Statistics
| Branch: | Revision:

ffmpeg / libavcodec / x86 / dsputil_mmx.c @ 2912e87a

History | View | Annotate | Download (117 KB)

1
/*
2
 * MMX optimized DSP utils
3
 * Copyright (c) 2000, 2001 Fabrice Bellard
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6
 * This file is part of Libav.
7
 *
8
 * Libav 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
 * Libav 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 Libav; 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/cpu.h"
26
#include "libavutil/x86_cpu.h"
27
#include "libavcodec/dsputil.h"
28
#include "libavcodec/h264dsp.h"
29
#include "libavcodec/mpegvideo.h"
30
#include "libavcodec/simple_idct.h"
31
#include "libavcodec/ac3dec.h"
32
#include "dsputil_mmx.h"
33
#include "idct_xvid.h"
34

    
35
//#undef NDEBUG
36
//#include <assert.h>
37

    
38
/* pixel operations */
39
DECLARE_ALIGNED(8,  const uint64_t, ff_bone) = 0x0101010101010101ULL;
40
DECLARE_ALIGNED(8,  const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
41

    
42
DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
43
{0x8000000080000000ULL, 0x8000000080000000ULL};
44

    
45
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_1  ) = 0x0001000100010001ULL;
46
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_3  ) = {0x0003000300030003ULL, 0x0003000300030003ULL};
47
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_4  ) = {0x0004000400040004ULL, 0x0004000400040004ULL};
48
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
49
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_8  ) = {0x0008000800080008ULL, 0x0008000800080008ULL};
50
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_9  ) = {0x0009000900090009ULL, 0x0009000900090009ULL};
51
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
52
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};
53
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_17 ) = {0x0011001100110011ULL, 0x0011001100110011ULL};
54
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_18 ) = {0x0012001200120012ULL, 0x0012001200120012ULL};
55
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
56
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_27 ) = {0x001B001B001B001BULL, 0x001B001B001B001BULL};
57
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};
58
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};
59
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
60
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_53 ) = 0x0035003500350035ULL;
61
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_63 ) = {0x003F003F003F003FULL, 0x003F003F003F003FULL};
62
DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL};
63
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
64
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
65
DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
66

    
67
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_0  ) = {0x0000000000000000ULL, 0x0000000000000000ULL};
68
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_1  ) = {0x0101010101010101ULL, 0x0101010101010101ULL};
69
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_3  ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
70
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_4  ) = {0x0404040404040404ULL, 0x0404040404040404ULL};
71
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
72
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
73
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
74
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_80 ) = {0x8080808080808080ULL, 0x8080808080808080ULL};
75
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;
76
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_A1 ) = {0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL};
77
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_F8 ) = {0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL};
78
DECLARE_ALIGNED(8,  const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
79
DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_FE ) = {0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL};
80

    
81
DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
82
DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
83

    
84
#define JUMPALIGN() __asm__ volatile (".p2align 3"::)
85
#define MOVQ_ZERO(regd)  __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)
86

    
87
#define MOVQ_BFE(regd) \
88
    __asm__ volatile ( \
89
    "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
90
    "paddb %%" #regd ", %%" #regd " \n\t" ::)
91

    
92
#ifndef PIC
93
#define MOVQ_BONE(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
94
#define MOVQ_WTWO(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
95
#else
96
// for shared library it's better to use this way for accessing constants
97
// pcmpeqd -> -1
98
#define MOVQ_BONE(regd) \
99
    __asm__ volatile ( \
100
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
101
    "psrlw $15, %%" #regd " \n\t" \
102
    "packuswb %%" #regd ", %%" #regd " \n\t" ::)
103

    
104
#define MOVQ_WTWO(regd) \
105
    __asm__ volatile ( \
106
    "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
107
    "psrlw $15, %%" #regd " \n\t" \
108
    "psllw $1, %%" #regd " \n\t"::)
109

    
110
#endif
111

    
112
// using regr as temporary and for the output result
113
// first argument is unmodifed and second is trashed
114
// regfe is supposed to contain 0xfefefefefefefefe
115
#define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
116
    "movq " #rega ", " #regr "  \n\t"\
117
    "pand " #regb ", " #regr "  \n\t"\
118
    "pxor " #rega ", " #regb "  \n\t"\
119
    "pand " #regfe "," #regb "  \n\t"\
120
    "psrlq $1, " #regb "        \n\t"\
121
    "paddb " #regb ", " #regr " \n\t"
122

    
123
#define PAVGB_MMX(rega, regb, regr, regfe) \
124
    "movq " #rega ", " #regr "  \n\t"\
125
    "por  " #regb ", " #regr "  \n\t"\
126
    "pxor " #rega ", " #regb "  \n\t"\
127
    "pand " #regfe "," #regb "  \n\t"\
128
    "psrlq $1, " #regb "        \n\t"\
129
    "psubb " #regb ", " #regr " \n\t"
130

    
131
// mm6 is supposed to contain 0xfefefefefefefefe
132
#define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \
133
    "movq " #rega ", " #regr "  \n\t"\
134
    "movq " #regc ", " #regp "  \n\t"\
135
    "pand " #regb ", " #regr "  \n\t"\
136
    "pand " #regd ", " #regp "  \n\t"\
137
    "pxor " #rega ", " #regb "  \n\t"\
138
    "pxor " #regc ", " #regd "  \n\t"\
139
    "pand %%mm6, " #regb "      \n\t"\
140
    "pand %%mm6, " #regd "      \n\t"\
141
    "psrlq $1, " #regb "        \n\t"\
142
    "psrlq $1, " #regd "        \n\t"\
143
    "paddb " #regb ", " #regr " \n\t"\
144
    "paddb " #regd ", " #regp " \n\t"
145

    
146
#define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
147
    "movq " #rega ", " #regr "  \n\t"\
148
    "movq " #regc ", " #regp "  \n\t"\
149
    "por  " #regb ", " #regr "  \n\t"\
150
    "por  " #regd ", " #regp "  \n\t"\
151
    "pxor " #rega ", " #regb "  \n\t"\
152
    "pxor " #regc ", " #regd "  \n\t"\
153
    "pand %%mm6, " #regb "      \n\t"\
154
    "pand %%mm6, " #regd "      \n\t"\
155
    "psrlq $1, " #regd "        \n\t"\
156
    "psrlq $1, " #regb "        \n\t"\
157
    "psubb " #regb ", " #regr " \n\t"\
158
    "psubb " #regd ", " #regp " \n\t"
159

    
160
/***********************************/
161
/* MMX no rounding */
162
#define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
163
#define SET_RND  MOVQ_WONE
164
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
165
#define PAVGB(a, b, c, e)               PAVGB_MMX_NO_RND(a, b, c, e)
166
#define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)
167

    
168
#include "dsputil_mmx_rnd_template.c"
169

    
170
#undef DEF
171
#undef SET_RND
172
#undef PAVGBP
173
#undef PAVGB
174
/***********************************/
175
/* MMX rounding */
176

    
177
#define DEF(x, y) x ## _ ## y ##_mmx
178
#define SET_RND  MOVQ_WTWO
179
#define PAVGBP(a, b, c, d, e, f)        PAVGBP_MMX(a, b, c, d, e, f)
180
#define PAVGB(a, b, c, e)               PAVGB_MMX(a, b, c, e)
181

    
182
#include "dsputil_mmx_rnd_template.c"
183

    
184
#undef DEF
185
#undef SET_RND
186
#undef PAVGBP
187
#undef PAVGB
188
#undef OP_AVG
189

    
190
/***********************************/
191
/* 3Dnow specific */
192

    
193
#define DEF(x) x ## _3dnow
194
#define PAVGB "pavgusb"
195
#define OP_AVG PAVGB
196

    
197
#include "dsputil_mmx_avg_template.c"
198

    
199
#undef DEF
200
#undef PAVGB
201
#undef OP_AVG
202

    
203
/***********************************/
204
/* MMX2 specific */
205

    
206
#define DEF(x) x ## _mmx2
207

    
208
/* Introduced only in MMX2 set */
209
#define PAVGB "pavgb"
210
#define OP_AVG PAVGB
211

    
212
#include "dsputil_mmx_avg_template.c"
213

    
214
#undef DEF
215
#undef PAVGB
216
#undef OP_AVG
217

    
218
#define put_no_rnd_pixels16_mmx put_pixels16_mmx
219
#define put_no_rnd_pixels8_mmx put_pixels8_mmx
220
#define put_pixels16_mmx2 put_pixels16_mmx
221
#define put_pixels8_mmx2 put_pixels8_mmx
222
#define put_pixels4_mmx2 put_pixels4_mmx
223
#define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
224
#define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
225
#define put_pixels16_3dnow put_pixels16_mmx
226
#define put_pixels8_3dnow put_pixels8_mmx
227
#define put_pixels4_3dnow put_pixels4_mmx
228
#define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
229
#define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
230

    
231
/***********************************/
232
/* standard MMX */
233

    
234
void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
235
{
236
    const DCTELEM *p;
237
    uint8_t *pix;
238

    
239
    /* read the pixels */
240
    p = block;
241
    pix = pixels;
242
    /* unrolled loop */
243
        __asm__ volatile(
244
                "movq   %3, %%mm0               \n\t"
245
                "movq   8%3, %%mm1              \n\t"
246
                "movq   16%3, %%mm2             \n\t"
247
                "movq   24%3, %%mm3             \n\t"
248
                "movq   32%3, %%mm4             \n\t"
249
                "movq   40%3, %%mm5             \n\t"
250
                "movq   48%3, %%mm6             \n\t"
251
                "movq   56%3, %%mm7             \n\t"
252
                "packuswb %%mm1, %%mm0          \n\t"
253
                "packuswb %%mm3, %%mm2          \n\t"
254
                "packuswb %%mm5, %%mm4          \n\t"
255
                "packuswb %%mm7, %%mm6          \n\t"
256
                "movq   %%mm0, (%0)             \n\t"
257
                "movq   %%mm2, (%0, %1)         \n\t"
258
                "movq   %%mm4, (%0, %1, 2)      \n\t"
259
                "movq   %%mm6, (%0, %2)         \n\t"
260
                ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)
261
                :"memory");
262
        pix += line_size*4;
263
        p += 32;
264

    
265
    // if here would be an exact copy of the code above
266
    // compiler would generate some very strange code
267
    // thus using "r"
268
    __asm__ volatile(
269
            "movq       (%3), %%mm0             \n\t"
270
            "movq       8(%3), %%mm1            \n\t"
271
            "movq       16(%3), %%mm2           \n\t"
272
            "movq       24(%3), %%mm3           \n\t"
273
            "movq       32(%3), %%mm4           \n\t"
274
            "movq       40(%3), %%mm5           \n\t"
275
            "movq       48(%3), %%mm6           \n\t"
276
            "movq       56(%3), %%mm7           \n\t"
277
            "packuswb %%mm1, %%mm0              \n\t"
278
            "packuswb %%mm3, %%mm2              \n\t"
279
            "packuswb %%mm5, %%mm4              \n\t"
280
            "packuswb %%mm7, %%mm6              \n\t"
281
            "movq       %%mm0, (%0)             \n\t"
282
            "movq       %%mm2, (%0, %1)         \n\t"
283
            "movq       %%mm4, (%0, %1, 2)      \n\t"
284
            "movq       %%mm6, (%0, %2)         \n\t"
285
            ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)
286
            :"memory");
287
}
288

    
289
#define put_signed_pixels_clamped_mmx_half(off) \
290
            "movq    "#off"(%2), %%mm1          \n\t"\
291
            "movq 16+"#off"(%2), %%mm2          \n\t"\
292
            "movq 32+"#off"(%2), %%mm3          \n\t"\
293
            "movq 48+"#off"(%2), %%mm4          \n\t"\
294
            "packsswb  8+"#off"(%2), %%mm1      \n\t"\
295
            "packsswb 24+"#off"(%2), %%mm2      \n\t"\
296
            "packsswb 40+"#off"(%2), %%mm3      \n\t"\
297
            "packsswb 56+"#off"(%2), %%mm4      \n\t"\
298
            "paddb %%mm0, %%mm1                 \n\t"\
299
            "paddb %%mm0, %%mm2                 \n\t"\
300
            "paddb %%mm0, %%mm3                 \n\t"\
301
            "paddb %%mm0, %%mm4                 \n\t"\
302
            "movq %%mm1, (%0)                   \n\t"\
303
            "movq %%mm2, (%0, %3)               \n\t"\
304
            "movq %%mm3, (%0, %3, 2)            \n\t"\
305
            "movq %%mm4, (%0, %1)               \n\t"
306

    
307
void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
308
{
309
    x86_reg line_skip = line_size;
310
    x86_reg line_skip3;
311

    
312
    __asm__ volatile (
313
            "movq "MANGLE(ff_pb_80)", %%mm0 \n\t"
314
            "lea (%3, %3, 2), %1                \n\t"
315
            put_signed_pixels_clamped_mmx_half(0)
316
            "lea (%0, %3, 4), %0                \n\t"
317
            put_signed_pixels_clamped_mmx_half(64)
318
            :"+&r" (pixels), "=&r" (line_skip3)
319
            :"r" (block), "r"(line_skip)
320
            :"memory");
321
}
322

    
323
void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
324
{
325
    const DCTELEM *p;
326
    uint8_t *pix;
327
    int i;
328

    
329
    /* read the pixels */
330
    p = block;
331
    pix = pixels;
332
    MOVQ_ZERO(mm7);
333
    i = 4;
334
    do {
335
        __asm__ volatile(
336
                "movq   (%2), %%mm0     \n\t"
337
                "movq   8(%2), %%mm1    \n\t"
338
                "movq   16(%2), %%mm2   \n\t"
339
                "movq   24(%2), %%mm3   \n\t"
340
                "movq   %0, %%mm4       \n\t"
341
                "movq   %1, %%mm6       \n\t"
342
                "movq   %%mm4, %%mm5    \n\t"
343
                "punpcklbw %%mm7, %%mm4 \n\t"
344
                "punpckhbw %%mm7, %%mm5 \n\t"
345
                "paddsw %%mm4, %%mm0    \n\t"
346
                "paddsw %%mm5, %%mm1    \n\t"
347
                "movq   %%mm6, %%mm5    \n\t"
348
                "punpcklbw %%mm7, %%mm6 \n\t"
349
                "punpckhbw %%mm7, %%mm5 \n\t"
350
                "paddsw %%mm6, %%mm2    \n\t"
351
                "paddsw %%mm5, %%mm3    \n\t"
352
                "packuswb %%mm1, %%mm0  \n\t"
353
                "packuswb %%mm3, %%mm2  \n\t"
354
                "movq   %%mm0, %0       \n\t"
355
                "movq   %%mm2, %1       \n\t"
356
                :"+m"(*pix), "+m"(*(pix+line_size))
357
                :"r"(p)
358
                :"memory");
359
        pix += line_size*2;
360
        p += 16;
361
    } while (--i);
362
}
363

    
364
static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
365
{
366
    __asm__ volatile(
367
         "lea (%3, %3), %%"REG_a"       \n\t"
368
         ".p2align 3                    \n\t"
369
         "1:                            \n\t"
370
         "movd (%1), %%mm0              \n\t"
371
         "movd (%1, %3), %%mm1          \n\t"
372
         "movd %%mm0, (%2)              \n\t"
373
         "movd %%mm1, (%2, %3)          \n\t"
374
         "add %%"REG_a", %1             \n\t"
375
         "add %%"REG_a", %2             \n\t"
376
         "movd (%1), %%mm0              \n\t"
377
         "movd (%1, %3), %%mm1          \n\t"
378
         "movd %%mm0, (%2)              \n\t"
379
         "movd %%mm1, (%2, %3)          \n\t"
380
         "add %%"REG_a", %1             \n\t"
381
         "add %%"REG_a", %2             \n\t"
382
         "subl $4, %0                   \n\t"
383
         "jnz 1b                        \n\t"
384
         : "+g"(h), "+r" (pixels),  "+r" (block)
385
         : "r"((x86_reg)line_size)
386
         : "%"REG_a, "memory"
387
        );
388
}
389

    
390
static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
391
{
392
    __asm__ volatile(
393
         "lea (%3, %3), %%"REG_a"       \n\t"
394
         ".p2align 3                    \n\t"
395
         "1:                            \n\t"
396
         "movq (%1), %%mm0              \n\t"
397
         "movq (%1, %3), %%mm1          \n\t"
398
         "movq %%mm0, (%2)              \n\t"
399
         "movq %%mm1, (%2, %3)          \n\t"
400
         "add %%"REG_a", %1             \n\t"
401
         "add %%"REG_a", %2             \n\t"
402
         "movq (%1), %%mm0              \n\t"
403
         "movq (%1, %3), %%mm1          \n\t"
404
         "movq %%mm0, (%2)              \n\t"
405
         "movq %%mm1, (%2, %3)          \n\t"
406
         "add %%"REG_a", %1             \n\t"
407
         "add %%"REG_a", %2             \n\t"
408
         "subl $4, %0                   \n\t"
409
         "jnz 1b                        \n\t"
410
         : "+g"(h), "+r" (pixels),  "+r" (block)
411
         : "r"((x86_reg)line_size)
412
         : "%"REG_a, "memory"
413
        );
414
}
415

    
416
static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
417
{
418
    __asm__ volatile(
419
         "lea (%3, %3), %%"REG_a"       \n\t"
420
         ".p2align 3                    \n\t"
421
         "1:                            \n\t"
422
         "movq (%1), %%mm0              \n\t"
423
         "movq 8(%1), %%mm4             \n\t"
424
         "movq (%1, %3), %%mm1          \n\t"
425
         "movq 8(%1, %3), %%mm5         \n\t"
426
         "movq %%mm0, (%2)              \n\t"
427
         "movq %%mm4, 8(%2)             \n\t"
428
         "movq %%mm1, (%2, %3)          \n\t"
429
         "movq %%mm5, 8(%2, %3)         \n\t"
430
         "add %%"REG_a", %1             \n\t"
431
         "add %%"REG_a", %2             \n\t"
432
         "movq (%1), %%mm0              \n\t"
433
         "movq 8(%1), %%mm4             \n\t"
434
         "movq (%1, %3), %%mm1          \n\t"
435
         "movq 8(%1, %3), %%mm5         \n\t"
436
         "movq %%mm0, (%2)              \n\t"
437
         "movq %%mm4, 8(%2)             \n\t"
438
         "movq %%mm1, (%2, %3)          \n\t"
439
         "movq %%mm5, 8(%2, %3)         \n\t"
440
         "add %%"REG_a", %1             \n\t"
441
         "add %%"REG_a", %2             \n\t"
442
         "subl $4, %0                   \n\t"
443
         "jnz 1b                        \n\t"
444
         : "+g"(h), "+r" (pixels),  "+r" (block)
445
         : "r"((x86_reg)line_size)
446
         : "%"REG_a, "memory"
447
        );
448
}
449

    
450
static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
451
{
452
    __asm__ volatile(
453
         "1:                            \n\t"
454
         "movdqu (%1), %%xmm0           \n\t"
455
         "movdqu (%1,%3), %%xmm1        \n\t"
456
         "movdqu (%1,%3,2), %%xmm2      \n\t"
457
         "movdqu (%1,%4), %%xmm3        \n\t"
458
         "movdqa %%xmm0, (%2)           \n\t"
459
         "movdqa %%xmm1, (%2,%3)        \n\t"
460
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
461
         "movdqa %%xmm3, (%2,%4)        \n\t"
462
         "subl $4, %0                   \n\t"
463
         "lea (%1,%3,4), %1             \n\t"
464
         "lea (%2,%3,4), %2             \n\t"
465
         "jnz 1b                        \n\t"
466
         : "+g"(h), "+r" (pixels),  "+r" (block)
467
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
468
         : "memory"
469
        );
470
}
471

    
472
static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
473
{
474
    __asm__ volatile(
475
         "1:                            \n\t"
476
         "movdqu (%1), %%xmm0           \n\t"
477
         "movdqu (%1,%3), %%xmm1        \n\t"
478
         "movdqu (%1,%3,2), %%xmm2      \n\t"
479
         "movdqu (%1,%4), %%xmm3        \n\t"
480
         "pavgb  (%2), %%xmm0           \n\t"
481
         "pavgb  (%2,%3), %%xmm1        \n\t"
482
         "pavgb  (%2,%3,2), %%xmm2      \n\t"
483
         "pavgb  (%2,%4), %%xmm3        \n\t"
484
         "movdqa %%xmm0, (%2)           \n\t"
485
         "movdqa %%xmm1, (%2,%3)        \n\t"
486
         "movdqa %%xmm2, (%2,%3,2)      \n\t"
487
         "movdqa %%xmm3, (%2,%4)        \n\t"
488
         "subl $4, %0                   \n\t"
489
         "lea (%1,%3,4), %1             \n\t"
490
         "lea (%2,%3,4), %2             \n\t"
491
         "jnz 1b                        \n\t"
492
         : "+g"(h), "+r" (pixels),  "+r" (block)
493
         : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)
494
         : "memory"
495
        );
496
}
497

    
498
#define CLEAR_BLOCKS(name,n) \
499
static void name(DCTELEM *blocks)\
500
{\
501
    __asm__ volatile(\
502
                "pxor %%mm7, %%mm7              \n\t"\
503
                "mov     %1, %%"REG_a"          \n\t"\
504
                "1:                             \n\t"\
505
                "movq %%mm7, (%0, %%"REG_a")    \n\t"\
506
                "movq %%mm7, 8(%0, %%"REG_a")   \n\t"\
507
                "movq %%mm7, 16(%0, %%"REG_a")  \n\t"\
508
                "movq %%mm7, 24(%0, %%"REG_a")  \n\t"\
509
                "add $32, %%"REG_a"             \n\t"\
510
                " js 1b                         \n\t"\
511
                : : "r" (((uint8_t *)blocks)+128*n),\
512
                    "i" (-128*n)\
513
                : "%"REG_a\
514
        );\
515
}
516
CLEAR_BLOCKS(clear_blocks_mmx, 6)
517
CLEAR_BLOCKS(clear_block_mmx, 1)
518

    
519
static void clear_block_sse(DCTELEM *block)
520
{
521
    __asm__ volatile(
522
        "xorps  %%xmm0, %%xmm0  \n"
523
        "movaps %%xmm0,    (%0) \n"
524
        "movaps %%xmm0,  16(%0) \n"
525
        "movaps %%xmm0,  32(%0) \n"
526
        "movaps %%xmm0,  48(%0) \n"
527
        "movaps %%xmm0,  64(%0) \n"
528
        "movaps %%xmm0,  80(%0) \n"
529
        "movaps %%xmm0,  96(%0) \n"
530
        "movaps %%xmm0, 112(%0) \n"
531
        :: "r"(block)
532
        : "memory"
533
    );
534
}
535

    
536
static void clear_blocks_sse(DCTELEM *blocks)
537
{\
538
    __asm__ volatile(
539
        "xorps  %%xmm0, %%xmm0  \n"
540
        "mov     %1, %%"REG_a"  \n"
541
        "1:                     \n"
542
        "movaps %%xmm0,    (%0, %%"REG_a") \n"
543
        "movaps %%xmm0,  16(%0, %%"REG_a") \n"
544
        "movaps %%xmm0,  32(%0, %%"REG_a") \n"
545
        "movaps %%xmm0,  48(%0, %%"REG_a") \n"
546
        "movaps %%xmm0,  64(%0, %%"REG_a") \n"
547
        "movaps %%xmm0,  80(%0, %%"REG_a") \n"
548
        "movaps %%xmm0,  96(%0, %%"REG_a") \n"
549
        "movaps %%xmm0, 112(%0, %%"REG_a") \n"
550
        "add $128, %%"REG_a"    \n"
551
        " js 1b                 \n"
552
        : : "r" (((uint8_t *)blocks)+128*6),
553
            "i" (-128*6)
554
        : "%"REG_a
555
    );
556
}
557

    
558
static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
559
    x86_reg i=0;
560
    __asm__ volatile(
561
        "jmp 2f                         \n\t"
562
        "1:                             \n\t"
563
        "movq  (%1, %0), %%mm0          \n\t"
564
        "movq  (%2, %0), %%mm1          \n\t"
565
        "paddb %%mm0, %%mm1             \n\t"
566
        "movq %%mm1, (%2, %0)           \n\t"
567
        "movq 8(%1, %0), %%mm0          \n\t"
568
        "movq 8(%2, %0), %%mm1          \n\t"
569
        "paddb %%mm0, %%mm1             \n\t"
570
        "movq %%mm1, 8(%2, %0)          \n\t"
571
        "add $16, %0                    \n\t"
572
        "2:                             \n\t"
573
        "cmp %3, %0                     \n\t"
574
        " js 1b                         \n\t"
575
        : "+r" (i)
576
        : "r"(src), "r"(dst), "r"((x86_reg)w-15)
577
    );
578
    for(; i<w; i++)
579
        dst[i+0] += src[i+0];
580
}
581

    
582
static void add_bytes_l2_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
583
    x86_reg i=0;
584
    __asm__ volatile(
585
        "jmp 2f                         \n\t"
586
        "1:                             \n\t"
587
        "movq   (%2, %0), %%mm0         \n\t"
588
        "movq  8(%2, %0), %%mm1         \n\t"
589
        "paddb  (%3, %0), %%mm0         \n\t"
590
        "paddb 8(%3, %0), %%mm1         \n\t"
591
        "movq %%mm0,  (%1, %0)          \n\t"
592
        "movq %%mm1, 8(%1, %0)          \n\t"
593
        "add $16, %0                    \n\t"
594
        "2:                             \n\t"
595
        "cmp %4, %0                     \n\t"
596
        " js 1b                         \n\t"
597
        : "+r" (i)
598
        : "r"(dst), "r"(src1), "r"(src2), "r"((x86_reg)w-15)
599
    );
600
    for(; i<w; i++)
601
        dst[i] = src1[i] + src2[i];
602
}
603

    
604
#if HAVE_7REGS && HAVE_TEN_OPERANDS
605
static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top) {
606
    x86_reg w2 = -w;
607
    x86_reg x;
608
    int l = *left & 0xff;
609
    int tl = *left_top & 0xff;
610
    int t;
611
    __asm__ volatile(
612
        "mov    %7, %3 \n"
613
        "1: \n"
614
        "movzbl (%3,%4), %2 \n"
615
        "mov    %2, %k3 \n"
616
        "sub   %b1, %b3 \n"
617
        "add   %b0, %b3 \n"
618
        "mov    %2, %1 \n"
619
        "cmp    %0, %2 \n"
620
        "cmovg  %0, %2 \n"
621
        "cmovg  %1, %0 \n"
622
        "cmp   %k3, %0 \n"
623
        "cmovg %k3, %0 \n"
624
        "mov    %7, %3 \n"
625
        "cmp    %2, %0 \n"
626
        "cmovl  %2, %0 \n"
627
        "add (%6,%4), %b0 \n"
628
        "mov   %b0, (%5,%4) \n"
629
        "inc    %4 \n"
630
        "jl 1b \n"
631
        :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
632
        :"r"(dst+w), "r"(diff+w), "rm"(top+w)
633
    );
634
    *left = l;
635
    *left_top = tl;
636
}
637
#endif
638

    
639
#define H263_LOOP_FILTER \
640
        "pxor %%mm7, %%mm7              \n\t"\
641
        "movq  %0, %%mm0                \n\t"\
642
        "movq  %0, %%mm1                \n\t"\
643
        "movq  %3, %%mm2                \n\t"\
644
        "movq  %3, %%mm3                \n\t"\
645
        "punpcklbw %%mm7, %%mm0         \n\t"\
646
        "punpckhbw %%mm7, %%mm1         \n\t"\
647
        "punpcklbw %%mm7, %%mm2         \n\t"\
648
        "punpckhbw %%mm7, %%mm3         \n\t"\
649
        "psubw %%mm2, %%mm0             \n\t"\
650
        "psubw %%mm3, %%mm1             \n\t"\
651
        "movq  %1, %%mm2                \n\t"\
652
        "movq  %1, %%mm3                \n\t"\
653
        "movq  %2, %%mm4                \n\t"\
654
        "movq  %2, %%mm5                \n\t"\
655
        "punpcklbw %%mm7, %%mm2         \n\t"\
656
        "punpckhbw %%mm7, %%mm3         \n\t"\
657
        "punpcklbw %%mm7, %%mm4         \n\t"\
658
        "punpckhbw %%mm7, %%mm5         \n\t"\
659
        "psubw %%mm2, %%mm4             \n\t"\
660
        "psubw %%mm3, %%mm5             \n\t"\
661
        "psllw $2, %%mm4                \n\t"\
662
        "psllw $2, %%mm5                \n\t"\
663
        "paddw %%mm0, %%mm4             \n\t"\
664
        "paddw %%mm1, %%mm5             \n\t"\
665
        "pxor %%mm6, %%mm6              \n\t"\
666
        "pcmpgtw %%mm4, %%mm6           \n\t"\
667
        "pcmpgtw %%mm5, %%mm7           \n\t"\
668
        "pxor %%mm6, %%mm4              \n\t"\
669
        "pxor %%mm7, %%mm5              \n\t"\
670
        "psubw %%mm6, %%mm4             \n\t"\
671
        "psubw %%mm7, %%mm5             \n\t"\
672
        "psrlw $3, %%mm4                \n\t"\
673
        "psrlw $3, %%mm5                \n\t"\
674
        "packuswb %%mm5, %%mm4          \n\t"\
675
        "packsswb %%mm7, %%mm6          \n\t"\
676
        "pxor %%mm7, %%mm7              \n\t"\
677
        "movd %4, %%mm2                 \n\t"\
678
        "punpcklbw %%mm2, %%mm2         \n\t"\
679
        "punpcklbw %%mm2, %%mm2         \n\t"\
680
        "punpcklbw %%mm2, %%mm2         \n\t"\
681
        "psubusb %%mm4, %%mm2           \n\t"\
682
        "movq %%mm2, %%mm3              \n\t"\
683
        "psubusb %%mm4, %%mm3           \n\t"\
684
        "psubb %%mm3, %%mm2             \n\t"\
685
        "movq %1, %%mm3                 \n\t"\
686
        "movq %2, %%mm4                 \n\t"\
687
        "pxor %%mm6, %%mm3              \n\t"\
688
        "pxor %%mm6, %%mm4              \n\t"\
689
        "paddusb %%mm2, %%mm3           \n\t"\
690
        "psubusb %%mm2, %%mm4           \n\t"\
691
        "pxor %%mm6, %%mm3              \n\t"\
692
        "pxor %%mm6, %%mm4              \n\t"\
693
        "paddusb %%mm2, %%mm2           \n\t"\
694
        "packsswb %%mm1, %%mm0          \n\t"\
695
        "pcmpgtb %%mm0, %%mm7           \n\t"\
696
        "pxor %%mm7, %%mm0              \n\t"\
697
        "psubb %%mm7, %%mm0             \n\t"\
698
        "movq %%mm0, %%mm1              \n\t"\
699
        "psubusb %%mm2, %%mm0           \n\t"\
700
        "psubb %%mm0, %%mm1             \n\t"\
701
        "pand %5, %%mm1                 \n\t"\
702
        "psrlw $2, %%mm1                \n\t"\
703
        "pxor %%mm7, %%mm1              \n\t"\
704
        "psubb %%mm7, %%mm1             \n\t"\
705
        "movq %0, %%mm5                 \n\t"\
706
        "movq %3, %%mm6                 \n\t"\
707
        "psubb %%mm1, %%mm5             \n\t"\
708
        "paddb %%mm1, %%mm6             \n\t"
709

    
710
static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
711
    if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
712
    const int strength= ff_h263_loop_filter_strength[qscale];
713

    
714
    __asm__ volatile(
715

    
716
        H263_LOOP_FILTER
717

    
718
        "movq %%mm3, %1                 \n\t"
719
        "movq %%mm4, %2                 \n\t"
720
        "movq %%mm5, %0                 \n\t"
721
        "movq %%mm6, %3                 \n\t"
722
        : "+m" (*(uint64_t*)(src - 2*stride)),
723
          "+m" (*(uint64_t*)(src - 1*stride)),
724
          "+m" (*(uint64_t*)(src + 0*stride)),
725
          "+m" (*(uint64_t*)(src + 1*stride))
726
        : "g" (2*strength), "m"(ff_pb_FC)
727
    );
728
    }
729
}
730

    
731
static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
732
    if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
733
    const int strength= ff_h263_loop_filter_strength[qscale];
734
    DECLARE_ALIGNED(8, uint64_t, temp)[4];
735
    uint8_t *btemp= (uint8_t*)temp;
736

    
737
    src -= 2;
738

    
739
    transpose4x4(btemp  , src           , 8, stride);
740
    transpose4x4(btemp+4, src + 4*stride, 8, stride);
741
    __asm__ volatile(
742
        H263_LOOP_FILTER // 5 3 4 6
743

    
744
        : "+m" (temp[0]),
745
          "+m" (temp[1]),
746
          "+m" (temp[2]),
747
          "+m" (temp[3])
748
        : "g" (2*strength), "m"(ff_pb_FC)
749
    );
750

    
751
    __asm__ volatile(
752
        "movq %%mm5, %%mm1              \n\t"
753
        "movq %%mm4, %%mm0              \n\t"
754
        "punpcklbw %%mm3, %%mm5         \n\t"
755
        "punpcklbw %%mm6, %%mm4         \n\t"
756
        "punpckhbw %%mm3, %%mm1         \n\t"
757
        "punpckhbw %%mm6, %%mm0         \n\t"
758
        "movq %%mm5, %%mm3              \n\t"
759
        "movq %%mm1, %%mm6              \n\t"
760
        "punpcklwd %%mm4, %%mm5         \n\t"
761
        "punpcklwd %%mm0, %%mm1         \n\t"
762
        "punpckhwd %%mm4, %%mm3         \n\t"
763
        "punpckhwd %%mm0, %%mm6         \n\t"
764
        "movd %%mm5, (%0)               \n\t"
765
        "punpckhdq %%mm5, %%mm5         \n\t"
766
        "movd %%mm5, (%0,%2)            \n\t"
767
        "movd %%mm3, (%0,%2,2)          \n\t"
768
        "punpckhdq %%mm3, %%mm3         \n\t"
769
        "movd %%mm3, (%0,%3)            \n\t"
770
        "movd %%mm1, (%1)               \n\t"
771
        "punpckhdq %%mm1, %%mm1         \n\t"
772
        "movd %%mm1, (%1,%2)            \n\t"
773
        "movd %%mm6, (%1,%2,2)          \n\t"
774
        "punpckhdq %%mm6, %%mm6         \n\t"
775
        "movd %%mm6, (%1,%3)            \n\t"
776
        :: "r" (src),
777
           "r" (src + 4*stride),
778
           "r" ((x86_reg)   stride ),
779
           "r" ((x86_reg)(3*stride))
780
    );
781
    }
782
}
783

    
784
/* draw the edges of width 'w' of an image of size width, height
785
   this mmx version can only handle w==8 || w==16 */
786
static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
787
{
788
    uint8_t *ptr, *last_line;
789
    int i;
790

    
791
    last_line = buf + (height - 1) * wrap;
792
    /* left and right */
793
    ptr = buf;
794
    if(w==8)
795
    {
796
        __asm__ volatile(
797
                "1:                             \n\t"
798
                "movd (%0), %%mm0               \n\t"
799
                "punpcklbw %%mm0, %%mm0         \n\t"
800
                "punpcklwd %%mm0, %%mm0         \n\t"
801
                "punpckldq %%mm0, %%mm0         \n\t"
802
                "movq %%mm0, -8(%0)             \n\t"
803
                "movq -8(%0, %2), %%mm1         \n\t"
804
                "punpckhbw %%mm1, %%mm1         \n\t"
805
                "punpckhwd %%mm1, %%mm1         \n\t"
806
                "punpckhdq %%mm1, %%mm1         \n\t"
807
                "movq %%mm1, (%0, %2)           \n\t"
808
                "add %1, %0                     \n\t"
809
                "cmp %3, %0                     \n\t"
810
                " jb 1b                         \n\t"
811
                : "+r" (ptr)
812
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
813
        );
814
    }
815
    else
816
    {
817
        __asm__ volatile(
818
                "1:                             \n\t"
819
                "movd (%0), %%mm0               \n\t"
820
                "punpcklbw %%mm0, %%mm0         \n\t"
821
                "punpcklwd %%mm0, %%mm0         \n\t"
822
                "punpckldq %%mm0, %%mm0         \n\t"
823
                "movq %%mm0, -8(%0)             \n\t"
824
                "movq %%mm0, -16(%0)            \n\t"
825
                "movq -8(%0, %2), %%mm1         \n\t"
826
                "punpckhbw %%mm1, %%mm1         \n\t"
827
                "punpckhwd %%mm1, %%mm1         \n\t"
828
                "punpckhdq %%mm1, %%mm1         \n\t"
829
                "movq %%mm1, (%0, %2)           \n\t"
830
                "movq %%mm1, 8(%0, %2)          \n\t"
831
                "add %1, %0                     \n\t"
832
                "cmp %3, %0                     \n\t"
833
                " jb 1b                         \n\t"
834
                : "+r" (ptr)
835
                : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)
836
        );
837
    }
838

    
839
    for(i=0;i<w;i+=4) {
840
        /* top and bottom (and hopefully also the corners) */
841
        ptr= buf - (i + 1) * wrap - w;
842
        __asm__ volatile(
843
                "1:                             \n\t"
844
                "movq (%1, %0), %%mm0           \n\t"
845
                "movq %%mm0, (%0)               \n\t"
846
                "movq %%mm0, (%0, %2)           \n\t"
847
                "movq %%mm0, (%0, %2, 2)        \n\t"
848
                "movq %%mm0, (%0, %3)           \n\t"
849
                "add $8, %0                     \n\t"
850
                "cmp %4, %0                     \n\t"
851
                " jb 1b                         \n\t"
852
                : "+r" (ptr)
853
                : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)
854
        );
855
        ptr= last_line + (i + 1) * wrap - w;
856
        __asm__ volatile(
857
                "1:                             \n\t"
858
                "movq (%1, %0), %%mm0           \n\t"
859
                "movq %%mm0, (%0)               \n\t"
860
                "movq %%mm0, (%0, %2)           \n\t"
861
                "movq %%mm0, (%0, %2, 2)        \n\t"
862
                "movq %%mm0, (%0, %3)           \n\t"
863
                "add $8, %0                     \n\t"
864
                "cmp %4, %0                     \n\t"
865
                " jb 1b                         \n\t"
866
                : "+r" (ptr)
867
                : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)
868
        );
869
    }
870
}
871

    
872
#define PAETH(cpu, abs3)\
873
static void add_png_paeth_prediction_##cpu(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)\
874
{\
875
    x86_reg i = -bpp;\
876
    x86_reg end = w-3;\
877
    __asm__ volatile(\
878
        "pxor      %%mm7, %%mm7 \n"\
879
        "movd    (%1,%0), %%mm0 \n"\
880
        "movd    (%2,%0), %%mm1 \n"\
881
        "punpcklbw %%mm7, %%mm0 \n"\
882
        "punpcklbw %%mm7, %%mm1 \n"\
883
        "add       %4, %0 \n"\
884
        "1: \n"\
885
        "movq      %%mm1, %%mm2 \n"\
886
        "movd    (%2,%0), %%mm1 \n"\
887
        "movq      %%mm2, %%mm3 \n"\
888
        "punpcklbw %%mm7, %%mm1 \n"\
889
        "movq      %%mm2, %%mm4 \n"\
890
        "psubw     %%mm1, %%mm3 \n"\
891
        "psubw     %%mm0, %%mm4 \n"\
892
        "movq      %%mm3, %%mm5 \n"\
893
        "paddw     %%mm4, %%mm5 \n"\
894
        abs3\
895
        "movq      %%mm4, %%mm6 \n"\
896
        "pminsw    %%mm5, %%mm6 \n"\
897
        "pcmpgtw   %%mm6, %%mm3 \n"\
898
        "pcmpgtw   %%mm5, %%mm4 \n"\
899
        "movq      %%mm4, %%mm6 \n"\
900
        "pand      %%mm3, %%mm4 \n"\
901
        "pandn     %%mm3, %%mm6 \n"\
902
        "pandn     %%mm0, %%mm3 \n"\
903
        "movd    (%3,%0), %%mm0 \n"\
904
        "pand      %%mm1, %%mm6 \n"\
905
        "pand      %%mm4, %%mm2 \n"\
906
        "punpcklbw %%mm7, %%mm0 \n"\
907
        "movq      %6,    %%mm5 \n"\
908
        "paddw     %%mm6, %%mm0 \n"\
909
        "paddw     %%mm2, %%mm3 \n"\
910
        "paddw     %%mm3, %%mm0 \n"\
911
        "pand      %%mm5, %%mm0 \n"\
912
        "movq      %%mm0, %%mm3 \n"\
913
        "packuswb  %%mm3, %%mm3 \n"\
914
        "movd      %%mm3, (%1,%0) \n"\
915
        "add       %4, %0 \n"\
916
        "cmp       %5, %0 \n"\
917
        "jle 1b \n"\
918
        :"+r"(i)\
919
        :"r"(dst), "r"(top), "r"(src), "r"((x86_reg)bpp), "g"(end),\
920
         "m"(ff_pw_255)\
921
        :"memory"\
922
    );\
923
}
924

    
925
#define ABS3_MMX2\
926
        "psubw     %%mm5, %%mm7 \n"\
927
        "pmaxsw    %%mm7, %%mm5 \n"\
928
        "pxor      %%mm6, %%mm6 \n"\
929
        "pxor      %%mm7, %%mm7 \n"\
930
        "psubw     %%mm3, %%mm6 \n"\
931
        "psubw     %%mm4, %%mm7 \n"\
932
        "pmaxsw    %%mm6, %%mm3 \n"\
933
        "pmaxsw    %%mm7, %%mm4 \n"\
934
        "pxor      %%mm7, %%mm7 \n"
935

    
936
#define ABS3_SSSE3\
937
        "pabsw     %%mm3, %%mm3 \n"\
938
        "pabsw     %%mm4, %%mm4 \n"\
939
        "pabsw     %%mm5, %%mm5 \n"
940

    
941
PAETH(mmx2, ABS3_MMX2)
942
#if HAVE_SSSE3
943
PAETH(ssse3, ABS3_SSSE3)
944
#endif
945

    
946
#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
947
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\
948
        "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\
949
        "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\
950
        "movq "#in7", " #m3 "             \n\t" /* d */\
951
        "movq "#in0", %%mm5               \n\t" /* D */\
952
        "paddw " #m3 ", %%mm5             \n\t" /* x4 */\
953
        "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\
954
        "movq "#in1", %%mm5               \n\t" /* C */\
955
        "movq "#in2", %%mm6               \n\t" /* B */\
956
        "paddw " #m6 ", %%mm5             \n\t" /* x3 */\
957
        "paddw " #m5 ", %%mm6             \n\t" /* x2 */\
958
        "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\
959
        "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\
960
        "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\
961
        "paddw " #rnd ", %%mm4            \n\t" /* x2 */\
962
        "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
963
        "psraw $5, %%mm5                  \n\t"\
964
        "packuswb %%mm5, %%mm5            \n\t"\
965
        OP(%%mm5, out, %%mm7, d)
966

    
967
#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
968
static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
969
    uint64_t temp;\
970
\
971
    __asm__ volatile(\
972
        "pxor %%mm7, %%mm7                \n\t"\
973
        "1:                               \n\t"\
974
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
975
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
976
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
977
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
978
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
979
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
980
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
981
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
982
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
983
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
984
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
985
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
986
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
987
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
988
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
989
        "paddw %%mm3, %%mm5               \n\t" /* b */\
990
        "paddw %%mm2, %%mm6               \n\t" /* c */\
991
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
992
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
993
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
994
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
995
        "paddw %%mm4, %%mm0               \n\t" /* a */\
996
        "paddw %%mm1, %%mm5               \n\t" /* d */\
997
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
998
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
999
        "paddw %6, %%mm6                  \n\t"\
1000
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1001
        "psraw $5, %%mm0                  \n\t"\
1002
        "movq %%mm0, %5                   \n\t"\
1003
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1004
        \
1005
        "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\
1006
        "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\
1007
        "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\
1008
        "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\
1009
        "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\
1010
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\
1011
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\
1012
        "paddw %%mm0, %%mm2               \n\t" /* b */\
1013
        "paddw %%mm5, %%mm3               \n\t" /* c */\
1014
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1015
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1016
        "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\
1017
        "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\
1018
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\
1019
        "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\
1020
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1021
        "paddw %%mm2, %%mm1               \n\t" /* a */\
1022
        "paddw %%mm6, %%mm4               \n\t" /* d */\
1023
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1024
        "psubw %%mm4, %%mm3               \n\t" /* - 6b +3c - d */\
1025
        "paddw %6, %%mm1                  \n\t"\
1026
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b +3c - d */\
1027
        "psraw $5, %%mm3                  \n\t"\
1028
        "movq %5, %%mm1                   \n\t"\
1029
        "packuswb %%mm3, %%mm1            \n\t"\
1030
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
1031
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
1032
        \
1033
        "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\
1034
        "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\
1035
        "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\
1036
        "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\
1037
        "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\
1038
        "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\
1039
        "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\
1040
        "paddw %%mm1, %%mm5               \n\t" /* b */\
1041
        "paddw %%mm4, %%mm0               \n\t" /* c */\
1042
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1043
        "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\
1044
        "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\
1045
        "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\
1046
        "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\
1047
        "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\
1048
        "paddw %%mm3, %%mm2               \n\t" /* d */\
1049
        "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\
1050
        "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\
1051
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\
1052
        "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\
1053
        "paddw %%mm2, %%mm6               \n\t" /* a */\
1054
        "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
1055
        "paddw %6, %%mm0                  \n\t"\
1056
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1057
        "psraw $5, %%mm0                  \n\t"\
1058
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
1059
        \
1060
        "paddw %%mm5, %%mm3               \n\t" /* a */\
1061
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\
1062
        "paddw %%mm4, %%mm6               \n\t" /* b */\
1063
        "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\
1064
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\
1065
        "paddw %%mm1, %%mm4               \n\t" /* c */\
1066
        "paddw %%mm2, %%mm5               \n\t" /* d */\
1067
        "paddw %%mm6, %%mm6               \n\t" /* 2b */\
1068
        "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\
1069
        "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
1070
        "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\
1071
        "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\
1072
        "paddw %6, %%mm4                  \n\t"\
1073
        "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\
1074
        "psraw $5, %%mm4                  \n\t"\
1075
        "packuswb %%mm4, %%mm0            \n\t"\
1076
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
1077
        \
1078
        "add %3, %0                       \n\t"\
1079
        "add %4, %1                       \n\t"\
1080
        "decl %2                          \n\t"\
1081
        " jnz 1b                          \n\t"\
1082
        : "+a"(src), "+c"(dst), "+D"(h)\
1083
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
1084
        : "memory"\
1085
    );\
1086
}\
1087
\
1088
static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1089
    int i;\
1090
    int16_t temp[16];\
1091
    /* quick HACK, XXX FIXME MUST be optimized */\
1092
    for(i=0; i<h; i++)\
1093
    {\
1094
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1095
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1096
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1097
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1098
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1099
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
1100
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
1101
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
1102
        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
1103
        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
1104
        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
1105
        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
1106
        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
1107
        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
1108
        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
1109
        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
1110
        __asm__ volatile(\
1111
            "movq (%0), %%mm0               \n\t"\
1112
            "movq 8(%0), %%mm1              \n\t"\
1113
            "paddw %2, %%mm0                \n\t"\
1114
            "paddw %2, %%mm1                \n\t"\
1115
            "psraw $5, %%mm0                \n\t"\
1116
            "psraw $5, %%mm1                \n\t"\
1117
            "packuswb %%mm1, %%mm0          \n\t"\
1118
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1119
            "movq 16(%0), %%mm0             \n\t"\
1120
            "movq 24(%0), %%mm1             \n\t"\
1121
            "paddw %2, %%mm0                \n\t"\
1122
            "paddw %2, %%mm1                \n\t"\
1123
            "psraw $5, %%mm0                \n\t"\
1124
            "psraw $5, %%mm1                \n\t"\
1125
            "packuswb %%mm1, %%mm0          \n\t"\
1126
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
1127
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1128
            : "memory"\
1129
        );\
1130
        dst+=dstStride;\
1131
        src+=srcStride;\
1132
    }\
1133
}\
1134
\
1135
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1136
    __asm__ volatile(\
1137
        "pxor %%mm7, %%mm7                \n\t"\
1138
        "1:                               \n\t"\
1139
        "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\
1140
        "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\
1141
        "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\
1142
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\
1143
        "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\
1144
        "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\
1145
        "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\
1146
        "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\
1147
        "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\
1148
        "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\
1149
        "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\
1150
        "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\
1151
        "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\
1152
        "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\
1153
        "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\
1154
        "paddw %%mm3, %%mm5               \n\t" /* b */\
1155
        "paddw %%mm2, %%mm6               \n\t" /* c */\
1156
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
1157
        "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\
1158
        "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\
1159
        "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\
1160
        "paddw %%mm4, %%mm0               \n\t" /* a */\
1161
        "paddw %%mm1, %%mm5               \n\t" /* d */\
1162
        "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
1163
        "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\
1164
        "paddw %5, %%mm6                  \n\t"\
1165
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
1166
        "psraw $5, %%mm0                  \n\t"\
1167
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
1168
        \
1169
        "movd 5(%0), %%mm5                \n\t" /* FGHI */\
1170
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\
1171
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\
1172
        "paddw %%mm5, %%mm1               \n\t" /* a */\
1173
        "paddw %%mm6, %%mm2               \n\t" /* b */\
1174
        "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\
1175
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\
1176
        "paddw %%mm6, %%mm3               \n\t" /* c */\
1177
        "paddw %%mm5, %%mm4               \n\t" /* d */\
1178
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
1179
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
1180
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
1181
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
1182
        "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\
1183
        "paddw %5, %%mm1                  \n\t"\
1184
        "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\
1185
        "psraw $5, %%mm3                  \n\t"\
1186
        "packuswb %%mm3, %%mm0            \n\t"\
1187
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
1188
        \
1189
        "add %3, %0                       \n\t"\
1190
        "add %4, %1                       \n\t"\
1191
        "decl %2                          \n\t"\
1192
        " jnz 1b                          \n\t"\
1193
        : "+a"(src), "+c"(dst), "+d"(h)\
1194
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\
1195
        : "memory"\
1196
    );\
1197
}\
1198
\
1199
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1200
    int i;\
1201
    int16_t temp[8];\
1202
    /* quick HACK, XXX FIXME MUST be optimized */\
1203
    for(i=0; i<h; i++)\
1204
    {\
1205
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
1206
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
1207
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
1208
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
1209
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
1210
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
1211
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
1212
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
1213
        __asm__ volatile(\
1214
            "movq (%0), %%mm0           \n\t"\
1215
            "movq 8(%0), %%mm1          \n\t"\
1216
            "paddw %2, %%mm0            \n\t"\
1217
            "paddw %2, %%mm1            \n\t"\
1218
            "psraw $5, %%mm0            \n\t"\
1219
            "psraw $5, %%mm1            \n\t"\
1220
            "packuswb %%mm1, %%mm0      \n\t"\
1221
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
1222
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1223
            :"memory"\
1224
        );\
1225
        dst+=dstStride;\
1226
        src+=srcStride;\
1227
    }\
1228
}
1229

    
1230
#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1231
\
1232
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1233
    uint64_t temp[17*4];\
1234
    uint64_t *temp_ptr= temp;\
1235
    int count= 17;\
1236
\
1237
    /*FIXME unroll */\
1238
    __asm__ volatile(\
1239
        "pxor %%mm7, %%mm7              \n\t"\
1240
        "1:                             \n\t"\
1241
        "movq (%0), %%mm0               \n\t"\
1242
        "movq (%0), %%mm1               \n\t"\
1243
        "movq 8(%0), %%mm2              \n\t"\
1244
        "movq 8(%0), %%mm3              \n\t"\
1245
        "punpcklbw %%mm7, %%mm0         \n\t"\
1246
        "punpckhbw %%mm7, %%mm1         \n\t"\
1247
        "punpcklbw %%mm7, %%mm2         \n\t"\
1248
        "punpckhbw %%mm7, %%mm3         \n\t"\
1249
        "movq %%mm0, (%1)               \n\t"\
1250
        "movq %%mm1, 17*8(%1)           \n\t"\
1251
        "movq %%mm2, 2*17*8(%1)         \n\t"\
1252
        "movq %%mm3, 3*17*8(%1)         \n\t"\
1253
        "add $8, %1                     \n\t"\
1254
        "add %3, %0                     \n\t"\
1255
        "decl %2                        \n\t"\
1256
        " jnz 1b                        \n\t"\
1257
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1258
        : "r" ((x86_reg)srcStride)\
1259
        : "memory"\
1260
    );\
1261
    \
1262
    temp_ptr= temp;\
1263
    count=4;\
1264
    \
1265
/*FIXME reorder for speed */\
1266
    __asm__ volatile(\
1267
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1268
        "1:                             \n\t"\
1269
        "movq (%0), %%mm0               \n\t"\
1270
        "movq 8(%0), %%mm1              \n\t"\
1271
        "movq 16(%0), %%mm2             \n\t"\
1272
        "movq 24(%0), %%mm3             \n\t"\
1273
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1274
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1275
        "add %4, %1                     \n\t"\
1276
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1277
        \
1278
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1279
        "add %4, %1                     \n\t"\
1280
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1281
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
1282
        "add %4, %1                     \n\t"\
1283
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
1284
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
1285
        "add %4, %1                     \n\t"\
1286
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
1287
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
1288
        "add %4, %1                     \n\t"\
1289
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
1290
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
1291
        "add %4, %1                     \n\t"\
1292
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
1293
        \
1294
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
1295
        "add %4, %1                     \n\t"  \
1296
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
1297
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
1298
        \
1299
        "add $136, %0                   \n\t"\
1300
        "add %6, %1                     \n\t"\
1301
        "decl %2                        \n\t"\
1302
        " jnz 1b                        \n\t"\
1303
        \
1304
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1305
        : "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)\
1306
        :"memory"\
1307
    );\
1308
}\
1309
\
1310
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1311
    uint64_t temp[9*2];\
1312
    uint64_t *temp_ptr= temp;\
1313
    int count= 9;\
1314
\
1315
    /*FIXME unroll */\
1316
    __asm__ volatile(\
1317
        "pxor %%mm7, %%mm7              \n\t"\
1318
        "1:                             \n\t"\
1319
        "movq (%0), %%mm0               \n\t"\
1320
        "movq (%0), %%mm1               \n\t"\
1321
        "punpcklbw %%mm7, %%mm0         \n\t"\
1322
        "punpckhbw %%mm7, %%mm1         \n\t"\
1323
        "movq %%mm0, (%1)               \n\t"\
1324
        "movq %%mm1, 9*8(%1)            \n\t"\
1325
        "add $8, %1                     \n\t"\
1326
        "add %3, %0                     \n\t"\
1327
        "decl %2                        \n\t"\
1328
        " jnz 1b                        \n\t"\
1329
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1330
        : "r" ((x86_reg)srcStride)\
1331
        : "memory"\
1332
    );\
1333
    \
1334
    temp_ptr= temp;\
1335
    count=2;\
1336
    \
1337
/*FIXME reorder for speed */\
1338
    __asm__ volatile(\
1339
        /*"pxor %%mm7, %%mm7              \n\t"*/\
1340
        "1:                             \n\t"\
1341
        "movq (%0), %%mm0               \n\t"\
1342
        "movq 8(%0), %%mm1              \n\t"\
1343
        "movq 16(%0), %%mm2             \n\t"\
1344
        "movq 24(%0), %%mm3             \n\t"\
1345
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
1346
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
1347
        "add %4, %1                     \n\t"\
1348
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
1349
        \
1350
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1351
        "add %4, %1                     \n\t"\
1352
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1353
        \
1354
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1355
        "add %4, %1                     \n\t"\
1356
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1357
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1358
                \
1359
        "add $72, %0                    \n\t"\
1360
        "add %6, %1                     \n\t"\
1361
        "decl %2                        \n\t"\
1362
        " jnz 1b                        \n\t"\
1363
         \
1364
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1365
        : "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)\
1366
        : "memory"\
1367
   );\
1368
}\
1369
\
1370
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1371
    OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\
1372
}\
1373
\
1374
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1375
    uint64_t temp[8];\
1376
    uint8_t * const half= (uint8_t*)temp;\
1377
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1378
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1379
}\
1380
\
1381
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1382
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1383
}\
1384
\
1385
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1386
    uint64_t temp[8];\
1387
    uint8_t * const half= (uint8_t*)temp;\
1388
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1389
    OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
1390
}\
1391
\
1392
static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1393
    uint64_t temp[8];\
1394
    uint8_t * const half= (uint8_t*)temp;\
1395
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1396
    OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
1397
}\
1398
\
1399
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1400
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
1401
}\
1402
\
1403
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1404
    uint64_t temp[8];\
1405
    uint8_t * const half= (uint8_t*)temp;\
1406
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
1407
    OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
1408
}\
1409
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1410
    uint64_t half[8 + 9];\
1411
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1412
    uint8_t * const halfHV= ((uint8_t*)half);\
1413
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1414
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1415
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1416
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1417
}\
1418
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1419
    uint64_t half[8 + 9];\
1420
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1421
    uint8_t * const halfHV= ((uint8_t*)half);\
1422
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1423
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1424
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1425
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1426
}\
1427
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1428
    uint64_t half[8 + 9];\
1429
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1430
    uint8_t * const halfHV= ((uint8_t*)half);\
1431
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1432
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1433
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1434
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1435
}\
1436
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1437
    uint64_t half[8 + 9];\
1438
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1439
    uint8_t * const halfHV= ((uint8_t*)half);\
1440
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1441
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1442
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1443
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1444
}\
1445
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1446
    uint64_t half[8 + 9];\
1447
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1448
    uint8_t * const halfHV= ((uint8_t*)half);\
1449
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1450
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1451
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
1452
}\
1453
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1454
    uint64_t half[8 + 9];\
1455
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
1456
    uint8_t * const halfHV= ((uint8_t*)half);\
1457
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1458
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
1459
    OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
1460
}\
1461
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1462
    uint64_t half[8 + 9];\
1463
    uint8_t * const halfH= ((uint8_t*)half);\
1464
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1465
    put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
1466
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1467
}\
1468
static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1469
    uint64_t half[8 + 9];\
1470
    uint8_t * const halfH= ((uint8_t*)half);\
1471
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1472
    put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
1473
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1474
}\
1475
static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1476
    uint64_t half[9];\
1477
    uint8_t * const halfH= ((uint8_t*)half);\
1478
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1479
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
1480
}\
1481
static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1482
    OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\
1483
}\
1484
\
1485
static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1486
    uint64_t temp[32];\
1487
    uint8_t * const half= (uint8_t*)temp;\
1488
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1489
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1490
}\
1491
\
1492
static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1493
    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1494
}\
1495
\
1496
static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1497
    uint64_t temp[32];\
1498
    uint8_t * const half= (uint8_t*)temp;\
1499
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1500
    OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
1501
}\
1502
\
1503
static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1504
    uint64_t temp[32];\
1505
    uint8_t * const half= (uint8_t*)temp;\
1506
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1507
    OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
1508
}\
1509
\
1510
static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1511
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
1512
}\
1513
\
1514
static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1515
    uint64_t temp[32];\
1516
    uint8_t * const half= (uint8_t*)temp;\
1517
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
1518
    OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
1519
}\
1520
static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1521
    uint64_t half[16*2 + 17*2];\
1522
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1523
    uint8_t * const halfHV= ((uint8_t*)half);\
1524
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1525
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1526
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1527
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1528
}\
1529
static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1530
    uint64_t half[16*2 + 17*2];\
1531
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1532
    uint8_t * const halfHV= ((uint8_t*)half);\
1533
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1534
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1535
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1536
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1537
}\
1538
static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1539
    uint64_t half[16*2 + 17*2];\
1540
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1541
    uint8_t * const halfHV= ((uint8_t*)half);\
1542
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1543
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1544
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1545
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1546
}\
1547
static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1548
    uint64_t half[16*2 + 17*2];\
1549
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1550
    uint8_t * const halfHV= ((uint8_t*)half);\
1551
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1552
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1553
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1554
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1555
}\
1556
static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1557
    uint64_t half[16*2 + 17*2];\
1558
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1559
    uint8_t * const halfHV= ((uint8_t*)half);\
1560
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1561
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1562
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
1563
}\
1564
static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1565
    uint64_t half[16*2 + 17*2];\
1566
    uint8_t * const halfH= ((uint8_t*)half) + 256;\
1567
    uint8_t * const halfHV= ((uint8_t*)half);\
1568
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1569
    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
1570
    OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
1571
}\
1572
static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1573
    uint64_t half[17*2];\
1574
    uint8_t * const halfH= ((uint8_t*)half);\
1575
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1576
    put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
1577
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1578
}\
1579
static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1580
    uint64_t half[17*2];\
1581
    uint8_t * const halfH= ((uint8_t*)half);\
1582
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1583
    put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
1584
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1585
}\
1586
static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1587
    uint64_t half[17*2];\
1588
    uint8_t * const halfH= ((uint8_t*)half);\
1589
    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1590
    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
1591
}
1592

    
1593
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"
1594
#define AVG_3DNOW_OP(a,b,temp, size) \
1595
"mov" #size " " #b ", " #temp "   \n\t"\
1596
"pavgusb " #temp ", " #a "        \n\t"\
1597
"mov" #size " " #a ", " #b "      \n\t"
1598
#define AVG_MMX2_OP(a,b,temp, size) \
1599
"mov" #size " " #b ", " #temp "   \n\t"\
1600
"pavgb " #temp ", " #a "          \n\t"\
1601
"mov" #size " " #a ", " #b "      \n\t"
1602

    
1603
QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
1604
QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
1605
QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
1606
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
1607
QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
1608
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1609
QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
1610
QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
1611
QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1612

    
1613
/***********************************/
1614
/* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
1615

    
1616
#define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
1617
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1618
    OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
1619
}
1620
#define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
1621
static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1622
    OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
1623
}
1624

    
1625
#define QPEL_2TAP(OPNAME, SIZE, MMX)\
1626
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
1627
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
1628
QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
1629
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
1630
                          OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
1631
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
1632
                          OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
1633
static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
1634
                          OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
1635
static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1636
    OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
1637
}\
1638
static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1639
    OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
1640
}\
1641
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\
1642
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\
1643
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\
1644
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\
1645
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\
1646
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\
1647
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\
1648
QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
1649

    
1650
QPEL_2TAP(put_, 16, mmx2)
1651
QPEL_2TAP(avg_, 16, mmx2)
1652
QPEL_2TAP(put_,  8, mmx2)
1653
QPEL_2TAP(avg_,  8, mmx2)
1654
QPEL_2TAP(put_, 16, 3dnow)
1655
QPEL_2TAP(avg_, 16, 3dnow)
1656
QPEL_2TAP(put_,  8, 3dnow)
1657
QPEL_2TAP(avg_,  8, 3dnow)
1658

    
1659

    
1660
#if 0
1661
static void just_return(void) { return; }
1662
#endif
1663

    
1664
#if HAVE_YASM
1665
typedef void emu_edge_core_func (uint8_t *buf, const uint8_t *src,
1666
                                 x86_reg linesize, x86_reg start_y,
1667
                                 x86_reg end_y, x86_reg block_h,
1668
                                 x86_reg start_x, x86_reg end_x,
1669
                                 x86_reg block_w);
1670
extern emu_edge_core_func ff_emu_edge_core_mmx;
1671
extern emu_edge_core_func ff_emu_edge_core_sse;
1672

    
1673
static av_always_inline
1674
void emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
1675
                      int block_w, int block_h,
1676
                      int src_x, int src_y, int w, int h,
1677
                      emu_edge_core_func *core_fn)
1678
{
1679
    int start_y, start_x, end_y, end_x, src_y_add=0;
1680

    
1681
    if(src_y>= h){
1682
        src_y_add = h-1-src_y;
1683
        src_y=h-1;
1684
    }else if(src_y<=-block_h){
1685
        src_y_add = 1-block_h-src_y;
1686
        src_y=1-block_h;
1687
    }
1688
    if(src_x>= w){
1689
        src+= (w-1-src_x);
1690
        src_x=w-1;
1691
    }else if(src_x<=-block_w){
1692
        src+= (1-block_w-src_x);
1693
        src_x=1-block_w;
1694
    }
1695

    
1696
    start_y= FFMAX(0, -src_y);
1697
    start_x= FFMAX(0, -src_x);
1698
    end_y= FFMIN(block_h, h-src_y);
1699
    end_x= FFMIN(block_w, w-src_x);
1700
    assert(start_x < end_x && block_w > 0);
1701
    assert(start_y < end_y && block_h > 0);
1702

    
1703
    // fill in the to-be-copied part plus all above/below
1704
    src += (src_y_add+start_y)*linesize + start_x;
1705
    buf += start_x;
1706
    core_fn(buf, src, linesize, start_y, end_y, block_h, start_x, end_x, block_w);
1707
}
1708

    
1709
#if ARCH_X86_32
1710
static av_noinline
1711
void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src, int linesize,
1712
                          int block_w, int block_h,
1713
                          int src_x, int src_y, int w, int h)
1714
{
1715
    emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
1716
                     w, h, &ff_emu_edge_core_mmx);
1717
}
1718
#endif
1719
static av_noinline
1720
void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src, int linesize,
1721
                          int block_w, int block_h,
1722
                          int src_x, int src_y, int w, int h)
1723
{
1724
    emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
1725
                     w, h, &ff_emu_edge_core_sse);
1726
}
1727
#endif /* HAVE_YASM */
1728

    
1729
typedef void emulated_edge_mc_func (uint8_t *dst, const uint8_t *src,
1730
                                    int linesize, int block_w, int block_h,
1731
                                    int src_x, int src_y, int w, int h);
1732

    
1733
static av_always_inline
1734
void gmc(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1735
         int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height,
1736
         emulated_edge_mc_func *emu_edge_fn)
1737
{
1738
    const int w = 8;
1739
    const int ix = ox>>(16+shift);
1740
    const int iy = oy>>(16+shift);
1741
    const int oxs = ox>>4;
1742
    const int oys = oy>>4;
1743
    const int dxxs = dxx>>4;
1744
    const int dxys = dxy>>4;
1745
    const int dyxs = dyx>>4;
1746
    const int dyys = dyy>>4;
1747
    const uint16_t r4[4] = {r,r,r,r};
1748
    const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
1749
    const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
1750
    const uint64_t shift2 = 2*shift;
1751
    uint8_t edge_buf[(h+1)*stride];
1752
    int x, y;
1753

    
1754
    const int dxw = (dxx-(1<<(16+shift)))*(w-1);
1755
    const int dyh = (dyy-(1<<(16+shift)))*(h-1);
1756
    const int dxh = dxy*(h-1);
1757
    const int dyw = dyx*(w-1);
1758
    if( // non-constant fullpel offset (3% of blocks)
1759
        ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |
1760
         (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)
1761
        // uses more than 16 bits of subpel mv (only at huge resolution)
1762
        || (dxx|dxy|dyx|dyy)&15 )
1763
    {
1764
        //FIXME could still use mmx for some of the rows
1765
        ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
1766
        return;
1767
    }
1768

    
1769
    src += ix + iy*stride;
1770
    if( (unsigned)ix >= width-w ||
1771
        (unsigned)iy >= height-h )
1772
    {
1773
        emu_edge_fn(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
1774
        src = edge_buf;
1775
    }
1776

    
1777
    __asm__ volatile(
1778
        "movd         %0, %%mm6 \n\t"
1779
        "pxor      %%mm7, %%mm7 \n\t"
1780
        "punpcklwd %%mm6, %%mm6 \n\t"
1781
        "punpcklwd %%mm6, %%mm6 \n\t"
1782
        :: "r"(1<<shift)
1783
    );
1784

    
1785
    for(x=0; x<w; x+=4){
1786
        uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
1787
                            oxs - dxys + dxxs*(x+1),
1788
                            oxs - dxys + dxxs*(x+2),
1789
                            oxs - dxys + dxxs*(x+3) };
1790
        uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
1791
                            oys - dyys + dyxs*(x+1),
1792
                            oys - dyys + dyxs*(x+2),
1793
                            oys - dyys + dyxs*(x+3) };
1794

    
1795
        for(y=0; y<h; y++){
1796
            __asm__ volatile(
1797
                "movq   %0,  %%mm4 \n\t"
1798
                "movq   %1,  %%mm5 \n\t"
1799
                "paddw  %2,  %%mm4 \n\t"
1800
                "paddw  %3,  %%mm5 \n\t"
1801
                "movq   %%mm4, %0  \n\t"
1802
                "movq   %%mm5, %1  \n\t"
1803
                "psrlw  $12, %%mm4 \n\t"
1804
                "psrlw  $12, %%mm5 \n\t"
1805
                : "+m"(*dx4), "+m"(*dy4)
1806
                : "m"(*dxy4), "m"(*dyy4)
1807
            );
1808

    
1809
            __asm__ volatile(
1810
                "movq   %%mm6, %%mm2 \n\t"
1811
                "movq   %%mm6, %%mm1 \n\t"
1812
                "psubw  %%mm4, %%mm2 \n\t"
1813
                "psubw  %%mm5, %%mm1 \n\t"
1814
                "movq   %%mm2, %%mm0 \n\t"
1815
                "movq   %%mm4, %%mm3 \n\t"
1816
                "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
1817
                "pmullw %%mm5, %%mm3 \n\t" // dx*dy
1818
                "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
1819
                "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
1820

    
1821
                "movd   %4,    %%mm5 \n\t"
1822
                "movd   %3,    %%mm4 \n\t"
1823
                "punpcklbw %%mm7, %%mm5 \n\t"
1824
                "punpcklbw %%mm7, %%mm4 \n\t"
1825
                "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
1826
                "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
1827

    
1828
                "movd   %2,    %%mm5 \n\t"
1829
                "movd   %1,    %%mm4 \n\t"
1830
                "punpcklbw %%mm7, %%mm5 \n\t"
1831
                "punpcklbw %%mm7, %%mm4 \n\t"
1832
                "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
1833
                "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
1834
                "paddw  %5,    %%mm1 \n\t"
1835
                "paddw  %%mm3, %%mm2 \n\t"
1836
                "paddw  %%mm1, %%mm0 \n\t"
1837
                "paddw  %%mm2, %%mm0 \n\t"
1838

    
1839
                "psrlw    %6,    %%mm0 \n\t"
1840
                "packuswb %%mm0, %%mm0 \n\t"
1841
                "movd     %%mm0, %0    \n\t"
1842

    
1843
                : "=m"(dst[x+y*stride])
1844
                : "m"(src[0]), "m"(src[1]),
1845
                  "m"(src[stride]), "m"(src[stride+1]),
1846
                  "m"(*r4), "m"(shift2)
1847
            );
1848
            src += stride;
1849
        }
1850
        src += 4-h*stride;
1851
    }
1852
}
1853

    
1854
#if HAVE_YASM
1855
#if ARCH_X86_32
1856
static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1857
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
1858
{
1859
    gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
1860
        width, height, &emulated_edge_mc_mmx);
1861
}
1862
#endif
1863
static void gmc_sse(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1864
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
1865
{
1866
    gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
1867
        width, height, &emulated_edge_mc_sse);
1868
}
1869
#else
1870
static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
1871
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
1872
{
1873
    gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
1874
        width, height, &ff_emulated_edge_mc);
1875
}
1876
#endif
1877

    
1878
#define PREFETCH(name, op) \
1879
static void name(void *mem, int stride, int h){\
1880
    const uint8_t *p= mem;\
1881
    do{\
1882
        __asm__ volatile(#op" %0" :: "m"(*p));\
1883
        p+= stride;\
1884
    }while(--h);\
1885
}
1886
PREFETCH(prefetch_mmx2,  prefetcht0)
1887
PREFETCH(prefetch_3dnow, prefetch)
1888
#undef PREFETCH
1889

    
1890
#include "h264_qpel_mmx.c"
1891

    
1892
void ff_put_h264_chroma_mc8_mmx_rnd   (uint8_t *dst, uint8_t *src,
1893
                                       int stride, int h, int x, int y);
1894
void ff_put_rv40_chroma_mc8_mmx       (uint8_t *dst, uint8_t *src,
1895
                                       int stride, int h, int x, int y);
1896
void ff_avg_h264_chroma_mc8_mmx2_rnd  (uint8_t *dst, uint8_t *src,
1897
                                       int stride, int h, int x, int y);
1898
void ff_avg_rv40_chroma_mc8_mmx2      (uint8_t *dst, uint8_t *src,
1899
                                       int stride, int h, int x, int y);
1900
void ff_avg_h264_chroma_mc8_3dnow_rnd (uint8_t *dst, uint8_t *src,
1901
                                       int stride, int h, int x, int y);
1902
void ff_avg_rv40_chroma_mc8_3dnow     (uint8_t *dst, uint8_t *src,
1903
                                       int stride, int h, int x, int y);
1904

    
1905
void ff_put_h264_chroma_mc4_mmx       (uint8_t *dst, uint8_t *src,
1906
                                       int stride, int h, int x, int y);
1907
void ff_put_rv40_chroma_mc4_mmx       (uint8_t *dst, uint8_t *src,
1908
                                       int stride, int h, int x, int y);
1909
void ff_avg_h264_chroma_mc4_mmx2      (uint8_t *dst, uint8_t *src,
1910
                                       int stride, int h, int x, int y);
1911
void ff_avg_rv40_chroma_mc4_mmx2      (uint8_t *dst, uint8_t *src,
1912
                                       int stride, int h, int x, int y);
1913
void ff_avg_h264_chroma_mc4_3dnow     (uint8_t *dst, uint8_t *src,
1914
                                       int stride, int h, int x, int y);
1915
void ff_avg_rv40_chroma_mc4_3dnow     (uint8_t *dst, uint8_t *src,
1916
                                       int stride, int h, int x, int y);
1917

    
1918
void ff_put_h264_chroma_mc2_mmx2      (uint8_t *dst, uint8_t *src,
1919
                                       int stride, int h, int x, int y);
1920
void ff_avg_h264_chroma_mc2_mmx2      (uint8_t *dst, uint8_t *src,
1921
                                       int stride, int h, int x, int y);
1922

    
1923
void ff_put_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
1924
                                       int stride, int h, int x, int y);
1925
void ff_put_h264_chroma_mc4_ssse3     (uint8_t *dst, uint8_t *src,
1926
                                       int stride, int h, int x, int y);
1927

    
1928
void ff_avg_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
1929
                                       int stride, int h, int x, int y);
1930
void ff_avg_h264_chroma_mc4_ssse3     (uint8_t *dst, uint8_t *src,
1931
                                       int stride, int h, int x, int y);
1932

    
1933

    
1934
/* CAVS specific */
1935
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1936
    put_pixels8_mmx(dst, src, stride, 8);
1937
}
1938
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1939
    avg_pixels8_mmx(dst, src, stride, 8);
1940
}
1941
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1942
    put_pixels16_mmx(dst, src, stride, 16);
1943
}
1944
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
1945
    avg_pixels16_mmx(dst, src, stride, 16);
1946
}
1947

    
1948
/* VC1 specific */
1949
void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1950
    put_pixels8_mmx(dst, src, stride, 8);
1951
}
1952
void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
1953
    avg_pixels8_mmx2(dst, src, stride, 8);
1954
}
1955

    
1956
/* XXX: those functions should be suppressed ASAP when all IDCTs are
1957
   converted */
1958
#if CONFIG_GPL
1959
static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1960
{
1961
    ff_mmx_idct (block);
1962
    ff_put_pixels_clamped_mmx(block, dest, line_size);
1963
}
1964
static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1965
{
1966
    ff_mmx_idct (block);
1967
    ff_add_pixels_clamped_mmx(block, dest, line_size);
1968
}
1969
static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
1970
{
1971
    ff_mmxext_idct (block);
1972
    ff_put_pixels_clamped_mmx(block, dest, line_size);
1973
}
1974
static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
1975
{
1976
    ff_mmxext_idct (block);
1977
    ff_add_pixels_clamped_mmx(block, dest, line_size);
1978
}
1979
#endif
1980
static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
1981
{
1982
    ff_idct_xvid_mmx (block);
1983
    ff_put_pixels_clamped_mmx(block, dest, line_size);
1984
}
1985
static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
1986
{
1987
    ff_idct_xvid_mmx (block);
1988
    ff_add_pixels_clamped_mmx(block, dest, line_size);
1989
}
1990
static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
1991
{
1992
    ff_idct_xvid_mmx2 (block);
1993
    ff_put_pixels_clamped_mmx(block, dest, line_size);
1994
}
1995
static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
1996
{
1997
    ff_idct_xvid_mmx2 (block);
1998
    ff_add_pixels_clamped_mmx(block, dest, line_size);
1999
}
2000

    
2001
static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
2002
{
2003
    int i;
2004
    __asm__ volatile("pxor %%mm7, %%mm7":);
2005
    for(i=0; i<blocksize; i+=2) {
2006
        __asm__ volatile(
2007
            "movq    %0,    %%mm0 \n\t"
2008
            "movq    %1,    %%mm1 \n\t"
2009
            "movq    %%mm0, %%mm2 \n\t"
2010
            "movq    %%mm1, %%mm3 \n\t"
2011
            "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
2012
            "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
2013
            "pslld   $31,   %%mm2 \n\t" // keep only the sign bit
2014
            "pxor    %%mm2, %%mm1 \n\t"
2015
            "movq    %%mm3, %%mm4 \n\t"
2016
            "pand    %%mm1, %%mm3 \n\t"
2017
            "pandn   %%mm1, %%mm4 \n\t"
2018
            "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
2019
            "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
2020
            "movq    %%mm3, %1    \n\t"
2021
            "movq    %%mm0, %0    \n\t"
2022
            :"+m"(mag[i]), "+m"(ang[i])
2023
            ::"memory"
2024
        );
2025
    }
2026
    __asm__ volatile("femms");
2027
}
2028
static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
2029
{
2030
    int i;
2031

    
2032
    __asm__ volatile(
2033
            "movaps  %0,     %%xmm5 \n\t"
2034
        ::"m"(ff_pdw_80000000[0])
2035
    );
2036
    for(i=0; i<blocksize; i+=4) {
2037
        __asm__ volatile(
2038
            "movaps  %0,     %%xmm0 \n\t"
2039
            "movaps  %1,     %%xmm1 \n\t"
2040
            "xorps   %%xmm2, %%xmm2 \n\t"
2041
            "xorps   %%xmm3, %%xmm3 \n\t"
2042
            "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
2043
            "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
2044
            "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit
2045
            "xorps   %%xmm2, %%xmm1 \n\t"
2046
            "movaps  %%xmm3, %%xmm4 \n\t"
2047
            "andps   %%xmm1, %%xmm3 \n\t"
2048
            "andnps  %%xmm1, %%xmm4 \n\t"
2049
            "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
2050
            "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
2051
            "movaps  %%xmm3, %1     \n\t"
2052
            "movaps  %%xmm0, %0     \n\t"
2053
            :"+m"(mag[i]), "+m"(ang[i])
2054
            ::"memory"
2055
        );
2056
    }
2057
}
2058

    
2059
#define IF1(x) x
2060
#define IF0(x)
2061

    
2062
#define MIX5(mono,stereo)\
2063
    __asm__ volatile(\
2064
        "movss          0(%2), %%xmm5 \n"\
2065
        "movss          8(%2), %%xmm6 \n"\
2066
        "movss         24(%2), %%xmm7 \n"\
2067
        "shufps    $0, %%xmm5, %%xmm5 \n"\
2068
        "shufps    $0, %%xmm6, %%xmm6 \n"\
2069
        "shufps    $0, %%xmm7, %%xmm7 \n"\
2070
        "1: \n"\
2071
        "movaps       (%0,%1), %%xmm0 \n"\
2072
        "movaps  0x400(%0,%1), %%xmm1 \n"\
2073
        "movaps  0x800(%0,%1), %%xmm2 \n"\
2074
        "movaps  0xc00(%0,%1), %%xmm3 \n"\
2075
        "movaps 0x1000(%0,%1), %%xmm4 \n"\
2076
        "mulps         %%xmm5, %%xmm0 \n"\
2077
        "mulps         %%xmm6, %%xmm1 \n"\
2078
        "mulps         %%xmm5, %%xmm2 \n"\
2079
        "mulps         %%xmm7, %%xmm3 \n"\
2080
        "mulps         %%xmm7, %%xmm4 \n"\
2081
 stereo("addps         %%xmm1, %%xmm0 \n")\
2082
        "addps         %%xmm1, %%xmm2 \n"\
2083
        "addps         %%xmm3, %%xmm0 \n"\
2084
        "addps         %%xmm4, %%xmm2 \n"\
2085
   mono("addps         %%xmm2, %%xmm0 \n")\
2086
        "movaps  %%xmm0,      (%0,%1) \n"\
2087
 stereo("movaps  %%xmm2, 0x400(%0,%1) \n")\
2088
        "add $16, %0 \n"\
2089
        "jl 1b \n"\
2090
        :"+&r"(i)\
2091
        :"r"(samples[0]+len), "r"(matrix)\
2092
        :XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
2093
                      "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
2094
         "memory"\
2095
    );
2096

    
2097
#define MIX_MISC(stereo)\
2098
    __asm__ volatile(\
2099
        "1: \n"\
2100
        "movaps  (%3,%0), %%xmm0 \n"\
2101
 stereo("movaps   %%xmm0, %%xmm1 \n")\
2102
        "mulps    %%xmm4, %%xmm0 \n"\
2103
 stereo("mulps    %%xmm5, %%xmm1 \n")\
2104
        "lea 1024(%3,%0), %1 \n"\
2105
        "mov %5, %2 \n"\
2106
        "2: \n"\
2107
        "movaps   (%1),   %%xmm2 \n"\
2108
 stereo("movaps   %%xmm2, %%xmm3 \n")\
2109
        "mulps   (%4,%2), %%xmm2 \n"\
2110
 stereo("mulps 16(%4,%2), %%xmm3 \n")\
2111
        "addps    %%xmm2, %%xmm0 \n"\
2112
 stereo("addps    %%xmm3, %%xmm1 \n")\
2113
        "add $1024, %1 \n"\
2114
        "add $32, %2 \n"\
2115
        "jl 2b \n"\
2116
        "movaps   %%xmm0,     (%3,%0) \n"\
2117
 stereo("movaps   %%xmm1, 1024(%3,%0) \n")\
2118
        "add $16, %0 \n"\
2119
        "jl 1b \n"\
2120
        :"+&r"(i), "=&r"(j), "=&r"(k)\
2121
        :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\
2122
        :"memory"\
2123
    );
2124

    
2125
static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
2126
{
2127
    int (*matrix_cmp)[2] = (int(*)[2])matrix;
2128
    intptr_t i,j,k;
2129

    
2130
    i = -len*sizeof(float);
2131
    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]))) {
2132
        MIX5(IF0,IF1);
2133
    } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {
2134
        MIX5(IF1,IF0);
2135
    } else {
2136
        DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
2137
        j = 2*in_ch*sizeof(float);
2138
        __asm__ volatile(
2139
            "1: \n"
2140
            "sub $8, %0 \n"
2141
            "movss     (%2,%0), %%xmm4 \n"
2142
            "movss    4(%2,%0), %%xmm5 \n"
2143
            "shufps $0, %%xmm4, %%xmm4 \n"
2144
            "shufps $0, %%xmm5, %%xmm5 \n"
2145
            "movaps %%xmm4,   (%1,%0,4) \n"
2146
            "movaps %%xmm5, 16(%1,%0,4) \n"
2147
            "jg 1b \n"
2148
            :"+&r"(j)
2149
            :"r"(matrix_simd), "r"(matrix)
2150
            :"memory"
2151
        );
2152
        if(out_ch == 2) {
2153
            MIX_MISC(IF1);
2154
        } else {
2155
            MIX_MISC(IF0);
2156
        }
2157
    }
2158
}
2159

    
2160
static void vector_fmul_3dnow(float *dst, const float *src0, const float *src1, int len){
2161
    x86_reg i = (len-4)*4;
2162
    __asm__ volatile(
2163
        "1: \n\t"
2164
        "movq    (%2,%0), %%mm0 \n\t"
2165
        "movq   8(%2,%0), %%mm1 \n\t"
2166
        "pfmul   (%3,%0), %%mm0 \n\t"
2167
        "pfmul  8(%3,%0), %%mm1 \n\t"
2168
        "movq   %%mm0,  (%1,%0) \n\t"
2169
        "movq   %%mm1, 8(%1,%0) \n\t"
2170
        "sub  $16, %0 \n\t"
2171
        "jge 1b \n\t"
2172
        "femms  \n\t"
2173
        :"+r"(i)
2174
        :"r"(dst), "r"(src0), "r"(src1)
2175
        :"memory"
2176
    );
2177
}
2178
static void vector_fmul_sse(float *dst, const float *src0, const float *src1, int len){
2179
    x86_reg i = (len-8)*4;
2180
    __asm__ volatile(
2181
        "1: \n\t"
2182
        "movaps    (%2,%0), %%xmm0 \n\t"
2183
        "movaps  16(%2,%0), %%xmm1 \n\t"
2184
        "mulps     (%3,%0), %%xmm0 \n\t"
2185
        "mulps   16(%3,%0), %%xmm1 \n\t"
2186
        "movaps  %%xmm0,   (%1,%0) \n\t"
2187
        "movaps  %%xmm1, 16(%1,%0) \n\t"
2188
        "sub  $32, %0 \n\t"
2189
        "jge 1b \n\t"
2190
        :"+r"(i)
2191
        :"r"(dst), "r"(src0), "r"(src1)
2192
        :"memory"
2193
    );
2194
}
2195

    
2196
static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
2197
    x86_reg i = len*4-16;
2198
    __asm__ volatile(
2199
        "1: \n\t"
2200
        "pswapd   8(%1), %%mm0 \n\t"
2201
        "pswapd    (%1), %%mm1 \n\t"
2202
        "pfmul  (%3,%0), %%mm0 \n\t"
2203
        "pfmul 8(%3,%0), %%mm1 \n\t"
2204
        "movq  %%mm0,  (%2,%0) \n\t"
2205
        "movq  %%mm1, 8(%2,%0) \n\t"
2206
        "add   $16, %1 \n\t"
2207
        "sub   $16, %0 \n\t"
2208
        "jge   1b \n\t"
2209
        :"+r"(i), "+r"(src1)
2210
        :"r"(dst), "r"(src0)
2211
    );
2212
    __asm__ volatile("femms");
2213
}
2214
static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
2215
    x86_reg i = len*4-32;
2216
    __asm__ volatile(
2217
        "1: \n\t"
2218
        "movaps        16(%1), %%xmm0 \n\t"
2219
        "movaps          (%1), %%xmm1 \n\t"
2220
        "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
2221
        "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
2222
        "mulps        (%3,%0), %%xmm0 \n\t"
2223
        "mulps      16(%3,%0), %%xmm1 \n\t"
2224
        "movaps     %%xmm0,   (%2,%0) \n\t"
2225
        "movaps     %%xmm1, 16(%2,%0) \n\t"
2226
        "add    $32, %1 \n\t"
2227
        "sub    $32, %0 \n\t"
2228
        "jge    1b \n\t"
2229
        :"+r"(i), "+r"(src1)
2230
        :"r"(dst), "r"(src0)
2231
    );
2232
}
2233

    
2234
static void vector_fmul_add_3dnow(float *dst, const float *src0, const float *src1,
2235
                                  const float *src2, int len){
2236
    x86_reg i = (len-4)*4;
2237
    __asm__ volatile(
2238
        "1: \n\t"
2239
        "movq    (%2,%0), %%mm0 \n\t"
2240
        "movq   8(%2,%0), %%mm1 \n\t"
2241
        "pfmul   (%3,%0), %%mm0 \n\t"
2242
        "pfmul  8(%3,%0), %%mm1 \n\t"
2243
        "pfadd   (%4,%0), %%mm0 \n\t"
2244
        "pfadd  8(%4,%0), %%mm1 \n\t"
2245
        "movq  %%mm0,   (%1,%0) \n\t"
2246
        "movq  %%mm1,  8(%1,%0) \n\t"
2247
        "sub  $16, %0 \n\t"
2248
        "jge  1b \n\t"
2249
        :"+r"(i)
2250
        :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2251
        :"memory"
2252
    );
2253
    __asm__ volatile("femms");
2254
}
2255
static void vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
2256
                                const float *src2, int len){
2257
    x86_reg i = (len-8)*4;
2258
    __asm__ volatile(
2259
        "1: \n\t"
2260
        "movaps   (%2,%0), %%xmm0 \n\t"
2261
        "movaps 16(%2,%0), %%xmm1 \n\t"
2262
        "mulps    (%3,%0), %%xmm0 \n\t"
2263
        "mulps  16(%3,%0), %%xmm1 \n\t"
2264
        "addps    (%4,%0), %%xmm0 \n\t"
2265
        "addps  16(%4,%0), %%xmm1 \n\t"
2266
        "movaps %%xmm0,   (%1,%0) \n\t"
2267
        "movaps %%xmm1, 16(%1,%0) \n\t"
2268
        "sub  $32, %0 \n\t"
2269
        "jge  1b \n\t"
2270
        :"+r"(i)
2271
        :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
2272
        :"memory"
2273
    );
2274
}
2275

    
2276
#if HAVE_6REGS
2277
static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,
2278
                                      const float *win, int len){
2279
    x86_reg i = -len*4;
2280
    x86_reg j = len*4-8;
2281
    __asm__ volatile(
2282
        "1: \n"
2283
        "pswapd  (%5,%1), %%mm1 \n"
2284
        "movq    (%5,%0), %%mm0 \n"
2285
        "pswapd  (%4,%1), %%mm5 \n"
2286
        "movq    (%3,%0), %%mm4 \n"
2287
        "movq      %%mm0, %%mm2 \n"
2288
        "movq      %%mm1, %%mm3 \n"
2289
        "pfmul     %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]
2290
        "pfmul     %%mm5, %%mm3 \n" // src1[    j]*win[len+j]
2291
        "pfmul     %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]
2292
        "pfmul     %%mm5, %%mm0 \n" // src1[    j]*win[len+i]
2293
        "pfadd     %%mm3, %%mm2 \n"
2294
        "pfsub     %%mm0, %%mm1 \n"
2295
        "pswapd    %%mm2, %%mm2 \n"
2296
        "movq      %%mm1, (%2,%0) \n"
2297
        "movq      %%mm2, (%2,%1) \n"
2298
        "sub $8, %1 \n"
2299
        "add $8, %0 \n"
2300
        "jl 1b \n"
2301
        "femms \n"
2302
        :"+r"(i), "+r"(j)
2303
        :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2304
    );
2305
}
2306

    
2307
static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,
2308
                                   const float *win, int len){
2309
    x86_reg i = -len*4;
2310
    x86_reg j = len*4-16;
2311
    __asm__ volatile(
2312
        "1: \n"
2313
        "movaps       (%5,%1), %%xmm1 \n"
2314
        "movaps       (%5,%0), %%xmm0 \n"
2315
        "movaps       (%4,%1), %%xmm5 \n"
2316
        "movaps       (%3,%0), %%xmm4 \n"
2317
        "shufps $0x1b, %%xmm1, %%xmm1 \n"
2318
        "shufps $0x1b, %%xmm5, %%xmm5 \n"
2319
        "movaps        %%xmm0, %%xmm2 \n"
2320
        "movaps        %%xmm1, %%xmm3 \n"
2321
        "mulps         %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]
2322
        "mulps         %%xmm5, %%xmm3 \n" // src1[    j]*win[len+j]
2323
        "mulps         %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]
2324
        "mulps         %%xmm5, %%xmm0 \n" // src1[    j]*win[len+i]
2325
        "addps         %%xmm3, %%xmm2 \n"
2326
        "subps         %%xmm0, %%xmm1 \n"
2327
        "shufps $0x1b, %%xmm2, %%xmm2 \n"
2328
        "movaps        %%xmm1, (%2,%0) \n"
2329
        "movaps        %%xmm2, (%2,%1) \n"
2330
        "sub $16, %1 \n"
2331
        "add $16, %0 \n"
2332
        "jl 1b \n"
2333
        :"+r"(i), "+r"(j)
2334
        :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)
2335
    );
2336
}
2337
#endif /* HAVE_6REGS */
2338

    
2339
static void vector_clipf_sse(float *dst, const float *src, float min, float max,
2340
                             int len)
2341
{
2342
    x86_reg i = (len-16)*4;
2343
    __asm__ volatile(
2344
        "movss  %3, %%xmm4 \n"
2345
        "movss  %4, %%xmm5 \n"
2346
        "shufps $0, %%xmm4, %%xmm4 \n"
2347
        "shufps $0, %%xmm5, %%xmm5 \n"
2348
        "1: \n\t"
2349
        "movaps    (%2,%0), %%xmm0 \n\t" // 3/1 on intel
2350
        "movaps  16(%2,%0), %%xmm1 \n\t"
2351
        "movaps  32(%2,%0), %%xmm2 \n\t"
2352
        "movaps  48(%2,%0), %%xmm3 \n\t"
2353
        "maxps      %%xmm4, %%xmm0 \n\t"
2354
        "maxps      %%xmm4, %%xmm1 \n\t"
2355
        "maxps      %%xmm4, %%xmm2 \n\t"
2356
        "maxps      %%xmm4, %%xmm3 \n\t"
2357
        "minps      %%xmm5, %%xmm0 \n\t"
2358
        "minps      %%xmm5, %%xmm1 \n\t"
2359
        "minps      %%xmm5, %%xmm2 \n\t"
2360
        "minps      %%xmm5, %%xmm3 \n\t"
2361
        "movaps  %%xmm0,   (%1,%0) \n\t"
2362
        "movaps  %%xmm1, 16(%1,%0) \n\t"
2363
        "movaps  %%xmm2, 32(%1,%0) \n\t"
2364
        "movaps  %%xmm3, 48(%1,%0) \n\t"
2365
        "sub  $64, %0 \n\t"
2366
        "jge 1b \n\t"
2367
        :"+&r"(i)
2368
        :"r"(dst), "r"(src), "m"(min), "m"(max)
2369
        :"memory"
2370
    );
2371
}
2372

    
2373
void ff_vp3_idct_mmx(int16_t *input_data);
2374
void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block);
2375
void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block);
2376

    
2377
void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int line_size, const DCTELEM *block);
2378

    
2379
void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
2380
void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
2381

    
2382
void ff_vp3_idct_sse2(int16_t *input_data);
2383
void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block);
2384
void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block);
2385

    
2386
int32_t ff_scalarproduct_int16_mmx2(const int16_t *v1, const int16_t *v2, int order, int shift);
2387
int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, int order, int shift);
2388
int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2389
int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2390
int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
2391
void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
2392
int  ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, int w, int left);
2393
int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, int left);
2394

    
2395
float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
2396

    
2397
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
2398
{
2399
    int mm_flags = av_get_cpu_flags();
2400

    
2401
    if (avctx->dsp_mask) {
2402
        if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)
2403
            mm_flags |= (avctx->dsp_mask & 0xffff);
2404
        else
2405
            mm_flags &= ~(avctx->dsp_mask & 0xffff);
2406
    }
2407

    
2408
#if 0
2409
    av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
2410
    if (mm_flags & AV_CPU_FLAG_MMX)
2411
        av_log(avctx, AV_LOG_INFO, " mmx");
2412
    if (mm_flags & AV_CPU_FLAG_MMX2)
2413
        av_log(avctx, AV_LOG_INFO, " mmx2");
2414
    if (mm_flags & AV_CPU_FLAG_3DNOW)
2415
        av_log(avctx, AV_LOG_INFO, " 3dnow");
2416
    if (mm_flags & AV_CPU_FLAG_SSE)
2417
        av_log(avctx, AV_LOG_INFO, " sse");
2418
    if (mm_flags & AV_CPU_FLAG_SSE2)
2419
        av_log(avctx, AV_LOG_INFO, " sse2");
2420
    av_log(avctx, AV_LOG_INFO, "\n");
2421
#endif
2422

    
2423
    if (mm_flags & AV_CPU_FLAG_MMX) {
2424
        const int idct_algo= avctx->idct_algo;
2425

    
2426
        if(avctx->lowres==0){
2427
            if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
2428
                c->idct_put= ff_simple_idct_put_mmx;
2429
                c->idct_add= ff_simple_idct_add_mmx;
2430
                c->idct    = ff_simple_idct_mmx;
2431
                c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
2432
#if CONFIG_GPL
2433
            }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
2434
                if(mm_flags & AV_CPU_FLAG_MMX2){
2435
                    c->idct_put= ff_libmpeg2mmx2_idct_put;
2436
                    c->idct_add= ff_libmpeg2mmx2_idct_add;
2437
                    c->idct    = ff_mmxext_idct;
2438
                }else{
2439
                    c->idct_put= ff_libmpeg2mmx_idct_put;
2440
                    c->idct_add= ff_libmpeg2mmx_idct_add;
2441
                    c->idct    = ff_mmx_idct;
2442
                }
2443
                c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
2444
#endif
2445
            }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&
2446
                     idct_algo==FF_IDCT_VP3 && HAVE_YASM){
2447
                if(mm_flags & AV_CPU_FLAG_SSE2){
2448
                    c->idct_put= ff_vp3_idct_put_sse2;
2449
                    c->idct_add= ff_vp3_idct_add_sse2;
2450
                    c->idct    = ff_vp3_idct_sse2;
2451
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2452
                }else{
2453
                    c->idct_put= ff_vp3_idct_put_mmx;
2454
                    c->idct_add= ff_vp3_idct_add_mmx;
2455
                    c->idct    = ff_vp3_idct_mmx;
2456
                    c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
2457
                }
2458
            }else if(idct_algo==FF_IDCT_CAVS){
2459
                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2460
            }else if(idct_algo==FF_IDCT_XVIDMMX){
2461
                if(mm_flags & AV_CPU_FLAG_SSE2){
2462
                    c->idct_put= ff_idct_xvid_sse2_put;
2463
                    c->idct_add= ff_idct_xvid_sse2_add;
2464
                    c->idct    = ff_idct_xvid_sse2;
2465
                    c->idct_permutation_type= FF_SSE2_IDCT_PERM;
2466
                }else if(mm_flags & AV_CPU_FLAG_MMX2){
2467
                    c->idct_put= ff_idct_xvid_mmx2_put;
2468
                    c->idct_add= ff_idct_xvid_mmx2_add;
2469
                    c->idct    = ff_idct_xvid_mmx2;
2470
                }else{
2471
                    c->idct_put= ff_idct_xvid_mmx_put;
2472
                    c->idct_add= ff_idct_xvid_mmx_add;
2473
                    c->idct    = ff_idct_xvid_mmx;
2474
                }
2475
            }
2476
        }
2477

    
2478
        c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
2479
        c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
2480
        c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
2481
        c->clear_block  = clear_block_mmx;
2482
        c->clear_blocks = clear_blocks_mmx;
2483
        if ((mm_flags & AV_CPU_FLAG_SSE) &&
2484
            !(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)){
2485
            /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
2486
            c->clear_block  = clear_block_sse;
2487
            c->clear_blocks = clear_blocks_sse;
2488
        }
2489

    
2490
#define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2491
        c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
2492
        c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
2493
        c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
2494
        c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU
2495

    
2496
        SET_HPEL_FUNCS(put, 0, 16, mmx);
2497
        SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
2498
        SET_HPEL_FUNCS(avg, 0, 16, mmx);
2499
        SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
2500
        SET_HPEL_FUNCS(put, 1, 8, mmx);
2501
        SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
2502
        SET_HPEL_FUNCS(avg, 1, 8, mmx);
2503
        SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
2504

    
2505
#if ARCH_X86_32 || !HAVE_YASM
2506
        c->gmc= gmc_mmx;
2507
#endif
2508
#if ARCH_X86_32 && HAVE_YASM
2509
        c->emulated_edge_mc = emulated_edge_mc_mmx;
2510
#endif
2511

    
2512
        c->add_bytes= add_bytes_mmx;
2513
        c->add_bytes_l2= add_bytes_l2_mmx;
2514

    
2515
        c->draw_edges = draw_edges_mmx;
2516

    
2517
        if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
2518
            c->h263_v_loop_filter= h263_v_loop_filter_mmx;
2519
            c->h263_h_loop_filter= h263_h_loop_filter_mmx;
2520
        }
2521

    
2522
#if HAVE_YASM
2523
        c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_mmx_rnd;
2524
        c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_mmx;
2525

    
2526
        c->put_rv40_chroma_pixels_tab[0]= ff_put_rv40_chroma_mc8_mmx;
2527
        c->put_rv40_chroma_pixels_tab[1]= ff_put_rv40_chroma_mc4_mmx;
2528
#endif
2529

    
2530
        if (mm_flags & AV_CPU_FLAG_MMX2) {
2531
            c->prefetch = prefetch_mmx2;
2532

    
2533
            c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
2534
            c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
2535

    
2536
            c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
2537
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
2538
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
2539

    
2540
            c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
2541
            c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
2542

    
2543
            c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
2544
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
2545
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
2546

    
2547
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2548
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
2549
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
2550
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
2551
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
2552
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
2553
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
2554

    
2555
                if (CONFIG_VP3_DECODER && HAVE_YASM) {
2556
                    c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2;
2557
                    c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;
2558
                }
2559
            }
2560
            if (CONFIG_VP3_DECODER && HAVE_YASM) {
2561
                c->vp3_idct_dc_add = ff_vp3_idct_dc_add_mmx2;
2562
            }
2563

    
2564
            if (CONFIG_VP3_DECODER
2565
                && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {
2566
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_mmx2;
2567
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_mmx2;
2568
            }
2569

    
2570
#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \
2571
            c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \
2572
            c->PFX ## _pixels_tab[IDX][ 1] = PFX ## SIZE ## _mc10_ ## CPU; \
2573
            c->PFX ## _pixels_tab[IDX][ 2] = PFX ## SIZE ## _mc20_ ## CPU; \
2574
            c->PFX ## _pixels_tab[IDX][ 3] = PFX ## SIZE ## _mc30_ ## CPU; \
2575
            c->PFX ## _pixels_tab[IDX][ 4] = PFX ## SIZE ## _mc01_ ## CPU; \
2576
            c->PFX ## _pixels_tab[IDX][ 5] = PFX ## SIZE ## _mc11_ ## CPU; \
2577
            c->PFX ## _pixels_tab[IDX][ 6] = PFX ## SIZE ## _mc21_ ## CPU; \
2578
            c->PFX ## _pixels_tab[IDX][ 7] = PFX ## SIZE ## _mc31_ ## CPU; \
2579
            c->PFX ## _pixels_tab[IDX][ 8] = PFX ## SIZE ## _mc02_ ## CPU; \
2580
            c->PFX ## _pixels_tab[IDX][ 9] = PFX ## SIZE ## _mc12_ ## CPU; \
2581
            c->PFX ## _pixels_tab[IDX][10] = PFX ## SIZE ## _mc22_ ## CPU; \
2582
            c->PFX ## _pixels_tab[IDX][11] = PFX ## SIZE ## _mc32_ ## CPU; \
2583
            c->PFX ## _pixels_tab[IDX][12] = PFX ## SIZE ## _mc03_ ## CPU; \
2584
            c->PFX ## _pixels_tab[IDX][13] = PFX ## SIZE ## _mc13_ ## CPU; \
2585
            c->PFX ## _pixels_tab[IDX][14] = PFX ## SIZE ## _mc23_ ## CPU; \
2586
            c->PFX ## _pixels_tab[IDX][15] = PFX ## SIZE ## _mc33_ ## CPU
2587

    
2588
            SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2);
2589
            SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2);
2590
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2);
2591
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2);
2592
            SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2);
2593
            SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2);
2594

    
2595
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2);
2596
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2);
2597
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2);
2598
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2);
2599
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2);
2600
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2);
2601

    
2602
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2);
2603
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2);
2604
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2);
2605
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2);
2606

    
2607
#if HAVE_YASM
2608
            c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_mmx2;
2609
            c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_mmx2;
2610

    
2611
            c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_mmx2_rnd;
2612
            c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_mmx2;
2613
            c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_mmx2;
2614
            c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_mmx2;
2615

    
2616
            c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
2617
#endif
2618
#if HAVE_7REGS && HAVE_TEN_OPERANDS
2619
            if( mm_flags&AV_CPU_FLAG_3DNOW )
2620
                c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
2621
#endif
2622

    
2623
            c->add_png_paeth_prediction= add_png_paeth_prediction_mmx2;
2624
        } else if (mm_flags & AV_CPU_FLAG_3DNOW) {
2625
            c->prefetch = prefetch_3dnow;
2626

    
2627
            c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
2628
            c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
2629

    
2630
            c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
2631
            c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
2632
            c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
2633

    
2634
            c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
2635
            c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
2636

    
2637
            c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
2638
            c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
2639
            c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
2640

    
2641
            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
2642
                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
2643
                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
2644
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
2645
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
2646
                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
2647
                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
2648
            }
2649

    
2650
            if (CONFIG_VP3_DECODER
2651
                && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {
2652
                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_3dnow;
2653
                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_3dnow;
2654
            }
2655

    
2656
            SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow);
2657
            SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow);
2658
            SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow);
2659
            SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow);
2660
            SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow);
2661
            SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow);
2662

    
2663
            SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow);
2664
            SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow);
2665
            SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow);
2666
            SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow);
2667
            SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow);
2668
            SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow);
2669

    
2670
            SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow);
2671
            SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow);
2672
            SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow);
2673
            SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow);
2674

    
2675
#if HAVE_YASM
2676
            c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_3dnow_rnd;
2677
            c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_3dnow;
2678

    
2679
            c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_3dnow;
2680
            c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_3dnow;
2681
#endif
2682
        }
2683

    
2684

    
2685
#define H264_QPEL_FUNCS(x, y, CPU)\
2686
            c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\
2687
            c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\
2688
            c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\
2689
            c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;
2690
        if((mm_flags & AV_CPU_FLAG_SSE2) && !(mm_flags & AV_CPU_FLAG_3DNOW)){
2691
            // these functions are slower than mmx on AMD, but faster on Intel
2692
            c->put_pixels_tab[0][0] = put_pixels16_sse2;
2693
            c->put_no_rnd_pixels_tab[0][0] = put_pixels16_sse2;
2694
            c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
2695
            H264_QPEL_FUNCS(0, 0, sse2);
2696
        }
2697
        if(mm_flags & AV_CPU_FLAG_SSE2){
2698
            H264_QPEL_FUNCS(0, 1, sse2);
2699
            H264_QPEL_FUNCS(0, 2, sse2);
2700
            H264_QPEL_FUNCS(0, 3, sse2);
2701
            H264_QPEL_FUNCS(1, 1, sse2);
2702
            H264_QPEL_FUNCS(1, 2, sse2);
2703
            H264_QPEL_FUNCS(1, 3, sse2);
2704
            H264_QPEL_FUNCS(2, 1, sse2);
2705
            H264_QPEL_FUNCS(2, 2, sse2);
2706
            H264_QPEL_FUNCS(2, 3, sse2);
2707
            H264_QPEL_FUNCS(3, 1, sse2);
2708
            H264_QPEL_FUNCS(3, 2, sse2);
2709
            H264_QPEL_FUNCS(3, 3, sse2);
2710
        }
2711
#if HAVE_SSSE3
2712
        if(mm_flags & AV_CPU_FLAG_SSSE3){
2713
            H264_QPEL_FUNCS(1, 0, ssse3);
2714
            H264_QPEL_FUNCS(1, 1, ssse3);
2715
            H264_QPEL_FUNCS(1, 2, ssse3);
2716
            H264_QPEL_FUNCS(1, 3, ssse3);
2717
            H264_QPEL_FUNCS(2, 0, ssse3);
2718
            H264_QPEL_FUNCS(2, 1, ssse3);
2719
            H264_QPEL_FUNCS(2, 2, ssse3);
2720
            H264_QPEL_FUNCS(2, 3, ssse3);
2721
            H264_QPEL_FUNCS(3, 0, ssse3);
2722
            H264_QPEL_FUNCS(3, 1, ssse3);
2723
            H264_QPEL_FUNCS(3, 2, ssse3);
2724
            H264_QPEL_FUNCS(3, 3, ssse3);
2725
            c->add_png_paeth_prediction= add_png_paeth_prediction_ssse3;
2726
#if HAVE_YASM
2727
            c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_ssse3_rnd;
2728
            c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_ssse3_rnd;
2729
            c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_ssse3;
2730
            c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_ssse3;
2731
            c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
2732
            if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe
2733
                c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
2734
#endif
2735
        }
2736
#endif
2737

    
2738
        if(mm_flags & AV_CPU_FLAG_3DNOW){
2739
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
2740
            c->vector_fmul = vector_fmul_3dnow;
2741
        }
2742
        if(mm_flags & AV_CPU_FLAG_3DNOWEXT){
2743
            c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
2744
#if HAVE_6REGS
2745
            c->vector_fmul_window = vector_fmul_window_3dnow2;
2746
#endif
2747
        }
2748
        if(mm_flags & AV_CPU_FLAG_MMX2){
2749
#if HAVE_YASM
2750
            c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;
2751
            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;
2752
#endif
2753
        }
2754
        if(mm_flags & AV_CPU_FLAG_SSE){
2755
            c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
2756
            c->ac3_downmix = ac3_downmix_sse;
2757
            c->vector_fmul = vector_fmul_sse;
2758
            c->vector_fmul_reverse = vector_fmul_reverse_sse;
2759
            c->vector_fmul_add = vector_fmul_add_sse;
2760
#if HAVE_6REGS
2761
            c->vector_fmul_window = vector_fmul_window_sse;
2762
#endif
2763
            c->vector_clipf = vector_clipf_sse;
2764
#if HAVE_YASM
2765
            c->scalarproduct_float = ff_scalarproduct_float_sse;
2766
#endif
2767
        }
2768
        if(mm_flags & AV_CPU_FLAG_3DNOW)
2769
            c->vector_fmul_add = vector_fmul_add_3dnow; // faster than sse
2770
        if(mm_flags & AV_CPU_FLAG_SSE2){
2771
#if HAVE_YASM
2772
            c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
2773
            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
2774

    
2775
            c->emulated_edge_mc = emulated_edge_mc_sse;
2776
            c->gmc= gmc_sse;
2777
#endif
2778
        }
2779
        if((mm_flags & AV_CPU_FLAG_SSSE3) && !(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW)) && HAVE_YASM) // cachesplit
2780
            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
2781
    }
2782

    
2783
    if (CONFIG_ENCODERS)
2784
        dsputilenc_init_mmx(c, avctx);
2785

    
2786
#if 0
2787
    // for speed testing
2788
    get_pixels = just_return;
2789
    put_pixels_clamped = just_return;
2790
    add_pixels_clamped = just_return;
2791

2792
    pix_abs16x16 = just_return;
2793
    pix_abs16x16_x2 = just_return;
2794
    pix_abs16x16_y2 = just_return;
2795
    pix_abs16x16_xy2 = just_return;
2796

2797
    put_pixels_tab[0] = just_return;
2798
    put_pixels_tab[1] = just_return;
2799
    put_pixels_tab[2] = just_return;
2800
    put_pixels_tab[3] = just_return;
2801

2802
    put_no_rnd_pixels_tab[0] = just_return;
2803
    put_no_rnd_pixels_tab[1] = just_return;
2804
    put_no_rnd_pixels_tab[2] = just_return;
2805
    put_no_rnd_pixels_tab[3] = just_return;
2806

2807
    avg_pixels_tab[0] = just_return;
2808
    avg_pixels_tab[1] = just_return;
2809
    avg_pixels_tab[2] = just_return;
2810
    avg_pixels_tab[3] = just_return;
2811

2812
    avg_no_rnd_pixels_tab[0] = just_return;
2813
    avg_no_rnd_pixels_tab[1] = just_return;
2814
    avg_no_rnd_pixels_tab[2] = just_return;
2815
    avg_no_rnd_pixels_tab[3] = just_return;
2816

2817
    //av_fdct = just_return;
2818
    //ff_idct = just_return;
2819
#endif
2820
}