Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / h264dsp_mmx.c @ be449fca

History | View | Annotate | Download (84.6 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

    
380
/***********************************/
381
/* deblocking */
382

    
383
// out: o = |x-y|>a
384
// clobbers: t
385
#define DIFF_GT_MMX(x,y,a,o,t)\
386
    "movq     "#y", "#t"  \n\t"\
387
    "movq     "#x", "#o"  \n\t"\
388
    "psubusb  "#x", "#t"  \n\t"\
389
    "psubusb  "#y", "#o"  \n\t"\
390
    "por      "#t", "#o"  \n\t"\
391
    "psubusb  "#a", "#o"  \n\t"
392

    
393
// out: o = |x-y|>a
394
// clobbers: t
395
#define DIFF_GT2_MMX(x,y,a,o,t)\
396
    "movq     "#y", "#t"  \n\t"\
397
    "movq     "#x", "#o"  \n\t"\
398
    "psubusb  "#x", "#t"  \n\t"\
399
    "psubusb  "#y", "#o"  \n\t"\
400
    "psubusb  "#a", "#t"  \n\t"\
401
    "psubusb  "#a", "#o"  \n\t"\
402
    "pcmpeqb  "#t", "#o"  \n\t"\
403

    
404
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
405
// out: mm5=beta-1, mm7=mask
406
// clobbers: mm4,mm6
407
#define H264_DEBLOCK_MASK(alpha1, beta1) \
408
    "pshufw $0, "#alpha1", %%mm4 \n\t"\
409
    "pshufw $0, "#beta1 ", %%mm5 \n\t"\
410
    "packuswb  %%mm4, %%mm4      \n\t"\
411
    "packuswb  %%mm5, %%mm5      \n\t"\
412
    DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
413
    DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
414
    "por       %%mm4, %%mm7      \n\t"\
415
    DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
416
    "por       %%mm4, %%mm7      \n\t"\
417
    "pxor      %%mm6, %%mm6      \n\t"\
418
    "pcmpeqb   %%mm6, %%mm7      \n\t"
419

    
420
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
421
// out: mm1=p0' mm2=q0'
422
// clobbers: mm0,3-6
423
#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
424
        "movq    %%mm1              , %%mm5 \n\t"\
425
        "pxor    %%mm2              , %%mm5 \n\t" /* p0^q0*/\
426
        "pand    "#pb_01"           , %%mm5 \n\t" /* (p0^q0)&1*/\
427
        "pcmpeqb %%mm4              , %%mm4 \n\t"\
428
        "pxor    %%mm4              , %%mm3 \n\t"\
429
        "pavgb   %%mm0              , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
430
        "pavgb   "MANGLE(ff_pb_3)"  , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
431
        "pxor    %%mm1              , %%mm4 \n\t"\
432
        "pavgb   %%mm2              , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
433
        "pavgb   %%mm5              , %%mm3 \n\t"\
434
        "paddusb %%mm4              , %%mm3 \n\t" /* d+128+33*/\
435
        "movq    "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
436
        "psubusb %%mm3              , %%mm6 \n\t"\
437
        "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
438
        "pminub  %%mm7              , %%mm6 \n\t"\
439
        "pminub  %%mm7              , %%mm3 \n\t"\
440
        "psubusb %%mm6              , %%mm1 \n\t"\
441
        "psubusb %%mm3              , %%mm2 \n\t"\
442
        "paddusb %%mm3              , %%mm1 \n\t"\
443
        "paddusb %%mm6              , %%mm2 \n\t"
444

    
445
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
446
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
447
// clobbers: q2, tmp, tc0
448
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
449
        "movq     %%mm1,  "#tmp"   \n\t"\
450
        "pavgb    %%mm2,  "#tmp"   \n\t"\
451
        "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
452
        "pxor   "q2addr", "#tmp"   \n\t"\
453
        "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
454
        "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
455
        "movq     "#p1",  "#tmp"   \n\t"\
456
        "psubusb  "#tc0", "#tmp"   \n\t"\
457
        "paddusb  "#p1",  "#tc0"   \n\t"\
458
        "pmaxub   "#tmp", "#q2"    \n\t"\
459
        "pminub   "#tc0", "#q2"    \n\t"\
460
        "movq     "#q2",  "q1addr" \n\t"
461

    
462
static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
463
{
464
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);
465

    
466
    __asm__ volatile(
467
        "movq    (%1,%3), %%mm0    \n\t" //p1
468
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
469
        "movq    (%2),    %%mm2    \n\t" //q0
470
        "movq    (%2,%3), %%mm3    \n\t" //q1
471
        H264_DEBLOCK_MASK(%6, %7)
472

    
473
        "movd      %5,    %%mm4    \n\t"
474
        "punpcklbw %%mm4, %%mm4    \n\t"
475
        "punpcklwd %%mm4, %%mm4    \n\t"
476
        "pcmpeqb   %%mm3, %%mm3    \n\t"
477
        "movq      %%mm4, %%mm6    \n\t"
478
        "pcmpgtb   %%mm3, %%mm4    \n\t"
479
        "movq      %%mm6, 8+%0     \n\t"
480
        "pand      %%mm4, %%mm7    \n\t"
481
        "movq      %%mm7, %0       \n\t"
482

    
483
        /* filter p1 */
484
        "movq     (%1),   %%mm3    \n\t" //p2
485
        DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
486
        "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
487
        "pand     8+%0,   %%mm7    \n\t" // mask & tc0
488
        "movq     %%mm7,  %%mm4    \n\t"
489
        "psubb    %%mm6,  %%mm7    \n\t"
490
        "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
491
        H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
492

    
493
        /* filter q1 */
494
        "movq    (%2,%3,2), %%mm4  \n\t" //q2
495
        DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
496
        "pand     %0,     %%mm6    \n\t"
497
        "movq     8+%0,   %%mm5    \n\t" // can be merged with the and below but is slower then
498
        "pand     %%mm6,  %%mm5    \n\t"
499
        "psubb    %%mm6,  %%mm7    \n\t"
500
        "movq    (%2,%3), %%mm3    \n\t"
501
        H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
502

    
503
        /* filter p0, q0 */
504
        H264_DEBLOCK_P0_Q0(%8, unused)
505
        "movq      %%mm1, (%1,%3,2) \n\t"
506
        "movq      %%mm2, (%2)      \n\t"
507

    
508
        : "=m"(*tmp0)
509
        : "r"(pix-3*stride), "r"(pix), "r"((x86_reg)stride),
510
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
511
          "m"(ff_bone)
512
    );
513
}
514

    
515
static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
516
{
517
    if((tc0[0] & tc0[1]) >= 0)
518
        h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
519
    if((tc0[2] & tc0[3]) >= 0)
520
        h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
521
}
522
static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
523
{
524
    //FIXME: could cut some load/stores by merging transpose with filter
525
    // also, it only needs to transpose 6x8
526
    DECLARE_ALIGNED_8(uint8_t, trans[8*8]);
527
    int i;
528
    for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
529
        if((tc0[0] & tc0[1]) < 0)
530
            continue;
531
        transpose4x4(trans,       pix-4,          8, stride);
532
        transpose4x4(trans  +4*8, pix,            8, stride);
533
        transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
534
        transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
535
        h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
536
        transpose4x4(pix-2,          trans  +2*8, stride, 8);
537
        transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
538
    }
539
}
540

    
541
static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
542
{
543
    __asm__ volatile(
544
        "movq    (%0),    %%mm0     \n\t" //p1
545
        "movq    (%0,%2), %%mm1     \n\t" //p0
546
        "movq    (%1),    %%mm2     \n\t" //q0
547
        "movq    (%1,%2), %%mm3     \n\t" //q1
548
        H264_DEBLOCK_MASK(%4, %5)
549
        "movd      %3,    %%mm6     \n\t"
550
        "punpcklbw %%mm6, %%mm6     \n\t"
551
        "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
552
        H264_DEBLOCK_P0_Q0(%6, %7)
553
        "movq      %%mm1, (%0,%2)   \n\t"
554
        "movq      %%mm2, (%1)      \n\t"
555

    
556
        :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
557
           "r"(*(uint32_t*)tc0),
558
           "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
559
    );
560
}
561

    
562
static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
563
{
564
    h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
565
}
566

    
567
static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
568
{
569
    //FIXME: could cut some load/stores by merging transpose with filter
570
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
571
    transpose4x4(trans, pix-2, 8, stride);
572
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
573
    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
574
    transpose4x4(pix-2, trans, stride, 8);
575
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
576
}
577

    
578
// p0 = (p0 + q1 + 2*p1 + 2) >> 2
579
#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
580
    "movq    "#p0", %%mm4  \n\t"\
581
    "pxor    "#q1", %%mm4  \n\t"\
582
    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
583
    "pavgb   "#q1", "#p0"  \n\t"\
584
    "psubusb %%mm4, "#p0"  \n\t"\
585
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
586

    
587
static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
588
{
589
    __asm__ volatile(
590
        "movq    (%0),    %%mm0     \n\t"
591
        "movq    (%0,%2), %%mm1     \n\t"
592
        "movq    (%1),    %%mm2     \n\t"
593
        "movq    (%1,%2), %%mm3     \n\t"
594
        H264_DEBLOCK_MASK(%3, %4)
595
        "movq    %%mm1,   %%mm5     \n\t"
596
        "movq    %%mm2,   %%mm6     \n\t"
597
        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
598
        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
599
        "psubb   %%mm5,   %%mm1     \n\t"
600
        "psubb   %%mm6,   %%mm2     \n\t"
601
        "pand    %%mm7,   %%mm1     \n\t"
602
        "pand    %%mm7,   %%mm2     \n\t"
603
        "paddb   %%mm5,   %%mm1     \n\t"
604
        "paddb   %%mm6,   %%mm2     \n\t"
605
        "movq    %%mm1,   (%0,%2)   \n\t"
606
        "movq    %%mm2,   (%1)      \n\t"
607
        :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
608
           "m"(alpha1), "m"(beta1), "m"(ff_bone)
609
    );
610
}
611

    
612
static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
613
{
614
    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
615
}
616

    
617
static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
618
{
619
    //FIXME: could cut some load/stores by merging transpose with filter
620
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
621
    transpose4x4(trans, pix-2, 8, stride);
622
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
623
    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
624
    transpose4x4(pix-2, trans, stride, 8);
625
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
626
}
627

    
628
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],
629
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
630
    int dir;
631
    __asm__ volatile(
632
        "pxor %%mm7, %%mm7 \n\t"
633
        "movq %0, %%mm6 \n\t"
634
        "movq %1, %%mm5 \n\t"
635
        "movq %2, %%mm4 \n\t"
636
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
637
    );
638
    if(field)
639
        __asm__ volatile(
640
            "movq %0, %%mm5 \n\t"
641
            "movq %1, %%mm4 \n\t"
642
            ::"m"(ff_pb_3_1), "m"(ff_pb_7_3)
643
        );
644

    
645
    // could do a special case for dir==0 && edges==1, but it only reduces the
646
    // average filter time by 1.2%
647
    for( dir=1; dir>=0; dir-- ) {
648
        const int d_idx = dir ? -8 : -1;
649
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
650
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
651
        int b_idx, edge, l;
652
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
653
            __asm__ volatile(
654
                "pand %0, %%mm0 \n\t"
655
                ::"m"(mask_dir)
656
            );
657
            if(!(mask_mv & edge)) {
658
                __asm__ volatile("pxor %%mm0, %%mm0 \n\t":);
659
                for( l = bidir; l >= 0; l-- ) {
660
                    __asm__ volatile(
661
                        "movd %0, %%mm1 \n\t"
662
                        "punpckldq %1, %%mm1 \n\t"
663
                        "movq %%mm1, %%mm2 \n\t"
664
                        "psrlw $7, %%mm2 \n\t"
665
                        "pand %%mm6, %%mm2 \n\t"
666
                        "por %%mm2, %%mm1 \n\t" // ref_cache with -2 mapped to -1
667
                        "punpckldq %%mm1, %%mm2 \n\t"
668
                        "pcmpeqb %%mm2, %%mm1 \n\t"
669
                        "paddb %%mm6, %%mm1 \n\t"
670
                        "punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn]
671
                        "por %%mm1, %%mm0 \n\t"
672

    
673
                        "movq %2, %%mm1 \n\t"
674
                        "movq %3, %%mm2 \n\t"
675
                        "psubw %4, %%mm1 \n\t"
676
                        "psubw %5, %%mm2 \n\t"
677
                        "packsswb %%mm2, %%mm1 \n\t"
678
                        "paddb %%mm5, %%mm1 \n\t"
679
                        "pminub %%mm4, %%mm1 \n\t"
680
                        "pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit
681
                        "por %%mm1, %%mm0 \n\t"
682
                        ::"m"(ref[l][b_idx]),
683
                          "m"(ref[l][b_idx+d_idx]),
684
                          "m"(mv[l][b_idx][0]),
685
                          "m"(mv[l][b_idx+2][0]),
686
                          "m"(mv[l][b_idx+d_idx][0]),
687
                          "m"(mv[l][b_idx+d_idx+2][0])
688
                    );
689
                }
690
            }
691
            __asm__ volatile(
692
                "movd %0, %%mm1 \n\t"
693
                "por  %1, %%mm1 \n\t"
694
                "punpcklbw %%mm7, %%mm1 \n\t"
695
                "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]
696
                ::"m"(nnz[b_idx]),
697
                  "m"(nnz[b_idx+d_idx])
698
            );
699
            __asm__ volatile(
700
                "pcmpeqw %%mm7, %%mm0 \n\t"
701
                "pcmpeqw %%mm7, %%mm0 \n\t"
702
                "psrlw $15, %%mm0 \n\t" // nonzero -> 1
703
                "psrlw $14, %%mm1 \n\t"
704
                "movq %%mm0, %%mm2 \n\t"
705
                "por %%mm1, %%mm2 \n\t"
706
                "psrlw $1, %%mm1 \n\t"
707
                "pandn %%mm2, %%mm1 \n\t"
708
                "movq %%mm1, %0 \n\t"
709
                :"=m"(*bS[dir][edge])
710
                ::"memory"
711
            );
712
        }
713
        edges = 4;
714
        step = 1;
715
    }
716
    __asm__ volatile(
717
        "movq   (%0), %%mm0 \n\t"
718
        "movq  8(%0), %%mm1 \n\t"
719
        "movq 16(%0), %%mm2 \n\t"
720
        "movq 24(%0), %%mm3 \n\t"
721
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
722
        "movq %%mm0,   (%0) \n\t"
723
        "movq %%mm3,  8(%0) \n\t"
724
        "movq %%mm4, 16(%0) \n\t"
725
        "movq %%mm2, 24(%0) \n\t"
726
        ::"r"(bS[0])
727
        :"memory"
728
    );
729
}
730

    
731
/***********************************/
732
/* motion compensation */
733

    
734
#define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
735
        "mov"#q" "#C", "#T"         \n\t"\
736
        "mov"#d" (%0), "#F"         \n\t"\
737
        "paddw "#D", "#T"           \n\t"\
738
        "psllw $2, "#T"             \n\t"\
739
        "psubw "#B", "#T"           \n\t"\
740
        "psubw "#E", "#T"           \n\t"\
741
        "punpcklbw "#Z", "#F"       \n\t"\
742
        "pmullw %4, "#T"            \n\t"\
743
        "paddw %5, "#A"             \n\t"\
744
        "add %2, %0                 \n\t"\
745
        "paddw "#F", "#A"           \n\t"\
746
        "paddw "#A", "#T"           \n\t"\
747
        "psraw $5, "#T"             \n\t"\
748
        "packuswb "#T", "#T"        \n\t"\
749
        OP(T, (%1), A, d)\
750
        "add %3, %1                 \n\t"
751

    
752
#define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
753
        "mov"#q" "#C", "#T"         \n\t"\
754
        "mov"#d" (%0), "#F"         \n\t"\
755
        "paddw "#D", "#T"           \n\t"\
756
        "psllw $2, "#T"             \n\t"\
757
        "paddw %4, "#A"             \n\t"\
758
        "psubw "#B", "#T"           \n\t"\
759
        "psubw "#E", "#T"           \n\t"\
760
        "punpcklbw "#Z", "#F"       \n\t"\
761
        "pmullw %3, "#T"            \n\t"\
762
        "paddw "#F", "#A"           \n\t"\
763
        "add %2, %0                 \n\t"\
764
        "paddw "#A", "#T"           \n\t"\
765
        "mov"#q" "#T", "#OF"(%1)    \n\t"
766

    
767
#define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
768
#define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
769
#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)
770
#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)
771

    
772

    
773
#define QPEL_H264(OPNAME, OP, MMX)\
774
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
775
    int h=4;\
776
\
777
    __asm__ volatile(\
778
        "pxor %%mm7, %%mm7          \n\t"\
779
        "movq %5, %%mm4             \n\t"\
780
        "movq %6, %%mm5             \n\t"\
781
        "1:                         \n\t"\
782
        "movd  -1(%0), %%mm1        \n\t"\
783
        "movd    (%0), %%mm2        \n\t"\
784
        "movd   1(%0), %%mm3        \n\t"\
785
        "movd   2(%0), %%mm0        \n\t"\
786
        "punpcklbw %%mm7, %%mm1     \n\t"\
787
        "punpcklbw %%mm7, %%mm2     \n\t"\
788
        "punpcklbw %%mm7, %%mm3     \n\t"\
789
        "punpcklbw %%mm7, %%mm0     \n\t"\
790
        "paddw %%mm0, %%mm1         \n\t"\
791
        "paddw %%mm3, %%mm2         \n\t"\
792
        "movd  -2(%0), %%mm0        \n\t"\
793
        "movd   3(%0), %%mm3        \n\t"\
794
        "punpcklbw %%mm7, %%mm0     \n\t"\
795
        "punpcklbw %%mm7, %%mm3     \n\t"\
796
        "paddw %%mm3, %%mm0         \n\t"\
797
        "psllw $2, %%mm2            \n\t"\
798
        "psubw %%mm1, %%mm2         \n\t"\
799
        "pmullw %%mm4, %%mm2        \n\t"\
800
        "paddw %%mm5, %%mm0         \n\t"\
801
        "paddw %%mm2, %%mm0         \n\t"\
802
        "psraw $5, %%mm0            \n\t"\
803
        "packuswb %%mm0, %%mm0      \n\t"\
804
        OP(%%mm0, (%1),%%mm6, d)\
805
        "add %3, %0                 \n\t"\
806
        "add %4, %1                 \n\t"\
807
        "decl %2                    \n\t"\
808
        " jnz 1b                    \n\t"\
809
        : "+a"(src), "+c"(dst), "+g"(h)\
810
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
811
        : "memory"\
812
    );\
813
}\
814
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
815
    int h=4;\
816
    __asm__ volatile(\
817
        "pxor %%mm7, %%mm7          \n\t"\
818
        "movq %0, %%mm4             \n\t"\
819
        "movq %1, %%mm5             \n\t"\
820
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
821
    );\
822
    do{\
823
    __asm__ volatile(\
824
        "movd  -1(%0), %%mm1        \n\t"\
825
        "movd    (%0), %%mm2        \n\t"\
826
        "movd   1(%0), %%mm3        \n\t"\
827
        "movd   2(%0), %%mm0        \n\t"\
828
        "punpcklbw %%mm7, %%mm1     \n\t"\
829
        "punpcklbw %%mm7, %%mm2     \n\t"\
830
        "punpcklbw %%mm7, %%mm3     \n\t"\
831
        "punpcklbw %%mm7, %%mm0     \n\t"\
832
        "paddw %%mm0, %%mm1         \n\t"\
833
        "paddw %%mm3, %%mm2         \n\t"\
834
        "movd  -2(%0), %%mm0        \n\t"\
835
        "movd   3(%0), %%mm3        \n\t"\
836
        "punpcklbw %%mm7, %%mm0     \n\t"\
837
        "punpcklbw %%mm7, %%mm3     \n\t"\
838
        "paddw %%mm3, %%mm0         \n\t"\
839
        "psllw $2, %%mm2            \n\t"\
840
        "psubw %%mm1, %%mm2         \n\t"\
841
        "pmullw %%mm4, %%mm2        \n\t"\
842
        "paddw %%mm5, %%mm0         \n\t"\
843
        "paddw %%mm2, %%mm0         \n\t"\
844
        "movd   (%2), %%mm3         \n\t"\
845
        "psraw $5, %%mm0            \n\t"\
846
        "packuswb %%mm0, %%mm0      \n\t"\
847
        PAVGB" %%mm3, %%mm0         \n\t"\
848
        OP(%%mm0, (%1),%%mm6, d)\
849
        "add %4, %0                 \n\t"\
850
        "add %4, %1                 \n\t"\
851
        "add %3, %2                 \n\t"\
852
        : "+a"(src), "+c"(dst), "+d"(src2)\
853
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
854
        : "memory"\
855
    );\
856
    }while(--h);\
857
}\
858
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
859
    src -= 2*srcStride;\
860
    __asm__ volatile(\
861
        "pxor %%mm7, %%mm7          \n\t"\
862
        "movd (%0), %%mm0           \n\t"\
863
        "add %2, %0                 \n\t"\
864
        "movd (%0), %%mm1           \n\t"\
865
        "add %2, %0                 \n\t"\
866
        "movd (%0), %%mm2           \n\t"\
867
        "add %2, %0                 \n\t"\
868
        "movd (%0), %%mm3           \n\t"\
869
        "add %2, %0                 \n\t"\
870
        "movd (%0), %%mm4           \n\t"\
871
        "add %2, %0                 \n\t"\
872
        "punpcklbw %%mm7, %%mm0     \n\t"\
873
        "punpcklbw %%mm7, %%mm1     \n\t"\
874
        "punpcklbw %%mm7, %%mm2     \n\t"\
875
        "punpcklbw %%mm7, %%mm3     \n\t"\
876
        "punpcklbw %%mm7, %%mm4     \n\t"\
877
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
878
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
879
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
880
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
881
         \
882
        : "+a"(src), "+c"(dst)\
883
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
884
        : "memory"\
885
    );\
886
}\
887
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){\
888
    int h=4;\
889
    int w=3;\
890
    src -= 2*srcStride+2;\
891
    while(w--){\
892
        __asm__ volatile(\
893
            "pxor %%mm7, %%mm7      \n\t"\
894
            "movd (%0), %%mm0       \n\t"\
895
            "add %2, %0             \n\t"\
896
            "movd (%0), %%mm1       \n\t"\
897
            "add %2, %0             \n\t"\
898
            "movd (%0), %%mm2       \n\t"\
899
            "add %2, %0             \n\t"\
900
            "movd (%0), %%mm3       \n\t"\
901
            "add %2, %0             \n\t"\
902
            "movd (%0), %%mm4       \n\t"\
903
            "add %2, %0             \n\t"\
904
            "punpcklbw %%mm7, %%mm0 \n\t"\
905
            "punpcklbw %%mm7, %%mm1 \n\t"\
906
            "punpcklbw %%mm7, %%mm2 \n\t"\
907
            "punpcklbw %%mm7, %%mm3 \n\t"\
908
            "punpcklbw %%mm7, %%mm4 \n\t"\
909
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
910
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
911
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
912
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
913
             \
914
            : "+a"(src)\
915
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
916
            : "memory"\
917
        );\
918
        tmp += 4;\
919
        src += 4 - 9*srcStride;\
920
    }\
921
    tmp -= 3*4;\
922
    __asm__ volatile(\
923
        "1:                         \n\t"\
924
        "movq     (%0), %%mm0       \n\t"\
925
        "paddw  10(%0), %%mm0       \n\t"\
926
        "movq    2(%0), %%mm1       \n\t"\
927
        "paddw   8(%0), %%mm1       \n\t"\
928
        "movq    4(%0), %%mm2       \n\t"\
929
        "paddw   6(%0), %%mm2       \n\t"\
930
        "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
931
        "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
932
        "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
933
        "paddsw %%mm2, %%mm0        \n\t"\
934
        "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
935
        "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 */\
936
        "psraw $6, %%mm0            \n\t"\
937
        "packuswb %%mm0, %%mm0      \n\t"\
938
        OP(%%mm0, (%1),%%mm7, d)\
939
        "add $24, %0                \n\t"\
940
        "add %3, %1                 \n\t"\
941
        "decl %2                    \n\t"\
942
        " jnz 1b                    \n\t"\
943
        : "+a"(tmp), "+c"(dst), "+g"(h)\
944
        : "S"((x86_reg)dstStride)\
945
        : "memory"\
946
    );\
947
}\
948
\
949
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
950
    int h=8;\
951
    __asm__ volatile(\
952
        "pxor %%mm7, %%mm7          \n\t"\
953
        "movq %5, %%mm6             \n\t"\
954
        "1:                         \n\t"\
955
        "movq    (%0), %%mm0        \n\t"\
956
        "movq   1(%0), %%mm2        \n\t"\
957
        "movq %%mm0, %%mm1          \n\t"\
958
        "movq %%mm2, %%mm3          \n\t"\
959
        "punpcklbw %%mm7, %%mm0     \n\t"\
960
        "punpckhbw %%mm7, %%mm1     \n\t"\
961
        "punpcklbw %%mm7, %%mm2     \n\t"\
962
        "punpckhbw %%mm7, %%mm3     \n\t"\
963
        "paddw %%mm2, %%mm0         \n\t"\
964
        "paddw %%mm3, %%mm1         \n\t"\
965
        "psllw $2, %%mm0            \n\t"\
966
        "psllw $2, %%mm1            \n\t"\
967
        "movq   -1(%0), %%mm2       \n\t"\
968
        "movq    2(%0), %%mm4       \n\t"\
969
        "movq %%mm2, %%mm3          \n\t"\
970
        "movq %%mm4, %%mm5          \n\t"\
971
        "punpcklbw %%mm7, %%mm2     \n\t"\
972
        "punpckhbw %%mm7, %%mm3     \n\t"\
973
        "punpcklbw %%mm7, %%mm4     \n\t"\
974
        "punpckhbw %%mm7, %%mm5     \n\t"\
975
        "paddw %%mm4, %%mm2         \n\t"\
976
        "paddw %%mm3, %%mm5         \n\t"\
977
        "psubw %%mm2, %%mm0         \n\t"\
978
        "psubw %%mm5, %%mm1         \n\t"\
979
        "pmullw %%mm6, %%mm0        \n\t"\
980
        "pmullw %%mm6, %%mm1        \n\t"\
981
        "movd   -2(%0), %%mm2       \n\t"\
982
        "movd    7(%0), %%mm5       \n\t"\
983
        "punpcklbw %%mm7, %%mm2     \n\t"\
984
        "punpcklbw %%mm7, %%mm5     \n\t"\
985
        "paddw %%mm3, %%mm2         \n\t"\
986
        "paddw %%mm5, %%mm4         \n\t"\
987
        "movq %6, %%mm5             \n\t"\
988
        "paddw %%mm5, %%mm2         \n\t"\
989
        "paddw %%mm5, %%mm4         \n\t"\
990
        "paddw %%mm2, %%mm0         \n\t"\
991
        "paddw %%mm4, %%mm1         \n\t"\
992
        "psraw $5, %%mm0            \n\t"\
993
        "psraw $5, %%mm1            \n\t"\
994
        "packuswb %%mm1, %%mm0      \n\t"\
995
        OP(%%mm0, (%1),%%mm5, q)\
996
        "add %3, %0                 \n\t"\
997
        "add %4, %1                 \n\t"\
998
        "decl %2                    \n\t"\
999
        " jnz 1b                    \n\t"\
1000
        : "+a"(src), "+c"(dst), "+g"(h)\
1001
        : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1002
        : "memory"\
1003
    );\
1004
}\
1005
\
1006
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1007
    int h=8;\
1008
    __asm__ volatile(\
1009
        "pxor %%mm7, %%mm7          \n\t"\
1010
        "movq %0, %%mm6             \n\t"\
1011
        :: "m"(ff_pw_5)\
1012
    );\
1013
    do{\
1014
    __asm__ volatile(\
1015
        "movq    (%0), %%mm0        \n\t"\
1016
        "movq   1(%0), %%mm2        \n\t"\
1017
        "movq %%mm0, %%mm1          \n\t"\
1018
        "movq %%mm2, %%mm3          \n\t"\
1019
        "punpcklbw %%mm7, %%mm0     \n\t"\
1020
        "punpckhbw %%mm7, %%mm1     \n\t"\
1021
        "punpcklbw %%mm7, %%mm2     \n\t"\
1022
        "punpckhbw %%mm7, %%mm3     \n\t"\
1023
        "paddw %%mm2, %%mm0         \n\t"\
1024
        "paddw %%mm3, %%mm1         \n\t"\
1025
        "psllw $2, %%mm0            \n\t"\
1026
        "psllw $2, %%mm1            \n\t"\
1027
        "movq   -1(%0), %%mm2       \n\t"\
1028
        "movq    2(%0), %%mm4       \n\t"\
1029
        "movq %%mm2, %%mm3          \n\t"\
1030
        "movq %%mm4, %%mm5          \n\t"\
1031
        "punpcklbw %%mm7, %%mm2     \n\t"\
1032
        "punpckhbw %%mm7, %%mm3     \n\t"\
1033
        "punpcklbw %%mm7, %%mm4     \n\t"\
1034
        "punpckhbw %%mm7, %%mm5     \n\t"\
1035
        "paddw %%mm4, %%mm2         \n\t"\
1036
        "paddw %%mm3, %%mm5         \n\t"\
1037
        "psubw %%mm2, %%mm0         \n\t"\
1038
        "psubw %%mm5, %%mm1         \n\t"\
1039
        "pmullw %%mm6, %%mm0        \n\t"\
1040
        "pmullw %%mm6, %%mm1        \n\t"\
1041
        "movd   -2(%0), %%mm2       \n\t"\
1042
        "movd    7(%0), %%mm5       \n\t"\
1043
        "punpcklbw %%mm7, %%mm2     \n\t"\
1044
        "punpcklbw %%mm7, %%mm5     \n\t"\
1045
        "paddw %%mm3, %%mm2         \n\t"\
1046
        "paddw %%mm5, %%mm4         \n\t"\
1047
        "movq %5, %%mm5             \n\t"\
1048
        "paddw %%mm5, %%mm2         \n\t"\
1049
        "paddw %%mm5, %%mm4         \n\t"\
1050
        "paddw %%mm2, %%mm0         \n\t"\
1051
        "paddw %%mm4, %%mm1         \n\t"\
1052
        "psraw $5, %%mm0            \n\t"\
1053
        "psraw $5, %%mm1            \n\t"\
1054
        "movq (%2), %%mm4           \n\t"\
1055
        "packuswb %%mm1, %%mm0      \n\t"\
1056
        PAVGB" %%mm4, %%mm0         \n\t"\
1057
        OP(%%mm0, (%1),%%mm5, q)\
1058
        "add %4, %0                 \n\t"\
1059
        "add %4, %1                 \n\t"\
1060
        "add %3, %2                 \n\t"\
1061
        : "+a"(src), "+c"(dst), "+d"(src2)\
1062
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1063
          "m"(ff_pw_16)\
1064
        : "memory"\
1065
    );\
1066
    }while(--h);\
1067
}\
1068
\
1069
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1070
    int w= 2;\
1071
    src -= 2*srcStride;\
1072
    \
1073
    while(w--){\
1074
      __asm__ volatile(\
1075
        "pxor %%mm7, %%mm7          \n\t"\
1076
        "movd (%0), %%mm0           \n\t"\
1077
        "add %2, %0                 \n\t"\
1078
        "movd (%0), %%mm1           \n\t"\
1079
        "add %2, %0                 \n\t"\
1080
        "movd (%0), %%mm2           \n\t"\
1081
        "add %2, %0                 \n\t"\
1082
        "movd (%0), %%mm3           \n\t"\
1083
        "add %2, %0                 \n\t"\
1084
        "movd (%0), %%mm4           \n\t"\
1085
        "add %2, %0                 \n\t"\
1086
        "punpcklbw %%mm7, %%mm0     \n\t"\
1087
        "punpcklbw %%mm7, %%mm1     \n\t"\
1088
        "punpcklbw %%mm7, %%mm2     \n\t"\
1089
        "punpcklbw %%mm7, %%mm3     \n\t"\
1090
        "punpcklbw %%mm7, %%mm4     \n\t"\
1091
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1092
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1093
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1094
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1095
        QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1096
        QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1097
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1098
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1099
         \
1100
        : "+a"(src), "+c"(dst)\
1101
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1102
        : "memory"\
1103
     );\
1104
     if(h==16){\
1105
        __asm__ volatile(\
1106
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1107
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1108
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1109
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1110
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1111
            QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1112
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1113
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1114
            \
1115
           : "+a"(src), "+c"(dst)\
1116
           : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1117
           : "memory"\
1118
        );\
1119
     }\
1120
     src += 4-(h+5)*srcStride;\
1121
     dst += 4-h*dstStride;\
1122
   }\
1123
}\
1124
static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
1125
    int w = (size+8)>>2;\
1126
    src -= 2*srcStride+2;\
1127
    while(w--){\
1128
        __asm__ volatile(\
1129
            "pxor %%mm7, %%mm7      \n\t"\
1130
            "movd (%0), %%mm0       \n\t"\
1131
            "add %2, %0             \n\t"\
1132
            "movd (%0), %%mm1       \n\t"\
1133
            "add %2, %0             \n\t"\
1134
            "movd (%0), %%mm2       \n\t"\
1135
            "add %2, %0             \n\t"\
1136
            "movd (%0), %%mm3       \n\t"\
1137
            "add %2, %0             \n\t"\
1138
            "movd (%0), %%mm4       \n\t"\
1139
            "add %2, %0             \n\t"\
1140
            "punpcklbw %%mm7, %%mm0 \n\t"\
1141
            "punpcklbw %%mm7, %%mm1 \n\t"\
1142
            "punpcklbw %%mm7, %%mm2 \n\t"\
1143
            "punpcklbw %%mm7, %%mm3 \n\t"\
1144
            "punpcklbw %%mm7, %%mm4 \n\t"\
1145
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
1146
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
1147
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
1148
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
1149
            QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
1150
            QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
1151
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
1152
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
1153
            : "+a"(src)\
1154
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1155
            : "memory"\
1156
        );\
1157
        if(size==16){\
1158
            __asm__ volatile(\
1159
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
1160
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
1161
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
1162
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
1163
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
1164
                QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
1165
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
1166
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
1167
                : "+a"(src)\
1168
                : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1169
                : "memory"\
1170
            );\
1171
        }\
1172
        tmp += 4;\
1173
        src += 4 - (size+5)*srcStride;\
1174
    }\
1175
}\
1176
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1177
    int w = size>>4;\
1178
    do{\
1179
    int h = size;\
1180
    __asm__ volatile(\
1181
        "1:                         \n\t"\
1182
        "movq     (%0), %%mm0       \n\t"\
1183
        "movq    8(%0), %%mm3       \n\t"\
1184
        "movq    2(%0), %%mm1       \n\t"\
1185
        "movq   10(%0), %%mm4       \n\t"\
1186
        "paddw   %%mm4, %%mm0       \n\t"\
1187
        "paddw   %%mm3, %%mm1       \n\t"\
1188
        "paddw  18(%0), %%mm3       \n\t"\
1189
        "paddw  16(%0), %%mm4       \n\t"\
1190
        "movq    4(%0), %%mm2       \n\t"\
1191
        "movq   12(%0), %%mm5       \n\t"\
1192
        "paddw   6(%0), %%mm2       \n\t"\
1193
        "paddw  14(%0), %%mm5       \n\t"\
1194
        "psubw %%mm1, %%mm0         \n\t"\
1195
        "psubw %%mm4, %%mm3         \n\t"\
1196
        "psraw $2, %%mm0            \n\t"\
1197
        "psraw $2, %%mm3            \n\t"\
1198
        "psubw %%mm1, %%mm0         \n\t"\
1199
        "psubw %%mm4, %%mm3         \n\t"\
1200
        "paddsw %%mm2, %%mm0        \n\t"\
1201
        "paddsw %%mm5, %%mm3        \n\t"\
1202
        "psraw $2, %%mm0            \n\t"\
1203
        "psraw $2, %%mm3            \n\t"\
1204
        "paddw %%mm2, %%mm0         \n\t"\
1205
        "paddw %%mm5, %%mm3         \n\t"\
1206
        "psraw $6, %%mm0            \n\t"\
1207
        "psraw $6, %%mm3            \n\t"\
1208
        "packuswb %%mm3, %%mm0      \n\t"\
1209
        OP(%%mm0, (%1),%%mm7, q)\
1210
        "add $48, %0                \n\t"\
1211
        "add %3, %1                 \n\t"\
1212
        "decl %2                    \n\t"\
1213
        " jnz 1b                    \n\t"\
1214
        : "+a"(tmp), "+c"(dst), "+g"(h)\
1215
        : "S"((x86_reg)dstStride)\
1216
        : "memory"\
1217
    );\
1218
    tmp += 8 - size*24;\
1219
    dst += 8 - size*dstStride;\
1220
    }while(w--);\
1221
}\
1222
\
1223
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1224
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1225
}\
1226
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1227
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1228
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1229
}\
1230
\
1231
static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1232
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1233
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1234
    src += 8*srcStride;\
1235
    dst += 8*dstStride;\
1236
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1237
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1238
}\
1239
\
1240
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1241
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1242
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1243
    src += 8*dstStride;\
1244
    dst += 8*dstStride;\
1245
    src2 += 8*src2Stride;\
1246
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1247
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1248
}\
1249
\
1250
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){\
1251
          put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
1252
    OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1253
}\
1254
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1255
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
1256
}\
1257
\
1258
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1259
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
1260
}\
1261
\
1262
static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1263
{\
1264
    __asm__ volatile(\
1265
        "movq      (%1), %%mm0          \n\t"\
1266
        "movq    24(%1), %%mm1          \n\t"\
1267
        "psraw      $5,  %%mm0          \n\t"\
1268
        "psraw      $5,  %%mm1          \n\t"\
1269
        "packuswb %%mm0, %%mm0          \n\t"\
1270
        "packuswb %%mm1, %%mm1          \n\t"\
1271
        PAVGB"     (%0), %%mm0          \n\t"\
1272
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1273
        OP(%%mm0, (%2),    %%mm4, d)\
1274
        OP(%%mm1, (%2,%4), %%mm5, d)\
1275
        "lea  (%0,%3,2), %0             \n\t"\
1276
        "lea  (%2,%4,2), %2             \n\t"\
1277
        "movq    48(%1), %%mm0          \n\t"\
1278
        "movq    72(%1), %%mm1          \n\t"\
1279
        "psraw      $5,  %%mm0          \n\t"\
1280
        "psraw      $5,  %%mm1          \n\t"\
1281
        "packuswb %%mm0, %%mm0          \n\t"\
1282
        "packuswb %%mm1, %%mm1          \n\t"\
1283
        PAVGB"     (%0), %%mm0          \n\t"\
1284
        PAVGB"  (%0,%3), %%mm1          \n\t"\
1285
        OP(%%mm0, (%2),    %%mm4, d)\
1286
        OP(%%mm1, (%2,%4), %%mm5, d)\
1287
        :"+a"(src8), "+c"(src16), "+d"(dst)\
1288
        :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
1289
        :"memory");\
1290
}\
1291
static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1292
{\
1293
    do{\
1294
    __asm__ volatile(\
1295
        "movq      (%1), %%mm0          \n\t"\
1296
        "movq     8(%1), %%mm1          \n\t"\
1297
        "movq    48(%1), %%mm2          \n\t"\
1298
        "movq  8+48(%1), %%mm3          \n\t"\
1299
        "psraw      $5,  %%mm0          \n\t"\
1300
        "psraw      $5,  %%mm1          \n\t"\
1301
        "psraw      $5,  %%mm2          \n\t"\
1302
        "psraw      $5,  %%mm3          \n\t"\
1303
        "packuswb %%mm1, %%mm0          \n\t"\
1304
        "packuswb %%mm3, %%mm2          \n\t"\
1305
        PAVGB"     (%0), %%mm0          \n\t"\
1306
        PAVGB"  (%0,%3), %%mm2          \n\t"\
1307
        OP(%%mm0, (%2), %%mm5, q)\
1308
        OP(%%mm2, (%2,%4), %%mm5, q)\
1309
        ::"a"(src8), "c"(src16), "d"(dst),\
1310
          "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
1311
        :"memory");\
1312
        src8 += 2L*src8Stride;\
1313
        src16 += 48;\
1314
        dst += 2L*dstStride;\
1315
    }while(h-=2);\
1316
}\
1317
static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1318
{\
1319
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
1320
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1321
}\
1322

    
1323

    
1324
#ifdef ARCH_X86_64
1325
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1326
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1327
    int h=16;\
1328
    __asm__ volatile(\
1329
        "pxor %%xmm15, %%xmm15      \n\t"\
1330
        "movdqa %6, %%xmm14         \n\t"\
1331
        "movdqa %7, %%xmm13         \n\t"\
1332
        "1:                         \n\t"\
1333
        "lddqu    3(%0), %%xmm1     \n\t"\
1334
        "lddqu   -5(%0), %%xmm7     \n\t"\
1335
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1336
        "punpckhbw %%xmm15, %%xmm1  \n\t"\
1337
        "punpcklbw %%xmm15, %%xmm0  \n\t"\
1338
        "punpcklbw %%xmm15, %%xmm7  \n\t"\
1339
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1340
        "movdqa  %%xmm0, %%xmm6     \n\t"\
1341
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1342
        "movdqa  %%xmm0, %%xmm8     \n\t"\
1343
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1344
        "movdqa  %%xmm0, %%xmm9     \n\t"\
1345
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1346
        "movdqa  %%xmm0, %%xmm10    \n\t"\
1347
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1348
        "palignr $6, %%xmm7, %%xmm10\n\t"\
1349
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1350
        "palignr $8, %%xmm7, %%xmm9 \n\t"\
1351
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1352
        "palignr $10,%%xmm7, %%xmm8 \n\t"\
1353
        "paddw   %%xmm1, %%xmm5     \n\t"\
1354
        "paddw   %%xmm0, %%xmm10    \n\t"\
1355
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1356
        "palignr $12,%%xmm7, %%xmm6 \n\t"\
1357
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1358
        "palignr $14,%%xmm7, %%xmm0 \n\t"\
1359
        "paddw   %%xmm3, %%xmm2     \n\t"\
1360
        "paddw   %%xmm8, %%xmm6     \n\t"\
1361
        "paddw   %%xmm4, %%xmm1     \n\t"\
1362
        "paddw   %%xmm9, %%xmm0     \n\t"\
1363
        "psllw   $2,     %%xmm2     \n\t"\
1364
        "psllw   $2,     %%xmm6     \n\t"\
1365
        "psubw   %%xmm1, %%xmm2     \n\t"\
1366
        "psubw   %%xmm0, %%xmm6     \n\t"\
1367
        "paddw   %%xmm13,%%xmm5     \n\t"\
1368
        "paddw   %%xmm13,%%xmm10    \n\t"\
1369
        "pmullw  %%xmm14,%%xmm2     \n\t"\
1370
        "pmullw  %%xmm14,%%xmm6     \n\t"\
1371
        "lddqu   (%2),   %%xmm3     \n\t"\
1372
        "paddw   %%xmm5, %%xmm2     \n\t"\
1373
        "paddw   %%xmm10,%%xmm6     \n\t"\
1374
        "psraw   $5,     %%xmm2     \n\t"\
1375
        "psraw   $5,     %%xmm6     \n\t"\
1376
        "packuswb %%xmm2,%%xmm6     \n\t"\
1377
        "pavgb   %%xmm3, %%xmm6     \n\t"\
1378
        OP(%%xmm6, (%1), %%xmm4, dqa)\
1379
        "add %5, %0                 \n\t"\
1380
        "add %5, %1                 \n\t"\
1381
        "add %4, %2                 \n\t"\
1382
        "decl %3                    \n\t"\
1383
        "jg 1b                      \n\t"\
1384
        : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
1385
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1386
          "m"(ff_pw_5), "m"(ff_pw_16)\
1387
        : "memory"\
1388
    );\
1389
}
1390
#else // ARCH_X86_64
1391
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1392
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1393
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1394
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1395
    src += 8*dstStride;\
1396
    dst += 8*dstStride;\
1397
    src2 += 8*src2Stride;\
1398
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
1399
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1400
}
1401
#endif // ARCH_X86_64
1402

    
1403
#define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
1404
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1405
    int h=8;\
1406
    __asm__ volatile(\
1407
        "pxor %%xmm7, %%xmm7        \n\t"\
1408
        "movdqa %0, %%xmm6          \n\t"\
1409
        :: "m"(ff_pw_5)\
1410
    );\
1411
    do{\
1412
    __asm__ volatile(\
1413
        "lddqu   -5(%0), %%xmm1     \n\t"\
1414
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1415
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
1416
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1417
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1418
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1419
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1420
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1421
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1422
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1423
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1424
        "paddw   %%xmm1, %%xmm5     \n\t"\
1425
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1426
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1427
        "paddw   %%xmm3, %%xmm2     \n\t"\
1428
        "paddw   %%xmm4, %%xmm1     \n\t"\
1429
        "psllw   $2,     %%xmm2     \n\t"\
1430
        "movq    (%2),   %%xmm3     \n\t"\
1431
        "psubw   %%xmm1, %%xmm2     \n\t"\
1432
        "paddw   %5,     %%xmm5     \n\t"\
1433
        "pmullw  %%xmm6, %%xmm2     \n\t"\
1434
        "paddw   %%xmm5, %%xmm2     \n\t"\
1435
        "psraw   $5,     %%xmm2     \n\t"\
1436
        "packuswb %%xmm2, %%xmm2    \n\t"\
1437
        "pavgb   %%xmm3, %%xmm2     \n\t"\
1438
        OP(%%xmm2, (%1), %%xmm4, q)\
1439
        "add %4, %0                 \n\t"\
1440
        "add %4, %1                 \n\t"\
1441
        "add %3, %2                 \n\t"\
1442
        : "+a"(src), "+c"(dst), "+d"(src2)\
1443
        : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1444
          "m"(ff_pw_16)\
1445
        : "memory"\
1446
    );\
1447
    }while(--h);\
1448
}\
1449
QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1450
\
1451
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1452
    int h=8;\
1453
    __asm__ volatile(\
1454
        "pxor %%xmm7, %%xmm7        \n\t"\
1455
        "movdqa %5, %%xmm6          \n\t"\
1456
        "1:                         \n\t"\
1457
        "lddqu   -5(%0), %%xmm1     \n\t"\
1458
        "movdqa  %%xmm1, %%xmm0     \n\t"\
1459
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
1460
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1461
        "movdqa  %%xmm1, %%xmm2     \n\t"\
1462
        "movdqa  %%xmm1, %%xmm3     \n\t"\
1463
        "movdqa  %%xmm1, %%xmm4     \n\t"\
1464
        "movdqa  %%xmm1, %%xmm5     \n\t"\
1465
        "palignr $6, %%xmm0, %%xmm5 \n\t"\
1466
        "palignr $8, %%xmm0, %%xmm4 \n\t"\
1467
        "palignr $10,%%xmm0, %%xmm3 \n\t"\
1468
        "paddw   %%xmm1, %%xmm5     \n\t"\
1469
        "palignr $12,%%xmm0, %%xmm2 \n\t"\
1470
        "palignr $14,%%xmm0, %%xmm1 \n\t"\
1471
        "paddw   %%xmm3, %%xmm2     \n\t"\
1472
        "paddw   %%xmm4, %%xmm1     \n\t"\
1473
        "psllw   $2,     %%xmm2     \n\t"\
1474
        "psubw   %%xmm1, %%xmm2     \n\t"\
1475
        "paddw   %6,     %%xmm5     \n\t"\
1476
        "pmullw  %%xmm6, %%xmm2     \n\t"\
1477
        "paddw   %%xmm5, %%xmm2     \n\t"\
1478
        "psraw   $5,     %%xmm2     \n\t"\
1479
        "packuswb %%xmm2, %%xmm2    \n\t"\
1480
        OP(%%xmm2, (%1), %%xmm4, q)\
1481
        "add %3, %0                 \n\t"\
1482
        "add %4, %1                 \n\t"\
1483
        "decl %2                    \n\t"\
1484
        " jnz 1b                    \n\t"\
1485
        : "+a"(src), "+c"(dst), "+g"(h)\
1486
        : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\
1487
          "m"(ff_pw_5), "m"(ff_pw_16)\
1488
        : "memory"\
1489
    );\
1490
}\
1491
static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1492
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1493
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1494
    src += 8*srcStride;\
1495
    dst += 8*dstStride;\
1496
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
1497
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1498
}\
1499

    
1500
#define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
1501
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1502
    src -= 2*srcStride;\
1503
    \
1504
    __asm__ volatile(\
1505
        "pxor %%xmm7, %%xmm7        \n\t"\
1506
        "movq (%0), %%xmm0          \n\t"\
1507
        "add %2, %0                 \n\t"\
1508
        "movq (%0), %%xmm1          \n\t"\
1509
        "add %2, %0                 \n\t"\
1510
        "movq (%0), %%xmm2          \n\t"\
1511
        "add %2, %0                 \n\t"\
1512
        "movq (%0), %%xmm3          \n\t"\
1513
        "add %2, %0                 \n\t"\
1514
        "movq (%0), %%xmm4          \n\t"\
1515
        "add %2, %0                 \n\t"\
1516
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
1517
        "punpcklbw %%xmm7, %%xmm1   \n\t"\
1518
        "punpcklbw %%xmm7, %%xmm2   \n\t"\
1519
        "punpcklbw %%xmm7, %%xmm3   \n\t"\
1520
        "punpcklbw %%xmm7, %%xmm4   \n\t"\
1521
        QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1522
        QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1523
        QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1524
        QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1525
        QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1526
        QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1527
        QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1528
        QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1529
         \
1530
        : "+a"(src), "+c"(dst)\
1531
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1532
        : "memory"\
1533
    );\
1534
    if(h==16){\
1535
        __asm__ volatile(\
1536
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1537
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1538
            QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1539
            QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1540
            QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1541
            QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1542
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1543
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1544
            \
1545
            : "+a"(src), "+c"(dst)\
1546
            : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1547
            : "memory"\
1548
        );\
1549
    }\
1550
}\
1551
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1552
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
1553
}\
1554
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1555
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
1556
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1557
}
1558

    
1559
static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
1560
    int w = (size+8)>>3;
1561
    src -= 2*srcStride+2;
1562
    while(w--){
1563
        __asm__ volatile(
1564
            "pxor %%xmm7, %%xmm7        \n\t"
1565
            "movq (%0), %%xmm0          \n\t"
1566
            "add %2, %0                 \n\t"
1567
            "movq (%0), %%xmm1          \n\t"
1568
            "add %2, %0                 \n\t"
1569
            "movq (%0), %%xmm2          \n\t"
1570
            "add %2, %0                 \n\t"
1571
            "movq (%0), %%xmm3          \n\t"
1572
            "add %2, %0                 \n\t"
1573
            "movq (%0), %%xmm4          \n\t"
1574
            "add %2, %0                 \n\t"
1575
            "punpcklbw %%xmm7, %%xmm0   \n\t"
1576
            "punpcklbw %%xmm7, %%xmm1   \n\t"
1577
            "punpcklbw %%xmm7, %%xmm2   \n\t"
1578
            "punpcklbw %%xmm7, %%xmm3   \n\t"
1579
            "punpcklbw %%xmm7, %%xmm4   \n\t"
1580
            QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
1581
            QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
1582
            QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
1583
            QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
1584
            QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
1585
            QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
1586
            QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
1587
            QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
1588
            : "+a"(src)
1589
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1590
            : "memory"
1591
        );
1592
        if(size==16){
1593
            __asm__ volatile(
1594
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
1595
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
1596
                QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
1597
                QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
1598
                QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
1599
                QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
1600
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
1601
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
1602
                : "+a"(src)
1603
                : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1604
                : "memory"
1605
            );
1606
        }
1607
        tmp += 8;
1608
        src += 8 - (size+5)*srcStride;
1609
    }
1610
}
1611

    
1612
#define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
1613
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1614
    int h = size;\
1615
    if(size == 16){\
1616
        __asm__ volatile(\
1617
            "1:                         \n\t"\
1618
            "movdqa 32(%0), %%xmm4      \n\t"\
1619
            "movdqa 16(%0), %%xmm5      \n\t"\
1620
            "movdqa   (%0), %%xmm7      \n\t"\
1621
            "movdqa %%xmm4, %%xmm3      \n\t"\
1622
            "movdqa %%xmm4, %%xmm2      \n\t"\
1623
            "movdqa %%xmm4, %%xmm1      \n\t"\
1624
            "movdqa %%xmm4, %%xmm0      \n\t"\
1625
            "palignr $10, %%xmm5, %%xmm0 \n\t"\
1626
            "palignr  $8, %%xmm5, %%xmm1 \n\t"\
1627
            "palignr  $6, %%xmm5, %%xmm2 \n\t"\
1628
            "palignr  $4, %%xmm5, %%xmm3 \n\t"\
1629
            "palignr  $2, %%xmm5, %%xmm4 \n\t"\
1630
            "paddw  %%xmm5, %%xmm0      \n\t"\
1631
            "paddw  %%xmm4, %%xmm1      \n\t"\
1632
            "paddw  %%xmm3, %%xmm2      \n\t"\
1633
            "movdqa %%xmm5, %%xmm6      \n\t"\
1634
            "movdqa %%xmm5, %%xmm4      \n\t"\
1635
            "movdqa %%xmm5, %%xmm3      \n\t"\
1636
            "palignr  $8, %%xmm7, %%xmm4 \n\t"\
1637
            "palignr  $2, %%xmm7, %%xmm6 \n\t"\
1638
            "palignr $10, %%xmm7, %%xmm3 \n\t"\
1639
            "paddw  %%xmm6, %%xmm4      \n\t"\
1640
            "movdqa %%xmm5, %%xmm6      \n\t"\
1641
            "palignr  $6, %%xmm7, %%xmm5 \n\t"\
1642
            "palignr  $4, %%xmm7, %%xmm6 \n\t"\
1643
            "paddw  %%xmm7, %%xmm3      \n\t"\
1644
            "paddw  %%xmm6, %%xmm5      \n\t"\
1645
            \
1646
            "psubw  %%xmm1, %%xmm0      \n\t"\
1647
            "psubw  %%xmm4, %%xmm3      \n\t"\
1648
            "psraw      $2, %%xmm0      \n\t"\
1649
            "psraw      $2, %%xmm3      \n\t"\
1650
            "psubw  %%xmm1, %%xmm0      \n\t"\
1651
            "psubw  %%xmm4, %%xmm3      \n\t"\
1652
            "paddw  %%xmm2, %%xmm0      \n\t"\
1653
            "paddw  %%xmm5, %%xmm3      \n\t"\
1654
            "psraw      $2, %%xmm0      \n\t"\
1655
            "psraw      $2, %%xmm3      \n\t"\
1656
            "paddw  %%xmm2, %%xmm0      \n\t"\
1657
            "paddw  %%xmm5, %%xmm3      \n\t"\
1658
            "psraw      $6, %%xmm0      \n\t"\
1659
            "psraw      $6, %%xmm3      \n\t"\
1660
            "packuswb %%xmm0, %%xmm3    \n\t"\
1661
            OP(%%xmm3, (%1), %%xmm7, dqa)\
1662
            "add $48, %0                \n\t"\
1663
            "add %3, %1                 \n\t"\
1664
            "decl %2                    \n\t"\
1665
            " jnz 1b                    \n\t"\
1666
            : "+a"(tmp), "+c"(dst), "+g"(h)\
1667
            : "S"((x86_reg)dstStride)\
1668
            : "memory"\
1669
        );\
1670
    }else{\
1671
        __asm__ volatile(\
1672
            "1:                         \n\t"\
1673
            "movdqa 16(%0), %%xmm1      \n\t"\
1674
            "movdqa   (%0), %%xmm0      \n\t"\
1675
            "movdqa %%xmm1, %%xmm2      \n\t"\
1676
            "movdqa %%xmm1, %%xmm3      \n\t"\
1677
            "movdqa %%xmm1, %%xmm4      \n\t"\
1678
            "movdqa %%xmm1, %%xmm5      \n\t"\
1679
            "palignr $10, %%xmm0, %%xmm5 \n\t"\
1680
            "palignr  $8, %%xmm0, %%xmm4 \n\t"\
1681
            "palignr  $6, %%xmm0, %%xmm3 \n\t"\
1682
            "palignr  $4, %%xmm0, %%xmm2 \n\t"\
1683
            "palignr  $2, %%xmm0, %%xmm1 \n\t"\
1684
            "paddw  %%xmm5, %%xmm0      \n\t"\
1685
            "paddw  %%xmm4, %%xmm1      \n\t"\
1686
            "paddw  %%xmm3, %%xmm2      \n\t"\
1687
            "psubw  %%xmm1, %%xmm0      \n\t"\
1688
            "psraw      $2, %%xmm0      \n\t"\
1689
            "psubw  %%xmm1, %%xmm0      \n\t"\
1690
            "paddw  %%xmm2, %%xmm0      \n\t"\
1691
            "psraw      $2, %%xmm0      \n\t"\
1692
            "paddw  %%xmm2, %%xmm0      \n\t"\
1693
            "psraw      $6, %%xmm0      \n\t"\
1694
            "packuswb %%xmm0, %%xmm0    \n\t"\
1695
            OP(%%xmm0, (%1), %%xmm7, q)\
1696
            "add $48, %0                \n\t"\
1697
            "add %3, %1                 \n\t"\
1698
            "decl %2                    \n\t"\
1699
            " jnz 1b                    \n\t"\
1700
            : "+a"(tmp), "+c"(dst), "+g"(h)\
1701
            : "S"((x86_reg)dstStride)\
1702
            : "memory"\
1703
        );\
1704
    }\
1705
}
1706

    
1707
#define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
1708
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){\
1709
          put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
1710
    OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1711
}\
1712
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1713
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
1714
}\
1715
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1716
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1717
}\
1718

    
1719
#define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1720
#define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1721
#define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1722
#define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1723
#define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1724
#define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1725
#define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1726
#define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1727

    
1728
#define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1729
#define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1730
#define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1731
#define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1732
#define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1733
#define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1734
#define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1735
#define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1736

    
1737
#define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1738
#define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1739
#define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1740
#define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1741

    
1742
#define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1743
#define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1744
#define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1745
#define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1746

    
1747
#define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1748
#define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1749

    
1750
#define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1751
H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1752
H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1753
H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1754
H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1755

    
1756
static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1757
    put_pixels16_sse2(dst, src, stride, 16);
1758
}
1759
static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1760
    avg_pixels16_sse2(dst, src, stride, 16);
1761
}
1762
#define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1763
#define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1764

    
1765
#define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1766
static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1767
    OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1768
}\
1769

    
1770
#define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1771
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1772
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1773
}\
1774
\
1775
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1776
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1777
}\
1778
\
1779
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1780
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1781
}\
1782

    
1783
#define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1784
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1785
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1786
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1787
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1788
}\
1789
\
1790
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1791
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1792
}\
1793
\
1794
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1795
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1796
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1797
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1798
}\
1799

    
1800
#define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1801
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1802
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1803
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1804
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1805
}\
1806
\
1807
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1808
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1809
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1810
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1811
}\
1812
\
1813
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1814
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1815
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1816
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1817
}\
1818
\
1819
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1820
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*SIZE]);\
1821
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1822
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1823
}\
1824
\
1825
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1826
    DECLARE_ALIGNED(ALIGN, uint16_t, temp[SIZE*(SIZE<8?12:24)]);\
1827
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
1828
}\
1829
\
1830
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1831
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1832
    uint8_t * const halfHV= temp;\
1833
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1834
    assert(((int)temp & 7) == 0);\
1835
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1836
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1837
}\
1838
\
1839
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1840
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1841
    uint8_t * const halfHV= temp;\
1842
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1843
    assert(((int)temp & 7) == 0);\
1844
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1845
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1846
}\
1847
\
1848
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1849
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1850
    uint8_t * const halfHV= temp;\
1851
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1852
    assert(((int)temp & 7) == 0);\
1853
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1854
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1855
}\
1856
\
1857
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1858
    DECLARE_ALIGNED(ALIGN, uint8_t, temp[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE]);\
1859
    uint8_t * const halfHV= temp;\
1860
    int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1861
    assert(((int)temp & 7) == 0);\
1862
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1863
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1864
}\
1865

    
1866
#define H264_MC_4816(MMX)\
1867
H264_MC(put_, 4, MMX, 8)\
1868
H264_MC(put_, 8, MMX, 8)\
1869
H264_MC(put_, 16,MMX, 8)\
1870
H264_MC(avg_, 4, MMX, 8)\
1871
H264_MC(avg_, 8, MMX, 8)\
1872
H264_MC(avg_, 16,MMX, 8)\
1873

    
1874
#define H264_MC_816(QPEL, XMM)\
1875
QPEL(put_, 8, XMM, 16)\
1876
QPEL(put_, 16,XMM, 16)\
1877
QPEL(avg_, 8, XMM, 16)\
1878
QPEL(avg_, 16,XMM, 16)\
1879

    
1880

    
1881
#define AVG_3DNOW_OP(a,b,temp, size) \
1882
"mov" #size " " #b ", " #temp "   \n\t"\
1883
"pavgusb " #temp ", " #a "        \n\t"\
1884
"mov" #size " " #a ", " #b "      \n\t"
1885
#define AVG_MMX2_OP(a,b,temp, size) \
1886
"mov" #size " " #b ", " #temp "   \n\t"\
1887
"pavgb " #temp ", " #a "          \n\t"\
1888
"mov" #size " " #a ", " #b "      \n\t"
1889

    
1890
#define PAVGB "pavgusb"
1891
QPEL_H264(put_,       PUT_OP, 3dnow)
1892
QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1893
#undef PAVGB
1894
#define PAVGB "pavgb"
1895
QPEL_H264(put_,       PUT_OP, mmx2)
1896
QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
1897
QPEL_H264_V_XMM(put_,       PUT_OP, sse2)
1898
QPEL_H264_V_XMM(avg_,  AVG_MMX2_OP, sse2)
1899
QPEL_H264_HV_XMM(put_,       PUT_OP, sse2)
1900
QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, sse2)
1901
#ifdef HAVE_SSSE3
1902
QPEL_H264_H_XMM(put_,       PUT_OP, ssse3)
1903
QPEL_H264_H_XMM(avg_,  AVG_MMX2_OP, ssse3)
1904
QPEL_H264_HV2_XMM(put_,       PUT_OP, ssse3)
1905
QPEL_H264_HV2_XMM(avg_,  AVG_MMX2_OP, ssse3)
1906
QPEL_H264_HV_XMM(put_,       PUT_OP, ssse3)
1907
QPEL_H264_HV_XMM(avg_,  AVG_MMX2_OP, ssse3)
1908
#endif
1909
#undef PAVGB
1910

    
1911
H264_MC_4816(3dnow)
1912
H264_MC_4816(mmx2)
1913
H264_MC_816(H264_MC_V, sse2)
1914
H264_MC_816(H264_MC_HV, sse2)
1915
#ifdef HAVE_SSSE3
1916
H264_MC_816(H264_MC_H, ssse3)
1917
H264_MC_816(H264_MC_HV, ssse3)
1918
#endif
1919

    
1920

    
1921
#define H264_CHROMA_OP(S,D)
1922
#define H264_CHROMA_OP4(S,D,T)
1923
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
1924
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
1925
#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
1926
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1927
#include "dsputil_h264_template_mmx.c"
1928

    
1929
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)
1930
{
1931
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 1);
1932
}
1933
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)
1934
{
1935
    put_h264_chroma_mc8_mmx(dst, src, stride, h, x, y, 0);
1936
}
1937

    
1938
#undef H264_CHROMA_OP
1939
#undef H264_CHROMA_OP4
1940
#undef H264_CHROMA_MC8_TMPL
1941
#undef H264_CHROMA_MC4_TMPL
1942
#undef H264_CHROMA_MC2_TMPL
1943
#undef H264_CHROMA_MC8_MV0
1944

    
1945
#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
1946
#define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\
1947
                               "pavgb " #T ", " #D " \n\t"
1948
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
1949
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
1950
#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
1951
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
1952
#include "dsputil_h264_template_mmx.c"
1953
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)
1954
{
1955
    avg_h264_chroma_mc8_mmx2(dst, src, stride, h, x, y, 1);
1956
}
1957
#undef H264_CHROMA_OP
1958
#undef H264_CHROMA_OP4
1959
#undef H264_CHROMA_MC8_TMPL
1960
#undef H264_CHROMA_MC4_TMPL
1961
#undef H264_CHROMA_MC2_TMPL
1962
#undef H264_CHROMA_MC8_MV0
1963

    
1964
#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
1965
#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
1966
                               "pavgusb " #T ", " #D " \n\t"
1967
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
1968
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
1969
#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
1970
#include "dsputil_h264_template_mmx.c"
1971
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)
1972
{
1973
    avg_h264_chroma_mc8_3dnow(dst, src, stride, h, x, y, 1);
1974
}
1975
#undef H264_CHROMA_OP
1976
#undef H264_CHROMA_OP4
1977
#undef H264_CHROMA_MC8_TMPL
1978
#undef H264_CHROMA_MC4_TMPL
1979
#undef H264_CHROMA_MC8_MV0
1980

    
1981
#ifdef HAVE_SSSE3
1982
#define AVG_OP(X)
1983
#undef H264_CHROMA_MC8_TMPL
1984
#undef H264_CHROMA_MC4_TMPL
1985
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_ssse3
1986
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_ssse3
1987
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1988
#include "dsputil_h264_template_ssse3.c"
1989
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)
1990
{
1991
    put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
1992
}
1993
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)
1994
{
1995
    put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 0);
1996
}
1997

    
1998
#undef AVG_OP
1999
#undef H264_CHROMA_MC8_TMPL
2000
#undef H264_CHROMA_MC4_TMPL
2001
#undef H264_CHROMA_MC8_MV0
2002
#define AVG_OP(X) X
2003
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_ssse3
2004
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_ssse3
2005
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
2006
#include "dsputil_h264_template_ssse3.c"
2007
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)
2008
{
2009
    avg_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
2010
}
2011
#undef AVG_OP
2012
#undef H264_CHROMA_MC8_TMPL
2013
#undef H264_CHROMA_MC4_TMPL
2014
#undef H264_CHROMA_MC8_MV0
2015
#endif
2016

    
2017
/***********************************/
2018
/* weighted prediction */
2019

    
2020
static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
2021
{
2022
    int x, y;
2023
    offset <<= log2_denom;
2024
    offset += (1 << log2_denom) >> 1;
2025
    __asm__ volatile(
2026
        "movd    %0, %%mm4        \n\t"
2027
        "movd    %1, %%mm5        \n\t"
2028
        "movd    %2, %%mm6        \n\t"
2029
        "pshufw  $0, %%mm4, %%mm4 \n\t"
2030
        "pshufw  $0, %%mm5, %%mm5 \n\t"
2031
        "pxor    %%mm7, %%mm7     \n\t"
2032
        :: "g"(weight), "g"(offset), "g"(log2_denom)
2033
    );
2034
    for(y=0; y<h; y+=2){
2035
        for(x=0; x<w; x+=4){
2036
            __asm__ volatile(
2037
                "movd      %0,    %%mm0 \n\t"
2038
                "movd      %1,    %%mm1 \n\t"
2039
                "punpcklbw %%mm7, %%mm0 \n\t"
2040
                "punpcklbw %%mm7, %%mm1 \n\t"
2041
                "pmullw    %%mm4, %%mm0 \n\t"
2042
                "pmullw    %%mm4, %%mm1 \n\t"
2043
                "paddsw    %%mm5, %%mm0 \n\t"
2044
                "paddsw    %%mm5, %%mm1 \n\t"
2045
                "psraw     %%mm6, %%mm0 \n\t"
2046
                "psraw     %%mm6, %%mm1 \n\t"
2047
                "packuswb  %%mm7, %%mm0 \n\t"
2048
                "packuswb  %%mm7, %%mm1 \n\t"
2049
                "movd      %%mm0, %0    \n\t"
2050
                "movd      %%mm1, %1    \n\t"
2051
                : "+m"(*(uint32_t*)(dst+x)),
2052
                  "+m"(*(uint32_t*)(dst+x+stride))
2053
            );
2054
        }
2055
        dst += 2*stride;
2056
    }
2057
}
2058

    
2059
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)
2060
{
2061
    int x, y;
2062
    offset = ((offset + 1) | 1) << log2_denom;
2063
    __asm__ volatile(
2064
        "movd    %0, %%mm3        \n\t"
2065
        "movd    %1, %%mm4        \n\t"
2066
        "movd    %2, %%mm5        \n\t"
2067
        "movd    %3, %%mm6        \n\t"
2068
        "pshufw  $0, %%mm3, %%mm3 \n\t"
2069
        "pshufw  $0, %%mm4, %%mm4 \n\t"
2070
        "pshufw  $0, %%mm5, %%mm5 \n\t"
2071
        "pxor    %%mm7, %%mm7     \n\t"
2072
        :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
2073
    );
2074
    for(y=0; y<h; y++){
2075
        for(x=0; x<w; x+=4){
2076
            __asm__ volatile(
2077
                "movd      %0,    %%mm0 \n\t"
2078
                "movd      %1,    %%mm1 \n\t"
2079
                "punpcklbw %%mm7, %%mm0 \n\t"
2080
                "punpcklbw %%mm7, %%mm1 \n\t"
2081
                "pmullw    %%mm3, %%mm0 \n\t"
2082
                "pmullw    %%mm4, %%mm1 \n\t"
2083
                "paddsw    %%mm1, %%mm0 \n\t"
2084
                "paddsw    %%mm5, %%mm0 \n\t"
2085
                "psraw     %%mm6, %%mm0 \n\t"
2086
                "packuswb  %%mm0, %%mm0 \n\t"
2087
                "movd      %%mm0, %0    \n\t"
2088
                : "+m"(*(uint32_t*)(dst+x))
2089
                :  "m"(*(uint32_t*)(src+x))
2090
            );
2091
        }
2092
        src += stride;
2093
        dst += stride;
2094
    }
2095
}
2096

    
2097
#define H264_WEIGHT(W,H) \
2098
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){ \
2099
    ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
2100
} \
2101
static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
2102
    ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
2103
}
2104

    
2105
H264_WEIGHT(16,16)
2106
H264_WEIGHT(16, 8)
2107
H264_WEIGHT( 8,16)
2108
H264_WEIGHT( 8, 8)
2109
H264_WEIGHT( 8, 4)
2110
H264_WEIGHT( 4, 8)
2111
H264_WEIGHT( 4, 4)
2112
H264_WEIGHT( 4, 2)
2113