Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / h264dsp_mmx.c @ c4ff7c53

History | View | Annotate | Download (88.2 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
DECLARE_ALIGNED_8 (static const uint64_t, ff_pb_3_1  ) = 0x0103010301030103ULL;
24
DECLARE_ALIGNED_8 (static const uint64_t, ff_pb_7_3  ) = 0x0307030703070307ULL;
25

    
26
/***********************************/
27
/* IDCT */
28

    
29
#define SUMSUB_BADC( a, b, c, d ) \
30
    "paddw "#b", "#a" \n\t"\
31
    "paddw "#d", "#c" \n\t"\
32
    "paddw "#b", "#b" \n\t"\
33
    "paddw "#d", "#d" \n\t"\
34
    "psubw "#a", "#b" \n\t"\
35
    "psubw "#c", "#d" \n\t"
36

    
37
#define SUMSUBD2_AB( a, b, t ) \
38
    "movq  "#b", "#t" \n\t"\
39
    "psraw  $1 , "#b" \n\t"\
40
    "paddw "#a", "#b" \n\t"\
41
    "psraw  $1 , "#a" \n\t"\
42
    "psubw "#t", "#a" \n\t"
43

    
44
#define IDCT4_1D( s02, s13, d02, d13, t ) \
45
    SUMSUB_BA  ( s02, d02 )\
46
    SUMSUBD2_AB( s13, d13, t )\
47
    SUMSUB_BADC( d13, s02, s13, d02 )
48

    
49
#define STORE_DIFF_4P( p, t, z ) \
50
    "psraw      $6,     "#p" \n\t"\
51
    "movd       (%0),   "#t" \n\t"\
52
    "punpcklbw "#z",    "#t" \n\t"\
53
    "paddsw    "#t",    "#p" \n\t"\
54
    "packuswb  "#z",    "#p" \n\t"\
55
    "movd      "#p",    (%0) \n\t"
56

    
57
static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
58
{
59
    /* Load dct coeffs */
60
    __asm__ volatile(
61
        "movq   (%0), %%mm0 \n\t"
62
        "movq  8(%0), %%mm1 \n\t"
63
        "movq 16(%0), %%mm2 \n\t"
64
        "movq 24(%0), %%mm3 \n\t"
65
    :: "r"(block) );
66

    
67
    __asm__ volatile(
68
        /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
69
        IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
70

    
71
        "movq      %0,    %%mm6 \n\t"
72
        /* in: 1,4,0,2  out: 1,2,3,0 */
73
        TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
74

    
75
        "paddw     %%mm6, %%mm3 \n\t"
76

    
77
        /* mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13 */
78
        IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
79

    
80
        "pxor %%mm7, %%mm7    \n\t"
81
    :: "m"(ff_pw_32));
82

    
83
    __asm__ volatile(
84
    STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
85
        "add %1, %0             \n\t"
86
    STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
87
        "add %1, %0             \n\t"
88
    STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
89
        "add %1, %0             \n\t"
90
    STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
91
        : "+r"(dst)
92
        : "r" ((x86_reg)stride)
93
    );
94
}
95

    
96
static inline void h264_idct8_1d(int16_t *block)
97
{
98
    __asm__ volatile(
99
        "movq 112(%0), %%mm7  \n\t"
100
        "movq  80(%0), %%mm0  \n\t"
101
        "movq  48(%0), %%mm3  \n\t"
102
        "movq  16(%0), %%mm5  \n\t"
103

    
104
        "movq   %%mm0, %%mm4  \n\t"
105
        "movq   %%mm5, %%mm1  \n\t"
106
        "psraw  $1,    %%mm4  \n\t"
107
        "psraw  $1,    %%mm1  \n\t"
108
        "paddw  %%mm0, %%mm4  \n\t"
109
        "paddw  %%mm5, %%mm1  \n\t"
110
        "paddw  %%mm7, %%mm4  \n\t"
111
        "paddw  %%mm0, %%mm1  \n\t"
112
        "psubw  %%mm5, %%mm4  \n\t"
113
        "paddw  %%mm3, %%mm1  \n\t"
114

    
115
        "psubw  %%mm3, %%mm5  \n\t"
116
        "psubw  %%mm3, %%mm0  \n\t"
117
        "paddw  %%mm7, %%mm5  \n\t"
118
        "psubw  %%mm7, %%mm0  \n\t"
119
        "psraw  $1,    %%mm3  \n\t"
120
        "psraw  $1,    %%mm7  \n\t"
121
        "psubw  %%mm3, %%mm5  \n\t"
122
        "psubw  %%mm7, %%mm0  \n\t"
123

    
124
        "movq   %%mm4, %%mm3  \n\t"
125
        "movq   %%mm1, %%mm7  \n\t"
126
        "psraw  $2,    %%mm1  \n\t"
127
        "psraw  $2,    %%mm3  \n\t"
128
        "paddw  %%mm5, %%mm3  \n\t"
129
        "psraw  $2,    %%mm5  \n\t"
130
        "paddw  %%mm0, %%mm1  \n\t"
131
        "psraw  $2,    %%mm0  \n\t"
132
        "psubw  %%mm4, %%mm5  \n\t"
133
        "psubw  %%mm0, %%mm7  \n\t"
134

    
135
        "movq  32(%0), %%mm2  \n\t"
136
        "movq  96(%0), %%mm6  \n\t"
137
        "movq   %%mm2, %%mm4  \n\t"
138
        "movq   %%mm6, %%mm0  \n\t"
139
        "psraw  $1,    %%mm4  \n\t"
140
        "psraw  $1,    %%mm6  \n\t"
141
        "psubw  %%mm0, %%mm4  \n\t"
142
        "paddw  %%mm2, %%mm6  \n\t"
143

    
144
        "movq    (%0), %%mm2  \n\t"
145
        "movq  64(%0), %%mm0  \n\t"
146
        SUMSUB_BA( %%mm0, %%mm2 )
147
        SUMSUB_BA( %%mm6, %%mm0 )
148
        SUMSUB_BA( %%mm4, %%mm2 )
149
        SUMSUB_BA( %%mm7, %%mm6 )
150
        SUMSUB_BA( %%mm5, %%mm4 )
151
        SUMSUB_BA( %%mm3, %%mm2 )
152
        SUMSUB_BA( %%mm1, %%mm0 )
153
        :: "r"(block)
154
    );
155
}
156

    
157
static void ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
158
{
159
    int i;
160
    int16_t __attribute__ ((aligned(8))) b2[64];
161

    
162
    block[0] += 32;
163

    
164
    for(i=0; i<2; i++){
165
        DECLARE_ALIGNED_8(uint64_t, tmp);
166

    
167
        h264_idct8_1d(block+4*i);
168

    
169
        __asm__ volatile(
170
            "movq   %%mm7,    %0   \n\t"
171
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
172
            "movq   %%mm0,  8(%1)  \n\t"
173
            "movq   %%mm6, 24(%1)  \n\t"
174
            "movq   %%mm7, 40(%1)  \n\t"
175
            "movq   %%mm4, 56(%1)  \n\t"
176
            "movq    %0,    %%mm7  \n\t"
177
            TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
178
            "movq   %%mm7,   (%1)  \n\t"
179
            "movq   %%mm1, 16(%1)  \n\t"
180
            "movq   %%mm0, 32(%1)  \n\t"
181
            "movq   %%mm3, 48(%1)  \n\t"
182
            : "=m"(tmp)
183
            : "r"(b2+32*i)
184
            : "memory"
185
        );
186
    }
187

    
188
    for(i=0; i<2; i++){
189
        h264_idct8_1d(b2+4*i);
190

    
191
        __asm__ volatile(
192
            "psraw     $6, %%mm7  \n\t"
193
            "psraw     $6, %%mm6  \n\t"
194
            "psraw     $6, %%mm5  \n\t"
195
            "psraw     $6, %%mm4  \n\t"
196
            "psraw     $6, %%mm3  \n\t"
197
            "psraw     $6, %%mm2  \n\t"
198
            "psraw     $6, %%mm1  \n\t"
199
            "psraw     $6, %%mm0  \n\t"
200

    
201
            "movq   %%mm7,    (%0)  \n\t"
202
            "movq   %%mm5,  16(%0)  \n\t"
203
            "movq   %%mm3,  32(%0)  \n\t"
204
            "movq   %%mm1,  48(%0)  \n\t"
205
            "movq   %%mm0,  64(%0)  \n\t"
206
            "movq   %%mm2,  80(%0)  \n\t"
207
            "movq   %%mm4,  96(%0)  \n\t"
208
            "movq   %%mm6, 112(%0)  \n\t"
209
            :: "r"(b2+4*i)
210
            : "memory"
211
        );
212
    }
213

    
214
    add_pixels_clamped_mmx(b2, dst, stride);
215
}
216

    
217
#define STORE_DIFF_8P( p, d, t, z )\
218
        "movq       "#d", "#t" \n"\
219
        "psraw       $6,  "#p" \n"\
220
        "punpcklbw  "#z", "#t" \n"\
221
        "paddsw     "#t", "#p" \n"\
222
        "packuswb   "#p", "#p" \n"\
223
        "movq       "#p", "#d" \n"
224

    
225
#define H264_IDCT8_1D_SSE2(a,b,c,d,e,f,g,h)\
226
        "movdqa     "#c", "#a" \n"\
227
        "movdqa     "#g", "#e" \n"\
228
        "psraw       $1,  "#c" \n"\
229
        "psraw       $1,  "#g" \n"\
230
        "psubw      "#e", "#c" \n"\
231
        "paddw      "#a", "#g" \n"\
232
        "movdqa     "#b", "#e" \n"\
233
        "psraw       $1,  "#e" \n"\
234
        "paddw      "#b", "#e" \n"\
235
        "paddw      "#d", "#e" \n"\
236
        "paddw      "#f", "#e" \n"\
237
        "movdqa     "#f", "#a" \n"\
238
        "psraw       $1,  "#a" \n"\
239
        "paddw      "#f", "#a" \n"\
240
        "paddw      "#h", "#a" \n"\
241
        "psubw      "#b", "#a" \n"\
242
        "psubw      "#d", "#b" \n"\
243
        "psubw      "#d", "#f" \n"\
244
        "paddw      "#h", "#b" \n"\
245
        "psubw      "#h", "#f" \n"\
246
        "psraw       $1,  "#d" \n"\
247
        "psraw       $1,  "#h" \n"\
248
        "psubw      "#d", "#b" \n"\
249
        "psubw      "#h", "#f" \n"\
250
        "movdqa     "#e", "#d" \n"\
251
        "movdqa     "#a", "#h" \n"\
252
        "psraw       $2,  "#d" \n"\
253
        "psraw       $2,  "#h" \n"\
254
        "paddw      "#f", "#d" \n"\
255
        "paddw      "#b", "#h" \n"\
256
        "psraw       $2,  "#f" \n"\
257
        "psraw       $2,  "#b" \n"\
258
        "psubw      "#f", "#e" \n"\
259
        "psubw      "#a", "#b" \n"\
260
        "movdqa 0x00(%1), "#a" \n"\
261
        "movdqa 0x40(%1), "#f" \n"\
262
        SUMSUB_BA(f, a)\
263
        SUMSUB_BA(g, f)\
264
        SUMSUB_BA(c, a)\
265
        SUMSUB_BA(e, g)\
266
        SUMSUB_BA(b, c)\
267
        SUMSUB_BA(h, a)\
268
        SUMSUB_BA(d, f)
269

    
270
static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
271
{
272
    __asm__ volatile(
273
        "movdqa   0x10(%1), %%xmm1 \n"
274
        "movdqa   0x20(%1), %%xmm2 \n"
275
        "movdqa   0x30(%1), %%xmm3 \n"
276
        "movdqa   0x50(%1), %%xmm5 \n"
277
        "movdqa   0x60(%1), %%xmm6 \n"
278
        "movdqa   0x70(%1), %%xmm7 \n"
279
        H264_IDCT8_1D_SSE2(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)
280
        TRANSPOSE8(%%xmm4, %%xmm1, %%xmm7, %%xmm3, %%xmm5, %%xmm0, %%xmm2, %%xmm6, (%1))
281
        "paddw          %4, %%xmm4 \n"
282
        "movdqa     %%xmm4, 0x00(%1) \n"
283
        "movdqa     %%xmm2, 0x40(%1) \n"
284
        H264_IDCT8_1D_SSE2(%%xmm4, %%xmm0, %%xmm6, %%xmm3, %%xmm2, %%xmm5, %%xmm7, %%xmm1)
285
        "movdqa     %%xmm6, 0x60(%1) \n"
286
        "movdqa     %%xmm7, 0x70(%1) \n"
287
        "pxor       %%xmm7, %%xmm7 \n"
288
        STORE_DIFF_8P(%%xmm2, (%0),      %%xmm6, %%xmm7)
289
        STORE_DIFF_8P(%%xmm0, (%0,%2),   %%xmm6, %%xmm7)
290
        STORE_DIFF_8P(%%xmm1, (%0,%2,2), %%xmm6, %%xmm7)
291
        STORE_DIFF_8P(%%xmm3, (%0,%3),   %%xmm6, %%xmm7)
292
        "lea     (%0,%2,4), %0 \n"
293
        STORE_DIFF_8P(%%xmm5, (%0),      %%xmm6, %%xmm7)
294
        STORE_DIFF_8P(%%xmm4, (%0,%2),   %%xmm6, %%xmm7)
295
        "movdqa   0x60(%1), %%xmm0 \n"
296
        "movdqa   0x70(%1), %%xmm1 \n"
297
        STORE_DIFF_8P(%%xmm0, (%0,%2,2), %%xmm6, %%xmm7)
298
        STORE_DIFF_8P(%%xmm1, (%0,%3),   %%xmm6, %%xmm7)
299
        :"+r"(dst)
300
        :"r"(block), "r"((x86_reg)stride), "r"((x86_reg)3L*stride), "m"(ff_pw_32)
301
    );
302
}
303

    
304
static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
305
{
306
    int dc = (block[0] + 32) >> 6;
307
    __asm__ volatile(
308
        "movd          %0, %%mm0 \n\t"
309
        "pshufw $0, %%mm0, %%mm0 \n\t"
310
        "pxor       %%mm1, %%mm1 \n\t"
311
        "psubw      %%mm0, %%mm1 \n\t"
312
        "packuswb   %%mm0, %%mm0 \n\t"
313
        "packuswb   %%mm1, %%mm1 \n\t"
314
        ::"r"(dc)
315
    );
316
    __asm__ volatile(
317
        "movd          %0, %%mm2 \n\t"
318
        "movd          %1, %%mm3 \n\t"
319
        "movd          %2, %%mm4 \n\t"
320
        "movd          %3, %%mm5 \n\t"
321
        "paddusb    %%mm0, %%mm2 \n\t"
322
        "paddusb    %%mm0, %%mm3 \n\t"
323
        "paddusb    %%mm0, %%mm4 \n\t"
324
        "paddusb    %%mm0, %%mm5 \n\t"
325
        "psubusb    %%mm1, %%mm2 \n\t"
326
        "psubusb    %%mm1, %%mm3 \n\t"
327
        "psubusb    %%mm1, %%mm4 \n\t"
328
        "psubusb    %%mm1, %%mm5 \n\t"
329
        "movd       %%mm2, %0    \n\t"
330
        "movd       %%mm3, %1    \n\t"
331
        "movd       %%mm4, %2    \n\t"
332
        "movd       %%mm5, %3    \n\t"
333
        :"+m"(*(uint32_t*)(dst+0*stride)),
334
         "+m"(*(uint32_t*)(dst+1*stride)),
335
         "+m"(*(uint32_t*)(dst+2*stride)),
336
         "+m"(*(uint32_t*)(dst+3*stride))
337
    );
338
}
339

    
340
static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
341
{
342
    int dc = (block[0] + 32) >> 6;
343
    int y;
344
    __asm__ volatile(
345
        "movd          %0, %%mm0 \n\t"
346
        "pshufw $0, %%mm0, %%mm0 \n\t"
347
        "pxor       %%mm1, %%mm1 \n\t"
348
        "psubw      %%mm0, %%mm1 \n\t"
349
        "packuswb   %%mm0, %%mm0 \n\t"
350
        "packuswb   %%mm1, %%mm1 \n\t"
351
        ::"r"(dc)
352
    );
353
    for(y=2; y--; dst += 4*stride){
354
    __asm__ volatile(
355
        "movq          %0, %%mm2 \n\t"
356
        "movq          %1, %%mm3 \n\t"
357
        "movq          %2, %%mm4 \n\t"
358
        "movq          %3, %%mm5 \n\t"
359
        "paddusb    %%mm0, %%mm2 \n\t"
360
        "paddusb    %%mm0, %%mm3 \n\t"
361
        "paddusb    %%mm0, %%mm4 \n\t"
362
        "paddusb    %%mm0, %%mm5 \n\t"
363
        "psubusb    %%mm1, %%mm2 \n\t"
364
        "psubusb    %%mm1, %%mm3 \n\t"
365
        "psubusb    %%mm1, %%mm4 \n\t"
366
        "psubusb    %%mm1, %%mm5 \n\t"
367
        "movq       %%mm2, %0    \n\t"
368
        "movq       %%mm3, %1    \n\t"
369
        "movq       %%mm4, %2    \n\t"
370
        "movq       %%mm5, %3    \n\t"
371
        :"+m"(*(uint64_t*)(dst+0*stride)),
372
         "+m"(*(uint64_t*)(dst+1*stride)),
373
         "+m"(*(uint64_t*)(dst+2*stride)),
374
         "+m"(*(uint64_t*)(dst+3*stride))
375
    );
376
    }
377
}
378

    
379
//FIXME this table is a duplicate from h264data.h, and will be removed once the tables from, h264 have been split
380
static const uint8_t scan8[16 + 2*4]={
381
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
382
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
383
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
384
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
385
 1+1*8, 2+1*8,
386
 1+2*8, 2+2*8,
387
 1+4*8, 2+4*8,
388
 1+5*8, 2+5*8,
389
};
390

    
391
static void ff_h264_idct_add16_mmx(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
392
    int i;
393
    for(i=0; i<16; i++){
394
        if(nnzc[ scan8[i] ])
395
            ff_h264_idct_add_mmx(dst + block_offset[i], block + i*16, stride);
396
    }
397
}
398

    
399
static void ff_h264_idct8_add4_mmx(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
400
    int i;
401
    for(i=0; i<16; i+=4){
402
        if(nnzc[ scan8[i] ])
403
            ff_h264_idct8_add_mmx(dst + block_offset[i], block + i*16, stride);
404
    }
405
}
406

    
407

    
408
static void ff_h264_idct_add16_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
409
    int i;
410
    for(i=0; i<16; i++){
411
        int nnz = nnzc[ scan8[i] ];
412
        if(nnz){
413
            if(nnz==1 && block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
414
            else                      ff_h264_idct_add_mmx    (dst + block_offset[i], block + i*16, stride);
415
        }
416
    }
417
}
418

    
419
static void ff_h264_idct_add16intra_mmx(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
420
    int i;
421
    for(i=0; i<16; i++){
422
        if(nnzc[ scan8[i] ] || block[i*16])
423
            ff_h264_idct_add_mmx(dst + block_offset[i], block + i*16, stride);
424
    }
425
}
426

    
427
static void ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
428
    int i;
429
    for(i=0; i<16; i++){
430
        if(nnzc[ scan8[i] ]) ff_h264_idct_add_mmx    (dst + block_offset[i], block + i*16, stride);
431
        else if(block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
432
    }
433
}
434

    
435
static void ff_h264_idct8_add4_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
436
    int i;
437
    for(i=0; i<16; i+=4){
438
        int nnz = nnzc[ scan8[i] ];
439
        if(nnz){
440
            if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
441
            else                      ff_h264_idct8_add_mmx    (dst + block_offset[i], block + i*16, stride);
442
        }
443
    }
444
}
445

    
446
static void ff_h264_idct8_add4_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
447
    int i;
448
    for(i=0; i<16; i+=4){
449
        int nnz = nnzc[ scan8[i] ];
450
        if(nnz){
451
            if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
452
            else                      ff_h264_idct8_add_sse2   (dst + block_offset[i], block + i*16, stride);
453
        }
454
    }
455
}
456

    
457
static void ff_h264_idct_add8_mmx(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
458
    int i;
459
    for(i=16; i<16+8; i++){
460
        if(nnzc[ scan8[i] ] || block[i*16])
461
            ff_h264_idct_add_mmx    (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
462
    }
463
}
464

    
465
static void ff_h264_idct_add8_mmx2(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
466
    int i;
467
    for(i=16; i<16+8; i++){
468
        if(nnzc[ scan8[i] ])
469
            ff_h264_idct_add_mmx    (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
470
        else if(block[i*16])
471
            ff_h264_idct_dc_add_mmx2(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
472
    }
473
}
474

    
475
/***********************************/
476
/* deblocking */
477

    
478
// out: o = |x-y|>a
479
// clobbers: t
480
#define DIFF_GT_MMX(x,y,a,o,t)\
481
    "movq     "#y", "#t"  \n\t"\
482
    "movq     "#x", "#o"  \n\t"\
483
    "psubusb  "#x", "#t"  \n\t"\
484
    "psubusb  "#y", "#o"  \n\t"\
485
    "por      "#t", "#o"  \n\t"\
486
    "psubusb  "#a", "#o"  \n\t"
487

    
488
// out: o = |x-y|>a
489
// clobbers: t
490
#define DIFF_GT2_MMX(x,y,a,o,t)\
491
    "movq     "#y", "#t"  \n\t"\
492
    "movq     "#x", "#o"  \n\t"\
493
    "psubusb  "#x", "#t"  \n\t"\
494
    "psubusb  "#y", "#o"  \n\t"\
495
    "psubusb  "#a", "#t"  \n\t"\
496
    "psubusb  "#a", "#o"  \n\t"\
497
    "pcmpeqb  "#t", "#o"  \n\t"\
498

    
499
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
500
// out: mm5=beta-1, mm7=mask
501
// clobbers: mm4,mm6
502
#define H264_DEBLOCK_MASK(alpha1, beta1) \
503
    "pshufw $0, "#alpha1", %%mm4 \n\t"\
504
    "pshufw $0, "#beta1 ", %%mm5 \n\t"\
505
    "packuswb  %%mm4, %%mm4      \n\t"\
506
    "packuswb  %%mm5, %%mm5      \n\t"\
507
    DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
508
    DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
509
    "por       %%mm4, %%mm7      \n\t"\
510
    DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
511
    "por       %%mm4, %%mm7      \n\t"\
512
    "pxor      %%mm6, %%mm6      \n\t"\
513
    "pcmpeqb   %%mm6, %%mm7      \n\t"
514

    
515
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
516
// out: mm1=p0' mm2=q0'
517
// clobbers: mm0,3-6
518
#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
519
        "movq    %%mm1              , %%mm5 \n\t"\
520
        "pxor    %%mm2              , %%mm5 \n\t" /* p0^q0*/\
521
        "pand    "#pb_01"           , %%mm5 \n\t" /* (p0^q0)&1*/\
522
        "pcmpeqb %%mm4              , %%mm4 \n\t"\
523
        "pxor    %%mm4              , %%mm3 \n\t"\
524
        "pavgb   %%mm0              , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
525
        "pavgb   "MANGLE(ff_pb_3)"  , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
526
        "pxor    %%mm1              , %%mm4 \n\t"\
527
        "pavgb   %%mm2              , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
528
        "pavgb   %%mm5              , %%mm3 \n\t"\
529
        "paddusb %%mm4              , %%mm3 \n\t" /* d+128+33*/\
530
        "movq    "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
531
        "psubusb %%mm3              , %%mm6 \n\t"\
532
        "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
533
        "pminub  %%mm7              , %%mm6 \n\t"\
534
        "pminub  %%mm7              , %%mm3 \n\t"\
535
        "psubusb %%mm6              , %%mm1 \n\t"\
536
        "psubusb %%mm3              , %%mm2 \n\t"\
537
        "paddusb %%mm3              , %%mm1 \n\t"\
538
        "paddusb %%mm6              , %%mm2 \n\t"
539

    
540
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
541
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
542
// clobbers: q2, tmp, tc0
543
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
544
        "movq     %%mm1,  "#tmp"   \n\t"\
545
        "pavgb    %%mm2,  "#tmp"   \n\t"\
546
        "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
547
        "pxor   "q2addr", "#tmp"   \n\t"\
548
        "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
549
        "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
550
        "movq     "#p1",  "#tmp"   \n\t"\
551
        "psubusb  "#tc0", "#tmp"   \n\t"\
552
        "paddusb  "#p1",  "#tc0"   \n\t"\
553
        "pmaxub   "#tmp", "#q2"    \n\t"\
554
        "pminub   "#tc0", "#q2"    \n\t"\
555
        "movq     "#q2",  "q1addr" \n\t"
556

    
557
static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
558
{
559
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);
560

    
561
    __asm__ volatile(
562
        "movq    (%1,%3), %%mm0    \n\t" //p1
563
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
564
        "movq    (%2),    %%mm2    \n\t" //q0
565
        "movq    (%2,%3), %%mm3    \n\t" //q1
566
        H264_DEBLOCK_MASK(%6, %7)
567

    
568
        "movd      %5,    %%mm4    \n\t"
569
        "punpcklbw %%mm4, %%mm4    \n\t"
570
        "punpcklwd %%mm4, %%mm4    \n\t"
571
        "pcmpeqb   %%mm3, %%mm3    \n\t"
572
        "movq      %%mm4, %%mm6    \n\t"
573
        "pcmpgtb   %%mm3, %%mm4    \n\t"
574
        "movq      %%mm6, 8+%0     \n\t"
575
        "pand      %%mm4, %%mm7    \n\t"
576
        "movq      %%mm7, %0       \n\t"
577

    
578
        /* filter p1 */
579
        "movq     (%1),   %%mm3    \n\t" //p2
580
        DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
581
        "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
582
        "pand     8+%0,   %%mm7    \n\t" // mask & tc0
583
        "movq     %%mm7,  %%mm4    \n\t"
584
        "psubb    %%mm6,  %%mm7    \n\t"
585
        "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
586
        H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
587

    
588
        /* filter q1 */
589
        "movq    (%2,%3,2), %%mm4  \n\t" //q2
590
        DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
591
        "pand     %0,     %%mm6    \n\t"
592
        "movq     8+%0,   %%mm5    \n\t" // can be merged with the and below but is slower then
593
        "pand     %%mm6,  %%mm5    \n\t"
594
        "psubb    %%mm6,  %%mm7    \n\t"
595
        "movq    (%2,%3), %%mm3    \n\t"
596
        H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
597

    
598
        /* filter p0, q0 */
599
        H264_DEBLOCK_P0_Q0(%8, unused)
600
        "movq      %%mm1, (%1,%3,2) \n\t"
601
        "movq      %%mm2, (%2)      \n\t"
602

    
603
        : "=m"(*tmp0)
604
        : "r"(pix-3*stride), "r"(pix), "r"((x86_reg)stride),
605
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
606
          "m"(ff_bone)
607
    );
608
}
609

    
610
static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
611
{
612
    if((tc0[0] & tc0[1]) >= 0)
613
        h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
614
    if((tc0[2] & tc0[3]) >= 0)
615
        h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
616
}
617
static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
618
{
619
    //FIXME: could cut some load/stores by merging transpose with filter
620
    // also, it only needs to transpose 6x8
621
    DECLARE_ALIGNED_8(uint8_t, trans[8*8]);
622
    int i;
623
    for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
624
        if((tc0[0] & tc0[1]) < 0)
625
            continue;
626
        transpose4x4(trans,       pix-4,          8, stride);
627
        transpose4x4(trans  +4*8, pix,            8, stride);
628
        transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
629
        transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
630
        h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
631
        transpose4x4(pix-2,          trans  +2*8, stride, 8);
632
        transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
633
    }
634
}
635

    
636
static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
637
{
638
    __asm__ volatile(
639
        "movq    (%0),    %%mm0     \n\t" //p1
640
        "movq    (%0,%2), %%mm1     \n\t" //p0
641
        "movq    (%1),    %%mm2     \n\t" //q0
642
        "movq    (%1,%2), %%mm3     \n\t" //q1
643
        H264_DEBLOCK_MASK(%4, %5)
644
        "movd      %3,    %%mm6     \n\t"
645
        "punpcklbw %%mm6, %%mm6     \n\t"
646
        "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
647
        H264_DEBLOCK_P0_Q0(%6, %7)
648
        "movq      %%mm1, (%0,%2)   \n\t"
649
        "movq      %%mm2, (%1)      \n\t"
650

    
651
        :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
652
           "r"(*(uint32_t*)tc0),
653
           "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
654
    );
655
}
656

    
657
static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
658
{
659
    h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
660
}
661

    
662
static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
663
{
664
    //FIXME: could cut some load/stores by merging transpose with filter
665
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
666
    transpose4x4(trans, pix-2, 8, stride);
667
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
668
    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
669
    transpose4x4(pix-2, trans, stride, 8);
670
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
671
}
672

    
673
// p0 = (p0 + q1 + 2*p1 + 2) >> 2
674
#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
675
    "movq    "#p0", %%mm4  \n\t"\
676
    "pxor    "#q1", %%mm4  \n\t"\
677
    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
678
    "pavgb   "#q1", "#p0"  \n\t"\
679
    "psubusb %%mm4, "#p0"  \n\t"\
680
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
681

    
682
static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
683
{
684
    __asm__ volatile(
685
        "movq    (%0),    %%mm0     \n\t"
686
        "movq    (%0,%2), %%mm1     \n\t"
687
        "movq    (%1),    %%mm2     \n\t"
688
        "movq    (%1,%2), %%mm3     \n\t"
689
        H264_DEBLOCK_MASK(%3, %4)
690
        "movq    %%mm1,   %%mm5     \n\t"
691
        "movq    %%mm2,   %%mm6     \n\t"
692
        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
693
        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
694
        "psubb   %%mm5,   %%mm1     \n\t"
695
        "psubb   %%mm6,   %%mm2     \n\t"
696
        "pand    %%mm7,   %%mm1     \n\t"
697
        "pand    %%mm7,   %%mm2     \n\t"
698
        "paddb   %%mm5,   %%mm1     \n\t"
699
        "paddb   %%mm6,   %%mm2     \n\t"
700
        "movq    %%mm1,   (%0,%2)   \n\t"
701
        "movq    %%mm2,   (%1)      \n\t"
702
        :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
703
           "m"(alpha1), "m"(beta1), "m"(ff_bone)
704
    );
705
}
706

    
707
static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
708
{
709
    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
710
}
711

    
712
static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
713
{
714
    //FIXME: could cut some load/stores by merging transpose with filter
715
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
716
    transpose4x4(trans, pix-2, 8, stride);
717
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
718
    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
719
    transpose4x4(pix-2, trans, stride, 8);
720
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
721
}
722

    
723
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],
724
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
725
    int dir;
726
    __asm__ volatile(
727
        "pxor %%mm7, %%mm7 \n\t"
728
        "movq %0, %%mm6 \n\t"
729
        "movq %1, %%mm5 \n\t"
730
        "movq %2, %%mm4 \n\t"
731
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
732
    );
733
    if(field)
734
        __asm__ volatile(
735
            "movq %0, %%mm5 \n\t"
736
            "movq %1, %%mm4 \n\t"
737
            ::"m"(ff_pb_3_1), "m"(ff_pb_7_3)
738
        );
739

    
740
    // could do a special case for dir==0 && edges==1, but it only reduces the
741
    // average filter time by 1.2%
742
    for( dir=1; dir>=0; dir-- ) {
743
        const int d_idx = dir ? -8 : -1;
744
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
745
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
746
        int b_idx, edge, l;
747
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
748
            __asm__ volatile(
749
                "pand %0, %%mm0 \n\t"
750
                ::"m"(mask_dir)
751
            );
752
            if(!(mask_mv & edge)) {
753
                __asm__ volatile("pxor %%mm0, %%mm0 \n\t":);
754
                for( l = bidir; l >= 0; l-- ) {
755
                    __asm__ volatile(
756
                        "movd %0, %%mm1 \n\t"
757
                        "punpckldq %1, %%mm1 \n\t"
758
                        "movq %%mm1, %%mm2 \n\t"
759
                        "psrlw $7, %%mm2 \n\t"
760
                        "pand %%mm6, %%mm2 \n\t"
761
                        "por %%mm2, %%mm1 \n\t" // ref_cache with -2 mapped to -1
762
                        "punpckldq %%mm1, %%mm2 \n\t"
763
                        "pcmpeqb %%mm2, %%mm1 \n\t"
764
                        "paddb %%mm6, %%mm1 \n\t"
765
                        "punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn]
766
                        "por %%mm1, %%mm0 \n\t"
767

    
768
                        "movq %2, %%mm1 \n\t"
769
                        "movq %3, %%mm2 \n\t"
770
                        "psubw %4, %%mm1 \n\t"
771
                        "psubw %5, %%mm2 \n\t"
772
                        "packsswb %%mm2, %%mm1 \n\t"
773
                        "paddb %%mm5, %%mm1 \n\t"
774
                        "pminub %%mm4, %%mm1 \n\t"
775
                        "pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit
776
                        "por %%mm1, %%mm0 \n\t"
777
                        ::"m"(ref[l][b_idx]),
778
                          "m"(ref[l][b_idx+d_idx]),
779
                          "m"(mv[l][b_idx][0]),
780
                          "m"(mv[l][b_idx+2][0]),
781
                          "m"(mv[l][b_idx+d_idx][0]),
782
                          "m"(mv[l][b_idx+d_idx+2][0])
783
                    );
784
                }
785
            }
786
            __asm__ volatile(
787
                "movd %0, %%mm1 \n\t"
788
                "por  %1, %%mm1 \n\t"
789
                "punpcklbw %%mm7, %%mm1 \n\t"
790
                "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]
791
                ::"m"(nnz[b_idx]),
792
                  "m"(nnz[b_idx+d_idx])
793
            );
794
            __asm__ volatile(
795
                "pcmpeqw %%mm7, %%mm0 \n\t"
796
                "pcmpeqw %%mm7, %%mm0 \n\t"
797
                "psrlw $15, %%mm0 \n\t" // nonzero -> 1
798
                "psrlw $14, %%mm1 \n\t"
799
                "movq %%mm0, %%mm2 \n\t"
800
                "por %%mm1, %%mm2 \n\t"
801
                "psrlw $1, %%mm1 \n\t"
802
                "pandn %%mm2, %%mm1 \n\t"
803
                "movq %%mm1, %0 \n\t"
804
                :"=m"(*bS[dir][edge])
805
                ::"memory"
806
            );
807
        }
808
        edges = 4;
809
        step = 1;
810
    }
811
    __asm__ volatile(
812
        "movq   (%0), %%mm0 \n\t"
813
        "movq  8(%0), %%mm1 \n\t"
814
        "movq 16(%0), %%mm2 \n\t"
815
        "movq 24(%0), %%mm3 \n\t"
816
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
817
        "movq %%mm0,   (%0) \n\t"
818
        "movq %%mm3,  8(%0) \n\t"
819
        "movq %%mm4, 16(%0) \n\t"
820
        "movq %%mm2, 24(%0) \n\t"
821
        ::"r"(bS[0])
822
        :"memory"
823
    );
824
}
825

    
826
/***********************************/
827
/* motion compensation */
828

    
829
#define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
830
        "mov"#q" "#C", "#T"         \n\t"\
831
        "mov"#d" (%0), "#F"         \n\t"\
832
        "paddw "#D", "#T"           \n\t"\
833
        "psllw $2, "#T"             \n\t"\
834
        "psubw "#B", "#T"           \n\t"\
835
        "psubw "#E", "#T"           \n\t"\
836
        "punpcklbw "#Z", "#F"       \n\t"\
837
        "pmullw %4, "#T"            \n\t"\
838
        "paddw %5, "#A"             \n\t"\
839
        "add %2, %0                 \n\t"\
840
        "paddw "#F", "#A"           \n\t"\
841
        "paddw "#A", "#T"           \n\t"\
842
        "psraw $5, "#T"             \n\t"\
843
        "packuswb "#T", "#T"        \n\t"\
844
        OP(T, (%1), A, d)\
845
        "add %3, %1                 \n\t"
846

    
847
#define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
848
        "mov"#q" "#C", "#T"         \n\t"\
849
        "mov"#d" (%0), "#F"         \n\t"\
850
        "paddw "#D", "#T"           \n\t"\
851
        "psllw $2, "#T"             \n\t"\
852
        "paddw %4, "#A"             \n\t"\
853
        "psubw "#B", "#T"           \n\t"\
854
        "psubw "#E", "#T"           \n\t"\
855
        "punpcklbw "#Z", "#F"       \n\t"\
856
        "pmullw %3, "#T"            \n\t"\
857
        "paddw "#F", "#A"           \n\t"\
858
        "add %2, %0                 \n\t"\
859
        "paddw "#A", "#T"           \n\t"\
860
        "mov"#q" "#T", "#OF"(%1)    \n\t"
861

    
862
#define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
863
#define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
864
#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)
865
#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)
866

    
867

    
868
#define QPEL_H264(OPNAME, OP, MMX)\
869
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
870
    int h=4;\
871
\
872
    __asm__ volatile(\
873
        "pxor %%mm7, %%mm7          \n\t"\
874
        "movq %5, %%mm4             \n\t"\
875
        "movq %6, %%mm5             \n\t"\
876
        "1:                         \n\t"\
877
        "movd  -1(%0), %%mm1        \n\t"\
878
        "movd    (%0), %%mm2        \n\t"\
879
        "movd   1(%0), %%mm3        \n\t"\
880
        "movd   2(%0), %%mm0        \n\t"\
881
        "punpcklbw %%mm7, %%mm1     \n\t"\
882
        "punpcklbw %%mm7, %%mm2     \n\t"\
883
        "punpcklbw %%mm7, %%mm3     \n\t"\
884
        "punpcklbw %%mm7, %%mm0     \n\t"\
885
        "paddw %%mm0, %%mm1         \n\t"\
886
        "paddw %%mm3, %%mm2         \n\t"\
887
        "movd  -2(%0), %%mm0        \n\t"\
888
        "movd   3(%0), %%mm3        \n\t"\
889
        "punpcklbw %%mm7, %%mm0     \n\t"\
890
        "punpcklbw %%mm7, %%mm3     \n\t"\
891
        "paddw %%mm3, %%mm0         \n\t"\
892
        "psllw $2, %%mm2            \n\t"\
893
        "psubw %%mm1, %%mm2         \n\t"\
894
        "pmullw %%mm4, %%mm2        \n\t"\
895
        "paddw %%mm5, %%mm0         \n\t"\
896
        "paddw %%mm2, %%mm0         \n\t"\
897
        "psraw $5, %%mm0            \n\t"\
898
        "packuswb %%mm0, %%mm0      \n\t"\
899
        OP(%%mm0, (%1),%%mm6, d)\
900
        "add %3, %0                 \n\t"\
901
        "add %4, %1                 \n\t"\
902
        "decl %2                    \n\t"\
903
        " jnz 1b                    \n\t"\
904
        : "+a"(src), "+c"(dst), "+g"(h)\
905
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
906
        : "memory"\
907
    );\
908
}\
909
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
910
    int h=4;\
911
    __asm__ volatile(\
912
        "pxor %%mm7, %%mm7          \n\t"\
913
        "movq %0, %%mm4             \n\t"\
914
        "movq %1, %%mm5             \n\t"\
915
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
916
    );\
917
    do{\
918
    __asm__ volatile(\
919
        "movd  -1(%0), %%mm1        \n\t"\
920
        "movd    (%0), %%mm2        \n\t"\
921
        "movd   1(%0), %%mm3        \n\t"\
922
        "movd   2(%0), %%mm0        \n\t"\
923
        "punpcklbw %%mm7, %%mm1     \n\t"\
924
        "punpcklbw %%mm7, %%mm2     \n\t"\
925
        "punpcklbw %%mm7, %%mm3     \n\t"\
926
        "punpcklbw %%mm7, %%mm0     \n\t"\
927
        "paddw %%mm0, %%mm1         \n\t"\
928
        "paddw %%mm3, %%mm2         \n\t"\
929
        "movd  -2(%0), %%mm0        \n\t"\
930
        "movd   3(%0), %%mm3        \n\t"\
931
        "punpcklbw %%mm7, %%mm0     \n\t"\
932
        "punpcklbw %%mm7, %%mm3     \n\t"\
933
        "paddw %%mm3, %%mm0         \n\t"\
934
        "psllw $2, %%mm2            \n\t"\
935
        "psubw %%mm1, %%mm2         \n\t"\
936
        "pmullw %%mm4, %%mm2        \n\t"\
937
        "paddw %%mm5, %%mm0         \n\t"\
938
        "paddw %%mm2, %%mm0         \n\t"\
939
        "movd   (%2), %%mm3         \n\t"\
940
        "psraw $5, %%mm0            \n\t"\
941
        "packuswb %%mm0, %%mm0      \n\t"\
942
        PAVGB" %%mm3, %%mm0         \n\t"\
943
        OP(%%mm0, (%1),%%mm6, d)\
944
        "add %4, %0                 \n\t"\
945
        "add %4, %1                 \n\t"\
946
        "add %3, %2                 \n\t"\
947
        : "+a"(src), "+c"(dst), "+d"(src2)\
948
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
949
        : "memory"\
950
    );\
951
    }while(--h);\
952
}\
953
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
954
    src -= 2*srcStride;\
955
    __asm__ volatile(\
956
        "pxor %%mm7, %%mm7          \n\t"\
957
        "movd (%0), %%mm0           \n\t"\
958
        "add %2, %0                 \n\t"\
959
        "movd (%0), %%mm1           \n\t"\
960
        "add %2, %0                 \n\t"\
961
        "movd (%0), %%mm2           \n\t"\
962
        "add %2, %0                 \n\t"\
963
        "movd (%0), %%mm3           \n\t"\
964
        "add %2, %0                 \n\t"\
965
        "movd (%0), %%mm4           \n\t"\
966
        "add %2, %0                 \n\t"\
967
        "punpcklbw %%mm7, %%mm0     \n\t"\
968
        "punpcklbw %%mm7, %%mm1     \n\t"\
969
        "punpcklbw %%mm7, %%mm2     \n\t"\
970
        "punpcklbw %%mm7, %%mm3     \n\t"\
971
        "punpcklbw %%mm7, %%mm4     \n\t"\
972
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
973
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
974
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
975
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
976
         \
977
        : "+a"(src), "+c"(dst)\
978
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
979
        : "memory"\
980
    );\
981
}\
982
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){\
983
    int h=4;\
984
    int w=3;\
985
    src -= 2*srcStride+2;\
986
    while(w--){\
987
        __asm__ volatile(\
988
            "pxor %%mm7, %%mm7      \n\t"\
989
            "movd (%0), %%mm0       \n\t"\
990
            "add %2, %0             \n\t"\
991
            "movd (%0), %%mm1       \n\t"\
992
            "add %2, %0             \n\t"\
993
            "movd (%0), %%mm2       \n\t"\
994
            "add %2, %0             \n\t"\
995
            "movd (%0), %%mm3       \n\t"\
996
            "add %2, %0             \n\t"\
997
            "movd (%0), %%mm4       \n\t"\
998
            "add %2, %0             \n\t"\
999
            "punpcklbw %%mm7, %%mm0 \n\t"\
1000
            "punpcklbw %%mm7, %%mm1 \n\t"\
1001
            "punpcklbw %%mm7, %%mm2 \n\t"\
1002
            "punpcklbw %%mm7, %%mm3 \n\t"\
1003
            "punpcklbw %%mm7, %%mm4 \n\t"\
1004
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
1005
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
1006
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
1007
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
1008
             \
1009
            : "+a"(src)\
1010
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1011
            : "memory"\
1012
        );\
1013
        tmp += 4;\
1014
        src += 4 - 9*srcStride;\
1015
    }\
1016
    tmp -= 3*4;\
1017
    __asm__ volatile(\
1018
        "1:                         \n\t"\
1019
        "movq     (%0), %%mm0       \n\t"\
1020
        "paddw  10(%0), %%mm0       \n\t"\
1021
        "movq    2(%0), %%mm1       \n\t"\
1022
        "paddw   8(%0), %%mm1       \n\t"\
1023
        "movq    4(%0), %%mm2       \n\t"\
1024
        "paddw   6(%0), %%mm2       \n\t"\
1025
        "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
1026
        "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
1027
        "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
1028
        "paddsw %%mm2, %%mm0        \n\t"\
1029
        "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
1030
        "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
1031
        "psraw $6, %%mm0            \n\t"\
1032
        "packuswb %%mm0, %%mm0      \n\t"\
1033
        OP(%%mm0, (%1),%%mm7, d)\
1034
        "add $24, %0                \n\t"\
1035
        "add %3, %1                 \n\t"\
1036
        "decl %2                    \n\t"\
1037
        " jnz 1b                    \n\t"\
1038
        : "+a"(tmp), "+c"(dst), "+g"(h)\
1039
        : "S"((x86_reg)dstStride)\
1040
        : "memory"\
1041
    );\
1042
}\
1043
\
1044
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1045
    int h=8;\
1046
    __asm__ volatile(\
1047
        "pxor %%mm7, %%mm7          \n\t"\
1048
        "movq %5, %%mm6             \n\t"\
1049
        "1:                         \n\t"\
1050
        "movq    (%0), %%mm0        \n\t"\
1051
        "movq   1(%0), %%mm2        \n\t"\
1052
        "movq %%mm0, %%mm1          \n\t"\
1053
        "movq %%mm2, %%mm3          \n\t"\
1054
        "punpcklbw %%mm7, %%mm0     \n\t"\
1055
        "punpckhbw %%mm7, %%mm1     \n\t"\
1056
        "punpcklbw %%mm7, %%mm2     \n\t"\
1057
        "punpckhbw %%mm7, %%mm3     \n\t"\
1058
        "paddw %%mm2, %%mm0         \n\t"\
1059
        "paddw %%mm3, %%mm1         \n\t"\
1060
        "psllw $2, %%mm0            \n\t"\
1061
        "psllw $2, %%mm1            \n\t"\
1062
        "movq   -1(%0), %%mm2       \n\t"\
1063
        "movq    2(%0), %%mm4       \n\t"\
1064
        "movq %%mm2, %%mm3          \n\t"\
1065
        "movq %%mm4, %%mm5          \n\t"\
1066
        "punpcklbw %%mm7, %%mm2     \n\t"\
1067
        "punpckhbw %%mm7, %%mm3     \n\t"\
1068
        "punpcklbw %%mm7, %%mm4     \n\t"\
1069
        "punpckhbw %%mm7, %%mm5     \n\t"\
1070
        "paddw %%mm4, %%mm2         \n\t"\
1071
        "paddw %%mm3, %%mm5         \n\t"\
1072
        "psubw %%mm2, %%mm0         \n\t"\
1073
        "psubw %%mm5, %%mm1         \n\t"\
1074
        "pmullw %%mm6, %%mm0        \n\t"\
1075
        "pmullw %%mm6, %%mm1        \n\t"\
1076
        "movd   -2(%0), %%mm2       \n\t"\
1077
        "movd    7(%0), %%mm5       \n\t"\
1078
        "punpcklbw %%mm7, %%mm2     \n\t"\
1079
        "punpcklbw %%mm7, %%mm5     \n\t"\
1080
        "paddw %%mm3, %%mm2         \n\t"\
1081
        "paddw %%mm5, %%mm4         \n\t"\
1082
        "movq %6, %%mm5             \n\t"\
1083
        "paddw %%mm5, %%mm2         \n\t"\
1084
        "paddw %%mm5, %%mm4         \n\t"\
1085
        "paddw %%mm2, %%mm0         \n\t"\
1086
        "paddw %%mm4, %%mm1         \n\t"\
1087
        "psraw $5, %%mm0            \n\t"\
1088
        "psraw $5, %%mm1            \n\t"\
1089
        "packuswb %%mm1, %%mm0      \n\t"\
1090
        OP(%%mm0, (%1),%%mm5, q)\
1091
        "add %3, %0                 \n\t"\
1092
        "add %4, %1                 \n\t"\
1093
        "decl %2                    \n\t"\
1094
        " jnz 1b                    \n\t"\
1095
        : "+a"(src), "+c"(dst), "+g"(h)\
1096
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1097
        : "memory"\
1098
    );\
1099
}\
1100
\
1101
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1102
    int h=8;\
1103
    __asm__ volatile(\
1104
        "pxor %%mm7, %%mm7          \n\t"\
1105
        "movq %0, %%mm6             \n\t"\
1106
        :: "m"(ff_pw_5)\
1107
    );\
1108
    do{\
1109
    __asm__ volatile(\
1110
        "movq    (%0), %%mm0        \n\t"\
1111
        "movq   1(%0), %%mm2        \n\t"\
1112
        "movq %%mm0, %%mm1          \n\t"\
1113
        "movq %%mm2, %%mm3          \n\t"\
1114
        "punpcklbw %%mm7, %%mm0     \n\t"\
1115
        "punpckhbw %%mm7, %%mm1     \n\t"\
1116
        "punpcklbw %%mm7, %%mm2     \n\t"\
1117
        "punpckhbw %%mm7, %%mm3     \n\t"\
1118
        "paddw %%mm2, %%mm0         \n\t"\
1119
        "paddw %%mm3, %%mm1         \n\t"\
1120
        "psllw $2, %%mm0            \n\t"\
1121
        "psllw $2, %%mm1            \n\t"\
1122
        "movq   -1(%0), %%mm2       \n\t"\
1123
        "movq    2(%0), %%mm4       \n\t"\
1124
        "movq %%mm2, %%mm3          \n\t"\
1125
        "movq %%mm4, %%mm5          \n\t"\
1126
        "punpcklbw %%mm7, %%mm2     \n\t"\
1127
        "punpckhbw %%mm7, %%mm3     \n\t"\
1128
        "punpcklbw %%mm7, %%mm4     \n\t"\
1129
        "punpckhbw %%mm7, %%mm5     \n\t"\
1130
        "paddw %%mm4, %%mm2         \n\t"\
1131
        "paddw %%mm3, %%mm5         \n\t"\
1132
        "psubw %%mm2, %%mm0         \n\t"\
1133
        "psubw %%mm5, %%mm1         \n\t"\
1134
        "pmullw %%mm6, %%mm0        \n\t"\
1135
        "pmullw %%mm6, %%mm1        \n\t"\
1136
        "movd   -2(%0), %%mm2       \n\t"\
1137
        "movd    7(%0), %%mm5       \n\t"\
1138
        "punpcklbw %%mm7, %%mm2     \n\t"\
1139
        "punpcklbw %%mm7, %%mm5     \n\t"\
1140
        "paddw %%mm3, %%mm2         \n\t"\
1141
        "paddw %%mm5, %%mm4         \n\t"\
1142
        "movq %5, %%mm5             \n\t"\
1143
        "paddw %%mm5, %%mm2         \n\t"\
1144
        "paddw %%mm5, %%mm4         \n\t"\
1145
        "paddw %%mm2, %%mm0         \n\t"\
1146
        "paddw %%mm4, %%mm1         \n\t"\
1147
        "psraw $5, %%mm0            \n\t"\
1148
        "psraw $5, %%mm1            \n\t"\
1149
        "movq (%2), %%mm4           \n\t"\
1150
        "packuswb %%mm1, %%mm0      \n\t"\
1151
        PAVGB" %%mm4, %%mm0         \n\t"\
1152
        OP(%%mm0, (%1),%%mm5, q)\
1153
        "add %4, %0                 \n\t"\
1154
        "add %4, %1                 \n\t"\
1155
        "add %3, %2                 \n\t"\
1156
        : "+a"(src), "+c"(dst), "+d"(src2)\
1157
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1158
          "m"(ff_pw_16)\
1159
        : "memory"\
1160
    );\
1161
    }while(--h);\
1162
}\
1163
\
1164
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1165
    int w= 2;\
1166
    src -= 2*srcStride;\
1167
    \
1168
    while(w--){\
1169
      __asm__ volatile(\
1170
        "pxor %%mm7, %%mm7          \n\t"\
1171
        "movd (%0), %%mm0           \n\t"\
1172
        "add %2, %0                 \n\t"\
1173
        "movd (%0), %%mm1           \n\t"\
1174
        "add %2, %0                 \n\t"\
1175
        "movd (%0), %%mm2           \n\t"\
1176
        "add %2, %0                 \n\t"\
1177
        "movd (%0), %%mm3           \n\t"\
1178
        "add %2, %0                 \n\t"\
1179
        "movd (%0), %%mm4           \n\t"\
1180
        "add %2, %0                 \n\t"\
1181
        "punpcklbw %%mm7, %%mm0     \n\t"\
1182
        "punpcklbw %%mm7, %%mm1     \n\t"\
1183
        "punpcklbw %%mm7, %%mm2     \n\t"\
1184
        "punpcklbw %%mm7, %%mm3     \n\t"\
1185
        "punpcklbw %%mm7, %%mm4     \n\t"\
1186
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1187
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1188
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1189
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1190
        QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1191
        QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1192
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1193
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1194
         \
1195
        : "+a"(src), "+c"(dst)\
1196
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1197
        : "memory"\
1198
     );\
1199
     if(h==16){\
1200
        __asm__ volatile(\
1201
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1202
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1203
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1204
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1205
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1206
            QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1207
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1208
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1209
            \
1210
           : "+a"(src), "+c"(dst)\
1211
           : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1212
           : "memory"\
1213
        );\
1214
     }\
1215
     src += 4-(h+5)*srcStride;\
1216
     dst += 4-h*dstStride;\
1217
   }\
1218
}\
1219
static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
1220
    int w = (size+8)>>2;\
1221
    src -= 2*srcStride+2;\
1222
    while(w--){\
1223
        __asm__ volatile(\
1224
            "pxor %%mm7, %%mm7      \n\t"\
1225
            "movd (%0), %%mm0       \n\t"\
1226
            "add %2, %0             \n\t"\
1227
            "movd (%0), %%mm1       \n\t"\
1228
            "add %2, %0             \n\t"\
1229
            "movd (%0), %%mm2       \n\t"\
1230
            "add %2, %0             \n\t"\
1231
            "movd (%0), %%mm3       \n\t"\
1232
            "add %2, %0             \n\t"\
1233
            "movd (%0), %%mm4       \n\t"\
1234
            "add %2, %0             \n\t"\
1235
            "punpcklbw %%mm7, %%mm0 \n\t"\
1236
            "punpcklbw %%mm7, %%mm1 \n\t"\
1237
            "punpcklbw %%mm7, %%mm2 \n\t"\
1238
            "punpcklbw %%mm7, %%mm3 \n\t"\
1239
            "punpcklbw %%mm7, %%mm4 \n\t"\
1240
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
1241
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
1242
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
1243
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
1244
            QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
1245
            QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
1246
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
1247
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
1248
            : "+a"(src)\
1249
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1250
            : "memory"\
1251
        );\
1252
        if(size==16){\
1253
            __asm__ volatile(\
1254
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
1255
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
1256
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
1257
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
1258
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
1259
                QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
1260
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
1261
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
1262
                : "+a"(src)\
1263
                : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1264
                : "memory"\
1265
            );\
1266
        }\
1267
        tmp += 4;\
1268
        src += 4 - (size+5)*srcStride;\
1269
    }\
1270
}\
1271
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1272
    int w = size>>4;\
1273
    do{\
1274
    int h = size;\
1275
    __asm__ volatile(\
1276
        "1:                         \n\t"\
1277
        "movq     (%0), %%mm0       \n\t"\
1278
        "movq    8(%0), %%mm3       \n\t"\
1279
        "movq    2(%0), %%mm1       \n\t"\
1280
        "movq   10(%0), %%mm4       \n\t"\
1281
        "paddw   %%mm4, %%mm0       \n\t"\
1282
        "paddw   %%mm3, %%mm1       \n\t"\
1283
        "paddw  18(%0), %%mm3       \n\t"\
1284
        "paddw  16(%0), %%mm4       \n\t"\
1285
        "movq    4(%0), %%mm2       \n\t"\
1286
        "movq   12(%0), %%mm5       \n\t"\
1287
        "paddw   6(%0), %%mm2       \n\t"\
1288
        "paddw  14(%0), %%mm5       \n\t"\
1289
        "psubw %%mm1, %%mm0         \n\t"\
1290
        "psubw %%mm4, %%mm3         \n\t"\
1291
        "psraw $2, %%mm0            \n\t"\
1292
        "psraw $2, %%mm3            \n\t"\
1293
        "psubw %%mm1, %%mm0         \n\t"\
1294
        "psubw %%mm4, %%mm3         \n\t"\
1295
        "paddsw %%mm2, %%mm0        \n\t"\
1296
        "paddsw %%mm5, %%mm3        \n\t"\
1297
        "psraw $2, %%mm0            \n\t"\
1298
        "psraw $2, %%mm3            \n\t"\
1299
        "paddw %%mm2, %%mm0         \n\t"\
1300
        "paddw %%mm5, %%mm3         \n\t"\
1301
        "psraw $6, %%mm0            \n\t"\
1302
        "psraw $6, %%mm3            \n\t"\
1303
        "packuswb %%mm3, %%mm0      \n\t"\
1304
        OP(%%mm0, (%1),%%mm7, q)\
1305
        "add $48, %0                \n\t"\
1306
        "add %3, %1                 \n\t"\
1307
        "decl %2                    \n\t"\
1308
        " jnz 1b                    \n\t"\
1309
        : "+a"(tmp), "+c"(dst), "+g"(h)\
1310
        : "S"((x86_reg)dstStride)\
1311
        : "memory"\
1312
    );\
1313
    tmp += 8 - size*24;\
1314
    dst += 8 - size*dstStride;\
1315
    }while(w--);\
1316
}\
1317
\
1318
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1319
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1320
}\
1321
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1322
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1323
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1324
}\
1325
\
1326
static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1327
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1328
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1329
    src += 8*srcStride;\
1330
    dst += 8*dstStride;\
1331
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1332
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1333
}\
1334
\
1335
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1336
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1337
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1338
    src += 8*dstStride;\
1339
    dst += 8*dstStride;\
1340
    src2 += 8*src2Stride;\
1341
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1342
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1343
}\
1344
\
1345
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){\
1346
          put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
1347
    OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1348
}\
1349
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1350
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
1351
}\
1352
\
1353
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1354
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
1355
}\
1356
\
1357
static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1358
{\
1359
    __asm__ volatile(\
1360
        "movq      (%1), %%mm0          \n\t"\
1361
        "movq    24(%1), %%mm1          \n\t"\
1362
        "psraw      $5,  %%mm0          \n\t"\
1363
        "psraw      $5,  %%mm1          \n\t"\
1364
        "packuswb %%mm0, %%mm0          \n\t"\
1365
        "packuswb %%mm1, %%mm1          \n\t"\
1366
        PAVGB"     (%0), %%mm0          \n\t"\
1367
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1368
        OP(%%mm0, (%2),    %%mm4, d)\
1369
        OP(%%mm1, (%2,%4), %%mm5, d)\
1370
        "lea  (%0,%3,2), %0             \n\t"\
1371
        "lea  (%2,%4,2), %2             \n\t"\
1372
        "movq    48(%1), %%mm0          \n\t"\
1373
        "movq    72(%1), %%mm1          \n\t"\
1374
        "psraw      $5,  %%mm0          \n\t"\
1375
        "psraw      $5,  %%mm1          \n\t"\
1376
        "packuswb %%mm0, %%mm0          \n\t"\
1377
        "packuswb %%mm1, %%mm1          \n\t"\
1378
        PAVGB"     (%0), %%mm0          \n\t"\
1379
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1380
        OP(%%mm0, (%2),    %%mm4, d)\
1381
        OP(%%mm1, (%2,%4), %%mm5, d)\
1382
        :"+a"(src8), "+c"(src16), "+d"(dst)\
1383
        :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
1384
        :"memory");\
1385
}\
1386
static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1387
{\
1388
    do{\
1389
    __asm__ volatile(\
1390
        "movq      (%1), %%mm0          \n\t"\
1391
        "movq     8(%1), %%mm1          \n\t"\
1392
        "movq    48(%1), %%mm2          \n\t"\
1393
        "movq  8+48(%1), %%mm3          \n\t"\
1394
        "psraw      $5,  %%mm0          \n\t"\
1395
        "psraw      $5,  %%mm1          \n\t"\
1396
        "psraw      $5,  %%mm2          \n\t"\
1397
        "psraw      $5,  %%mm3          \n\t"\
1398
        "packuswb %%mm1, %%mm0          \n\t"\
1399
        "packuswb %%mm3, %%mm2          \n\t"\
1400
        PAVGB"     (%0), %%mm0          \n\t"\
1401
        PAVGB"  (%0,%3), %%mm2          \n\t"\
1402
        OP(%%mm0, (%2), %%mm5, q)\
1403
        OP(%%mm2, (%2,%4), %%mm5, q)\
1404
        ::"a"(src8), "c"(src16), "d"(dst),\
1405
          "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
1406
        :"memory");\
1407
        src8 += 2L*src8Stride;\
1408
        src16 += 48;\
1409
        dst += 2L*dstStride;\
1410
    }while(h-=2);\
1411
}\
1412
static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1413
{\
1414
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
1415
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1416
}\
1417

    
1418

    
1419
#ifdef ARCH_X86_64
1420
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1421
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1422
    int h=16;\
1423
    __asm__ volatile(\
1424
        "pxor %%xmm15, %%xmm15      \n\t"\
1425
        "movdqa %6, %%xmm14         \n\t"\
1426
        "movdqa %7, %%xmm13         \n\t"\
1427
        "1:                         \n\t"\
1428
        "lddqu    3(%0), %%xmm1     \n\t"\
1429
        "lddqu   -5(%0), %%xmm7     \n\t"\
1430
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1431
        "punpckhbw %%xmm15, %%xmm1  \n\t"\
1432
        "punpcklbw %%xmm15, %%xmm0  \n\t"\
1433
        "punpcklbw %%xmm15, %%xmm7  \n\t"\
1434
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1435
        "movdqa  %%xmm0, %%xmm6     \n\t"\
1436
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1437
        "movdqa  %%xmm0, %%xmm8     \n\t"\
1438
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1439
        "movdqa  %%xmm0, %%xmm9     \n\t"\
1440
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1441
        "movdqa  %%xmm0, %%xmm10    \n\t"\
1442
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1443
        "palignr $6, %%xmm7, %%xmm10\n\t"\
1444
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1445
        "palignr $8, %%xmm7, %%xmm9 \n\t"\
1446
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1447
        "palignr $10,%%xmm7, %%xmm8 \n\t"\
1448
        "paddw   %%xmm1, %%xmm5     \n\t"\
1449
        "paddw   %%xmm0, %%xmm10    \n\t"\
1450
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1451
        "palignr $12,%%xmm7, %%xmm6 \n\t"\
1452
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1453
        "palignr $14,%%xmm7, %%xmm0 \n\t"\
1454
        "paddw   %%xmm3, %%xmm2     \n\t"\
1455
        "paddw   %%xmm8, %%xmm6     \n\t"\
1456
        "paddw   %%xmm4, %%xmm1     \n\t"\
1457
        "paddw   %%xmm9, %%xmm0     \n\t"\
1458
        "psllw   $2,     %%xmm2     \n\t"\
1459
        "psllw   $2,     %%xmm6     \n\t"\
1460
        "psubw   %%xmm1, %%xmm2     \n\t"\
1461
        "psubw   %%xmm0, %%xmm6     \n\t"\
1462
        "paddw   %%xmm13,%%xmm5     \n\t"\
1463
        "paddw   %%xmm13,%%xmm10    \n\t"\
1464
        "pmullw  %%xmm14,%%xmm2     \n\t"\
1465
        "pmullw  %%xmm14,%%xmm6     \n\t"\
1466
        "lddqu   (%2),   %%xmm3     \n\t"\
1467
        "paddw   %%xmm5, %%xmm2     \n\t"\
1468
        "paddw   %%xmm10,%%xmm6     \n\t"\
1469
        "psraw   $5,     %%xmm2     \n\t"\
1470
        "psraw   $5,     %%xmm6     \n\t"\
1471
        "packuswb %%xmm2,%%xmm6     \n\t"\
1472
        "pavgb   %%xmm3, %%xmm6     \n\t"\
1473
        OP(%%xmm6, (%1), %%xmm4, dqa)\
1474
        "add %5, %0                 \n\t"\
1475
        "add %5, %1                 \n\t"\
1476
        "add %4, %2                 \n\t"\
1477
        "decl %3                    \n\t"\
1478
        "jg 1b                      \n\t"\
1479
        : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
1480
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1481
          "m"(ff_pw_5), "m"(ff_pw_16)\
1482
        : "memory"\
1483
    );\
1484
}
1485
#else // ARCH_X86_64
1486
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1487
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1488
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1489
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1490
    src += 8*dstStride;\
1491
    dst += 8*dstStride;\
1492
    src2 += 8*src2Stride;\
1493
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1494
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1495
}
1496
#endif // ARCH_X86_64
1497

    
1498
#define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
1499
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1500
    int h=8;\
1501
    __asm__ volatile(\
1502
        "pxor %%xmm7, %%xmm7        \n\t"\
1503
        "movdqa %0, %%xmm6          \n\t"\
1504
        :: "m"(ff_pw_5)\
1505
    );\
1506
    do{\
1507
    __asm__ volatile(\
1508
        "lddqu   -5(%0), %%xmm1     \n\t"\
1509
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1510
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
1511
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1512
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1513
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1514
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1515
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1516
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1517
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1518
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1519
        "paddw   %%xmm1, %%xmm5     \n\t"\
1520
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1521
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1522
        "paddw   %%xmm3, %%xmm2     \n\t"\
1523
        "paddw   %%xmm4, %%xmm1     \n\t"\
1524
        "psllw   $2,     %%xmm2     \n\t"\
1525
        "movq    (%2),   %%xmm3     \n\t"\
1526
        "psubw   %%xmm1, %%xmm2     \n\t"\
1527
        "paddw   %5,     %%xmm5     \n\t"\
1528
        "pmullw  %%xmm6, %%xmm2     \n\t"\
1529
        "paddw   %%xmm5, %%xmm2     \n\t"\
1530
        "psraw   $5,     %%xmm2     \n\t"\
1531
        "packuswb %%xmm2, %%xmm2    \n\t"\
1532
        "pavgb   %%xmm3, %%xmm2     \n\t"\
1533
        OP(%%xmm2, (%1), %%xmm4, q)\
1534
        "add %4, %0                 \n\t"\
1535
        "add %4, %1                 \n\t"\
1536
        "add %3, %2                 \n\t"\
1537
        : "+a"(src), "+c"(dst), "+d"(src2)\
1538
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1539
          "m"(ff_pw_16)\
1540
        : "memory"\
1541
    );\
1542
    }while(--h);\
1543
}\
1544
QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1545
\
1546
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1547
    int h=8;\
1548
    __asm__ volatile(\
1549
        "pxor %%xmm7, %%xmm7        \n\t"\
1550
        "movdqa %5, %%xmm6          \n\t"\
1551
        "1:                         \n\t"\
1552
        "lddqu   -5(%0), %%xmm1     \n\t"\
1553
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1554
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
1555
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1556
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1557
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1558
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1559
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1560
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1561
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1562
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1563
        "paddw   %%xmm1, %%xmm5     \n\t"\
1564
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1565
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1566
        "paddw   %%xmm3, %%xmm2     \n\t"\
1567
        "paddw   %%xmm4, %%xmm1     \n\t"\
1568
        "psllw   $2,     %%xmm2     \n\t"\
1569
        "psubw   %%xmm1, %%xmm2     \n\t"\
1570
        "paddw   %6,     %%xmm5     \n\t"\
1571
        "pmullw  %%xmm6, %%xmm2     \n\t"\
1572
        "paddw   %%xmm5, %%xmm2     \n\t"\
1573
        "psraw   $5,     %%xmm2     \n\t"\
1574
        "packuswb %%xmm2, %%xmm2    \n\t"\
1575
        OP(%%xmm2, (%1), %%xmm4, q)\
1576
        "add %3, %0                 \n\t"\
1577
        "add %4, %1                 \n\t"\
1578
        "decl %2                    \n\t"\
1579
        " jnz 1b                    \n\t"\
1580
        : "+a"(src), "+c"(dst), "+g"(h)\
1581
        : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\
1582
          "m"(ff_pw_5), "m"(ff_pw_16)\
1583
        : "memory"\
1584
    );\
1585
}\
1586
static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1587
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1588
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1589
    src += 8*srcStride;\
1590
    dst += 8*dstStride;\
1591
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1592
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1593
}\
1594

    
1595
#define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
1596
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1597
    src -= 2*srcStride;\
1598
    \
1599
    __asm__ volatile(\
1600
        "pxor %%xmm7, %%xmm7        \n\t"\
1601
        "movq (%0), %%xmm0          \n\t"\
1602
        "add %2, %0                 \n\t"\
1603
        "movq (%0), %%xmm1          \n\t"\
1604
        "add %2, %0                 \n\t"\
1605
        "movq (%0), %%xmm2          \n\t"\
1606
        "add %2, %0                 \n\t"\
1607
        "movq (%0), %%xmm3          \n\t"\
1608
        "add %2, %0                 \n\t"\
1609
        "movq (%0), %%xmm4          \n\t"\
1610
        "add %2, %0                 \n\t"\
1611
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1612
        "punpcklbw %%xmm7, %%xmm1   \n\t"\
1613
        "punpcklbw %%xmm7, %%xmm2   \n\t"\
1614
        "punpcklbw %%xmm7, %%xmm3   \n\t"\
1615
        "punpcklbw %%xmm7, %%xmm4   \n\t"\
1616
        QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1617
        QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1618
        QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1619
        QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1620
        QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1621
        QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1622
        QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1623
        QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1624
         \
1625
        : "+a"(src), "+c"(dst)\
1626
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1627
        : "memory"\
1628
    );\
1629
    if(h==16){\
1630
        __asm__ volatile(\
1631
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1632
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1633
            QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1634
            QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1635
            QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1636
            QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1637
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1638
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1639
            \
1640
            : "+a"(src), "+c"(dst)\
1641
            : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1642
            : "memory"\
1643
        );\
1644
    }\
1645
}\
1646
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1647
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1648
}\
1649
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1650
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1651
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1652
}
1653

    
1654
static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
1655
    int w = (size+8)>>3;
1656
    src -= 2*srcStride+2;
1657
    while(w--){
1658
        __asm__ volatile(
1659
            "pxor %%xmm7, %%xmm7        \n\t"
1660
            "movq (%0), %%xmm0          \n\t"
1661
            "add %2, %0                 \n\t"
1662
            "movq (%0), %%xmm1          \n\t"
1663
            "add %2, %0                 \n\t"
1664
            "movq (%0), %%xmm2          \n\t"
1665
            "add %2, %0                 \n\t"
1666
            "movq (%0), %%xmm3          \n\t"
1667
            "add %2, %0                 \n\t"
1668
            "movq (%0), %%xmm4          \n\t"
1669
            "add %2, %0                 \n\t"
1670
            "punpcklbw %%xmm7, %%xmm0   \n\t"
1671
            "punpcklbw %%xmm7, %%xmm1   \n\t"
1672
            "punpcklbw %%xmm7, %%xmm2   \n\t"
1673
            "punpcklbw %%xmm7, %%xmm3   \n\t"
1674
            "punpcklbw %%xmm7, %%xmm4   \n\t"
1675
            QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
1676
            QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
1677
            QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
1678
            QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
1679
            QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
1680
            QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
1681
            QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
1682
            QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
1683
            : "+a"(src)
1684
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1685
            : "memory"
1686
        );
1687
        if(size==16){
1688
            __asm__ volatile(
1689
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
1690
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
1691
                QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
1692
                QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
1693
                QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
1694
                QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
1695
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
1696
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
1697
                : "+a"(src)
1698
                : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1699
                : "memory"
1700
            );
1701
        }
1702
        tmp += 8;
1703
        src += 8 - (size+5)*srcStride;
1704
    }
1705
}
1706

    
1707
#define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
1708
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1709
    int h = size;\
1710
    if(size == 16){\
1711
        __asm__ volatile(\
1712
            "1:                         \n\t"\
1713
            "movdqa 32(%0), %%xmm4      \n\t"\
1714
            "movdqa 16(%0), %%xmm5      \n\t"\
1715
            "movdqa   (%0), %%xmm7      \n\t"\
1716
            "movdqa %%xmm4, %%xmm3      \n\t"\
1717
            "movdqa %%xmm4, %%xmm2      \n\t"\
1718
            "movdqa %%xmm4, %%xmm1      \n\t"\
1719
            "movdqa %%xmm4, %%xmm0      \n\t"\
1720
            "palignr $10, %%xmm5, %%xmm0 \n\t"\
1721
            "palignr  $8, %%xmm5, %%xmm1 \n\t"\
1722
            "palignr  $6, %%xmm5, %%xmm2 \n\t"\
1723
            "palignr  $4, %%xmm5, %%xmm3 \n\t"\
1724
            "palignr  $2, %%xmm5, %%xmm4 \n\t"\
1725
            "paddw  %%xmm5, %%xmm0      \n\t"\
1726
            "paddw  %%xmm4, %%xmm1      \n\t"\
1727
            "paddw  %%xmm3, %%xmm2      \n\t"\
1728
            "movdqa %%xmm5, %%xmm6      \n\t"\
1729
            "movdqa %%xmm5, %%xmm4      \n\t"\
1730
            "movdqa %%xmm5, %%xmm3      \n\t"\
1731
            "palignr  $8, %%xmm7, %%xmm4 \n\t"\
1732
            "palignr  $2, %%xmm7, %%xmm6 \n\t"\
1733
            "palignr $10, %%xmm7, %%xmm3 \n\t"\
1734
            "paddw  %%xmm6, %%xmm4      \n\t"\
1735
            "movdqa %%xmm5, %%xmm6      \n\t"\
1736
            "palignr  $6, %%xmm7, %%xmm5 \n\t"\
1737
            "palignr  $4, %%xmm7, %%xmm6 \n\t"\
1738
            "paddw  %%xmm7, %%xmm3      \n\t"\
1739
            "paddw  %%xmm6, %%xmm5      \n\t"\
1740
            \
1741
            "psubw  %%xmm1, %%xmm0      \n\t"\
1742
            "psubw  %%xmm4, %%xmm3      \n\t"\
1743
            "psraw      $2, %%xmm0      \n\t"\
1744
            "psraw      $2, %%xmm3      \n\t"\
1745
            "psubw  %%xmm1, %%xmm0      \n\t"\
1746
            "psubw  %%xmm4, %%xmm3      \n\t"\
1747
            "paddw  %%xmm2, %%xmm0      \n\t"\
1748
            "paddw  %%xmm5, %%xmm3      \n\t"\
1749
            "psraw      $2, %%xmm0      \n\t"\
1750
            "psraw      $2, %%xmm3      \n\t"\
1751
            "paddw  %%xmm2, %%xmm0      \n\t"\
1752
            "paddw  %%xmm5, %%xmm3      \n\t"\
1753
            "psraw      $6, %%xmm0      \n\t"\
1754
            "psraw      $6, %%xmm3      \n\t"\
1755
            "packuswb %%xmm0, %%xmm3    \n\t"\
1756
            OP(%%xmm3, (%1), %%xmm7, dqa)\
1757
            "add $48, %0                \n\t"\
1758
            "add %3, %1                 \n\t"\
1759
            "decl %2                    \n\t"\
1760
            " jnz 1b                    \n\t"\
1761
            : "+a"(tmp), "+c"(dst), "+g"(h)\
1762
            : "S"((x86_reg)dstStride)\
1763
            : "memory"\
1764
        );\
1765
    }else{\
1766
        __asm__ volatile(\
1767
            "1:                         \n\t"\
1768
            "movdqa 16(%0), %%xmm1      \n\t"\
1769
            "movdqa   (%0), %%xmm0      \n\t"\
1770
            "movdqa %%xmm1, %%xmm2      \n\t"\
1771
            "movdqa %%xmm1, %%xmm3      \n\t"\
1772
            "movdqa %%xmm1, %%xmm4      \n\t"\
1773
            "movdqa %%xmm1, %%xmm5      \n\t"\
1774
            "palignr $10, %%xmm0, %%xmm5 \n\t"\
1775
            "palignr  $8, %%xmm0, %%xmm4 \n\t"\
1776
            "palignr  $6, %%xmm0, %%xmm3 \n\t"\
1777
            "palignr  $4, %%xmm0, %%xmm2 \n\t"\
1778
            "palignr  $2, %%xmm0, %%xmm1 \n\t"\
1779
            "paddw  %%xmm5, %%xmm0      \n\t"\
1780
            "paddw  %%xmm4, %%xmm1      \n\t"\
1781
            "paddw  %%xmm3, %%xmm2      \n\t"\
1782
            "psubw  %%xmm1, %%xmm0      \n\t"\
1783
            "psraw      $2, %%xmm0      \n\t"\
1784
            "psubw  %%xmm1, %%xmm0      \n\t"\
1785
            "paddw  %%xmm2, %%xmm0      \n\t"\
1786
            "psraw      $2, %%xmm0      \n\t"\
1787
            "paddw  %%xmm2, %%xmm0      \n\t"\
1788
            "psraw      $6, %%xmm0      \n\t"\
1789
            "packuswb %%xmm0, %%xmm0    \n\t"\
1790
            OP(%%xmm0, (%1), %%xmm7, q)\
1791
            "add $48, %0                \n\t"\
1792
            "add %3, %1                 \n\t"\
1793
            "decl %2                    \n\t"\
1794
            " jnz 1b                    \n\t"\
1795
            : "+a"(tmp), "+c"(dst), "+g"(h)\
1796
            : "S"((x86_reg)dstStride)\
1797
            : "memory"\
1798
        );\
1799
    }\
1800
}
1801

    
1802
#define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
1803
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){\
1804
          put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
1805
    OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1806
}\
1807
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1808
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
1809
}\
1810
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1811
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1812
}\
1813

    
1814
#define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1815
#define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1816
#define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1817
#define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1818
#define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1819
#define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1820
#define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1821
#define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1822

    
1823
#define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1824
#define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1825
#define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1826
#define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1827
#define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1828
#define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1829
#define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1830
#define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1831

    
1832
#define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1833
#define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1834
#define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1835
#define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1836

    
1837
#define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1838
#define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1839
#define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1840
#define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1841

    
1842
#define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1843
#define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1844

    
1845
#define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1846
H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1847
H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1848
H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1849
H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1850

    
1851
static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1852
    put_pixels16_sse2(dst, src, stride, 16);
1853
}
1854
static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1855
    avg_pixels16_sse2(dst, src, stride, 16);
1856
}
1857
#define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1858
#define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1859

    
1860
#define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1861
static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1862
    OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1863
}\
1864

    
1865
#define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1866
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1867
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1868
}\
1869
\
1870
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1871
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1872
}\
1873
\
1874
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1875
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1876
}\
1877

    
1878
#define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1879
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1880
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1881
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1882
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1883
}\
1884
\
1885
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1886
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1887
}\
1888
\
1889
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1890
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1891
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1892
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1893
}\
1894

    
1895
#define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1896
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1897
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1898
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1899
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1900
}\
1901
\
1902
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1903
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1904
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1905
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1906
}\
1907
\
1908
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1909
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1910
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1911
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1912
}\
1913
\
1914
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1915
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1916
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1917
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1918
}\
1919
\
1920
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1921
    DECLARE_ALIGNED(ALIGN, uint16_t, temp[SIZE*(SIZE<8?12:24)]);\
1922
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
1923
}\
1924
\
1925
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1926
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1927
    uint8_t * const halfHV= temp;\
1928
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1929
    assert(((int)temp & 7) == 0);\
1930
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1931
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1932
}\
1933
\
1934
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1935
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1936
    uint8_t * const halfHV= temp;\
1937
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1938
    assert(((int)temp & 7) == 0);\
1939
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1940
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1941
}\
1942
\
1943
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1944
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1945
    uint8_t * const halfHV= temp;\
1946
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1947
    assert(((int)temp & 7) == 0);\
1948
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1949
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1950
}\
1951
\
1952
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1953
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1954
    uint8_t * const halfHV= temp;\
1955
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1956
    assert(((int)temp & 7) == 0);\
1957
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1958
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1959
}\
1960

    
1961
#define H264_MC_4816(MMX)\
1962
H264_MC(put_, 4, MMX, 8)\
1963
H264_MC(put_, 8, MMX, 8)\
1964
H264_MC(put_, 16,MMX, 8)\
1965
H264_MC(avg_, 4, MMX, 8)\
1966
H264_MC(avg_, 8, MMX, 8)\
1967
H264_MC(avg_, 16,MMX, 8)\
1968

    
1969
#define H264_MC_816(QPEL, XMM)\
1970
QPEL(put_, 8, XMM, 16)\
1971
QPEL(put_, 16,XMM, 16)\
1972
QPEL(avg_, 8, XMM, 16)\
1973
QPEL(avg_, 16,XMM, 16)\
1974

    
1975

    
1976
#define AVG_3DNOW_OP(a,b,temp, size) \
1977
"mov" #size " " #b ", " #temp "   \n\t"\
1978
"pavgusb " #temp ", " #a "        \n\t"\
1979
"mov" #size " " #a ", " #b "      \n\t"
1980
#define AVG_MMX2_OP(a,b,temp, size) \
1981
"mov" #size " " #b ", " #temp "   \n\t"\
1982
"pavgb " #temp ", " #a "          \n\t"\
1983
"mov" #size " " #a ", " #b "      \n\t"
1984

    
1985
#define PAVGB "pavgusb"
1986
QPEL_H264(put_,       PUT_OP, 3dnow)
1987
QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1988
#undef PAVGB
1989
#define PAVGB "pavgb"
1990
QPEL_H264(put_,       PUT_OP, mmx2)
1991
QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
1992
QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
1993
QPEL_H264_V_XMM(avg_,  AVG_MMX2_OP, sse2)
1994
QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
1995
QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, sse2)
1996
#ifdef HAVE_SSSE3
1997
QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
1998
QPEL_H264_H_XMM(avg_,  AVG_MMX2_OP, ssse3)
1999
QPEL_H264_HV2_XMM(put_,       PUT_OP, ssse3)
2000
QPEL_H264_HV2_XMM(avg_,  AVG_MMX2_OP, ssse3)
2001
QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
2002
QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, ssse3)
2003
#endif
2004
#undef PAVGB
2005

    
2006
H264_MC_4816(3dnow)
2007
H264_MC_4816(mmx2)
2008
H264_MC_816(H264_MC_V, sse2)
2009
H264_MC_816(H264_MC_HV, sse2)
2010
#ifdef HAVE_SSSE3
2011
H264_MC_816(H264_MC_H, ssse3)
2012
H264_MC_816(H264_MC_HV, ssse3)
2013
#endif
2014

    
2015

    
2016
#define H264_CHROMA_OP(S,D)
2017
#define H264_CHROMA_OP4(S,D,T)
2018
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
2019
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
2020
#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
2021
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
2022
#include "dsputil_h264_template_mmx.c"
2023

    
2024
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)
2025
{
2026
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 1);
2027
}
2028
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)
2029
{
2030
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 0);
2031
}
2032

    
2033
#undef H264_CHROMA_OP
2034
#undef H264_CHROMA_OP4
2035
#undef H264_CHROMA_MC8_TMPL
2036
#undef H264_CHROMA_MC4_TMPL
2037
#undef H264_CHROMA_MC2_TMPL
2038
#undef H264_CHROMA_MC8_MV0
2039

    
2040
#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
2041
#define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\
2042
                               "pavgb " #T ", " #D " \n\t"
2043
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
2044
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
2045
#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
2046
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
2047
#include "dsputil_h264_template_mmx.c"
2048
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)
2049
{
2050
    avg_h264_chroma_mc8_mmx2(dst, src, stride, h, x, y, 1);
2051
}
2052
#undef H264_CHROMA_OP
2053
#undef H264_CHROMA_OP4
2054
#undef H264_CHROMA_MC8_TMPL
2055
#undef H264_CHROMA_MC4_TMPL
2056
#undef H264_CHROMA_MC2_TMPL
2057
#undef H264_CHROMA_MC8_MV0
2058

    
2059
#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
2060
#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
2061
                               "pavgusb " #T ", " #D " \n\t"
2062
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
2063
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
2064
#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
2065
#include "dsputil_h264_template_mmx.c"
2066
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)
2067
{
2068
    avg_h264_chroma_mc8_3dnow(dst, src, stride, h, x, y, 1);
2069
}
2070
#undef H264_CHROMA_OP
2071
#undef H264_CHROMA_OP4
2072
#undef H264_CHROMA_MC8_TMPL
2073
#undef H264_CHROMA_MC4_TMPL
2074
#undef H264_CHROMA_MC8_MV0
2075

    
2076
#ifdef HAVE_SSSE3
2077
#define AVG_OP(X)
2078
#undef H264_CHROMA_MC8_TMPL
2079
#undef H264_CHROMA_MC4_TMPL
2080
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_ssse3
2081
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_ssse3
2082
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
2083
#include "dsputil_h264_template_ssse3.c"
2084
static void put_h264_chroma_mc8_ssse3_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2085
{
2086
    put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
2087
}
2088
static void put_h264_chroma_mc8_ssse3_nornd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2089
{
2090
    put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 0);
2091
}
2092

    
2093
#undef AVG_OP
2094
#undef H264_CHROMA_MC8_TMPL
2095
#undef H264_CHROMA_MC4_TMPL
2096
#undef H264_CHROMA_MC8_MV0
2097
#define AVG_OP(X) X
2098
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_ssse3
2099
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_ssse3
2100
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
2101
#include "dsputil_h264_template_ssse3.c"
2102
static void avg_h264_chroma_mc8_ssse3_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2103
{
2104
    avg_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
2105
}
2106
#undef AVG_OP
2107
#undef H264_CHROMA_MC8_TMPL
2108
#undef H264_CHROMA_MC4_TMPL
2109
#undef H264_CHROMA_MC8_MV0
2110
#endif
2111

    
2112
/***********************************/
2113
/* weighted prediction */
2114

    
2115
static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
2116
{
2117
    int x, y;
2118
    offset <<= log2_denom;
2119
    offset += (1 << log2_denom) >> 1;
2120
    __asm__ volatile(
2121
        "movd    %0, %%mm4        \n\t"
2122
        "movd    %1, %%mm5        \n\t"
2123
        "movd    %2, %%mm6        \n\t"
2124
        "pshufw  $0, %%mm4, %%mm4 \n\t"
2125
        "pshufw  $0, %%mm5, %%mm5 \n\t"
2126
        "pxor    %%mm7, %%mm7     \n\t"
2127
        :: "g"(weight), "g"(offset), "g"(log2_denom)
2128
    );
2129
    for(y=0; y<h; y+=2){
2130
        for(x=0; x<w; x+=4){
2131
            __asm__ volatile(
2132
                "movd      %0,    %%mm0 \n\t"
2133
                "movd      %1,    %%mm1 \n\t"
2134
                "punpcklbw %%mm7, %%mm0 \n\t"
2135
                "punpcklbw %%mm7, %%mm1 \n\t"
2136
                "pmullw    %%mm4, %%mm0 \n\t"
2137
                "pmullw    %%mm4, %%mm1 \n\t"
2138
                "paddsw    %%mm5, %%mm0 \n\t"
2139
                "paddsw    %%mm5, %%mm1 \n\t"
2140
                "psraw     %%mm6, %%mm0 \n\t"
2141
                "psraw     %%mm6, %%mm1 \n\t"
2142
                "packuswb  %%mm7, %%mm0 \n\t"
2143
                "packuswb  %%mm7, %%mm1 \n\t"
2144
                "movd      %%mm0, %0    \n\t"
2145
                "movd      %%mm1, %1    \n\t"
2146
                : "+m"(*(uint32_t*)(dst+x)),
2147
                  "+m"(*(uint32_t*)(dst+x+stride))
2148
            );
2149
        }
2150
        dst += 2*stride;
2151
    }
2152
}
2153

    
2154
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)
2155
{
2156
    int x, y;
2157
    offset = ((offset + 1) | 1) << log2_denom;
2158
    __asm__ volatile(
2159
        "movd    %0, %%mm3        \n\t"
2160
        "movd    %1, %%mm4        \n\t"
2161
        "movd    %2, %%mm5        \n\t"
2162
        "movd    %3, %%mm6        \n\t"
2163
        "pshufw  $0, %%mm3, %%mm3 \n\t"
2164
        "pshufw  $0, %%mm4, %%mm4 \n\t"
2165
        "pshufw  $0, %%mm5, %%mm5 \n\t"
2166
        "pxor    %%mm7, %%mm7     \n\t"
2167
        :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
2168
    );
2169
    for(y=0; y<h; y++){
2170
        for(x=0; x<w; x+=4){
2171
            __asm__ volatile(
2172
                "movd      %0,    %%mm0 \n\t"
2173
                "movd      %1,    %%mm1 \n\t"
2174
                "punpcklbw %%mm7, %%mm0 \n\t"
2175
                "punpcklbw %%mm7, %%mm1 \n\t"
2176
                "pmullw    %%mm3, %%mm0 \n\t"
2177
                "pmullw    %%mm4, %%mm1 \n\t"
2178
                "paddsw    %%mm1, %%mm0 \n\t"
2179
                "paddsw    %%mm5, %%mm0 \n\t"
2180
                "psraw     %%mm6, %%mm0 \n\t"
2181
                "packuswb  %%mm0, %%mm0 \n\t"
2182
                "movd      %%mm0, %0    \n\t"
2183
                : "+m"(*(uint32_t*)(dst+x))
2184
                :  "m"(*(uint32_t*)(src+x))
2185
            );
2186
        }
2187
        src += stride;
2188
        dst += stride;
2189
    }
2190
}
2191

    
2192
#define H264_WEIGHT(W,H) \
2193
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){ \
2194
    ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
2195
} \
2196
static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
2197
    ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
2198
}
2199

    
2200
H264_WEIGHT(16,16)
2201
H264_WEIGHT(16, 8)
2202
H264_WEIGHT( 8,16)
2203
H264_WEIGHT( 8, 8)
2204
H264_WEIGHT( 8, 4)
2205
H264_WEIGHT( 4, 8)
2206
H264_WEIGHT( 4, 4)
2207
H264_WEIGHT( 4, 2)
2208