Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > VSs > VSs__H264__App

view libavcodec/x86/h264dsp_mmx.c @ 2:897f711a7157

find changesets by author, revision, files, or words in the commit message
rearrange to work with autoconf
author Nina Engelhardt <nengel@mailbox.tu-berlin.de>
date Tue, 25 Sep 2012 15:55:33 +0200
parents
children
line source
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_mmx2(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 int nnz = nnzc[ scan8[i] ];
395 if(nnz){
396 if(nnz==1 && block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
397 else ff_h264_idct_add_mmx (dst + block_offset[i], block + i*16, stride);
398 }
399 }
400 }
401
402 static void ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
403 int i;
404 for(i=0; i<16; i++){
405 if(nnzc[ scan8[i] ]) ff_h264_idct_add_mmx (dst + block_offset[i], block + i*16, stride);
406 else if(block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
407 }
408 }
409
410 static void ff_h264_idct8_add4_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
411 int i;
412 for(i=0; i<16; i+=4){
413 int nnz = nnzc[ scan8[i] ];
414 if(nnz){
415 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
416 else ff_h264_idct8_add_mmx (dst + block_offset[i], block + i*16, stride);
417 }
418 }
419 }
420
421 static void ff_h264_idct8_add4_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
422 int i;
423 for(i=0; i<16; i+=4){
424 int nnz = nnzc[ scan8[i] ];
425 if(nnz){
426 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
427 else ff_h264_idct8_add_sse2 (dst + block_offset[i], block + i*16, stride);
428 }
429 }
430 }
431
432 static void ff_h264_idct_add8_mmx2(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
433 int i;
434 for(i=16; i<16+8; i++){
435 if(nnzc[ scan8[i] ])
436 ff_h264_idct_add_mmx (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
437 else if(block[i*16])
438 ff_h264_idct_dc_add_mmx2(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
439 }
440 }
441
442 /***********************************/
443 /* deblocking */
444
445 // out: o = |x-y|>a
446 // clobbers: t
447 #define DIFF_GT_MMX(x,y,a,o,t)\
448 "movq "#y", "#t" \n\t"\
449 "movq "#x", "#o" \n\t"\
450 "psubusb "#x", "#t" \n\t"\
451 "psubusb "#y", "#o" \n\t"\
452 "por "#t", "#o" \n\t"\
453 "psubusb "#a", "#o" \n\t"
454
455 // out: o = |x-y|>a
456 // clobbers: t
457 #define DIFF_GT2_MMX(x,y,a,o,t)\
458 "movq "#y", "#t" \n\t"\
459 "movq "#x", "#o" \n\t"\
460 "psubusb "#x", "#t" \n\t"\
461 "psubusb "#y", "#o" \n\t"\
462 "psubusb "#a", "#t" \n\t"\
463 "psubusb "#a", "#o" \n\t"\
464 "pcmpeqb "#t", "#o" \n\t"\
465
466 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
467 // out: mm5=beta-1, mm7=mask
468 // clobbers: mm4,mm6
469 #define H264_DEBLOCK_MASK(alpha1, beta1) \
470 "pshufw $0, "#alpha1", %%mm4 \n\t"\
471 "pshufw $0, "#beta1 ", %%mm5 \n\t"\
472 "packuswb %%mm4, %%mm4 \n\t"\
473 "packuswb %%mm5, %%mm5 \n\t"\
474 DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
475 DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
476 "por %%mm4, %%mm7 \n\t"\
477 DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
478 "por %%mm4, %%mm7 \n\t"\
479 "pxor %%mm6, %%mm6 \n\t"\
480 "pcmpeqb %%mm6, %%mm7 \n\t"
481
482 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
483 // out: mm1=p0' mm2=q0'
484 // clobbers: mm0,3-6
485 #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
486 "movq %%mm1 , %%mm5 \n\t"\
487 "pxor %%mm2 , %%mm5 \n\t" /* p0^q0*/\
488 "pand "#pb_01" , %%mm5 \n\t" /* (p0^q0)&1*/\
489 "pcmpeqb %%mm4 , %%mm4 \n\t"\
490 "pxor %%mm4 , %%mm3 \n\t"\
491 "pavgb %%mm0 , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
492 "pavgb "MANGLE(ff_pb_3)" , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
493 "pxor %%mm1 , %%mm4 \n\t"\
494 "pavgb %%mm2 , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
495 "pavgb %%mm5 , %%mm3 \n\t"\
496 "paddusb %%mm4 , %%mm3 \n\t" /* d+128+33*/\
497 "movq "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
498 "psubusb %%mm3 , %%mm6 \n\t"\
499 "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
500 "pminub %%mm7 , %%mm6 \n\t"\
501 "pminub %%mm7 , %%mm3 \n\t"\
502 "psubusb %%mm6 , %%mm1 \n\t"\
503 "psubusb %%mm3 , %%mm2 \n\t"\
504 "paddusb %%mm3 , %%mm1 \n\t"\
505 "paddusb %%mm6 , %%mm2 \n\t"
506
507 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
508 // out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
509 // clobbers: q2, tmp, tc0
510 #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
511 "movq %%mm1, "#tmp" \n\t"\
512 "pavgb %%mm2, "#tmp" \n\t"\
513 "pavgb "#tmp", "#q2" \n\t" /* avg(p2,avg(p0,q0)) */\
514 "pxor "q2addr", "#tmp" \n\t"\
515 "pand %9, "#tmp" \n\t" /* (p2^avg(p0,q0))&1 */\
516 "psubusb "#tmp", "#q2" \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
517 "movq "#p1", "#tmp" \n\t"\
518 "psubusb "#tc0", "#tmp" \n\t"\
519 "paddusb "#p1", "#tc0" \n\t"\
520 "pmaxub "#tmp", "#q2" \n\t"\
521 "pminub "#tc0", "#q2" \n\t"\
522 "movq "#q2", "q1addr" \n\t"
523
524 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
525 {
526 DECLARE_ALIGNED(8, uint64_t, tmp0)[2];
527
528 __asm__ volatile(
529 "movq (%2,%4), %%mm0 \n\t" //p1
530 "movq (%2,%4,2), %%mm1 \n\t" //p0
531 "movq (%3), %%mm2 \n\t" //q0
532 "movq (%3,%4), %%mm3 \n\t" //q1
533 H264_DEBLOCK_MASK(%7, %8)
534
535 "movd %6, %%mm4 \n\t"
536 "punpcklbw %%mm4, %%mm4 \n\t"
537 "punpcklwd %%mm4, %%mm4 \n\t"
538 "pcmpeqb %%mm3, %%mm3 \n\t"
539 "movq %%mm4, %%mm6 \n\t"
540 "pcmpgtb %%mm3, %%mm4 \n\t"
541 "movq %%mm6, %1 \n\t"
542 "pand %%mm4, %%mm7 \n\t"
543 "movq %%mm7, %0 \n\t"
544
545 /* filter p1 */
546 "movq (%2), %%mm3 \n\t" //p2
547 DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
548 "pand %%mm7, %%mm6 \n\t" // mask & |p2-p0|<beta
549 "pand %1, %%mm7 \n\t" // mask & tc0
550 "movq %%mm7, %%mm4 \n\t"
551 "psubb %%mm6, %%mm7 \n\t"
552 "pand %%mm4, %%mm6 \n\t" // mask & |p2-p0|<beta & tc0
553 H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%2)", "(%2,%4)", %%mm6, %%mm4)
554
555 /* filter q1 */
556 "movq (%3,%4,2), %%mm4 \n\t" //q2
557 DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
558 "pand %0, %%mm6 \n\t"
559 "movq %1, %%mm5 \n\t" // can be merged with the and below but is slower then
560 "pand %%mm6, %%mm5 \n\t"
561 "psubb %%mm6, %%mm7 \n\t"
562 "movq (%3,%4), %%mm3 \n\t"
563 H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%3,%4,2)", "(%3,%4)", %%mm5, %%mm6)
564
565 /* filter p0, q0 */
566 H264_DEBLOCK_P0_Q0(%9, unused)
567 "movq %%mm1, (%2,%4,2) \n\t"
568 "movq %%mm2, (%3) \n\t"
569
570 : "=m"(tmp0[0]), "=m"(tmp0[1])
571 : "r"(pix-3*stride), "r"(pix), "r"((x86_reg)stride),
572 "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
573 "m"(ff_bone)
574 );
575 }
576
577 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
578 {
579 if((tc0[0] & tc0[1]) >= 0)
580 h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
581 if((tc0[2] & tc0[3]) >= 0)
582 h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
583 }
584 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
585 {
586 //FIXME: could cut some load/stores by merging transpose with filter
587 // also, it only needs to transpose 6x8
588 DECLARE_ALIGNED(8, uint8_t, trans)[8*8];
589 int i;
590 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
591 if((tc0[0] & tc0[1]) < 0)
592 continue;
593 transpose4x4(trans, pix-4, 8, stride);
594 transpose4x4(trans +4*8, pix, 8, stride);
595 transpose4x4(trans+4, pix-4+4*stride, 8, stride);
596 transpose4x4(trans+4+4*8, pix +4*stride, 8, stride);
597 h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
598 transpose4x4(pix-2, trans +2*8, stride, 8);
599 transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
600 }
601 }
602
603 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
604 {
605 __asm__ volatile(
606 "movq (%0), %%mm0 \n\t" //p1
607 "movq (%0,%2), %%mm1 \n\t" //p0
608 "movq (%1), %%mm2 \n\t" //q0
609 "movq (%1,%2), %%mm3 \n\t" //q1
610 H264_DEBLOCK_MASK(%4, %5)
611 "movd %3, %%mm6 \n\t"
612 "punpcklbw %%mm6, %%mm6 \n\t"
613 "pand %%mm6, %%mm7 \n\t" // mm7 = tc&mask
614 H264_DEBLOCK_P0_Q0(%6, %7)
615 "movq %%mm1, (%0,%2) \n\t"
616 "movq %%mm2, (%1) \n\t"
617
618 :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
619 "r"(*(uint32_t*)tc0),
620 "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
621 );
622 }
623
624 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
625 {
626 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
627 }
628
629 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
630 {
631 //FIXME: could cut some load/stores by merging transpose with filter
632 DECLARE_ALIGNED(8, uint8_t, trans)[8*4];
633 transpose4x4(trans, pix-2, 8, stride);
634 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
635 h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
636 transpose4x4(pix-2, trans, stride, 8);
637 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
638 }
639
640 // p0 = (p0 + q1 + 2*p1 + 2) >> 2
641 #define H264_FILTER_CHROMA4(p0, p1, q1, one) \
642 "movq "#p0", %%mm4 \n\t"\
643 "pxor "#q1", %%mm4 \n\t"\
644 "pand "#one", %%mm4 \n\t" /* mm4 = (p0^q1)&1 */\
645 "pavgb "#q1", "#p0" \n\t"\
646 "psubusb %%mm4, "#p0" \n\t"\
647 "pavgb "#p1", "#p0" \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
648
649 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
650 {
651 __asm__ volatile(
652 "movq (%0), %%mm0 \n\t"
653 "movq (%0,%2), %%mm1 \n\t"
654 "movq (%1), %%mm2 \n\t"
655 "movq (%1,%2), %%mm3 \n\t"
656 H264_DEBLOCK_MASK(%3, %4)
657 "movq %%mm1, %%mm5 \n\t"
658 "movq %%mm2, %%mm6 \n\t"
659 H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
660 H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
661 "psubb %%mm5, %%mm1 \n\t"
662 "psubb %%mm6, %%mm2 \n\t"
663 "pand %%mm7, %%mm1 \n\t"
664 "pand %%mm7, %%mm2 \n\t"
665 "paddb %%mm5, %%mm1 \n\t"
666 "paddb %%mm6, %%mm2 \n\t"
667 "movq %%mm1, (%0,%2) \n\t"
668 "movq %%mm2, (%1) \n\t"
669 :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
670 "m"(alpha1), "m"(beta1), "m"(ff_bone)
671 );
672 }
673
674 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
675 {
676 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
677 }
678
679 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
680 {
681 //FIXME: could cut some load/stores by merging transpose with filter
682 DECLARE_ALIGNED(8, uint8_t, trans)[8*4];
683 transpose4x4(trans, pix-2, 8, stride);
684 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
685 h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
686 transpose4x4(pix-2, trans, stride, 8);
687 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
688 }
689
690 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],
691 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
692 int dir;
693 __asm__ volatile(
694 "movq %0, %%mm7 \n"
695 "movq %1, %%mm6 \n"
696 ::"m"(ff_pb_1), "m"(ff_pb_3)
697 );
698 if(field)
699 __asm__ volatile(
700 "movq %0, %%mm6 \n"
701 ::"m"(ff_pb_3_1)
702 );
703 __asm__ volatile(
704 "movq %%mm6, %%mm5 \n"
705 "paddb %%mm5, %%mm5 \n"
706 :);
707
708 // could do a special case for dir==0 && edges==1, but it only reduces the
709 // average filter time by 1.2%
710 for( dir=1; dir>=0; dir-- ) {
711 const x86_reg d_idx = dir ? -8 : -1;
712 const int mask_mv = dir ? mask_mv1 : mask_mv0;
713 DECLARE_ALIGNED(8, const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
714 int b_idx, edge;
715 for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
716 __asm__ volatile(
717 "pand %0, %%mm0 \n\t"
718 ::"m"(mask_dir)
719 );
720 if(!(mask_mv & edge)) {
721 if(bidir) {
722 __asm__ volatile(
723 "movd (%1,%0), %%mm2 \n"
724 "punpckldq 40(%1,%0), %%mm2 \n" // { ref0[bn], ref1[bn] }
725 "pshufw $0x44, (%1), %%mm0 \n" // { ref0[b], ref0[b] }
726 "pshufw $0x44, 40(%1), %%mm1 \n" // { ref1[b], ref1[b] }
727 "pshufw $0x4E, %%mm2, %%mm3 \n"
728 "psubb %%mm2, %%mm0 \n" // { ref0[b]!=ref0[bn], ref0[b]!=ref1[bn] }
729 "psubb %%mm3, %%mm1 \n" // { ref1[b]!=ref1[bn], ref1[b]!=ref0[bn] }
730 "1: \n"
731 "por %%mm1, %%mm0 \n"
732 "movq (%2,%0,4), %%mm1 \n"
733 "movq 8(%2,%0,4), %%mm2 \n"
734 "movq %%mm1, %%mm3 \n"
735 "movq %%mm2, %%mm4 \n"
736 "psubw (%2), %%mm1 \n"
737 "psubw 8(%2), %%mm2 \n"
738 "psubw 160(%2), %%mm3 \n"
739 "psubw 168(%2), %%mm4 \n"
740 "packsswb %%mm2, %%mm1 \n"
741 "packsswb %%mm4, %%mm3 \n"
742 "paddb %%mm6, %%mm1 \n"
743 "paddb %%mm6, %%mm3 \n"
744 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
745 "psubusb %%mm5, %%mm3 \n"
746 "packsswb %%mm3, %%mm1 \n"
747 "add $40, %0 \n"
748 "cmp $40, %0 \n"
749 "jl 1b \n"
750 "sub $80, %0 \n"
751 "pshufw $0x4E, %%mm1, %%mm1 \n"
752 "por %%mm1, %%mm0 \n"
753 "pshufw $0x4E, %%mm0, %%mm1 \n"
754 "pminub %%mm1, %%mm0 \n"
755 ::"r"(d_idx),
756 "r"(ref[0]+b_idx),
757 "r"(mv[0]+b_idx)
758 );
759 } else {
760 __asm__ volatile(
761 "movd (%1), %%mm0 \n"
762 "psubb (%1,%0), %%mm0 \n" // ref[b] != ref[bn]
763 "movq (%2), %%mm1 \n"
764 "movq 8(%2), %%mm2 \n"
765 "psubw (%2,%0,4), %%mm1 \n"
766 "psubw 8(%2,%0,4), %%mm2 \n"
767 "packsswb %%mm2, %%mm1 \n"
768 "paddb %%mm6, %%mm1 \n"
769 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
770 "packsswb %%mm1, %%mm1 \n"
771 "por %%mm1, %%mm0 \n"
772 ::"r"(d_idx),
773 "r"(ref[0]+b_idx),
774 "r"(mv[0]+b_idx)
775 );
776 }
777 }
778 __asm__ volatile(
779 "movd %0, %%mm1 \n"
780 "por %1, %%mm1 \n" // nnz[b] || nnz[bn]
781 ::"m"(nnz[b_idx]),
782 "m"(nnz[b_idx+d_idx])
783 );
784 __asm__ volatile(
785 "pminub %%mm7, %%mm1 \n"
786 "pminub %%mm7, %%mm0 \n"
787 "psllw $1, %%mm1 \n"
788 "pxor %%mm2, %%mm2 \n"
789 "pmaxub %%mm0, %%mm1 \n"
790 "punpcklbw %%mm2, %%mm1 \n"
791 "movq %%mm1, %0 \n"
792 :"=m"(*bS[dir][edge])
793 ::"memory"
794 );
795 }
796 edges = 4;
797 step = 1;
798 }
799 __asm__ volatile(
800 "movq (%0), %%mm0 \n\t"
801 "movq 8(%0), %%mm1 \n\t"
802 "movq 16(%0), %%mm2 \n\t"
803 "movq 24(%0), %%mm3 \n\t"
804 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
805 "movq %%mm0, (%0) \n\t"
806 "movq %%mm3, 8(%0) \n\t"
807 "movq %%mm4, 16(%0) \n\t"
808 "movq %%mm2, 24(%0) \n\t"
809 ::"r"(bS[0])
810 :"memory"
811 );
812 }
813
814 /***********************************/
815 /* motion compensation */
816
817 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
818 "mov"#q" "#C", "#T" \n\t"\
819 "mov"#d" (%0), "#F" \n\t"\
820 "paddw "#D", "#T" \n\t"\
821 "psllw $2, "#T" \n\t"\
822 "psubw "#B", "#T" \n\t"\
823 "psubw "#E", "#T" \n\t"\
824 "punpcklbw "#Z", "#F" \n\t"\
825 "pmullw %4, "#T" \n\t"\
826 "paddw %5, "#A" \n\t"\
827 "add %2, %0 \n\t"\
828 "paddw "#F", "#A" \n\t"\
829 "paddw "#A", "#T" \n\t"\
830 "psraw $5, "#T" \n\t"\
831 "packuswb "#T", "#T" \n\t"\
832 OP(T, (%1), A, d)\
833 "add %3, %1 \n\t"
834
835 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
836 "mov"#q" "#C", "#T" \n\t"\
837 "mov"#d" (%0), "#F" \n\t"\
838 "paddw "#D", "#T" \n\t"\
839 "psllw $2, "#T" \n\t"\
840 "paddw %4, "#A" \n\t"\
841 "psubw "#B", "#T" \n\t"\
842 "psubw "#E", "#T" \n\t"\
843 "punpcklbw "#Z", "#F" \n\t"\
844 "pmullw %3, "#T" \n\t"\
845 "paddw "#F", "#A" \n\t"\
846 "add %2, %0 \n\t"\
847 "paddw "#A", "#T" \n\t"\
848 "mov"#q" "#T", "#OF"(%1) \n\t"
849
850 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
851 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
852 #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)
853 #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)
854
855
856 #define QPEL_H264(OPNAME, OP, MMX)\
857 \
858 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
859 int h=8;\
860 __asm__ volatile(\
861 "pxor %%mm7, %%mm7 \n\t"\
862 "movq %0, %%mm6 \n\t"\
863 :: "m"(ff_pw_5)\
864 );\
865 do{\
866 __asm__ volatile(\
867 "movq (%0), %%mm0 \n\t"\
868 "movq 1(%0), %%mm2 \n\t"\
869 "movq %%mm0, %%mm1 \n\t"\
870 "movq %%mm2, %%mm3 \n\t"\
871 "punpcklbw %%mm7, %%mm0 \n\t"\
872 "punpckhbw %%mm7, %%mm1 \n\t"\
873 "punpcklbw %%mm7, %%mm2 \n\t"\
874 "punpckhbw %%mm7, %%mm3 \n\t"\
875 "paddw %%mm2, %%mm0 \n\t"\
876 "paddw %%mm3, %%mm1 \n\t"\
877 "psllw $2, %%mm0 \n\t"\
878 "psllw $2, %%mm1 \n\t"\
879 "movq -1(%0), %%mm2 \n\t"\
880 "movq 2(%0), %%mm4 \n\t"\
881 "movq %%mm2, %%mm3 \n\t"\
882 "movq %%mm4, %%mm5 \n\t"\
883 "punpcklbw %%mm7, %%mm2 \n\t"\
884 "punpckhbw %%mm7, %%mm3 \n\t"\
885 "punpcklbw %%mm7, %%mm4 \n\t"\
886 "punpckhbw %%mm7, %%mm5 \n\t"\
887 "paddw %%mm4, %%mm2 \n\t"\
888 "paddw %%mm3, %%mm5 \n\t"\
889 "psubw %%mm2, %%mm0 \n\t"\
890 "psubw %%mm5, %%mm1 \n\t"\
891 "pmullw %%mm6, %%mm0 \n\t"\
892 "pmullw %%mm6, %%mm1 \n\t"\
893 "movd -2(%0), %%mm2 \n\t"\
894 "movd 7(%0), %%mm5 \n\t"\
895 "punpcklbw %%mm7, %%mm2 \n\t"\
896 "punpcklbw %%mm7, %%mm5 \n\t"\
897 "paddw %%mm3, %%mm2 \n\t"\
898 "paddw %%mm5, %%mm4 \n\t"\
899 "movq %5, %%mm5 \n\t"\
900 "paddw %%mm5, %%mm2 \n\t"\
901 "paddw %%mm5, %%mm4 \n\t"\
902 "paddw %%mm2, %%mm0 \n\t"\
903 "paddw %%mm4, %%mm1 \n\t"\
904 "psraw $5, %%mm0 \n\t"\
905 "psraw $5, %%mm1 \n\t"\
906 "movq (%2), %%mm4 \n\t"\
907 "packuswb %%mm1, %%mm0 \n\t"\
908 PAVGB" %%mm4, %%mm0 \n\t"\
909 OP(%%mm0, (%1),%%mm5, q)\
910 "add %4, %0 \n\t"\
911 "add %4, %1 \n\t"\
912 "add %3, %2 \n\t"\
913 : "+a"(src), "+c"(dst), "+d"(src2)\
914 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
915 "m"(ff_pw_16)\
916 : "memory"\
917 );\
918 }while(--h);\
919 }\
920 \
921 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
922 int w = size>>4;\
923 do{\
924 int h = size;\
925 __asm__ volatile(\
926 "1: \n\t"\
927 "movq (%0), %%mm0 \n\t"\
928 "movq 8(%0), %%mm3 \n\t"\
929 "movq 2(%0), %%mm1 \n\t"\
930 "movq 10(%0), %%mm4 \n\t"\
931 "paddw %%mm4, %%mm0 \n\t"\
932 "paddw %%mm3, %%mm1 \n\t"\
933 "paddw 18(%0), %%mm3 \n\t"\
934 "paddw 16(%0), %%mm4 \n\t"\
935 "movq 4(%0), %%mm2 \n\t"\
936 "movq 12(%0), %%mm5 \n\t"\
937 "paddw 6(%0), %%mm2 \n\t"\
938 "paddw 14(%0), %%mm5 \n\t"\
939 "psubw %%mm1, %%mm0 \n\t"\
940 "psubw %%mm4, %%mm3 \n\t"\
941 "psraw $2, %%mm0 \n\t"\
942 "psraw $2, %%mm3 \n\t"\
943 "psubw %%mm1, %%mm0 \n\t"\
944 "psubw %%mm4, %%mm3 \n\t"\
945 "paddsw %%mm2, %%mm0 \n\t"\
946 "paddsw %%mm5, %%mm3 \n\t"\
947 "psraw $2, %%mm0 \n\t"\
948 "psraw $2, %%mm3 \n\t"\
949 "paddw %%mm2, %%mm0 \n\t"\
950 "paddw %%mm5, %%mm3 \n\t"\
951 "psraw $6, %%mm0 \n\t"\
952 "psraw $6, %%mm3 \n\t"\
953 "packuswb %%mm3, %%mm0 \n\t"\
954 OP(%%mm0, (%1),%%mm7, q)\
955 "add $48, %0 \n\t"\
956 "add %3, %1 \n\t"\
957 "decl %2 \n\t"\
958 " jnz 1b \n\t"\
959 : "+a"(tmp), "+c"(dst), "+g"(h)\
960 : "S"((x86_reg)dstStride)\
961 : "memory"\
962 );\
963 tmp += 8 - size*24;\
964 dst += 8 - size*dstStride;\
965 }while(w--);\
966 }\
967 \
968 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
969 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
970 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
971 src += 8*dstStride;\
972 dst += 8*dstStride;\
973 src2 += 8*src2Stride;\
974 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
975 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
976 }\
977 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
978 {\
979 do{\
980 __asm__ volatile(\
981 "movq (%1), %%mm0 \n\t"\
982 "movq 8(%1), %%mm1 \n\t"\
983 "movq 48(%1), %%mm2 \n\t"\
984 "movq 8+48(%1), %%mm3 \n\t"\
985 "psraw $5, %%mm0 \n\t"\
986 "psraw $5, %%mm1 \n\t"\
987 "psraw $5, %%mm2 \n\t"\
988 "psraw $5, %%mm3 \n\t"\
989 "packuswb %%mm1, %%mm0 \n\t"\
990 "packuswb %%mm3, %%mm2 \n\t"\
991 PAVGB" (%0), %%mm0 \n\t"\
992 PAVGB" (%0,%3), %%mm2 \n\t"\
993 OP(%%mm0, (%2), %%mm5, q)\
994 OP(%%mm2, (%2,%4), %%mm5, q)\
995 ::"a"(src8), "c"(src16), "d"(dst),\
996 "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
997 :"memory");\
998 src8 += 2L*src8Stride;\
999 src16 += 48;\
1000 dst += 2L*dstStride;\
1001 }while(h-=2);\
1002 }\
1003 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1004 {\
1005 OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
1006 OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1007 }\
1008
1009
1010 #if ARCH_X86_64
1011 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1012 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1013 int h=16;\
1014 __asm__ volatile(\
1015 "pxor %%xmm15, %%xmm15 \n\t"\
1016 "movdqa %6, %%xmm14 \n\t"\
1017 "movdqa %7, %%xmm13 \n\t"\
1018 "1: \n\t"\
1019 "lddqu 6(%0), %%xmm1 \n\t"\
1020 "lddqu -2(%0), %%xmm7 \n\t"\
1021 "movdqa %%xmm1, %%xmm0 \n\t"\
1022 "punpckhbw %%xmm15, %%xmm1 \n\t"\
1023 "punpcklbw %%xmm15, %%xmm0 \n\t"\
1024 "punpcklbw %%xmm15, %%xmm7 \n\t"\
1025 "movdqa %%xmm1, %%xmm2 \n\t"\
1026 "movdqa %%xmm0, %%xmm6 \n\t"\
1027 "movdqa %%xmm1, %%xmm3 \n\t"\
1028 "movdqa %%xmm0, %%xmm8 \n\t"\
1029 "movdqa %%xmm1, %%xmm4 \n\t"\
1030 "movdqa %%xmm0, %%xmm9 \n\t"\
1031 "movdqa %%xmm0, %%xmm12 \n\t"\
1032 "movdqa %%xmm1, %%xmm11 \n\t"\
1033 "palignr $10,%%xmm0, %%xmm11\n\t"\
1034 "palignr $10,%%xmm7, %%xmm12\n\t"\
1035 "palignr $2, %%xmm0, %%xmm4 \n\t"\
1036 "palignr $2, %%xmm7, %%xmm9 \n\t"\
1037 "palignr $4, %%xmm0, %%xmm3 \n\t"\
1038 "palignr $4, %%xmm7, %%xmm8 \n\t"\
1039 "palignr $6, %%xmm0, %%xmm2 \n\t"\
1040 "palignr $6, %%xmm7, %%xmm6 \n\t"\
1041 "paddw %%xmm0 ,%%xmm11 \n\t"\
1042 "palignr $8, %%xmm0, %%xmm1 \n\t"\
1043 "palignr $8, %%xmm7, %%xmm0 \n\t"\
1044 "paddw %%xmm12,%%xmm7 \n\t"\
1045 "paddw %%xmm3, %%xmm2 \n\t"\
1046 "paddw %%xmm8, %%xmm6 \n\t"\
1047 "paddw %%xmm4, %%xmm1 \n\t"\
1048 "paddw %%xmm9, %%xmm0 \n\t"\
1049 "psllw $2, %%xmm2 \n\t"\
1050 "psllw $2, %%xmm6 \n\t"\
1051 "psubw %%xmm1, %%xmm2 \n\t"\
1052 "psubw %%xmm0, %%xmm6 \n\t"\
1053 "paddw %%xmm13,%%xmm11 \n\t"\
1054 "paddw %%xmm13,%%xmm7 \n\t"\
1055 "pmullw %%xmm14,%%xmm2 \n\t"\
1056 "pmullw %%xmm14,%%xmm6 \n\t"\
1057 "lddqu (%2), %%xmm3 \n\t"\
1058 "paddw %%xmm11,%%xmm2 \n\t"\
1059 "paddw %%xmm7, %%xmm6 \n\t"\
1060 "psraw $5, %%xmm2 \n\t"\
1061 "psraw $5, %%xmm6 \n\t"\
1062 "packuswb %%xmm2,%%xmm6 \n\t"\
1063 "pavgb %%xmm3, %%xmm6 \n\t"\
1064 OP(%%xmm6, (%1), %%xmm4, dqa)\
1065 "add %5, %0 \n\t"\
1066 "add %5, %1 \n\t"\
1067 "add %4, %2 \n\t"\
1068 "decl %3 \n\t"\
1069 "jg 1b \n\t"\
1070 : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
1071 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1072 "m"(ff_pw_5), "m"(ff_pw_16)\
1073 : "memory"\
1074 );\
1075 }
1076 #else // ARCH_X86_64
1077 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
1078 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1079 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1080 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1081 src += 8*dstStride;\
1082 dst += 8*dstStride;\
1083 src2 += 8*src2Stride;\
1084 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1085 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1086 }
1087 #endif // ARCH_X86_64
1088
1089 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
1090 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1091 int h=8;\
1092 __asm__ volatile(\
1093 "pxor %%xmm7, %%xmm7 \n\t"\
1094 "movdqa %0, %%xmm6 \n\t"\
1095 :: "m"(ff_pw_5)\
1096 );\
1097 do{\
1098 __asm__ volatile(\
1099 "lddqu -2(%0), %%xmm1 \n\t"\
1100 "movdqa %%xmm1, %%xmm0 \n\t"\
1101 "punpckhbw %%xmm7, %%xmm1 \n\t"\
1102 "punpcklbw %%xmm7, %%xmm0 \n\t"\
1103 "movdqa %%xmm1, %%xmm2 \n\t"\
1104 "movdqa %%xmm1, %%xmm3 \n\t"\
1105 "movdqa %%xmm1, %%xmm4 \n\t"\
1106 "movdqa %%xmm1, %%xmm5 \n\t"\
1107 "palignr $2, %%xmm0, %%xmm4 \n\t"\
1108 "palignr $4, %%xmm0, %%xmm3 \n\t"\
1109 "palignr $6, %%xmm0, %%xmm2 \n\t"\
1110 "palignr $8, %%xmm0, %%xmm1 \n\t"\
1111 "palignr $10,%%xmm0, %%xmm5 \n\t"\
1112 "paddw %%xmm5, %%xmm0 \n\t"\
1113 "paddw %%xmm3, %%xmm2 \n\t"\
1114 "paddw %%xmm4, %%xmm1 \n\t"\
1115 "psllw $2, %%xmm2 \n\t"\
1116 "movq (%2), %%xmm3 \n\t"\
1117 "psubw %%xmm1, %%xmm2 \n\t"\
1118 "paddw %5, %%xmm0 \n\t"\
1119 "pmullw %%xmm6, %%xmm2 \n\t"\
1120 "paddw %%xmm0, %%xmm2 \n\t"\
1121 "psraw $5, %%xmm2 \n\t"\
1122 "packuswb %%xmm2, %%xmm2 \n\t"\
1123 "pavgb %%xmm3, %%xmm2 \n\t"\
1124 OP(%%xmm2, (%1), %%xmm4, q)\
1125 "add %4, %0 \n\t"\
1126 "add %4, %1 \n\t"\
1127 "add %3, %2 \n\t"\
1128 : "+a"(src), "+c"(dst), "+d"(src2)\
1129 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
1130 "m"(ff_pw_16)\
1131 : "memory"\
1132 );\
1133 }while(--h);\
1134 }\
1135 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
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 %%xmm7, %%xmm7 \n\t"\
1141 "movdqa %5, %%xmm6 \n\t"\
1142 "1: \n\t"\
1143 "lddqu -2(%0), %%xmm1 \n\t"\
1144 "movdqa %%xmm1, %%xmm0 \n\t"\
1145 "punpckhbw %%xmm7, %%xmm1 \n\t"\
1146 "punpcklbw %%xmm7, %%xmm0 \n\t"\
1147 "movdqa %%xmm1, %%xmm2 \n\t"\
1148 "movdqa %%xmm1, %%xmm3 \n\t"\
1149 "movdqa %%xmm1, %%xmm4 \n\t"\
1150 "movdqa %%xmm1, %%xmm5 \n\t"\
1151 "palignr $2, %%xmm0, %%xmm4 \n\t"\
1152 "palignr $4, %%xmm0, %%xmm3 \n\t"\
1153 "palignr $6, %%xmm0, %%xmm2 \n\t"\
1154 "palignr $8, %%xmm0, %%xmm1 \n\t"\
1155 "palignr $10,%%xmm0, %%xmm5 \n\t"\
1156 "paddw %%xmm5, %%xmm0 \n\t"\
1157 "paddw %%xmm3, %%xmm2 \n\t"\
1158 "paddw %%xmm4, %%xmm1 \n\t"\
1159 "psllw $2, %%xmm2 \n\t"\
1160 "psubw %%xmm1, %%xmm2 \n\t"\
1161 "paddw %6, %%xmm0 \n\t"\
1162 "pmullw %%xmm6, %%xmm2 \n\t"\
1163 "paddw %%xmm0, %%xmm2 \n\t"\
1164 "psraw $5, %%xmm2 \n\t"\
1165 "packuswb %%xmm2, %%xmm2 \n\t"\
1166 OP(%%xmm2, (%1), %%xmm4, q)\
1167 "add %3, %0 \n\t"\
1168 "add %4, %1 \n\t"\
1169 "decl %2 \n\t"\
1170 " jnz 1b \n\t"\
1171 : "+a"(src), "+c"(dst), "+g"(h)\
1172 : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\
1173 "m"(ff_pw_5), "m"(ff_pw_16)\
1174 : "memory"\
1175 );\
1176 }\
1177 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1178 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1179 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1180 src += 8*srcStride;\
1181 dst += 8*dstStride;\
1182 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1183 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1184 }\
1185
1186 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
1187 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
1188 src -= 2*srcStride;\
1189 \
1190 __asm__ volatile(\
1191 "pxor %%xmm7, %%xmm7 \n\t"\
1192 "movq (%0), %%xmm0 \n\t"\
1193 "add %2, %0 \n\t"\
1194 "movq (%0), %%xmm1 \n\t"\
1195 "add %2, %0 \n\t"\
1196 "movq (%0), %%xmm2 \n\t"\
1197 "add %2, %0 \n\t"\
1198 "movq (%0), %%xmm3 \n\t"\
1199 "add %2, %0 \n\t"\
1200 "movq (%0), %%xmm4 \n\t"\
1201 "add %2, %0 \n\t"\
1202 "punpcklbw %%xmm7, %%xmm0 \n\t"\
1203 "punpcklbw %%xmm7, %%xmm1 \n\t"\
1204 "punpcklbw %%xmm7, %%xmm2 \n\t"\
1205 "punpcklbw %%xmm7, %%xmm3 \n\t"\
1206 "punpcklbw %%xmm7, %%xmm4 \n\t"\
1207 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1208 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1209 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1210 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1211 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1212 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1213 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1214 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1215 \
1216 : "+a"(src), "+c"(dst)\
1217 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1218 : "memory"\
1219 );\
1220 if(h==16){\
1221 __asm__ volatile(\
1222 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1223 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1224 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
1225 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
1226 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
1227 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
1228 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
1229 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
1230 \
1231 : "+a"(src), "+c"(dst)\
1232 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
1233 : "memory"\
1234 );\
1235 }\
1236 }\
1237 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1238 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
1239 }\
1240 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1241 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
1242 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1243 }
1244
1245 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
1246 int w = (size+8)>>3;
1247 src -= 2*srcStride+2;
1248 while(w--){
1249 __asm__ volatile(
1250 "pxor %%xmm7, %%xmm7 \n\t"
1251 "movq (%0), %%xmm0 \n\t"
1252 "add %2, %0 \n\t"
1253 "movq (%0), %%xmm1 \n\t"
1254 "add %2, %0 \n\t"
1255 "movq (%0), %%xmm2 \n\t"
1256 "add %2, %0 \n\t"
1257 "movq (%0), %%xmm3 \n\t"
1258 "add %2, %0 \n\t"
1259 "movq (%0), %%xmm4 \n\t"
1260 "add %2, %0 \n\t"
1261 "punpcklbw %%xmm7, %%xmm0 \n\t"
1262 "punpcklbw %%xmm7, %%xmm1 \n\t"
1263 "punpcklbw %%xmm7, %%xmm2 \n\t"
1264 "punpcklbw %%xmm7, %%xmm3 \n\t"
1265 "punpcklbw %%xmm7, %%xmm4 \n\t"
1266 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
1267 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
1268 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
1269 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
1270 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
1271 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
1272 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
1273 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
1274 : "+a"(src)
1275 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1276 : "memory"
1277 );
1278 if(size==16){
1279 __asm__ volatile(
1280 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 8*48)
1281 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 9*48)
1282 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
1283 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
1284 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
1285 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
1286 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
1287 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
1288 : "+a"(src)
1289 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
1290 : "memory"
1291 );
1292 }
1293 tmp += 8;
1294 src += 8 - (size+5)*srcStride;
1295 }
1296 }
1297
1298 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
1299 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
1300 int h = size;\
1301 if(size == 16){\
1302 __asm__ volatile(\
1303 "1: \n\t"\
1304 "movdqa 32(%0), %%xmm4 \n\t"\
1305 "movdqa 16(%0), %%xmm5 \n\t"\
1306 "movdqa (%0), %%xmm7 \n\t"\
1307 "movdqa %%xmm4, %%xmm3 \n\t"\
1308 "movdqa %%xmm4, %%xmm2 \n\t"\
1309 "movdqa %%xmm4, %%xmm1 \n\t"\
1310 "movdqa %%xmm4, %%xmm0 \n\t"\
1311 "palignr $10, %%xmm5, %%xmm0 \n\t"\
1312 "palignr $8, %%xmm5, %%xmm1 \n\t"\
1313 "palignr $6, %%xmm5, %%xmm2 \n\t"\
1314 "palignr $4, %%xmm5, %%xmm3 \n\t"\
1315 "palignr $2, %%xmm5, %%xmm4 \n\t"\
1316 "paddw %%xmm5, %%xmm0 \n\t"\
1317 "paddw %%xmm4, %%xmm1 \n\t"\
1318 "paddw %%xmm3, %%xmm2 \n\t"\
1319 "movdqa %%xmm5, %%xmm6 \n\t"\
1320 "movdqa %%xmm5, %%xmm4 \n\t"\
1321 "movdqa %%xmm5, %%xmm3 \n\t"\
1322 "palignr $8, %%xmm7, %%xmm4 \n\t"\
1323 "palignr $2, %%xmm7, %%xmm6 \n\t"\
1324 "palignr $10, %%xmm7, %%xmm3 \n\t"\
1325 "paddw %%xmm6, %%xmm4 \n\t"\
1326 "movdqa %%xmm5, %%xmm6 \n\t"\
1327 "palignr $6, %%xmm7, %%xmm5 \n\t"\
1328 "palignr $4, %%xmm7, %%xmm6 \n\t"\
1329 "paddw %%xmm7, %%xmm3 \n\t"\
1330 "paddw %%xmm6, %%xmm5 \n\t"\
1331 \
1332 "psubw %%xmm1, %%xmm0 \n\t"\
1333 "psubw %%xmm4, %%xmm3 \n\t"\
1334 "psraw $2, %%xmm0 \n\t"\
1335 "psraw $2, %%xmm3 \n\t"\
1336 "psubw %%xmm1, %%xmm0 \n\t"\
1337 "psubw %%xmm4, %%xmm3 \n\t"\
1338 "paddw %%xmm2, %%xmm0 \n\t"\
1339 "paddw %%xmm5, %%xmm3 \n\t"\
1340 "psraw $2, %%xmm0 \n\t"\
1341 "psraw $2, %%xmm3 \n\t"\
1342 "paddw %%xmm2, %%xmm0 \n\t"\
1343 "paddw %%xmm5, %%xmm3 \n\t"\
1344 "psraw $6, %%xmm0 \n\t"\
1345 "psraw $6, %%xmm3 \n\t"\
1346 "packuswb %%xmm0, %%xmm3 \n\t"\
1347 OP(%%xmm3, (%1), %%xmm7, dqa)\
1348 "add $48, %0 \n\t"\
1349 "add %3, %1 \n\t"\
1350 "decl %2 \n\t"\
1351 " jnz 1b \n\t"\
1352 : "+a"(tmp), "+c"(dst), "+g"(h)\
1353 : "S"((x86_reg)dstStride)\
1354 : "memory"\
1355 );\
1356 }else{\
1357 __asm__ volatile(\
1358 "1: \n\t"\
1359 "movdqa 16(%0), %%xmm1 \n\t"\
1360 "movdqa (%0), %%xmm0 \n\t"\
1361 "movdqa %%xmm1, %%xmm2 \n\t"\
1362 "movdqa %%xmm1, %%xmm3 \n\t"\
1363 "movdqa %%xmm1, %%xmm4 \n\t"\
1364 "movdqa %%xmm1, %%xmm5 \n\t"\
1365 "palignr $10, %%xmm0, %%xmm5 \n\t"\
1366 "palignr $8, %%xmm0, %%xmm4 \n\t"\
1367 "palignr $6, %%xmm0, %%xmm3 \n\t"\
1368 "palignr $4, %%xmm0, %%xmm2 \n\t"\
1369 "palignr $2, %%xmm0, %%xmm1 \n\t"\
1370 "paddw %%xmm5, %%xmm0 \n\t"\
1371 "paddw %%xmm4, %%xmm1 \n\t"\
1372 "paddw %%xmm3, %%xmm2 \n\t"\
1373 "psubw %%xmm1, %%xmm0 \n\t"\
1374 "psraw $2, %%xmm0 \n\t"\
1375 "psubw %%xmm1, %%xmm0 \n\t"\
1376 "paddw %%xmm2, %%xmm0 \n\t"\
1377 "psraw $2, %%xmm0 \n\t"\
1378 "paddw %%xmm2, %%xmm0 \n\t"\
1379 "psraw $6, %%xmm0 \n\t"\
1380 "packuswb %%xmm0, %%xmm0 \n\t"\
1381 OP(%%xmm0, (%1), %%xmm7, q)\
1382 "add $48, %0 \n\t"\
1383 "add %3, %1 \n\t"\
1384 "decl %2 \n\t"\
1385 " jnz 1b \n\t"\
1386 : "+a"(tmp), "+c"(dst), "+g"(h)\
1387 : "S"((x86_reg)dstStride)\
1388 : "memory"\
1389 );\
1390 }\
1391 }
1392
1393 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
1394 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){\
1395 put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
1396 OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1397 }\
1398 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1399 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
1400 }\
1401 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1402 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1403 }\
1404
1405 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1406 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1407 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1408 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1409 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1410 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1411 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1412 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1413
1414 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1415 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1416 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1417 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1418 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1419 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1420 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1421 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1422
1423 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1424 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1425 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1426 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1427
1428 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1429 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1430 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1431 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1432
1433 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1434 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1435
1436 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1437 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1438 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1439 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1440 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1441
1442 // static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1443 // put_pixels16_sse2(dst, src, stride, 16);
1444 // }
1445 // static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1446 // avg_pixels16_sse2(dst, src, stride, 16);
1447 // }
1448 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1449 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1450
1451 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1452 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1453 OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1454 }\
1455
1456 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1457 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1458 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1459 }\
1460 \
1461 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1462 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1463 }\
1464 \
1465 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1466 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1467 }\
1468
1469 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1470 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1471 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1472 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1473 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1474 }\
1475 \
1476 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1477 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1478 }\
1479 \
1480 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1481 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1482 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1483 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1484 }\
1485
1486 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1487 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1488 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1489 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1490 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1491 }\
1492 \
1493 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1494 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1495 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1496 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1497 }\
1498 \
1499 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1500 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1501 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1502 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1503 }\
1504 \
1505 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1506 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1507 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1508 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1509 }\
1510 \
1511 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1512 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
1513 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
1514 }\
1515 \
1516 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1517 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1518 uint8_t * const halfHV= temp;\
1519 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1520 assert(((int)temp & 7) == 0);\
1521 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1522 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1523 }\
1524 \
1525 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1526 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1527 uint8_t * const halfHV= temp;\
1528 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1529 assert(((int)temp & 7) == 0);\
1530 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1531 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1532 }\
1533 \
1534 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1535 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1536 uint8_t * const halfHV= temp;\
1537 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1538 assert(((int)temp & 7) == 0);\
1539 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1540 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1541 }\
1542 \
1543 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1544 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1545 uint8_t * const halfHV= temp;\
1546 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1547 assert(((int)temp & 7) == 0);\
1548 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1549 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1550 }\
1551
1552 #define H264_MC_4816(MMX)\
1553 H264_MC(put_, 4, MMX, 8)\
1554 H264_MC(put_, 8, MMX, 8)\
1555 H264_MC(put_, 16,MMX, 8)\
1556 H264_MC(avg_, 4, MMX, 8)\
1557 H264_MC(avg_, 8, MMX, 8)\
1558 H264_MC(avg_, 16,MMX, 8)\
1559
1560 #define H264_MC_816(QPEL, XMM)\
1561 QPEL(put_, 8, XMM, 16)\
1562 QPEL(put_, 16,XMM, 16)\
1563 QPEL(avg_, 8, XMM, 16)\
1564 QPEL(avg_, 16,XMM, 16)\
1565
1566
1567 #define AVG_3DNOW_OP(a,b,temp, size) \
1568 "mov" #size " " #b ", " #temp " \n\t"\
1569 "pavgusb " #temp ", " #a " \n\t"\
1570 "mov" #size " " #a ", " #b " \n\t"
1571 #define AVG_MMX2_OP(a,b,temp, size) \
1572 "mov" #size " " #b ", " #temp " \n\t"\
1573 "pavgb " #temp ", " #a " \n\t"\
1574 "mov" #size " " #a ", " #b " \n\t"
1575
1576 ///this does not get detected correctly, uncomment on AMD machine
1577 #ifdef HAVE_AMD3DNOW
1578 #define PAVGB "pavgusb"
1579 //QPEL_H264(put_, PUT_OP, 3dnow)
1580 //QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1581 #undef PAVGB
1582 #endif
1583
1584 #define PAVGB "pavgb"
1585 QPEL_H264(put_, PUT_OP, mmx2)
1586 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
1587 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
1588 QPEL_H264_V_XMM(avg_, AVG_MMX2_OP, sse2)
1589 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
1590 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, sse2)
1591 #if HAVE_SSSE3
1592 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
1593 QPEL_H264_H_XMM(avg_, AVG_MMX2_OP, ssse3)
1594 QPEL_H264_HV2_XMM(put_, PUT_OP, ssse3)
1595 QPEL_H264_HV2_XMM(avg_, AVG_MMX2_OP, ssse3)
1596 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
1597 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, ssse3)
1598 #endif
1599 #undef PAVGB
1600
1601 H264_MC_816(H264_MC_V, sse2)
1602 H264_MC_816(H264_MC_HV, sse2)
1603 #if HAVE_SSSE3
1604 H264_MC_816(H264_MC_H, ssse3)
1605 H264_MC_816(H264_MC_HV, ssse3)
1606 #endif
1607
1608 /* rnd interleaved with rnd div 8, use p+1 to access rnd div 8 */
1609 DECLARE_ALIGNED(8, static const uint64_t, h264_rnd_reg)[4] = {
1610 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL
1611 };
1612
1613 #if HAVE_SSSE3
1614 #define AVG_OP(X)
1615 #undef H264_CHROMA_MC8_TMPL
1616 #undef H264_CHROMA_MC4_TMPL
1617 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_ssse3
1618 #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_ssse3
1619 #define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1620 #include "dsputil_h264_template_ssse3.c"
1621 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)
1622 {
1623 put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
1624 }
1625
1626 #undef AVG_OP
1627 #undef H264_CHROMA_MC8_TMPL
1628 #undef H264_CHROMA_MC4_TMPL
1629 #undef H264_CHROMA_MC8_MV0
1630 #define AVG_OP(X) X
1631 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_ssse3
1632 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_ssse3
1633 #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
1634 #include "dsputil_h264_template_ssse3.c"
1635 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)
1636 {
1637 avg_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
1638 }
1639 #undef AVG_OP
1640 #undef H264_CHROMA_MC8_TMPL
1641 #undef H264_CHROMA_MC4_TMPL
1642 #undef H264_CHROMA_MC8_MV0
1643 #endif
1644
1645 /***********************************/
1646 /* weighted prediction */
1647
1648 static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
1649 {
1650 int x, y;
1651 offset <<= log2_denom;
1652 offset += (1 << log2_denom) >> 1;
1653 __asm__ volatile(
1654 "movd %0, %%mm4 \n\t"
1655 "movd %1, %%mm5 \n\t"
1656 "movd %2, %%mm6 \n\t"
1657 "pshufw $0, %%mm4, %%mm4 \n\t"
1658 "pshufw $0, %%mm5, %%mm5 \n\t"
1659 "pxor %%mm7, %%mm7 \n\t"
1660 :: "g"(weight), "g"(offset), "g"(log2_denom)
1661 );
1662 for(y=0; y<h; y+=2){
1663 for(x=0; x<w; x+=4){
1664 __asm__ volatile(
1665 "movd %0, %%mm0 \n\t"
1666 "movd %1, %%mm1 \n\t"
1667 "punpcklbw %%mm7, %%mm0 \n\t"
1668 "punpcklbw %%mm7, %%mm1 \n\t"
1669 "pmullw %%mm4, %%mm0 \n\t"
1670 "pmullw %%mm4, %%mm1 \n\t"
1671 "paddsw %%mm5, %%mm0 \n\t"
1672 "paddsw %%mm5, %%mm1 \n\t"
1673 "psraw %%mm6, %%mm0 \n\t"
1674 "psraw %%mm6, %%mm1 \n\t"
1675 "packuswb %%mm7, %%mm0 \n\t"
1676 "packuswb %%mm7, %%mm1 \n\t"
1677 "movd %%mm0, %0 \n\t"
1678 "movd %%mm1, %1 \n\t"
1679 : "+m"(*(uint32_t*)(dst+x)),
1680 "+m"(*(uint32_t*)(dst+x+stride))
1681 );
1682 }
1683 dst += 2*stride;
1684 }
1685 }
1686
1687 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)
1688 {
1689 int x, y;
1690 offset = ((offset + 1) | 1) << log2_denom;
1691 __asm__ volatile(
1692 "movd %0, %%mm3 \n\t"
1693 "movd %1, %%mm4 \n\t"
1694 "movd %2, %%mm5 \n\t"
1695 "movd %3, %%mm6 \n\t"
1696 "pshufw $0, %%mm3, %%mm3 \n\t"
1697 "pshufw $0, %%mm4, %%mm4 \n\t"
1698 "pshufw $0, %%mm5, %%mm5 \n\t"
1699 "pxor %%mm7, %%mm7 \n\t"
1700 :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
1701 );
1702 for(y=0; y<h; y++){
1703 for(x=0; x<w; x+=4){
1704 __asm__ volatile(
1705 "movd %0, %%mm0 \n\t"
1706 "movd %1, %%mm1 \n\t"
1707 "punpcklbw %%mm7, %%mm0 \n\t"
1708 "punpcklbw %%mm7, %%mm1 \n\t"
1709 "pmullw %%mm3, %%mm0 \n\t"
1710 "pmullw %%mm4, %%mm1 \n\t"
1711 "paddsw %%mm1, %%mm0 \n\t"
1712 "paddsw %%mm5, %%mm0 \n\t"
1713 "psraw %%mm6, %%mm0 \n\t"
1714 "packuswb %%mm0, %%mm0 \n\t"
1715 "movd %%mm0, %0 \n\t"
1716 : "+m"(*(uint32_t*)(dst+x))
1717 : "m"(*(uint32_t*)(src+x))
1718 );
1719 }
1720 src += stride;
1721 dst += stride;
1722 }
1723 }
1724
1725 #define H264_WEIGHT(W,H) \
1726 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){ \
1727 ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
1728 } \
1729 static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
1730 ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
1731 }
1732
1733 H264_WEIGHT(16,16)
1734 H264_WEIGHT(16, 8)
1735 H264_WEIGHT( 8,16)
1736 H264_WEIGHT( 8, 8)
1737 H264_WEIGHT( 8, 4)
1738 H264_WEIGHT( 4, 8)
1739 H264_WEIGHT( 4, 4)
1740 H264_WEIGHT( 4, 2)
1741