Replace more "m" constraints with MANGLE to fix compilation issues
[libav.git] / libavcodec / x86 / h264dsp_mmx.c
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 DECLARE_ALIGNED(8, int16_t, b2)[64];
161
162 block[0] += 32;
163
164 for(i=0; i<2; i++){
165 DECLARE_ALIGNED(8, uint64_t, tmp);
166
167 h264_idct8_1d(block+4*i);
168
169 __asm__ volatile(
170 "movq %%mm7, %0 \n\t"
171 TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
172 "movq %%mm0, 8(%1) \n\t"
173 "movq %%mm6, 24(%1) \n\t"
174 "movq %%mm7, 40(%1) \n\t"
175 "movq %%mm4, 56(%1) \n\t"
176 "movq %0, %%mm7 \n\t"
177 TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
178 "movq %%mm7, (%1) \n\t"
179 "movq %%mm1, 16(%1) \n\t"
180 "movq %%mm0, 32(%1) \n\t"
181 "movq %%mm3, 48(%1) \n\t"
182 : "=m"(tmp)
183 : "r"(b2+32*i)
184 : "memory"
185 );
186 }
187
188 for(i=0; i<2; i++){
189 h264_idct8_1d(b2+4*i);
190
191 __asm__ volatile(
192 "psraw $6, %%mm7 \n\t"
193 "psraw $6, %%mm6 \n\t"
194 "psraw $6, %%mm5 \n\t"
195 "psraw $6, %%mm4 \n\t"
196 "psraw $6, %%mm3 \n\t"
197 "psraw $6, %%mm2 \n\t"
198 "psraw $6, %%mm1 \n\t"
199 "psraw $6, %%mm0 \n\t"
200
201 "movq %%mm7, (%0) \n\t"
202 "movq %%mm5, 16(%0) \n\t"
203 "movq %%mm3, 32(%0) \n\t"
204 "movq %%mm1, 48(%0) \n\t"
205 "movq %%mm0, 64(%0) \n\t"
206 "movq %%mm2, 80(%0) \n\t"
207 "movq %%mm4, 96(%0) \n\t"
208 "movq %%mm6, 112(%0) \n\t"
209 :: "r"(b2+4*i)
210 : "memory"
211 );
212 }
213
214 add_pixels_clamped_mmx(b2, dst, stride);
215 }
216
217 #define STORE_DIFF_8P( p, d, t, z )\
218 "movq "#d", "#t" \n"\
219 "psraw $6, "#p" \n"\
220 "punpcklbw "#z", "#t" \n"\
221 "paddsw "#t", "#p" \n"\
222 "packuswb "#p", "#p" \n"\
223 "movq "#p", "#d" \n"
224
225 #define H264_IDCT8_1D_SSE2(a,b,c,d,e,f,g,h)\
226 "movdqa "#c", "#a" \n"\
227 "movdqa "#g", "#e" \n"\
228 "psraw $1, "#c" \n"\
229 "psraw $1, "#g" \n"\
230 "psubw "#e", "#c" \n"\
231 "paddw "#a", "#g" \n"\
232 "movdqa "#b", "#e" \n"\
233 "psraw $1, "#e" \n"\
234 "paddw "#b", "#e" \n"\
235 "paddw "#d", "#e" \n"\
236 "paddw "#f", "#e" \n"\
237 "movdqa "#f", "#a" \n"\
238 "psraw $1, "#a" \n"\
239 "paddw "#f", "#a" \n"\
240 "paddw "#h", "#a" \n"\
241 "psubw "#b", "#a" \n"\
242 "psubw "#d", "#b" \n"\
243 "psubw "#d", "#f" \n"\
244 "paddw "#h", "#b" \n"\
245 "psubw "#h", "#f" \n"\
246 "psraw $1, "#d" \n"\
247 "psraw $1, "#h" \n"\
248 "psubw "#d", "#b" \n"\
249 "psubw "#h", "#f" \n"\
250 "movdqa "#e", "#d" \n"\
251 "movdqa "#a", "#h" \n"\
252 "psraw $2, "#d" \n"\
253 "psraw $2, "#h" \n"\
254 "paddw "#f", "#d" \n"\
255 "paddw "#b", "#h" \n"\
256 "psraw $2, "#f" \n"\
257 "psraw $2, "#b" \n"\
258 "psubw "#f", "#e" \n"\
259 "psubw "#a", "#b" \n"\
260 "movdqa 0x00(%1), "#a" \n"\
261 "movdqa 0x40(%1), "#f" \n"\
262 SUMSUB_BA(f, a)\
263 SUMSUB_BA(g, f)\
264 SUMSUB_BA(c, a)\
265 SUMSUB_BA(e, g)\
266 SUMSUB_BA(b, c)\
267 SUMSUB_BA(h, a)\
268 SUMSUB_BA(d, f)
269
270 static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
271 {
272 __asm__ volatile(
273 "movdqa 0x10(%1), %%xmm1 \n"
274 "movdqa 0x20(%1), %%xmm2 \n"
275 "movdqa 0x30(%1), %%xmm3 \n"
276 "movdqa 0x50(%1), %%xmm5 \n"
277 "movdqa 0x60(%1), %%xmm6 \n"
278 "movdqa 0x70(%1), %%xmm7 \n"
279 H264_IDCT8_1D_SSE2(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)
280 TRANSPOSE8(%%xmm4, %%xmm1, %%xmm7, %%xmm3, %%xmm5, %%xmm0, %%xmm2, %%xmm6, (%1))
281 "paddw %4, %%xmm4 \n"
282 "movdqa %%xmm4, 0x00(%1) \n"
283 "movdqa %%xmm2, 0x40(%1) \n"
284 H264_IDCT8_1D_SSE2(%%xmm4, %%xmm0, %%xmm6, %%xmm3, %%xmm2, %%xmm5, %%xmm7, %%xmm1)
285 "movdqa %%xmm6, 0x60(%1) \n"
286 "movdqa %%xmm7, 0x70(%1) \n"
287 "pxor %%xmm7, %%xmm7 \n"
288 STORE_DIFF_8P(%%xmm2, (%0), %%xmm6, %%xmm7)
289 STORE_DIFF_8P(%%xmm0, (%0,%2), %%xmm6, %%xmm7)
290 STORE_DIFF_8P(%%xmm1, (%0,%2,2), %%xmm6, %%xmm7)
291 STORE_DIFF_8P(%%xmm3, (%0,%3), %%xmm6, %%xmm7)
292 "lea (%0,%2,4), %0 \n"
293 STORE_DIFF_8P(%%xmm5, (%0), %%xmm6, %%xmm7)
294 STORE_DIFF_8P(%%xmm4, (%0,%2), %%xmm6, %%xmm7)
295 "movdqa 0x60(%1), %%xmm0 \n"
296 "movdqa 0x70(%1), %%xmm1 \n"
297 STORE_DIFF_8P(%%xmm0, (%0,%2,2), %%xmm6, %%xmm7)
298 STORE_DIFF_8P(%%xmm1, (%0,%3), %%xmm6, %%xmm7)
299 :"+r"(dst)
300 :"r"(block), "r"((x86_reg)stride), "r"((x86_reg)3L*stride), "m"(ff_pw_32)
301 );
302 }
303
304 static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
305 {
306 int dc = (block[0] + 32) >> 6;
307 __asm__ volatile(
308 "movd %0, %%mm0 \n\t"
309 "pshufw $0, %%mm0, %%mm0 \n\t"
310 "pxor %%mm1, %%mm1 \n\t"
311 "psubw %%mm0, %%mm1 \n\t"
312 "packuswb %%mm0, %%mm0 \n\t"
313 "packuswb %%mm1, %%mm1 \n\t"
314 ::"r"(dc)
315 );
316 __asm__ volatile(
317 "movd %0, %%mm2 \n\t"
318 "movd %1, %%mm3 \n\t"
319 "movd %2, %%mm4 \n\t"
320 "movd %3, %%mm5 \n\t"
321 "paddusb %%mm0, %%mm2 \n\t"
322 "paddusb %%mm0, %%mm3 \n\t"
323 "paddusb %%mm0, %%mm4 \n\t"
324 "paddusb %%mm0, %%mm5 \n\t"
325 "psubusb %%mm1, %%mm2 \n\t"
326 "psubusb %%mm1, %%mm3 \n\t"
327 "psubusb %%mm1, %%mm4 \n\t"
328 "psubusb %%mm1, %%mm5 \n\t"
329 "movd %%mm2, %0 \n\t"
330 "movd %%mm3, %1 \n\t"
331 "movd %%mm4, %2 \n\t"
332 "movd %%mm5, %3 \n\t"
333 :"+m"(*(uint32_t*)(dst+0*stride)),
334 "+m"(*(uint32_t*)(dst+1*stride)),
335 "+m"(*(uint32_t*)(dst+2*stride)),
336 "+m"(*(uint32_t*)(dst+3*stride))
337 );
338 }
339
340 static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
341 {
342 int dc = (block[0] + 32) >> 6;
343 int y;
344 __asm__ volatile(
345 "movd %0, %%mm0 \n\t"
346 "pshufw $0, %%mm0, %%mm0 \n\t"
347 "pxor %%mm1, %%mm1 \n\t"
348 "psubw %%mm0, %%mm1 \n\t"
349 "packuswb %%mm0, %%mm0 \n\t"
350 "packuswb %%mm1, %%mm1 \n\t"
351 ::"r"(dc)
352 );
353 for(y=2; y--; dst += 4*stride){
354 __asm__ volatile(
355 "movq %0, %%mm2 \n\t"
356 "movq %1, %%mm3 \n\t"
357 "movq %2, %%mm4 \n\t"
358 "movq %3, %%mm5 \n\t"
359 "paddusb %%mm0, %%mm2 \n\t"
360 "paddusb %%mm0, %%mm3 \n\t"
361 "paddusb %%mm0, %%mm4 \n\t"
362 "paddusb %%mm0, %%mm5 \n\t"
363 "psubusb %%mm1, %%mm2 \n\t"
364 "psubusb %%mm1, %%mm3 \n\t"
365 "psubusb %%mm1, %%mm4 \n\t"
366 "psubusb %%mm1, %%mm5 \n\t"
367 "movq %%mm2, %0 \n\t"
368 "movq %%mm3, %1 \n\t"
369 "movq %%mm4, %2 \n\t"
370 "movq %%mm5, %3 \n\t"
371 :"+m"(*(uint64_t*)(dst+0*stride)),
372 "+m"(*(uint64_t*)(dst+1*stride)),
373 "+m"(*(uint64_t*)(dst+2*stride)),
374 "+m"(*(uint64_t*)(dst+3*stride))
375 );
376 }
377 }
378
379 //FIXME this table is a duplicate from h264data.h, and will be removed once the tables from, h264 have been split
380 static const uint8_t scan8[16 + 2*4]={
381 4+1*8, 5+1*8, 4+2*8, 5+2*8,
382 6+1*8, 7+1*8, 6+2*8, 7+2*8,
383 4+3*8, 5+3*8, 4+4*8, 5+4*8,
384 6+3*8, 7+3*8, 6+4*8, 7+4*8,
385 1+1*8, 2+1*8,
386 1+2*8, 2+2*8,
387 1+4*8, 2+4*8,
388 1+5*8, 2+5*8,
389 };
390
391 static void ff_h264_idct_add16_mmx(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
392 int i;
393 for(i=0; i<16; i++){
394 if(nnzc[ scan8[i] ])
395 ff_h264_idct_add_mmx(dst + block_offset[i], block + i*16, stride);
396 }
397 }
398
399 static void ff_h264_idct8_add4_mmx(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
400 int i;
401 for(i=0; i<16; i+=4){
402 if(nnzc[ scan8[i] ])
403 ff_h264_idct8_add_mmx(dst + block_offset[i], block + i*16, stride);
404 }
405 }
406
407
408 static void ff_h264_idct_add16_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
409 int i;
410 for(i=0; i<16; i++){
411 int nnz = nnzc[ scan8[i] ];
412 if(nnz){
413 if(nnz==1 && block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
414 else ff_h264_idct_add_mmx (dst + block_offset[i], block + i*16, stride);
415 }
416 }
417 }
418
419 static void ff_h264_idct_add16intra_mmx(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
420 int i;
421 for(i=0; i<16; i++){
422 if(nnzc[ scan8[i] ] || block[i*16])
423 ff_h264_idct_add_mmx(dst + block_offset[i], block + i*16, stride);
424 }
425 }
426
427 static void ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
428 int i;
429 for(i=0; i<16; i++){
430 if(nnzc[ scan8[i] ]) ff_h264_idct_add_mmx (dst + block_offset[i], block + i*16, stride);
431 else if(block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
432 }
433 }
434
435 static void ff_h264_idct8_add4_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
436 int i;
437 for(i=0; i<16; i+=4){
438 int nnz = nnzc[ scan8[i] ];
439 if(nnz){
440 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
441 else ff_h264_idct8_add_mmx (dst + block_offset[i], block + i*16, stride);
442 }
443 }
444 }
445
446 static void ff_h264_idct8_add4_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
447 int i;
448 for(i=0; i<16; i+=4){
449 int nnz = nnzc[ scan8[i] ];
450 if(nnz){
451 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
452 else ff_h264_idct8_add_sse2 (dst + block_offset[i], block + i*16, stride);
453 }
454 }
455 }
456
457 static void ff_h264_idct_add8_mmx(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
458 int i;
459 for(i=16; i<16+8; i++){
460 if(nnzc[ scan8[i] ] || block[i*16])
461 ff_h264_idct_add_mmx (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
462 }
463 }
464
465 static void ff_h264_idct_add8_mmx2(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
466 int i;
467 for(i=16; i<16+8; i++){
468 if(nnzc[ scan8[i] ])
469 ff_h264_idct_add_mmx (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
470 else if(block[i*16])
471 ff_h264_idct_dc_add_mmx2(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
472 }
473 }
474
475 #if CONFIG_GPL && HAVE_YASM
476 static void ff_h264_idct_dc_add8_mmx2(uint8_t *dst, int16_t *block, int stride)
477 {
478 __asm__ volatile(
479 "movd %0, %%mm0 \n\t" // 0 0 X D
480 "punpcklwd %1, %%mm0 \n\t" // x X d D
481 "paddsw %2, %%mm0 \n\t"
482 "psraw $6, %%mm0 \n\t"
483 "punpcklwd %%mm0, %%mm0 \n\t" // d d D D
484 "pxor %%mm1, %%mm1 \n\t" // 0 0 0 0
485 "psubw %%mm0, %%mm1 \n\t" // -d-d-D-D
486 "packuswb %%mm1, %%mm0 \n\t" // -d-d-D-D d d D D
487 "pshufw $0xFA, %%mm0, %%mm1 \n\t" // -d-d-d-d-D-D-D-D
488 "punpcklwd %%mm0, %%mm0 \n\t" // d d d d D D D D
489 ::"m"(block[ 0]),
490 "m"(block[16]),
491 "m"(ff_pw_32)
492 );
493 __asm__ volatile(
494 "movq %0, %%mm2 \n\t"
495 "movq %1, %%mm3 \n\t"
496 "movq %2, %%mm4 \n\t"
497 "movq %3, %%mm5 \n\t"
498 "paddusb %%mm0, %%mm2 \n\t"
499 "paddusb %%mm0, %%mm3 \n\t"
500 "paddusb %%mm0, %%mm4 \n\t"
501 "paddusb %%mm0, %%mm5 \n\t"
502 "psubusb %%mm1, %%mm2 \n\t"
503 "psubusb %%mm1, %%mm3 \n\t"
504 "psubusb %%mm1, %%mm4 \n\t"
505 "psubusb %%mm1, %%mm5 \n\t"
506 "movq %%mm2, %0 \n\t"
507 "movq %%mm3, %1 \n\t"
508 "movq %%mm4, %2 \n\t"
509 "movq %%mm5, %3 \n\t"
510 :"+m"(*(uint64_t*)(dst+0*stride)),
511 "+m"(*(uint64_t*)(dst+1*stride)),
512 "+m"(*(uint64_t*)(dst+2*stride)),
513 "+m"(*(uint64_t*)(dst+3*stride))
514 );
515 }
516
517 extern void ff_x264_add8x4_idct_sse2(uint8_t *dst, int16_t *block, int stride);
518
519 static void ff_h264_idct_add16_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
520 int i;
521 for(i=0; i<16; i+=2)
522 if(nnzc[ scan8[i+0] ]|nnzc[ scan8[i+1] ])
523 ff_x264_add8x4_idct_sse2 (dst + block_offset[i], block + i*16, stride);
524 }
525
526 static void ff_h264_idct_add16intra_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
527 int i;
528 for(i=0; i<16; i+=2){
529 if(nnzc[ scan8[i+0] ]|nnzc[ scan8[i+1] ])
530 ff_x264_add8x4_idct_sse2 (dst + block_offset[i], block + i*16, stride);
531 else if(block[i*16]|block[i*16+16])
532 ff_h264_idct_dc_add8_mmx2(dst + block_offset[i], block + i*16, stride);
533 }
534 }
535
536 static void ff_h264_idct_add8_sse2(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
537 int i;
538 for(i=16; i<16+8; i+=2){
539 if(nnzc[ scan8[i+0] ]|nnzc[ scan8[i+1] ])
540 ff_x264_add8x4_idct_sse2 (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
541 else if(block[i*16]|block[i*16+16])
542 ff_h264_idct_dc_add8_mmx2(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
543 }
544 }
545 #endif
546
547 /***********************************/
548 /* deblocking */
549
550 // out: o = |x-y|>a
551 // clobbers: t
552 #define DIFF_GT_MMX(x,y,a,o,t)\
553 "movq "#y", "#t" \n\t"\
554 "movq "#x", "#o" \n\t"\
555 "psubusb "#x", "#t" \n\t"\
556 "psubusb "#y", "#o" \n\t"\
557 "por "#t", "#o" \n\t"\
558 "psubusb "#a", "#o" \n\t"
559
560 // out: o = |x-y|>a
561 // clobbers: t
562 #define DIFF_GT2_MMX(x,y,a,o,t)\
563 "movq "#y", "#t" \n\t"\
564 "movq "#x", "#o" \n\t"\
565 "psubusb "#x", "#t" \n\t"\
566 "psubusb "#y", "#o" \n\t"\
567 "psubusb "#a", "#t" \n\t"\
568 "psubusb "#a", "#o" \n\t"\
569 "pcmpeqb "#t", "#o" \n\t"\
570
571 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
572 // out: mm5=beta-1, mm7=mask
573 // clobbers: mm4,mm6
574 #define H264_DEBLOCK_MASK(alpha1, beta1) \
575 "pshufw $0, "#alpha1", %%mm4 \n\t"\
576 "pshufw $0, "#beta1 ", %%mm5 \n\t"\
577 "packuswb %%mm4, %%mm4 \n\t"\
578 "packuswb %%mm5, %%mm5 \n\t"\
579 DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
580 DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
581 "por %%mm4, %%mm7 \n\t"\
582 DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
583 "por %%mm4, %%mm7 \n\t"\
584 "pxor %%mm6, %%mm6 \n\t"\
585 "pcmpeqb %%mm6, %%mm7 \n\t"
586
587 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
588 // out: mm1=p0' mm2=q0'
589 // clobbers: mm0,3-6
590 #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
591 "movq %%mm1 , %%mm5 \n\t"\
592 "pxor %%mm2 , %%mm5 \n\t" /* p0^q0*/\
593 "pand "#pb_01" , %%mm5 \n\t" /* (p0^q0)&1*/\
594 "pcmpeqb %%mm4 , %%mm4 \n\t"\
595 "pxor %%mm4 , %%mm3 \n\t"\
596 "pavgb %%mm0 , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
597 "pavgb "MANGLE(ff_pb_3)" , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
598 "pxor %%mm1 , %%mm4 \n\t"\
599 "pavgb %%mm2 , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
600 "pavgb %%mm5 , %%mm3 \n\t"\
601 "paddusb %%mm4 , %%mm3 \n\t" /* d+128+33*/\
602 "movq "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
603 "psubusb %%mm3 , %%mm6 \n\t"\
604 "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
605 "pminub %%mm7 , %%mm6 \n\t"\
606 "pminub %%mm7 , %%mm3 \n\t"\
607 "psubusb %%mm6 , %%mm1 \n\t"\
608 "psubusb %%mm3 , %%mm2 \n\t"\
609 "paddusb %%mm3 , %%mm1 \n\t"\
610 "paddusb %%mm6 , %%mm2 \n\t"
611
612 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
613 // out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
614 // clobbers: q2, tmp, tc0
615 #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
616 "movq %%mm1, "#tmp" \n\t"\
617 "pavgb %%mm2, "#tmp" \n\t"\
618 "pavgb "#tmp", "#q2" \n\t" /* avg(p2,avg(p0,q0)) */\
619 "pxor "q2addr", "#tmp" \n\t"\
620 "pand %9, "#tmp" \n\t" /* (p2^avg(p0,q0))&1 */\
621 "psubusb "#tmp", "#q2" \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
622 "movq "#p1", "#tmp" \n\t"\
623 "psubusb "#tc0", "#tmp" \n\t"\
624 "paddusb "#p1", "#tc0" \n\t"\
625 "pmaxub "#tmp", "#q2" \n\t"\
626 "pminub "#tc0", "#q2" \n\t"\
627 "movq "#q2", "q1addr" \n\t"
628
629 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
630 {
631 DECLARE_ALIGNED(8, uint64_t, tmp0)[2];
632
633 __asm__ volatile(
634 "movq (%2,%4), %%mm0 \n\t" //p1
635 "movq (%2,%4,2), %%mm1 \n\t" //p0
636 "movq (%3), %%mm2 \n\t" //q0
637 "movq (%3,%4), %%mm3 \n\t" //q1
638 H264_DEBLOCK_MASK(%7, %8)
639
640 "movd %6, %%mm4 \n\t"
641 "punpcklbw %%mm4, %%mm4 \n\t"
642 "punpcklwd %%mm4, %%mm4 \n\t"
643 "pcmpeqb %%mm3, %%mm3 \n\t"
644 "movq %%mm4, %%mm6 \n\t"
645 "pcmpgtb %%mm3, %%mm4 \n\t"
646 "movq %%mm6, %1 \n\t"
647 "pand %%mm4, %%mm7 \n\t"
648 "movq %%mm7, %0 \n\t"
649
650 /* filter p1 */
651 "movq (%2), %%mm3 \n\t" //p2
652 DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
653 "pand %%mm7, %%mm6 \n\t" // mask & |p2-p0|<beta
654 "pand %1, %%mm7 \n\t" // mask & tc0
655 "movq %%mm7, %%mm4 \n\t"
656 "psubb %%mm6, %%mm7 \n\t"
657 "pand %%mm4, %%mm6 \n\t" // mask & |p2-p0|<beta & tc0
658 H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%2)", "(%2,%4)", %%mm6, %%mm4)
659
660 /* filter q1 */
661 "movq (%3,%4,2), %%mm4 \n\t" //q2
662 DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
663 "pand %0, %%mm6 \n\t"
664 "movq %1, %%mm5 \n\t" // can be merged with the and below but is slower then
665 "pand %%mm6, %%mm5 \n\t"
666 "psubb %%mm6, %%mm7 \n\t"
667 "movq (%3,%4), %%mm3 \n\t"
668 H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%3,%4,2)", "(%3,%4)", %%mm5, %%mm6)
669
670 /* filter p0, q0 */
671 H264_DEBLOCK_P0_Q0(%9, unused)
672 "movq %%mm1, (%2,%4,2) \n\t"
673 "movq %%mm2, (%3) \n\t"
674
675 : "=m"(tmp0[0]), "=m"(tmp0[1])
676 : "r"(pix-3*stride), "r"(pix), "r"((x86_reg)stride),
677 "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
678 "m"(ff_bone)
679 );
680 }
681
682 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
683 {
684 if((tc0[0] & tc0[1]) >= 0)
685 h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
686 if((tc0[2] & tc0[3]) >= 0)
687 h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
688 }
689 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
690 {
691 //FIXME: could cut some load/stores by merging transpose with filter
692 // also, it only needs to transpose 6x8
693 DECLARE_ALIGNED(8, uint8_t, trans)[8*8];
694 int i;
695 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
696 if((tc0[0] & tc0[1]) < 0)
697 continue;
698 transpose4x4(trans, pix-4, 8, stride);
699 transpose4x4(trans +4*8, pix, 8, stride);
700 transpose4x4(trans+4, pix-4+4*stride, 8, stride);
701 transpose4x4(trans+4+4*8, pix +4*stride, 8, stride);
702 h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
703 transpose4x4(pix-2, trans +2*8, stride, 8);
704 transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
705 }
706 }
707
708 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
709 {
710 __asm__ volatile(
711 "movq (%0), %%mm0 \n\t" //p1
712 "movq (%0,%2), %%mm1 \n\t" //p0
713 "movq (%1), %%mm2 \n\t" //q0
714 "movq (%1,%2), %%mm3 \n\t" //q1
715 H264_DEBLOCK_MASK(%4, %5)
716 "movd %3, %%mm6 \n\t"
717 "punpcklbw %%mm6, %%mm6 \n\t"
718 "pand %%mm6, %%mm7 \n\t" // mm7 = tc&mask
719 H264_DEBLOCK_P0_Q0(%6, %7)
720 "movq %%mm1, (%0,%2) \n\t"
721 "movq %%mm2, (%1) \n\t"
722
723 :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
724 "r"(*(uint32_t*)tc0),
725 "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
726 );
727 }
728
729 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
730 {
731 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
732 }
733
734 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
735 {
736 //FIXME: could cut some load/stores by merging transpose with filter
737 DECLARE_ALIGNED(8, uint8_t, trans)[8*4];
738 transpose4x4(trans, pix-2, 8, stride);
739 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
740 h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
741 transpose4x4(pix-2, trans, stride, 8);
742 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
743 }
744
745 // p0 = (p0 + q1 + 2*p1 + 2) >> 2
746 #define H264_FILTER_CHROMA4(p0, p1, q1, one) \
747 "movq "#p0", %%mm4 \n\t"\
748 "pxor "#q1", %%mm4 \n\t"\
749 "pand "#one", %%mm4 \n\t" /* mm4 = (p0^q1)&1 */\
750 "pavgb "#q1", "#p0" \n\t"\
751 "psubusb %%mm4, "#p0" \n\t"\
752 "pavgb "#p1", "#p0" \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
753
754 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
755 {
756 __asm__ volatile(
757 "movq (%0), %%mm0 \n\t"
758 "movq (%0,%2), %%mm1 \n\t"
759 "movq (%1), %%mm2 \n\t"
760 "movq (%1,%2), %%mm3 \n\t"
761 H264_DEBLOCK_MASK(%3, %4)
762 "movq %%mm1, %%mm5 \n\t"
763 "movq %%mm2, %%mm6 \n\t"
764 H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
765 H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
766 "psubb %%mm5, %%mm1 \n\t"
767 "psubb %%mm6, %%mm2 \n\t"
768 "pand %%mm7, %%mm1 \n\t"
769 "pand %%mm7, %%mm2 \n\t"
770 "paddb %%mm5, %%mm1 \n\t"
771 "paddb %%mm6, %%mm2 \n\t"
772 "movq %%mm1, (%0,%2) \n\t"
773 "movq %%mm2, (%1) \n\t"
774 :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
775 "m"(alpha1), "m"(beta1), "m"(ff_bone)
776 );
777 }
778
779 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
780 {
781 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
782 }
783
784 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
785 {
786 //FIXME: could cut some load/stores by merging transpose with filter
787 DECLARE_ALIGNED(8, uint8_t, trans)[8*4];
788 transpose4x4(trans, pix-2, 8, stride);
789 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
790 h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
791 transpose4x4(pix-2, trans, stride, 8);
792 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
793 }
794
795 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],
796 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
797 int dir;
798 __asm__ volatile(
799 "movq %0, %%mm7 \n"
800 "movq %1, %%mm6 \n"
801 ::"m"(ff_pb_1), "m"(ff_pb_3)
802 );
803 if(field)
804 __asm__ volatile(
805 "movq %0, %%mm6 \n"
806 ::"m"(ff_pb_3_1)
807 );
808 __asm__ volatile(
809 "movq %%mm6, %%mm5 \n"
810 "paddb %%mm5, %%mm5 \n"
811 :);
812
813 // could do a special case for dir==0 && edges==1, but it only reduces the
814 // average filter time by 1.2%
815 for( dir=1; dir>=0; dir-- ) {
816 const x86_reg d_idx = dir ? -8 : -1;
817 const int mask_mv = dir ? mask_mv1 : mask_mv0;
818 DECLARE_ALIGNED(8, const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
819 int b_idx, edge;
820 for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
821 __asm__ volatile(
822 "pand %0, %%mm0 \n\t"
823 ::"m"(mask_dir)
824 );
825 if(!(mask_mv & edge)) {
826 if(bidir) {
827 __asm__ volatile(
828 "movd (%1,%0), %%mm2 \n"
829 "punpckldq 40(%1,%0), %%mm2 \n" // { ref0[bn], ref1[bn] }
830 "pshufw $0x44, (%1), %%mm0 \n" // { ref0[b], ref0[b] }
831 "pshufw $0x44, 40(%1), %%mm1 \n" // { ref1[b], ref1[b] }
832 "pshufw $0x4E, %%mm2, %%mm3 \n"
833 "psubb %%mm2, %%mm0 \n" // { ref0[b]!=ref0[bn], ref0[b]!=ref1[bn] }
834 "psubb %%mm3, %%mm1 \n" // { ref1[b]!=ref1[bn], ref1[b]!=ref0[bn] }
835 "1: \n"
836 "por %%mm1, %%mm0 \n"
837 "movq (%2,%0,4), %%mm1 \n"
838 "movq 8(%2,%0,4), %%mm2 \n"
839 "movq %%mm1, %%mm3 \n"
840 "movq %%mm2, %%mm4 \n"
841 "psubw (%2), %%mm1 \n"
842 "psubw 8(%2), %%mm2 \n"
843 "psubw 160(%2), %%mm3 \n"
844 "psubw 168(%2), %%mm4 \n"
845 "packsswb %%mm2, %%mm1 \n"
846 "packsswb %%mm4, %%mm3 \n"
847 "paddb %%mm6, %%mm1 \n"
848 "paddb %%mm6, %%mm3 \n"
849 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
850 "psubusb %%mm5, %%mm3 \n"
851 "packsswb %%mm3, %%mm1 \n"
852 "add $40, %0 \n"
853 "cmp $40, %0 \n"
854 "jl 1b \n"
855 "sub $80, %0 \n"
856 "pshufw $0x4E, %%mm1, %%mm1 \n"
857 "por %%mm1, %%mm0 \n"
858 "pshufw $0x4E, %%mm0, %%mm1 \n"
859 "pminub %%mm1, %%mm0 \n"
860 ::"r"(d_idx),
861 "r"(ref[0]+b_idx),
862 "r"(mv[0]+b_idx)
863 );
864 } else {
865 __asm__ volatile(
866 "movd (%1), %%mm0 \n"
867 "psubb (%1,%0), %%mm0 \n" // ref[b] != ref[bn]
868 "movq (%2), %%mm1 \n"
869 "movq 8(%2), %%mm2 \n"
870 "psubw (%2,%0,4), %%mm1 \n"
871 "psubw 8(%2,%0,4), %%mm2 \n"
872 "packsswb %%mm2, %%mm1 \n"
873 "paddb %%mm6, %%mm1 \n"
874 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
875 "packsswb %%mm1, %%mm1 \n"
876 "por %%mm1, %%mm0 \n"
877 ::"r"(d_idx),
878 "r"(ref[0]+b_idx),
879 "r"(mv[0]+b_idx)
880 );
881 }
882 }
883 __asm__ volatile(
884 "movd %0, %%mm1 \n"
885 "por %1, %%mm1 \n" // nnz[b] || nnz[bn]
886 ::"m"(nnz[b_idx]),
887 "m"(nnz[b_idx+d_idx])
888 );
889 __asm__ volatile(
890 "pminub %%mm7, %%mm1 \n"
891 "pminub %%mm7, %%mm0 \n"
892 "psllw $1, %%mm1 \n"
893 "pxor %%mm2, %%mm2 \n"
894 "pmaxub %%mm0, %%mm1 \n"
895 "punpcklbw %%mm2, %%mm1 \n"
896 "movq %%mm1, %0 \n"
897 :"=m"(*bS[dir][edge])
898 ::"memory"
899 );
900 }
901 edges = 4;
902 step = 1;
903 }
904 __asm__ volatile(
905 "movq (%0), %%mm0 \n\t"
906 "movq 8(%0), %%mm1 \n\t"
907 "movq 16(%0), %%mm2 \n\t"
908 "movq 24(%0), %%mm3 \n\t"
909 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
910 "movq %%mm0, (%0) \n\t"
911 "movq %%mm3, 8(%0) \n\t"
912 "movq %%mm4, 16(%0) \n\t"
913 "movq %%mm2, 24(%0) \n\t"
914 ::"r"(bS[0])
915 :"memory"
916 );
917 }
918
919 /***********************************/
920 /* motion compensation */
921
922 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
923 "mov"#q" "#C", "#T" \n\t"\
924 "mov"#d" (%0), "#F" \n\t"\
925 "paddw "#D", "#T" \n\t"\
926 "psllw $2, "#T" \n\t"\
927 "psubw "#B", "#T" \n\t"\
928 "psubw "#E", "#T" \n\t"\
929 "punpcklbw "#Z", "#F" \n\t"\
930 "pmullw %4, "#T" \n\t"\
931 "paddw %5, "#A" \n\t"\
932 "add %2, %0 \n\t"\
933 "paddw "#F", "#A" \n\t"\
934 "paddw "#A", "#T" \n\t"\
935 "psraw $5, "#T" \n\t"\
936 "packuswb "#T", "#T" \n\t"\
937 OP(T, (%1), A, d)\
938 "add %3, %1 \n\t"
939
940 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
941 "mov"#q" "#C", "#T" \n\t"\
942 "mov"#d" (%0), "#F" \n\t"\
943 "paddw "#D", "#T" \n\t"\
944 "psllw $2, "#T" \n\t"\
945 "paddw %4, "#A" \n\t"\
946 "psubw "#B", "#T" \n\t"\
947 "psubw "#E", "#T" \n\t"\
948 "punpcklbw "#Z", "#F" \n\t"\
949 "pmullw %3, "#T" \n\t"\
950 "paddw "#F", "#A" \n\t"\
951 "add %2, %0 \n\t"\
952 "paddw "#A", "#T" \n\t"\
953 "mov"#q" "#T", "#OF"(%1) \n\t"
954
955 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
956 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
957 #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)
958 #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)
959
960
961 #define QPEL_H264(OPNAME, OP, MMX)\
962 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
963 int h=4;\
964 \
965 __asm__ volatile(\
966 "pxor %%mm7, %%mm7 \n\t"\
967 "movq "MANGLE(ff_pw_5) ", %%mm4\n\t"\
968 "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
969 "1: \n\t"\
970 "movd -1(%0), %%mm1 \n\t"\
971 "movd (%0), %%mm2 \n\t"\
972 "movd 1(%0), %%mm3 \n\t"\
973 "movd 2(%0), %%mm0 \n\t"\
974 "punpcklbw %%mm7, %%mm1 \n\t"\
975 "punpcklbw %%mm7, %%mm2 \n\t"\
976 "punpcklbw %%mm7, %%mm3 \n\t"\
977 "punpcklbw %%mm7, %%mm0 \n\t"\
978 "paddw %%mm0, %%mm1 \n\t"\
979 "paddw %%mm3, %%mm2 \n\t"\
980 "movd -2(%0), %%mm0 \n\t"\
981 "movd 3(%0), %%mm3 \n\t"\
982 "punpcklbw %%mm7, %%mm0 \n\t"\
983 "punpcklbw %%mm7, %%mm3 \n\t"\
984 "paddw %%mm3, %%mm0 \n\t"\
985 "psllw $2, %%mm2 \n\t"\
986 "psubw %%mm1, %%mm2 \n\t"\
987 "pmullw %%mm4, %%mm2 \n\t"\
988 "paddw %%mm5, %%mm0 \n\t"\
989 "paddw %%mm2, %%mm0 \n\t"\
990 "psraw $5, %%mm0 \n\t"\
991 "packuswb %%mm0, %%mm0 \n\t"\
992 OP(%%mm0, (%1),%%mm6, d)\
993 "add %3, %0 \n\t"\
994 "add %4, %1 \n\t"\
995 "decl %2 \n\t"\
996 " jnz 1b \n\t"\
997 : "+a"(src), "+c"(dst), "+g"(h)\
998 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
999 : "memory"\
1000 );\
1001 }\
1002 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1003 int h=4;\
1004 __asm__ volatile(\
1005 "pxor %%mm7, %%mm7 \n\t"\
1006 "movq %0, %%mm4 \n\t"\
1007 "movq %1, %%mm5 \n\t"\
1008 :: "m"(ff_pw_5), "m"(ff_pw_16)\
1009 );\
1010 do{\
1011 __asm__ volatile(\
1012 "movd -1(%0), %%mm1 \n\t"\
1013 "movd (%0), %%mm2 \n\t"\
1014 "movd 1(%0), %%mm3 \n\t"\
1015 "movd 2(%0), %%mm0 \n\t"\
1016 "punpcklbw %%mm7, %%mm1 \n\t"\
1017 "punpcklbw %%mm7, %%mm2 \n\t"\
1018 "punpcklbw %%mm7, %%mm3 \n\t"\
1019 "punpcklbw %%mm7, %%mm0 \n\t"\
1020 "paddw %%mm0, %%mm1 \n\t"\
1021 "paddw %%mm3, %%mm2 \n\t"\
1022 "movd -2(%0), %%mm0 \n\t"\
1023 "movd 3(%0), %%mm3 \n\t"\
1024 "punpcklbw %%mm7, %%mm0 \n\t"\
1025 "punpcklbw %%mm7, %%mm3 \n\t"\
1026 "paddw %%mm3, %%mm0 \n\t"\
1027 "psllw $2, %%mm2 \n\t"\
1028 "psubw %%mm1, %%mm2 \n\t"\
1029 "pmullw %%mm4, %%mm2 \n\t"\
1030 "paddw %%mm5, %%mm0 \n\t"\
1031 "paddw %%mm2, %%mm0 \n\t"\
1032 "movd (%2), %%mm3 \n\t"\
1033 "psraw $5, %%mm0 \n\t"\
1034 "packuswb %%mm0, %%mm0 \n\t"\
1035 PAVGB" %%mm3, %%mm0 \n\t"\
1036 OP(%%mm0, (%1),%%mm6, d)\
1037 "add %4, %0 \n\t"\
1038 "add %4, %1 \n\t"\
1039 "add %3, %2 \n\t"\
1040 : "+a"(src), "+c"(dst), "+d"(src2)\
1041 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
1042 : "memory"\
1043 );\
1044 }while(--h);\
1045 }\
1046 static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1047 src -= 2*srcStride;\
1048 __asm__ volatile(\
1049 "pxor %%mm7, %%mm7 \n\t"\
1050 "movd (%0), %%mm0 \n\t"\
1051 "add %2, %0 \n\t"\
1052 "movd (%0), %%mm1 \n\t"\
1053 "add %2, %0 \n\t"\
1054 "movd (%0), %%mm2 \n\t"\
1055 "add %2, %0 \n\t"\
1056 "movd (%0), %%mm3 \n\t"\
1057 "add %2, %0 \n\t"\
1058 "movd (%0), %%mm4 \n\t"\
1059 "add %2, %0 \n\t"\
1060 "punpcklbw %%mm7, %%mm0 \n\t"\
1061 "punpcklbw %%mm7, %%mm1 \n\t"\
1062 "punpcklbw %%mm7, %%mm2 \n\t"\
1063 "punpcklbw %%mm7, %%mm3 \n\t"\
1064 "punpcklbw %%mm7, %%mm4 \n\t"\
1065 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1066 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1067 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1068 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1069 \
1070 : "+a"(src), "+c"(dst)\
1071 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1072 : "memory"\
1073 );\
1074 }\
1075 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){\
1076 int h=4;\
1077 int w=3;\
1078 src -= 2*srcStride+2;\
1079 while(w--){\
1080 __asm__ volatile(\
1081 "pxor %%mm7, %%mm7 \n\t"\
1082 "movd (%0), %%mm0 \n\t"\
1083 "add %2, %0 \n\t"\
1084 "movd (%0), %%mm1 \n\t"\
1085 "add %2, %0 \n\t"\
1086 "movd (%0), %%mm2 \n\t"\
1087 "add %2, %0 \n\t"\
1088 "movd (%0), %%mm3 \n\t"\
1089 "add %2, %0 \n\t"\
1090 "movd (%0), %%mm4 \n\t"\
1091 "add %2, %0 \n\t"\
1092 "punpcklbw %%mm7, %%mm0 \n\t"\
1093 "punpcklbw %%mm7, %%mm1 \n\t"\
1094 "punpcklbw %%mm7, %%mm2 \n\t"\
1095 "punpcklbw %%mm7, %%mm3 \n\t"\
1096 "punpcklbw %%mm7, %%mm4 \n\t"\
1097 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
1098 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
1099 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
1100 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
1101 \
1102 : "+a"(src)\
1103 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1104 : "memory"\
1105 );\
1106 tmp += 4;\
1107 src += 4 - 9*srcStride;\
1108 }\
1109 tmp -= 3*4;\
1110 __asm__ volatile(\
1111 "1: \n\t"\
1112 "movq (%0), %%mm0 \n\t"\
1113 "paddw 10(%0), %%mm0 \n\t"\
1114 "movq 2(%0), %%mm1 \n\t"\
1115 "paddw 8(%0), %%mm1 \n\t"\
1116 "movq 4(%0), %%mm2 \n\t"\
1117 "paddw 6(%0), %%mm2 \n\t"\
1118 "psubw %%mm1, %%mm0 \n\t"/*a-b (abccba)*/\
1119 "psraw $2, %%mm0 \n\t"/*(a-b)/4 */\
1120 "psubw %%mm1, %%mm0 \n\t"/*(a-b)/4-b */\
1121 "paddsw %%mm2, %%mm0 \n\t"\
1122 "psraw $2, %%mm0 \n\t"/*((a-b)/4-b+c)/4 */\
1123 "paddw %%mm2, %%mm0 \n\t"/*(a-5*b+20*c)/16 */\
1124 "psraw $6, %%mm0 \n\t"\
1125 "packuswb %%mm0, %%mm0 \n\t"\
1126 OP(%%mm0, (%1),%%mm7, d)\
1127 "add $24, %0 \n\t"\
1128 "add %3, %1 \n\t"\
1129 "decl %2 \n\t"\
1130 " jnz 1b \n\t"\
1131 : "+a"(tmp), "+c"(dst), "+g"(h)\
1132 : "S"((x86_reg)dstStride)\
1133 : "memory"\
1134 );\
1135 }\
1136 \
1137 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1138 int h=8;\
1139 __asm__ volatile(\
1140 "pxor %%mm7, %%mm7 \n\t"\
1141 "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
1142 "1: \n\t"\
1143 "movq (%0), %%mm0 \n\t"\
1144 "movq 1(%0), %%mm2 \n\t"\
1145 "movq %%mm0, %%mm1 \n\t"\
1146 "movq %%mm2, %%mm3 \n\t"\
1147 "punpcklbw %%mm7, %%mm0 \n\t"\
1148 "punpckhbw %%mm7, %%mm1 \n\t"\
1149 "punpcklbw %%mm7, %%mm2 \n\t"\
1150 "punpckhbw %%mm7, %%mm3 \n\t"\
1151 "paddw %%mm2, %%mm0 \n\t"\
1152 "paddw %%mm3, %%mm1 \n\t"\
1153 "psllw $2, %%mm0 \n\t"\
1154 "psllw $2, %%mm1 \n\t"\
1155 "movq -1(%0), %%mm2 \n\t"\
1156 "movq 2(%0), %%mm4 \n\t"\
1157 "movq %%mm2, %%mm3 \n\t"\
1158 "movq %%mm4, %%mm5 \n\t"\
1159 "punpcklbw %%mm7, %%mm2 \n\t"\
1160 "punpckhbw %%mm7, %%mm3 \n\t"\
1161 "punpcklbw %%mm7, %%mm4 \n\t"\
1162 "punpckhbw %%mm7, %%mm5 \n\t"\
1163 "paddw %%mm4, %%mm2 \n\t"\
1164 "paddw %%mm3, %%mm5 \n\t"\
1165 "psubw %%mm2, %%mm0 \n\t"\
1166 "psubw %%mm5, %%mm1 \n\t"\
1167 "pmullw %%mm6, %%mm0 \n\t"\
1168 "pmullw %%mm6, %%mm1 \n\t"\
1169 "movd -2(%0), %%mm2 \n\t"\
1170 "movd 7(%0), %%mm5 \n\t"\
1171 "punpcklbw %%mm7, %%mm2 \n\t"\
1172 "punpcklbw %%mm7, %%mm5 \n\t"\
1173 "paddw %%mm3, %%mm2 \n\t"\
1174 "paddw %%mm5, %%mm4 \n\t"\
1175 "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
1176 "paddw %%mm5, %%mm2 \n\t"\
1177 "paddw %%mm5, %%mm4 \n\t"\
1178 "paddw %%mm2, %%mm0 \n\t"\
1179 "paddw %%mm4, %%mm1 \n\t"\
1180 "psraw $5, %%mm0 \n\t"\
1181 "psraw $5, %%mm1 \n\t"\
1182 "packuswb %%mm1, %%mm0 \n\t"\
1183 OP(%%mm0, (%1),%%mm5, q)\
1184 "add %3, %0 \n\t"\
1185 "add %4, %1 \n\t"\
1186 "decl %2 \n\t"\
1187 " jnz 1b \n\t"\
1188 : "+a"(src), "+c"(dst), "+g"(h)\
1189 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
1190 : "memory"\
1191 );\
1192 }\
1193 \
1194 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1195 int h=8;\
1196 __asm__ volatile(\
1197 "pxor %%mm7, %%mm7 \n\t"\
1198 "movq %0, %%mm6 \n\t"\
1199 :: "m"(ff_pw_5)\
1200 );\
1201 do{\
1202 __asm__ volatile(\
1203 "movq (%0), %%mm0 \n\t"\
1204 "movq 1(%0), %%mm2 \n\t"\
1205 "movq %%mm0, %%mm1 \n\t"\
1206 "movq %%mm2, %%mm3 \n\t"\
1207 "punpcklbw %%mm7, %%mm0 \n\t"\
1208 "punpckhbw %%mm7, %%mm1 \n\t"\
1209 "punpcklbw %%mm7, %%mm2 \n\t"\
1210 "punpckhbw %%mm7, %%mm3 \n\t"\
1211 "paddw %%mm2, %%mm0 \n\t"\
1212 "paddw %%mm3, %%mm1 \n\t"\
1213 "psllw $2, %%mm0 \n\t"\
1214 "psllw $2, %%mm1 \n\t"\
1215 "movq -1(%0), %%mm2 \n\t"\
1216 "movq 2(%0), %%mm4 \n\t"\
1217 "movq %%mm2, %%mm3 \n\t"\
1218 "movq %%mm4, %%mm5 \n\t"\
1219 "punpcklbw %%mm7, %%mm2 \n\t"\
1220 "punpckhbw %%mm7, %%mm3 \n\t"\
1221 "punpcklbw %%mm7, %%mm4 \n\t"\
1222 "punpckhbw %%mm7, %%mm5 \n\t"\
1223 "paddw %%mm4, %%mm2 \n\t"\
1224 "paddw %%mm3, %%mm5 \n\t"\
1225 "psubw %%mm2, %%mm0 \n\t"\
1226 "psubw %%mm5, %%mm1 \n\t"\
1227 "pmullw %%mm6, %%mm0 \n\t"\
1228 "pmullw %%mm6, %%mm1 \n\t"\
1229 "movd -2(%0), %%mm2 \n\t"\
1230 "movd 7(%0), %%mm5 \n\t"\
1231 "punpcklbw %%mm7, %%mm2 \n\t"\
1232 "punpcklbw %%mm7, %%mm5 \n\t"\
1233 "paddw %%mm3, %%mm2 \n\t"\
1234 "paddw %%mm5, %%mm4 \n\t"\
1235 "movq %5, %%mm5 \n\t"\
1236 "paddw %%mm5, %%mm2 \n\t"\
1237 "paddw %%mm5, %%mm4 \n\t"\
1238 "paddw %%mm2, %%mm0 \n\t"\
1239 "paddw %%mm4, %%mm1 \n\t"\
1240 "psraw $5, %%mm0 \n\t"\
1241 "psraw $5, %%mm1 \n\t"\
1242 "movq (%2), %%mm4 \n\t"\
1243 "packuswb %%mm1, %%mm0 \n\t"\
1244 PAVGB" %%mm4, %%mm0 \n\t"\
1245 OP(%%mm0, (%1),%%mm5, q)\
1246 "add %4, %0 \n\t"\
1247 "add %4, %1 \n\t"\
1248 "add %3, %2 \n\t"\
1249 : "+a"(src), "+c"(dst), "+d"(src2)\
1250 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1251 "m"(ff_pw_16)\
1252 : "memory"\
1253 );\
1254 }while(--h);\
1255 }\
1256 \
1257 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1258 int w= 2;\
1259 src -= 2*srcStride;\
1260 \
1261 while(w--){\
1262 __asm__ volatile(\
1263 "pxor %%mm7, %%mm7 \n\t"\
1264 "movd (%0), %%mm0 \n\t"\
1265 "add %2, %0 \n\t"\
1266 "movd (%0), %%mm1 \n\t"\
1267 "add %2, %0 \n\t"\
1268 "movd (%0), %%mm2 \n\t"\
1269 "add %2, %0 \n\t"\
1270 "movd (%0), %%mm3 \n\t"\
1271 "add %2, %0 \n\t"\
1272 "movd (%0), %%mm4 \n\t"\
1273 "add %2, %0 \n\t"\
1274 "punpcklbw %%mm7, %%mm0 \n\t"\
1275 "punpcklbw %%mm7, %%mm1 \n\t"\
1276 "punpcklbw %%mm7, %%mm2 \n\t"\
1277 "punpcklbw %%mm7, %%mm3 \n\t"\
1278 "punpcklbw %%mm7, %%mm4 \n\t"\
1279 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1280 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1281 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1282 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1283 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1284 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1285 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1286 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1287 \
1288 : "+a"(src), "+c"(dst)\
1289 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1290 : "memory"\
1291 );\
1292 if(h==16){\
1293 __asm__ volatile(\
1294 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1295 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1296 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
1297 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
1298 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
1299 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
1300 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
1301 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
1302 \
1303 : "+a"(src), "+c"(dst)\
1304 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1305 : "memory"\
1306 );\
1307 }\
1308 src += 4-(h+5)*srcStride;\
1309 dst += 4-h*dstStride;\
1310 }\
1311 }\
1312 static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
1313 int w = (size+8)>>2;\
1314 src -= 2*srcStride+2;\
1315 while(w--){\
1316 __asm__ volatile(\
1317 "pxor %%mm7, %%mm7 \n\t"\
1318 "movd (%0), %%mm0 \n\t"\
1319 "add %2, %0 \n\t"\
1320 "movd (%0), %%mm1 \n\t"\
1321 "add %2, %0 \n\t"\
1322 "movd (%0), %%mm2 \n\t"\
1323 "add %2, %0 \n\t"\
1324 "movd (%0), %%mm3 \n\t"\
1325 "add %2, %0 \n\t"\
1326 "movd (%0), %%mm4 \n\t"\
1327 "add %2, %0 \n\t"\
1328 "punpcklbw %%mm7, %%mm0 \n\t"\
1329 "punpcklbw %%mm7, %%mm1 \n\t"\
1330 "punpcklbw %%mm7, %%mm2 \n\t"\
1331 "punpcklbw %%mm7, %%mm3 \n\t"\
1332 "punpcklbw %%mm7, %%mm4 \n\t"\
1333 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
1334 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
1335 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
1336 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
1337 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
1338 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
1339 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
1340 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
1341 : "+a"(src)\
1342 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1343 : "memory"\
1344 );\
1345 if(size==16){\
1346 __asm__ volatile(\
1347 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 8*48)\
1348 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 9*48)\
1349 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
1350 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
1351 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
1352 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
1353 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
1354 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
1355 : "+a"(src)\
1356 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1357 : "memory"\
1358 );\
1359 }\
1360 tmp += 4;\
1361 src += 4 - (size+5)*srcStride;\
1362 }\
1363 }\
1364 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1365 int w = size>>4;\
1366 do{\
1367 int h = size;\
1368 __asm__ volatile(\
1369 "1: \n\t"\
1370 "movq (%0), %%mm0 \n\t"\
1371 "movq 8(%0), %%mm3 \n\t"\
1372 "movq 2(%0), %%mm1 \n\t"\
1373 "movq 10(%0), %%mm4 \n\t"\
1374 "paddw %%mm4, %%mm0 \n\t"\
1375 "paddw %%mm3, %%mm1 \n\t"\
1376 "paddw 18(%0), %%mm3 \n\t"\
1377 "paddw 16(%0), %%mm4 \n\t"\
1378 "movq 4(%0), %%mm2 \n\t"\
1379 "movq 12(%0), %%mm5 \n\t"\
1380 "paddw 6(%0), %%mm2 \n\t"\
1381 "paddw 14(%0), %%mm5 \n\t"\
1382 "psubw %%mm1, %%mm0 \n\t"\
1383 "psubw %%mm4, %%mm3 \n\t"\
1384 "psraw $2, %%mm0 \n\t"\
1385 "psraw $2, %%mm3 \n\t"\
1386 "psubw %%mm1, %%mm0 \n\t"\
1387 "psubw %%mm4, %%mm3 \n\t"\
1388 "paddsw %%mm2, %%mm0 \n\t"\
1389 "paddsw %%mm5, %%mm3 \n\t"\
1390 "psraw $2, %%mm0 \n\t"\
1391 "psraw $2, %%mm3 \n\t"\
1392 "paddw %%mm2, %%mm0 \n\t"\
1393 "paddw %%mm5, %%mm3 \n\t"\
1394 "psraw $6, %%mm0 \n\t"\
1395 "psraw $6, %%mm3 \n\t"\
1396 "packuswb %%mm3, %%mm0 \n\t"\
1397 OP(%%mm0, (%1),%%mm7, q)\
1398 "add $48, %0 \n\t"\
1399 "add %3, %1 \n\t"\
1400 "decl %2 \n\t"\
1401 " jnz 1b \n\t"\
1402 : "+a"(tmp), "+c"(dst), "+g"(h)\
1403 : "S"((x86_reg)dstStride)\
1404 : "memory"\
1405 );\
1406 tmp += 8 - size*24;\
1407 dst += 8 - size*dstStride;\
1408 }while(w--);\
1409 }\
1410 \
1411 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1412 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
1413 }\
1414 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1415 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
1416 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1417 }\
1418 \
1419 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1420 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1421 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1422 src += 8*srcStride;\
1423 dst += 8*dstStride;\
1424 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1425 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1426 }\
1427 \
1428 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1429 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1430 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1431 src += 8*dstStride;\
1432 dst += 8*dstStride;\
1433 src2 += 8*src2Stride;\
1434 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1435 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1436 }\
1437 \
1438 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){\
1439 put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
1440 OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1441 }\
1442 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1443 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
1444 }\
1445 \
1446 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1447 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
1448 }\
1449 \
1450 static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1451 {\
1452 __asm__ volatile(\
1453 "movq (%1), %%mm0 \n\t"\
1454 "movq 24(%1), %%mm1 \n\t"\
1455 "psraw $5, %%mm0 \n\t"\
1456 "psraw $5, %%mm1 \n\t"\
1457 "packuswb %%mm0, %%mm0 \n\t"\
1458 "packuswb %%mm1, %%mm1 \n\t"\
1459 PAVGB" (%0), %%mm0 \n\t"\
1460 PAVGB" (%0,%3), %%mm1 \n\t"\
1461 OP(%%mm0, (%2), %%mm4, d)\
1462 OP(%%mm1, (%2,%4), %%mm5, d)\
1463 "lea (%0,%3,2), %0 \n\t"\
1464 "lea (%2,%4,2), %2 \n\t"\
1465 "movq 48(%1), %%mm0 \n\t"\
1466 "movq 72(%1), %%mm1 \n\t"\
1467 "psraw $5, %%mm0 \n\t"\
1468 "psraw $5, %%mm1 \n\t"\
1469 "packuswb %%mm0, %%mm0 \n\t"\
1470 "packuswb %%mm1, %%mm1 \n\t"\
1471 PAVGB" (%0), %%mm0 \n\t"\
1472 PAVGB" (%0,%3), %%mm1 \n\t"\
1473 OP(%%mm0, (%2), %%mm4, d)\
1474 OP(%%mm1, (%2,%4), %%mm5, d)\
1475 :"+a"(src8), "+c"(src16), "+d"(dst)\
1476 :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
1477 :"memory");\
1478 }\
1479 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1480 {\
1481 do{\
1482 __asm__ volatile(\
1483 "movq (%1), %%mm0 \n\t"\
1484 "movq 8(%1), %%mm1 \n\t"\
1485 "movq 48(%1), %%mm2 \n\t"\
1486 "movq 8+48(%1), %%mm3 \n\t"\
1487 "psraw $5, %%mm0 \n\t"\
1488 "psraw $5, %%mm1 \n\t"\
1489 "psraw $5, %%mm2 \n\t"\
1490 "psraw $5, %%mm3 \n\t"\
1491 "packuswb %%mm1, %%mm0 \n\t"\
1492 "packuswb %%mm3, %%mm2 \n\t"\
1493 PAVGB" (%0), %%mm0 \n\t"\
1494 PAVGB" (%0,%3), %%mm2 \n\t"\
1495 OP(%%mm0, (%2), %%mm5, q)\
1496 OP(%%mm2, (%2,%4), %%mm5, q)\
1497 ::"a"(src8), "c"(src16), "d"(dst),\
1498 "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
1499 :"memory");\
1500 src8 += 2L*src8Stride;\
1501 src16 += 48;\
1502 dst += 2L*dstStride;\
1503 }while(h-=2);\
1504 }\
1505 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1506 {\
1507 OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
1508 OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1509 }\
1510
1511
1512 #if ARCH_X86_64
1513 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1514 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1515 int h=16;\
1516 __asm__ volatile(\
1517 "pxor %%xmm15, %%xmm15 \n\t"\
1518 "movdqa %6, %%xmm14 \n\t"\
1519 "movdqa %7, %%xmm13 \n\t"\
1520 "1: \n\t"\
1521 "lddqu 6(%0), %%xmm1 \n\t"\
1522 "lddqu -2(%0), %%xmm7 \n\t"\
1523 "movdqa %%xmm1, %%xmm0 \n\t"\
1524 "punpckhbw %%xmm15, %%xmm1 \n\t"\
1525 "punpcklbw %%xmm15, %%xmm0 \n\t"\
1526 "punpcklbw %%xmm15, %%xmm7 \n\t"\
1527 "movdqa %%xmm1, %%xmm2 \n\t"\
1528 "movdqa %%xmm0, %%xmm6 \n\t"\
1529 "movdqa %%xmm1, %%xmm3 \n\t"\
1530 "movdqa %%xmm0, %%xmm8 \n\t"\
1531 "movdqa %%xmm1, %%xmm4 \n\t"\
1532 "movdqa %%xmm0, %%xmm9 \n\t"\
1533 "movdqa %%xmm0, %%xmm12 \n\t"\
1534 "movdqa %%xmm1, %%xmm11 \n\t"\
1535 "palignr $10,%%xmm0, %%xmm11\n\t"\
1536 "palignr $10,%%xmm7, %%xmm12\n\t"\
1537 "palignr $2, %%xmm0, %%xmm4 \n\t"\
1538 "palignr $2, %%xmm7, %%xmm9 \n\t"\
1539 "palignr $4, %%xmm0, %%xmm3 \n\t"\
1540 "palignr $4, %%xmm7, %%xmm8 \n\t"\
1541 "palignr $6, %%xmm0, %%xmm2 \n\t"\
1542 "palignr $6, %%xmm7, %%xmm6 \n\t"\
1543 "paddw %%xmm0 ,%%xmm11 \n\t"\
1544 "palignr $8, %%xmm0, %%xmm1 \n\t"\
1545 "palignr $8, %%xmm7, %%xmm0 \n\t"\
1546 "paddw %%xmm12,%%xmm7 \n\t"\
1547 "paddw %%xmm3, %%xmm2 \n\t"\
1548 "paddw %%xmm8, %%xmm6 \n\t"\
1549 "paddw %%xmm4, %%xmm1 \n\t"\
1550 "paddw %%xmm9, %%xmm0 \n\t"\
1551 "psllw $2, %%xmm2 \n\t"\
1552 "psllw $2, %%xmm6 \n\t"\
1553 "psubw %%xmm1, %%xmm2 \n\t"\
1554 "psubw %%xmm0, %%xmm6 \n\t"\
1555 "paddw %%xmm13,%%xmm11 \n\t"\
1556 "paddw %%xmm13,%%xmm7 \n\t"\
1557 "pmullw %%xmm14,%%xmm2 \n\t"\
1558 "pmullw %%xmm14,%%xmm6 \n\t"\
1559 "lddqu (%2), %%xmm3 \n\t"\
1560 "paddw %%xmm11,%%xmm2 \n\t"\
1561 "paddw %%xmm7, %%xmm6 \n\t"\
1562 "psraw $5, %%xmm2 \n\t"\
1563 "psraw $5, %%xmm6 \n\t"\
1564 "packuswb %%xmm2,%%xmm6 \n\t"\
1565 "pavgb %%xmm3, %%xmm6 \n\t"\
1566 OP(%%xmm6, (%1), %%xmm4, dqa)\
1567 "add %5, %0 \n\t"\
1568 "add %5, %1 \n\t"\
1569 "add %4, %2 \n\t"\
1570 "decl %3 \n\t"\
1571 "jg 1b \n\t"\
1572 : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
1573 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1574 "m"(ff_pw_5), "m"(ff_pw_16)\
1575 : "memory"\
1576 );\
1577 }
1578 #else // ARCH_X86_64
1579 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1580 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1581 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1582 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1583 src += 8*dstStride;\
1584 dst += 8*dstStride;\
1585 src2 += 8*src2Stride;\
1586 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1587 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1588 }
1589 #endif // ARCH_X86_64
1590
1591 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
1592 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1593 int h=8;\
1594 __asm__ volatile(\
1595 "pxor %%xmm7, %%xmm7 \n\t"\
1596 "movdqa %0, %%xmm6 \n\t"\
1597 :: "m"(ff_pw_5)\
1598 );\
1599 do{\
1600 __asm__ volatile(\
1601 "lddqu -2(%0), %%xmm1 \n\t"\
1602 "movdqa %%xmm1, %%xmm0 \n\t"\
1603 "punpckhbw %%xmm7, %%xmm1 \n\t"\
1604 "punpcklbw %%xmm7, %%xmm0 \n\t"\
1605 "movdqa %%xmm1, %%xmm2 \n\t"\
1606 "movdqa %%xmm1, %%xmm3 \n\t"\
1607 "movdqa %%xmm1, %%xmm4 \n\t"\
1608 "movdqa %%xmm1, %%xmm5 \n\t"\
1609 "palignr $2, %%xmm0, %%xmm4 \n\t"\
1610 "palignr $4, %%xmm0, %%xmm3 \n\t"\
1611 "palignr $6, %%xmm0, %%xmm2 \n\t"\
1612 "palignr $8, %%xmm0, %%xmm1 \n\t"\
1613 "palignr $10,%%xmm0, %%xmm5 \n\t"\
1614 "paddw %%xmm5, %%xmm0 \n\t"\
1615 "paddw %%xmm3, %%xmm2 \n\t"\
1616 "paddw %%xmm4, %%xmm1 \n\t"\
1617 "psllw $2, %%xmm2 \n\t"\
1618 "movq (%2), %%xmm3 \n\t"\
1619 "psubw %%xmm1, %%xmm2 \n\t"\
1620 "paddw %5, %%xmm0 \n\t"\
1621 "pmullw %%xmm6, %%xmm2 \n\t"\
1622 "paddw %%xmm0, %%xmm2 \n\t"\
1623 "psraw $5, %%xmm2 \n\t"\
1624 "packuswb %%xmm2, %%xmm2 \n\t"\
1625 "pavgb %%xmm3, %%xmm2 \n\t"\
1626 OP(%%xmm2, (%1), %%xmm4, q)\
1627 "add %4, %0 \n\t"\
1628 "add %4, %1 \n\t"\
1629 "add %3, %2 \n\t"\
1630 : "+a"(src), "+c"(dst), "+d"(src2)\
1631 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1632 "m"(ff_pw_16)\
1633 : "memory"\
1634 );\
1635 }while(--h);\
1636 }\
1637 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1638 \
1639 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1640 int h=8;\
1641 __asm__ volatile(\
1642 "pxor %%xmm7, %%xmm7 \n\t"\
1643 "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
1644 "1: \n\t"\
1645 "lddqu -2(%0), %%xmm1 \n\t"\
1646 "movdqa %%xmm1, %%xmm0 \n\t"\
1647 "punpckhbw %%xmm7, %%xmm1 \n\t"\
1648 "punpcklbw %%xmm7, %%xmm0 \n\t"\
1649 "movdqa %%xmm1, %%xmm2 \n\t"\
1650 "movdqa %%xmm1, %%xmm3 \n\t"\
1651 "movdqa %%xmm1, %%xmm4 \n\t"\
1652 "movdqa %%xmm1, %%xmm5 \n\t"\
1653 "palignr $2, %%xmm0, %%xmm4 \n\t"\
1654 "palignr $4, %%xmm0, %%xmm3 \n\t"\
1655 "palignr $6, %%xmm0, %%xmm2 \n\t"\
1656 "palignr $8, %%xmm0, %%xmm1 \n\t"\
1657 "palignr $10,%%xmm0, %%xmm5 \n\t"\
1658 "paddw %%xmm5, %%xmm0 \n\t"\
1659 "paddw %%xmm3, %%xmm2 \n\t"\
1660 "paddw %%xmm4, %%xmm1 \n\t"\
1661 "psllw $2, %%xmm2 \n\t"\
1662 "psubw %%xmm1, %%xmm2 \n\t"\
1663 "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
1664 "pmullw %%xmm6, %%xmm2 \n\t"\
1665 "paddw %%xmm0, %%xmm2 \n\t"\
1666 "psraw $5, %%xmm2 \n\t"\
1667 "packuswb %%xmm2, %%xmm2 \n\t"\
1668 OP(%%xmm2, (%1), %%xmm4, q)\
1669 "add %3, %0 \n\t"\
1670 "add %4, %1 \n\t"\
1671 "decl %2 \n\t"\
1672 " jnz 1b \n\t"\
1673 : "+a"(src), "+c"(dst), "+g"(h)\
1674 : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
1675 : "memory"\
1676 );\
1677 }\
1678 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1679 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1680 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1681 src += 8*srcStride;\
1682 dst += 8*dstStride;\
1683 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1684 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1685 }\
1686
1687 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
1688 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1689 src -= 2*srcStride;\
1690 \
1691 __asm__ volatile(\
1692 "pxor %%xmm7, %%xmm7 \n\t"\
1693 "movq (%0), %%xmm0 \n\t"\
1694 "add %2, %0 \n\t"\
1695 "movq (%0), %%xmm1 \n\t"\
1696 "add %2, %0 \n\t"\
1697 "movq (%0), %%xmm2 \n\t"\
1698 "add %2, %0 \n\t"\
1699 "movq (%0), %%xmm3 \n\t"\
1700 "add %2, %0 \n\t"\
1701 "movq (%0), %%xmm4 \n\t"\
1702 "add %2, %0 \n\t"\
1703 "punpcklbw %%xmm7, %%xmm0 \n\t"\
1704 "punpcklbw %%xmm7, %%xmm1 \n\t"\
1705 "punpcklbw %%xmm7, %%xmm2 \n\t"\
1706 "punpcklbw %%xmm7, %%xmm3 \n\t"\
1707 "punpcklbw %%xmm7, %%xmm4 \n\t"\
1708 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1709 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1710 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1711 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1712 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1713 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1714 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1715 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1716 \
1717 : "+a"(src), "+c"(dst)\
1718 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1719 : "memory"\
1720 );\
1721 if(h==16){\
1722 __asm__ volatile(\
1723 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1724 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1725 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1726 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1727 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1728 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1729 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1730 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1731 \
1732 : "+a"(src), "+c"(dst)\
1733 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1734 : "memory"\
1735 );\
1736 }\
1737 }\
1738 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1739 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
1740 }\
1741 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1742 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
1743 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1744 }
1745
1746 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
1747 int w = (size+8)>>3;
1748 src -= 2*srcStride+2;
1749 while(w--){
1750 __asm__ volatile(
1751 "pxor %%xmm7, %%xmm7 \n\t"
1752 "movq (%0), %%xmm0 \n\t"
1753 "add %2, %0 \n\t"
1754 "movq (%0), %%xmm1 \n\t"
1755 "add %2, %0 \n\t"
1756 "movq (%0), %%xmm2 \n\t"
1757 "add %2, %0 \n\t"
1758 "movq (%0), %%xmm3 \n\t"
1759 "add %2, %0 \n\t"
1760 "movq (%0), %%xmm4 \n\t"
1761 "add %2, %0 \n\t"
1762 "punpcklbw %%xmm7, %%xmm0 \n\t"
1763 "punpcklbw %%xmm7, %%xmm1 \n\t"
1764 "punpcklbw %%xmm7, %%xmm2 \n\t"
1765 "punpcklbw %%xmm7, %%xmm3 \n\t"
1766 "punpcklbw %%xmm7, %%xmm4 \n\t"
1767 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
1768 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
1769 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
1770 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
1771 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
1772 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
1773 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
1774 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
1775 : "+a"(src)
1776 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1777 : "memory"
1778 );
1779 if(size==16){
1780 __asm__ volatile(
1781 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 8*48)
1782 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 9*48)
1783 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
1784 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
1785 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
1786 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
1787 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
1788 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
1789 : "+a"(src)
1790 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1791 : "memory"
1792 );
1793 }
1794 tmp += 8;
1795 src += 8 - (size+5)*srcStride;
1796 }
1797 }
1798
1799 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
1800 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1801 int h = size;\
1802 if(size == 16){\
1803 __asm__ volatile(\
1804 "1: \n\t"\
1805 "movdqa 32(%0), %%xmm4 \n\t"\
1806 "movdqa 16(%0), %%xmm5 \n\t"\
1807 "movdqa (%0), %%xmm7 \n\t"\
1808 "movdqa %%xmm4, %%xmm3 \n\t"\
1809 "movdqa %%xmm4, %%xmm2 \n\t"\
1810 "movdqa %%xmm4, %%xmm1 \n\t"\
1811 "movdqa %%xmm4, %%xmm0 \n\t"\
1812 "palignr $10, %%xmm5, %%xmm0 \n\t"\
1813 "palignr $8, %%xmm5, %%xmm1 \n\t"\
1814 "palignr $6, %%xmm5, %%xmm2 \n\t"\
1815 "palignr $4, %%xmm5, %%xmm3 \n\t"\
1816 "palignr $2, %%xmm5, %%xmm4 \n\t"\
1817 "paddw %%xmm5, %%xmm0 \n\t"\
1818 "paddw %%xmm4, %%xmm1 \n\t"\
1819 "paddw %%xmm3, %%xmm2 \n\t"\
1820 "movdqa %%xmm5, %%xmm6 \n\t"\
1821 "movdqa %%xmm5, %%xmm4 \n\t"\
1822 "movdqa %%xmm5, %%xmm3 \n\t"\
1823 "palignr $8, %%xmm7, %%xmm4 \n\t"\
1824 "palignr $2, %%xmm7, %%xmm6 \n\t"\
1825 "palignr $10, %%xmm7, %%xmm3 \n\t"\
1826 "paddw %%xmm6, %%xmm4 \n\t"\
1827 "movdqa %%xmm5, %%xmm6 \n\t"\
1828 "palignr $6, %%xmm7, %%xmm5 \n\t"\
1829 "palignr $4, %%xmm7, %%xmm6 \n\t"\
1830 "paddw %%xmm7, %%xmm3 \n\t"\
1831 "paddw %%xmm6, %%xmm5 \n\t"\
1832 \
1833 "psubw %%xmm1, %%xmm0 \n\t"\
1834 "psubw %%xmm4, %%xmm3 \n\t"\
1835 "psraw $2, %%xmm0 \n\t"\
1836 "psraw $2, %%xmm3 \n\t"\
1837 "psubw %%xmm1, %%xmm0 \n\t"\
1838 "psubw %%xmm4, %%xmm3 \n\t"\
1839 "paddw %%xmm2, %%xmm0 \n\t"\
1840 "paddw %%xmm5, %%xmm3 \n\t"\
1841 "psraw $2, %%xmm0 \n\t"\
1842 "psraw $2, %%xmm3 \n\t"\
1843 "paddw %%xmm2, %%xmm0 \n\t"\
1844 "paddw %%xmm5, %%xmm3 \n\t"\
1845 "psraw $6, %%xmm0 \n\t"\
1846 "psraw $6, %%xmm3 \n\t"\
1847 "packuswb %%xmm0, %%xmm3 \n\t"\
1848 OP(%%xmm3, (%1), %%xmm7, dqa)\
1849 "add $48, %0 \n\t"\
1850 "add %3, %1 \n\t"\
1851 "decl %2 \n\t"\
1852 " jnz 1b \n\t"\
1853 : "+a"(tmp), "+c"(dst), "+g"(h)\
1854 : "S"((x86_reg)dstStride)\
1855 : "memory"\
1856 );\
1857 }else{\
1858 __asm__ volatile(\
1859 "1: \n\t"\
1860 "movdqa 16(%0), %%xmm1 \n\t"\
1861 "movdqa (%0), %%xmm0 \n\t"\
1862 "movdqa %%xmm1, %%xmm2 \n\t"\
1863 "movdqa %%xmm1, %%xmm3 \n\t"\
1864 "movdqa %%xmm1, %%xmm4 \n\t"\
1865 "movdqa %%xmm1, %%xmm5 \n\t"\
1866 "palignr $10, %%xmm0, %%xmm5 \n\t"\
1867 "palignr $8, %%xmm0, %%xmm4 \n\t"\
1868 "palignr $6, %%xmm0, %%xmm3 \n\t"\
1869 "palignr $4, %%xmm0, %%xmm2 \n\t"\
1870 "palignr $2, %%xmm0, %%xmm1 \n\t"\
1871 "paddw %%xmm5, %%xmm0 \n\t"\
1872 "paddw %%xmm4, %%xmm1 \n\t"\
1873 "paddw %%xmm3, %%xmm2 \n\t"\
1874 "psubw %%xmm1, %%xmm0 \n\t"\
1875 "psraw $2, %%xmm0 \n\t"\
1876 "psubw %%xmm1, %%xmm0 \n\t"\
1877 "paddw %%xmm2, %%xmm0 \n\t"\
1878 "psraw $2, %%xmm0 \n\t"\
1879 "paddw %%xmm2, %%xmm0 \n\t"\
1880 "psraw $6, %%xmm0 \n\t"\
1881 "packuswb %%xmm0, %%xmm0 \n\t"\
1882 OP(%%xmm0, (%1), %%xmm7, q)\
1883 "add $48, %0 \n\t"\
1884 "add %3, %1 \n\t"\
1885 "decl %2 \n\t"\
1886 " jnz 1b \n\t"\
1887 : "+a"(tmp), "+c"(dst), "+g"(h)\
1888 : "S"((x86_reg)dstStride)\
1889 : "memory"\
1890 );\
1891 }\
1892 }
1893
1894 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
1895 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){\
1896 put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
1897 OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1898 }\
1899 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1900 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
1901 }\
1902 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1903 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1904 }\
1905
1906 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1907 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1908 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1909 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1910 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1911 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1912 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1913 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1914
1915 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1916 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1917 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1918 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1919 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1920 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1921 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1922 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1923
1924 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1925 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1926 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1927 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1928
1929 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1930 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1931 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1932 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1933
1934 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1935 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1936
1937 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1938 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1939 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1940 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1941 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1942
1943 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1944 put_pixels16_sse2(dst, src, stride, 16);
1945 }
1946 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1947 avg_pixels16_sse2(dst, src, stride, 16);
1948 }
1949 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1950 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1951
1952 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1953 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1954 OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1955 }\
1956
1957 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1958 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1959 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1960 }\
1961 \
1962 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1963 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1964 }\
1965 \
1966 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1967 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1968 }\
1969
1970 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1971 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1972 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1973 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1974 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1975 }\
1976 \
1977 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1978 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1979 }\
1980 \
1981 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1982 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1983 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1984 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1985 }\
1986
1987 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1988 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1989 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1990 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1991 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1992 }\
1993 \
1994 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1995 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1996 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1997 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1998 }\
1999 \
2000 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2001 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
2002 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
2003 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
2004 }\
2005 \
2006 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2007 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
2008 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
2009 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
2010 }\
2011 \
2012 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2013 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
2014 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
2015 }\
2016 \
2017 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2018 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
2019 uint8_t * const halfHV= temp;\
2020 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
2021 assert(((int)temp & 7) == 0);\
2022 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
2023 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
2024 }\
2025 \
2026 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2027 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
2028 uint8_t * const halfHV= temp;\
2029 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
2030 assert(((int)temp & 7) == 0);\
2031 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
2032 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
2033 }\
2034 \
2035 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2036 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
2037 uint8_t * const halfHV= temp;\
2038 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
2039 assert(((int)temp & 7) == 0);\
2040 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
2041 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
2042 }\
2043 \
2044 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
2045 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
2046 uint8_t * const halfHV= temp;\
2047 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
2048 assert(((int)temp & 7) == 0);\
2049 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
2050 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
2051 }\
2052
2053 #define H264_MC_4816(MMX)\
2054 H264_MC(put_, 4, MMX, 8)\
2055 H264_MC(put_, 8, MMX, 8)\
2056 H264_MC(put_, 16,MMX, 8)\
2057 H264_MC(avg_, 4, MMX, 8)\
2058 H264_MC(avg_, 8, MMX, 8)\
2059 H264_MC(avg_, 16,MMX, 8)\
2060
2061 #define H264_MC_816(QPEL, XMM)\
2062 QPEL(put_, 8, XMM, 16)\
2063 QPEL(put_, 16,XMM, 16)\
2064 QPEL(avg_, 8, XMM, 16)\
2065 QPEL(avg_, 16,XMM, 16)\
2066
2067
2068 #define AVG_3DNOW_OP(a,b,temp, size) \
2069 "mov" #size " " #b ", " #temp " \n\t"\
2070 "pavgusb " #temp ", " #a " \n\t"\
2071 "mov" #size " " #a ", " #b " \n\t"
2072 #define AVG_MMX2_OP(a,b,temp, size) \
2073 "mov" #size " " #b ", " #temp " \n\t"\
2074 "pavgb " #temp ", " #a " \n\t"\
2075 "mov" #size " " #a ", " #b " \n\t"
2076
2077 #define PAVGB "pavgusb"
2078 QPEL_H264(put_, PUT_OP, 3dnow)
2079 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
2080 #undef PAVGB
2081 #define PAVGB "pavgb"
2082 QPEL_H264(put_, PUT_OP, mmx2)
2083 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
2084 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
2085 QPEL_H264_V_XMM(avg_, AVG_MMX2_OP, sse2)
2086 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
2087 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, sse2)
2088 #if HAVE_SSSE3
2089 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
2090 QPEL_H264_H_XMM(avg_, AVG_MMX2_OP, ssse3)
2091 QPEL_H264_HV2_XMM(put_, PUT_OP, ssse3)
2092 QPEL_H264_HV2_XMM(avg_, AVG_MMX2_OP, ssse3)
2093 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
2094 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, ssse3)
2095 #endif
2096 #undef PAVGB
2097
2098 H264_MC_4816(3dnow)
2099 H264_MC_4816(mmx2)
2100 H264_MC_816(H264_MC_V, sse2)
2101 H264_MC_816(H264_MC_HV, sse2)
2102 #if HAVE_SSSE3
2103 H264_MC_816(H264_MC_H, ssse3)
2104 H264_MC_816(H264_MC_HV, ssse3)
2105 #endif
2106
2107 /* rnd interleaved with rnd div 8, use p+1 to access rnd div 8 */
2108 DECLARE_ALIGNED(8, static const uint64_t, h264_rnd_reg)[4] = {
2109 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL
2110 };
2111
2112 #define H264_CHROMA_OP(S,D)
2113 #define H264_CHROMA_OP4(S,D,T)
2114 #define H264_CHROMA_MC8_TMPL put_h264_chroma_generic_mc8_mmx
2115 #define H264_CHROMA_MC4_TMPL put_h264_chroma_generic_mc4_mmx
2116 #define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
2117 #define H264_CHROMA_MC8_MV0 put_pixels8_mmx
2118 #include "dsputil_h264_template_mmx.c"
2119
2120 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)
2121 {
2122 put_h264_chroma_generic_mc8_mmx(dst, src, stride, h, x, y, h264_rnd_reg);
2123 }
2124 static void put_vc1_chroma_mc8_mmx_nornd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2125 {
2126 put_h264_chroma_generic_mc8_mmx(dst, src, stride, h, x, y, h264_rnd_reg+2);
2127 }
2128 static void put_h264_chroma_mc4_mmx(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2129 {
2130 put_h264_chroma_generic_mc4_mmx(dst, src, stride, h, x, y, h264_rnd_reg);
2131 }
2132
2133 #undef H264_CHROMA_OP
2134 #undef H264_CHROMA_OP4
2135 #undef H264_CHROMA_MC8_TMPL
2136 #undef H264_CHROMA_MC4_TMPL
2137 #undef H264_CHROMA_MC2_TMPL
2138 #undef H264_CHROMA_MC8_MV0
2139
2140 #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
2141 #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
2142 "pavgb " #T ", " #D " \n\t"
2143 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_generic_mc8_mmx2
2144 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_generic_mc4_mmx2
2145 #define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
2146 #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
2147 #include "dsputil_h264_template_mmx.c"
2148 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)
2149 {
2150 avg_h264_chroma_generic_mc8_mmx2(dst, src, stride, h, x, y, h264_rnd_reg);
2151 }
2152 static void avg_vc1_chroma_mc8_mmx2_nornd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2153 {
2154 avg_h264_chroma_generic_mc8_mmx2(dst, src, stride, h, x, y, h264_rnd_reg+2);
2155 }
2156 static void avg_h264_chroma_mc4_mmx2(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2157 {
2158 avg_h264_chroma_generic_mc4_mmx2(dst, src, stride, h, x, y, h264_rnd_reg);
2159 }
2160 #undef H264_CHROMA_OP
2161 #undef H264_CHROMA_OP4
2162 #undef H264_CHROMA_MC8_TMPL
2163 #undef H264_CHROMA_MC4_TMPL
2164 #undef H264_CHROMA_MC2_TMPL
2165 #undef H264_CHROMA_MC8_MV0
2166
2167 #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
2168 #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
2169 "pavgusb " #T ", " #D " \n\t"
2170 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_generic_mc8_3dnow
2171 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_generic_mc4_3dnow
2172 #define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
2173 #include "dsputil_h264_template_mmx.c"
2174 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)
2175 {
2176 avg_h264_chroma_generic_mc8_3dnow(dst, src, stride, h, x, y, h264_rnd_reg);
2177 }
2178 static void avg_h264_chroma_mc4_3dnow(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2179 {
2180 avg_h264_chroma_generic_mc4_3dnow(dst, src, stride, h, x, y, h264_rnd_reg);
2181 }
2182 #undef H264_CHROMA_OP
2183 #undef H264_CHROMA_OP4
2184 #undef H264_CHROMA_MC8_TMPL
2185 #undef H264_CHROMA_MC4_TMPL
2186 #undef H264_CHROMA_MC8_MV0
2187
2188 #if HAVE_SSSE3
2189 #define AVG_OP(X)
2190 #undef H264_CHROMA_MC8_TMPL
2191 #undef H264_CHROMA_MC4_TMPL
2192 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_ssse3
2193 #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_ssse3
2194 #define H264_CHROMA_MC8_MV0 put_pixels8_mmx
2195 #include "dsputil_h264_template_ssse3.c"
2196 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)
2197 {
2198 put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
2199 }
2200 static void put_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2201 {
2202 put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 0);
2203 }
2204
2205 #undef AVG_OP
2206 #undef H264_CHROMA_MC8_TMPL
2207 #undef H264_CHROMA_MC4_TMPL
2208 #undef H264_CHROMA_MC8_MV0
2209 #define AVG_OP(X) X
2210 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_ssse3
2211 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_ssse3
2212 #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
2213 #include "dsputil_h264_template_ssse3.c"
2214 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)
2215 {
2216 avg_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
2217 }
2218 static void avg_vc1_chroma_mc8_ssse3_nornd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
2219 {
2220 avg_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 0);
2221 }
2222 #undef AVG_OP
2223 #undef H264_CHROMA_MC8_TMPL
2224 #undef H264_CHROMA_MC4_TMPL
2225 #undef H264_CHROMA_MC8_MV0
2226 #endif
2227
2228 /***********************************/
2229 /* weighted prediction */
2230
2231 static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
2232 {
2233 int x, y;
2234 offset <<= log2_denom;
2235 offset += (1 << log2_denom) >> 1;
2236 __asm__ volatile(
2237 "movd %0, %%mm4 \n\t"
2238 "movd %1, %%mm5 \n\t"
2239 "movd %2, %%mm6 \n\t"
2240 "pshufw $0, %%mm4, %%mm4 \n\t"
2241 "pshufw $0, %%mm5, %%mm5 \n\t"
2242 "pxor %%mm7, %%mm7 \n\t"
2243 :: "g"(weight), "g"(offset), "g"(log2_denom)
2244 );
2245 for(y=0; y<h; y+=2){
2246 for(x=0; x<w; x+=4){
2247 __asm__ volatile(
2248 "movd %0, %%mm0 \n\t"
2249 "movd %1, %%mm1 \n\t"
2250 "punpcklbw %%mm7, %%mm0 \n\t"
2251 "punpcklbw %%mm7, %%mm1 \n\t"
2252 "pmullw %%mm4, %%mm0 \n\t"
2253 "pmullw %%mm4, %%mm1 \n\t"
2254 "paddsw %%mm5, %%mm0 \n\t"
2255 "paddsw %%mm5, %%mm1 \n\t"
2256 "psraw %%mm6, %%mm0 \n\t"
2257 "psraw %%mm6, %%mm1 \n\t"
2258 "packuswb %%mm7, %%mm0 \n\t"
2259 "packuswb %%mm7, %%mm1 \n\t"
2260 "movd %%mm0, %0 \n\t"
2261 "movd %%mm1, %1 \n\t"
2262 : "+m"(*(uint32_t*)(dst+x)),
2263 "+m"(*(uint32_t*)(dst+x+stride))
2264 );
2265 }
2266 dst += 2*stride;
2267 }
2268 }
2269
2270 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)
2271 {
2272 int x, y;
2273 offset = ((offset + 1) | 1) << log2_denom;
2274 __asm__ volatile(
2275 "movd %0, %%mm3 \n\t"
2276 "movd %1, %%mm4 \n\t"
2277 "movd %2, %%mm5 \n\t"
2278 "movd %3, %%mm6 \n\t"
2279 "pshufw $0, %%mm3, %%mm3 \n\t"
2280 "pshufw $0, %%mm4, %%mm4 \n\t"
2281 "pshufw $0, %%mm5, %%mm5 \n\t"
2282 "pxor %%mm7, %%mm7 \n\t"
2283 :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
2284 );
2285 for(y=0; y<h; y++){
2286 for(x=0; x<w; x+=4){
2287 __asm__ volatile(
2288 "movd %0, %%mm0 \n\t"
2289 "movd %1, %%mm1 \n\t"
2290 "punpcklbw %%mm7, %%mm0 \n\t"
2291 "punpcklbw %%mm7, %%mm1 \n\t"
2292 "pmullw %%mm3, %%mm0 \n\t"
2293 "pmullw %%mm4, %%mm1 \n\t"
2294 "paddsw %%mm1, %%mm0 \n\t"
2295 "paddsw %%mm5, %%mm0 \n\t"
2296 "psraw %%mm6, %%mm0 \n\t"
2297 "packuswb %%mm0, %%mm0 \n\t"
2298 "movd %%mm0, %0 \n\t"
2299 : "+m"(*(uint32_t*)(dst+x))
2300 : "m"(*(uint32_t*)(src+x))
2301 );
2302 }
2303 src += stride;
2304 dst += stride;
2305 }
2306 }
2307
2308 #define H264_WEIGHT(W,H) \
2309 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){ \
2310 ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
2311 } \
2312 static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
2313 ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
2314 }
2315
2316 H264_WEIGHT(16,16)
2317 H264_WEIGHT(16, 8)
2318 H264_WEIGHT( 8,16)
2319 H264_WEIGHT( 8, 8)
2320 H264_WEIGHT( 8, 4)
2321 H264_WEIGHT( 4, 8)
2322 H264_WEIGHT( 4, 4)
2323 H264_WEIGHT( 4, 2)
2324