VirtualBox

source: vbox/trunk/src/libs/ffmpeg-20060710/libavcodec/i386/h264dsp_mmx.c@ 9384

Last change on this file since 9384 was 5776, checked in by vboxsync, 17 years ago

ffmpeg: exported to OSE
File size: 54.3 KB
Line 
1/*
2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3 *
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
8 *
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
13 *
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18
19
20/***********************************/
21/* IDCT */
22
23/* in/out: mma=mma+mmb, mmb=mmb-mma */
24#define SUMSUB_BA( a, b ) \
25 "paddw "#b", "#a" \n\t"\
26 "paddw "#b", "#b" \n\t"\
27 "psubw "#a", "#b" \n\t"
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 TRANSPOSE4(a,b,c,d,t)\
50 SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
51 SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
52 SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
53 SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
54
55#define STORE_DIFF_4P( p, t, z ) \
56 "psraw $6, "#p" \n\t"\
57 "movd (%0), "#t" \n\t"\
58 "punpcklbw "#z", "#t" \n\t"\
59 "paddsw "#t", "#p" \n\t"\
60 "packuswb "#z", "#p" \n\t"\
61 "movd "#p", (%0) \n\t"
62
63static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
64{
65 /* Load dct coeffs */
66 asm volatile(
67 "movq (%0), %%mm0 \n\t"
68 "movq 8(%0), %%mm1 \n\t"
69 "movq 16(%0), %%mm2 \n\t"
70 "movq 24(%0), %%mm3 \n\t"
71 :: "r"(block) );
72
73 asm volatile(
74 /* mm1=s02+s13 mm2=s02-s13 mm4=d02+d13 mm0=d02-d13 */
75 IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
76
77 "movq %0, %%mm6 \n\t"
78 /* in: 1,4,0,2 out: 1,2,3,0 */
79 TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
80
81 "paddw %%mm6, %%mm3 \n\t"
82
83 /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */
84 IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
85
86 "pxor %%mm7, %%mm7 \n\t"
87 :: "m"(ff_pw_32));
88
89 asm volatile(
90 STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
91 "add %1, %0 \n\t"
92 STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
93 "add %1, %0 \n\t"
94 STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
95 "add %1, %0 \n\t"
96 STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
97 : "+r"(dst)
98 : "r" ((long)stride)
99 );
100}
101
102static inline void h264_idct8_1d(int16_t *block)
103{
104 asm volatile(
105 "movq 112(%0), %%mm7 \n\t"
106 "movq 80(%0), %%mm5 \n\t"
107 "movq 48(%0), %%mm3 \n\t"
108 "movq 16(%0), %%mm1 \n\t"
109
110 "movq %%mm7, %%mm4 \n\t"
111 "movq %%mm3, %%mm6 \n\t"
112 "movq %%mm5, %%mm0 \n\t"
113 "movq %%mm7, %%mm2 \n\t"
114 "psraw $1, %%mm4 \n\t"
115 "psraw $1, %%mm6 \n\t"
116 "psubw %%mm7, %%mm0 \n\t"
117 "psubw %%mm6, %%mm2 \n\t"
118 "psubw %%mm4, %%mm0 \n\t"
119 "psubw %%mm3, %%mm2 \n\t"
120 "psubw %%mm3, %%mm0 \n\t"
121 "paddw %%mm1, %%mm2 \n\t"
122
123 "movq %%mm5, %%mm4 \n\t"
124 "movq %%mm1, %%mm6 \n\t"
125 "psraw $1, %%mm4 \n\t"
126 "psraw $1, %%mm6 \n\t"
127 "paddw %%mm5, %%mm4 \n\t"
128 "paddw %%mm1, %%mm6 \n\t"
129 "paddw %%mm7, %%mm4 \n\t"
130 "paddw %%mm5, %%mm6 \n\t"
131 "psubw %%mm1, %%mm4 \n\t"
132 "paddw %%mm3, %%mm6 \n\t"
133
134 "movq %%mm0, %%mm1 \n\t"
135 "movq %%mm4, %%mm3 \n\t"
136 "movq %%mm2, %%mm5 \n\t"
137 "movq %%mm6, %%mm7 \n\t"
138 "psraw $2, %%mm6 \n\t"
139 "psraw $2, %%mm3 \n\t"
140 "psraw $2, %%mm5 \n\t"
141 "psraw $2, %%mm0 \n\t"
142 "paddw %%mm6, %%mm1 \n\t"
143 "paddw %%mm2, %%mm3 \n\t"
144 "psubw %%mm4, %%mm5 \n\t"
145 "psubw %%mm0, %%mm7 \n\t"
146
147 "movq 32(%0), %%mm2 \n\t"
148 "movq 96(%0), %%mm6 \n\t"
149 "movq %%mm2, %%mm4 \n\t"
150 "movq %%mm6, %%mm0 \n\t"
151 "psraw $1, %%mm4 \n\t"
152 "psraw $1, %%mm6 \n\t"
153 "psubw %%mm0, %%mm4 \n\t"
154 "paddw %%mm2, %%mm6 \n\t"
155
156 "movq (%0), %%mm2 \n\t"
157 "movq 64(%0), %%mm0 \n\t"
158 SUMSUB_BA( %%mm0, %%mm2 )
159 SUMSUB_BA( %%mm6, %%mm0 )
160 SUMSUB_BA( %%mm4, %%mm2 )
161 SUMSUB_BA( %%mm7, %%mm6 )
162 SUMSUB_BA( %%mm5, %%mm4 )
163 SUMSUB_BA( %%mm3, %%mm2 )
164 SUMSUB_BA( %%mm1, %%mm0 )
165 :: "r"(block)
166 );
167}
168
169static void ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
170{
171 int i;
172 int16_t __attribute__ ((aligned(8))) b2[64];
173
174 block[0] += 32;
175
176 for(i=0; i<2; i++){
177 uint64_t tmp;
178
179 h264_idct8_1d(block+4*i);
180
181 asm volatile(
182 "movq %%mm7, %0 \n\t"
183 TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
184 "movq %%mm0, 8(%1) \n\t"
185 "movq %%mm6, 24(%1) \n\t"
186 "movq %%mm7, 40(%1) \n\t"
187 "movq %%mm4, 56(%1) \n\t"
188 "movq %0, %%mm7 \n\t"
189 TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
190 "movq %%mm7, (%1) \n\t"
191 "movq %%mm1, 16(%1) \n\t"
192 "movq %%mm0, 32(%1) \n\t"
193 "movq %%mm3, 48(%1) \n\t"
194 : "=m"(tmp)
195 : "r"(b2+32*i)
196 : "memory"
197 );
198 }
199
200 for(i=0; i<2; i++){
201 h264_idct8_1d(b2+4*i);
202
203 asm volatile(
204 "psraw $6, %%mm7 \n\t"
205 "psraw $6, %%mm6 \n\t"
206 "psraw $6, %%mm5 \n\t"
207 "psraw $6, %%mm4 \n\t"
208 "psraw $6, %%mm3 \n\t"
209 "psraw $6, %%mm2 \n\t"
210 "psraw $6, %%mm1 \n\t"
211 "psraw $6, %%mm0 \n\t"
212
213 "movq %%mm7, (%0) \n\t"
214 "movq %%mm5, 16(%0) \n\t"
215 "movq %%mm3, 32(%0) \n\t"
216 "movq %%mm1, 48(%0) \n\t"
217 "movq %%mm0, 64(%0) \n\t"
218 "movq %%mm2, 80(%0) \n\t"
219 "movq %%mm4, 96(%0) \n\t"
220 "movq %%mm6, 112(%0) \n\t"
221 :: "r"(b2+4*i)
222 : "memory"
223 );
224 }
225
226 add_pixels_clamped_mmx(b2, dst, stride);
227}
228
229static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
230{
231 int dc = (block[0] + 32) >> 6;
232 asm volatile(
233 "movd %0, %%mm0 \n\t"
234 "pshufw $0, %%mm0, %%mm0 \n\t"
235 "pxor %%mm1, %%mm1 \n\t"
236 "psubw %%mm0, %%mm1 \n\t"
237 "packuswb %%mm0, %%mm0 \n\t"
238 "packuswb %%mm1, %%mm1 \n\t"
239 ::"r"(dc)
240 );
241 asm volatile(
242 "movd %0, %%mm2 \n\t"
243 "movd %1, %%mm3 \n\t"
244 "movd %2, %%mm4 \n\t"
245 "movd %3, %%mm5 \n\t"
246 "paddusb %%mm0, %%mm2 \n\t"
247 "paddusb %%mm0, %%mm3 \n\t"
248 "paddusb %%mm0, %%mm4 \n\t"
249 "paddusb %%mm0, %%mm5 \n\t"
250 "psubusb %%mm1, %%mm2 \n\t"
251 "psubusb %%mm1, %%mm3 \n\t"
252 "psubusb %%mm1, %%mm4 \n\t"
253 "psubusb %%mm1, %%mm5 \n\t"
254 "movd %%mm2, %0 \n\t"
255 "movd %%mm3, %1 \n\t"
256 "movd %%mm4, %2 \n\t"
257 "movd %%mm5, %3 \n\t"
258 :"+m"(*(uint32_t*)(dst+0*stride)),
259 "+m"(*(uint32_t*)(dst+1*stride)),
260 "+m"(*(uint32_t*)(dst+2*stride)),
261 "+m"(*(uint32_t*)(dst+3*stride))
262 );
263}
264
265static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
266{
267 int dc = (block[0] + 32) >> 6;
268 int y;
269 asm volatile(
270 "movd %0, %%mm0 \n\t"
271 "pshufw $0, %%mm0, %%mm0 \n\t"
272 "pxor %%mm1, %%mm1 \n\t"
273 "psubw %%mm0, %%mm1 \n\t"
274 "packuswb %%mm0, %%mm0 \n\t"
275 "packuswb %%mm1, %%mm1 \n\t"
276 ::"r"(dc)
277 );
278 for(y=2; y--; dst += 4*stride){
279 asm volatile(
280 "movq %0, %%mm2 \n\t"
281 "movq %1, %%mm3 \n\t"
282 "movq %2, %%mm4 \n\t"
283 "movq %3, %%mm5 \n\t"
284 "paddusb %%mm0, %%mm2 \n\t"
285 "paddusb %%mm0, %%mm3 \n\t"
286 "paddusb %%mm0, %%mm4 \n\t"
287 "paddusb %%mm0, %%mm5 \n\t"
288 "psubusb %%mm1, %%mm2 \n\t"
289 "psubusb %%mm1, %%mm3 \n\t"
290 "psubusb %%mm1, %%mm4 \n\t"
291 "psubusb %%mm1, %%mm5 \n\t"
292 "movq %%mm2, %0 \n\t"
293 "movq %%mm3, %1 \n\t"
294 "movq %%mm4, %2 \n\t"
295 "movq %%mm5, %3 \n\t"
296 :"+m"(*(uint64_t*)(dst+0*stride)),
297 "+m"(*(uint64_t*)(dst+1*stride)),
298 "+m"(*(uint64_t*)(dst+2*stride)),
299 "+m"(*(uint64_t*)(dst+3*stride))
300 );
301 }
302}
303
304
305/***********************************/
306/* deblocking */
307
308// out: o = |x-y|>a
309// clobbers: t
310#define DIFF_GT_MMX(x,y,a,o,t)\
311 "movq "#y", "#t" \n\t"\
312 "movq "#x", "#o" \n\t"\
313 "psubusb "#x", "#t" \n\t"\
314 "psubusb "#y", "#o" \n\t"\
315 "por "#t", "#o" \n\t"\
316 "psubusb "#a", "#o" \n\t"
317
318// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
319// out: mm5=beta-1, mm7=mask
320// clobbers: mm4,mm6
321#define H264_DEBLOCK_MASK(alpha1, beta1) \
322 "pshufw $0, "#alpha1", %%mm4 \n\t"\
323 "pshufw $0, "#beta1 ", %%mm5 \n\t"\
324 "packuswb %%mm4, %%mm4 \n\t"\
325 "packuswb %%mm5, %%mm5 \n\t"\
326 DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
327 DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
328 "por %%mm4, %%mm7 \n\t"\
329 DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
330 "por %%mm4, %%mm7 \n\t"\
331 "pxor %%mm6, %%mm6 \n\t"\
332 "pcmpeqb %%mm6, %%mm7 \n\t"
333
334// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
335// out: mm1=p0' mm2=q0'
336// clobbers: mm0,3-6
337#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
338 /* a = q0^p0^((p1-q1)>>2) */\
339 "movq %%mm0, %%mm4 \n\t"\
340 "psubb %%mm3, %%mm4 \n\t"\
341 "psrlw $2, %%mm4 \n\t"\
342 "pxor %%mm1, %%mm4 \n\t"\
343 "pxor %%mm2, %%mm4 \n\t"\
344 /* b = p0^(q1>>2) */\
345 "psrlw $2, %%mm3 \n\t"\
346 "pand "#pb_3f", %%mm3 \n\t"\
347 "movq %%mm1, %%mm5 \n\t"\
348 "pxor %%mm3, %%mm5 \n\t"\
349 /* c = q0^(p1>>2) */\
350 "psrlw $2, %%mm0 \n\t"\
351 "pand "#pb_3f", %%mm0 \n\t"\
352 "movq %%mm2, %%mm6 \n\t"\
353 "pxor %%mm0, %%mm6 \n\t"\
354 /* d = (c^b) & ~(b^a) & 1 */\
355 "pxor %%mm5, %%mm6 \n\t"\
356 "pxor %%mm4, %%mm5 \n\t"\
357 "pandn %%mm6, %%mm5 \n\t"\
358 "pand "#pb_01", %%mm5 \n\t"\
359 /* delta = (avg(q0, p1>>2) + (d&a))
360 * - (avg(p0, q1>>2) + (d&~a)) */\
361 "pavgb %%mm2, %%mm0 \n\t"\
362 "pand %%mm5, %%mm4 \n\t"\
363 "paddusb %%mm4, %%mm0 \n\t"\
364 "pavgb %%mm1, %%mm3 \n\t"\
365 "pxor %%mm5, %%mm4 \n\t"\
366 "paddusb %%mm4, %%mm3 \n\t"\
367 /* p0 += clip(delta, -tc0, tc0)
368 * q0 -= clip(delta, -tc0, tc0) */\
369 "movq %%mm0, %%mm4 \n\t"\
370 "psubusb %%mm3, %%mm0 \n\t"\
371 "psubusb %%mm4, %%mm3 \n\t"\
372 "pminub %%mm7, %%mm0 \n\t"\
373 "pminub %%mm7, %%mm3 \n\t"\
374 "paddusb %%mm0, %%mm1 \n\t"\
375 "paddusb %%mm3, %%mm2 \n\t"\
376 "psubusb %%mm3, %%mm1 \n\t"\
377 "psubusb %%mm0, %%mm2 \n\t"
378
379// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
380// out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
381// clobbers: q2, tmp, tc0
382#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
383 "movq %%mm1, "#tmp" \n\t"\
384 "pavgb %%mm2, "#tmp" \n\t"\
385 "pavgb "#tmp", "#q2" \n\t" /* avg(p2,avg(p0,q0)) */\
386 "pxor "q2addr", "#tmp" \n\t"\
387 "pand %8, "#tmp" \n\t" /* (p2^avg(p0,q0))&1 */\
388 "psubusb "#tmp", "#q2" \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
389 "movq "#p1", "#tmp" \n\t"\
390 "psubusb "#tc0", "#tmp" \n\t"\
391 "paddusb "#p1", "#tc0" \n\t"\
392 "pmaxub "#tmp", "#q2" \n\t"\
393 "pminub "#tc0", "#q2" \n\t"\
394 "movq "#q2", "q1addr" \n\t"
395
396static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
397{
398 uint64_t tmp0;
399 uint64_t tc = (uint8_t)tc0[1]*0x01010000 | (uint8_t)tc0[0]*0x0101;
400 // with luma, tc0=0 doesn't mean no filtering, so we need a separate input mask
401 uint32_t mask[2] = { (tc0[0]>=0)*0xffffffff, (tc0[1]>=0)*0xffffffff };
402
403 asm volatile(
404 "movq (%1,%3), %%mm0 \n\t" //p1
405 "movq (%1,%3,2), %%mm1 \n\t" //p0
406 "movq (%2), %%mm2 \n\t" //q0
407 "movq (%2,%3), %%mm3 \n\t" //q1
408 H264_DEBLOCK_MASK(%6, %7)
409 "pand %5, %%mm7 \n\t"
410 "movq %%mm7, %0 \n\t"
411
412 /* filter p1 */
413 "movq (%1), %%mm3 \n\t" //p2
414 DIFF_GT_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
415 "pandn %%mm7, %%mm6 \n\t"
416 "pcmpeqb %%mm7, %%mm6 \n\t"
417 "pand %%mm7, %%mm6 \n\t" // mask & |p2-p0|<beta
418 "pshufw $80, %4, %%mm4 \n\t"
419 "pand %%mm7, %%mm4 \n\t" // mask & tc0
420 "movq %8, %%mm7 \n\t"
421 "pand %%mm6, %%mm7 \n\t" // mask & |p2-p0|<beta & 1
422 "pand %%mm4, %%mm6 \n\t" // mask & |p2-p0|<beta & tc0
423 "paddb %%mm4, %%mm7 \n\t" // tc++
424 H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
425
426 /* filter q1 */
427 "movq (%2,%3,2), %%mm4 \n\t" //q2
428 DIFF_GT_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
429 "pandn %0, %%mm6 \n\t"
430 "pcmpeqb %0, %%mm6 \n\t"
431 "pand %0, %%mm6 \n\t"
432 "pshufw $80, %4, %%mm5 \n\t"
433 "pand %%mm6, %%mm5 \n\t"
434 "pand %8, %%mm6 \n\t"
435 "paddb %%mm6, %%mm7 \n\t"
436 "movq (%2,%3), %%mm3 \n\t"
437 H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
438
439 /* filter p0, q0 */
440 H264_DEBLOCK_P0_Q0(%8, %9)
441 "movq %%mm1, (%1,%3,2) \n\t"
442 "movq %%mm2, (%2) \n\t"
443
444 : "=m"(tmp0)
445 : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
446 "m"(tc), "m"(*(uint64_t*)mask), "m"(alpha1), "m"(beta1),
447 "m"(mm_bone), "m"(ff_pb_3F)
448 );
449}
450
451static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
452{
453 if((tc0[0] & tc0[1]) >= 0)
454 h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
455 if((tc0[2] & tc0[3]) >= 0)
456 h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
457}
458static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
459{
460 //FIXME: could cut some load/stores by merging transpose with filter
461 // also, it only needs to transpose 6x8
462 uint8_t trans[8*8];
463 int i;
464 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
465 if((tc0[0] & tc0[1]) < 0)
466 continue;
467 transpose4x4(trans, pix-4, 8, stride);
468 transpose4x4(trans +4*8, pix, 8, stride);
469 transpose4x4(trans+4, pix-4+4*stride, 8, stride);
470 transpose4x4(trans+4+4*8, pix +4*stride, 8, stride);
471 h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
472 transpose4x4(pix-2, trans +2*8, stride, 8);
473 transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
474 }
475}
476
477static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
478{
479 asm volatile(
480 "movq (%0), %%mm0 \n\t" //p1
481 "movq (%0,%2), %%mm1 \n\t" //p0
482 "movq (%1), %%mm2 \n\t" //q0
483 "movq (%1,%2), %%mm3 \n\t" //q1
484 H264_DEBLOCK_MASK(%4, %5)
485 "movd %3, %%mm6 \n\t"
486 "punpcklbw %%mm6, %%mm6 \n\t"
487 "pand %%mm6, %%mm7 \n\t" // mm7 = tc&mask
488 H264_DEBLOCK_P0_Q0(%6, %7)
489 "movq %%mm1, (%0,%2) \n\t"
490 "movq %%mm2, (%1) \n\t"
491
492 :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
493 "r"(*(uint32_t*)tc0),
494 "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
495 );
496}
497
498static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
499{
500 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
501}
502
503static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
504{
505 //FIXME: could cut some load/stores by merging transpose with filter
506 uint8_t trans[8*4];
507 transpose4x4(trans, pix-2, 8, stride);
508 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
509 h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
510 transpose4x4(pix-2, trans, stride, 8);
511 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
512}
513
514// p0 = (p0 + q1 + 2*p1 + 2) >> 2
515#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
516 "movq "#p0", %%mm4 \n\t"\
517 "pxor "#q1", %%mm4 \n\t"\
518 "pand "#one", %%mm4 \n\t" /* mm4 = (p0^q1)&1 */\
519 "pavgb "#q1", "#p0" \n\t"\
520 "psubusb %%mm4, "#p0" \n\t"\
521 "pavgb "#p1", "#p0" \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
522
523static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
524{
525 asm volatile(
526 "movq (%0), %%mm0 \n\t"
527 "movq (%0,%2), %%mm1 \n\t"
528 "movq (%1), %%mm2 \n\t"
529 "movq (%1,%2), %%mm3 \n\t"
530 H264_DEBLOCK_MASK(%3, %4)
531 "movq %%mm1, %%mm5 \n\t"
532 "movq %%mm2, %%mm6 \n\t"
533 H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
534 H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
535 "psubb %%mm5, %%mm1 \n\t"
536 "psubb %%mm6, %%mm2 \n\t"
537 "pand %%mm7, %%mm1 \n\t"
538 "pand %%mm7, %%mm2 \n\t"
539 "paddb %%mm5, %%mm1 \n\t"
540 "paddb %%mm6, %%mm2 \n\t"
541 "movq %%mm1, (%0,%2) \n\t"
542 "movq %%mm2, (%1) \n\t"
543 :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
544 "m"(alpha1), "m"(beta1), "m"(mm_bone)
545 );
546}
547
548static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
549{
550 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
551}
552
553static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
554{
555 //FIXME: could cut some load/stores by merging transpose with filter
556 uint8_t trans[8*4];
557 transpose4x4(trans, pix-2, 8, stride);
558 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
559 h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
560 transpose4x4(pix-2, trans, stride, 8);
561 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
562}
563
564
565/***********************************/
566/* motion compensation */
567
568#define QPEL_H264V(A,B,C,D,E,F,OP)\
569 "movd (%0), "#F" \n\t"\
570 "movq "#C", %%mm6 \n\t"\
571 "paddw "#D", %%mm6 \n\t"\
572 "psllw $2, %%mm6 \n\t"\
573 "psubw "#B", %%mm6 \n\t"\
574 "psubw "#E", %%mm6 \n\t"\
575 "pmullw %4, %%mm6 \n\t"\
576 "add %2, %0 \n\t"\
577 "punpcklbw %%mm7, "#F" \n\t"\
578 "paddw %5, "#A" \n\t"\
579 "paddw "#F", "#A" \n\t"\
580 "paddw "#A", %%mm6 \n\t"\
581 "psraw $5, %%mm6 \n\t"\
582 "packuswb %%mm6, %%mm6 \n\t"\
583 OP(%%mm6, (%1), A, d)\
584 "add %3, %1 \n\t"
585
586#define QPEL_H264HV(A,B,C,D,E,F,OF)\
587 "movd (%0), "#F" \n\t"\
588 "movq "#C", %%mm6 \n\t"\
589 "paddw "#D", %%mm6 \n\t"\
590 "psllw $2, %%mm6 \n\t"\
591 "psubw "#B", %%mm6 \n\t"\
592 "psubw "#E", %%mm6 \n\t"\
593 "pmullw %3, %%mm6 \n\t"\
594 "add %2, %0 \n\t"\
595 "punpcklbw %%mm7, "#F" \n\t"\
596 "paddw "#F", "#A" \n\t"\
597 "paddw "#A", %%mm6 \n\t"\
598 "movq %%mm6, "#OF"(%1) \n\t"
599
600#define QPEL_H264(OPNAME, OP, MMX)\
601static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
602 int h=4;\
603\
604 asm volatile(\
605 "pxor %%mm7, %%mm7 \n\t"\
606 "movq %5, %%mm4 \n\t"\
607 "movq %6, %%mm5 \n\t"\
608 "1: \n\t"\
609 "movd -1(%0), %%mm1 \n\t"\
610 "movd (%0), %%mm2 \n\t"\
611 "movd 1(%0), %%mm3 \n\t"\
612 "movd 2(%0), %%mm0 \n\t"\
613 "punpcklbw %%mm7, %%mm1 \n\t"\
614 "punpcklbw %%mm7, %%mm2 \n\t"\
615 "punpcklbw %%mm7, %%mm3 \n\t"\
616 "punpcklbw %%mm7, %%mm0 \n\t"\
617 "paddw %%mm0, %%mm1 \n\t"\
618 "paddw %%mm3, %%mm2 \n\t"\
619 "movd -2(%0), %%mm0 \n\t"\
620 "movd 3(%0), %%mm3 \n\t"\
621 "punpcklbw %%mm7, %%mm0 \n\t"\
622 "punpcklbw %%mm7, %%mm3 \n\t"\
623 "paddw %%mm3, %%mm0 \n\t"\
624 "psllw $2, %%mm2 \n\t"\
625 "psubw %%mm1, %%mm2 \n\t"\
626 "pmullw %%mm4, %%mm2 \n\t"\
627 "paddw %%mm5, %%mm0 \n\t"\
628 "paddw %%mm2, %%mm0 \n\t"\
629 "psraw $5, %%mm0 \n\t"\
630 "packuswb %%mm0, %%mm0 \n\t"\
631 OP(%%mm0, (%1),%%mm6, d)\
632 "add %3, %0 \n\t"\
633 "add %4, %1 \n\t"\
634 "decl %2 \n\t"\
635 " jnz 1b \n\t"\
636 : "+a"(src), "+c"(dst), "+m"(h)\
637 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
638 : "memory"\
639 );\
640}\
641static void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
642 int h=4;\
643 asm volatile(\
644 "pxor %%mm7, %%mm7 \n\t"\
645 "movq %0, %%mm4 \n\t"\
646 "movq %1, %%mm5 \n\t"\
647 :: "m"(ff_pw_5), "m"(ff_pw_16)\
648 );\
649 do{\
650 asm volatile(\
651 "movd -1(%0), %%mm1 \n\t"\
652 "movd (%0), %%mm2 \n\t"\
653 "movd 1(%0), %%mm3 \n\t"\
654 "movd 2(%0), %%mm0 \n\t"\
655 "punpcklbw %%mm7, %%mm1 \n\t"\
656 "punpcklbw %%mm7, %%mm2 \n\t"\
657 "punpcklbw %%mm7, %%mm3 \n\t"\
658 "punpcklbw %%mm7, %%mm0 \n\t"\
659 "paddw %%mm0, %%mm1 \n\t"\
660 "paddw %%mm3, %%mm2 \n\t"\
661 "movd -2(%0), %%mm0 \n\t"\
662 "movd 3(%0), %%mm3 \n\t"\
663 "punpcklbw %%mm7, %%mm0 \n\t"\
664 "punpcklbw %%mm7, %%mm3 \n\t"\
665 "paddw %%mm3, %%mm0 \n\t"\
666 "psllw $2, %%mm2 \n\t"\
667 "psubw %%mm1, %%mm2 \n\t"\
668 "pmullw %%mm4, %%mm2 \n\t"\
669 "paddw %%mm5, %%mm0 \n\t"\
670 "paddw %%mm2, %%mm0 \n\t"\
671 "movd (%2), %%mm3 \n\t"\
672 "psraw $5, %%mm0 \n\t"\
673 "packuswb %%mm0, %%mm0 \n\t"\
674 PAVGB" %%mm3, %%mm0 \n\t"\
675 OP(%%mm0, (%1),%%mm6, d)\
676 "add %4, %0 \n\t"\
677 "add %4, %1 \n\t"\
678 "add %3, %2 \n\t"\
679 : "+a"(src), "+c"(dst), "+d"(src2)\
680 : "D"((long)src2Stride), "S"((long)dstStride)\
681 : "memory"\
682 );\
683 }while(--h);\
684}\
685static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
686 src -= 2*srcStride;\
687 asm volatile(\
688 "pxor %%mm7, %%mm7 \n\t"\
689 "movd (%0), %%mm0 \n\t"\
690 "add %2, %0 \n\t"\
691 "movd (%0), %%mm1 \n\t"\
692 "add %2, %0 \n\t"\
693 "movd (%0), %%mm2 \n\t"\
694 "add %2, %0 \n\t"\
695 "movd (%0), %%mm3 \n\t"\
696 "add %2, %0 \n\t"\
697 "movd (%0), %%mm4 \n\t"\
698 "add %2, %0 \n\t"\
699 "punpcklbw %%mm7, %%mm0 \n\t"\
700 "punpcklbw %%mm7, %%mm1 \n\t"\
701 "punpcklbw %%mm7, %%mm2 \n\t"\
702 "punpcklbw %%mm7, %%mm3 \n\t"\
703 "punpcklbw %%mm7, %%mm4 \n\t"\
704 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
705 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
706 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
707 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
708 \
709 : "+a"(src), "+c"(dst)\
710 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
711 : "memory"\
712 );\
713}\
714static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
715 int h=4;\
716 int w=3;\
717 src -= 2*srcStride+2;\
718 while(w--){\
719 asm volatile(\
720 "pxor %%mm7, %%mm7 \n\t"\
721 "movd (%0), %%mm0 \n\t"\
722 "add %2, %0 \n\t"\
723 "movd (%0), %%mm1 \n\t"\
724 "add %2, %0 \n\t"\
725 "movd (%0), %%mm2 \n\t"\
726 "add %2, %0 \n\t"\
727 "movd (%0), %%mm3 \n\t"\
728 "add %2, %0 \n\t"\
729 "movd (%0), %%mm4 \n\t"\
730 "add %2, %0 \n\t"\
731 "punpcklbw %%mm7, %%mm0 \n\t"\
732 "punpcklbw %%mm7, %%mm1 \n\t"\
733 "punpcklbw %%mm7, %%mm2 \n\t"\
734 "punpcklbw %%mm7, %%mm3 \n\t"\
735 "punpcklbw %%mm7, %%mm4 \n\t"\
736 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
737 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
738 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
739 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
740 \
741 : "+a"(src)\
742 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
743 : "memory"\
744 );\
745 tmp += 4;\
746 src += 4 - 9*srcStride;\
747 }\
748 tmp -= 3*4;\
749 asm volatile(\
750 "movq %4, %%mm6 \n\t"\
751 "1: \n\t"\
752 "movq (%0), %%mm0 \n\t"\
753 "paddw 10(%0), %%mm0 \n\t"\
754 "movq 2(%0), %%mm1 \n\t"\
755 "paddw 8(%0), %%mm1 \n\t"\
756 "movq 4(%0), %%mm2 \n\t"\
757 "paddw 6(%0), %%mm2 \n\t"\
758 "psubw %%mm1, %%mm0 \n\t"/*a-b (abccba)*/\
759 "psraw $2, %%mm0 \n\t"/*(a-b)/4 */\
760 "psubw %%mm1, %%mm0 \n\t"/*(a-b)/4-b */\
761 "paddsw %%mm2, %%mm0 \n\t"\
762 "psraw $2, %%mm0 \n\t"/*((a-b)/4-b+c)/4 */\
763 "paddw %%mm6, %%mm2 \n\t"\
764 "paddw %%mm2, %%mm0 \n\t"/*(a-5*b+20*c)/16 +32 */\
765 "psraw $6, %%mm0 \n\t"\
766 "packuswb %%mm0, %%mm0 \n\t"\
767 OP(%%mm0, (%1),%%mm7, d)\
768 "add $24, %0 \n\t"\
769 "add %3, %1 \n\t"\
770 "decl %2 \n\t"\
771 " jnz 1b \n\t"\
772 : "+a"(tmp), "+c"(dst), "+m"(h)\
773 : "S"((long)dstStride), "m"(ff_pw_32)\
774 : "memory"\
775 );\
776}\
777\
778static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
779 int h=8;\
780 asm volatile(\
781 "pxor %%mm7, %%mm7 \n\t"\
782 "movq %5, %%mm6 \n\t"\
783 "1: \n\t"\
784 "movq (%0), %%mm0 \n\t"\
785 "movq 1(%0), %%mm2 \n\t"\
786 "movq %%mm0, %%mm1 \n\t"\
787 "movq %%mm2, %%mm3 \n\t"\
788 "punpcklbw %%mm7, %%mm0 \n\t"\
789 "punpckhbw %%mm7, %%mm1 \n\t"\
790 "punpcklbw %%mm7, %%mm2 \n\t"\
791 "punpckhbw %%mm7, %%mm3 \n\t"\
792 "paddw %%mm2, %%mm0 \n\t"\
793 "paddw %%mm3, %%mm1 \n\t"\
794 "psllw $2, %%mm0 \n\t"\
795 "psllw $2, %%mm1 \n\t"\
796 "movq -1(%0), %%mm2 \n\t"\
797 "movq 2(%0), %%mm4 \n\t"\
798 "movq %%mm2, %%mm3 \n\t"\
799 "movq %%mm4, %%mm5 \n\t"\
800 "punpcklbw %%mm7, %%mm2 \n\t"\
801 "punpckhbw %%mm7, %%mm3 \n\t"\
802 "punpcklbw %%mm7, %%mm4 \n\t"\
803 "punpckhbw %%mm7, %%mm5 \n\t"\
804 "paddw %%mm4, %%mm2 \n\t"\
805 "paddw %%mm3, %%mm5 \n\t"\
806 "psubw %%mm2, %%mm0 \n\t"\
807 "psubw %%mm5, %%mm1 \n\t"\
808 "pmullw %%mm6, %%mm0 \n\t"\
809 "pmullw %%mm6, %%mm1 \n\t"\
810 "movd -2(%0), %%mm2 \n\t"\
811 "movd 7(%0), %%mm5 \n\t"\
812 "punpcklbw %%mm7, %%mm2 \n\t"\
813 "punpcklbw %%mm7, %%mm5 \n\t"\
814 "paddw %%mm3, %%mm2 \n\t"\
815 "paddw %%mm5, %%mm4 \n\t"\
816 "movq %6, %%mm5 \n\t"\
817 "paddw %%mm5, %%mm2 \n\t"\
818 "paddw %%mm5, %%mm4 \n\t"\
819 "paddw %%mm2, %%mm0 \n\t"\
820 "paddw %%mm4, %%mm1 \n\t"\
821 "psraw $5, %%mm0 \n\t"\
822 "psraw $5, %%mm1 \n\t"\
823 "packuswb %%mm1, %%mm0 \n\t"\
824 OP(%%mm0, (%1),%%mm5, q)\
825 "add %3, %0 \n\t"\
826 "add %4, %1 \n\t"\
827 "decl %2 \n\t"\
828 " jnz 1b \n\t"\
829 : "+a"(src), "+c"(dst), "+m"(h)\
830 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
831 : "memory"\
832 );\
833}\
834\
835static void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
836 int h=8;\
837 asm volatile(\
838 "pxor %%mm7, %%mm7 \n\t"\
839 "movq %0, %%mm6 \n\t"\
840 :: "m"(ff_pw_5)\
841 );\
842 do{\
843 asm volatile(\
844 "movq (%0), %%mm0 \n\t"\
845 "movq 1(%0), %%mm2 \n\t"\
846 "movq %%mm0, %%mm1 \n\t"\
847 "movq %%mm2, %%mm3 \n\t"\
848 "punpcklbw %%mm7, %%mm0 \n\t"\
849 "punpckhbw %%mm7, %%mm1 \n\t"\
850 "punpcklbw %%mm7, %%mm2 \n\t"\
851 "punpckhbw %%mm7, %%mm3 \n\t"\
852 "paddw %%mm2, %%mm0 \n\t"\
853 "paddw %%mm3, %%mm1 \n\t"\
854 "psllw $2, %%mm0 \n\t"\
855 "psllw $2, %%mm1 \n\t"\
856 "movq -1(%0), %%mm2 \n\t"\
857 "movq 2(%0), %%mm4 \n\t"\
858 "movq %%mm2, %%mm3 \n\t"\
859 "movq %%mm4, %%mm5 \n\t"\
860 "punpcklbw %%mm7, %%mm2 \n\t"\
861 "punpckhbw %%mm7, %%mm3 \n\t"\
862 "punpcklbw %%mm7, %%mm4 \n\t"\
863 "punpckhbw %%mm7, %%mm5 \n\t"\
864 "paddw %%mm4, %%mm2 \n\t"\
865 "paddw %%mm3, %%mm5 \n\t"\
866 "psubw %%mm2, %%mm0 \n\t"\
867 "psubw %%mm5, %%mm1 \n\t"\
868 "pmullw %%mm6, %%mm0 \n\t"\
869 "pmullw %%mm6, %%mm1 \n\t"\
870 "movd -2(%0), %%mm2 \n\t"\
871 "movd 7(%0), %%mm5 \n\t"\
872 "punpcklbw %%mm7, %%mm2 \n\t"\
873 "punpcklbw %%mm7, %%mm5 \n\t"\
874 "paddw %%mm3, %%mm2 \n\t"\
875 "paddw %%mm5, %%mm4 \n\t"\
876 "movq %5, %%mm5 \n\t"\
877 "paddw %%mm5, %%mm2 \n\t"\
878 "paddw %%mm5, %%mm4 \n\t"\
879 "paddw %%mm2, %%mm0 \n\t"\
880 "paddw %%mm4, %%mm1 \n\t"\
881 "psraw $5, %%mm0 \n\t"\
882 "psraw $5, %%mm1 \n\t"\
883 "movq (%2), %%mm4 \n\t"\
884 "packuswb %%mm1, %%mm0 \n\t"\
885 PAVGB" %%mm4, %%mm0 \n\t"\
886 OP(%%mm0, (%1),%%mm5, q)\
887 "add %4, %0 \n\t"\
888 "add %4, %1 \n\t"\
889 "add %3, %2 \n\t"\
890 : "+a"(src), "+c"(dst), "+d"(src2)\
891 : "D"((long)src2Stride), "S"((long)dstStride),\
892 "m"(ff_pw_16)\
893 : "memory"\
894 );\
895 }while(--h);\
896}\
897\
898static inline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
899 int w= 2;\
900 src -= 2*srcStride;\
901 \
902 while(w--){\
903 asm volatile(\
904 "pxor %%mm7, %%mm7 \n\t"\
905 "movd (%0), %%mm0 \n\t"\
906 "add %2, %0 \n\t"\
907 "movd (%0), %%mm1 \n\t"\
908 "add %2, %0 \n\t"\
909 "movd (%0), %%mm2 \n\t"\
910 "add %2, %0 \n\t"\
911 "movd (%0), %%mm3 \n\t"\
912 "add %2, %0 \n\t"\
913 "movd (%0), %%mm4 \n\t"\
914 "add %2, %0 \n\t"\
915 "punpcklbw %%mm7, %%mm0 \n\t"\
916 "punpcklbw %%mm7, %%mm1 \n\t"\
917 "punpcklbw %%mm7, %%mm2 \n\t"\
918 "punpcklbw %%mm7, %%mm3 \n\t"\
919 "punpcklbw %%mm7, %%mm4 \n\t"\
920 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
921 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
922 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
923 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
924 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
925 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
926 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
927 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
928 \
929 : "+a"(src), "+c"(dst)\
930 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
931 : "memory"\
932 );\
933 if(h==16){\
934 asm volatile(\
935 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
936 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
937 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
938 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
939 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
940 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
941 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
942 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
943 \
944 : "+a"(src), "+c"(dst)\
945 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
946 : "memory"\
947 );\
948 }\
949 src += 4-(h+5)*srcStride;\
950 dst += 4-h*dstStride;\
951 }\
952}\
953static inline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
954 int h = size;\
955 int w = (size+8)>>2;\
956 src -= 2*srcStride+2;\
957 while(w--){\
958 asm volatile(\
959 "pxor %%mm7, %%mm7 \n\t"\
960 "movd (%0), %%mm0 \n\t"\
961 "add %2, %0 \n\t"\
962 "movd (%0), %%mm1 \n\t"\
963 "add %2, %0 \n\t"\
964 "movd (%0), %%mm2 \n\t"\
965 "add %2, %0 \n\t"\
966 "movd (%0), %%mm3 \n\t"\
967 "add %2, %0 \n\t"\
968 "movd (%0), %%mm4 \n\t"\
969 "add %2, %0 \n\t"\
970 "punpcklbw %%mm7, %%mm0 \n\t"\
971 "punpcklbw %%mm7, %%mm1 \n\t"\
972 "punpcklbw %%mm7, %%mm2 \n\t"\
973 "punpcklbw %%mm7, %%mm3 \n\t"\
974 "punpcklbw %%mm7, %%mm4 \n\t"\
975 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
976 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
977 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
978 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
979 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
980 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
981 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
982 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
983 : "+a"(src)\
984 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
985 : "memory"\
986 );\
987 if(size==16){\
988 asm volatile(\
989 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 8*48)\
990 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 9*48)\
991 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
992 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
993 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
994 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
995 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
996 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
997 : "+a"(src)\
998 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
999 : "memory"\
1000 );\
1001 }\
1002 tmp += 4;\
1003 src += 4 - (size+5)*srcStride;\
1004 }\
1005 tmp -= size+8;\
1006 w = size>>4;\
1007 do{\
1008 h = size;\
1009 asm volatile(\
1010 "movq %4, %%mm6 \n\t"\
1011 "1: \n\t"\
1012 "movq (%0), %%mm0 \n\t"\
1013 "movq 8(%0), %%mm3 \n\t"\
1014 "movq 2(%0), %%mm1 \n\t"\
1015 "movq 10(%0), %%mm4 \n\t"\
1016 "paddw %%mm4, %%mm0 \n\t"\
1017 "paddw %%mm3, %%mm1 \n\t"\
1018 "paddw 18(%0), %%mm3 \n\t"\
1019 "paddw 16(%0), %%mm4 \n\t"\
1020 "movq 4(%0), %%mm2 \n\t"\
1021 "movq 12(%0), %%mm5 \n\t"\
1022 "paddw 6(%0), %%mm2 \n\t"\
1023 "paddw 14(%0), %%mm5 \n\t"\
1024 "psubw %%mm1, %%mm0 \n\t"\
1025 "psubw %%mm4, %%mm3 \n\t"\
1026 "psraw $2, %%mm0 \n\t"\
1027 "psraw $2, %%mm3 \n\t"\
1028 "psubw %%mm1, %%mm0 \n\t"\
1029 "psubw %%mm4, %%mm3 \n\t"\
1030 "paddsw %%mm2, %%mm0 \n\t"\
1031 "paddsw %%mm5, %%mm3 \n\t"\
1032 "psraw $2, %%mm0 \n\t"\
1033 "psraw $2, %%mm3 \n\t"\
1034 "paddw %%mm6, %%mm2 \n\t"\
1035 "paddw %%mm6, %%mm5 \n\t"\
1036 "paddw %%mm2, %%mm0 \n\t"\
1037 "paddw %%mm5, %%mm3 \n\t"\
1038 "psraw $6, %%mm0 \n\t"\
1039 "psraw $6, %%mm3 \n\t"\
1040 "packuswb %%mm3, %%mm0 \n\t"\
1041 OP(%%mm0, (%1),%%mm7, q)\
1042 "add $48, %0 \n\t"\
1043 "add %3, %1 \n\t"\
1044 "decl %2 \n\t"\
1045 " jnz 1b \n\t"\
1046 : "+a"(tmp), "+c"(dst), "+m"(h)\
1047 : "S"((long)dstStride), "m"(ff_pw_32)\
1048 : "memory"\
1049 );\
1050 tmp += 8 - size*24;\
1051 dst += 8 - size*dstStride;\
1052 }while(w--);\
1053}\
1054\
1055static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1056 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
1057}\
1058static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1059 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
1060 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1061}\
1062\
1063static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1064 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1065 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1066 src += 8*srcStride;\
1067 dst += 8*dstStride;\
1068 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
1069 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
1070}\
1071\
1072static void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
1073 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1074 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1075 src += 8*dstStride;\
1076 dst += 8*dstStride;\
1077 src2 += 8*src2Stride;\
1078 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
1079 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
1080}\
1081\
1082static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1083 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
1084}\
1085\
1086static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1087 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
1088}\
1089\
1090static void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1091{\
1092 asm volatile(\
1093 "movq %5, %%mm6 \n\t"\
1094 "movq (%1), %%mm0 \n\t"\
1095 "movq 24(%1), %%mm1 \n\t"\
1096 "paddw %%mm6, %%mm0 \n\t"\
1097 "paddw %%mm6, %%mm1 \n\t"\
1098 "psraw $5, %%mm0 \n\t"\
1099 "psraw $5, %%mm1 \n\t"\
1100 "packuswb %%mm0, %%mm0 \n\t"\
1101 "packuswb %%mm1, %%mm1 \n\t"\
1102 PAVGB" (%0), %%mm0 \n\t"\
1103 PAVGB" (%0,%3), %%mm1 \n\t"\
1104 OP(%%mm0, (%2), %%mm4, d)\
1105 OP(%%mm1, (%2,%4), %%mm5, d)\
1106 "lea (%0,%3,2), %0 \n\t"\
1107 "lea (%2,%4,2), %2 \n\t"\
1108 "movq 48(%1), %%mm0 \n\t"\
1109 "movq 72(%1), %%mm1 \n\t"\
1110 "paddw %%mm6, %%mm0 \n\t"\
1111 "paddw %%mm6, %%mm1 \n\t"\
1112 "psraw $5, %%mm0 \n\t"\
1113 "psraw $5, %%mm1 \n\t"\
1114 "packuswb %%mm0, %%mm0 \n\t"\
1115 "packuswb %%mm1, %%mm1 \n\t"\
1116 PAVGB" (%0), %%mm0 \n\t"\
1117 PAVGB" (%0,%3), %%mm1 \n\t"\
1118 OP(%%mm0, (%2), %%mm4, d)\
1119 OP(%%mm1, (%2,%4), %%mm5, d)\
1120 :"+a"(src8), "+c"(src16), "+d"(dst)\
1121 :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\
1122 :"memory");\
1123}\
1124static void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1125{\
1126 asm volatile(\
1127 "movq %0, %%mm6 \n\t"\
1128 ::"m"(ff_pw_16)\
1129 );\
1130 while(h--){\
1131 asm volatile(\
1132 "movq (%1), %%mm0 \n\t"\
1133 "movq 8(%1), %%mm1 \n\t"\
1134 "paddw %%mm6, %%mm0 \n\t"\
1135 "paddw %%mm6, %%mm1 \n\t"\
1136 "psraw $5, %%mm0 \n\t"\
1137 "psraw $5, %%mm1 \n\t"\
1138 "packuswb %%mm1, %%mm0 \n\t"\
1139 PAVGB" (%0), %%mm0 \n\t"\
1140 OP(%%mm0, (%2), %%mm5, q)\
1141 ::"a"(src8), "c"(src16), "d"(dst)\
1142 :"memory");\
1143 src8 += src8Stride;\
1144 src16 += 24;\
1145 dst += dstStride;\
1146 }\
1147}\
1148static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1149{\
1150 OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
1151 OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
1152}\
1153
1154
1155#define H264_MC(OPNAME, SIZE, MMX) \
1156static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1157 OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
1158}\
1159\
1160static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1161 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1162}\
1163\
1164static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1165 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1166}\
1167\
1168static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1169 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1170}\
1171\
1172static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1173 uint64_t temp[SIZE*SIZE/8];\
1174 uint8_t * const half= (uint8_t*)temp;\
1175 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
1176 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
1177}\
1178\
1179static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1180 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1181}\
1182\
1183static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1184 uint64_t temp[SIZE*SIZE/8];\
1185 uint8_t * const half= (uint8_t*)temp;\
1186 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
1187 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
1188}\
1189\
1190static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1191 uint64_t temp[SIZE*SIZE/8];\
1192 uint8_t * const halfV= (uint8_t*)temp;\
1193 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
1194 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
1195}\
1196\
1197static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1198 uint64_t temp[SIZE*SIZE/8];\
1199 uint8_t * const halfV= (uint8_t*)temp;\
1200 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
1201 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
1202}\
1203\
1204static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1205 uint64_t temp[SIZE*SIZE/8];\
1206 uint8_t * const halfV= (uint8_t*)temp;\
1207 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
1208 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
1209}\
1210\
1211static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1212 uint64_t temp[SIZE*SIZE/8];\
1213 uint8_t * const halfV= (uint8_t*)temp;\
1214 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
1215 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
1216}\
1217\
1218static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1219 uint64_t temp[SIZE*(SIZE<8?12:24)/4];\
1220 int16_t * const tmp= (int16_t*)temp;\
1221 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
1222}\
1223\
1224static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1225 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1226 uint8_t * const halfHV= (uint8_t*)temp;\
1227 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
1228 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
1229 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1230}\
1231\
1232static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1233 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1234 uint8_t * const halfHV= (uint8_t*)temp;\
1235 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
1236 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
1237 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1238}\
1239\
1240static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1241 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1242 int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
1243 uint8_t * const halfHV= ((uint8_t*)temp);\
1244 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1245 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1246}\
1247\
1248static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1249 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
1250 int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
1251 uint8_t * const halfHV= ((uint8_t*)temp);\
1252 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1253 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1254}\
1255
1256
1257#define AVG_3DNOW_OP(a,b,temp, size) \
1258"mov" #size " " #b ", " #temp " \n\t"\
1259"pavgusb " #temp ", " #a " \n\t"\
1260"mov" #size " " #a ", " #b " \n\t"
1261#define AVG_MMX2_OP(a,b,temp, size) \
1262"mov" #size " " #b ", " #temp " \n\t"\
1263"pavgb " #temp ", " #a " \n\t"\
1264"mov" #size " " #a ", " #b " \n\t"
1265
1266#define PAVGB "pavgusb"
1267QPEL_H264(put_, PUT_OP, 3dnow)
1268QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1269#undef PAVGB
1270#define PAVGB "pavgb"
1271QPEL_H264(put_, PUT_OP, mmx2)
1272QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
1273#undef PAVGB
1274
1275H264_MC(put_, 4, 3dnow)
1276H264_MC(put_, 8, 3dnow)
1277H264_MC(put_, 16,3dnow)
1278H264_MC(avg_, 4, 3dnow)
1279H264_MC(avg_, 8, 3dnow)
1280H264_MC(avg_, 16,3dnow)
1281H264_MC(put_, 4, mmx2)
1282H264_MC(put_, 8, mmx2)
1283H264_MC(put_, 16,mmx2)
1284H264_MC(avg_, 4, mmx2)
1285H264_MC(avg_, 8, mmx2)
1286H264_MC(avg_, 16,mmx2)
1287
1288
1289#define H264_CHROMA_OP(S,D)
1290#define H264_CHROMA_OP4(S,D,T)
1291#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
1292#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
1293#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
1294#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1295#include "dsputil_h264_template_mmx.c"
1296#undef H264_CHROMA_OP
1297#undef H264_CHROMA_OP4
1298#undef H264_CHROMA_MC8_TMPL
1299#undef H264_CHROMA_MC4_TMPL
1300#undef H264_CHROMA_MC2_TMPL
1301#undef H264_CHROMA_MC8_MV0
1302
1303#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
1304#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
1305 "pavgb " #T ", " #D " \n\t"
1306#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
1307#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
1308#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
1309#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
1310#include "dsputil_h264_template_mmx.c"
1311#undef H264_CHROMA_OP
1312#undef H264_CHROMA_OP4
1313#undef H264_CHROMA_MC8_TMPL
1314#undef H264_CHROMA_MC4_TMPL
1315#undef H264_CHROMA_MC2_TMPL
1316#undef H264_CHROMA_MC8_MV0
1317
1318#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
1319#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
1320 "pavgusb " #T ", " #D " \n\t"
1321#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
1322#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
1323#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
1324#include "dsputil_h264_template_mmx.c"
1325#undef H264_CHROMA_OP
1326#undef H264_CHROMA_OP4
1327#undef H264_CHROMA_MC8_TMPL
1328#undef H264_CHROMA_MC4_TMPL
1329#undef H264_CHROMA_MC8_MV0
1330
1331/***********************************/
1332/* weighted prediction */
1333
1334static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
1335{
1336 int x, y;
1337 offset <<= log2_denom;
1338 offset += (1 << log2_denom) >> 1;
1339 asm volatile(
1340 "movd %0, %%mm4 \n\t"
1341 "movd %1, %%mm5 \n\t"
1342 "movd %2, %%mm6 \n\t"
1343 "pshufw $0, %%mm4, %%mm4 \n\t"
1344 "pshufw $0, %%mm5, %%mm5 \n\t"
1345 "pxor %%mm7, %%mm7 \n\t"
1346 :: "g"(weight), "g"(offset), "g"(log2_denom)
1347 );
1348 for(y=0; y<h; y+=2){
1349 for(x=0; x<w; x+=4){
1350 asm volatile(
1351 "movd %0, %%mm0 \n\t"
1352 "movd %1, %%mm1 \n\t"
1353 "punpcklbw %%mm7, %%mm0 \n\t"
1354 "punpcklbw %%mm7, %%mm1 \n\t"
1355 "pmullw %%mm4, %%mm0 \n\t"
1356 "pmullw %%mm4, %%mm1 \n\t"
1357 "paddsw %%mm5, %%mm0 \n\t"
1358 "paddsw %%mm5, %%mm1 \n\t"
1359 "psraw %%mm6, %%mm0 \n\t"
1360 "psraw %%mm6, %%mm1 \n\t"
1361 "packuswb %%mm7, %%mm0 \n\t"
1362 "packuswb %%mm7, %%mm1 \n\t"
1363 "movd %%mm0, %0 \n\t"
1364 "movd %%mm1, %1 \n\t"
1365 : "+m"(*(uint32_t*)(dst+x)),
1366 "+m"(*(uint32_t*)(dst+x+stride))
1367 );
1368 }
1369 dst += 2*stride;
1370 }
1371}
1372
1373static 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)
1374{
1375 int x, y;
1376 offset = ((offset + 1) | 1) << log2_denom;
1377 asm volatile(
1378 "movd %0, %%mm3 \n\t"
1379 "movd %1, %%mm4 \n\t"
1380 "movd %2, %%mm5 \n\t"
1381 "movd %3, %%mm6 \n\t"
1382 "pshufw $0, %%mm3, %%mm3 \n\t"
1383 "pshufw $0, %%mm4, %%mm4 \n\t"
1384 "pshufw $0, %%mm5, %%mm5 \n\t"
1385 "pxor %%mm7, %%mm7 \n\t"
1386 :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
1387 );
1388 for(y=0; y<h; y++){
1389 for(x=0; x<w; x+=4){
1390 asm volatile(
1391 "movd %0, %%mm0 \n\t"
1392 "movd %1, %%mm1 \n\t"
1393 "punpcklbw %%mm7, %%mm0 \n\t"
1394 "punpcklbw %%mm7, %%mm1 \n\t"
1395 "pmullw %%mm3, %%mm0 \n\t"
1396 "pmullw %%mm4, %%mm1 \n\t"
1397 "paddsw %%mm1, %%mm0 \n\t"
1398 "paddsw %%mm5, %%mm0 \n\t"
1399 "psraw %%mm6, %%mm0 \n\t"
1400 "packuswb %%mm0, %%mm0 \n\t"
1401 "movd %%mm0, %0 \n\t"
1402 : "+m"(*(uint32_t*)(dst+x))
1403 : "m"(*(uint32_t*)(src+x))
1404 );
1405 }
1406 src += stride;
1407 dst += stride;
1408 }
1409}
1410
1411#define H264_WEIGHT(W,H) \
1412static 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){ \
1413 ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
1414} \
1415static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
1416 ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
1417}
1418
1419H264_WEIGHT(16,16)
1420H264_WEIGHT(16, 8)
1421H264_WEIGHT( 8,16)
1422H264_WEIGHT( 8, 8)
1423H264_WEIGHT( 8, 4)
1424H264_WEIGHT( 4, 8)
1425H264_WEIGHT( 4, 4)
1426H264_WEIGHT( 4, 2)
1427
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette