Statistics
| Branch: | Revision:

ffmpeg / libavcodec / x86 / dsputil_mmx.c @ 50e23ae9

History | View | Annotate | Download (127 KB)

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

    
25
#include "libavutil/x86_cpu.h"
26
#include "libavcodec/dsputil.h"
27
#include "libavcodec/h263.h"
28
#include "libavcodec/mpegvideo.h"
29
#include "libavcodec/simple_idct.h"
30
#include "dsputil_mmx.h"
31
#include "vp3dsp_mmx.h"
32
#include "vp3dsp_sse2.h"
33
#include "vp6dsp_mmx.h"
34
#include "vp6dsp_sse2.h"
35
#include "idct_xvid.h"
36

    
37
//#undef NDEBUG
38
//#include <assert.h>
39

    
40
int mm_flags; /* multimedia extension flags */
41

    
42
/* pixel operations */
43
DECLARE_ALIGNED_8 (const uint64_t, ff_bone) = 0x0101010101010101ULL;
44
DECLARE_ALIGNED_8 (const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
45

    
46
DECLARE_ALIGNED_16(const uint64_t, ff_pdw_80000000[2]) =
47
{0x8000000080000000ULL, 0x8000000080000000ULL};
48

    
49
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_3  ) = 0x0003000300030003ULL;
50
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_4  ) = 0x0004000400040004ULL;
51
DECLARE_ALIGNED_16(const xmm_reg,  ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
52
DECLARE_ALIGNED_16(const xmm_reg,  ff_pw_8  ) = {0x0008000800080008ULL, 0x0008000800080008ULL};
53
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
54
DECLARE_ALIGNED_16(const xmm_reg,  ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
55
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
56
DECLARE_ALIGNED_16(const xmm_reg,  ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
57
DECLARE_ALIGNED_16(const xmm_reg,  ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
58
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
59
DECLARE_ALIGNED_16(const xmm_reg,  ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL};
60
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
61
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
62
DECLARE_ALIGNED_8 (const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
63

    
64
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1  ) = 0x0101010101010101ULL;
65
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3  ) = 0x0303030303030303ULL;
66
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
67
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
68
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
69
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
70
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL;
71
DECLARE_ALIGNED_8 (const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
72

    
73
DECLARE_ALIGNED_16(const double, ff_pd_1[2]) = { 1.0, 1.0 };
74
DECLARE_ALIGNED_16(const double, ff_pd_2[2]) = { 2.0, 2.0 };
75

    
76
#define JUMPALIGN() __asm__ volatile (ASMALIGN(3)::)
77
#define MOVQ_ZERO(regd)  __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
78

    
79
#define MOVQ_BFE(regd) \
80
    __asm__ volatile ( \
81
    "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
82
    "paddb %%" #regd ", %%" #regd " \n\t" ::)
83

    
84
#ifndef PIC
85
#define MOVQ_BONE(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
86
#define MOVQ_WTWO(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
87
#else
88
// for shared library it's better to use this way for accessing constants
89
// pcmpeqd -> -1
90
#define MOVQ_BONE(regd) \
91
    __asm__ volatile ( \
92
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
93
    "psrlw $15, %%" #regd " \n\t" \
94
    "packuswb %%" #regd ", %%" #regd " \n\t" ::)
95

    
96
#define MOVQ_WTWO(regd) \
97
    __asm__ volatile ( \
98
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
99
    "psrlw $15, %%" #regd " \n\t" \
100
    "psllw $1, %%" #regd " \n\t"::)
101

    
102
#endif
103

    
104
// using regr as temporary and for the output result
105
// first argument is unmodifed and second is trashed
106
// regfe is supposed to contain 0xfefefefefefefefe
107
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
108
    "movq " #rega ", " #regr "  \n\t"\
109
    "pand " #regb ", " #regr "  \n\t"\
110
    "pxor " #rega ", " #regb "  \n\t"\
111
    "pand " #regfe "," #regb "  \n\t"\
112
    "psrlq $1, " #regb "        \n\t"\
113
    "paddb " #regb ", " #regr " \n\t"
114

    
115
#define PAVGB_MMX(rega, regb, regr, regfe) \
116
    "movq " #rega ", " #regr "  \n\t"\
117
    "por  " #regb ", " #regr "  \n\t"\
118
    "pxor " #rega ", " #regb "  \n\t"\
119
    "pand " #regfe "," #regb "  \n\t"\
120
    "psrlq $1, " #regb "        \n\t"\
121
    "psubb " #regb ", " #regr " \n\t"
122

    
123
// mm6 is supposed to contain 0xfefefefefefefefe
124
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \
125
    "movq " #rega ", " #regr "  \n\t"\
126
    "movq " #regc ", " #regp "  \n\t"\
127
    "pand " #regb ", " #regr "  \n\t"\
128
    "pand " #regd ", " #regp "  \n\t"\
129
    "pxor " #rega ", " #regb "  \n\t"\
130
    "pxor " #regc ", " #regd "  \n\t"\
131
    "pand %%mm6, " #regb "      \n\t"\
132
    "pand %%mm6, " #regd "      \n\t"\
133
    "psrlq $1, " #regb "        \n\t"\
134
    "psrlq $1, " #regd "        \n\t"\
135
    "paddb " #regb ", " #regr " \n\t"\
136
    "paddb " #regd ", " #regp " \n\t"
137

    
138
#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
139
    "movq " #rega ", " #regr "  \n\t"\
140
    "movq " #regc ", " #regp "  \n\t"\
141
    "por  " #regb ", " #regr "  \n\t"\
142
    "por  " #regd ", " #regp "  \n\t"\
143
    "pxor " #rega ", " #regb "  \n\t"\
144
    "pxor " #regc ", " #regd "  \n\t"\
145
    "pand %%mm6, " #regb "      \n\t"\
146
    "pand %%mm6, " #regd "      \n\t"\
147
    "psrlq $1, " #regd "        \n\t"\
148
    "psrlq $1, " #regb "        \n\t"\
149
    "psubb " #regb ", " #regr " \n\t"\
150
    "psubb " #regd ", " #regp " \n\t"
151

    
152
/***********************************/
153
/* MMX no rounding */
154
#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
155
#define SET_RND  MOVQ_WONE
156
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
157
#define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
158
#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
159

    
160
#include "dsputil_mmx_rnd_template.c"
161

    
162
#undef DEF
163
#undef SET_RND
164
#undef PAVGBP
165
#undef PAVGB
166
/***********************************/
167
/* MMX rounding */
168

    
169
#define DEF(x, y) x ## _ ## y ##_mmx
170
#define SET_RND  MOVQ_WTWO
171
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
172
#define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)
173

    
174
#include "dsputil_mmx_rnd_template.c"
175

    
176
#undef DEF
177
#undef SET_RND
178
#undef PAVGBP
179
#undef PAVGB
180
#undef OP_AVG
181

    
182
/***********************************/
183
/* 3Dnow specific */
184

    
185
#define DEF(x) x ## _3dnow
186
#define PAVGB "pavgusb"
187
#define OP_AVG PAVGB
188

    
189
#include "dsputil_mmx_avg_template.c"
190

    
191
#undef DEF
192
#undef PAVGB
193
#undef OP_AVG
194

    
195
/***********************************/
196
/* MMX2 specific */
197

    
198
#define DEF(x) x ## _mmx2
199

    
200
/* Introduced only in MMX2 set */
201
#define PAVGB "pavgb"
202
#define OP_AVG PAVGB
203

    
204
#include "dsputil_mmx_avg_template.c"
205

    
206
#undef DEF
207
#undef PAVGB
208
#undef OP_AVG
209

    
210
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
211
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
212
#define put_pixels16_mmx2 put_pixels16_mmx
213
#define put_pixels8_mmx2 put_pixels8_mmx
214
#define put_pixels4_mmx2 put_pixels4_mmx
215
#define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
216
#define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
217
#define put_pixels16_3dnow put_pixels16_mmx
218
#define put_pixels8_3dnow put_pixels8_mmx
219
#define put_pixels4_3dnow put_pixels4_mmx
220
#define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
221
#define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
222

    
223
/***********************************/
224
/* standard MMX */
225

    
226
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
227
{
228
    const DCTELEM *p;
229
    uint8_t *pix;
230

    
231
    /* read the pixels */
232
    p = block;
233
    pix = pixels;
234
    /* unrolled loop */
235
        __asm__ volatile(
236
                "movq   %3, %%mm0               \n\t"
237
                "movq   8%3, %%mm1              \n\t"
238
                "movq   16%3, %%mm2             \n\t"
239
                "movq   24%3, %%mm3             \n\t"
240
                "movq   32%3, %%mm4             \n\t"
241
                "movq   40%3, %%mm5             \n\t"
242
                "movq   48%3, %%mm6             \n\t"
243
                "movq   56%3, %%mm7             \n\t"
244
                "packuswb %%mm1, %%mm0          \n\t"
245
                "packuswb %%mm3, %%mm2          \n\t"
246
                "packuswb %%mm5, %%mm4          \n\t"
247
                "packuswb %%mm7, %%mm6          \n\t"
248
                "movq   %%mm0, (%0)             \n\t"
249
                "movq   %%mm2, (%0, %1)         \n\t"
250
                "movq   %%mm4, (%0, %1, 2)      \n\t"
251
                "movq   %%mm6, (%0, %2)         \n\t"
252
                ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
253
                :"memory");
254
        pix += line_size*4;
255
        p += 32;
256

    
257
    // if here would be an exact copy of the code above
258
    // compiler would generate some very strange code
259
    // thus using "r"
260
    __asm__ volatile(
261
            "movq       (%3), %%mm0             \n\t"
262
            "movq       8(%3), %%mm1            \n\t"
263
            "movq       16(%3), %%mm2           \n\t"
264
            "movq       24(%3), %%mm3           \n\t"
265
            "movq       32(%3), %%mm4           \n\t"
266
            "movq       40(%3), %%mm5           \n\t"
267
            "movq       48(%3), %%mm6           \n\t"
268
            "movq       56(%3), %%mm7           \n\t"
269
            "packuswb %%mm1, %%mm0              \n\t"
270
            "packuswb %%mm3, %%mm2              \n\t"
271
            "packuswb %%mm5, %%mm4              \n\t"
272
            "packuswb %%mm7, %%mm6              \n\t"
273
            "movq       %%mm0, (%0)             \n\t"
274
            "movq       %%mm2, (%0, %1)         \n\t"
275
            "movq       %%mm4, (%0, %1, 2)      \n\t"
276
            "movq       %%mm6, (%0, %2)         \n\t"
277
            ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
278
            :"memory");
279
}
280

    
281
DECLARE_ASM_CONST(8, uint8_t, ff_vector128[8]) =
282
  { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
283

    
284
#define put_signed_pixels_clamped_mmx_half(off) \
285
            "movq    "#off"(%2), %%mm1          \n\t"\
286
            "movq 16+"#off"(%2), %%mm2          \n\t"\
287
            "movq 32+"#off"(%2), %%mm3          \n\t"\
288
            "movq 48+"#off"(%2), %%mm4          \n\t"\
289
            "packsswb  8+"#off"(%2), %%mm1      \n\t"\
290
            "packsswb 24+"#off"(%2), %%mm2      \n\t"\
291
            "packsswb 40+"#off"(%2), %%mm3      \n\t"\
292
            "packsswb 56+"#off"(%2), %%mm4      \n\t"\
293
            "paddb %%mm0, %%mm1                 \n\t"\
294
            "paddb %%mm0, %%mm2                 \n\t"\
295
            "paddb %%mm0, %%mm3                 \n\t"\
296
            "paddb %%mm0, %%mm4                 \n\t"\
297
            "movq %%mm1, (%0)                   \n\t"\
298
            "movq %%mm2, (%0, %3)               \n\t"\
299
            "movq %%mm3, (%0, %3, 2)            \n\t"\
300
            "movq %%mm4, (%0, %1)               \n\t"
301

    
302
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
303
{
304
    x86_reg line_skip = line_size;
305
    x86_reg line_skip3;
306

    
307
    __asm__ volatile (
308
            "movq "MANGLE(ff_vector128)", %%mm0 \n\t"
309
            "lea (%3, %3, 2), %1                \n\t"
310
            put_signed_pixels_clamped_mmx_half(0)
311
            "lea (%0, %3, 4), %0                \n\t"
312
            put_signed_pixels_clamped_mmx_half(64)
313
            :"+&r" (pixels), "=&r" (line_skip3)
314
            :"r" (block), "r"(line_skip)
315
            :"memory");
316
}
317

    
318
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
319
{
320
    const DCTELEM *p;
321
    uint8_t *pix;
322
    int i;
323

    
324
    /* read the pixels */
325
    p = block;
326
    pix = pixels;
327
    MOVQ_ZERO(mm7);
328
    i = 4;
329
    do {
330
        __asm__ volatile(
331
                "movq   (%2), %%mm0     \n\t"
332
                "movq   8(%2), %%mm1    \n\t"
333
                "movq   16(%2), %%mm2   \n\t"
334
                "movq   24(%2), %%mm3   \n\t"
335
                "movq   %0, %%mm4       \n\t"
336
                "movq   %1, %%mm6       \n\t"
337
                "movq   %%mm4, %%mm5    \n\t"
338
                "punpcklbw %%mm7, %%mm4 \n\t"
339
                "punpckhbw %%mm7, %%mm5 \n\t"
340
                "paddsw %%mm4, %%mm0    \n\t"
341
                "paddsw %%mm5, %%mm1    \n\t"
342
                "movq   %%mm6, %%mm5    \n\t"
343
                "punpcklbw %%mm7, %%mm6 \n\t"
344
                "punpckhbw %%mm7, %%mm5 \n\t"
345
                "paddsw %%mm6, %%mm2    \n\t"
346
                "paddsw %%mm5, %%mm3    \n\t"
347
                "packuswb %%mm1, %%mm0  \n\t"
348
                "packuswb %%mm3, %%mm2  \n\t"
349
                "movq   %%mm0, %0       \n\t"
350
                "movq   %%mm2, %1       \n\t"
351
                :"+m"(*pix), "+m"(*(pix+line_size))
352
                :"r"(p)
353
                :"memory");
354
        pix += line_size*2;
355
        p += 16;
356
    } while (--i);
357
}
358

    
359
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
360
{
361
    __asm__ volatile(
362
         "lea (%3, %3), %%"REG_a"       \n\t"
363
         ASMALIGN(3)
364
         "1:                            \n\t"
365
         "movd (%1), %%mm0              \n\t"
366
         "movd (%1, %3), %%mm1          \n\t"
367
         "movd %%mm0, (%2)              \n\t"
368
         "movd %%mm1, (%2, %3)          \n\t"
369
         "add %%"REG_a", %1             \n\t"
370
         "add %%"REG_a", %2             \n\t"
371
         "movd (%1), %%mm0              \n\t"
372
         "movd (%1, %3), %%mm1          \n\t"
373
         "movd %%mm0, (%2)              \n\t"
374
         "movd %%mm1, (%2, %3)          \n\t"
375
         "add %%"REG_a", %1             \n\t"
376
         "add %%"REG_a", %2             \n\t"
377
         "subl $4, %0                   \n\t"
378
         "jnz 1b                        \n\t"
379
         : "+g"(h), "+r" (pixels),  "+r" (block)
380
         : "r"((x86_reg)line_size)
381
         : "%"REG_a, "memory"
382
        );
383
}
384

    
385
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
386
{
387
    __asm__ volatile(
388
         "lea (%3, %3), %%"REG_a"       \n\t"
389
         ASMALIGN(3)
390
         "1:                            \n\t"
391
         "movq (%1), %%mm0              \n\t"
392
         "movq (%1, %3), %%mm1          \n\t"
393
         "movq %%mm0, (%2)              \n\t"
394
         "movq %%mm1, (%2, %3)          \n\t"
395
         "add %%"REG_a", %1             \n\t"
396
         "add %%"REG_a", %2             \n\t"
397
         "movq (%1), %%mm0              \n\t"
398
         "movq (%1, %3), %%mm1          \n\t"
399
         "movq %%mm0, (%2)              \n\t"
400
         "movq %%mm1, (%2, %3)          \n\t"
401
         "add %%"REG_a", %1             \n\t"
402
         "add %%"REG_a", %2             \n\t"
403
         "subl $4, %0                   \n\t"
404
         "jnz 1b                        \n\t"
405
         : "+g"(h), "+r" (pixels),  "+r" (block)
406
         : "r"((x86_reg)line_size)
407
         : "%"REG_a, "memory"
408
        );
409
}
410

    
411
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
412
{
413
    __asm__ volatile(
414
         "lea (%3, %3), %%"REG_a"       \n\t"
415
         ASMALIGN(3)
416
         "1:                            \n\t"
417
         "movq (%1), %%mm0              \n\t"
418
         "movq 8(%1), %%mm4             \n\t"
419
         "movq (%1, %3), %%mm1          \n\t"
420
         "movq 8(%1, %3), %%mm5         \n\t"
421
         "movq %%mm0, (%2)              \n\t"
422
         "movq %%mm4, 8(%2)             \n\t"
423
         "movq %%mm1, (%2, %3)          \n\t"
424
         "movq %%mm5, 8(%2, %3)         \n\t"
425
         "add %%"REG_a", %1             \n\t"
426
         "add %%"REG_a", %2             \n\t"
427
         "movq (%1), %%mm0              \n\t"
428
         "movq 8(%1), %%mm4             \n\t"
429
         "movq (%1, %3), %%mm1          \n\t"
430
         "movq 8(%1, %3), %%mm5         \n\t"
431
         "movq %%mm0, (%2)              \n\t"
432
         "movq %%mm4, 8(%2)             \n\t"
433
         "movq %%mm1, (%2, %3)          \n\t"
434
         "movq %%mm5, 8(%2, %3)         \n\t"
435
         "add %%"REG_a", %1             \n\t"
436
         "add %%"REG_a", %2             \n\t"
437
         "subl $4, %0                   \n\t"
438
         "jnz 1b                        \n\t"
439
         : "+g"(h), "+r" (pixels),  "+r" (block)
440
         : "r"((x86_reg)line_size)
441
         : "%"REG_a, "memory"
442
        );
443
}
444

    
445
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
446
{
447
    __asm__ volatile(
448
         "1:                            \n\t"
449
         "movdqu (%1), %%xmm0           \n\t"
450
         "movdqu (%1,%3), %%xmm1        \n\t"
451
         "movdqu (%1,%3,2), %%xmm2      \n\t"
452
         "movdqu (%1,%4), %%xmm3        \n\t"
453
         "movdqa %%xmm0, (%2)           \n\t"
454
         "movdqa %%xmm1, (%2,%3)        \n\t"
455
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
456
         "movdqa %%xmm3, (%2,%4)        \n\t"
457
         "subl $4, %0                   \n\t"
458
         "lea (%1,%3,4), %1             \n\t"
459
         "lea (%2,%3,4), %2             \n\t"
460
         "jnz 1b                        \n\t"
461
         : "+g"(h), "+r" (pixels),  "+r" (block)
462
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
463
         : "memory"
464
        );
465
}
466

    
467
static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
468
{
469
    __asm__ volatile(
470
         "1:                            \n\t"
471
         "movdqu (%1), %%xmm0           \n\t"
472
         "movdqu (%1,%3), %%xmm1        \n\t"
473
         "movdqu (%1,%3,2), %%xmm2      \n\t"
474
         "movdqu (%1,%4), %%xmm3        \n\t"
475
         "pavgb  (%2), %%xmm0           \n\t"
476
         "pavgb  (%2,%3), %%xmm1        \n\t"
477
         "pavgb  (%2,%3,2), %%xmm2      \n\t"
478
         "pavgb  (%2,%4), %%xmm3        \n\t"
479
         "movdqa %%xmm0, (%2)           \n\t"
480
         "movdqa %%xmm1, (%2,%3)        \n\t"
481
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
482
         "movdqa %%xmm3, (%2,%4)        \n\t"
483
         "subl $4, %0                   \n\t"
484
         "lea (%1,%3,4), %1             \n\t"
485
         "lea (%2,%3,4), %2             \n\t"
486
         "jnz 1b                        \n\t"
487
         : "+g"(h), "+r" (pixels),  "+r" (block)
488
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
489
         : "memory"
490
        );
491
}
492

    
493
#define CLEAR_BLOCKS(name,n) \
494
static void name(DCTELEM *blocks)\
495
{\
496
    __asm__ volatile(\
497
                "pxor %%mm7, %%mm7              \n\t"\
498
                "mov     %1, %%"REG_a"          \n\t"\
499
                "1:                             \n\t"\
500
                "movq %%mm7, (%0, %%"REG_a")    \n\t"\
501
                "movq %%mm7, 8(%0, %%"REG_a")   \n\t"\
502
                "movq %%mm7, 16(%0, %%"REG_a")  \n\t"\
503
                "movq %%mm7, 24(%0, %%"REG_a")  \n\t"\
504
                "add $32, %%"REG_a"             \n\t"\
505
                " js 1b                         \n\t"\
506
                : : "r" (((uint8_t *)blocks)+128*n),\
507
                    "i" (-128*n)\
508
                : "%"REG_a\
509
        );\
510
}
511
CLEAR_BLOCKS(clear_blocks_mmx, 6)
512
CLEAR_BLOCKS(clear_block_mmx, 1)
513

    
514
static void clear_block_sse(DCTELEM *block)
515
{
516
    __asm__ volatile(
517
        "xorps  %%xmm0, %%xmm0  \n"
518
        "movaps %%xmm0,    (%0) \n"
519
        "movaps %%xmm0,  16(%0) \n"
520
        "movaps %%xmm0,  32(%0) \n"
521
        "movaps %%xmm0,  48(%0) \n"
522
        "movaps %%xmm0,  64(%0) \n"
523
        "movaps %%xmm0,  80(%0) \n"
524
        "movaps %%xmm0,  96(%0) \n"
525
        "movaps %%xmm0, 112(%0) \n"
526
        :: "r"(block)
527
        : "memory"
528
    );
529
}
530

    
531
static void clear_blocks_sse(DCTELEM *blocks)
532
{\
533
    __asm__ volatile(
534
        "xorps  %%xmm0, %%xmm0  \n"
535
        "mov     %1, %%"REG_a"  \n"
536
        "1:                     \n"
537
        "movaps %%xmm0,    (%0, %%"REG_a") \n"
538
        "movaps %%xmm0,  16(%0, %%"REG_a") \n"
539
        "movaps %%xmm0,  32(%0, %%"REG_a") \n"
540
        "movaps %%xmm0,  48(%0, %%"REG_a") \n"
541
        "movaps %%xmm0,  64(%0, %%"REG_a") \n"
542
        "movaps %%xmm0,  80(%0, %%"REG_a") \n"
543
        "movaps %%xmm0,  96(%0, %%"REG_a") \n"
544
        "movaps %%xmm0, 112(%0, %%"REG_a") \n"
545
        "add $128, %%"REG_a"    \n"
546
        " js 1b                 \n"
547
        : : "r" (((uint8_t *)blocks)+128*6),
548
            "i" (-128*6)
549
        : "%"REG_a
550
    );
551
}
552

    
553
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
554
    x86_reg i=0;
555
    __asm__ volatile(
556
        "jmp 2f                         \n\t"
557
        "1:                             \n\t"
558
        "movq  (%1, %0), %%mm0          \n\t"
559
        "movq  (%2, %0), %%mm1          \n\t"
560
        "paddb %%mm0, %%mm1             \n\t"
561
        "movq %%mm1, (%2, %0)           \n\t"
562
        "movq 8(%1, %0), %%mm0          \n\t"
563
        "movq 8(%2, %0), %%mm1          \n\t"
564
        "paddb %%mm0, %%mm1             \n\t"
565
        "movq %%mm1, 8(%2, %0)          \n\t"
566
        "add $16, %0                    \n\t"
567
        "2:                             \n\t"
568
        "cmp %3, %0                     \n\t"
569
        " js 1b                         \n\t"
570
        : "+r" (i)
571
        : "r"(src), "r"(dst), "r"((x86_reg)w-15)
572
    );
573
    for(; i<w; i++)
574
        dst[i+0] += src[i+0];
575
}
576

    
577
static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
578
    x86_reg i=0;
579
    __asm__ volatile(
580
        "jmp 2f                         \n\t"
581
        "1:                             \n\t"
582
        "movq   (%2, %0), %%mm0         \n\t"
583
        "movq  8(%2, %0), %%mm1         \n\t"
584
        "paddb  (%3, %0), %%mm0         \n\t"
585
        "paddb 8(%3, %0), %%mm1         \n\t"
586
        "movq %%mm0,  (%1, %0)          \n\t"
587
        "movq %%mm1, 8(%1, %0)          \n\t"
588
        "add $16, %0                    \n\t"
589
        "2:                             \n\t"
590
        "cmp %4, %0                     \n\t"
591
        " js 1b                         \n\t"
592
        : "+r" (i)
593
        : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
594
    );
595
    for(; i<w; i++)
596
        dst[i] = src1[i] + src2[i];
597
}
598

    
599
#if HAVE_7REGS && HAVE_TEN_OPERANDS
600
static void add_hfyu_median_prediction_cmov(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top) {
601
    x86_reg w2 = -w;
602
    x86_reg x;
603
    int l = *left & 0xff;
604
    int tl = *left_top & 0xff;
605
    int t;
606
    __asm__ volatile(
607
        "mov    %7, %3 \n"
608
        "1: \n"
609
        "movzx (%3,%4), %2 \n"
610
        "mov    %2, %k3 \n"
611
        "sub   %b1, %b3 \n"
612
        "add   %b0, %b3 \n"
613
        "mov    %2, %1 \n"
614
        "cmp    %0, %2 \n"
615
        "cmovg  %0, %2 \n"
616
        "cmovg  %1, %0 \n"
617
        "cmp   %k3, %0 \n"
618
        "cmovg %k3, %0 \n"
619
        "mov    %7, %3 \n"
620
        "cmp    %2, %0 \n"
621
        "cmovl  %2, %0 \n"
622
        "add (%6,%4), %b0 \n"
623
        "mov   %b0, (%5,%4) \n"
624
        "inc    %4 \n"
625
        "jl 1b \n"
626
        :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
627
        :"r"(dst+w), "r"(diff+w), "rm"(top+w)
628
    );
629
    *left = l;
630
    *left_top = tl;
631
}
632
#endif
633

    
634
#define H263_LOOP_FILTER \
635
        "pxor %%mm7, %%mm7              \n\t"\
636
        "movq  %0, %%mm0                \n\t"\
637
        "movq  %0, %%mm1                \n\t"\
638
        "movq  %3, %%mm2                \n\t"\
639
        "movq  %3, %%mm3                \n\t"\
640
        "punpcklbw %%mm7, %%mm0         \n\t"\
641
        "punpckhbw %%mm7, %%mm1         \n\t"\
642
        "punpcklbw %%mm7, %%mm2         \n\t"\
643
        "punpckhbw %%mm7, %%mm3         \n\t"\
644
        "psubw %%mm2, %%mm0             \n\t"\
645
        "psubw %%mm3, %%mm1             \n\t"\
646
        "movq  %1, %%mm2                \n\t"\
647
        "movq  %1, %%mm3                \n\t"\
648
        "movq  %2, %%mm4                \n\t"\
649
        "movq  %2, %%mm5                \n\t"\
650
        "punpcklbw %%mm7, %%mm2         \n\t"\
651
        "punpckhbw %%mm7, %%mm3         \n\t"\
652
        "punpcklbw %%mm7, %%mm4         \n\t"\
653
        "punpckhbw %%mm7, %%mm5         \n\t"\
654
        "psubw %%mm2, %%mm4             \n\t"\
655
        "psubw %%mm3, %%mm5             \n\t"\
656
        "psllw $2, %%mm4                \n\t"\
657
        "psllw $2, %%mm5                \n\t"\
658
        "paddw %%mm0, %%mm4             \n\t"\
659
        "paddw %%mm1, %%mm5             \n\t"\
660
        "pxor %%mm6, %%mm6              \n\t"\
661
        "pcmpgtw %%mm4, %%mm6           \n\t"\
662
        "pcmpgtw %%mm5, %%mm7           \n\t"\
663
        "pxor %%mm6, %%mm4              \n\t"\
664
        "pxor %%mm7, %%mm5              \n\t"\
665
        "psubw %%mm6, %%mm4             \n\t"\
666
        "psubw %%mm7, %%mm5             \n\t"\
667
        "psrlw $3, %%mm4                \n\t"\
668
        "psrlw $3, %%mm5                \n\t"\
669
        "packuswb %%mm5, %%mm4          \n\t"\
670
        "packsswb %%mm7, %%mm6          \n\t"\
671
        "pxor %%mm7, %%mm7              \n\t"\
672
        "movd %4, %%mm2                 \n\t"\
673
        "punpcklbw %%mm2, %%mm2         \n\t"\
674
        "punpcklbw %%mm2, %%mm2         \n\t"\
675
        "punpcklbw %%mm2, %%mm2         \n\t"\
676
        "psubusb %%mm4, %%mm2           \n\t"\
677
        "movq %%mm2, %%mm3              \n\t"\
678
        "psubusb %%mm4, %%mm3           \n\t"\
679
        "psubb %%mm3, %%mm2             \n\t"\
680
        "movq %1, %%mm3                 \n\t"\
681
        "movq %2, %%mm4                 \n\t"\
682
        "pxor %%mm6, %%mm3              \n\t"\
683
        "pxor %%mm6, %%mm4              \n\t"\
684
        "paddusb %%mm2, %%mm3           \n\t"\
685
        "psubusb %%mm2, %%mm4           \n\t"\
686
        "pxor %%mm6, %%mm3              \n\t"\
687
        "pxor %%mm6, %%mm4              \n\t"\
688
        "paddusb %%mm2, %%mm2           \n\t"\
689
        "packsswb %%mm1, %%mm0          \n\t"\
690
        "pcmpgtb %%mm0, %%mm7           \n\t"\
691
        "pxor %%mm7, %%mm0              \n\t"\
692
        "psubb %%mm7, %%mm0             \n\t"\
693
        "movq %%mm0, %%mm1              \n\t"\
694
        "psubusb %%mm2, %%mm0           \n\t"\
695
        "psubb %%mm0, %%mm1             \n\t"\
696
        "pand %5, %%mm1                 \n\t"\
697
        "psrlw $2, %%mm1                \n\t"\
698
        "pxor %%mm7, %%mm1              \n\t"\
699
        "psubb %%mm7, %%mm1             \n\t"\
700
        "movq %0, %%mm5                 \n\t"\
701
        "movq %3, %%mm6                 \n\t"\
702
        "psubb %%mm1, %%mm5             \n\t"\
703
        "paddb %%mm1, %%mm6             \n\t"
704

    
705
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
706
    if(CONFIG_ANY_H263) {
707
    const int strength= ff_h263_loop_filter_strength[qscale];
708

    
709
    __asm__ volatile(
710

    
711
        H263_LOOP_FILTER
712

    
713
        "movq %%mm3, %1                 \n\t"
714
        "movq %%mm4, %2                 \n\t"
715
        "movq %%mm5, %0                 \n\t"
716
        "movq %%mm6, %3                 \n\t"
717
        : "+m" (*(uint64_t*)(src - 2*stride)),
718
          "+m" (*(uint64_t*)(src - 1*stride)),
719
          "+m" (*(uint64_t*)(src + 0*stride)),
720
          "+m" (*(uint64_t*)(src + 1*stride))
721
        : "g" (2*strength), "m"(ff_pb_FC)
722
    );
723
    }
724
}
725

    
726
static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
727
    __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
728
        "movd  %4, %%mm0                \n\t"
729
        "movd  %5, %%mm1                \n\t"
730
        "movd  %6, %%mm2                \n\t"
731
        "movd  %7, %%mm3                \n\t"
732
        "punpcklbw %%mm1, %%mm0         \n\t"
733
        "punpcklbw %%mm3, %%mm2         \n\t"
734
        "movq %%mm0, %%mm1              \n\t"
735
        "punpcklwd %%mm2, %%mm0         \n\t"
736
        "punpckhwd %%mm2, %%mm1         \n\t"
737
        "movd  %%mm0, %0                \n\t"
738
        "punpckhdq %%mm0, %%mm0         \n\t"
739
        "movd  %%mm0, %1                \n\t"
740
        "movd  %%mm1, %2                \n\t"
741
        "punpckhdq %%mm1, %%mm1         \n\t"
742
        "movd  %%mm1, %3                \n\t"
743

    
744
        : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
745
          "=m" (*(uint32_t*)(dst + 1*dst_stride)),
746
          "=m" (*(uint32_t*)(dst + 2*dst_stride)),
747
          "=m" (*(uint32_t*)(dst + 3*dst_stride))
748
        :  "m" (*(uint32_t*)(src + 0*src_stride)),
749
           "m" (*(uint32_t*)(src + 1*src_stride)),
750
           "m" (*(uint32_t*)(src + 2*src_stride)),
751
           "m" (*(uint32_t*)(src + 3*src_stride))
752
    );
753
}
754

    
755
static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
756
    if(CONFIG_ANY_H263) {
757
    const int strength= ff_h263_loop_filter_strength[qscale];
758
    DECLARE_ALIGNED(8, uint64_t, temp[4]);
759
    uint8_t *btemp= (uint8_t*)temp;
760

    
761
    src -= 2;
762

    
763
    transpose4x4(btemp  , src           , 8, stride);
764
    transpose4x4(btemp+4, src + 4*stride, 8, stride);
765
    __asm__ volatile(
766
        H263_LOOP_FILTER // 5 3 4 6
767

    
768
        : "+m" (temp[0]),
769
          "+m" (temp[1]),
770
          "+m" (temp[2]),
771
          "+m" (temp[3])
772
        : "g" (2*strength), "m"(ff_pb_FC)
773
    );
774

    
775
    __asm__ volatile(
776
        "movq %%mm5, %%mm1              \n\t"
777
        "movq %%mm4, %%mm0              \n\t"
778
        "punpcklbw %%mm3, %%mm5         \n\t"
779
        "punpcklbw %%mm6, %%mm4         \n\t"
780
        "punpckhbw %%mm3, %%mm1         \n\t"
781
        "punpckhbw %%mm6, %%mm0         \n\t"
782
        "movq %%mm5, %%mm3              \n\t"
783
        "movq %%mm1, %%mm6              \n\t"
784
        "punpcklwd %%mm4, %%mm5         \n\t"
785
        "punpcklwd %%mm0, %%mm1         \n\t"
786
        "punpckhwd %%mm4, %%mm3         \n\t"
787
        "punpckhwd %%mm0, %%mm6         \n\t"
788
        "movd %%mm5, (%0)               \n\t"
789
        "punpckhdq %%mm5, %%mm5         \n\t"
790
        "movd %%mm5, (%0,%2)            \n\t"
791
        "movd %%mm3, (%0,%2,2)          \n\t"
792
        "punpckhdq %%mm3, %%mm3         \n\t"
793
        "movd %%mm3, (%0,%3)            \n\t"
794
        "movd %%mm1, (%1)               \n\t"
795
        "punpckhdq %%mm1, %%mm1         \n\t"
796
        "movd %%mm1, (%1,%2)            \n\t"
797
        "movd %%mm6, (%1,%2,2)          \n\t"
798
        "punpckhdq %%mm6, %%mm6         \n\t"
799
        "movd %%mm6, (%1,%3)            \n\t"
800
        :: "r" (src),
801
           "r" (src + 4*stride),
802
           "r" ((x86_reg)   stride ),
803
           "r" ((x86_reg)(3*stride))
804
    );
805
    }
806
}
807

    
808
/* draw the edges of width 'w' of an image of size width, height
809
   this mmx version can only handle w==8 || w==16 */
810
static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
811
{
812
    uint8_t *ptr, *last_line;
813
    int i;
814

    
815
    last_line = buf + (height - 1) * wrap;
816
    /* left and right */
817
    ptr = buf;
818
    if(w==8)
819
    {
820
        __asm__ volatile(
821
                "1:                             \n\t"
822
                "movd (%0), %%mm0               \n\t"
823
                "punpcklbw %%mm0, %%mm0         \n\t"
824
                "punpcklwd %%mm0, %%mm0         \n\t"
825
                "punpckldq %%mm0, %%mm0         \n\t"
826
                "movq %%mm0, -8(%0)             \n\t"
827
                "movq -8(%0, %2), %%mm1         \n\t"
828
                "punpckhbw %%mm1, %%mm1         \n\t"
829
                "punpckhwd %%mm1, %%mm1         \n\t"
830
                "punpckhdq %%mm1, %%mm1         \n\t"
831
                "movq %%mm1, (%0, %2)           \n\t"
832
                "add %1, %0                     \n\t"
833
                "cmp %3, %0                     \n\t"
834
                " jb 1b                         \n\t"
835
                : "+r" (ptr)
836
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
837
        );
838
    }
839
    else
840
    {
841
        __asm__ volatile(
842
                "1:                             \n\t"
843
                "movd (%0), %%mm0               \n\t"
844
                "punpcklbw %%mm0, %%mm0         \n\t"
845
                "punpcklwd %%mm0, %%mm0         \n\t"
846
                "punpckldq %%mm0, %%mm0         \n\t"
847
                "movq %%mm0, -8(%0)             \n\t"
848
                "movq %%mm0, -16(%0)            \n\t"
849
                "movq -8(%0, %2), %%mm1         \n\t"
850
                "punpckhbw %%mm1, %%mm1         \n\t"
851
                "punpckhwd %%mm1, %%mm1         \n\t"
852
                "punpckhdq %%mm1, %%mm1         \n\t"
853
                "movq %%mm1, (%0, %2)           \n\t"
854
                "movq %%mm1, 8(%0, %2)          \n\t"
855
                "add %1, %0                     \n\t"
856
                "cmp %3, %0                     \n\t"
857
                " jb 1b                         \n\t"
858
                : "+r" (ptr)
859
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
860
        );
861
    }
862

    
863
    for(i=0;i<w;i+=4) {
864
        /* top and bottom (and hopefully also the corners) */
865
        ptr= buf - (i + 1) * wrap - w;
866
        __asm__ volatile(
867
                "1:                             \n\t"
868
                "movq (%1, %0), %%mm0           \n\t"
869
                "movq %%mm0, (%0)               \n\t"
870
                "movq %%mm0, (%0, %2)           \n\t"
871
                "movq %%mm0, (%0, %2, 2)        \n\t"
872
                "movq %%mm0, (%0, %3)           \n\t"
873
                "add $8, %0                     \n\t"
874
                "cmp %4, %0                     \n\t"
875
                " jb 1b                         \n\t"
876
                : "+r" (ptr)
877
                : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
878
        );
879
        ptr= last_line + (i + 1) * wrap - w;
880
        __asm__ volatile(
881
                "1:                             \n\t"
882
                "movq (%1, %0), %%mm0           \n\t"
883
                "movq %%mm0, (%0)               \n\t"
884
                "movq %%mm0, (%0, %2)           \n\t"
885
                "movq %%mm0, (%0, %2, 2)        \n\t"
886
                "movq %%mm0, (%0, %3)           \n\t"
887
                "add $8, %0                     \n\t"
888
                "cmp %4, %0                     \n\t"
889
                " jb 1b                         \n\t"
890
                : "+r" (ptr)
891
                : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
892
        );
893
    }
894
}
895

    
896
#define PAETH(cpu, abs3)\
897
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
898
{\
899
    x86_reg i = -bpp;\
900
    x86_reg end = w-3;\
901
    __asm__ volatile(\
902
        "pxor      %%mm7, %%mm7 \n"\
903
        "movd    (%1,%0), %%mm0 \n"\
904
        "movd    (%2,%0), %%mm1 \n"\
905
        "punpcklbw %%mm7, %%mm0 \n"\
906
        "punpcklbw %%mm7, %%mm1 \n"\
907
        "add       %4, %0 \n"\
908
        "1: \n"\
909
        "movq      %%mm1, %%mm2 \n"\
910
        "movd    (%2,%0), %%mm1 \n"\
911
        "movq      %%mm2, %%mm3 \n"\
912
        "punpcklbw %%mm7, %%mm1 \n"\
913
        "movq      %%mm2, %%mm4 \n"\
914
        "psubw     %%mm1, %%mm3 \n"\
915
        "psubw     %%mm0, %%mm4 \n"\
916
        "movq      %%mm3, %%mm5 \n"\
917
        "paddw     %%mm4, %%mm5 \n"\
918
        abs3\
919
        "movq      %%mm4, %%mm6 \n"\
920
        "pminsw    %%mm5, %%mm6 \n"\
921
        "pcmpgtw   %%mm6, %%mm3 \n"\
922
        "pcmpgtw   %%mm5, %%mm4 \n"\
923
        "movq      %%mm4, %%mm6 \n"\
924
        "pand      %%mm3, %%mm4 \n"\
925
        "pandn     %%mm3, %%mm6 \n"\
926
        "pandn     %%mm0, %%mm3 \n"\
927
        "movd    (%3,%0), %%mm0 \n"\
928
        "pand      %%mm1, %%mm6 \n"\
929
        "pand      %%mm4, %%mm2 \n"\
930
        "punpcklbw %%mm7, %%mm0 \n"\
931
        "movq      %6,    %%mm5 \n"\
932
        "paddw     %%mm6, %%mm0 \n"\
933
        "paddw     %%mm2, %%mm3 \n"\
934
        "paddw     %%mm3, %%mm0 \n"\
935
        "pand      %%mm5, %%mm0 \n"\
936
        "movq      %%mm0, %%mm3 \n"\
937
        "packuswb  %%mm3, %%mm3 \n"\
938
        "movd      %%mm3, (%1,%0) \n"\
939
        "add       %4, %0 \n"\
940
        "cmp       %5, %0 \n"\
941
        "jle 1b \n"\
942
        :"+r"(i)\
943
        :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
944
         "m"(ff_pw_255)\
945
        :"memory"\
946
    );\
947
}
948

    
949
#define ABS3_MMX2\
950
        "psubw     %%mm5, %%mm7 \n"\
951
        "pmaxsw    %%mm7, %%mm5 \n"\
952
        "pxor      %%mm6, %%mm6 \n"\
953
        "pxor      %%mm7, %%mm7 \n"\
954
        "psubw     %%mm3, %%mm6 \n"\
955
        "psubw     %%mm4, %%mm7 \n"\
956
        "pmaxsw    %%mm6, %%mm3 \n"\
957
        "pmaxsw    %%mm7, %%mm4 \n"\
958
        "pxor      %%mm7, %%mm7 \n"
959

    
960
#define ABS3_SSSE3\
961
        "pabsw     %%mm3, %%mm3 \n"\
962
        "pabsw     %%mm4, %%mm4 \n"\
963
        "pabsw     %%mm5, %%mm5 \n"
964

    
965
PAETH(mmx2, ABS3_MMX2)
966
#if HAVE_SSSE3
967
PAETH(ssse3, ABS3_SSSE3)
968
#endif
969

    
970
#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
971
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\
972
        "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\
973
        "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\
974
        "movq "#in7", " #m3 "             \n\t" /* d */\
975
        "movq "#in0", %%mm5               \n\t" /* D */\
976
        "paddw " #m3 ", %%mm5             \n\t" /* x4 */\
977
        "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\
978
        "movq "#in1", %%mm5               \n\t" /* C */\
979
        "movq "#in2", %%mm6               \n\t" /* B */\
980
        "paddw " #m6 ", %%mm5             \n\t" /* x3 */\
981
        "paddw " #m5 ", %%mm6             \n\t" /* x2 */\
982
        "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\
983
        "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\
984
        "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\
985
        "paddw " #rnd ", %%mm4            \n\t" /* x2 */\
986
        "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
987
        "psraw $5, %%mm5                  \n\t"\
988
        "packuswb %%mm5, %%mm5            \n\t"\
989
        OP(%%mm5, out, %%mm7, d)
990

    
991
#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
992
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
993
    uint64_t temp;\
994
\
995
    __asm__ volatile(\
996
        "pxor %%mm7, %%mm7                \n\t"\
997
        "1:                               \n\t"\
998
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
999
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
1000
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1001
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
1002
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
1003
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
1004
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
1005
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1006
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1007
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
1008
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
1009
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
1010
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
1011
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
1012
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
1013
        "paddw %%mm3, %%mm5               \n\t" /* b */\
1014
        "paddw %%mm2, %%mm6               \n\t" /* c */\
1015
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1016
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
1017
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
1018
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
1019
        "paddw %%mm4, %%mm0               \n\t" /* a */\
1020
        "paddw %%mm1, %%mm5               \n\t" /* d */\
1021
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1022
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
1023
        "paddw %6, %%mm6                  \n\t"\
1024
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1025
        "psraw $5, %%mm0                  \n\t"\
1026
        "movq %%mm0, %5                   \n\t"\
1027
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1028
        \
1029
        "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\
1030
        "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\
1031
        "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\
1032
        "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\
1033
        "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\
1034
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\
1035
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\
1036
        "paddw %%mm0, %%mm2               \n\t" /* b */\
1037
        "paddw %%mm5, %%mm3               \n\t" /* c */\
1038
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1039
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1040
        "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\
1041
        "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\
1042
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\
1043
        "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\
1044
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1045
        "paddw %%mm2, %%mm1               \n\t" /* a */\
1046
        "paddw %%mm6, %%mm4               \n\t" /* d */\
1047
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1048
        "psubw %%mm4, %%mm3               \n\t" /* - 6b +3c - d */\
1049
        "paddw %6, %%mm1                  \n\t"\
1050
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b +3c - d */\
1051
        "psraw $5, %%mm3                  \n\t"\
1052
        "movq %5, %%mm1                   \n\t"\
1053
        "packuswb %%mm3, %%mm1            \n\t"\
1054
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
1055
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
1056
        \
1057
        "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\
1058
        "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\
1059
        "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\
1060
        "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\
1061
        "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\
1062
        "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\
1063
        "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\
1064
        "paddw %%mm1, %%mm5               \n\t" /* b */\
1065
        "paddw %%mm4, %%mm0               \n\t" /* c */\
1066
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1067
        "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\
1068
        "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\
1069
        "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\
1070
        "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\
1071
        "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\
1072
        "paddw %%mm3, %%mm2               \n\t" /* d */\
1073
        "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\
1074
        "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\
1075
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\
1076
        "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\
1077
        "paddw %%mm2, %%mm6               \n\t" /* a */\
1078
        "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
1079
        "paddw %6, %%mm0                  \n\t"\
1080
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1081
        "psraw $5, %%mm0                  \n\t"\
1082
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
1083
        \
1084
        "paddw %%mm5, %%mm3               \n\t" /* a */\
1085
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\
1086
        "paddw %%mm4, %%mm6               \n\t" /* b */\
1087
        "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\
1088
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\
1089
        "paddw %%mm1, %%mm4               \n\t" /* c */\
1090
        "paddw %%mm2, %%mm5               \n\t" /* d */\
1091
        "paddw %%mm6, %%mm6               \n\t" /* 2b */\
1092
        "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\
1093
        "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
1094
        "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\
1095
        "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\
1096
        "paddw %6, %%mm4                  \n\t"\
1097
        "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\
1098
        "psraw $5, %%mm4                  \n\t"\
1099
        "packuswb %%mm4, %%mm0            \n\t"\
1100
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
1101
        \
1102
        "add %3, %0                       \n\t"\
1103
        "add %4, %1                       \n\t"\
1104
        "decl %2                          \n\t"\
1105
        " jnz 1b                          \n\t"\
1106
        : "+a"(src), "+c"(dst), "+D"(h)\
1107
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1108
        : "memory"\
1109
    );\
1110
}\
1111
\
1112
static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1113
    int i;\
1114
    int16_t temp[16];\
1115
    /* quick HACK, XXX FIXME MUST be optimized */\
1116
    for(i=0; i<h; i++)\
1117
    {\
1118
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1119
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1120
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1121
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1122
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1123
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1124
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1125
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1126
        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1127
        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1128
        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1129
        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1130
        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1131
        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1132
        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1133
        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1134
        __asm__ volatile(\
1135
            "movq (%0), %%mm0               \n\t"\
1136
            "movq 8(%0), %%mm1              \n\t"\
1137
            "paddw %2, %%mm0                \n\t"\
1138
            "paddw %2, %%mm1                \n\t"\
1139
            "psraw $5, %%mm0                \n\t"\
1140
            "psraw $5, %%mm1                \n\t"\
1141
            "packuswb %%mm1, %%mm0          \n\t"\
1142
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1143
            "movq 16(%0), %%mm0             \n\t"\
1144
            "movq 24(%0), %%mm1             \n\t"\
1145
            "paddw %2, %%mm0                \n\t"\
1146
            "paddw %2, %%mm1                \n\t"\
1147
            "psraw $5, %%mm0                \n\t"\
1148
            "psraw $5, %%mm1                \n\t"\
1149
            "packuswb %%mm1, %%mm0          \n\t"\
1150
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1151
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1152
            : "memory"\
1153
        );\
1154
        dst+=dstStride;\
1155
        src+=srcStride;\
1156
    }\
1157
}\
1158
\
1159
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1160
    __asm__ volatile(\
1161
        "pxor %%mm7, %%mm7                \n\t"\
1162
        "1:                               \n\t"\
1163
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
1164
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
1165
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1166
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
1167
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
1168
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
1169
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
1170
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1171
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1172
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
1173
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
1174
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
1175
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
1176
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
1177
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
1178
        "paddw %%mm3, %%mm5               \n\t" /* b */\
1179
        "paddw %%mm2, %%mm6               \n\t" /* c */\
1180
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1181
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
1182
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
1183
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
1184
        "paddw %%mm4, %%mm0               \n\t" /* a */\
1185
        "paddw %%mm1, %%mm5               \n\t" /* d */\
1186
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1187
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
1188
        "paddw %5, %%mm6                  \n\t"\
1189
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1190
        "psraw $5, %%mm0                  \n\t"\
1191
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1192
        \
1193
        "movd 5(%0), %%mm5                \n\t" /* FGHI */\
1194
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\
1195
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\
1196
        "paddw %%mm5, %%mm1               \n\t" /* a */\
1197
        "paddw %%mm6, %%mm2               \n\t" /* b */\
1198
        "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\
1199
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\
1200
        "paddw %%mm6, %%mm3               \n\t" /* c */\
1201
        "paddw %%mm5, %%mm4               \n\t" /* d */\
1202
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1203
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1204
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1205
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1206
        "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\
1207
        "paddw %5, %%mm1                  \n\t"\
1208
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\
1209
        "psraw $5, %%mm3                  \n\t"\
1210
        "packuswb %%mm3, %%mm0            \n\t"\
1211
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
1212
        \
1213
        "add %3, %0                       \n\t"\
1214
        "add %4, %1                       \n\t"\
1215
        "decl %2                          \n\t"\
1216
        " jnz 1b                          \n\t"\
1217
        : "+a"(src), "+c"(dst), "+d"(h)\
1218
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
1219
        : "memory"\
1220
    );\
1221
}\
1222
\
1223
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1224
    int i;\
1225
    int16_t temp[8];\
1226
    /* quick HACK, XXX FIXME MUST be optimized */\
1227
    for(i=0; i<h; i++)\
1228
    {\
1229
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1230
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1231
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1232
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1233
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1234
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1235
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1236
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1237
        __asm__ volatile(\
1238
            "movq (%0), %%mm0           \n\t"\
1239
            "movq 8(%0), %%mm1          \n\t"\
1240
            "paddw %2, %%mm0            \n\t"\
1241
            "paddw %2, %%mm1            \n\t"\
1242
            "psraw $5, %%mm0            \n\t"\
1243
            "psraw $5, %%mm1            \n\t"\
1244
            "packuswb %%mm1, %%mm0      \n\t"\
1245
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1246
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1247
            :"memory"\
1248
        );\
1249
        dst+=dstStride;\
1250
        src+=srcStride;\
1251
    }\
1252
}
1253

    
1254
#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1255
\
1256
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1257
    uint64_t temp[17*4];\
1258
    uint64_t *temp_ptr= temp;\
1259
    int count= 17;\
1260
\
1261
    /*FIXME unroll */\
1262
    __asm__ volatile(\
1263
        "pxor %%mm7, %%mm7              \n\t"\
1264
        "1:                             \n\t"\
1265
        "movq (%0), %%mm0               \n\t"\
1266
        "movq (%0), %%mm1               \n\t"\
1267
        "movq 8(%0), %%mm2              \n\t"\
1268
        "movq 8(%0), %%mm3              \n\t"\
1269
        "punpcklbw %%mm7, %%mm0         \n\t"\
1270
        "punpckhbw %%mm7, %%mm1         \n\t"\
1271
        "punpcklbw %%mm7, %%mm2         \n\t"\
1272
        "punpckhbw %%mm7, %%mm3         \n\t"\
1273
        "movq %%mm0, (%1)               \n\t"\
1274
        "movq %%mm1, 17*8(%1)           \n\t"\
1275
        "movq %%mm2, 2*17*8(%1)         \n\t"\
1276
        "movq %%mm3, 3*17*8(%1)         \n\t"\
1277
        "add $8, %1                     \n\t"\
1278
        "add %3, %0                     \n\t"\
1279
        "decl %2                        \n\t"\
1280
        " jnz 1b                        \n\t"\
1281
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1282
        : "r" ((x86_reg)srcStride)\
1283
        : "memory"\
1284
    );\
1285
    \
1286
    temp_ptr= temp;\
1287
    count=4;\
1288
    \
1289
/*FIXME reorder for speed */\
1290
    __asm__ volatile(\
1291
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1292
        "1:                             \n\t"\
1293
        "movq (%0), %%mm0               \n\t"\
1294
        "movq 8(%0), %%mm1              \n\t"\
1295
        "movq 16(%0), %%mm2             \n\t"\
1296
        "movq 24(%0), %%mm3             \n\t"\
1297
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1298
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1299
        "add %4, %1                     \n\t"\
1300
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1301
        \
1302
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1303
        "add %4, %1                     \n\t"\
1304
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1305
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1306
        "add %4, %1                     \n\t"\
1307
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1308
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1309
        "add %4, %1                     \n\t"\
1310
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1311
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1312
        "add %4, %1                     \n\t"\
1313
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1314
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1315
        "add %4, %1                     \n\t"\
1316
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1317
        \
1318
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1319
        "add %4, %1                     \n\t"  \
1320
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1321
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1322
        \
1323
        "add $136, %0                   \n\t"\
1324
        "add %6, %1                     \n\t"\
1325
        "decl %2                        \n\t"\
1326
        " jnz 1b                        \n\t"\
1327
        \
1328
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1329
        : "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)\
1330
        :"memory"\
1331
    );\
1332
}\
1333
\
1334
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1335
    uint64_t temp[9*2];\
1336
    uint64_t *temp_ptr= temp;\
1337
    int count= 9;\
1338
\
1339
    /*FIXME unroll */\
1340
    __asm__ volatile(\
1341
        "pxor %%mm7, %%mm7              \n\t"\
1342
        "1:                             \n\t"\
1343
        "movq (%0), %%mm0               \n\t"\
1344
        "movq (%0), %%mm1               \n\t"\
1345
        "punpcklbw %%mm7, %%mm0         \n\t"\
1346
        "punpckhbw %%mm7, %%mm1         \n\t"\
1347
        "movq %%mm0, (%1)               \n\t"\
1348
        "movq %%mm1, 9*8(%1)            \n\t"\
1349
        "add $8, %1                     \n\t"\
1350
        "add %3, %0                     \n\t"\
1351
        "decl %2                        \n\t"\
1352
        " jnz 1b                        \n\t"\
1353
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1354
        : "r" ((x86_reg)srcStride)\
1355
        : "memory"\
1356
    );\
1357
    \
1358
    temp_ptr= temp;\
1359
    count=2;\
1360
    \
1361
/*FIXME reorder for speed */\
1362
    __asm__ volatile(\
1363
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1364
        "1:                             \n\t"\
1365
        "movq (%0), %%mm0               \n\t"\
1366
        "movq 8(%0), %%mm1              \n\t"\
1367
        "movq 16(%0), %%mm2             \n\t"\
1368
        "movq 24(%0), %%mm3             \n\t"\
1369
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1370
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1371
        "add %4, %1                     \n\t"\
1372
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1373
        \
1374
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1375
        "add %4, %1                     \n\t"\
1376
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1377
        \
1378
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1379
        "add %4, %1                     \n\t"\
1380
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1381
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1382
                \
1383
        "add $72, %0                    \n\t"\
1384
        "add %6, %1                     \n\t"\
1385
        "decl %2                        \n\t"\
1386
        " jnz 1b                        \n\t"\
1387
         \
1388
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1389
        : "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)\
1390
        : "memory"\
1391
   );\
1392
}\
1393
\
1394
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1395
    OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\
1396
}\
1397
\
1398
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1399
    uint64_t temp[8];\
1400
    uint8_t * const half= (uint8_t*)temp;\
1401
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1402
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1403
}\
1404
\
1405
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1406
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1407
}\
1408
\
1409
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1410
    uint64_t temp[8];\
1411
    uint8_t * const half= (uint8_t*)temp;\
1412
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1413
    OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
1414
}\
1415
\
1416
static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1417
    uint64_t temp[8];\
1418
    uint8_t * const half= (uint8_t*)temp;\
1419
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1420
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1421
}\
1422
\
1423
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1424
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1425
}\
1426
\
1427
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1428
    uint64_t temp[8];\
1429
    uint8_t * const half= (uint8_t*)temp;\
1430
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1431
    OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
1432
}\
1433
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1434
    uint64_t half[8 + 9];\
1435
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1436
    uint8_t * const halfHV= ((uint8_t*)half);\
1437
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1438
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1439
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1440
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1441
}\
1442
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1443
    uint64_t half[8 + 9];\
1444
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1445
    uint8_t * const halfHV= ((uint8_t*)half);\
1446
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1447
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1448
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1449
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1450
}\
1451
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1452
    uint64_t half[8 + 9];\
1453
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1454
    uint8_t * const halfHV= ((uint8_t*)half);\
1455
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1456
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1457
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1458
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1459
}\
1460
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1461
    uint64_t half[8 + 9];\
1462
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1463
    uint8_t * const halfHV= ((uint8_t*)half);\
1464
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1465
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1466
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1467
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1468
}\
1469
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1470
    uint64_t half[8 + 9];\
1471
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1472
    uint8_t * const halfHV= ((uint8_t*)half);\
1473
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1474
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1475
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1476
}\
1477
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1478
    uint64_t half[8 + 9];\
1479
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1480
    uint8_t * const halfHV= ((uint8_t*)half);\
1481
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1482
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1483
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1484
}\
1485
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1486
    uint64_t half[8 + 9];\
1487
    uint8_t * const halfH= ((uint8_t*)half);\
1488
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1489
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1490
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1491
}\
1492
static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1493
    uint64_t half[8 + 9];\
1494
    uint8_t * const halfH= ((uint8_t*)half);\
1495
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1496
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1497
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1498
}\
1499
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1500
    uint64_t half[9];\
1501
    uint8_t * const halfH= ((uint8_t*)half);\
1502
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1503
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1504
}\
1505
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1506
    OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\
1507
}\
1508
\
1509
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1510
    uint64_t temp[32];\
1511
    uint8_t * const half= (uint8_t*)temp;\
1512
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1513
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1514
}\
1515
\
1516
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1517
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1518
}\
1519
\
1520
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1521
    uint64_t temp[32];\
1522
    uint8_t * const half= (uint8_t*)temp;\
1523
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1524
    OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
1525
}\
1526
\
1527
static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1528
    uint64_t temp[32];\
1529
    uint8_t * const half= (uint8_t*)temp;\
1530
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1531
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1532
}\
1533
\
1534
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1535
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1536
}\
1537
\
1538
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1539
    uint64_t temp[32];\
1540
    uint8_t * const half= (uint8_t*)temp;\
1541
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1542
    OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
1543
}\
1544
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1545
    uint64_t half[16*2 + 17*2];\
1546
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1547
    uint8_t * const halfHV= ((uint8_t*)half);\
1548
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1549
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1550
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1551
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1552
}\
1553
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1554
    uint64_t half[16*2 + 17*2];\
1555
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1556
    uint8_t * const halfHV= ((uint8_t*)half);\
1557
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1558
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1559
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1560
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1561
}\
1562
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1563
    uint64_t half[16*2 + 17*2];\
1564
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1565
    uint8_t * const halfHV= ((uint8_t*)half);\
1566
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1567
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1568
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1569
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1570
}\
1571
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1572
    uint64_t half[16*2 + 17*2];\
1573
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1574
    uint8_t * const halfHV= ((uint8_t*)half);\
1575
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1576
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1577
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1578
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1579
}\
1580
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1581
    uint64_t half[16*2 + 17*2];\
1582
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1583
    uint8_t * const halfHV= ((uint8_t*)half);\
1584
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1585
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1586
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1587
}\
1588
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1589
    uint64_t half[16*2 + 17*2];\
1590
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1591
    uint8_t * const halfHV= ((uint8_t*)half);\
1592
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1593
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1594
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1595
}\
1596
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1597
    uint64_t half[17*2];\
1598
    uint8_t * const halfH= ((uint8_t*)half);\
1599
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1600
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1601
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1602
}\
1603
static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1604
    uint64_t half[17*2];\
1605
    uint8_t * const halfH= ((uint8_t*)half);\
1606
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1607
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1608
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1609
}\
1610
static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1611
    uint64_t half[17*2];\
1612
    uint8_t * const halfH= ((uint8_t*)half);\
1613
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1614
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1615
}
1616

    
1617
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"
1618
#define AVG_3DNOW_OP(a,b,temp, size) \
1619
"mov" #size " " #b ", " #temp "   \n\t"\
1620
"pavgusb " #temp ", " #a "        \n\t"\
1621
"mov" #size " " #a ", " #b "      \n\t"
1622
#define AVG_MMX2_OP(a,b,temp, size) \
1623
"mov" #size " " #b ", " #temp "   \n\t"\
1624
"pavgb " #temp ", " #a "          \n\t"\
1625
"mov" #size " " #a ", " #b "      \n\t"
1626

    
1627
QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
1628
QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
1629
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1630
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
1631
QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
1632
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1633
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
1634
QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
1635
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1636

    
1637
/***********************************/
1638
/* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
1639

    
1640
#define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
1641
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1642
    OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
1643
}
1644
#define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
1645
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1646
    OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
1647
}
1648

    
1649
#define QPEL_2TAP(OPNAME, SIZE, MMX)\
1650
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
1651
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
1652
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
1653
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
1654
                          OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
1655
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
1656
                          OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
1657
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
1658
                          OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
1659
static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1660
    OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
1661
}\
1662
static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1663
    OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
1664
}\
1665
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\
1666
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\
1667
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\
1668
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\
1669
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\
1670
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\
1671
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\
1672
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
1673

    
1674
QPEL_2TAP(put_, 16, mmx2)
1675
QPEL_2TAP(avg_, 16, mmx2)
1676
QPEL_2TAP(put_,  8, mmx2)
1677
QPEL_2TAP(avg_,  8, mmx2)
1678
QPEL_2TAP(put_, 16, 3dnow)
1679
QPEL_2TAP(avg_, 16, 3dnow)
1680
QPEL_2TAP(put_,  8, 3dnow)
1681
QPEL_2TAP(avg_,  8, 3dnow)
1682

    
1683

    
1684
#if 0
1685
static void just_return(void) { return; }
1686
#endif
1687

    
1688
static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1689
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){
1690
    const int w = 8;
1691
    const int ix = ox>>(16+shift);
1692
    const int iy = oy>>(16+shift);
1693
    const int oxs = ox>>4;
1694
    const int oys = oy>>4;
1695
    const int dxxs = dxx>>4;
1696
    const int dxys = dxy>>4;
1697
    const int dyxs = dyx>>4;
1698
    const int dyys = dyy>>4;
1699
    const uint16_t r4[4] = {r,r,r,r};
1700
    const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
1701
    const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
1702
    const uint64_t shift2 = 2*shift;
1703
    uint8_t edge_buf[(h+1)*stride];
1704
    int x, y;
1705

    
1706
    const int dxw = (dxx-(1<<(16+shift)))*(w-1);
1707
    const int dyh = (dyy-(1<<(16+shift)))*(h-1);
1708
    const int dxh = dxy*(h-1);
1709
    const int dyw = dyx*(w-1);
1710
    if( // non-constant fullpel offset (3% of blocks)
1711
        ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |
1712
         (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)
1713
        // uses more than 16 bits of subpel mv (only at huge resolution)
1714
        || (dxx|dxy|dyx|dyy)&15 )
1715
    {
1716
        //FIXME could still use mmx for some of the rows
1717
        ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
1718
        return;
1719
    }
1720

    
1721
    src += ix + iy*stride;
1722
    if( (unsigned)ix >= width-w ||
1723
        (unsigned)iy >= height-h )
1724
    {
1725
        ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
1726
        src = edge_buf;
1727
    }
1728

    
1729
    __asm__ volatile(
1730
        "movd         %0, %%mm6 \n\t"
1731
        "pxor      %%mm7, %%mm7 \n\t"
1732
        "punpcklwd %%mm6, %%mm6 \n\t"
1733
        "punpcklwd %%mm6, %%mm6 \n\t"
1734
        :: "r"(1<<shift)
1735
    );
1736

    
1737
    for(x=0; x<w; x+=4){
1738
        uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
1739
                            oxs - dxys + dxxs*(x+1),
1740
                            oxs - dxys + dxxs*(x+2),
1741
                            oxs - dxys + dxxs*(x+3) };
1742
        uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
1743
                            oys - dyys + dyxs*(x+1),
1744
                            oys - dyys + dyxs*(x+2),
1745
                            oys - dyys + dyxs*(x+3) };
1746

    
1747
        for(y=0; y<h; y++){
1748
            __asm__ volatile(
1749
                "movq   %0,  %%mm4 \n\t"
1750
                "movq   %1,  %%mm5 \n\t"
1751
                "paddw  %2,  %%mm4 \n\t"
1752
                "paddw  %3,  %%mm5 \n\t"
1753
                "movq   %%mm4, %0  \n\t"
1754
                "movq   %%mm5, %1  \n\t"
1755
                "psrlw  $12, %%mm4 \n\t"
1756
                "psrlw  $12, %%mm5 \n\t"
1757
                : "+m"(*dx4), "+m"(*dy4)
1758
                : "m"(*dxy4), "m"(*dyy4)
1759
            );
1760

    
1761
            __asm__ volatile(
1762
                "movq   %%mm6, %%mm2 \n\t"
1763
                "movq   %%mm6, %%mm1 \n\t"
1764
                "psubw  %%mm4, %%mm2 \n\t"
1765
                "psubw  %%mm5, %%mm1 \n\t"
1766
                "movq   %%mm2, %%mm0 \n\t"
1767
                "movq   %%mm4, %%mm3 \n\t"
1768
                "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
1769
                "pmullw %%mm5, %%mm3 \n\t" // dx*dy
1770
                "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
1771
                "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
1772

    
1773
                "movd   %4,    %%mm5 \n\t"
1774
                "movd   %3,    %%mm4 \n\t"
1775
                "punpcklbw %%mm7, %%mm5 \n\t"
1776
                "punpcklbw %%mm7, %%mm4 \n\t"
1777
                "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
1778
                "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
1779

    
1780
                "movd   %2,    %%mm5 \n\t"
1781
                "movd   %1,    %%mm4 \n\t"
1782
                "punpcklbw %%mm7, %%mm5 \n\t"
1783
                "punpcklbw %%mm7, %%mm4 \n\t"
1784
                "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
1785
                "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
1786
                "paddw  %5,    %%mm1 \n\t"
1787
                "paddw  %%mm3, %%mm2 \n\t"
1788
                "paddw  %%mm1, %%mm0 \n\t"
1789
                "paddw  %%mm2, %%mm0 \n\t"
1790

    
1791
                "psrlw    %6,    %%mm0 \n\t"
1792
                "packuswb %%mm0, %%mm0 \n\t"
1793
                "movd     %%mm0, %0    \n\t"
1794

    
1795
                : "=m"(dst[x+y*stride])
1796
                : "m"(src[0]), "m"(src[1]),
1797
                  "m"(src[stride]), "m"(src[stride+1]),
1798
                  "m"(*r4), "m"(shift2)
1799
            );
1800
            src += stride;
1801
        }
1802
        src += 4-h*stride;
1803
    }
1804
}
1805

    
1806
#define PREFETCH(name, op) \
1807
static void name(void *mem, int stride, int h){\
1808
    const uint8_t *p= mem;\
1809
    do{\
1810
        __asm__ volatile(#op" %0" :: "m"(*p));\
1811
        p+= stride;\
1812
    }while(--h);\
1813
}
1814
PREFETCH(prefetch_mmx2,  prefetcht0)
1815
PREFETCH(prefetch_3dnow, prefetch)
1816
#undef PREFETCH
1817

    
1818
#include "h264dsp_mmx.c"
1819
#include "rv40dsp_mmx.c"
1820

    
1821
/* CAVS specific */
1822
void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);
1823
void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx);
1824

    
1825
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1826
    put_pixels8_mmx(dst, src, stride, 8);
1827
}
1828
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1829
    avg_pixels8_mmx(dst, src, stride, 8);
1830
}
1831
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1832
    put_pixels16_mmx(dst, src, stride, 16);
1833
}
1834
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1835
    avg_pixels16_mmx(dst, src, stride, 16);
1836
}
1837

    
1838
/* VC1 specific */
1839
void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx);
1840

    
1841
void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1842
    put_pixels8_mmx(dst, src, stride, 8);
1843
}
1844
void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1845
    avg_pixels8_mmx2(dst, src, stride, 8);
1846
}
1847

    
1848
/* external functions, from idct_mmx.c */
1849
void ff_mmx_idct(DCTELEM *block);
1850
void ff_mmxext_idct(DCTELEM *block);
1851

    
1852
/* XXX: those functions should be suppressed ASAP when all IDCTs are
1853
   converted */
1854
#if CONFIG_GPL
1855
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1856
{
1857
    ff_mmx_idct (block);
1858
    put_pixels_clamped_mmx(block, dest, line_size);
1859
}
1860
static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1861
{
1862
    ff_mmx_idct (block);
1863
    add_pixels_clamped_mmx(block, dest, line_size);
1864
}
1865
static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1866
{
1867
    ff_mmxext_idct (block);
1868
    put_pixels_clamped_mmx(block, dest, line_size);
1869
}
1870
static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1871
{
1872
    ff_mmxext_idct (block);
1873
    add_pixels_clamped_mmx(block, dest, line_size);
1874
}
1875
#endif
1876
static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
1877
{
1878
    ff_idct_xvid_mmx (block);
1879
    put_pixels_clamped_mmx(block, dest, line_size);
1880
}
1881
static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
1882
{
1883
    ff_idct_xvid_mmx (block);
1884
    add_pixels_clamped_mmx(block, dest, line_size);
1885
}
1886
static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
1887
{
1888
    ff_idct_xvid_mmx2 (block);
1889
    put_pixels_clamped_mmx(block, dest, line_size);
1890
}
1891
static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
1892
{
1893
    ff_idct_xvid_mmx2 (block);
1894
    add_pixels_clamped_mmx(block, dest, line_size);
1895
}
1896

    
1897
static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
1898
{
1899
    int i;
1900
    __asm__ volatile("pxor %%mm7, %%mm7":);
1901
    for(i=0; i<blocksize; i+=2) {
1902
        __asm__ volatile(
1903
            "movq    %0,    %%mm0 \n\t"
1904
            "movq    %1,    %%mm1 \n\t"
1905
            "movq    %%mm0, %%mm2 \n\t"
1906
            "movq    %%mm1, %%mm3 \n\t"
1907
            "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
1908
            "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
1909
            "pslld   $31,   %%mm2 \n\t" // keep only the sign bit
1910
            "pxor    %%mm2, %%mm1 \n\t"
1911
            "movq    %%mm3, %%mm4 \n\t"
1912
            "pand    %%mm1, %%mm3 \n\t"
1913
            "pandn   %%mm1, %%mm4 \n\t"
1914
            "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1915
            "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1916
            "movq    %%mm3, %1    \n\t"
1917
            "movq    %%mm0, %0    \n\t"
1918
            :"+m"(mag[i]), "+m"(ang[i])
1919
            ::"memory"
1920
        );
1921
    }
1922
    __asm__ volatile("femms");
1923
}
1924
static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
1925
{
1926
    int i;
1927

    
1928
    __asm__ volatile(
1929
            "movaps  %0,     %%xmm5 \n\t"
1930
        ::"m"(ff_pdw_80000000[0])
1931
    );
1932
    for(i=0; i<blocksize; i+=4) {
1933
        __asm__ volatile(
1934
            "movaps  %0,     %%xmm0 \n\t"
1935
            "movaps  %1,     %%xmm1 \n\t"
1936
            "xorps   %%xmm2, %%xmm2 \n\t"
1937
            "xorps   %%xmm3, %%xmm3 \n\t"
1938
            "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
1939
            "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
1940
            "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit
1941
            "xorps   %%xmm2, %%xmm1 \n\t"
1942
            "movaps  %%xmm3, %%xmm4 \n\t"
1943
            "andps   %%xmm1, %%xmm3 \n\t"
1944
            "andnps  %%xmm1, %%xmm4 \n\t"
1945
            "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
1946
            "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
1947
            "movaps  %%xmm3, %1     \n\t"
1948
            "movaps  %%xmm0, %0     \n\t"
1949
            :"+m"(mag[i]), "+m"(ang[i])
1950
            ::"memory"
1951
        );
1952
    }
1953
}
1954

    
1955
#define IF1(x) x
1956
#define IF0(x)
1957

    
1958
#define MIX5(mono,stereo)\
1959
    __asm__ volatile(\
1960
        "movss          0(%2), %%xmm5 \n"\
1961
        "movss          8(%2), %%xmm6 \n"\
1962
        "movss         24(%2), %%xmm7 \n"\
1963
        "shufps    $0, %%xmm5, %%xmm5 \n"\
1964
        "shufps    $0, %%xmm6, %%xmm6 \n"\
1965
        "shufps    $0, %%xmm7, %%xmm7 \n"\
1966
        "1: \n"\
1967
        "movaps       (%0,%1), %%xmm0 \n"\
1968
        "movaps  0x400(%0,%1), %%xmm1 \n"\
1969
        "movaps  0x800(%0,%1), %%xmm2 \n"\
1970
        "movaps  0xc00(%0,%1), %%xmm3 \n"\
1971
        "movaps 0x1000(%0,%1), %%xmm4 \n"\
1972
        "mulps         %%xmm5, %%xmm0 \n"\
1973
        "mulps         %%xmm6, %%xmm1 \n"\
1974
        "mulps         %%xmm5, %%xmm2 \n"\
1975
        "mulps         %%xmm7, %%xmm3 \n"\
1976
        "mulps         %%xmm7, %%xmm4 \n"\
1977
 stereo("addps         %%xmm1, %%xmm0 \n")\
1978
        "addps         %%xmm1, %%xmm2 \n"\
1979
        "addps         %%xmm3, %%xmm0 \n"\
1980
        "addps         %%xmm4, %%xmm2 \n"\
1981
   mono("addps         %%xmm2, %%xmm0 \n")\
1982
        "movaps  %%xmm0,      (%0,%1) \n"\
1983
 stereo("movaps  %%xmm2, 0x400(%0,%1) \n")\
1984
        "add $16, %0 \n"\
1985
        "jl 1b \n"\
1986
        :"+&r"(i)\
1987
        :"r"(samples[0]+len), "r"(matrix)\
1988
        :"memory"\
1989
    );
1990

    
1991
#define MIX_MISC(stereo)\
1992
    __asm__ volatile(\
1993
        "1: \n"\
1994
        "movaps  (%3,%0), %%xmm0 \n"\
1995
 stereo("movaps   %%xmm0, %%xmm1 \n")\
1996
        "mulps    %%xmm6, %%xmm0 \n"\
1997
 stereo("mulps    %%xmm7, %%xmm1 \n")\
1998
        "lea 1024(%3,%0), %1 \n"\
1999
        "mov %5, %2 \n"\
2000
        "2: \n"\
2001
        "movaps   (%1),   %%xmm2 \n"\
2002
 stereo("movaps   %%xmm2, %%xmm3 \n")\
2003
        "mulps   (%4,%2), %%xmm2 \n"\
2004
 stereo("mulps 16(%4,%2), %%xmm3 \n")\
2005
        "addps    %%xmm2, %%xmm0 \n"\
2006
 stereo("addps    %%xmm3, %%xmm1 \n")\
2007
        "add $1024, %1 \n"\
2008
        "add $32, %2 \n"\
2009
        "jl 2b \n"\
2010
        "movaps   %%xmm0,     (%3,%0) \n"\
2011
 stereo("movaps   %%xmm1, 1024(%3,%0) \n")\
2012
        "add $16, %0 \n"\
2013
        "jl 1b \n"\
2014
        :"+&r"(i), "=&r"(j), "=&r"(k)\
2015
        :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\
2016
        :"memory"\
2017
    );
2018

    
2019
static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
2020
{
2021
    int (*matrix_cmp)[2] = (int(*)[2])matrix;
2022
    intptr_t i,j,k;
2023

    
2024
    i = -len*sizeof(float);
2025
    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]))) {
2026
        MIX5(IF0,IF1);
2027
    } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {
2028
        MIX5(IF1,IF0);
2029
    } else {
2030
        DECLARE_ALIGNED_16(float, matrix_simd[in_ch][2][4]);
2031
        j = 2*in_ch*sizeof(float);
2032
        __asm__ volatile(
2033
            "1: \n"
2034
            "sub $8, %0 \n"
2035
            "movss     (%2,%0), %%xmm6 \n"
2036
            "movss    4(%2,%0), %%xmm7 \n"
2037
            "shufps $0, %%xmm6, %%xmm6 \n"
2038
            "shufps $0, %%xmm7, %%xmm7 \n"
2039
            "movaps %%xmm6,   (%1,%0,4) \n"
2040
            "movaps %%xmm7, 16(%1,%0,4) \n"
2041
            "jg 1b \n"
2042
            :"+&r"(j)
2043
            :"r"(matrix_simd), "r"(matrix)
2044
            :"memory"
2045
        );
2046
        if(out_ch == 2) {
2047
            MIX_MISC(IF1);
2048
        } else {
2049
            MIX_MISC(IF0);
2050
        }
2051
    }
2052
}
2053

    
2054
static void vector_fmul_3dnow(float *dst, const float *src, int len){
2055
    x86_reg i = (len-4)*4;
2056
    __asm__ volatile(
2057
        "1: \n\t"
2058
        "movq    (%1,%0), %%mm0 \n\t"
2059
        "movq   8(%1,%0), %%mm1 \n\t"
2060
        "pfmul   (%2,%0), %%mm0 \n\t"
2061
        "pfmul  8(%2,%0), %%mm1 \n\t"
2062
        "movq   %%mm0,  (%1,%0) \n\t"
2063
        "movq   %%mm1, 8(%1,%0) \n\t"
2064
        "sub  $16, %0 \n\t"
2065
        "jge 1b \n\t"
2066
        "femms  \n\t"
2067
        :"+r"(i)
2068
        :"r"(dst), "r"(src)
2069
        :"memory"
2070
    );
2071
}
2072
static void vector_fmul_sse(float *dst, const float *src, int len){
2073
    x86_reg i = (len-8)*4;
2074
    __asm__ volatile(
2075
        "1: \n\t"
2076
        "movaps    (%1,%0), %%xmm0 \n\t"
2077
        "movaps  16(%1,%0), %%xmm1 \n\t"
2078
        "mulps     (%2,%0), %%xmm0 \n\t"
2079
        "mulps   16(%2,%0), %%xmm1 \n\t"
2080
        "movaps  %%xmm0,   (%1,%0) \n\t"
2081
        "movaps  %%xmm1, 16(%1,%0) \n\t"
2082
        "sub  $32, %0 \n\t"
2083
        "jge 1b \n\t"
2084
        :"+r"(i)
2085
        :"r"(dst), "r"(src)
2086
        :"memory"
2087
    );
2088
}
2089

    
2090
static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
2091
    x86_reg i = len*4-16;
2092
    __asm__ volatile(
2093
        "1: \n\t"
2094
        "pswapd   8(%1), %%mm0 \n\t"
2095
        "pswapd    (%1), %%mm1 \n\t"
2096
        "pfmul  (%3,%0), %%mm0 \n\t"
2097
        "pfmul 8(%3,%0), %%mm1 \n\t"
2098
        "movq  %%mm0,  (%2,%0) \n\t"
2099
        "movq  %%mm1, 8(%2,%0) \n\t"
2100
        "add   $16, %1 \n\t"
2101
        "sub   $16, %0 \n\t"
2102
        "jge   1b \n\t"
2103
        :"+r"(i), "+r"(src1)
2104
        :"r"(dst), "r"(src0)
2105
    );
2106
    __asm__ volatile("femms");
2107
}
2108
static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
2109
    x86_reg i = len*4-32;
2110
    __asm__ volatile(
2111
        "1: \n\t"
2112
        "movaps        16(%1), %%xmm0 \n\t"
2113
        "movaps          (%1), %%xmm1 \n\t"
2114
        "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
2115
        "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
2116
        "mulps        (%3,%0), %%xmm0 \n\t"
2117
        "mulps      16(%3,%0), %%xmm1 \n\t"
2118
        "movaps     %%xmm0,   (%2,%0) \n\t"
2119
        "movaps     %%xmm1, 16(%2,%0) \n\t"
2120
        "add    $32, %1 \n\t"
2121
        "sub    $32, %0 \n\t"
2122
        "jge    1b \n\t"
2123
        :"+r"(i), "+r"(src1)
2124
        :"r"(dst), "r"(src0)
2125
    );
2126
}
2127

    
2128
static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1,
2129
                                      const float *src2, int src3, int len, int step){
2130
    x86_reg i = (len-4)*4;
2131
    if(step == 2 && src3 == 0){
2132
        dst += (len-4)*2;
2133
        __asm__ volatile(
2134
            "1: \n\t"
2135
            "movq   (%2,%0),  %%mm0 \n\t"
2136
            "movq  8(%2,%0),  %%mm1 \n\t"
2137
            "pfmul  (%3,%0),  %%mm0 \n\t"
2138
            "pfmul 8(%3,%0),  %%mm1 \n\t"
2139
            "pfadd  (%4,%0),  %%mm0 \n\t"
2140
            "pfadd 8(%4,%0),  %%mm1 \n\t"
2141
            "movd     %%mm0,   (%1) \n\t"
2142
            "movd     %%mm1, 16(%1) \n\t"
2143
            "psrlq      $32,  %%mm0 \n\t"
2144
            "psrlq      $32,  %%mm1 \n\t"
2145
            "movd     %%mm0,  8(%1) \n\t"
2146
            "movd     %%mm1, 24(%1) \n\t"
2147
            "sub  $32, %1 \n\t"
2148
            "sub  $16, %0 \n\t"
2149
            "jge  1b \n\t"
2150
            :"+r"(i), "+r"(dst)
2151
            :"r"(src0), "r"(src1), "r"(src2)
2152
            :"memory"
2153
        );
2154
    }
2155
    else if(step == 1 && src3 == 0){
2156
        __asm__ volatile(
2157
            "1: \n\t"
2158
            "movq    (%2,%0), %%mm0 \n\t"
2159
            "movq   8(%2,%0), %%mm1 \n\t"
2160
            "pfmul   (%3,%0), %%mm0 \n\t"
2161
            "pfmul  8(%3,%0), %%mm1 \n\t"
2162
            "pfadd   (%4,%0), %%mm0 \n\t"
2163
            "pfadd  8(%4,%0), %%mm1 \n\t"
2164
            "movq  %%mm0,   (%1,%0) \n\t"
2165
            "movq  %%mm1,  8(%1,%0) \n\t"
2166
            "sub  $16, %0 \n\t"
2167
            "jge  1b \n\t"
2168
            :"+r"(i)
2169
            :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2170
            :"memory"
2171
        );
2172
    }
2173
    else
2174
        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
2175
    __asm__ volatile("femms");
2176
}
2177
static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1,
2178
                                    const float *src2, int src3, int len, int step){
2179
    x86_reg i = (len-8)*4;
2180
    if(step == 2 && src3 == 0){
2181
        dst += (len-8)*2;
2182
        __asm__ volatile(
2183
            "1: \n\t"
2184
            "movaps   (%2,%0), %%xmm0 \n\t"
2185
            "movaps 16(%2,%0), %%xmm1 \n\t"
2186
            "mulps    (%3,%0), %%xmm0 \n\t"
2187
            "mulps  16(%3,%0), %%xmm1 \n\t"
2188
            "addps    (%4,%0), %%xmm0 \n\t"
2189
            "addps  16(%4,%0), %%xmm1 \n\t"
2190
            "movss     %%xmm0,   (%1) \n\t"
2191
            "movss     %%xmm1, 32(%1) \n\t"
2192
            "movhlps   %%xmm0, %%xmm2 \n\t"
2193
            "movhlps   %%xmm1, %%xmm3 \n\t"
2194
            "movss     %%xmm2, 16(%1) \n\t"
2195
            "movss     %%xmm3, 48(%1) \n\t"
2196
            "shufps $0xb1, %%xmm0, %%xmm0 \n\t"
2197
            "shufps $0xb1, %%xmm1, %%xmm1 \n\t"
2198
            "movss     %%xmm0,  8(%1) \n\t"
2199
            "movss     %%xmm1, 40(%1) \n\t"
2200
            "movhlps   %%xmm0, %%xmm2 \n\t"
2201
            "movhlps   %%xmm1, %%xmm3 \n\t"
2202
            "movss     %%xmm2, 24(%1) \n\t"
2203
            "movss     %%xmm3, 56(%1) \n\t"
2204
            "sub  $64, %1 \n\t"
2205
            "sub  $32, %0 \n\t"
2206
            "jge  1b \n\t"
2207
            :"+r"(i), "+r"(dst)
2208
            :"r"(src0), "r"(src1), "r"(src2)
2209
            :"memory"
2210
        );
2211
    }
2212
    else if(step == 1 && src3 == 0){
2213
        __asm__ volatile(
2214
            "1: \n\t"
2215
            "movaps   (%2,%0), %%xmm0 \n\t"
2216
            "movaps 16(%2,%0), %%xmm1 \n\t"
2217
            "mulps    (%3,%0), %%xmm0 \n\t"
2218
            "mulps  16(%3,%0), %%xmm1 \n\t"
2219
            "addps    (%4,%0), %%xmm0 \n\t"
2220
            "addps  16(%4,%0), %%xmm1 \n\t"
2221
            "movaps %%xmm0,   (%1,%0) \n\t"
2222
            "movaps %%xmm1, 16(%1,%0) \n\t"
2223
            "sub  $32, %0 \n\t"
2224
            "jge  1b \n\t"
2225
            :"+r"(i)
2226
            :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2227
            :"memory"
2228
        );
2229
    }
2230
    else
2231
        ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
2232
}
2233

    
2234
static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
2235
                                      const float *win, float add_bias, int len){
2236
#if HAVE_6REGS
2237
    if(add_bias == 0){
2238
        x86_reg i = -len*4;
2239
        x86_reg j = len*4-8;
2240
        __asm__ volatile(
2241
            "1: \n"
2242
            "pswapd  (%5,%1), %%mm1 \n"
2243
            "movq    (%5,%0), %%mm0 \n"
2244
            "pswapd  (%4,%1), %%mm5 \n"
2245
            "movq    (%3,%0), %%mm4 \n"
2246
            "movq      %%mm0, %%mm2 \n"
2247
            "movq      %%mm1, %%mm3 \n"
2248
            "pfmul     %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
2249
            "pfmul     %%mm5, %%mm3 \n" // src1[    j]*win[len+j]
2250
            "pfmul     %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
2251
            "pfmul     %%mm5, %%mm0 \n" // src1[    j]*win[len+i]
2252
            "pfadd     %%mm3, %%mm2 \n"
2253
            "pfsub     %%mm0, %%mm1 \n"
2254
            "pswapd    %%mm2, %%mm2 \n"
2255
            "movq      %%mm1, (%2,%0) \n"
2256
            "movq      %%mm2, (%2,%1) \n"
2257
            "sub $8, %1 \n"
2258
            "add $8, %0 \n"
2259
            "jl 1b \n"
2260
            "femms \n"
2261
            :"+r"(i), "+r"(j)
2262
            :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2263
        );
2264
    }else
2265
#endif
2266
        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2267
}
2268

    
2269
static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
2270
                                   const float *win, float add_bias, int len){
2271
#if HAVE_6REGS
2272
    if(add_bias == 0){
2273
        x86_reg i = -len*4;
2274
        x86_reg j = len*4-16;
2275
        __asm__ volatile(
2276
            "1: \n"
2277
            "movaps       (%5,%1), %%xmm1 \n"
2278
            "movaps       (%5,%0), %%xmm0 \n"
2279
            "movaps       (%4,%1), %%xmm5 \n"
2280
            "movaps       (%3,%0), %%xmm4 \n"
2281
            "shufps $0x1b, %%xmm1, %%xmm1 \n"
2282
            "shufps $0x1b, %%xmm5, %%xmm5 \n"
2283
            "movaps        %%xmm0, %%xmm2 \n"
2284
            "movaps        %%xmm1, %%xmm3 \n"
2285
            "mulps         %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
2286
            "mulps         %%xmm5, %%xmm3 \n" // src1[    j]*win[len+j]
2287
            "mulps         %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
2288
            "mulps         %%xmm5, %%xmm0 \n" // src1[    j]*win[len+i]
2289
            "addps         %%xmm3, %%xmm2 \n"
2290
            "subps         %%xmm0, %%xmm1 \n"
2291
            "shufps $0x1b, %%xmm2, %%xmm2 \n"
2292
            "movaps        %%xmm1, (%2,%0) \n"
2293
            "movaps        %%xmm2, (%2,%1) \n"
2294
            "sub $16, %1 \n"
2295
            "add $16, %0 \n"
2296
            "jl 1b \n"
2297
            :"+r"(i), "+r"(j)
2298
            :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2299
        );
2300
    }else
2301
#endif
2302
        ff_vector_fmul_window_c(dst, src0, src1, win, add_bias, len);
2303
}
2304

    
2305
static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len)
2306
{
2307
    x86_reg i = -4*len;
2308
    __asm__ volatile(
2309
        "movss  %3, %%xmm4 \n"
2310
        "shufps $0, %%xmm4, %%xmm4 \n"
2311
        "1: \n"
2312
        "cvtpi2ps   (%2,%0), %%xmm0 \n"
2313
        "cvtpi2ps  8(%2,%0), %%xmm1 \n"
2314
        "cvtpi2ps 16(%2,%0), %%xmm2 \n"
2315
        "cvtpi2ps 24(%2,%0), %%xmm3 \n"
2316
        "movlhps  %%xmm1,    %%xmm0 \n"
2317
        "movlhps  %%xmm3,    %%xmm2 \n"
2318
        "mulps    %%xmm4,    %%xmm0 \n"
2319
        "mulps    %%xmm4,    %%xmm2 \n"
2320
        "movaps   %%xmm0,   (%1,%0) \n"
2321
        "movaps   %%xmm2, 16(%1,%0) \n"
2322
        "add $32, %0 \n"
2323
        "jl 1b \n"
2324
        :"+r"(i)
2325
        :"r"(dst+len), "r"(src+len), "m"(mul)
2326
    );
2327
}
2328

    
2329
static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len)
2330
{
2331
    x86_reg i = -4*len;
2332
    __asm__ volatile(
2333
        "movss  %3, %%xmm4 \n"
2334
        "shufps $0, %%xmm4, %%xmm4 \n"
2335
        "1: \n"
2336
        "cvtdq2ps   (%2,%0), %%xmm0 \n"
2337
        "cvtdq2ps 16(%2,%0), %%xmm1 \n"
2338
        "mulps    %%xmm4,    %%xmm0 \n"
2339
        "mulps    %%xmm4,    %%xmm1 \n"
2340
        "movaps   %%xmm0,   (%1,%0) \n"
2341
        "movaps   %%xmm1, 16(%1,%0) \n"
2342
        "add $32, %0 \n"
2343
        "jl 1b \n"
2344
        :"+r"(i)
2345
        :"r"(dst+len), "r"(src+len), "m"(mul)
2346
    );
2347
}
2348

    
2349
static void vector_clipf_sse(float *dst, const float *src, float min, float max,
2350
                             int len)
2351
{
2352
    x86_reg i = (len-16)*4;
2353
    __asm__ volatile(
2354
        "movss  %3, %%xmm4 \n"
2355
        "movss  %4, %%xmm5 \n"
2356
        "shufps $0, %%xmm4, %%xmm4 \n"
2357
        "shufps $0, %%xmm5, %%xmm5 \n"
2358
        "1: \n\t"
2359
        "movaps    (%2,%0), %%xmm0 \n\t" // 3/1 on intel
2360
        "movaps  16(%2,%0), %%xmm1 \n\t"
2361
        "movaps  32(%2,%0), %%xmm2 \n\t"
2362
        "movaps  48(%2,%0), %%xmm3 \n\t"
2363
        "maxps      %%xmm4, %%xmm0 \n\t"
2364
        "maxps      %%xmm4, %%xmm1 \n\t"
2365
        "maxps      %%xmm4, %%xmm2 \n\t"
2366
        "maxps      %%xmm4, %%xmm3 \n\t"
2367
        "minps      %%xmm5, %%xmm0 \n\t"
2368
        "minps      %%xmm5, %%xmm1 \n\t"
2369
        "minps      %%xmm5, %%xmm2 \n\t"
2370
        "minps      %%xmm5, %%xmm3 \n\t"
2371
        "movaps  %%xmm0,   (%1,%0) \n\t"
2372
        "movaps  %%xmm1, 16(%1,%0) \n\t"
2373
        "movaps  %%xmm2, 32(%1,%0) \n\t"
2374
        "movaps  %%xmm3, 48(%1,%0) \n\t"
2375
        "sub  $64, %0 \n\t"
2376
        "jge 1b \n\t"
2377
        :"+r"(i)
2378
        :"r"(dst), "r"(src), "m"(min), "m"(max)
2379
        :"memory"
2380
    );
2381
}
2382

    
2383
static void float_to_int16_3dnow(int16_t *dst, const float *src, long len){
2384
    x86_reg reglen = len;
2385
    // not bit-exact: pf2id uses different rounding than C and SSE
2386
    __asm__ volatile(
2387
        "add        %0          , %0        \n\t"
2388
        "lea         (%2,%0,2)  , %2        \n\t"
2389
        "add        %0          , %1        \n\t"
2390
        "neg        %0                      \n\t"
2391
        "1:                                 \n\t"
2392
        "pf2id       (%2,%0,2)  , %%mm0     \n\t"
2393
        "pf2id      8(%2,%0,2)  , %%mm1     \n\t"
2394
        "pf2id     16(%2,%0,2)  , %%mm2     \n\t"
2395
        "pf2id     24(%2,%0,2)  , %%mm3     \n\t"
2396
        "packssdw   %%mm1       , %%mm0     \n\t"
2397
        "packssdw   %%mm3       , %%mm2     \n\t"
2398
        "movq       %%mm0       ,  (%1,%0)  \n\t"
2399
        "movq       %%mm2       , 8(%1,%0)  \n\t"
2400
        "add        $16         , %0        \n\t"
2401
        " js 1b                             \n\t"
2402
        "femms                              \n\t"
2403
        :"+r"(reglen), "+r"(dst), "+r"(src)
2404
    );
2405
}
2406
static void float_to_int16_sse(int16_t *dst, const float *src, long len){
2407
    x86_reg reglen = len;
2408
    __asm__ volatile(
2409
        "add        %0          , %0        \n\t"
2410
        "lea         (%2,%0,2)  , %2        \n\t"
2411
        "add        %0          , %1        \n\t"
2412
        "neg        %0                      \n\t"
2413
        "1:                                 \n\t"
2414
        "cvtps2pi    (%2,%0,2)  , %%mm0     \n\t"
2415
        "cvtps2pi   8(%2,%0,2)  , %%mm1     \n\t"
2416
        "cvtps2pi  16(%2,%0,2)  , %%mm2     \n\t"
2417
        "cvtps2pi  24(%2,%0,2)  , %%mm3     \n\t"
2418
        "packssdw   %%mm1       , %%mm0     \n\t"
2419
        "packssdw   %%mm3       , %%mm2     \n\t"
2420
        "movq       %%mm0       ,  (%1,%0)  \n\t"
2421
        "movq       %%mm2       , 8(%1,%0)  \n\t"
2422
        "add        $16         , %0        \n\t"
2423
        " js 1b                             \n\t"
2424
        "emms                               \n\t"
2425
        :"+r"(reglen), "+r"(dst), "+r"(src)
2426
    );
2427
}
2428

    
2429
static void float_to_int16_sse2(int16_t *dst, const float *src, long len){
2430
    x86_reg reglen = len;
2431
    __asm__ volatile(
2432
        "add        %0          , %0        \n\t"
2433
        "lea         (%2,%0,2)  , %2        \n\t"
2434
        "add        %0          , %1        \n\t"
2435
        "neg        %0                      \n\t"
2436
        "1:                                 \n\t"
2437
        "cvtps2dq    (%2,%0,2)  , %%xmm0    \n\t"
2438
        "cvtps2dq  16(%2,%0,2)  , %%xmm1    \n\t"
2439
        "packssdw   %%xmm1      , %%xmm0    \n\t"
2440
        "movdqa     %%xmm0      ,  (%1,%0)  \n\t"
2441
        "add        $16         , %0        \n\t"
2442
        " js 1b                             \n\t"
2443
        :"+r"(reglen), "+r"(dst), "+r"(src)
2444
    );
2445
}
2446

    
2447
#if HAVE_YASM
2448
void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len);
2449
void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len);
2450
void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len);
2451
void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top);
2452
void ff_x264_deblock_v_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
2453
void ff_x264_deblock_h_luma_sse2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
2454
void ff_x264_deblock_v8_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
2455
void ff_x264_deblock_h_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta);
2456
#if ARCH_X86_32
2457
static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
2458
{
2459
    ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
2460
    ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
2461
}
2462
#endif
2463
void ff_x264_deblock_v_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta);
2464
void ff_x264_deblock_h_luma_intra_sse2(uint8_t *pix, int stride, int alpha, int beta);
2465
#else
2466
#define ff_float_to_int16_interleave6_sse(a,b,c)   float_to_int16_interleave_misc_sse(a,b,c,6)
2467
#define ff_float_to_int16_interleave6_3dnow(a,b,c) float_to_int16_interleave_misc_3dnow(a,b,c,6)
2468
#define ff_float_to_int16_interleave6_3dn2(a,b,c)  float_to_int16_interleave_misc_3dnow(a,b,c,6)
2469
#endif
2470
#define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse
2471

    
2472
#define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \
2473
/* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\
2474
static av_noinline void float_to_int16_interleave_misc_##cpu(int16_t *dst, const float **src, long len, int channels){\
2475
    DECLARE_ALIGNED_16(int16_t, tmp[len]);\
2476
    int i,j,c;\
2477
    for(c=0; c<channels; c++){\
2478
        float_to_int16_##cpu(tmp, src[c], len);\
2479
        for(i=0, j=c; i<len; i++, j+=channels)\
2480
            dst[j] = tmp[i];\
2481
    }\
2482
}\
2483
\
2484
static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, long len, int channels){\
2485
    if(channels==1)\
2486
        float_to_int16_##cpu(dst, src[0], len);\
2487
    else if(channels==2){\
2488
        x86_reg reglen = len; \
2489
        const float *src0 = src[0];\
2490
        const float *src1 = src[1];\
2491
        __asm__ volatile(\
2492
            "shl $2, %0 \n"\
2493
            "add %0, %1 \n"\
2494
            "add %0, %2 \n"\
2495
            "add %0, %3 \n"\
2496
            "neg %0 \n"\
2497
            body\
2498
            :"+r"(reglen), "+r"(dst), "+r"(src0), "+r"(src1)\
2499
        );\
2500
    }else if(channels==6){\
2501
        ff_float_to_int16_interleave6_##cpu(dst, src, len);\
2502
    }else\
2503
        float_to_int16_interleave_misc_##cpu(dst, src, len, channels);\
2504
}
2505

    
2506
FLOAT_TO_INT16_INTERLEAVE(3dnow,
2507
    "1:                         \n"
2508
    "pf2id     (%2,%0), %%mm0   \n"
2509
    "pf2id    8(%2,%0), %%mm1   \n"
2510
    "pf2id     (%3,%0), %%mm2   \n"
2511
    "pf2id    8(%3,%0), %%mm3   \n"
2512
    "packssdw    %%mm1, %%mm0   \n"
2513
    "packssdw    %%mm3, %%mm2   \n"
2514
    "movq        %%mm0, %%mm1   \n"
2515
    "punpcklwd   %%mm2, %%mm0   \n"
2516
    "punpckhwd   %%mm2, %%mm1   \n"
2517
    "movq        %%mm0,  (%1,%0)\n"
2518
    "movq        %%mm1, 8(%1,%0)\n"
2519
    "add $16, %0                \n"
2520
    "js 1b                      \n"
2521
    "femms                      \n"
2522
)
2523

    
2524
FLOAT_TO_INT16_INTERLEAVE(sse,
2525
    "1:                         \n"
2526
    "cvtps2pi  (%2,%0), %%mm0   \n"
2527
    "cvtps2pi 8(%2,%0), %%mm1   \n"
2528
    "cvtps2pi  (%3,%0), %%mm2   \n"
2529
    "cvtps2pi 8(%3,%0), %%mm3   \n"
2530
    "packssdw    %%mm1, %%mm0   \n"
2531
    "packssdw    %%mm3, %%mm2   \n"
2532
    "movq        %%mm0, %%mm1   \n"
2533
    "punpcklwd   %%mm2, %%mm0   \n"
2534
    "punpckhwd   %%mm2, %%mm1   \n"
2535
    "movq        %%mm0,  (%1,%0)\n"
2536
    "movq        %%mm1, 8(%1,%0)\n"
2537
    "add $16, %0                \n"
2538
    "js 1b                      \n"
2539
    "emms                       \n"
2540
)
2541

    
2542
FLOAT_TO_INT16_INTERLEAVE(sse2,
2543
    "1:                         \n"
2544
    "cvtps2dq  (%2,%0), %%xmm0  \n"
2545
    "cvtps2dq  (%3,%0), %%xmm1  \n"
2546
    "packssdw   %%xmm1, %%xmm0  \n"
2547
    "movhlps    %%xmm0, %%xmm1  \n"
2548
    "punpcklwd  %%xmm1, %%xmm0  \n"
2549
    "movdqa     %%xmm0, (%1,%0) \n"
2550
    "add $16, %0                \n"
2551
    "js 1b                      \n"
2552
)
2553

    
2554
static void float_to_int16_interleave_3dn2(int16_t *dst, const float **src, long len, int channels){
2555
    if(channels==6)
2556
        ff_float_to_int16_interleave6_3dn2(dst, src, len);
2557
    else
2558
        float_to_int16_interleave_3dnow(dst, src, len, channels);
2559
}
2560

    
2561

    
2562
void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width);
2563
void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width);
2564
void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
2565
void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
2566
void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2567
                                   int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2568
void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2569
                                  int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
2570

    
2571

    
2572
static void add_int16_sse2(int16_t * v1, int16_t * v2, int order)
2573
{
2574
    x86_reg o = -(order << 1);
2575
    v1 += order;
2576
    v2 += order;
2577
    __asm__ volatile(
2578
        "1:                          \n\t"
2579
        "movdqu   (%1,%2),   %%xmm0  \n\t"
2580
        "movdqu 16(%1,%2),   %%xmm1  \n\t"
2581
        "paddw    (%0,%2),   %%xmm0  \n\t"
2582
        "paddw  16(%0,%2),   %%xmm1  \n\t"
2583
        "movdqa   %%xmm0,    (%0,%2) \n\t"
2584
        "movdqa   %%xmm1,  16(%0,%2) \n\t"
2585
        "add      $32,       %2      \n\t"
2586
        "js       1b                 \n\t"
2587
        : "+r"(v1), "+r"(v2), "+r"(o)
2588
    );
2589
}
2590

    
2591
static void sub_int16_sse2(int16_t * v1, int16_t * v2, int order)
2592
{
2593
    x86_reg o = -(order << 1);
2594
    v1 += order;
2595
    v2 += order;
2596
    __asm__ volatile(
2597
        "1:                           \n\t"
2598
        "movdqa    (%0,%2),   %%xmm0  \n\t"
2599
        "movdqa  16(%0,%2),   %%xmm2  \n\t"
2600
        "movdqu    (%1,%2),   %%xmm1  \n\t"
2601
        "movdqu  16(%1,%2),   %%xmm3  \n\t"
2602
        "psubw     %%xmm1,    %%xmm0  \n\t"
2603
        "psubw     %%xmm3,    %%xmm2  \n\t"
2604
        "movdqa    %%xmm0,    (%0,%2) \n\t"
2605
        "movdqa    %%xmm2,  16(%0,%2) \n\t"
2606
        "add       $32,       %2      \n\t"
2607
        "js        1b                 \n\t"
2608
        : "+r"(v1), "+r"(v2), "+r"(o)
2609
    );
2610
}
2611

    
2612
static int32_t scalarproduct_int16_sse2(int16_t * v1, int16_t * v2, int order, int shift)
2613
{
2614
    int res = 0;
2615
    DECLARE_ALIGNED_16(xmm_reg, sh);
2616
    x86_reg o = -(order << 1);
2617

    
2618
    v1 += order;
2619
    v2 += order;
2620
    sh.a = shift;
2621
    __asm__ volatile(
2622
        "pxor      %%xmm7,  %%xmm7        \n\t"
2623
        "1:                               \n\t"
2624
        "movdqu    (%0,%3), %%xmm0        \n\t"
2625
        "movdqu  16(%0,%3), %%xmm1        \n\t"
2626
        "pmaddwd   (%1,%3), %%xmm0        \n\t"
2627
        "pmaddwd 16(%1,%3), %%xmm1        \n\t"
2628
        "paddd     %%xmm0,  %%xmm7        \n\t"
2629
        "paddd     %%xmm1,  %%xmm7        \n\t"
2630
        "add       $32,     %3            \n\t"
2631
        "js        1b                     \n\t"
2632
        "movhlps   %%xmm7,  %%xmm2        \n\t"
2633
        "paddd     %%xmm2,  %%xmm7        \n\t"
2634
        "psrad     %4,      %%xmm7        \n\t"
2635
        "pshuflw   $0x4E,   %%xmm7,%%xmm2 \n\t"
2636
        "paddd     %%xmm2,  %%xmm7        \n\t"
2637
        "movd      %%xmm7,  %2            \n\t"
2638
        : "+r"(v1), "+r"(v2), "=r"(res), "+r"(o)
2639
        : "m"(sh)
2640
    );
2641
    return res;
2642
}
2643

    
2644
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2645
{
2646
    mm_flags = mm_support();
2647

    
2648
    if (avctx->dsp_mask) {
2649
        if (avctx->dsp_mask & FF_MM_FORCE)
2650
            mm_flags |= (avctx->dsp_mask & 0xffff);
2651
        else
2652
            mm_flags &= ~(avctx->dsp_mask & 0xffff);
2653
    }
2654

    
2655
#if 0
2656
    av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2657
    if (mm_flags & FF_MM_MMX)
2658
        av_log(avctx, AV_LOG_INFO, " mmx");
2659
    if (mm_flags & FF_MM_MMX2)
2660
        av_log(avctx, AV_LOG_INFO, " mmx2");
2661
    if (mm_flags & FF_MM_3DNOW)
2662
        av_log(avctx, AV_LOG_INFO, " 3dnow");
2663
    if (mm_flags & FF_MM_SSE)
2664
        av_log(avctx, AV_LOG_INFO, " sse");
2665
    if (mm_flags & FF_MM_SSE2)
2666
        av_log(avctx, AV_LOG_INFO, " sse2");
2667
    av_log(avctx, AV_LOG_INFO, "\n");
2668
#endif
2669

    
2670
    if (mm_flags & FF_MM_MMX) {
2671
        const int idct_algo= avctx->idct_algo;
2672

    
2673
        if(avctx->lowres==0){
2674
            if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2675
                c->idct_put= ff_simple_idct_put_mmx;
2676
                c->idct_add= ff_simple_idct_add_mmx;
2677
                c->idct    = ff_simple_idct_mmx;
2678
                c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2679
#if CONFIG_GPL
2680
            }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2681
                if(mm_flags & FF_MM_MMX2){
2682
                    c->idct_put= ff_libmpeg2mmx2_idct_put;
2683
                    c->idct_add= ff_libmpeg2mmx2_idct_add;
2684
                    c->idct    = ff_mmxext_idct;
2685
                }else{
2686
                    c->idct_put= ff_libmpeg2mmx_idct_put;
2687
                    c->idct_add= ff_libmpeg2mmx_idct_add;
2688
                    c->idct    = ff_mmx_idct;
2689
                }
2690
                c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2691
#endif
2692
            }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&
2693
                     idct_algo==FF_IDCT_VP3){
2694
                if(mm_flags & FF_MM_SSE2){
2695
                    c->idct_put= ff_vp3_idct_put_sse2;
2696
                    c->idct_add= ff_vp3_idct_add_sse2;
2697
                    c->idct    = ff_vp3_idct_sse2;
2698
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2699
                }else{
2700
                    c->idct_put= ff_vp3_idct_put_mmx;
2701
                    c->idct_add= ff_vp3_idct_add_mmx;
2702
                    c->idct    = ff_vp3_idct_mmx;
2703
                    c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2704
                }
2705
            }else if(idct_algo==FF_IDCT_CAVS){
2706
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2707
            }else if(idct_algo==FF_IDCT_XVIDMMX){
2708
                if(mm_flags & FF_MM_SSE2){
2709
                    c->idct_put= ff_idct_xvid_sse2_put;
2710
                    c->idct_add= ff_idct_xvid_sse2_add;
2711
                    c->idct    = ff_idct_xvid_sse2;
2712
                    c->idct_permutation_type= FF_SSE2_IDCT_PERM;
2713
                }else if(mm_flags & FF_MM_MMX2){
2714
                    c->idct_put= ff_idct_xvid_mmx2_put;
2715
                    c->idct_add= ff_idct_xvid_mmx2_add;
2716
                    c->idct    = ff_idct_xvid_mmx2;
2717
                }else{
2718
                    c->idct_put= ff_idct_xvid_mmx_put;
2719
                    c->idct_add= ff_idct_xvid_mmx_add;
2720
                    c->idct    = ff_idct_xvid_mmx;
2721
                }
2722
            }
2723
        }
2724

    
2725
        c->put_pixels_clamped = put_pixels_clamped_mmx;
2726
        c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;
2727
        c->add_pixels_clamped = add_pixels_clamped_mmx;
2728
        c->clear_block  = clear_block_mmx;
2729
        c->clear_blocks = clear_blocks_mmx;
2730
        if (mm_flags & FF_MM_SSE){
2731
            c->clear_block  = clear_block_sse;
2732
            c->clear_blocks = clear_blocks_sse;
2733
        }
2734

    
2735
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2736
        c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
2737
        c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
2738
        c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
2739
        c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU
2740

    
2741
        SET_HPEL_FUNCS(put, 0, 16, mmx);
2742
        SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2743
        SET_HPEL_FUNCS(avg, 0, 16, mmx);
2744
        SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2745
        SET_HPEL_FUNCS(put, 1, 8, mmx);
2746
        SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
2747
        SET_HPEL_FUNCS(avg, 1, 8, mmx);
2748
        SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
2749

    
2750
        c->gmc= gmc_mmx;
2751

    
2752
        c->add_bytes= add_bytes_mmx;
2753
        c->add_bytes_l2= add_bytes_l2_mmx;
2754

    
2755
        c->draw_edges = draw_edges_mmx;
2756

    
2757
        if (CONFIG_ANY_H263) {
2758
            c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2759
            c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2760
        }
2761
        c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_rnd;
2762
        c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;
2763
        c->put_no_rnd_vc1_chroma_pixels_tab[0]= put_vc1_chroma_mc8_mmx_nornd;
2764

    
2765
        c->put_rv40_chroma_pixels_tab[0]= put_rv40_chroma_mc8_mmx;
2766
        c->put_rv40_chroma_pixels_tab[1]= put_rv40_chroma_mc4_mmx;
2767

    
2768
        c->h264_idct_dc_add=
2769
        c->h264_idct_add= ff_h264_idct_add_mmx;
2770
        c->h264_idct8_dc_add=
2771
        c->h264_idct8_add= ff_h264_idct8_add_mmx;
2772

    
2773
        c->h264_idct_add16     = ff_h264_idct_add16_mmx;
2774
        c->h264_idct8_add4     = ff_h264_idct8_add4_mmx;
2775
        c->h264_idct_add8      = ff_h264_idct_add8_mmx;
2776
        c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx;
2777

    
2778
        if (CONFIG_VP6_DECODER) {
2779
            c->vp6_filter_diag4 = ff_vp6_filter_diag4_mmx;
2780
        }
2781

    
2782
        if (mm_flags & FF_MM_MMX2) {
2783
            c->prefetch = prefetch_mmx2;
2784

    
2785
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2786
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2787

    
2788
            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2789
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2790
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2791

    
2792
            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2793
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2794

    
2795
            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2796
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2797
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2798

    
2799
            c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;
2800
            c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;
2801
            c->h264_idct_add16     = ff_h264_idct_add16_mmx2;
2802
            c->h264_idct8_add4     = ff_h264_idct8_add4_mmx2;
2803
            c->h264_idct_add8      = ff_h264_idct_add8_mmx2;
2804
            c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2;
2805

    
2806
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2807
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2808
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2809
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2810
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2811
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2812
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2813

    
2814
                if (CONFIG_VP3_DECODER) {
2815
                    c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2;
2816
                    c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;
2817
                }
2818
            }
2819

    
2820
#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2821
            c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \
2822
            c->PFX ## _pixels_tab[IDX][ 1] = PFX ## SIZE ## _mc10_ ## CPU; \
2823
            c->PFX ## _pixels_tab[IDX][ 2] = PFX ## SIZE ## _mc20_ ## CPU; \
2824
            c->PFX ## _pixels_tab[IDX][ 3] = PFX ## SIZE ## _mc30_ ## CPU; \
2825
            c->PFX ## _pixels_tab[IDX][ 4] = PFX ## SIZE ## _mc01_ ## CPU; \
2826
            c->PFX ## _pixels_tab[IDX][ 5] = PFX ## SIZE ## _mc11_ ## CPU; \
2827
            c->PFX ## _pixels_tab[IDX][ 6] = PFX ## SIZE ## _mc21_ ## CPU; \
2828
            c->PFX ## _pixels_tab[IDX][ 7] = PFX ## SIZE ## _mc31_ ## CPU; \
2829
            c->PFX ## _pixels_tab[IDX][ 8] = PFX ## SIZE ## _mc02_ ## CPU; \
2830
            c->PFX ## _pixels_tab[IDX][ 9] = PFX ## SIZE ## _mc12_ ## CPU; \
2831
            c->PFX ## _pixels_tab[IDX][10] = PFX ## SIZE ## _mc22_ ## CPU; \
2832
            c->PFX ## _pixels_tab[IDX][11] = PFX ## SIZE ## _mc32_ ## CPU; \
2833
            c->PFX ## _pixels_tab[IDX][12] = PFX ## SIZE ## _mc03_ ## CPU; \
2834
            c->PFX ## _pixels_tab[IDX][13] = PFX ## SIZE ## _mc13_ ## CPU; \
2835
            c->PFX ## _pixels_tab[IDX][14] = PFX ## SIZE ## _mc23_ ## CPU; \
2836
            c->PFX ## _pixels_tab[IDX][15] = PFX ## SIZE ## _mc33_ ## CPU
2837

    
2838
            SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2);
2839
            SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2);
2840
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2);
2841
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2);
2842
            SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2);
2843
            SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2);
2844

    
2845
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2);
2846
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2);
2847
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2);
2848
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2);
2849
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2);
2850
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2);
2851

    
2852
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2);
2853
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2);
2854
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2);
2855
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2);
2856

    
2857
            c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_mmx2;
2858
            c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_mmx2;
2859

    
2860
            c->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_vc1_chroma_mc8_mmx2_nornd;
2861

    
2862
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2_rnd;
2863
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;
2864
            c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;
2865
            c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;
2866
            c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
2867
            c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
2868
            c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
2869
            c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;
2870
            c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;
2871
            c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;
2872
            c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
2873

    
2874
            c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
2875
            c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
2876
            c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
2877
            c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
2878
            c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
2879
            c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
2880
            c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
2881
            c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
2882

    
2883
            c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
2884
            c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
2885
            c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
2886
            c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
2887
            c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
2888
            c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
2889
            c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
2890
            c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
2891

    
2892
#if HAVE_YASM
2893
            c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
2894
#endif
2895
#if HAVE_7REGS && HAVE_TEN_OPERANDS
2896
            if( mm_flags&FF_MM_3DNOW )
2897
                c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
2898
#endif
2899

    
2900
            if (CONFIG_CAVS_DECODER)
2901
                ff_cavsdsp_init_mmx2(c, avctx);
2902

    
2903
            if (CONFIG_VC1_DECODER)
2904
                ff_vc1dsp_init_mmx(c, avctx);
2905

    
2906
            c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2;
2907
        } else if (mm_flags & FF_MM_3DNOW) {
2908
            c->prefetch = prefetch_3dnow;
2909

    
2910
            c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
2911
            c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
2912

    
2913
            c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
2914
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
2915
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
2916

    
2917
            c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
2918
            c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
2919

    
2920
            c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
2921
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
2922
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
2923

    
2924
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2925
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
2926
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
2927
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
2928
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
2929
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
2930
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
2931
            }
2932

    
2933
            SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow);
2934
            SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow);
2935
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow);
2936
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow);
2937
            SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow);
2938
            SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow);
2939

    
2940
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow);
2941
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow);
2942
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow);
2943
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow);
2944
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow);
2945
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow);
2946

    
2947
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow);
2948
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow);
2949
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow);
2950
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow);
2951

    
2952
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow_rnd;
2953
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;
2954

    
2955
            c->avg_rv40_chroma_pixels_tab[0]= avg_rv40_chroma_mc8_3dnow;
2956
            c->avg_rv40_chroma_pixels_tab[1]= avg_rv40_chroma_mc4_3dnow;
2957

    
2958
            if (CONFIG_CAVS_DECODER)
2959
                ff_cavsdsp_init_3dnow(c, avctx);
2960
        }
2961

    
2962

    
2963
#define H264_QPEL_FUNCS(x, y, CPU)\
2964
            c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\
2965
            c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\
2966
            c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\
2967
            c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;
2968
        if((mm_flags & FF_MM_SSE2) && !(mm_flags & FF_MM_3DNOW)){
2969
            // these functions are slower than mmx on AMD, but faster on Intel
2970
/* FIXME works in most codecs, but crashes svq1 due to unaligned chroma
2971
            c->put_pixels_tab[0][0] = put_pixels16_sse2;
2972
            c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
2973
*/
2974
            H264_QPEL_FUNCS(0, 0, sse2);
2975
        }
2976
        if(mm_flags & FF_MM_SSE2){
2977
            c->h264_idct8_add = ff_h264_idct8_add_sse2;
2978
            c->h264_idct8_add4= ff_h264_idct8_add4_sse2;
2979

    
2980
            H264_QPEL_FUNCS(0, 1, sse2);
2981
            H264_QPEL_FUNCS(0, 2, sse2);
2982
            H264_QPEL_FUNCS(0, 3, sse2);
2983
            H264_QPEL_FUNCS(1, 1, sse2);
2984
            H264_QPEL_FUNCS(1, 2, sse2);
2985
            H264_QPEL_FUNCS(1, 3, sse2);
2986
            H264_QPEL_FUNCS(2, 1, sse2);
2987
            H264_QPEL_FUNCS(2, 2, sse2);
2988
            H264_QPEL_FUNCS(2, 3, sse2);
2989
            H264_QPEL_FUNCS(3, 1, sse2);
2990
            H264_QPEL_FUNCS(3, 2, sse2);
2991
            H264_QPEL_FUNCS(3, 3, sse2);
2992

    
2993
            if (CONFIG_VP6_DECODER) {
2994
                c->vp6_filter_diag4 = ff_vp6_filter_diag4_sse2;
2995
            }
2996
        }
2997
#if HAVE_SSSE3
2998
        if(mm_flags & FF_MM_SSSE3){
2999
            H264_QPEL_FUNCS(1, 0, ssse3);
3000
            H264_QPEL_FUNCS(1, 1, ssse3);
3001
            H264_QPEL_FUNCS(1, 2, ssse3);
3002
            H264_QPEL_FUNCS(1, 3, ssse3);
3003
            H264_QPEL_FUNCS(2, 0, ssse3);
3004
            H264_QPEL_FUNCS(2, 1, ssse3);
3005
            H264_QPEL_FUNCS(2, 2, ssse3);
3006
            H264_QPEL_FUNCS(2, 3, ssse3);
3007
            H264_QPEL_FUNCS(3, 0, ssse3);
3008
            H264_QPEL_FUNCS(3, 1, ssse3);
3009
            H264_QPEL_FUNCS(3, 2, ssse3);
3010
            H264_QPEL_FUNCS(3, 3, ssse3);
3011
            c->put_no_rnd_vc1_chroma_pixels_tab[0]= put_vc1_chroma_mc8_ssse3_nornd;
3012
            c->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_vc1_chroma_mc8_ssse3_nornd;
3013
            c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_ssse3_rnd;
3014
            c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_ssse3_rnd;
3015
            c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_ssse3;
3016
            c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_ssse3;
3017
            c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
3018
        }
3019
#endif
3020

    
3021
#if CONFIG_GPL && HAVE_YASM
3022
        if (mm_flags & FF_MM_MMX2){
3023
#if ARCH_X86_32
3024
            c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext;
3025
            c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext;
3026
#endif
3027
            if( mm_flags&FF_MM_SSE2 ){
3028
#if ARCH_X86_64 || !defined(__ICC) || __ICC > 1110
3029
                c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2;
3030
                c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2;
3031
                c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2;
3032
                c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2;
3033
#endif
3034
                c->h264_idct_add16 = ff_h264_idct_add16_sse2;
3035
                c->h264_idct_add8  = ff_h264_idct_add8_sse2;
3036
                c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2;
3037
            }
3038
        }
3039
#endif
3040

    
3041
#if CONFIG_SNOW_DECODER
3042
        if(mm_flags & FF_MM_SSE2 & 0){
3043
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
3044
#if HAVE_7REGS
3045
            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
3046
#endif
3047
            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
3048
        }
3049
        else{
3050
            if(mm_flags & FF_MM_MMX2){
3051
            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
3052
#if HAVE_7REGS
3053
            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
3054
#endif
3055
            }
3056
            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
3057
        }
3058
#endif
3059

    
3060
        if(mm_flags & FF_MM_3DNOW){
3061
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
3062
            c->vector_fmul = vector_fmul_3dnow;
3063
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
3064
                c->float_to_int16 = float_to_int16_3dnow;
3065
                c->float_to_int16_interleave = float_to_int16_interleave_3dnow;
3066
            }
3067
        }
3068
        if(mm_flags & FF_MM_3DNOWEXT){
3069
            c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
3070
            c->vector_fmul_window = vector_fmul_window_3dnow2;
3071
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
3072
                c->float_to_int16_interleave = float_to_int16_interleave_3dn2;
3073
            }
3074
        }
3075
        if(mm_flags & FF_MM_SSE){
3076
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
3077
            c->ac3_downmix = ac3_downmix_sse;
3078
            c->vector_fmul = vector_fmul_sse;
3079
            c->vector_fmul_reverse = vector_fmul_reverse_sse;
3080
            c->vector_fmul_add_add = vector_fmul_add_add_sse;
3081
            c->vector_fmul_window = vector_fmul_window_sse;
3082
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse;
3083
            c->vector_clipf = vector_clipf_sse;
3084
            c->float_to_int16 = float_to_int16_sse;
3085
            c->float_to_int16_interleave = float_to_int16_interleave_sse;
3086
        }
3087
        if(mm_flags & FF_MM_3DNOW)
3088
            c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse
3089
        if(mm_flags & FF_MM_SSE2){
3090
            c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2;
3091
            c->float_to_int16 = float_to_int16_sse2;
3092
            c->float_to_int16_interleave = float_to_int16_interleave_sse2;
3093
            c->add_int16 = add_int16_sse2;
3094
            c->sub_int16 = sub_int16_sse2;
3095
            c->scalarproduct_int16 = scalarproduct_int16_sse2;
3096
        }
3097
    }
3098

    
3099
    if (CONFIG_ENCODERS)
3100
        dsputilenc_init_mmx(c, avctx);
3101

    
3102
#if 0
3103
    // for speed testing
3104
    get_pixels = just_return;
3105
    put_pixels_clamped = just_return;
3106
    add_pixels_clamped = just_return;
3107

3108
    pix_abs16x16 = just_return;
3109
    pix_abs16x16_x2 = just_return;
3110
    pix_abs16x16_y2 = just_return;
3111
    pix_abs16x16_xy2 = just_return;
3112

3113
    put_pixels_tab[0] = just_return;
3114
    put_pixels_tab[1] = just_return;
3115
    put_pixels_tab[2] = just_return;
3116
    put_pixels_tab[3] = just_return;
3117

3118
    put_no_rnd_pixels_tab[0] = just_return;
3119
    put_no_rnd_pixels_tab[1] = just_return;
3120
    put_no_rnd_pixels_tab[2] = just_return;
3121
    put_no_rnd_pixels_tab[3] = just_return;
3122

3123
    avg_pixels_tab[0] = just_return;
3124
    avg_pixels_tab[1] = just_return;
3125
    avg_pixels_tab[2] = just_return;
3126
    avg_pixels_tab[3] = just_return;
3127

3128
    avg_no_rnd_pixels_tab[0] = just_return;
3129
    avg_no_rnd_pixels_tab[1] = just_return;
3130
    avg_no_rnd_pixels_tab[2] = just_return;
3131
    avg_no_rnd_pixels_tab[3] = just_return;
3132

3133
    //av_fdct = just_return;
3134
    //ff_idct = just_return;
3135
#endif
3136
}