm_flags & FF_MM_SSE2){
2975
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2;
2976
            c->float_to_int16 = float_to_int16_sse2;
2977
            c->float_to_int16_interleave = float_to_int16_interleave_sse2;
2978
#if HAVE_YASM
2979
            c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
2980
            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
2981
#endif
2982
        }
2983
        if((mm_flags & FF_MM_SSSE3) && !(mm_flags & (FF_MM_SSE42|FF_MM_3DNOW)) && HAVE_YASM) // cachesplit
2984
            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
2985
    }
2986

    
2987
    if (CONFIG_ENCODERS)
2988
        dsputilenc_init_mmx(c, avctx);
2989

    
2990
#if 0
2991
    // for speed testing
2992
    get_pixels = just_return;
2993
    put_pixels_clamped = just_return;
2994
    add_pixels_clamped = just_return;
2995

2996
    pix_abs16x16 = just_return;
2997
    pix_abs16x16_x2 = just_return;
2998
    pix_abs16x16_y2 = just_return;
2999
    pix_abs16x16_xy2 = just_return;
3000

3001
    put_pixels_tab[0] = just_return;
3002
    put_pixels_tab[1] = just_return;
3003
    put_pixels_tab[2] = just_return;
3004
    put_pixels_tab[3] = just_return;
3005

3006
    put_no_rnd_pixels_tab[0] = just_return;
3007
    put_no_rnd_pixels_tab[1] = just_return;
3008
    put_no_rnd_pixels_tab[2] = just_return;
3009
    put_no_rnd_pixels_tab[3] = just_return;
3010

3011
    avg_pixels_tab[0] = just_return;
3012
    avg_pixels_tab[1] = just_return;
3013
    avg_pixels_tab[2] = just_return;
3014
    avg_pixels_tab[3] = just_return;
3015

3016
    avg_no_rnd_pixels_tab[0] = just_return;
3017
    avg_no_rnd_pixels_tab[1] = just_return;
3018
    avg_no_rnd_pixels_tab[2] = just_return;
3019
    avg_no_rnd_pixels_tab[3] = just_return;
3020

3021
    //av_fdct = just_return;
3022
    //ff_idct = just_return;
3023
#endif
3024
}