Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / h264dsp_mmx.c @ b64dfbb8

History | View | Annotate | Download (61 KB)

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

    
21
#include "dsputil_mmx.h"
22

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
159
    block[0] += 32;
160

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

    
164
        h264_idct8_1d(block+4*i);
165

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

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

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

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

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

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

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

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

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

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

    
376

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
724
#define QPEL_H264V(A,B,C,D,E,F,OP)\
725
        "movd (%0), "#F"            \n\t"\
726
        "movq "#C", %%mm6           \n\t"\
727
        "paddw "#D", %%mm6          \n\t"\
728
        "psllw $2, %%mm6            \n\t"\
729
        "psubw "#B", %%mm6          \n\t"\
730
        "psubw "#E", %%mm6          \n\t"\
731
        "pmullw %4, %%mm6           \n\t"\
732
        "add %2, %0                 \n\t"\
733
        "punpcklbw %%mm7, "#F"      \n\t"\
734
        "paddw %5, "#A"             \n\t"\
735
        "paddw "#F", "#A"           \n\t"\
736
        "paddw "#A", %%mm6          \n\t"\
737
        "psraw $5, %%mm6            \n\t"\
738
        "packuswb %%mm6, %%mm6      \n\t"\
739
        OP(%%mm6, (%1), A, d)\
740
        "add %3, %1                 \n\t"
741

    
742
#define QPEL_H264HV(A,B,C,D,E,F,OF)\
743
        "movd (%0), "#F"            \n\t"\
744
        "movq "#C", %%mm6           \n\t"\
745
        "paddw "#D", %%mm6          \n\t"\
746
        "psllw $2, %%mm6            \n\t"\
747
        "psubw "#B", %%mm6          \n\t"\
748
        "psubw "#E", %%mm6          \n\t"\
749
        "pmullw %3, %%mm6           \n\t"\
750
        "add %2, %0                 \n\t"\
751
        "punpcklbw %%mm7, "#F"      \n\t"\
752
        "paddw %4, "#A"             \n\t"\
753
        "paddw "#F", "#A"           \n\t"\
754
        "paddw "#A", %%mm6          \n\t"\
755
        "movq %%mm6, "#OF"(%1)      \n\t"
756

    
757
#define QPEL_H264(OPNAME, OP, MMX)\
758
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
759
    int h=4;\
760
\
761
    asm volatile(\
762
        "pxor %%mm7, %%mm7          \n\t"\
763
        "movq %5, %%mm4             \n\t"\
764
        "movq %6, %%mm5             \n\t"\
765
        "1:                         \n\t"\
766
        "movd  -1(%0), %%mm1        \n\t"\
767
        "movd    (%0), %%mm2        \n\t"\
768
        "movd   1(%0), %%mm3        \n\t"\
769
        "movd   2(%0), %%mm0        \n\t"\
770
        "punpcklbw %%mm7, %%mm1     \n\t"\
771
        "punpcklbw %%mm7, %%mm2     \n\t"\
772
        "punpcklbw %%mm7, %%mm3     \n\t"\
773
        "punpcklbw %%mm7, %%mm0     \n\t"\
774
        "paddw %%mm0, %%mm1         \n\t"\
775
        "paddw %%mm3, %%mm2         \n\t"\
776
        "movd  -2(%0), %%mm0        \n\t"\
777
        "movd   3(%0), %%mm3        \n\t"\
778
        "punpcklbw %%mm7, %%mm0     \n\t"\
779
        "punpcklbw %%mm7, %%mm3     \n\t"\
780
        "paddw %%mm3, %%mm0         \n\t"\
781
        "psllw $2, %%mm2            \n\t"\
782
        "psubw %%mm1, %%mm2         \n\t"\
783
        "pmullw %%mm4, %%mm2        \n\t"\
784
        "paddw %%mm5, %%mm0         \n\t"\
785
        "paddw %%mm2, %%mm0         \n\t"\
786
        "psraw $5, %%mm0            \n\t"\
787
        "packuswb %%mm0, %%mm0      \n\t"\
788
        OP(%%mm0, (%1),%%mm6, d)\
789
        "add %3, %0                 \n\t"\
790
        "add %4, %1                 \n\t"\
791
        "decl %2                    \n\t"\
792
        " jnz 1b                    \n\t"\
793
        : "+a"(src), "+c"(dst), "+m"(h)\
794
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
795
        : "memory"\
796
    );\
797
}\
798
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
799
    int h=4;\
800
    asm volatile(\
801
        "pxor %%mm7, %%mm7          \n\t"\
802
        "movq %0, %%mm4             \n\t"\
803
        "movq %1, %%mm5             \n\t"\
804
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
805
    );\
806
    do{\
807
    asm volatile(\
808
        "movd  -1(%0), %%mm1        \n\t"\
809
        "movd    (%0), %%mm2        \n\t"\
810
        "movd   1(%0), %%mm3        \n\t"\
811
        "movd   2(%0), %%mm0        \n\t"\
812
        "punpcklbw %%mm7, %%mm1     \n\t"\
813
        "punpcklbw %%mm7, %%mm2     \n\t"\
814
        "punpcklbw %%mm7, %%mm3     \n\t"\
815
        "punpcklbw %%mm7, %%mm0     \n\t"\
816
        "paddw %%mm0, %%mm1         \n\t"\
817
        "paddw %%mm3, %%mm2         \n\t"\
818
        "movd  -2(%0), %%mm0        \n\t"\
819
        "movd   3(%0), %%mm3        \n\t"\
820
        "punpcklbw %%mm7, %%mm0     \n\t"\
821
        "punpcklbw %%mm7, %%mm3     \n\t"\
822
        "paddw %%mm3, %%mm0         \n\t"\
823
        "psllw $2, %%mm2            \n\t"\
824
        "psubw %%mm1, %%mm2         \n\t"\
825
        "pmullw %%mm4, %%mm2        \n\t"\
826
        "paddw %%mm5, %%mm0         \n\t"\
827
        "paddw %%mm2, %%mm0         \n\t"\
828
        "movd   (%2), %%mm3         \n\t"\
829
        "psraw $5, %%mm0            \n\t"\
830
        "packuswb %%mm0, %%mm0      \n\t"\
831
        PAVGB" %%mm3, %%mm0         \n\t"\
832
        OP(%%mm0, (%1),%%mm6, d)\
833
        "add %4, %0                 \n\t"\
834
        "add %4, %1                 \n\t"\
835
        "add %3, %2                 \n\t"\
836
        : "+a"(src), "+c"(dst), "+d"(src2)\
837
        : "D"((long)src2Stride), "S"((long)dstStride)\
838
        : "memory"\
839
    );\
840
    }while(--h);\
841
}\
842
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
843
    src -= 2*srcStride;\
844
    asm volatile(\
845
        "pxor %%mm7, %%mm7          \n\t"\
846
        "movd (%0), %%mm0           \n\t"\
847
        "add %2, %0                 \n\t"\
848
        "movd (%0), %%mm1           \n\t"\
849
        "add %2, %0                 \n\t"\
850
        "movd (%0), %%mm2           \n\t"\
851
        "add %2, %0                 \n\t"\
852
        "movd (%0), %%mm3           \n\t"\
853
        "add %2, %0                 \n\t"\
854
        "movd (%0), %%mm4           \n\t"\
855
        "add %2, %0                 \n\t"\
856
        "punpcklbw %%mm7, %%mm0     \n\t"\
857
        "punpcklbw %%mm7, %%mm1     \n\t"\
858
        "punpcklbw %%mm7, %%mm2     \n\t"\
859
        "punpcklbw %%mm7, %%mm3     \n\t"\
860
        "punpcklbw %%mm7, %%mm4     \n\t"\
861
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
862
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
863
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
864
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
865
         \
866
        : "+a"(src), "+c"(dst)\
867
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
868
        : "memory"\
869
    );\
870
}\
871
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){\
872
    int h=4;\
873
    int w=3;\
874
    src -= 2*srcStride+2;\
875
    while(w--){\
876
        asm volatile(\
877
            "pxor %%mm7, %%mm7      \n\t"\
878
            "movd (%0), %%mm0       \n\t"\
879
            "add %2, %0             \n\t"\
880
            "movd (%0), %%mm1       \n\t"\
881
            "add %2, %0             \n\t"\
882
            "movd (%0), %%mm2       \n\t"\
883
            "add %2, %0             \n\t"\
884
            "movd (%0), %%mm3       \n\t"\
885
            "add %2, %0             \n\t"\
886
            "movd (%0), %%mm4       \n\t"\
887
            "add %2, %0             \n\t"\
888
            "punpcklbw %%mm7, %%mm0 \n\t"\
889
            "punpcklbw %%mm7, %%mm1 \n\t"\
890
            "punpcklbw %%mm7, %%mm2 \n\t"\
891
            "punpcklbw %%mm7, %%mm3 \n\t"\
892
            "punpcklbw %%mm7, %%mm4 \n\t"\
893
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
894
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
895
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
896
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
897
             \
898
            : "+a"(src)\
899
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
900
            : "memory"\
901
        );\
902
        tmp += 4;\
903
        src += 4 - 9*srcStride;\
904
    }\
905
    tmp -= 3*4;\
906
    asm volatile(\
907
        "1:                         \n\t"\
908
        "movq     (%0), %%mm0       \n\t"\
909
        "paddw  10(%0), %%mm0       \n\t"\
910
        "movq    2(%0), %%mm1       \n\t"\
911
        "paddw   8(%0), %%mm1       \n\t"\
912
        "movq    4(%0), %%mm2       \n\t"\
913
        "paddw   6(%0), %%mm2       \n\t"\
914
        "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
915
        "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
916
        "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
917
        "paddsw %%mm2, %%mm0        \n\t"\
918
        "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
919
        "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
920
        "psraw $6, %%mm0            \n\t"\
921
        "packuswb %%mm0, %%mm0      \n\t"\
922
        OP(%%mm0, (%1),%%mm7, d)\
923
        "add $24, %0                \n\t"\
924
        "add %3, %1                 \n\t"\
925
        "decl %2                    \n\t"\
926
        " jnz 1b                    \n\t"\
927
        : "+a"(tmp), "+c"(dst), "+m"(h)\
928
        : "S"((long)dstStride)\
929
        : "memory"\
930
    );\
931
}\
932
\
933
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
934
    int h=8;\
935
    asm volatile(\
936
        "pxor %%mm7, %%mm7          \n\t"\
937
        "movq %5, %%mm6             \n\t"\
938
        "1:                         \n\t"\
939
        "movq    (%0), %%mm0        \n\t"\
940
        "movq   1(%0), %%mm2        \n\t"\
941
        "movq %%mm0, %%mm1          \n\t"\
942
        "movq %%mm2, %%mm3          \n\t"\
943
        "punpcklbw %%mm7, %%mm0     \n\t"\
944
        "punpckhbw %%mm7, %%mm1     \n\t"\
945
        "punpcklbw %%mm7, %%mm2     \n\t"\
946
        "punpckhbw %%mm7, %%mm3     \n\t"\
947
        "paddw %%mm2, %%mm0         \n\t"\
948
        "paddw %%mm3, %%mm1         \n\t"\
949
        "psllw $2, %%mm0            \n\t"\
950
        "psllw $2, %%mm1            \n\t"\
951
        "movq   -1(%0), %%mm2       \n\t"\
952
        "movq    2(%0), %%mm4       \n\t"\
953
        "movq %%mm2, %%mm3          \n\t"\
954
        "movq %%mm4, %%mm5          \n\t"\
955
        "punpcklbw %%mm7, %%mm2     \n\t"\
956
        "punpckhbw %%mm7, %%mm3     \n\t"\
957
        "punpcklbw %%mm7, %%mm4     \n\t"\
958
        "punpckhbw %%mm7, %%mm5     \n\t"\
959
        "paddw %%mm4, %%mm2         \n\t"\
960
        "paddw %%mm3, %%mm5         \n\t"\
961
        "psubw %%mm2, %%mm0         \n\t"\
962
        "psubw %%mm5, %%mm1         \n\t"\
963
        "pmullw %%mm6, %%mm0        \n\t"\
964
        "pmullw %%mm6, %%mm1        \n\t"\
965
        "movd   -2(%0), %%mm2       \n\t"\
966
        "movd    7(%0), %%mm5       \n\t"\
967
        "punpcklbw %%mm7, %%mm2     \n\t"\
968
        "punpcklbw %%mm7, %%mm5     \n\t"\
969
        "paddw %%mm3, %%mm2         \n\t"\
970
        "paddw %%mm5, %%mm4         \n\t"\
971
        "movq %6, %%mm5             \n\t"\
972
        "paddw %%mm5, %%mm2         \n\t"\
973
        "paddw %%mm5, %%mm4         \n\t"\
974
        "paddw %%mm2, %%mm0         \n\t"\
975
        "paddw %%mm4, %%mm1         \n\t"\
976
        "psraw $5, %%mm0            \n\t"\
977
        "psraw $5, %%mm1            \n\t"\
978
        "packuswb %%mm1, %%mm0      \n\t"\
979
        OP(%%mm0, (%1),%%mm5, q)\
980
        "add %3, %0                 \n\t"\
981
        "add %4, %1                 \n\t"\
982
        "decl %2                    \n\t"\
983
        " jnz 1b                    \n\t"\
984
        : "+a"(src), "+c"(dst), "+m"(h)\
985
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
986
        : "memory"\
987
    );\
988
}\
989
\
990
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
991
    int h=8;\
992
    asm volatile(\
993
        "pxor %%mm7, %%mm7          \n\t"\
994
        "movq %0, %%mm6             \n\t"\
995
        :: "m"(ff_pw_5)\
996
    );\
997
    do{\
998
    asm volatile(\
999
        "movq    (%0), %%mm0        \n\t"\
1000
        "movq   1(%0), %%mm2        \n\t"\
1001
        "movq %%mm0, %%mm1          \n\t"\
1002
        "movq %%mm2, %%mm3          \n\t"\
1003
        "punpcklbw %%mm7, %%mm0     \n\t"\
1004
        "punpckhbw %%mm7, %%mm1     \n\t"\
1005
        "punpcklbw %%mm7, %%mm2     \n\t"\
1006
        "punpckhbw %%mm7, %%mm3     \n\t"\
1007
        "paddw %%mm2, %%mm0         \n\t"\
1008
        "paddw %%mm3, %%mm1         \n\t"\
1009
        "psllw $2, %%mm0            \n\t"\
1010
        "psllw $2, %%mm1            \n\t"\
1011
        "movq   -1(%0), %%mm2       \n\t"\
1012
        "movq    2(%0), %%mm4       \n\t"\
1013
        "movq %%mm2, %%mm3          \n\t"\
1014
        "movq %%mm4, %%mm5          \n\t"\
1015
        "punpcklbw %%mm7, %%mm2     \n\t"\
1016
        "punpckhbw %%mm7, %%mm3     \n\t"\
1017
        "punpcklbw %%mm7, %%mm4     \n\t"\
1018
        "punpckhbw %%mm7, %%mm5     \n\t"\
1019
        "paddw %%mm4, %%mm2         \n\t"\
1020
        "paddw %%mm3, %%mm5         \n\t"\
1021
        "psubw %%mm2, %%mm0         \n\t"\
1022
        "psubw %%mm5, %%mm1         \n\t"\
1023
        "pmullw %%mm6, %%mm0        \n\t"\
1024
        "pmullw %%mm6, %%mm1        \n\t"\
1025
        "movd   -2(%0), %%mm2       \n\t"\
1026
        "movd    7(%0), %%mm5       \n\t"\
1027
        "punpcklbw %%mm7, %%mm2     \n\t"\
1028
        "punpcklbw %%mm7, %%mm5     \n\t"\
1029
        "paddw %%mm3, %%mm2         \n\t"\
1030
        "paddw %%mm5, %%mm4         \n\t"\
1031
        "movq %5, %%mm5             \n\t"\
1032
        "paddw %%mm5, %%mm2         \n\t"\
1033
        "paddw %%mm5, %%mm4         \n\t"\
1034
        "paddw %%mm2, %%mm0         \n\t"\
1035
        "paddw %%mm4, %%mm1         \n\t"\
1036
        "psraw $5, %%mm0            \n\t"\
1037
        "psraw $5, %%mm1            \n\t"\
1038
        "movq (%2), %%mm4           \n\t"\
1039
        "packuswb %%mm1, %%mm0      \n\t"\
1040
        PAVGB" %%mm4, %%mm0         \n\t"\
1041
        OP(%%mm0, (%1),%%mm5, q)\
1042
        "add %4, %0                 \n\t"\
1043
        "add %4, %1                 \n\t"\
1044
        "add %3, %2                 \n\t"\
1045
        : "+a"(src), "+c"(dst), "+d"(src2)\
1046
        : "D"((long)src2Stride), "S"((long)dstStride),\
1047
          "m"(ff_pw_16)\
1048
        : "memory"\
1049
    );\
1050
    }while(--h);\
1051
}\
1052
\
1053
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1054
    int w= 2;\
1055
    src -= 2*srcStride;\
1056
    \
1057
    while(w--){\
1058
      asm volatile(\
1059
        "pxor %%mm7, %%mm7          \n\t"\
1060
        "movd (%0), %%mm0           \n\t"\
1061
        "add %2, %0                 \n\t"\
1062
        "movd (%0), %%mm1           \n\t"\
1063
        "add %2, %0                 \n\t"\
1064
        "movd (%0), %%mm2           \n\t"\
1065
        "add %2, %0                 \n\t"\
1066
        "movd (%0), %%mm3           \n\t"\
1067
        "add %2, %0                 \n\t"\
1068
        "movd (%0), %%mm4           \n\t"\
1069
        "add %2, %0                 \n\t"\
1070
        "punpcklbw %%mm7, %%mm0     \n\t"\
1071
        "punpcklbw %%mm7, %%mm1     \n\t"\
1072
        "punpcklbw %%mm7, %%mm2     \n\t"\
1073
        "punpcklbw %%mm7, %%mm3     \n\t"\
1074
        "punpcklbw %%mm7, %%mm4     \n\t"\
1075
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1076
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1077
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1078
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1079
        QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1080
        QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1081
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1082
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1083
         \
1084
        : "+a"(src), "+c"(dst)\
1085
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1086
        : "memory"\
1087
     );\
1088
     if(h==16){\
1089
        asm volatile(\
1090
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1091
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1092
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1093
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1094
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1095
            QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1096
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1097
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1098
            \
1099
           : "+a"(src), "+c"(dst)\
1100
           : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1101
           : "memory"\
1102
        );\
1103
     }\
1104
     src += 4-(h+5)*srcStride;\
1105
     dst += 4-h*dstStride;\
1106
   }\
1107
}\
1108
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){\
1109
    int h = size;\
1110
    int w = (size+8)>>2;\
1111
    src -= 2*srcStride+2;\
1112
    while(w--){\
1113
        asm volatile(\
1114
            "pxor %%mm7, %%mm7      \n\t"\
1115
            "movd (%0), %%mm0       \n\t"\
1116
            "add %2, %0             \n\t"\
1117
            "movd (%0), %%mm1       \n\t"\
1118
            "add %2, %0             \n\t"\
1119
            "movd (%0), %%mm2       \n\t"\
1120
            "add %2, %0             \n\t"\
1121
            "movd (%0), %%mm3       \n\t"\
1122
            "add %2, %0             \n\t"\
1123
            "movd (%0), %%mm4       \n\t"\
1124
            "add %2, %0             \n\t"\
1125
            "punpcklbw %%mm7, %%mm0 \n\t"\
1126
            "punpcklbw %%mm7, %%mm1 \n\t"\
1127
            "punpcklbw %%mm7, %%mm2 \n\t"\
1128
            "punpcklbw %%mm7, %%mm3 \n\t"\
1129
            "punpcklbw %%mm7, %%mm4 \n\t"\
1130
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
1131
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
1132
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
1133
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
1134
            QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
1135
            QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
1136
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
1137
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
1138
            : "+a"(src)\
1139
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1140
            : "memory"\
1141
        );\
1142
        if(size==16){\
1143
            asm volatile(\
1144
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
1145
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
1146
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
1147
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
1148
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
1149
                QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
1150
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
1151
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
1152
                : "+a"(src)\
1153
                : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1154
                : "memory"\
1155
            );\
1156
        }\
1157
        tmp += 4;\
1158
        src += 4 - (size+5)*srcStride;\
1159
    }\
1160
    tmp -= size+8;\
1161
    w = size>>4;\
1162
    do{\
1163
    h = size;\
1164
    asm volatile(\
1165
        "1:                         \n\t"\
1166
        "movq     (%0), %%mm0       \n\t"\
1167
        "movq    8(%0), %%mm3       \n\t"\
1168
        "movq    2(%0), %%mm1       \n\t"\
1169
        "movq   10(%0), %%mm4       \n\t"\
1170
        "paddw   %%mm4, %%mm0       \n\t"\
1171
        "paddw   %%mm3, %%mm1       \n\t"\
1172
        "paddw  18(%0), %%mm3       \n\t"\
1173
        "paddw  16(%0), %%mm4       \n\t"\
1174
        "movq    4(%0), %%mm2       \n\t"\
1175
        "movq   12(%0), %%mm5       \n\t"\
1176
        "paddw   6(%0), %%mm2       \n\t"\
1177
        "paddw  14(%0), %%mm5       \n\t"\
1178
        "psubw %%mm1, %%mm0         \n\t"\
1179
        "psubw %%mm4, %%mm3         \n\t"\
1180
        "psraw $2, %%mm0            \n\t"\
1181
        "psraw $2, %%mm3            \n\t"\
1182
        "psubw %%mm1, %%mm0         \n\t"\
1183
        "psubw %%mm4, %%mm3         \n\t"\
1184
        "paddsw %%mm2, %%mm0        \n\t"\
1185
        "paddsw %%mm5, %%mm3        \n\t"\
1186
        "psraw $2, %%mm0            \n\t"\
1187
        "psraw $2, %%mm3            \n\t"\
1188
        "paddw %%mm2, %%mm0         \n\t"\
1189
        "paddw %%mm5, %%mm3         \n\t"\
1190
        "psraw $6, %%mm0            \n\t"\
1191
        "psraw $6, %%mm3            \n\t"\
1192
        "packuswb %%mm3, %%mm0      \n\t"\
1193
        OP(%%mm0, (%1),%%mm7, q)\
1194
        "add $48, %0                \n\t"\
1195
        "add %3, %1                 \n\t"\
1196
        "decl %2                    \n\t"\
1197
        " jnz 1b                    \n\t"\
1198
        : "+a"(tmp), "+c"(dst), "+m"(h)\
1199
        : "S"((long)dstStride)\
1200
        : "memory"\
1201
    );\
1202
    tmp += 8 - size*24;\
1203
    dst += 8 - size*dstStride;\
1204
    }while(w--);\
1205
}\
1206
\
1207
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1208
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1209
}\
1210
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1211
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1212
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1213
}\
1214
\
1215
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1216
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1217
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1218
    src += 8*srcStride;\
1219
    dst += 8*dstStride;\
1220
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1221
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1222
}\
1223
\
1224
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1225
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1226
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1227
    src += 8*dstStride;\
1228
    dst += 8*dstStride;\
1229
    src2 += 8*src2Stride;\
1230
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1231
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1232
}\
1233
\
1234
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1235
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
1236
}\
1237
\
1238
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1239
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
1240
}\
1241
\
1242
static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1243
{\
1244
    asm volatile(\
1245
        "movq      (%1), %%mm0          \n\t"\
1246
        "movq    24(%1), %%mm1          \n\t"\
1247
        "psraw      $5,  %%mm0          \n\t"\
1248
        "psraw      $5,  %%mm1          \n\t"\
1249
        "packuswb %%mm0, %%mm0          \n\t"\
1250
        "packuswb %%mm1, %%mm1          \n\t"\
1251
        PAVGB"     (%0), %%mm0          \n\t"\
1252
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1253
        OP(%%mm0, (%2),    %%mm4, d)\
1254
        OP(%%mm1, (%2,%4), %%mm5, d)\
1255
        "lea  (%0,%3,2), %0             \n\t"\
1256
        "lea  (%2,%4,2), %2             \n\t"\
1257
        "movq    48(%1), %%mm0          \n\t"\
1258
        "movq    72(%1), %%mm1          \n\t"\
1259
        "psraw      $5,  %%mm0          \n\t"\
1260
        "psraw      $5,  %%mm1          \n\t"\
1261
        "packuswb %%mm0, %%mm0          \n\t"\
1262
        "packuswb %%mm1, %%mm1          \n\t"\
1263
        PAVGB"     (%0), %%mm0          \n\t"\
1264
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1265
        OP(%%mm0, (%2),    %%mm4, d)\
1266
        OP(%%mm1, (%2,%4), %%mm5, d)\
1267
        :"+a"(src8), "+c"(src16), "+d"(dst)\
1268
        :"S"((long)src8Stride), "D"((long)dstStride)\
1269
        :"memory");\
1270
}\
1271
static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1272
{\
1273
    while(h--){\
1274
    asm volatile(\
1275
        "movq      (%1), %%mm0          \n\t"\
1276
        "movq     8(%1), %%mm1          \n\t"\
1277
        "psraw      $5,  %%mm0          \n\t"\
1278
        "psraw      $5,  %%mm1          \n\t"\
1279
        "packuswb %%mm1, %%mm0          \n\t"\
1280
        PAVGB"     (%0), %%mm0          \n\t"\
1281
        OP(%%mm0, (%2), %%mm5, q)\
1282
        ::"a"(src8), "c"(src16), "d"(dst)\
1283
        :"memory");\
1284
        src8 += src8Stride;\
1285
        src16 += 24;\
1286
        dst += dstStride;\
1287
    }\
1288
}\
1289
static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1290
{\
1291
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
1292
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1293
}\
1294

    
1295

    
1296
#define H264_MC(OPNAME, SIZE, MMX) \
1297
static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1298
    OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1299
}\
1300
\
1301
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1302
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1303
}\
1304
\
1305
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1306
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1307
}\
1308
\
1309
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1310
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1311
}\
1312
\
1313
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1314
    uint64_t temp[SIZE*SIZE/8];\
1315
    uint8_t * const half= (uint8_t*)temp;\
1316
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
1317
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
1318
}\
1319
\
1320
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1321
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1322
}\
1323
\
1324
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1325
    uint64_t temp[SIZE*SIZE/8];\
1326
    uint8_t * const half= (uint8_t*)temp;\
1327
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
1328
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
1329
}\
1330
\
1331
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1332
    uint64_t temp[SIZE*SIZE/8];\
1333
    uint8_t * const halfV= (uint8_t*)temp;\
1334
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
1335
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
1336
}\
1337
\
1338
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1339
    uint64_t temp[SIZE*SIZE/8];\
1340
    uint8_t * const halfV= (uint8_t*)temp;\
1341
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
1342
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
1343
}\
1344
\
1345
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1346
    uint64_t temp[SIZE*SIZE/8];\
1347
    uint8_t * const halfV= (uint8_t*)temp;\
1348
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
1349
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
1350
}\
1351
\
1352
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1353
    uint64_t temp[SIZE*SIZE/8];\
1354
    uint8_t * const halfV= (uint8_t*)temp;\
1355
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
1356
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
1357
}\
1358
\
1359
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1360
    uint64_t temp[SIZE*(SIZE<8?12:24)/4];\
1361
    int16_t * const tmp= (int16_t*)temp;\
1362
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
1363
}\
1364
\
1365
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1366
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1367
    uint8_t * const halfHV= (uint8_t*)temp;\
1368
    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
1369
    assert(((int)temp & 7) == 0);\
1370
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
1371
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1372
}\
1373
\
1374
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1375
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1376
    uint8_t * const halfHV= (uint8_t*)temp;\
1377
    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
1378
    assert(((int)temp & 7) == 0);\
1379
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
1380
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1381
}\
1382
\
1383
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1384
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1385
    int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
1386
    uint8_t * const halfHV= ((uint8_t*)temp);\
1387
    assert(((int)temp & 7) == 0);\
1388
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1389
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1390
}\
1391
\
1392
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1393
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1394
    int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
1395
    uint8_t * const halfHV= ((uint8_t*)temp);\
1396
    assert(((int)temp & 7) == 0);\
1397
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1398
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1399
}\
1400

    
1401

    
1402
#define AVG_3DNOW_OP(a,b,temp, size) \
1403
"mov" #size " " #b ", " #temp "   \n\t"\
1404
"pavgusb " #temp ", " #a "        \n\t"\
1405
"mov" #size " " #a ", " #b "      \n\t"
1406
#define AVG_MMX2_OP(a,b,temp, size) \
1407
"mov" #size " " #b ", " #temp "   \n\t"\
1408
"pavgb " #temp ", " #a "          \n\t"\
1409
"mov" #size " " #a ", " #b "      \n\t"
1410

    
1411
#define PAVGB "pavgusb"
1412
QPEL_H264(put_,       PUT_OP, 3dnow)
1413
QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1414
#undef PAVGB
1415
#define PAVGB "pavgb"
1416
QPEL_H264(put_,       PUT_OP, mmx2)
1417
QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
1418
#undef PAVGB
1419

    
1420
H264_MC(put_, 4, 3dnow)
1421
H264_MC(put_, 8, 3dnow)
1422
H264_MC(put_, 16,3dnow)
1423
H264_MC(avg_, 4, 3dnow)
1424
H264_MC(avg_, 8, 3dnow)
1425
H264_MC(avg_, 16,3dnow)
1426
H264_MC(put_, 4, mmx2)
1427
H264_MC(put_, 8, mmx2)
1428
H264_MC(put_, 16,mmx2)
1429
H264_MC(avg_, 4, mmx2)
1430
H264_MC(avg_, 8, mmx2)
1431
H264_MC(avg_, 16,mmx2)
1432

    
1433

    
1434
#define H264_CHROMA_OP(S,D)
1435
#define H264_CHROMA_OP4(S,D,T)
1436
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
1437
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
1438
#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
1439
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1440
#include "dsputil_h264_template_mmx.c"
1441

    
1442
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)
1443
{
1444
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 1);
1445
}
1446
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)
1447
{
1448
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 0);
1449
}
1450

    
1451
#undef H264_CHROMA_OP
1452
#undef H264_CHROMA_OP4
1453
#undef H264_CHROMA_MC8_TMPL
1454
#undef H264_CHROMA_MC4_TMPL
1455
#undef H264_CHROMA_MC2_TMPL
1456
#undef H264_CHROMA_MC8_MV0
1457

    
1458
#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
1459
#define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\
1460
                               "pavgb " #T ", " #D " \n\t"
1461
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
1462
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
1463
#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
1464
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
1465
#include "dsputil_h264_template_mmx.c"
1466
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)
1467
{
1468
    avg_h264_chroma_mc8_mmx2(dst, src, stride, h, x, y, 1);
1469
}
1470
#undef H264_CHROMA_OP
1471
#undef H264_CHROMA_OP4
1472
#undef H264_CHROMA_MC8_TMPL
1473
#undef H264_CHROMA_MC4_TMPL
1474
#undef H264_CHROMA_MC2_TMPL
1475
#undef H264_CHROMA_MC8_MV0
1476

    
1477
#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
1478
#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
1479
                               "pavgusb " #T ", " #D " \n\t"
1480
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
1481
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
1482
#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
1483
#include "dsputil_h264_template_mmx.c"
1484
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)
1485
{
1486
    avg_h264_chroma_mc8_3dnow(dst, src, stride, h, x, y, 1);
1487
}
1488
#undef H264_CHROMA_OP
1489
#undef H264_CHROMA_OP4
1490
#undef H264_CHROMA_MC8_TMPL
1491
#undef H264_CHROMA_MC4_TMPL
1492
#undef H264_CHROMA_MC8_MV0
1493

    
1494
/***********************************/
1495
/* weighted prediction */
1496

    
1497
static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
1498
{
1499
    int x, y;
1500
    offset <<= log2_denom;
1501
    offset += (1 << log2_denom) >> 1;
1502
    asm volatile(
1503
        "movd    %0, %%mm4        \n\t"
1504
        "movd    %1, %%mm5        \n\t"
1505
        "movd    %2, %%mm6        \n\t"
1506
        "pshufw  $0, %%mm4, %%mm4 \n\t"
1507
        "pshufw  $0, %%mm5, %%mm5 \n\t"
1508
        "pxor    %%mm7, %%mm7     \n\t"
1509
        :: "g"(weight), "g"(offset), "g"(log2_denom)
1510
    );
1511
    for(y=0; y<h; y+=2){
1512
        for(x=0; x<w; x+=4){
1513
            asm volatile(
1514
                "movd      %0,    %%mm0 \n\t"
1515
                "movd      %1,    %%mm1 \n\t"
1516
                "punpcklbw %%mm7, %%mm0 \n\t"
1517
                "punpcklbw %%mm7, %%mm1 \n\t"
1518
                "pmullw    %%mm4, %%mm0 \n\t"
1519
                "pmullw    %%mm4, %%mm1 \n\t"
1520
                "paddsw    %%mm5, %%mm0 \n\t"
1521
                "paddsw    %%mm5, %%mm1 \n\t"
1522
                "psraw     %%mm6, %%mm0 \n\t"
1523
                "psraw     %%mm6, %%mm1 \n\t"
1524
                "packuswb  %%mm7, %%mm0 \n\t"
1525
                "packuswb  %%mm7, %%mm1 \n\t"
1526
                "movd      %%mm0, %0    \n\t"
1527
                "movd      %%mm1, %1    \n\t"
1528
                : "+m"(*(uint32_t*)(dst+x)),
1529
                  "+m"(*(uint32_t*)(dst+x+stride))
1530
            );
1531
        }
1532
        dst += 2*stride;
1533
    }
1534
}
1535

    
1536
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)
1537
{
1538
    int x, y;
1539
    offset = ((offset + 1) | 1) << log2_denom;
1540
    asm volatile(
1541
        "movd    %0, %%mm3        \n\t"
1542
        "movd    %1, %%mm4        \n\t"
1543
        "movd    %2, %%mm5        \n\t"
1544
        "movd    %3, %%mm6        \n\t"
1545
        "pshufw  $0, %%mm3, %%mm3 \n\t"
1546
        "pshufw  $0, %%mm4, %%mm4 \n\t"
1547
        "pshufw  $0, %%mm5, %%mm5 \n\t"
1548
        "pxor    %%mm7, %%mm7     \n\t"
1549
        :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
1550
    );
1551
    for(y=0; y<h; y++){
1552
        for(x=0; x<w; x+=4){
1553
            asm volatile(
1554
                "movd      %0,    %%mm0 \n\t"
1555
                "movd      %1,    %%mm1 \n\t"
1556
                "punpcklbw %%mm7, %%mm0 \n\t"
1557
                "punpcklbw %%mm7, %%mm1 \n\t"
1558
                "pmullw    %%mm3, %%mm0 \n\t"
1559
                "pmullw    %%mm4, %%mm1 \n\t"
1560
                "paddsw    %%mm1, %%mm0 \n\t"
1561
                "paddsw    %%mm5, %%mm0 \n\t"
1562
                "psraw     %%mm6, %%mm0 \n\t"
1563
                "packuswb  %%mm0, %%mm0 \n\t"
1564
                "movd      %%mm0, %0    \n\t"
1565
                : "+m"(*(uint32_t*)(dst+x))
1566
                :  "m"(*(uint32_t*)(src+x))
1567
            );
1568
        }
1569
        src += stride;
1570
        dst += stride;
1571
    }
1572
}
1573

    
1574
#define H264_WEIGHT(W,H) \
1575
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){ \
1576
    ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
1577
} \
1578
static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
1579
    ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
1580
}
1581

    
1582
H264_WEIGHT(16,16)
1583
H264_WEIGHT(16, 8)
1584
H264_WEIGHT( 8,16)
1585
H264_WEIGHT( 8, 8)
1586
H264_WEIGHT( 8, 4)
1587
H264_WEIGHT( 4, 8)
1588
H264_WEIGHT( 4, 4)
1589
H264_WEIGHT( 4, 2)
1590