Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / h264dsp_mmx.c @ ec199cc9

History | View | Annotate | Download (82.7 KB)

1
/*
2
 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

    
21
#include "dsputil_mmx.h"
22

    
23
/***********************************/
24
/* IDCT */
25

    
26
#define SUMSUB_BADC( a, b, c, d ) \
27
    "paddw "#b", "#a" \n\t"\
28
    "paddw "#d", "#c" \n\t"\
29
    "paddw "#b", "#b" \n\t"\
30
    "paddw "#d", "#d" \n\t"\
31
    "psubw "#a", "#b" \n\t"\
32
    "psubw "#c", "#d" \n\t"
33

    
34
#define SUMSUBD2_AB( a, b, t ) \
35
    "movq  "#b", "#t" \n\t"\
36
    "psraw  $1 , "#b" \n\t"\
37
    "paddw "#a", "#b" \n\t"\
38
    "psraw  $1 , "#a" \n\t"\
39
    "psubw "#t", "#a" \n\t"
40

    
41
#define IDCT4_1D( s02, s13, d02, d13, t ) \
42
    SUMSUB_BA  ( s02, d02 )\
43
    SUMSUBD2_AB( s13, d13, t )\
44
    SUMSUB_BADC( d13, s02, s13, d02 )
45

    
46
#define STORE_DIFF_4P( p, t, z ) \
47
    "psraw      $6,     "#p" \n\t"\
48
    "movd       (%0),   "#t" \n\t"\
49
    "punpcklbw "#z",    "#t" \n\t"\
50
    "paddsw    "#t",    "#p" \n\t"\
51
    "packuswb  "#z",    "#p" \n\t"\
52
    "movd      "#p",    (%0) \n\t"
53

    
54
static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
55
{
56
    /* Load dct coeffs */
57
    asm volatile(
58
        "movq   (%0), %%mm0 \n\t"
59
        "movq  8(%0), %%mm1 \n\t"
60
        "movq 16(%0), %%mm2 \n\t"
61
        "movq 24(%0), %%mm3 \n\t"
62
    :: "r"(block) );
63

    
64
    asm volatile(
65
        /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
66
        IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
67

    
68
        "movq      %0,    %%mm6 \n\t"
69
        /* in: 1,4,0,2  out: 1,2,3,0 */
70
        TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
71

    
72
        "paddw     %%mm6, %%mm3 \n\t"
73

    
74
        /* mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13 */
75
        IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
76

    
77
        "pxor %%mm7, %%mm7    \n\t"
78
    :: "m"(ff_pw_32));
79

    
80
    asm volatile(
81
    STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
82
        "add %1, %0             \n\t"
83
    STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
84
        "add %1, %0             \n\t"
85
    STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
86
        "add %1, %0             \n\t"
87
    STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
88
        : "+r"(dst)
89
        : "r" ((long)stride)
90
    );
91
}
92

    
93
static inline void h264_idct8_1d(int16_t *block)
94
{
95
    asm volatile(
96
        "movq 112(%0), %%mm7  \n\t"
97
        "movq  80(%0), %%mm0  \n\t"
98
        "movq  48(%0), %%mm3  \n\t"
99
        "movq  16(%0), %%mm5  \n\t"
100

    
101
        "movq   %%mm0, %%mm4  \n\t"
102
        "movq   %%mm5, %%mm1  \n\t"
103
        "psraw  $1,    %%mm4  \n\t"
104
        "psraw  $1,    %%mm1  \n\t"
105
        "paddw  %%mm0, %%mm4  \n\t"
106
        "paddw  %%mm5, %%mm1  \n\t"
107
        "paddw  %%mm7, %%mm4  \n\t"
108
        "paddw  %%mm0, %%mm1  \n\t"
109
        "psubw  %%mm5, %%mm4  \n\t"
110
        "paddw  %%mm3, %%mm1  \n\t"
111

    
112
        "psubw  %%mm3, %%mm5  \n\t"
113
        "psubw  %%mm3, %%mm0  \n\t"
114
        "paddw  %%mm7, %%mm5  \n\t"
115
        "psubw  %%mm7, %%mm0  \n\t"
116
        "psraw  $1,    %%mm3  \n\t"
117
        "psraw  $1,    %%mm7  \n\t"
118
        "psubw  %%mm3, %%mm5  \n\t"
119
        "psubw  %%mm7, %%mm0  \n\t"
120

    
121
        "movq   %%mm4, %%mm3  \n\t"
122
        "movq   %%mm1, %%mm7  \n\t"
123
        "psraw  $2,    %%mm1  \n\t"
124
        "psraw  $2,    %%mm3  \n\t"
125
        "paddw  %%mm5, %%mm3  \n\t"
126
        "psraw  $2,    %%mm5  \n\t"
127
        "paddw  %%mm0, %%mm1  \n\t"
128
        "psraw  $2,    %%mm0  \n\t"
129
        "psubw  %%mm4, %%mm5  \n\t"
130
        "psubw  %%mm0, %%mm7  \n\t"
131

    
132
        "movq  32(%0), %%mm2  \n\t"
133
        "movq  96(%0), %%mm6  \n\t"
134
        "movq   %%mm2, %%mm4  \n\t"
135
        "movq   %%mm6, %%mm0  \n\t"
136
        "psraw  $1,    %%mm4  \n\t"
137
        "psraw  $1,    %%mm6  \n\t"
138
        "psubw  %%mm0, %%mm4  \n\t"
139
        "paddw  %%mm2, %%mm6  \n\t"
140

    
141
        "movq    (%0), %%mm2  \n\t"
142
        "movq  64(%0), %%mm0  \n\t"
143
        SUMSUB_BA( %%mm0, %%mm2 )
144
        SUMSUB_BA( %%mm6, %%mm0 )
145
        SUMSUB_BA( %%mm4, %%mm2 )
146
        SUMSUB_BA( %%mm7, %%mm6 )
147
        SUMSUB_BA( %%mm5, %%mm4 )
148
        SUMSUB_BA( %%mm3, %%mm2 )
149
        SUMSUB_BA( %%mm1, %%mm0 )
150
        :: "r"(block)
151
    );
152
}
153

    
154
static void ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
155
{
156
    int i;
157
    int16_t __attribute__ ((aligned(8))) b2[64];
158

    
159
    block[0] += 32;
160

    
161
    for(i=0; i<2; i++){
162
        DECLARE_ALIGNED_8(uint64_t, tmp);
163

    
164
        h264_idct8_1d(block+4*i);
165

    
166
        asm volatile(
167
            "movq   %%mm7,    %0   \n\t"
168
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
169
            "movq   %%mm0,  8(%1)  \n\t"
170
            "movq   %%mm6, 24(%1)  \n\t"
171
            "movq   %%mm7, 40(%1)  \n\t"
172
            "movq   %%mm4, 56(%1)  \n\t"
173
            "movq    %0,    %%mm7  \n\t"
174
            TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
175
            "movq   %%mm7,   (%1)  \n\t"
176
            "movq   %%mm1, 16(%1)  \n\t"
177
            "movq   %%mm0, 32(%1)  \n\t"
178
            "movq   %%mm3, 48(%1)  \n\t"
179
            : "=m"(tmp)
180
            : "r"(b2+32*i)
181
            : "memory"
182
        );
183
    }
184

    
185
    for(i=0; i<2; i++){
186
        h264_idct8_1d(b2+4*i);
187

    
188
        asm volatile(
189
            "psraw     $6, %%mm7  \n\t"
190
            "psraw     $6, %%mm6  \n\t"
191
            "psraw     $6, %%mm5  \n\t"
192
            "psraw     $6, %%mm4  \n\t"
193
            "psraw     $6, %%mm3  \n\t"
194
            "psraw     $6, %%mm2  \n\t"
195
            "psraw     $6, %%mm1  \n\t"
196
            "psraw     $6, %%mm0  \n\t"
197

    
198
            "movq   %%mm7,    (%0)  \n\t"
199
            "movq   %%mm5,  16(%0)  \n\t"
200
            "movq   %%mm3,  32(%0)  \n\t"
201
            "movq   %%mm1,  48(%0)  \n\t"
202
            "movq   %%mm0,  64(%0)  \n\t"
203
            "movq   %%mm2,  80(%0)  \n\t"
204
            "movq   %%mm4,  96(%0)  \n\t"
205
            "movq   %%mm6, 112(%0)  \n\t"
206
            :: "r"(b2+4*i)
207
            : "memory"
208
        );
209
    }
210

    
211
    add_pixels_clamped_mmx(b2, dst, stride);
212
}
213

    
214
#define STORE_DIFF_8P( p, d, t, z )\
215
        "movq       "#d", "#t" \n"\
216
        "psraw       $6,  "#p" \n"\
217
        "punpcklbw  "#z", "#t" \n"\
218
        "paddsw     "#t", "#p" \n"\
219
        "packuswb   "#p", "#p" \n"\
220
        "movq       "#p", "#d" \n"
221

    
222
#define H264_IDCT8_1D_SSE2(a,b,c,d,e,f,g,h)\
223
        "movdqa     "#c", "#a" \n"\
224
        "movdqa     "#g", "#e" \n"\
225
        "psraw       $1,  "#c" \n"\
226
        "psraw       $1,  "#g" \n"\
227
        "psubw      "#e", "#c" \n"\
228
        "paddw      "#a", "#g" \n"\
229
        "movdqa     "#b", "#e" \n"\
230
        "psraw       $1,  "#e" \n"\
231
        "paddw      "#b", "#e" \n"\
232
        "paddw      "#d", "#e" \n"\
233
        "paddw      "#f", "#e" \n"\
234
        "movdqa     "#f", "#a" \n"\
235
        "psraw       $1,  "#a" \n"\
236
        "paddw      "#f", "#a" \n"\
237
        "paddw      "#h", "#a" \n"\
238
        "psubw      "#b", "#a" \n"\
239
        "psubw      "#d", "#b" \n"\
240
        "psubw      "#d", "#f" \n"\
241
        "paddw      "#h", "#b" \n"\
242
        "psubw      "#h", "#f" \n"\
243
        "psraw       $1,  "#d" \n"\
244
        "psraw       $1,  "#h" \n"\
245
        "psubw      "#d", "#b" \n"\
246
        "psubw      "#h", "#f" \n"\
247
        "movdqa     "#e", "#d" \n"\
248
        "movdqa     "#a", "#h" \n"\
249
        "psraw       $2,  "#d" \n"\
250
        "psraw       $2,  "#h" \n"\
251
        "paddw      "#f", "#d" \n"\
252
        "paddw      "#b", "#h" \n"\
253
        "psraw       $2,  "#f" \n"\
254
        "psraw       $2,  "#b" \n"\
255
        "psubw      "#f", "#e" \n"\
256
        "psubw      "#a", "#b" \n"\
257
        "movdqa 0x00(%1), "#a" \n"\
258
        "movdqa 0x40(%1), "#f" \n"\
259
        SUMSUB_BA(f, a)\
260
        SUMSUB_BA(g, f)\
261
        SUMSUB_BA(c, a)\
262
        SUMSUB_BA(e, g)\
263
        SUMSUB_BA(b, c)\
264
        SUMSUB_BA(h, a)\
265
        SUMSUB_BA(d, f)
266

    
267
static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
268
{
269
    asm volatile(
270
        "movdqa   0x10(%1), %%xmm1 \n"
271
        "movdqa   0x20(%1), %%xmm2 \n"
272
        "movdqa   0x30(%1), %%xmm3 \n"
273
        "movdqa   0x50(%1), %%xmm5 \n"
274
        "movdqa   0x60(%1), %%xmm6 \n"
275
        "movdqa   0x70(%1), %%xmm7 \n"
276
        H264_IDCT8_1D_SSE2(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)
277
        TRANSPOSE8(%%xmm4, %%xmm1, %%xmm7, %%xmm3, %%xmm5, %%xmm0, %%xmm2, %%xmm6, (%1))
278
        "paddw          %4, %%xmm4 \n"
279
        "movdqa     %%xmm4, 0x00(%1) \n"
280
        "movdqa     %%xmm2, 0x40(%1) \n"
281
        H264_IDCT8_1D_SSE2(%%xmm4, %%xmm0, %%xmm6, %%xmm3, %%xmm2, %%xmm5, %%xmm7, %%xmm1)
282
        "movdqa     %%xmm6, 0x60(%1) \n"
283
        "movdqa     %%xmm7, 0x70(%1) \n"
284
        "pxor       %%xmm7, %%xmm7 \n"
285
        STORE_DIFF_8P(%%xmm2, (%0),      %%xmm6, %%xmm7)
286
        STORE_DIFF_8P(%%xmm0, (%0,%2),   %%xmm6, %%xmm7)
287
        STORE_DIFF_8P(%%xmm1, (%0,%2,2), %%xmm6, %%xmm7)
288
        STORE_DIFF_8P(%%xmm3, (%0,%3),   %%xmm6, %%xmm7)
289
        "lea     (%0,%2,4), %0 \n"
290
        STORE_DIFF_8P(%%xmm5, (%0),      %%xmm6, %%xmm7)
291
        STORE_DIFF_8P(%%xmm4, (%0,%2),   %%xmm6, %%xmm7)
292
        "movdqa   0x60(%1), %%xmm0 \n"
293
        "movdqa   0x70(%1), %%xmm1 \n"
294
        STORE_DIFF_8P(%%xmm0, (%0,%2,2), %%xmm6, %%xmm7)
295
        STORE_DIFF_8P(%%xmm1, (%0,%3),   %%xmm6, %%xmm7)
296
        :"+r"(dst)
297
        :"r"(block), "r"((long)stride), "r"(3L*stride), "m"(ff_pw_32)
298
    );
299
}
300

    
301
static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
302
{
303
    int dc = (block[0] + 32) >> 6;
304
    asm volatile(
305
        "movd          %0, %%mm0 \n\t"
306
        "pshufw $0, %%mm0, %%mm0 \n\t"
307
        "pxor       %%mm1, %%mm1 \n\t"
308
        "psubw      %%mm0, %%mm1 \n\t"
309
        "packuswb   %%mm0, %%mm0 \n\t"
310
        "packuswb   %%mm1, %%mm1 \n\t"
311
        ::"r"(dc)
312
    );
313
    asm volatile(
314
        "movd          %0, %%mm2 \n\t"
315
        "movd          %1, %%mm3 \n\t"
316
        "movd          %2, %%mm4 \n\t"
317
        "movd          %3, %%mm5 \n\t"
318
        "paddusb    %%mm0, %%mm2 \n\t"
319
        "paddusb    %%mm0, %%mm3 \n\t"
320
        "paddusb    %%mm0, %%mm4 \n\t"
321
        "paddusb    %%mm0, %%mm5 \n\t"
322
        "psubusb    %%mm1, %%mm2 \n\t"
323
        "psubusb    %%mm1, %%mm3 \n\t"
324
        "psubusb    %%mm1, %%mm4 \n\t"
325
        "psubusb    %%mm1, %%mm5 \n\t"
326
        "movd       %%mm2, %0    \n\t"
327
        "movd       %%mm3, %1    \n\t"
328
        "movd       %%mm4, %2    \n\t"
329
        "movd       %%mm5, %3    \n\t"
330
        :"+m"(*(uint32_t*)(dst+0*stride)),
331
         "+m"(*(uint32_t*)(dst+1*stride)),
332
         "+m"(*(uint32_t*)(dst+2*stride)),
333
         "+m"(*(uint32_t*)(dst+3*stride))
334
    );
335
}
336

    
337
static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
338
{
339
    int dc = (block[0] + 32) >> 6;
340
    int y;
341
    asm volatile(
342
        "movd          %0, %%mm0 \n\t"
343
        "pshufw $0, %%mm0, %%mm0 \n\t"
344
        "pxor       %%mm1, %%mm1 \n\t"
345
        "psubw      %%mm0, %%mm1 \n\t"
346
        "packuswb   %%mm0, %%mm0 \n\t"
347
        "packuswb   %%mm1, %%mm1 \n\t"
348
        ::"r"(dc)
349
    );
350
    for(y=2; y--; dst += 4*stride){
351
    asm volatile(
352
        "movq          %0, %%mm2 \n\t"
353
        "movq          %1, %%mm3 \n\t"
354
        "movq          %2, %%mm4 \n\t"
355
        "movq          %3, %%mm5 \n\t"
356
        "paddusb    %%mm0, %%mm2 \n\t"
357
        "paddusb    %%mm0, %%mm3 \n\t"
358
        "paddusb    %%mm0, %%mm4 \n\t"
359
        "paddusb    %%mm0, %%mm5 \n\t"
360
        "psubusb    %%mm1, %%mm2 \n\t"
361
        "psubusb    %%mm1, %%mm3 \n\t"
362
        "psubusb    %%mm1, %%mm4 \n\t"
363
        "psubusb    %%mm1, %%mm5 \n\t"
364
        "movq       %%mm2, %0    \n\t"
365
        "movq       %%mm3, %1    \n\t"
366
        "movq       %%mm4, %2    \n\t"
367
        "movq       %%mm5, %3    \n\t"
368
        :"+m"(*(uint64_t*)(dst+0*stride)),
369
         "+m"(*(uint64_t*)(dst+1*stride)),
370
         "+m"(*(uint64_t*)(dst+2*stride)),
371
         "+m"(*(uint64_t*)(dst+3*stride))
372
    );
373
    }
374
}
375

    
376

    
377
/***********************************/
378
/* deblocking */
379

    
380
// out: o = |x-y|>a
381
// clobbers: t
382
#define DIFF_GT_MMX(x,y,a,o,t)\
383
    "movq     "#y", "#t"  \n\t"\
384
    "movq     "#x", "#o"  \n\t"\
385
    "psubusb  "#x", "#t"  \n\t"\
386
    "psubusb  "#y", "#o"  \n\t"\
387
    "por      "#t", "#o"  \n\t"\
388
    "psubusb  "#a", "#o"  \n\t"
389

    
390
// out: o = |x-y|>a
391
// clobbers: t
392
#define DIFF_GT2_MMX(x,y,a,o,t)\
393
    "movq     "#y", "#t"  \n\t"\
394
    "movq     "#x", "#o"  \n\t"\
395
    "psubusb  "#x", "#t"  \n\t"\
396
    "psubusb  "#y", "#o"  \n\t"\
397
    "psubusb  "#a", "#t"  \n\t"\
398
    "psubusb  "#a", "#o"  \n\t"\
399
    "pcmpeqb  "#t", "#o"  \n\t"\
400

    
401
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
402
// out: mm5=beta-1, mm7=mask
403
// clobbers: mm4,mm6
404
#define H264_DEBLOCK_MASK(alpha1, beta1) \
405
    "pshufw $0, "#alpha1", %%mm4 \n\t"\
406
    "pshufw $0, "#beta1 ", %%mm5 \n\t"\
407
    "packuswb  %%mm4, %%mm4      \n\t"\
408
    "packuswb  %%mm5, %%mm5      \n\t"\
409
    DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
410
    DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
411
    "por       %%mm4, %%mm7      \n\t"\
412
    DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
413
    "por       %%mm4, %%mm7      \n\t"\
414
    "pxor      %%mm6, %%mm6      \n\t"\
415
    "pcmpeqb   %%mm6, %%mm7      \n\t"
416

    
417
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
418
// out: mm1=p0' mm2=q0'
419
// clobbers: mm0,3-6
420
#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
421
        "movq    %%mm1              , %%mm5 \n\t"\
422
        "pxor    %%mm2              , %%mm5 \n\t" /* p0^q0*/\
423
        "pand    "#pb_01"           , %%mm5 \n\t" /* (p0^q0)&1*/\
424
        "pcmpeqb %%mm4              , %%mm4 \n\t"\
425
        "pxor    %%mm4              , %%mm3 \n\t"\
426
        "pavgb   %%mm0              , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
427
        "pavgb   "MANGLE(ff_pb_3)"  , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
428
        "pxor    %%mm1              , %%mm4 \n\t"\
429
        "pavgb   %%mm2              , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
430
        "pavgb   %%mm5              , %%mm3 \n\t"\
431
        "paddusb %%mm4              , %%mm3 \n\t" /* d+128+33*/\
432
        "movq    "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
433
        "psubusb %%mm3              , %%mm6 \n\t"\
434
        "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
435
        "pminub  %%mm7              , %%mm6 \n\t"\
436
        "pminub  %%mm7              , %%mm3 \n\t"\
437
        "psubusb %%mm6              , %%mm1 \n\t"\
438
        "psubusb %%mm3              , %%mm2 \n\t"\
439
        "paddusb %%mm3              , %%mm1 \n\t"\
440
        "paddusb %%mm6              , %%mm2 \n\t"
441

    
442
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
443
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
444
// clobbers: q2, tmp, tc0
445
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
446
        "movq     %%mm1,  "#tmp"   \n\t"\
447
        "pavgb    %%mm2,  "#tmp"   \n\t"\
448
        "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
449
        "pxor   "q2addr", "#tmp"   \n\t"\
450
        "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
451
        "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
452
        "movq     "#p1",  "#tmp"   \n\t"\
453
        "psubusb  "#tc0", "#tmp"   \n\t"\
454
        "paddusb  "#p1",  "#tc0"   \n\t"\
455
        "pmaxub   "#tmp", "#q2"    \n\t"\
456
        "pminub   "#tc0", "#q2"    \n\t"\
457
        "movq     "#q2",  "q1addr" \n\t"
458

    
459
static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
460
{
461
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);
462

    
463
    asm volatile(
464
        "movq    (%1,%3), %%mm0    \n\t" //p1
465
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
466
        "movq    (%2),    %%mm2    \n\t" //q0
467
        "movq    (%2,%3), %%mm3    \n\t" //q1
468
        H264_DEBLOCK_MASK(%6, %7)
469

    
470
        "movd      %5,    %%mm4    \n\t"
471
        "punpcklbw %%mm4, %%mm4    \n\t"
472
        "punpcklwd %%mm4, %%mm4    \n\t"
473
        "pcmpeqb   %%mm3, %%mm3    \n\t"
474
        "movq      %%mm4, %%mm6    \n\t"
475
        "pcmpgtb   %%mm3, %%mm4    \n\t"
476
        "movq      %%mm6, 8+%0     \n\t"
477
        "pand      %%mm4, %%mm7    \n\t"
478
        "movq      %%mm7, %0       \n\t"
479

    
480
        /* filter p1 */
481
        "movq     (%1),   %%mm3    \n\t" //p2
482
        DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
483
        "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
484
        "pand     8+%0,   %%mm7    \n\t" // mask & tc0
485
        "movq     %%mm7,  %%mm4    \n\t"
486
        "psubb    %%mm6,  %%mm7    \n\t"
487
        "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
488
        H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
489

    
490
        /* filter q1 */
491
        "movq    (%2,%3,2), %%mm4  \n\t" //q2
492
        DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
493
        "pand     %0,     %%mm6    \n\t"
494
        "movq     8+%0,   %%mm5    \n\t" // can be merged with the and below but is slower then
495
        "pand     %%mm6,  %%mm5    \n\t"
496
        "psubb    %%mm6,  %%mm7    \n\t"
497
        "movq    (%2,%3), %%mm3    \n\t"
498
        H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
499

    
500
        /* filter p0, q0 */
501
        H264_DEBLOCK_P0_Q0(%8, unused)
502
        "movq      %%mm1, (%1,%3,2) \n\t"
503
        "movq      %%mm2, (%2)      \n\t"
504

    
505
        : "=m"(*tmp0)
506
        : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
507
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
508
          "m"(ff_bone)
509
    );
510
}
511

    
512
static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
513
{
514
    if((tc0[0] & tc0[1]) >= 0)
515
        h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
516
    if((tc0[2] & tc0[3]) >= 0)
517
        h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
518
}
519
static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
520
{
521
    //FIXME: could cut some load/stores by merging transpose with filter
522
    // also, it only needs to transpose 6x8
523
    DECLARE_ALIGNED_8(uint8_t, trans[8*8]);
524
    int i;
525
    for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
526
        if((tc0[0] & tc0[1]) < 0)
527
            continue;
528
        transpose4x4(trans,       pix-4,          8, stride);
529
        transpose4x4(trans  +4*8, pix,            8, stride);
530
        transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
531
        transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
532
        h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
533
        transpose4x4(pix-2,          trans  +2*8, stride, 8);
534
        transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
535
    }
536
}
537

    
538
static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
539
{
540
    asm volatile(
541
        "movq    (%0),    %%mm0     \n\t" //p1
542
        "movq    (%0,%2), %%mm1     \n\t" //p0
543
        "movq    (%1),    %%mm2     \n\t" //q0
544
        "movq    (%1,%2), %%mm3     \n\t" //q1
545
        H264_DEBLOCK_MASK(%4, %5)
546
        "movd      %3,    %%mm6     \n\t"
547
        "punpcklbw %%mm6, %%mm6     \n\t"
548
        "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
549
        H264_DEBLOCK_P0_Q0(%6, %7)
550
        "movq      %%mm1, (%0,%2)   \n\t"
551
        "movq      %%mm2, (%1)      \n\t"
552

    
553
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
554
           "r"(*(uint32_t*)tc0),
555
           "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
556
    );
557
}
558

    
559
static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
560
{
561
    h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
562
}
563

    
564
static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
565
{
566
    //FIXME: could cut some load/stores by merging transpose with filter
567
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
568
    transpose4x4(trans, pix-2, 8, stride);
569
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
570
    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
571
    transpose4x4(pix-2, trans, stride, 8);
572
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
573
}
574

    
575
// p0 = (p0 + q1 + 2*p1 + 2) >> 2
576
#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
577
    "movq    "#p0", %%mm4  \n\t"\
578
    "pxor    "#q1", %%mm4  \n\t"\
579
    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
580
    "pavgb   "#q1", "#p0"  \n\t"\
581
    "psubusb %%mm4, "#p0"  \n\t"\
582
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
583

    
584
static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
585
{
586
    asm volatile(
587
        "movq    (%0),    %%mm0     \n\t"
588
        "movq    (%0,%2), %%mm1     \n\t"
589
        "movq    (%1),    %%mm2     \n\t"
590
        "movq    (%1,%2), %%mm3     \n\t"
591
        H264_DEBLOCK_MASK(%3, %4)
592
        "movq    %%mm1,   %%mm5     \n\t"
593
        "movq    %%mm2,   %%mm6     \n\t"
594
        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
595
        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
596
        "psubb   %%mm5,   %%mm1     \n\t"
597
        "psubb   %%mm6,   %%mm2     \n\t"
598
        "pand    %%mm7,   %%mm1     \n\t"
599
        "pand    %%mm7,   %%mm2     \n\t"
600
        "paddb   %%mm5,   %%mm1     \n\t"
601
        "paddb   %%mm6,   %%mm2     \n\t"
602
        "movq    %%mm1,   (%0,%2)   \n\t"
603
        "movq    %%mm2,   (%1)      \n\t"
604
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
605
           "m"(alpha1), "m"(beta1), "m"(ff_bone)
606
    );
607
}
608

    
609
static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
610
{
611
    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
612
}
613

    
614
static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
615
{
616
    //FIXME: could cut some load/stores by merging transpose with filter
617
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
618
    transpose4x4(trans, pix-2, 8, stride);
619
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
620
    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
621
    transpose4x4(pix-2, trans, stride, 8);
622
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
623
}
624

    
625
static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
626
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1 ) {
627
    int dir;
628
    asm volatile(
629
        "pxor %%mm7, %%mm7 \n\t"
630
        "movq %0, %%mm6 \n\t"
631
        "movq %1, %%mm5 \n\t"
632
        "movq %2, %%mm4 \n\t"
633
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
634
    );
635
    // could do a special case for dir==0 && edges==1, but it only reduces the
636
    // average filter time by 1.2%
637
    for( dir=1; dir>=0; dir-- ) {
638
        const int d_idx = dir ? -8 : -1;
639
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
640
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
641
        int b_idx, edge, l;
642
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
643
            asm volatile(
644
                "pand %0, %%mm0 \n\t"
645
                ::"m"(mask_dir)
646
            );
647
            if(!(mask_mv & edge)) {
648
                asm volatile("pxor %%mm0, %%mm0 \n\t":);
649
                for( l = bidir; l >= 0; l-- ) {
650
                    asm volatile(
651
                        "movd %0, %%mm1 \n\t"
652
                        "punpckldq %1, %%mm1 \n\t"
653
                        "movq %%mm1, %%mm2 \n\t"
654
                        "psrlw $7, %%mm2 \n\t"
655
                        "pand %%mm6, %%mm2 \n\t"
656
                        "por %%mm2, %%mm1 \n\t" // ref_cache with -2 mapped to -1
657
                        "punpckldq %%mm1, %%mm2 \n\t"
658
                        "pcmpeqb %%mm2, %%mm1 \n\t"
659
                        "paddb %%mm6, %%mm1 \n\t"
660
                        "punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn]
661
                        "por %%mm1, %%mm0 \n\t"
662

    
663
                        "movq %2, %%mm1 \n\t"
664
                        "movq %3, %%mm2 \n\t"
665
                        "psubw %4, %%mm1 \n\t"
666
                        "psubw %5, %%mm2 \n\t"
667
                        "packsswb %%mm2, %%mm1 \n\t"
668
                        "paddb %%mm5, %%mm1 \n\t"
669
                        "pminub %%mm4, %%mm1 \n\t"
670
                        "pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit
671
                        "por %%mm1, %%mm0 \n\t"
672
                        ::"m"(ref[l][b_idx]),
673
                          "m"(ref[l][b_idx+d_idx]),
674
                          "m"(mv[l][b_idx][0]),
675
                          "m"(mv[l][b_idx+2][0]),
676
                          "m"(mv[l][b_idx+d_idx][0]),
677
                          "m"(mv[l][b_idx+d_idx+2][0])
678
                    );
679
                }
680
            }
681
            asm volatile(
682
                "movd %0, %%mm1 \n\t"
683
                "por  %1, %%mm1 \n\t"
684
                "punpcklbw %%mm7, %%mm1 \n\t"
685
                "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]
686
                ::"m"(nnz[b_idx]),
687
                  "m"(nnz[b_idx+d_idx])
688
            );
689
            asm volatile(
690
                "pcmpeqw %%mm7, %%mm0 \n\t"
691
                "pcmpeqw %%mm7, %%mm0 \n\t"
692
                "psrlw $15, %%mm0 \n\t" // nonzero -> 1
693
                "psrlw $14, %%mm1 \n\t"
694
                "movq %%mm0, %%mm2 \n\t"
695
                "por %%mm1, %%mm2 \n\t"
696
                "psrlw $1, %%mm1 \n\t"
697
                "pandn %%mm2, %%mm1 \n\t"
698
                "movq %%mm1, %0 \n\t"
699
                :"=m"(*bS[dir][edge])
700
                ::"memory"
701
            );
702
        }
703
        edges = 4;
704
        step = 1;
705
    }
706
    asm volatile(
707
        "movq   (%0), %%mm0 \n\t"
708
        "movq  8(%0), %%mm1 \n\t"
709
        "movq 16(%0), %%mm2 \n\t"
710
        "movq 24(%0), %%mm3 \n\t"
711
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
712
        "movq %%mm0,   (%0) \n\t"
713
        "movq %%mm3,  8(%0) \n\t"
714
        "movq %%mm4, 16(%0) \n\t"
715
        "movq %%mm2, 24(%0) \n\t"
716
        ::"r"(bS[0])
717
        :"memory"
718
    );
719
}
720

    
721
/***********************************/
722
/* motion compensation */
723

    
724
#define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
725
        "mov"#q" "#C", "#T"         \n\t"\
726
        "mov"#d" (%0), "#F"         \n\t"\
727
        "paddw "#D", "#T"           \n\t"\
728
        "psllw $2, "#T"             \n\t"\
729
        "psubw "#B", "#T"           \n\t"\
730
        "psubw "#E", "#T"           \n\t"\
731
        "punpcklbw "#Z", "#F"       \n\t"\
732
        "pmullw %4, "#T"            \n\t"\
733
        "paddw %5, "#A"             \n\t"\
734
        "add %2, %0                 \n\t"\
735
        "paddw "#F", "#A"           \n\t"\
736
        "paddw "#A", "#T"           \n\t"\
737
        "psraw $5, "#T"             \n\t"\
738
        "packuswb "#T", "#T"        \n\t"\
739
        OP(T, (%1), A, d)\
740
        "add %3, %1                 \n\t"
741

    
742
#define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
743
        "mov"#q" "#C", "#T"         \n\t"\
744
        "mov"#d" (%0), "#F"         \n\t"\
745
        "paddw "#D", "#T"           \n\t"\
746
        "psllw $2, "#T"             \n\t"\
747
        "paddw %4, "#A"             \n\t"\
748
        "psubw "#B", "#T"           \n\t"\
749
        "psubw "#E", "#T"           \n\t"\
750
        "punpcklbw "#Z", "#F"       \n\t"\
751
        "pmullw %3, "#T"            \n\t"\
752
        "paddw "#F", "#A"           \n\t"\
753
        "add %2, %0                 \n\t"\
754
        "paddw "#A", "#T"           \n\t"\
755
        "mov"#q" "#T", "#OF"(%1)    \n\t"
756

    
757
#define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
758
#define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
759
#define QPEL_H264V_XMM(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%xmm6,%%xmm7,q,dqa)
760
#define QPEL_H264HV_XMM(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%xmm6,%%xmm7,q,dqa)
761

    
762

    
763
#define QPEL_H264(OPNAME, OP, MMX)\
764
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
765
    int h=4;\
766
\
767
    asm volatile(\
768
        "pxor %%mm7, %%mm7          \n\t"\
769
        "movq %5, %%mm4             \n\t"\
770
        "movq %6, %%mm5             \n\t"\
771
        "1:                         \n\t"\
772
        "movd  -1(%0), %%mm1        \n\t"\
773
        "movd    (%0), %%mm2        \n\t"\
774
        "movd   1(%0), %%mm3        \n\t"\
775
        "movd   2(%0), %%mm0        \n\t"\
776
        "punpcklbw %%mm7, %%mm1     \n\t"\
777
        "punpcklbw %%mm7, %%mm2     \n\t"\
778
        "punpcklbw %%mm7, %%mm3     \n\t"\
779
        "punpcklbw %%mm7, %%mm0     \n\t"\
780
        "paddw %%mm0, %%mm1         \n\t"\
781
        "paddw %%mm3, %%mm2         \n\t"\
782
        "movd  -2(%0), %%mm0        \n\t"\
783
        "movd   3(%0), %%mm3        \n\t"\
784
        "punpcklbw %%mm7, %%mm0     \n\t"\
785
        "punpcklbw %%mm7, %%mm3     \n\t"\
786
        "paddw %%mm3, %%mm0         \n\t"\
787
        "psllw $2, %%mm2            \n\t"\
788
        "psubw %%mm1, %%mm2         \n\t"\
789
        "pmullw %%mm4, %%mm2        \n\t"\
790
        "paddw %%mm5, %%mm0         \n\t"\
791
        "paddw %%mm2, %%mm0         \n\t"\
792
        "psraw $5, %%mm0            \n\t"\
793
        "packuswb %%mm0, %%mm0      \n\t"\
794
        OP(%%mm0, (%1),%%mm6, d)\
795
        "add %3, %0                 \n\t"\
796
        "add %4, %1                 \n\t"\
797
        "decl %2                    \n\t"\
798
        " jnz 1b                    \n\t"\
799
        : "+a"(src), "+c"(dst), "+g"(h)\
800
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
801
        : "memory"\
802
    );\
803
}\
804
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
805
    int h=4;\
806
    asm volatile(\
807
        "pxor %%mm7, %%mm7          \n\t"\
808
        "movq %0, %%mm4             \n\t"\
809
        "movq %1, %%mm5             \n\t"\
810
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
811
    );\
812
    do{\
813
    asm volatile(\
814
        "movd  -1(%0), %%mm1        \n\t"\
815
        "movd    (%0), %%mm2        \n\t"\
816
        "movd   1(%0), %%mm3        \n\t"\
817
        "movd   2(%0), %%mm0        \n\t"\
818
        "punpcklbw %%mm7, %%mm1     \n\t"\
819
        "punpcklbw %%mm7, %%mm2     \n\t"\
820
        "punpcklbw %%mm7, %%mm3     \n\t"\
821
        "punpcklbw %%mm7, %%mm0     \n\t"\
822
        "paddw %%mm0, %%mm1         \n\t"\
823
        "paddw %%mm3, %%mm2         \n\t"\
824
        "movd  -2(%0), %%mm0        \n\t"\
825
        "movd   3(%0), %%mm3        \n\t"\
826
        "punpcklbw %%mm7, %%mm0     \n\t"\
827
        "punpcklbw %%mm7, %%mm3     \n\t"\
828
        "paddw %%mm3, %%mm0         \n\t"\
829
        "psllw $2, %%mm2            \n\t"\
830
        "psubw %%mm1, %%mm2         \n\t"\
831
        "pmullw %%mm4, %%mm2        \n\t"\
832
        "paddw %%mm5, %%mm0         \n\t"\
833
        "paddw %%mm2, %%mm0         \n\t"\
834
        "movd   (%2), %%mm3         \n\t"\
835
        "psraw $5, %%mm0            \n\t"\
836
        "packuswb %%mm0, %%mm0      \n\t"\
837
        PAVGB" %%mm3, %%mm0         \n\t"\
838
        OP(%%mm0, (%1),%%mm6, d)\
839
        "add %4, %0                 \n\t"\
840
        "add %4, %1                 \n\t"\
841
        "add %3, %2                 \n\t"\
842
        : "+a"(src), "+c"(dst), "+d"(src2)\
843
        : "D"((long)src2Stride), "S"((long)dstStride)\
844
        : "memory"\
845
    );\
846
    }while(--h);\
847
}\
848
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
849
    src -= 2*srcStride;\
850
    asm volatile(\
851
        "pxor %%mm7, %%mm7          \n\t"\
852
        "movd (%0), %%mm0           \n\t"\
853
        "add %2, %0                 \n\t"\
854
        "movd (%0), %%mm1           \n\t"\
855
        "add %2, %0                 \n\t"\
856
        "movd (%0), %%mm2           \n\t"\
857
        "add %2, %0                 \n\t"\
858
        "movd (%0), %%mm3           \n\t"\
859
        "add %2, %0                 \n\t"\
860
        "movd (%0), %%mm4           \n\t"\
861
        "add %2, %0                 \n\t"\
862
        "punpcklbw %%mm7, %%mm0     \n\t"\
863
        "punpcklbw %%mm7, %%mm1     \n\t"\
864
        "punpcklbw %%mm7, %%mm2     \n\t"\
865
        "punpcklbw %%mm7, %%mm3     \n\t"\
866
        "punpcklbw %%mm7, %%mm4     \n\t"\
867
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
868
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
869
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
870
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
871
         \
872
        : "+a"(src), "+c"(dst)\
873
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
874
        : "memory"\
875
    );\
876
}\
877
static av_noinline void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
878
    int h=4;\
879
    int w=3;\
880
    src -= 2*srcStride+2;\
881
    while(w--){\
882
        asm volatile(\
883
            "pxor %%mm7, %%mm7      \n\t"\
884
            "movd (%0), %%mm0       \n\t"\
885
            "add %2, %0             \n\t"\
886
            "movd (%0), %%mm1       \n\t"\
887
            "add %2, %0             \n\t"\
888
            "movd (%0), %%mm2       \n\t"\
889
            "add %2, %0             \n\t"\
890
            "movd (%0), %%mm3       \n\t"\
891
            "add %2, %0             \n\t"\
892
            "movd (%0), %%mm4       \n\t"\
893
            "add %2, %0             \n\t"\
894
            "punpcklbw %%mm7, %%mm0 \n\t"\
895
            "punpcklbw %%mm7, %%mm1 \n\t"\
896
            "punpcklbw %%mm7, %%mm2 \n\t"\
897
            "punpcklbw %%mm7, %%mm3 \n\t"\
898
            "punpcklbw %%mm7, %%mm4 \n\t"\
899
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
900
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
901
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
902
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
903
             \
904
            : "+a"(src)\
905
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
906
            : "memory"\
907
        );\
908
        tmp += 4;\
909
        src += 4 - 9*srcStride;\
910
    }\
911
    tmp -= 3*4;\
912
    asm volatile(\
913
        "1:                         \n\t"\
914
        "movq     (%0), %%mm0       \n\t"\
915
        "paddw  10(%0), %%mm0       \n\t"\
916
        "movq    2(%0), %%mm1       \n\t"\
917
        "paddw   8(%0), %%mm1       \n\t"\
918
        "movq    4(%0), %%mm2       \n\t"\
919
        "paddw   6(%0), %%mm2       \n\t"\
920
        "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
921
        "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
922
        "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
923
        "paddsw %%mm2, %%mm0        \n\t"\
924
        "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
925
        "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
926
        "psraw $6, %%mm0            \n\t"\
927
        "packuswb %%mm0, %%mm0      \n\t"\
928
        OP(%%mm0, (%1),%%mm7, d)\
929
        "add $24, %0                \n\t"\
930
        "add %3, %1                 \n\t"\
931
        "decl %2                    \n\t"\
932
        " jnz 1b                    \n\t"\
933
        : "+a"(tmp), "+c"(dst), "+g"(h)\
934
        : "S"((long)dstStride)\
935
        : "memory"\
936
    );\
937
}\
938
\
939
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
940
    int h=8;\
941
    asm volatile(\
942
        "pxor %%mm7, %%mm7          \n\t"\
943
        "movq %5, %%mm6             \n\t"\
944
        "1:                         \n\t"\
945
        "movq    (%0), %%mm0        \n\t"\
946
        "movq   1(%0), %%mm2        \n\t"\
947
        "movq %%mm0, %%mm1          \n\t"\
948
        "movq %%mm2, %%mm3          \n\t"\
949
        "punpcklbw %%mm7, %%mm0     \n\t"\
950
        "punpckhbw %%mm7, %%mm1     \n\t"\
951
        "punpcklbw %%mm7, %%mm2     \n\t"\
952
        "punpckhbw %%mm7, %%mm3     \n\t"\
953
        "paddw %%mm2, %%mm0         \n\t"\
954
        "paddw %%mm3, %%mm1         \n\t"\
955
        "psllw $2, %%mm0            \n\t"\
956
        "psllw $2, %%mm1            \n\t"\
957
        "movq   -1(%0), %%mm2       \n\t"\
958
        "movq    2(%0), %%mm4       \n\t"\
959
        "movq %%mm2, %%mm3          \n\t"\
960
        "movq %%mm4, %%mm5          \n\t"\
961
        "punpcklbw %%mm7, %%mm2     \n\t"\
962
        "punpckhbw %%mm7, %%mm3     \n\t"\
963
        "punpcklbw %%mm7, %%mm4     \n\t"\
964
        "punpckhbw %%mm7, %%mm5     \n\t"\
965
        "paddw %%mm4, %%mm2         \n\t"\
966
        "paddw %%mm3, %%mm5         \n\t"\
967
        "psubw %%mm2, %%mm0         \n\t"\
968
        "psubw %%mm5, %%mm1         \n\t"\
969
        "pmullw %%mm6, %%mm0        \n\t"\
970
        "pmullw %%mm6, %%mm1        \n\t"\
971
        "movd   -2(%0), %%mm2       \n\t"\
972
        "movd    7(%0), %%mm5       \n\t"\
973
        "punpcklbw %%mm7, %%mm2     \n\t"\
974
        "punpcklbw %%mm7, %%mm5     \n\t"\
975
        "paddw %%mm3, %%mm2         \n\t"\
976
        "paddw %%mm5, %%mm4         \n\t"\
977
        "movq %6, %%mm5             \n\t"\
978
        "paddw %%mm5, %%mm2         \n\t"\
979
        "paddw %%mm5, %%mm4         \n\t"\
980
        "paddw %%mm2, %%mm0         \n\t"\
981
        "paddw %%mm4, %%mm1         \n\t"\
982
        "psraw $5, %%mm0            \n\t"\
983
        "psraw $5, %%mm1            \n\t"\
984
        "packuswb %%mm1, %%mm0      \n\t"\
985
        OP(%%mm0, (%1),%%mm5, q)\
986
        "add %3, %0                 \n\t"\
987
        "add %4, %1                 \n\t"\
988
        "decl %2                    \n\t"\
989
        " jnz 1b                    \n\t"\
990
        : "+a"(src), "+c"(dst), "+g"(h)\
991
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
992
        : "memory"\
993
    );\
994
}\
995
\
996
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
997
    int h=8;\
998
    asm volatile(\
999
        "pxor %%mm7, %%mm7          \n\t"\
1000
        "movq %0, %%mm6             \n\t"\
1001
        :: "m"(ff_pw_5)\
1002
    );\
1003
    do{\
1004
    asm volatile(\
1005
        "movq    (%0), %%mm0        \n\t"\
1006
        "movq   1(%0), %%mm2        \n\t"\
1007
        "movq %%mm0, %%mm1          \n\t"\
1008
        "movq %%mm2, %%mm3          \n\t"\
1009
        "punpcklbw %%mm7, %%mm0     \n\t"\
1010
        "punpckhbw %%mm7, %%mm1     \n\t"\
1011
        "punpcklbw %%mm7, %%mm2     \n\t"\
1012
        "punpckhbw %%mm7, %%mm3     \n\t"\
1013
        "paddw %%mm2, %%mm0         \n\t"\
1014
        "paddw %%mm3, %%mm1         \n\t"\
1015
        "psllw $2, %%mm0            \n\t"\
1016
        "psllw $2, %%mm1            \n\t"\
1017
        "movq   -1(%0), %%mm2       \n\t"\
1018
        "movq    2(%0), %%mm4       \n\t"\
1019
        "movq %%mm2, %%mm3          \n\t"\
1020
        "movq %%mm4, %%mm5          \n\t"\
1021
        "punpcklbw %%mm7, %%mm2     \n\t"\
1022
        "punpckhbw %%mm7, %%mm3     \n\t"\
1023
        "punpcklbw %%mm7, %%mm4     \n\t"\
1024
        "punpckhbw %%mm7, %%mm5     \n\t"\
1025
        "paddw %%mm4, %%mm2         \n\t"\
1026
        "paddw %%mm3, %%mm5         \n\t"\
1027
        "psubw %%mm2, %%mm0         \n\t"\
1028
        "psubw %%mm5, %%mm1         \n\t"\
1029
        "pmullw %%mm6, %%mm0        \n\t"\
1030
        "pmullw %%mm6, %%mm1        \n\t"\
1031
        "movd   -2(%0), %%mm2       \n\t"\
1032
        "movd    7(%0), %%mm5       \n\t"\
1033
        "punpcklbw %%mm7, %%mm2     \n\t"\
1034
        "punpcklbw %%mm7, %%mm5     \n\t"\
1035
        "paddw %%mm3, %%mm2         \n\t"\
1036
        "paddw %%mm5, %%mm4         \n\t"\
1037
        "movq %5, %%mm5             \n\t"\
1038
        "paddw %%mm5, %%mm2         \n\t"\
1039
        "paddw %%mm5, %%mm4         \n\t"\
1040
        "paddw %%mm2, %%mm0         \n\t"\
1041
        "paddw %%mm4, %%mm1         \n\t"\
1042
        "psraw $5, %%mm0            \n\t"\
1043
        "psraw $5, %%mm1            \n\t"\
1044
        "movq (%2), %%mm4           \n\t"\
1045
        "packuswb %%mm1, %%mm0      \n\t"\
1046
        PAVGB" %%mm4, %%mm0         \n\t"\
1047
        OP(%%mm0, (%1),%%mm5, q)\
1048
        "add %4, %0                 \n\t"\
1049
        "add %4, %1                 \n\t"\
1050
        "add %3, %2                 \n\t"\
1051
        : "+a"(src), "+c"(dst), "+d"(src2)\
1052
        : "D"((long)src2Stride), "S"((long)dstStride),\
1053
          "m"(ff_pw_16)\
1054
        : "memory"\
1055
    );\
1056
    }while(--h);\
1057
}\
1058
\
1059
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1060
    int w= 2;\
1061
    src -= 2*srcStride;\
1062
    \
1063
    while(w--){\
1064
      asm volatile(\
1065
        "pxor %%mm7, %%mm7          \n\t"\
1066
        "movd (%0), %%mm0           \n\t"\
1067
        "add %2, %0                 \n\t"\
1068
        "movd (%0), %%mm1           \n\t"\
1069
        "add %2, %0                 \n\t"\
1070
        "movd (%0), %%mm2           \n\t"\
1071
        "add %2, %0                 \n\t"\
1072
        "movd (%0), %%mm3           \n\t"\
1073
        "add %2, %0                 \n\t"\
1074
        "movd (%0), %%mm4           \n\t"\
1075
        "add %2, %0                 \n\t"\
1076
        "punpcklbw %%mm7, %%mm0     \n\t"\
1077
        "punpcklbw %%mm7, %%mm1     \n\t"\
1078
        "punpcklbw %%mm7, %%mm2     \n\t"\
1079
        "punpcklbw %%mm7, %%mm3     \n\t"\
1080
        "punpcklbw %%mm7, %%mm4     \n\t"\
1081
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1082
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1083
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1084
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1085
        QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1086
        QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1087
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1088
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1089
         \
1090
        : "+a"(src), "+c"(dst)\
1091
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1092
        : "memory"\
1093
     );\
1094
     if(h==16){\
1095
        asm volatile(\
1096
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1097
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1098
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1099
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1100
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1101
            QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1102
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1103
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1104
            \
1105
           : "+a"(src), "+c"(dst)\
1106
           : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1107
           : "memory"\
1108
        );\
1109
     }\
1110
     src += 4-(h+5)*srcStride;\
1111
     dst += 4-h*dstStride;\
1112
   }\
1113
}\
1114
static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
1115
    int w = (size+8)>>2;\
1116
    src -= 2*srcStride+2;\
1117
    while(w--){\
1118
        asm volatile(\
1119
            "pxor %%mm7, %%mm7      \n\t"\
1120
            "movd (%0), %%mm0       \n\t"\
1121
            "add %2, %0             \n\t"\
1122
            "movd (%0), %%mm1       \n\t"\
1123
            "add %2, %0             \n\t"\
1124
            "movd (%0), %%mm2       \n\t"\
1125
            "add %2, %0             \n\t"\
1126
            "movd (%0), %%mm3       \n\t"\
1127
            "add %2, %0             \n\t"\
1128
            "movd (%0), %%mm4       \n\t"\
1129
            "add %2, %0             \n\t"\
1130
            "punpcklbw %%mm7, %%mm0 \n\t"\
1131
            "punpcklbw %%mm7, %%mm1 \n\t"\
1132
            "punpcklbw %%mm7, %%mm2 \n\t"\
1133
            "punpcklbw %%mm7, %%mm3 \n\t"\
1134
            "punpcklbw %%mm7, %%mm4 \n\t"\
1135
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
1136
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
1137
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
1138
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
1139
            QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
1140
            QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
1141
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
1142
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
1143
            : "+a"(src)\
1144
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1145
            : "memory"\
1146
        );\
1147
        if(size==16){\
1148
            asm volatile(\
1149
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
1150
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
1151
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
1152
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
1153
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
1154
                QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
1155
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
1156
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
1157
                : "+a"(src)\
1158
                : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1159
                : "memory"\
1160
            );\
1161
        }\
1162
        tmp += 4;\
1163
        src += 4 - (size+5)*srcStride;\
1164
    }\
1165
}\
1166
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1167
    int w = size>>4;\
1168
    do{\
1169
    int h = size;\
1170
    asm volatile(\
1171
        "1:                         \n\t"\
1172
        "movq     (%0), %%mm0       \n\t"\
1173
        "movq    8(%0), %%mm3       \n\t"\
1174
        "movq    2(%0), %%mm1       \n\t"\
1175
        "movq   10(%0), %%mm4       \n\t"\
1176
        "paddw   %%mm4, %%mm0       \n\t"\
1177
        "paddw   %%mm3, %%mm1       \n\t"\
1178
        "paddw  18(%0), %%mm3       \n\t"\
1179
        "paddw  16(%0), %%mm4       \n\t"\
1180
        "movq    4(%0), %%mm2       \n\t"\
1181
        "movq   12(%0), %%mm5       \n\t"\
1182
        "paddw   6(%0), %%mm2       \n\t"\
1183
        "paddw  14(%0), %%mm5       \n\t"\
1184
        "psubw %%mm1, %%mm0         \n\t"\
1185
        "psubw %%mm4, %%mm3         \n\t"\
1186
        "psraw $2, %%mm0            \n\t"\
1187
        "psraw $2, %%mm3            \n\t"\
1188
        "psubw %%mm1, %%mm0         \n\t"\
1189
        "psubw %%mm4, %%mm3         \n\t"\
1190
        "paddsw %%mm2, %%mm0        \n\t"\
1191
        "paddsw %%mm5, %%mm3        \n\t"\
1192
        "psraw $2, %%mm0            \n\t"\
1193
        "psraw $2, %%mm3            \n\t"\
1194
        "paddw %%mm2, %%mm0         \n\t"\
1195
        "paddw %%mm5, %%mm3         \n\t"\
1196
        "psraw $6, %%mm0            \n\t"\
1197
        "psraw $6, %%mm3            \n\t"\
1198
        "packuswb %%mm3, %%mm0      \n\t"\
1199
        OP(%%mm0, (%1),%%mm7, q)\
1200
        "add $48, %0                \n\t"\
1201
        "add %3, %1                 \n\t"\
1202
        "decl %2                    \n\t"\
1203
        " jnz 1b                    \n\t"\
1204
        : "+a"(tmp), "+c"(dst), "+g"(h)\
1205
        : "S"((long)dstStride)\
1206
        : "memory"\
1207
    );\
1208
    tmp += 8 - size*24;\
1209
    dst += 8 - size*dstStride;\
1210
    }while(w--);\
1211
}\
1212
\
1213
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1214
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1215
}\
1216
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1217
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1218
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1219
}\
1220
\
1221
static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1222
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1223
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1224
    src += 8*srcStride;\
1225
    dst += 8*dstStride;\
1226
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1227
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1228
}\
1229
\
1230
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1231
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1232
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1233
    src += 8*dstStride;\
1234
    dst += 8*dstStride;\
1235
    src2 += 8*src2Stride;\
1236
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1237
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1238
}\
1239
\
1240
static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
1241
          put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
1242
    OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1243
}\
1244
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1245
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
1246
}\
1247
\
1248
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1249
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
1250
}\
1251
\
1252
static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1253
{\
1254
    asm volatile(\
1255
        "movq      (%1), %%mm0          \n\t"\
1256
        "movq    24(%1), %%mm1          \n\t"\
1257
        "psraw      $5,  %%mm0          \n\t"\
1258
        "psraw      $5,  %%mm1          \n\t"\
1259
        "packuswb %%mm0, %%mm0          \n\t"\
1260
        "packuswb %%mm1, %%mm1          \n\t"\
1261
        PAVGB"     (%0), %%mm0          \n\t"\
1262
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1263
        OP(%%mm0, (%2),    %%mm4, d)\
1264
        OP(%%mm1, (%2,%4), %%mm5, d)\
1265
        "lea  (%0,%3,2), %0             \n\t"\
1266
        "lea  (%2,%4,2), %2             \n\t"\
1267
        "movq    48(%1), %%mm0          \n\t"\
1268
        "movq    72(%1), %%mm1          \n\t"\
1269
        "psraw      $5,  %%mm0          \n\t"\
1270
        "psraw      $5,  %%mm1          \n\t"\
1271
        "packuswb %%mm0, %%mm0          \n\t"\
1272
        "packuswb %%mm1, %%mm1          \n\t"\
1273
        PAVGB"     (%0), %%mm0          \n\t"\
1274
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1275
        OP(%%mm0, (%2),    %%mm4, d)\
1276
        OP(%%mm1, (%2,%4), %%mm5, d)\
1277
        :"+a"(src8), "+c"(src16), "+d"(dst)\
1278
        :"S"((long)src8Stride), "D"((long)dstStride)\
1279
        :"memory");\
1280
}\
1281
static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1282
{\
1283
    do{\
1284
    asm volatile(\
1285
        "movq      (%1), %%mm0          \n\t"\
1286
        "movq     8(%1), %%mm1          \n\t"\
1287
        "movq    48(%1), %%mm2          \n\t"\
1288
        "movq  8+48(%1), %%mm3          \n\t"\
1289
        "psraw      $5,  %%mm0          \n\t"\
1290
        "psraw      $5,  %%mm1          \n\t"\
1291
        "psraw      $5,  %%mm2          \n\t"\
1292
        "psraw      $5,  %%mm3          \n\t"\
1293
        "packuswb %%mm1, %%mm0          \n\t"\
1294
        "packuswb %%mm3, %%mm2          \n\t"\
1295
        PAVGB"     (%0), %%mm0          \n\t"\
1296
        PAVGB"  (%0,%3), %%mm2          \n\t"\
1297
        OP(%%mm0, (%2), %%mm5, q)\
1298
        OP(%%mm2, (%2,%4), %%mm5, q)\
1299
        ::"a"(src8), "c"(src16), "d"(dst),\
1300
          "r"((long)src8Stride), "r"((long)dstStride)\
1301
        :"memory");\
1302
        src8 += 2L*src8Stride;\
1303
        src16 += 48;\
1304
        dst += 2L*dstStride;\
1305
    }while(h-=2);\
1306
}\
1307
static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1308
{\
1309
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
1310
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1311
}\
1312

    
1313

    
1314
#ifdef ARCH_X86_64
1315
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1316
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1317
    int h=16;\
1318
    asm volatile(\
1319
        "pxor %%xmm15, %%xmm15      \n\t"\
1320
        "movdqa %6, %%xmm14         \n\t"\
1321
        "movdqa %7, %%xmm13         \n\t"\
1322
        "1:                         \n\t"\
1323
        "lddqu    3(%0), %%xmm1     \n\t"\
1324
        "lddqu   -5(%0), %%xmm7     \n\t"\
1325
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1326
        "punpckhbw %%xmm15, %%xmm1  \n\t"\
1327
        "punpcklbw %%xmm15, %%xmm0  \n\t"\
1328
        "punpcklbw %%xmm15, %%xmm7  \n\t"\
1329
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1330
        "movdqa  %%xmm0, %%xmm6     \n\t"\
1331
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1332
        "movdqa  %%xmm0, %%xmm8     \n\t"\
1333
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1334
        "movdqa  %%xmm0, %%xmm9     \n\t"\
1335
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1336
        "movdqa  %%xmm0, %%xmm10    \n\t"\
1337
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1338
        "palignr $6, %%xmm7, %%xmm10\n\t"\
1339
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1340
        "palignr $8, %%xmm7, %%xmm9 \n\t"\
1341
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1342
        "palignr $10,%%xmm7, %%xmm8 \n\t"\
1343
        "paddw   %%xmm1, %%xmm5     \n\t"\
1344
        "paddw   %%xmm0, %%xmm10    \n\t"\
1345
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1346
        "palignr $12,%%xmm7, %%xmm6 \n\t"\
1347
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1348
        "palignr $14,%%xmm7, %%xmm0 \n\t"\
1349
        "paddw   %%xmm3, %%xmm2     \n\t"\
1350
        "paddw   %%xmm8, %%xmm6     \n\t"\
1351
        "paddw   %%xmm4, %%xmm1     \n\t"\
1352
        "paddw   %%xmm9, %%xmm0     \n\t"\
1353
        "psllw   $2,     %%xmm2     \n\t"\
1354
        "psllw   $2,     %%xmm6     \n\t"\
1355
        "psubw   %%xmm1, %%xmm2     \n\t"\
1356
        "psubw   %%xmm0, %%xmm6     \n\t"\
1357
        "paddw   %%xmm13,%%xmm5     \n\t"\
1358
        "paddw   %%xmm13,%%xmm10    \n\t"\
1359
        "pmullw  %%xmm14,%%xmm2     \n\t"\
1360
        "pmullw  %%xmm14,%%xmm6     \n\t"\
1361
        "lddqu   (%2),   %%xmm3     \n\t"\
1362
        "paddw   %%xmm5, %%xmm2     \n\t"\
1363
        "paddw   %%xmm10,%%xmm6     \n\t"\
1364
        "psraw   $5,     %%xmm2     \n\t"\
1365
        "psraw   $5,     %%xmm6     \n\t"\
1366
        "packuswb %%xmm2,%%xmm6     \n\t"\
1367
        "pavgb   %%xmm3, %%xmm6     \n\t"\
1368
        OP(%%xmm6, (%1), %%xmm4, dqa)\
1369
        "add %5, %0                 \n\t"\
1370
        "add %5, %1                 \n\t"\
1371
        "add %4, %2                 \n\t"\
1372
        "decl %3                    \n\t"\
1373
        "jg 1b                      \n\t"\
1374
        : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
1375
        : "D"((long)src2Stride), "S"((long)dstStride),\
1376
          "m"(ff_pw_5), "m"(ff_pw_16)\
1377
        : "memory"\
1378
    );\
1379
}
1380
#else // ARCH_X86_64
1381
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1382
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1383
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1384
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1385
    src += 8*dstStride;\
1386
    dst += 8*dstStride;\
1387
    src2 += 8*src2Stride;\
1388
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1389
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1390
}
1391
#endif // ARCH_X86_64
1392

    
1393
#define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
1394
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1395
    int h=8;\
1396
    asm volatile(\
1397
        "pxor %%xmm7, %%xmm7        \n\t"\
1398
        "movdqa %0, %%xmm6          \n\t"\
1399
        :: "m"(ff_pw_5)\
1400
    );\
1401
    do{\
1402
    asm volatile(\
1403
        "lddqu   -5(%0), %%xmm1     \n\t"\
1404
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1405
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
1406
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1407
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1408
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1409
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1410
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1411
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1412
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1413
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1414
        "paddw   %%xmm1, %%xmm5     \n\t"\
1415
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1416
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1417
        "paddw   %%xmm3, %%xmm2     \n\t"\
1418
        "paddw   %%xmm4, %%xmm1     \n\t"\
1419
        "psllw   $2,     %%xmm2     \n\t"\
1420
        "movq    (%2),   %%xmm3     \n\t"\
1421
        "psubw   %%xmm1, %%xmm2     \n\t"\
1422
        "paddw   %5,     %%xmm5     \n\t"\
1423
        "pmullw  %%xmm6, %%xmm2     \n\t"\
1424
        "paddw   %%xmm5, %%xmm2     \n\t"\
1425
        "psraw   $5,     %%xmm2     \n\t"\
1426
        "packuswb %%xmm2, %%xmm2    \n\t"\
1427
        "pavgb   %%xmm3, %%xmm2     \n\t"\
1428
        OP(%%xmm2, (%1), %%xmm4, q)\
1429
        "add %4, %0                 \n\t"\
1430
        "add %4, %1                 \n\t"\
1431
        "add %3, %2                 \n\t"\
1432
        : "+a"(src), "+c"(dst), "+d"(src2)\
1433
        : "D"((long)src2Stride), "S"((long)dstStride),\
1434
          "m"(ff_pw_16)\
1435
        : "memory"\
1436
    );\
1437
    }while(--h);\
1438
}\
1439
QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1440
\
1441
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1442
    int h=8;\
1443
    asm volatile(\
1444
        "pxor %%xmm7, %%xmm7        \n\t"\
1445
        "movdqa %5, %%xmm6          \n\t"\
1446
        "1:                         \n\t"\
1447
        "lddqu   -5(%0), %%xmm1     \n\t"\
1448
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1449
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
1450
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1451
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1452
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1453
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1454
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1455
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1456
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1457
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1458
        "paddw   %%xmm1, %%xmm5     \n\t"\
1459
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1460
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1461
        "paddw   %%xmm3, %%xmm2     \n\t"\
1462
        "paddw   %%xmm4, %%xmm1     \n\t"\
1463
        "psllw   $2,     %%xmm2     \n\t"\
1464
        "psubw   %%xmm1, %%xmm2     \n\t"\
1465
        "paddw   %6,     %%xmm5     \n\t"\
1466
        "pmullw  %%xmm6, %%xmm2     \n\t"\
1467
        "paddw   %%xmm5, %%xmm2     \n\t"\
1468
        "psraw   $5,     %%xmm2     \n\t"\
1469
        "packuswb %%xmm2, %%xmm2    \n\t"\
1470
        OP(%%xmm2, (%1), %%xmm4, q)\
1471
        "add %3, %0                 \n\t"\
1472
        "add %4, %1                 \n\t"\
1473
        "decl %2                    \n\t"\
1474
        " jnz 1b                    \n\t"\
1475
        : "+a"(src), "+c"(dst), "+g"(h)\
1476
        : "D"((long)srcStride), "S"((long)dstStride),\
1477
          "m"(ff_pw_5), "m"(ff_pw_16)\
1478
        : "memory"\
1479
    );\
1480
}\
1481
static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1482
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1483
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1484
    src += 8*srcStride;\
1485
    dst += 8*dstStride;\
1486
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1487
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1488
}\
1489

    
1490
#define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
1491
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1492
    src -= 2*srcStride;\
1493
    \
1494
    asm volatile(\
1495
        "pxor %%xmm7, %%xmm7        \n\t"\
1496
        "movq (%0), %%xmm0          \n\t"\
1497
        "add %2, %0                 \n\t"\
1498
        "movq (%0), %%xmm1          \n\t"\
1499
        "add %2, %0                 \n\t"\
1500
        "movq (%0), %%xmm2          \n\t"\
1501
        "add %2, %0                 \n\t"\
1502
        "movq (%0), %%xmm3          \n\t"\
1503
        "add %2, %0                 \n\t"\
1504
        "movq (%0), %%xmm4          \n\t"\
1505
        "add %2, %0                 \n\t"\
1506
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1507
        "punpcklbw %%xmm7, %%xmm1   \n\t"\
1508
        "punpcklbw %%xmm7, %%xmm2   \n\t"\
1509
        "punpcklbw %%xmm7, %%xmm3   \n\t"\
1510
        "punpcklbw %%xmm7, %%xmm4   \n\t"\
1511
        QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1512
        QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1513
        QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1514
        QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1515
        QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1516
        QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1517
        QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1518
        QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1519
         \
1520
        : "+a"(src), "+c"(dst)\
1521
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1522
        : "memory"\
1523
    );\
1524
    if(h==16){\
1525
        asm volatile(\
1526
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1527
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1528
            QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1529
            QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1530
            QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1531
            QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1532
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1533
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1534
            \
1535
            : "+a"(src), "+c"(dst)\
1536
            : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1537
            : "memory"\
1538
        );\
1539
    }\
1540
}\
1541
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1542
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1543
}\
1544
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1545
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1546
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1547
}
1548

    
1549
static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
1550
    int w = (size+8)>>3;
1551
    src -= 2*srcStride+2;
1552
    while(w--){
1553
        asm volatile(
1554
            "pxor %%xmm7, %%xmm7        \n\t"
1555
            "movq (%0), %%xmm0          \n\t"
1556
            "add %2, %0                 \n\t"
1557
            "movq (%0), %%xmm1          \n\t"
1558
            "add %2, %0                 \n\t"
1559
            "movq (%0), %%xmm2          \n\t"
1560
            "add %2, %0                 \n\t"
1561
            "movq (%0), %%xmm3          \n\t"
1562
            "add %2, %0                 \n\t"
1563
            "movq (%0), %%xmm4          \n\t"
1564
            "add %2, %0                 \n\t"
1565
            "punpcklbw %%xmm7, %%xmm0   \n\t"
1566
            "punpcklbw %%xmm7, %%xmm1   \n\t"
1567
            "punpcklbw %%xmm7, %%xmm2   \n\t"
1568
            "punpcklbw %%xmm7, %%xmm3   \n\t"
1569
            "punpcklbw %%xmm7, %%xmm4   \n\t"
1570
            QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
1571
            QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
1572
            QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
1573
            QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
1574
            QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
1575
            QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
1576
            QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
1577
            QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
1578
            : "+a"(src)
1579
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1580
            : "memory"
1581
        );
1582
        if(size==16){
1583
            asm volatile(
1584
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
1585
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
1586
                QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
1587
                QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
1588
                QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
1589
                QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
1590
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
1591
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
1592
                : "+a"(src)
1593
                : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1594
                : "memory"
1595
            );
1596
        }
1597
        tmp += 8;
1598
        src += 8 - (size+5)*srcStride;
1599
    }
1600
}
1601

    
1602
#define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
1603
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1604
    int h = size;\
1605
    if(size == 16){\
1606
        asm volatile(\
1607
            "1:                         \n\t"\
1608
            "movdqa 32(%0), %%xmm4      \n\t"\
1609
            "movdqa 16(%0), %%xmm5      \n\t"\
1610
            "movdqa   (%0), %%xmm7      \n\t"\
1611
            "movdqa %%xmm4, %%xmm3      \n\t"\
1612
            "movdqa %%xmm4, %%xmm2      \n\t"\
1613
            "movdqa %%xmm4, %%xmm1      \n\t"\
1614
            "movdqa %%xmm4, %%xmm0      \n\t"\
1615
            "palignr $10, %%xmm5, %%xmm0 \n\t"\
1616
            "palignr  $8, %%xmm5, %%xmm1 \n\t"\
1617
            "palignr  $6, %%xmm5, %%xmm2 \n\t"\
1618
            "palignr  $4, %%xmm5, %%xmm3 \n\t"\
1619
            "palignr  $2, %%xmm5, %%xmm4 \n\t"\
1620
            "paddw  %%xmm5, %%xmm0      \n\t"\
1621
            "paddw  %%xmm4, %%xmm1      \n\t"\
1622
            "paddw  %%xmm3, %%xmm2      \n\t"\
1623
            "movdqa %%xmm5, %%xmm6      \n\t"\
1624
            "movdqa %%xmm5, %%xmm4      \n\t"\
1625
            "movdqa %%xmm5, %%xmm3      \n\t"\
1626
            "palignr  $8, %%xmm7, %%xmm4 \n\t"\
1627
            "palignr  $2, %%xmm7, %%xmm6 \n\t"\
1628
            "palignr $10, %%xmm7, %%xmm3 \n\t"\
1629
            "paddw  %%xmm6, %%xmm4      \n\t"\
1630
            "movdqa %%xmm5, %%xmm6      \n\t"\
1631
            "palignr  $6, %%xmm7, %%xmm5 \n\t"\
1632
            "palignr  $4, %%xmm7, %%xmm6 \n\t"\
1633
            "paddw  %%xmm7, %%xmm3      \n\t"\
1634
            "paddw  %%xmm6, %%xmm5      \n\t"\
1635
            \
1636
            "psubw  %%xmm1, %%xmm0      \n\t"\
1637
            "psubw  %%xmm4, %%xmm3      \n\t"\
1638
            "psraw      $2, %%xmm0      \n\t"\
1639
            "psraw      $2, %%xmm3      \n\t"\
1640
            "psubw  %%xmm1, %%xmm0      \n\t"\
1641
            "psubw  %%xmm4, %%xmm3      \n\t"\
1642
            "paddw  %%xmm2, %%xmm0      \n\t"\
1643
            "paddw  %%xmm5, %%xmm3      \n\t"\
1644
            "psraw      $2, %%xmm0      \n\t"\
1645
            "psraw      $2, %%xmm3      \n\t"\
1646
            "paddw  %%xmm2, %%xmm0      \n\t"\
1647
            "paddw  %%xmm5, %%xmm3      \n\t"\
1648
            "psraw      $6, %%xmm0      \n\t"\
1649
            "psraw      $6, %%xmm3      \n\t"\
1650
            "packuswb %%xmm0, %%xmm3    \n\t"\
1651
            OP(%%xmm3, (%1), %%xmm7, dqa)\
1652
            "add $48, %0                \n\t"\
1653
            "add %3, %1                 \n\t"\
1654
            "decl %2                    \n\t"\
1655
            " jnz 1b                    \n\t"\
1656
            : "+a"(tmp), "+c"(dst), "+g"(h)\
1657
            : "S"((long)dstStride)\
1658
            : "memory"\
1659
        );\
1660
    }else{\
1661
        asm volatile(\
1662
            "1:                         \n\t"\
1663
            "movdqa 16(%0), %%xmm1      \n\t"\
1664
            "movdqa   (%0), %%xmm0      \n\t"\
1665
            "movdqa %%xmm1, %%xmm2      \n\t"\
1666
            "movdqa %%xmm1, %%xmm3      \n\t"\
1667
            "movdqa %%xmm1, %%xmm4      \n\t"\
1668
            "movdqa %%xmm1, %%xmm5      \n\t"\
1669
            "palignr $10, %%xmm0, %%xmm5 \n\t"\
1670
            "palignr  $8, %%xmm0, %%xmm4 \n\t"\
1671
            "palignr  $6, %%xmm0, %%xmm3 \n\t"\
1672
            "palignr  $4, %%xmm0, %%xmm2 \n\t"\
1673
            "palignr  $2, %%xmm0, %%xmm1 \n\t"\
1674
            "paddw  %%xmm5, %%xmm0      \n\t"\
1675
            "paddw  %%xmm4, %%xmm1      \n\t"\
1676
            "paddw  %%xmm3, %%xmm2      \n\t"\
1677
            "psubw  %%xmm1, %%xmm0      \n\t"\
1678
            "psraw      $2, %%xmm0      \n\t"\
1679
            "psubw  %%xmm1, %%xmm0      \n\t"\
1680
            "paddw  %%xmm2, %%xmm0      \n\t"\
1681
            "psraw      $2, %%xmm0      \n\t"\
1682
            "paddw  %%xmm2, %%xmm0      \n\t"\
1683
            "psraw      $6, %%xmm0      \n\t"\
1684
            "packuswb %%xmm0, %%xmm0    \n\t"\
1685
            OP(%%xmm0, (%1), %%xmm7, q)\
1686
            "add $48, %0                \n\t"\
1687
            "add %3, %1                 \n\t"\
1688
            "decl %2                    \n\t"\
1689
            " jnz 1b                    \n\t"\
1690
            : "+a"(tmp), "+c"(dst), "+g"(h)\
1691
            : "S"((long)dstStride)\
1692
            : "memory"\
1693
        );\
1694
    }\
1695
}
1696

    
1697
#define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
1698
static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
1699
          put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
1700
    OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1701
}\
1702
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1703
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
1704
}\
1705
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1706
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1707
}\
1708

    
1709
#define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1710
#define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1711
#define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1712
#define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1713
#define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1714
#define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1715
#define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1716
#define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1717

    
1718
#define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1719
#define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1720
#define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1721
#define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1722
#define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1723
#define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1724
#define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1725
#define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1726

    
1727
#define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1728
#define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1729
#define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1730
#define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1731

    
1732
#define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1733
#define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1734
#define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1735
#define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1736

    
1737
#define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1738
#define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1739

    
1740
#define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1741
H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1742
H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1743
H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1744
H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1745

    
1746
static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1747
    put_pixels16_sse2(dst, src, stride, 16);
1748
}
1749
static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1750
    avg_pixels16_sse2(dst, src, stride, 16);
1751
}
1752
#define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1753
#define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1754

    
1755
#define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1756
static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1757
    OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1758
}\
1759

    
1760
#define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1761
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1762
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1763
}\
1764
\
1765
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1766
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1767
}\
1768
\
1769
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1770
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1771
}\
1772

    
1773
#define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1774
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1775
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1776
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1777
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1778
}\
1779
\
1780
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1781
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1782
}\
1783
\
1784
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1785
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1786
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1787
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1788
}\
1789

    
1790
#define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1791
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1792
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1793
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1794
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1795
}\
1796
\
1797
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1798
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1799
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1800
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1801
}\
1802
\
1803
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1804
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1805
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1806
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1807
}\
1808
\
1809
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1810
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1811
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1812
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1813
}\
1814
\
1815
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1816
    DECLARE_ALIGNED(ALIGN, uint16_t, temp[SIZE*(SIZE<8?12:24)]);\
1817
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
1818
}\
1819
\
1820
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1821
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1822
    uint8_t * const halfHV= temp;\
1823
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1824
    assert(((int)temp & 7) == 0);\
1825
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1826
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1827
}\
1828
\
1829
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1830
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1831
    uint8_t * const halfHV= temp;\
1832
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1833
    assert(((int)temp & 7) == 0);\
1834
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1835
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1836
}\
1837
\
1838
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1839
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1840
    uint8_t * const halfHV= temp;\
1841
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1842
    assert(((int)temp & 7) == 0);\
1843
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1844
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1845
}\
1846
\
1847
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1848
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1849
    uint8_t * const halfHV= temp;\
1850
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1851
    assert(((int)temp & 7) == 0);\
1852
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1853
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1854
}\
1855

    
1856
#define H264_MC_4816(MMX)\
1857
H264_MC(put_, 4, MMX, 8)\
1858
H264_MC(put_, 8, MMX, 8)\
1859
H264_MC(put_, 16,MMX, 8)\
1860
H264_MC(avg_, 4, MMX, 8)\
1861
H264_MC(avg_, 8, MMX, 8)\
1862
H264_MC(avg_, 16,MMX, 8)\
1863

    
1864
#define H264_MC_816(QPEL, XMM)\
1865
QPEL(put_, 8, XMM, 16)\
1866
QPEL(put_, 16,XMM, 16)\
1867
QPEL(avg_, 8, XMM, 16)\
1868
QPEL(avg_, 16,XMM, 16)\
1869

    
1870

    
1871
#define AVG_3DNOW_OP(a,b,temp, size) \
1872
"mov" #size " " #b ", " #temp "   \n\t"\
1873
"pavgusb " #temp ", " #a "        \n\t"\
1874
"mov" #size " " #a ", " #b "      \n\t"
1875
#define AVG_MMX2_OP(a,b,temp, size) \
1876
"mov" #size " " #b ", " #temp "   \n\t"\
1877
"pavgb " #temp ", " #a "          \n\t"\
1878
"mov" #size " " #a ", " #b "      \n\t"
1879

    
1880
#define PAVGB "pavgusb"
1881
QPEL_H264(put_,       PUT_OP, 3dnow)
1882
QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1883
#undef PAVGB
1884
#define PAVGB "pavgb"
1885
QPEL_H264(put_,       PUT_OP, mmx2)
1886
QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
1887
QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
1888
QPEL_H264_V_XMM(avg_,  AVG_MMX2_OP, sse2)
1889
QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
1890
QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, sse2)
1891
#ifdef HAVE_SSSE3
1892
QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
1893
QPEL_H264_H_XMM(avg_,  AVG_MMX2_OP, ssse3)
1894
QPEL_H264_HV2_XMM(put_,       PUT_OP, ssse3)
1895
QPEL_H264_HV2_XMM(avg_,  AVG_MMX2_OP, ssse3)
1896
QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
1897
QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, ssse3)
1898
#endif
1899
#undef PAVGB
1900

    
1901
H264_MC_4816(3dnow)
1902
H264_MC_4816(mmx2)
1903
H264_MC_816(H264_MC_V, sse2)
1904
H264_MC_816(H264_MC_HV, sse2)
1905
#ifdef HAVE_SSSE3
1906
H264_MC_816(H264_MC_H, ssse3)
1907
H264_MC_816(H264_MC_HV, ssse3)
1908
#endif
1909

    
1910

    
1911
#define H264_CHROMA_OP(S,D)
1912
#define H264_CHROMA_OP4(S,D,T)
1913
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
1914
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
1915
#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
1916
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1917
#include "dsputil_h264_template_mmx.c"
1918

    
1919
static void put_h264_chroma_mc8_mmx_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
1920
{
1921
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 1);
1922
}
1923
static void put_h264_chroma_mc8_mmx_nornd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
1924
{
1925
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 0);
1926
}
1927

    
1928
#undef H264_CHROMA_OP
1929
#undef H264_CHROMA_OP4
1930
#undef H264_CHROMA_MC8_TMPL
1931
#undef H264_CHROMA_MC4_TMPL
1932
#undef H264_CHROMA_MC2_TMPL
1933
#undef H264_CHROMA_MC8_MV0
1934

    
1935
#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
1936
#define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\
1937
                               "pavgb " #T ", " #D " \n\t"
1938
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
1939
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
1940
#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
1941
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
1942
#include "dsputil_h264_template_mmx.c"
1943
static void avg_h264_chroma_mc8_mmx2_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
1944
{
1945
    avg_h264_chroma_mc8_mmx2(dst, src, stride, h, x, y, 1);
1946
}
1947
#undef H264_CHROMA_OP
1948
#undef H264_CHROMA_OP4
1949
#undef H264_CHROMA_MC8_TMPL
1950
#undef H264_CHROMA_MC4_TMPL
1951
#undef H264_CHROMA_MC2_TMPL
1952
#undef H264_CHROMA_MC8_MV0
1953

    
1954
#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
1955
#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
1956
                               "pavgusb " #T ", " #D " \n\t"
1957
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
1958
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
1959
#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
1960
#include "dsputil_h264_template_mmx.c"
1961
static void avg_h264_chroma_mc8_3dnow_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
1962
{
1963
    avg_h264_chroma_mc8_3dnow(dst, src, stride, h, x, y, 1);
1964
}
1965
#undef H264_CHROMA_OP
1966
#undef H264_CHROMA_OP4
1967
#undef H264_CHROMA_MC8_TMPL
1968
#undef H264_CHROMA_MC4_TMPL
1969
#undef H264_CHROMA_MC8_MV0
1970

    
1971
/***********************************/
1972
/* weighted prediction */
1973

    
1974
static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
1975
{
1976
    int x, y;
1977
    offset <<= log2_denom;
1978
    offset += (1 << log2_denom) >> 1;
1979
    asm volatile(
1980
        "movd    %0, %%mm4        \n\t"
1981
        "movd    %1, %%mm5        \n\t"
1982
        "movd    %2, %%mm6        \n\t"
1983
        "pshufw  $0, %%mm4, %%mm4 \n\t"
1984
        "pshufw  $0, %%mm5, %%mm5 \n\t"
1985
        "pxor    %%mm7, %%mm7     \n\t"
1986
        :: "g"(weight), "g"(offset), "g"(log2_denom)
1987
    );
1988
    for(y=0; y<h; y+=2){
1989
        for(x=0; x<w; x+=4){
1990
            asm volatile(
1991
                "movd      %0,    %%mm0 \n\t"
1992
                "movd      %1,    %%mm1 \n\t"
1993
                "punpcklbw %%mm7, %%mm0 \n\t"
1994
                "punpcklbw %%mm7, %%mm1 \n\t"
1995
                "pmullw    %%mm4, %%mm0 \n\t"
1996
                "pmullw    %%mm4, %%mm1 \n\t"
1997
                "paddsw    %%mm5, %%mm0 \n\t"
1998
                "paddsw    %%mm5, %%mm1 \n\t"
1999
                "psraw     %%mm6, %%mm0 \n\t"
2000
                "psraw     %%mm6, %%mm1 \n\t"
2001
                "packuswb  %%mm7, %%mm0 \n\t"
2002
                "packuswb  %%mm7, %%mm1 \n\t"
2003
                "movd      %%mm0, %0    \n\t"
2004
                "movd      %%mm1, %1    \n\t"
2005
                : "+m"(*(uint32_t*)(dst+x)),
2006
                  "+m"(*(uint32_t*)(dst+x+stride))
2007
            );
2008
        }
2009
        dst += 2*stride;
2010
    }
2011
}
2012

    
2013
static inline void ff_h264_biweight_WxH_mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset, int w, int h)
2014
{
2015
    int x, y;
2016
    offset = ((offset + 1) | 1) << log2_denom;
2017
    asm volatile(
2018
        "movd    %0, %%mm3        \n\t"
2019
        "movd    %1, %%mm4        \n\t"
2020
        "movd    %2, %%mm5        \n\t"
2021
        "movd    %3, %%mm6        \n\t"
2022
        "pshufw  $0, %%mm3, %%mm3 \n\t"
2023
        "pshufw  $0, %%mm4, %%mm4 \n\t"
2024
        "pshufw  $0, %%mm5, %%mm5 \n\t"
2025
        "pxor    %%mm7, %%mm7     \n\t"
2026
        :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
2027
    );
2028
    for(y=0; y<h; y++){
2029
        for(x=0; x<w; x+=4){
2030
            asm volatile(
2031
                "movd      %0,    %%mm0 \n\t"
2032
                "movd      %1,    %%mm1 \n\t"
2033
                "punpcklbw %%mm7, %%mm0 \n\t"
2034
                "punpcklbw %%mm7, %%mm1 \n\t"
2035
                "pmullw    %%mm3, %%mm0 \n\t"
2036
                "pmullw    %%mm4, %%mm1 \n\t"
2037
                "paddsw    %%mm1, %%mm0 \n\t"
2038
                "paddsw    %%mm5, %%mm0 \n\t"
2039
                "psraw     %%mm6, %%mm0 \n\t"
2040
                "packuswb  %%mm0, %%mm0 \n\t"
2041
                "movd      %%mm0, %0    \n\t"
2042
                : "+m"(*(uint32_t*)(dst+x))
2043
                :  "m"(*(uint32_t*)(src+x))
2044
            );
2045
        }
2046
        src += stride;
2047
        dst += stride;
2048
    }
2049
}
2050

    
2051
#define H264_WEIGHT(W,H) \
2052
static void ff_h264_biweight_ ## W ## x ## H ## _mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
2053
    ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
2054
} \
2055
static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
2056
    ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
2057
}
2058

    
2059
H264_WEIGHT(16,16)
2060
H264_WEIGHT(16, 8)
2061
H264_WEIGHT( 8,16)
2062
H264_WEIGHT( 8, 8)
2063
H264_WEIGHT( 8, 4)
2064
H264_WEIGHT( 4, 8)
2065
H264_WEIGHT( 4, 4)
2066
H264_WEIGHT( 4, 2)
2067