h264 mmx weighted prediction. up to 3% overall speedup.
[libav.git] / libavcodec / i386 / h264dsp_mmx.c
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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 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 SBUTTERFLY(a,b,t,n)\
50 "movq " #a ", " #t " \n\t" /* abcd */\
51 "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\
52 "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\
53
54 #define TRANSPOSE4(a,b,c,d,t)\
55 SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
56 SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
57 SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
58 SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
59
60 #define STORE_DIFF_4P( p, t, z ) \
61 "psraw $6, "#p" \n\t"\
62 "movd (%0), "#t" \n\t"\
63 "punpcklbw "#z", "#t" \n\t"\
64 "paddsw "#t", "#p" \n\t"\
65 "packuswb "#z", "#p" \n\t"\
66 "movd "#p", (%0) \n\t"
67
68 void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride)
69 {
70 /* Load dct coeffs */
71 asm volatile(
72 "movq (%0), %%mm0 \n\t"
73 "movq 8(%0), %%mm1 \n\t"
74 "movq 16(%0), %%mm2 \n\t"
75 "movq 24(%0), %%mm3 \n\t"
76 :: "r"(block) );
77
78 asm volatile(
79 /* mm1=s02+s13 mm2=s02-s13 mm4=d02+d13 mm0=d02-d13 */
80 IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
81
82 "movq %0, %%mm6 \n\t"
83 /* in: 1,4,0,2 out: 1,2,3,0 */
84 TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
85
86 "paddw %%mm6, %%mm3 \n\t"
87
88 /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */
89 IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
90
91 "pxor %%mm7, %%mm7 \n\t"
92 :: "m"(ff_pw_32));
93
94 asm volatile(
95 STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
96 "add %1, %0 \n\t"
97 STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
98 "add %1, %0 \n\t"
99 STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
100 "add %1, %0 \n\t"
101 STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
102 : "+r"(dst)
103 : "r" ((long)stride)
104 );
105 }
106
107
108 /***********************************/
109 /* deblocking */
110
111 // out: o = |x-y|>a
112 // clobbers: t
113 #define DIFF_GT_MMX(x,y,a,o,t)\
114 "movq "#y", "#t" \n\t"\
115 "movq "#x", "#o" \n\t"\
116 "psubusb "#x", "#t" \n\t"\
117 "psubusb "#y", "#o" \n\t"\
118 "por "#t", "#o" \n\t"\
119 "psubusb "#a", "#o" \n\t"
120
121 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
122 // out: mm5=beta-1, mm7=mask
123 // clobbers: mm4,mm6
124 #define H264_DEBLOCK_MASK(alpha1, beta1) \
125 "pshufw $0, "#alpha1", %%mm4 \n\t"\
126 "pshufw $0, "#beta1 ", %%mm5 \n\t"\
127 "packuswb %%mm4, %%mm4 \n\t"\
128 "packuswb %%mm5, %%mm5 \n\t"\
129 DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
130 DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
131 "por %%mm4, %%mm7 \n\t"\
132 DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
133 "por %%mm4, %%mm7 \n\t"\
134 "pxor %%mm6, %%mm6 \n\t"\
135 "pcmpeqb %%mm6, %%mm7 \n\t"
136
137 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
138 // out: mm1=p0' mm2=q0'
139 // clobbers: mm0,3-6
140 #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
141 /* a = q0^p0^((p1-q1)>>2) */\
142 "movq %%mm0, %%mm4 \n\t"\
143 "psubb %%mm3, %%mm4 \n\t"\
144 "psrlw $2, %%mm4 \n\t"\
145 "pxor %%mm1, %%mm4 \n\t"\
146 "pxor %%mm2, %%mm4 \n\t"\
147 /* b = p0^(q1>>2) */\
148 "psrlw $2, %%mm3 \n\t"\
149 "pand "#pb_3f", %%mm3 \n\t"\
150 "movq %%mm1, %%mm5 \n\t"\
151 "pxor %%mm3, %%mm5 \n\t"\
152 /* c = q0^(p1>>2) */\
153 "psrlw $2, %%mm0 \n\t"\
154 "pand "#pb_3f", %%mm0 \n\t"\
155 "movq %%mm2, %%mm6 \n\t"\
156 "pxor %%mm0, %%mm6 \n\t"\
157 /* d = (c^b) & ~(b^a) & 1 */\
158 "pxor %%mm5, %%mm6 \n\t"\
159 "pxor %%mm4, %%mm5 \n\t"\
160 "pandn %%mm6, %%mm5 \n\t"\
161 "pand "#pb_01", %%mm5 \n\t"\
162 /* delta = (avg(q0, p1>>2) + (d&a))
163 * - (avg(p0, q1>>2) + (d&~a)) */\
164 "pavgb %%mm2, %%mm0 \n\t"\
165 "pand %%mm5, %%mm4 \n\t"\
166 "paddusb %%mm4, %%mm0 \n\t"\
167 "pavgb %%mm1, %%mm3 \n\t"\
168 "pxor %%mm5, %%mm4 \n\t"\
169 "paddusb %%mm4, %%mm3 \n\t"\
170 /* p0 += clip(delta, -tc0, tc0)
171 * q0 -= clip(delta, -tc0, tc0) */\
172 "movq %%mm0, %%mm4 \n\t"\
173 "psubusb %%mm3, %%mm0 \n\t"\
174 "psubusb %%mm4, %%mm3 \n\t"\
175 "pminub %%mm7, %%mm0 \n\t"\
176 "pminub %%mm7, %%mm3 \n\t"\
177 "paddusb %%mm0, %%mm1 \n\t"\
178 "paddusb %%mm3, %%mm2 \n\t"\
179 "psubusb %%mm3, %%mm1 \n\t"\
180 "psubusb %%mm0, %%mm2 \n\t"
181
182 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
183 // out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
184 // clobbers: q2, tmp, tc0
185 #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
186 "movq %%mm1, "#tmp" \n\t"\
187 "pavgb %%mm2, "#tmp" \n\t"\
188 "pavgb "#tmp", "#q2" \n\t" /* avg(p2,avg(p0,q0)) */\
189 "pxor "q2addr", "#tmp" \n\t"\
190 "pand %8, "#tmp" \n\t" /* (p2^avg(p0,q0))&1 */\
191 "psubusb "#tmp", "#q2" \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
192 "movq "#p1", "#tmp" \n\t"\
193 "psubusb "#tc0", "#tmp" \n\t"\
194 "paddusb "#p1", "#tc0" \n\t"\
195 "pmaxub "#tmp", "#q2" \n\t"\
196 "pminub "#tc0", "#q2" \n\t"\
197 "movq "#q2", "q1addr" \n\t"
198
199 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
200 {
201 uint64_t tmp0;
202 uint64_t tc = (uint8_t)tc0[1]*0x01010000 | (uint8_t)tc0[0]*0x0101;
203 // with luma, tc0=0 doesn't mean no filtering, so we need a separate input mask
204 uint32_t mask[2] = { (tc0[0]>=0)*0xffffffff, (tc0[1]>=0)*0xffffffff };
205
206 asm volatile(
207 "movq (%1,%3), %%mm0 \n\t" //p1
208 "movq (%1,%3,2), %%mm1 \n\t" //p0
209 "movq (%2), %%mm2 \n\t" //q0
210 "movq (%2,%3), %%mm3 \n\t" //q1
211 H264_DEBLOCK_MASK(%6, %7)
212 "pand %5, %%mm7 \n\t"
213 "movq %%mm7, %0 \n\t"
214
215 /* filter p1 */
216 "movq (%1), %%mm3 \n\t" //p2
217 DIFF_GT_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
218 "pandn %%mm7, %%mm6 \n\t"
219 "pcmpeqb %%mm7, %%mm6 \n\t"
220 "pand %%mm7, %%mm6 \n\t" // mask & |p2-p0|<beta
221 "pshufw $80, %4, %%mm4 \n\t"
222 "pand %%mm7, %%mm4 \n\t" // mask & tc0
223 "movq %8, %%mm7 \n\t"
224 "pand %%mm6, %%mm7 \n\t" // mask & |p2-p0|<beta & 1
225 "pand %%mm4, %%mm6 \n\t" // mask & |p2-p0|<beta & tc0
226 "paddb %%mm4, %%mm7 \n\t" // tc++
227 H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
228
229 /* filter q1 */
230 "movq (%2,%3,2), %%mm4 \n\t" //q2
231 DIFF_GT_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
232 "pandn %0, %%mm6 \n\t"
233 "pcmpeqb %0, %%mm6 \n\t"
234 "pand %0, %%mm6 \n\t"
235 "pshufw $80, %4, %%mm5 \n\t"
236 "pand %%mm6, %%mm5 \n\t"
237 "pand %8, %%mm6 \n\t"
238 "paddb %%mm6, %%mm7 \n\t"
239 "movq (%2,%3), %%mm3 \n\t"
240 H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
241
242 /* filter p0, q0 */
243 H264_DEBLOCK_P0_Q0(%8, %9)
244 "movq %%mm1, (%1,%3,2) \n\t"
245 "movq %%mm2, (%2) \n\t"
246
247 : "=m"(tmp0)
248 : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
249 "m"(tc), "m"(*(uint64_t*)mask), "m"(alpha1), "m"(beta1),
250 "m"(mm_bone), "m"(ff_pb_3F)
251 );
252 }
253
254 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
255 {
256 if((tc0[0] & tc0[1]) >= 0)
257 h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
258 if((tc0[2] & tc0[3]) >= 0)
259 h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
260 }
261 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
262 {
263 //FIXME: could cut some load/stores by merging transpose with filter
264 // also, it only needs to transpose 6x8
265 uint8_t trans[8*8];
266 int i;
267 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
268 if((tc0[0] & tc0[1]) < 0)
269 continue;
270 transpose4x4(trans, pix-4, 8, stride);
271 transpose4x4(trans +4*8, pix, 8, stride);
272 transpose4x4(trans+4, pix-4+4*stride, 8, stride);
273 transpose4x4(trans+4+4*8, pix +4*stride, 8, stride);
274 h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
275 transpose4x4(pix-2, trans +2*8, stride, 8);
276 transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
277 }
278 }
279
280 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
281 {
282 asm volatile(
283 "movq (%0), %%mm0 \n\t" //p1
284 "movq (%0,%2), %%mm1 \n\t" //p0
285 "movq (%1), %%mm2 \n\t" //q0
286 "movq (%1,%2), %%mm3 \n\t" //q1
287 H264_DEBLOCK_MASK(%4, %5)
288 "movd %3, %%mm6 \n\t"
289 "punpcklbw %%mm6, %%mm6 \n\t"
290 "pand %%mm6, %%mm7 \n\t" // mm7 = tc&mask
291 H264_DEBLOCK_P0_Q0(%6, %7)
292 "movq %%mm1, (%0,%2) \n\t"
293 "movq %%mm2, (%1) \n\t"
294
295 :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
296 "r"(*(uint32_t*)tc0),
297 "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
298 );
299 }
300
301 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
302 {
303 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
304 }
305
306 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
307 {
308 //FIXME: could cut some load/stores by merging transpose with filter
309 uint8_t trans[8*4];
310 transpose4x4(trans, pix-2, 8, stride);
311 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
312 h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
313 transpose4x4(pix-2, trans, stride, 8);
314 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
315 }
316
317 // p0 = (p0 + q1 + 2*p1 + 2) >> 2
318 #define H264_FILTER_CHROMA4(p0, p1, q1, one) \
319 "movq "#p0", %%mm4 \n\t"\
320 "pxor "#q1", %%mm4 \n\t"\
321 "pand "#one", %%mm4 \n\t" /* mm4 = (p0^q1)&1 */\
322 "pavgb "#q1", "#p0" \n\t"\
323 "psubusb %%mm4, "#p0" \n\t"\
324 "pavgb "#p1", "#p0" \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
325
326 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
327 {
328 asm volatile(
329 "movq (%0), %%mm0 \n\t"
330 "movq (%0,%2), %%mm1 \n\t"
331 "movq (%1), %%mm2 \n\t"
332 "movq (%1,%2), %%mm3 \n\t"
333 H264_DEBLOCK_MASK(%3, %4)
334 "movq %%mm1, %%mm5 \n\t"
335 "movq %%mm2, %%mm6 \n\t"
336 H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
337 H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
338 "psubb %%mm5, %%mm1 \n\t"
339 "psubb %%mm6, %%mm2 \n\t"
340 "pand %%mm7, %%mm1 \n\t"
341 "pand %%mm7, %%mm2 \n\t"
342 "paddb %%mm5, %%mm1 \n\t"
343 "paddb %%mm6, %%mm2 \n\t"
344 "movq %%mm1, (%0,%2) \n\t"
345 "movq %%mm2, (%1) \n\t"
346 :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
347 "m"(alpha1), "m"(beta1), "m"(mm_bone)
348 );
349 }
350
351 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
352 {
353 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
354 }
355
356 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
357 {
358 //FIXME: could cut some load/stores by merging transpose with filter
359 uint8_t trans[8*4];
360 transpose4x4(trans, pix-2, 8, stride);
361 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
362 h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
363 transpose4x4(pix-2, trans, stride, 8);
364 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
365 }
366
367
368 /***********************************/
369 /* motion compensation */
370
371 #define QPEL_H264V(A,B,C,D,E,F,OP)\
372 "movd (%0), "#F" \n\t"\
373 "movq "#C", %%mm6 \n\t"\
374 "paddw "#D", %%mm6 \n\t"\
375 "psllw $2, %%mm6 \n\t"\
376 "psubw "#B", %%mm6 \n\t"\
377 "psubw "#E", %%mm6 \n\t"\
378 "pmullw %4, %%mm6 \n\t"\
379 "add %2, %0 \n\t"\
380 "punpcklbw %%mm7, "#F" \n\t"\
381 "paddw %5, "#A" \n\t"\
382 "paddw "#F", "#A" \n\t"\
383 "paddw "#A", %%mm6 \n\t"\
384 "psraw $5, %%mm6 \n\t"\
385 "packuswb %%mm6, %%mm6 \n\t"\
386 OP(%%mm6, (%1), A, d)\
387 "add %3, %1 \n\t"
388
389 #define QPEL_H264HV(A,B,C,D,E,F,OF)\
390 "movd (%0), "#F" \n\t"\
391 "movq "#C", %%mm6 \n\t"\
392 "paddw "#D", %%mm6 \n\t"\
393 "psllw $2, %%mm6 \n\t"\
394 "psubw "#B", %%mm6 \n\t"\
395 "psubw "#E", %%mm6 \n\t"\
396 "pmullw %3, %%mm6 \n\t"\
397 "add %2, %0 \n\t"\
398 "punpcklbw %%mm7, "#F" \n\t"\
399 "paddw "#F", "#A" \n\t"\
400 "paddw "#A", %%mm6 \n\t"\
401 "movq %%mm6, "#OF"(%1) \n\t"
402
403 #define QPEL_H264(OPNAME, OP, MMX)\
404 static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
405 int h=4;\
406 \
407 asm volatile(\
408 "pxor %%mm7, %%mm7 \n\t"\
409 "movq %5, %%mm4 \n\t"\
410 "movq %6, %%mm5 \n\t"\
411 "1: \n\t"\
412 "movd -1(%0), %%mm1 \n\t"\
413 "movd (%0), %%mm2 \n\t"\
414 "movd 1(%0), %%mm3 \n\t"\
415 "movd 2(%0), %%mm0 \n\t"\
416 "punpcklbw %%mm7, %%mm1 \n\t"\
417 "punpcklbw %%mm7, %%mm2 \n\t"\
418 "punpcklbw %%mm7, %%mm3 \n\t"\
419 "punpcklbw %%mm7, %%mm0 \n\t"\
420 "paddw %%mm0, %%mm1 \n\t"\
421 "paddw %%mm3, %%mm2 \n\t"\
422 "movd -2(%0), %%mm0 \n\t"\
423 "movd 3(%0), %%mm3 \n\t"\
424 "punpcklbw %%mm7, %%mm0 \n\t"\
425 "punpcklbw %%mm7, %%mm3 \n\t"\
426 "paddw %%mm3, %%mm0 \n\t"\
427 "psllw $2, %%mm2 \n\t"\
428 "psubw %%mm1, %%mm2 \n\t"\
429 "pmullw %%mm4, %%mm2 \n\t"\
430 "paddw %%mm5, %%mm0 \n\t"\
431 "paddw %%mm2, %%mm0 \n\t"\
432 "psraw $5, %%mm0 \n\t"\
433 "packuswb %%mm0, %%mm0 \n\t"\
434 OP(%%mm0, (%1),%%mm6, d)\
435 "add %3, %0 \n\t"\
436 "add %4, %1 \n\t"\
437 "decl %2 \n\t"\
438 " jnz 1b \n\t"\
439 : "+a"(src), "+c"(dst), "+m"(h)\
440 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
441 : "memory"\
442 );\
443 }\
444 static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
445 src -= 2*srcStride;\
446 asm volatile(\
447 "pxor %%mm7, %%mm7 \n\t"\
448 "movd (%0), %%mm0 \n\t"\
449 "add %2, %0 \n\t"\
450 "movd (%0), %%mm1 \n\t"\
451 "add %2, %0 \n\t"\
452 "movd (%0), %%mm2 \n\t"\
453 "add %2, %0 \n\t"\
454 "movd (%0), %%mm3 \n\t"\
455 "add %2, %0 \n\t"\
456 "movd (%0), %%mm4 \n\t"\
457 "add %2, %0 \n\t"\
458 "punpcklbw %%mm7, %%mm0 \n\t"\
459 "punpcklbw %%mm7, %%mm1 \n\t"\
460 "punpcklbw %%mm7, %%mm2 \n\t"\
461 "punpcklbw %%mm7, %%mm3 \n\t"\
462 "punpcklbw %%mm7, %%mm4 \n\t"\
463 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
464 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
465 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
466 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
467 \
468 : "+a"(src), "+c"(dst)\
469 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
470 : "memory"\
471 );\
472 }\
473 static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
474 int h=4;\
475 int w=3;\
476 src -= 2*srcStride+2;\
477 while(w--){\
478 asm volatile(\
479 "pxor %%mm7, %%mm7 \n\t"\
480 "movd (%0), %%mm0 \n\t"\
481 "add %2, %0 \n\t"\
482 "movd (%0), %%mm1 \n\t"\
483 "add %2, %0 \n\t"\
484 "movd (%0), %%mm2 \n\t"\
485 "add %2, %0 \n\t"\
486 "movd (%0), %%mm3 \n\t"\
487 "add %2, %0 \n\t"\
488 "movd (%0), %%mm4 \n\t"\
489 "add %2, %0 \n\t"\
490 "punpcklbw %%mm7, %%mm0 \n\t"\
491 "punpcklbw %%mm7, %%mm1 \n\t"\
492 "punpcklbw %%mm7, %%mm2 \n\t"\
493 "punpcklbw %%mm7, %%mm3 \n\t"\
494 "punpcklbw %%mm7, %%mm4 \n\t"\
495 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
496 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
497 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
498 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
499 \
500 : "+a"(src)\
501 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
502 : "memory"\
503 );\
504 tmp += 4;\
505 src += 4 - 9*srcStride;\
506 }\
507 tmp -= 3*4;\
508 asm volatile(\
509 "movq %4, %%mm6 \n\t"\
510 "1: \n\t"\
511 "movq (%0), %%mm0 \n\t"\
512 "paddw 10(%0), %%mm0 \n\t"\
513 "movq 2(%0), %%mm1 \n\t"\
514 "paddw 8(%0), %%mm1 \n\t"\
515 "movq 4(%0), %%mm2 \n\t"\
516 "paddw 6(%0), %%mm2 \n\t"\
517 "psubw %%mm1, %%mm0 \n\t"/*a-b (abccba)*/\
518 "psraw $2, %%mm0 \n\t"/*(a-b)/4 */\
519 "psubw %%mm1, %%mm0 \n\t"/*(a-b)/4-b */\
520 "paddsw %%mm2, %%mm0 \n\t"\
521 "psraw $2, %%mm0 \n\t"/*((a-b)/4-b)/4 */\
522 "paddw %%mm6, %%mm2 \n\t"\
523 "paddw %%mm2, %%mm0 \n\t"\
524 "psraw $6, %%mm0 \n\t"\
525 "packuswb %%mm0, %%mm0 \n\t"\
526 OP(%%mm0, (%1),%%mm7, d)\
527 "add $24, %0 \n\t"\
528 "add %3, %1 \n\t"\
529 "decl %2 \n\t"\
530 " jnz 1b \n\t"\
531 : "+a"(tmp), "+c"(dst), "+m"(h)\
532 : "S"((long)dstStride), "m"(ff_pw_32)\
533 : "memory"\
534 );\
535 }\
536 \
537 static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
538 int h=8;\
539 asm volatile(\
540 "pxor %%mm7, %%mm7 \n\t"\
541 "movq %5, %%mm6 \n\t"\
542 "1: \n\t"\
543 "movq (%0), %%mm0 \n\t"\
544 "movq 1(%0), %%mm2 \n\t"\
545 "movq %%mm0, %%mm1 \n\t"\
546 "movq %%mm2, %%mm3 \n\t"\
547 "punpcklbw %%mm7, %%mm0 \n\t"\
548 "punpckhbw %%mm7, %%mm1 \n\t"\
549 "punpcklbw %%mm7, %%mm2 \n\t"\
550 "punpckhbw %%mm7, %%mm3 \n\t"\
551 "paddw %%mm2, %%mm0 \n\t"\
552 "paddw %%mm3, %%mm1 \n\t"\
553 "psllw $2, %%mm0 \n\t"\
554 "psllw $2, %%mm1 \n\t"\
555 "movq -1(%0), %%mm2 \n\t"\
556 "movq 2(%0), %%mm4 \n\t"\
557 "movq %%mm2, %%mm3 \n\t"\
558 "movq %%mm4, %%mm5 \n\t"\
559 "punpcklbw %%mm7, %%mm2 \n\t"\
560 "punpckhbw %%mm7, %%mm3 \n\t"\
561 "punpcklbw %%mm7, %%mm4 \n\t"\
562 "punpckhbw %%mm7, %%mm5 \n\t"\
563 "paddw %%mm4, %%mm2 \n\t"\
564 "paddw %%mm3, %%mm5 \n\t"\
565 "psubw %%mm2, %%mm0 \n\t"\
566 "psubw %%mm5, %%mm1 \n\t"\
567 "pmullw %%mm6, %%mm0 \n\t"\
568 "pmullw %%mm6, %%mm1 \n\t"\
569 "movd -2(%0), %%mm2 \n\t"\
570 "movd 7(%0), %%mm5 \n\t"\
571 "punpcklbw %%mm7, %%mm2 \n\t"\
572 "punpcklbw %%mm7, %%mm5 \n\t"\
573 "paddw %%mm3, %%mm2 \n\t"\
574 "paddw %%mm5, %%mm4 \n\t"\
575 "movq %6, %%mm5 \n\t"\
576 "paddw %%mm5, %%mm2 \n\t"\
577 "paddw %%mm5, %%mm4 \n\t"\
578 "paddw %%mm2, %%mm0 \n\t"\
579 "paddw %%mm4, %%mm1 \n\t"\
580 "psraw $5, %%mm0 \n\t"\
581 "psraw $5, %%mm1 \n\t"\
582 "packuswb %%mm1, %%mm0 \n\t"\
583 OP(%%mm0, (%1),%%mm5, q)\
584 "add %3, %0 \n\t"\
585 "add %4, %1 \n\t"\
586 "decl %2 \n\t"\
587 " jnz 1b \n\t"\
588 : "+a"(src), "+c"(dst), "+m"(h)\
589 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
590 : "memory"\
591 );\
592 }\
593 \
594 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
595 int h= 2;\
596 src -= 2*srcStride;\
597 \
598 while(h--){\
599 asm volatile(\
600 "pxor %%mm7, %%mm7 \n\t"\
601 "movd (%0), %%mm0 \n\t"\
602 "add %2, %0 \n\t"\
603 "movd (%0), %%mm1 \n\t"\
604 "add %2, %0 \n\t"\
605 "movd (%0), %%mm2 \n\t"\
606 "add %2, %0 \n\t"\
607 "movd (%0), %%mm3 \n\t"\
608 "add %2, %0 \n\t"\
609 "movd (%0), %%mm4 \n\t"\
610 "add %2, %0 \n\t"\
611 "punpcklbw %%mm7, %%mm0 \n\t"\
612 "punpcklbw %%mm7, %%mm1 \n\t"\
613 "punpcklbw %%mm7, %%mm2 \n\t"\
614 "punpcklbw %%mm7, %%mm3 \n\t"\
615 "punpcklbw %%mm7, %%mm4 \n\t"\
616 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
617 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
618 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
619 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
620 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
621 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
622 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
623 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
624 \
625 : "+a"(src), "+c"(dst)\
626 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
627 : "memory"\
628 );\
629 src += 4-13*srcStride;\
630 dst += 4-8*dstStride;\
631 }\
632 }\
633 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
634 int h=8;\
635 int w=4;\
636 src -= 2*srcStride+2;\
637 while(w--){\
638 asm volatile(\
639 "pxor %%mm7, %%mm7 \n\t"\
640 "movd (%0), %%mm0 \n\t"\
641 "add %2, %0 \n\t"\
642 "movd (%0), %%mm1 \n\t"\
643 "add %2, %0 \n\t"\
644 "movd (%0), %%mm2 \n\t"\
645 "add %2, %0 \n\t"\
646 "movd (%0), %%mm3 \n\t"\
647 "add %2, %0 \n\t"\
648 "movd (%0), %%mm4 \n\t"\
649 "add %2, %0 \n\t"\
650 "punpcklbw %%mm7, %%mm0 \n\t"\
651 "punpcklbw %%mm7, %%mm1 \n\t"\
652 "punpcklbw %%mm7, %%mm2 \n\t"\
653 "punpcklbw %%mm7, %%mm3 \n\t"\
654 "punpcklbw %%mm7, %%mm4 \n\t"\
655 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*4)\
656 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*4)\
657 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*4)\
658 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*4)\
659 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*8*4)\
660 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*8*4)\
661 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*8*4)\
662 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*8*4)\
663 \
664 : "+a"(src)\
665 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
666 : "memory"\
667 );\
668 tmp += 4;\
669 src += 4 - 13*srcStride;\
670 }\
671 tmp -= 4*4;\
672 asm volatile(\
673 "movq %4, %%mm6 \n\t"\
674 "1: \n\t"\
675 "movq (%0), %%mm0 \n\t"\
676 "movq 8(%0), %%mm3 \n\t"\
677 "movq 2(%0), %%mm1 \n\t"\
678 "movq 10(%0), %%mm4 \n\t"\
679 "paddw %%mm4, %%mm0 \n\t"\
680 "paddw %%mm3, %%mm1 \n\t"\
681 "paddw 18(%0), %%mm3 \n\t"\
682 "paddw 16(%0), %%mm4 \n\t"\
683 "movq 4(%0), %%mm2 \n\t"\
684 "movq 12(%0), %%mm5 \n\t"\
685 "paddw 6(%0), %%mm2 \n\t"\
686 "paddw 14(%0), %%mm5 \n\t"\
687 "psubw %%mm1, %%mm0 \n\t"\
688 "psubw %%mm4, %%mm3 \n\t"\
689 "psraw $2, %%mm0 \n\t"\
690 "psraw $2, %%mm3 \n\t"\
691 "psubw %%mm1, %%mm0 \n\t"\
692 "psubw %%mm4, %%mm3 \n\t"\
693 "paddsw %%mm2, %%mm0 \n\t"\
694 "paddsw %%mm5, %%mm3 \n\t"\
695 "psraw $2, %%mm0 \n\t"\
696 "psraw $2, %%mm3 \n\t"\
697 "paddw %%mm6, %%mm2 \n\t"\
698 "paddw %%mm6, %%mm5 \n\t"\
699 "paddw %%mm2, %%mm0 \n\t"\
700 "paddw %%mm5, %%mm3 \n\t"\
701 "psraw $6, %%mm0 \n\t"\
702 "psraw $6, %%mm3 \n\t"\
703 "packuswb %%mm3, %%mm0 \n\t"\
704 OP(%%mm0, (%1),%%mm7, q)\
705 "add $32, %0 \n\t"\
706 "add %3, %1 \n\t"\
707 "decl %2 \n\t"\
708 " jnz 1b \n\t"\
709 : "+a"(tmp), "+c"(dst), "+m"(h)\
710 : "S"((long)dstStride), "m"(ff_pw_32)\
711 : "memory"\
712 );\
713 }\
714 static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
715 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
716 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
717 src += 8*srcStride;\
718 dst += 8*dstStride;\
719 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
720 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
721 }\
722 \
723 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
724 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
725 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
726 src += 8*srcStride;\
727 dst += 8*dstStride;\
728 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
729 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
730 }\
731 \
732 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
733 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride);\
734 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp , src+8, dstStride, tmpStride, srcStride);\
735 src += 8*srcStride;\
736 dst += 8*dstStride;\
737 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride);\
738 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp , src+8, dstStride, tmpStride, srcStride);\
739 }\
740
741 #define H264_MC(OPNAME, SIZE, MMX) \
742 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
743 OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
744 }\
745 \
746 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
747 uint64_t temp[SIZE*SIZE/8];\
748 uint8_t * const half= (uint8_t*)temp;\
749 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
750 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
751 }\
752 \
753 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
754 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
755 }\
756 \
757 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
758 uint64_t temp[SIZE*SIZE/8];\
759 uint8_t * const half= (uint8_t*)temp;\
760 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
761 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+1, half, stride, stride, SIZE);\
762 }\
763 \
764 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
765 uint64_t temp[SIZE*SIZE/8];\
766 uint8_t * const half= (uint8_t*)temp;\
767 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
768 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
769 }\
770 \
771 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
772 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
773 }\
774 \
775 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
776 uint64_t temp[SIZE*SIZE/8];\
777 uint8_t * const half= (uint8_t*)temp;\
778 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
779 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
780 }\
781 \
782 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
783 uint64_t temp[SIZE*SIZE/4];\
784 uint8_t * const halfH= (uint8_t*)temp;\
785 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
786 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
787 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
788 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
789 }\
790 \
791 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
792 uint64_t temp[SIZE*SIZE/4];\
793 uint8_t * const halfH= (uint8_t*)temp;\
794 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
795 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
796 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
797 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
798 }\
799 \
800 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
801 uint64_t temp[SIZE*SIZE/4];\
802 uint8_t * const halfH= (uint8_t*)temp;\
803 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
804 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
805 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
806 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
807 }\
808 \
809 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
810 uint64_t temp[SIZE*SIZE/4];\
811 uint8_t * const halfH= (uint8_t*)temp;\
812 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
813 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
814 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
815 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
816 }\
817 \
818 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
819 uint64_t temp[SIZE*(SIZE+8)/4];\
820 int16_t * const tmp= (int16_t*)temp;\
821 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
822 }\
823 \
824 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
825 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
826 uint8_t * const halfH= (uint8_t*)temp;\
827 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
828 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
829 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
830 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
831 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
832 }\
833 \
834 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
835 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
836 uint8_t * const halfH= (uint8_t*)temp;\
837 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
838 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
839 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
840 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
841 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
842 }\
843 \
844 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
845 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
846 uint8_t * const halfV= (uint8_t*)temp;\
847 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
848 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
849 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
850 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
851 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
852 }\
853 \
854 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
855 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
856 uint8_t * const halfV= (uint8_t*)temp;\
857 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
858 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
859 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
860 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
861 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
862 }\
863
864
865 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
866 #define AVG_3DNOW_OP(a,b,temp, size) \
867 "mov" #size " " #b ", " #temp " \n\t"\
868 "pavgusb " #temp ", " #a " \n\t"\
869 "mov" #size " " #a ", " #b " \n\t"
870 #define AVG_MMX2_OP(a,b,temp, size) \
871 "mov" #size " " #b ", " #temp " \n\t"\
872 "pavgb " #temp ", " #a " \n\t"\
873 "mov" #size " " #a ", " #b " \n\t"
874
875 QPEL_H264(put_, PUT_OP, 3dnow)
876 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
877 QPEL_H264(put_, PUT_OP, mmx2)
878 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
879
880 H264_MC(put_, 4, 3dnow)
881 H264_MC(put_, 8, 3dnow)
882 H264_MC(put_, 16,3dnow)
883 H264_MC(avg_, 4, 3dnow)
884 H264_MC(avg_, 8, 3dnow)
885 H264_MC(avg_, 16,3dnow)
886 H264_MC(put_, 4, mmx2)
887 H264_MC(put_, 8, mmx2)
888 H264_MC(put_, 16,mmx2)
889 H264_MC(avg_, 4, mmx2)
890 H264_MC(avg_, 8, mmx2)
891 H264_MC(avg_, 16,mmx2)
892
893
894 #define H264_CHROMA_OP(S,D)
895 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
896 #include "dsputil_h264_template_mmx.c"
897 #undef H264_CHROMA_OP
898 #undef H264_CHROMA_MC8_TMPL
899
900 #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
901 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
902 #include "dsputil_h264_template_mmx.c"
903 #undef H264_CHROMA_OP
904 #undef H264_CHROMA_MC8_TMPL
905
906 #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
907 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
908 #include "dsputil_h264_template_mmx.c"
909 #undef H264_CHROMA_OP
910 #undef H264_CHROMA_MC8_TMPL
911
912 /***********************************/
913 /* weighted prediction */
914
915 static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
916 {
917 int x, y;
918 offset <<= log2_denom;
919 offset += (1 << log2_denom) >> 1;
920 asm volatile(
921 "movd %0, %%mm4 \n\t"
922 "movd %1, %%mm5 \n\t"
923 "movd %2, %%mm6 \n\t"
924 "pshufw $0, %%mm4, %%mm4 \n\t"
925 "pshufw $0, %%mm5, %%mm5 \n\t"
926 "pxor %%mm7, %%mm7 \n\t"
927 :: "g"(weight), "g"(offset), "g"(log2_denom)
928 );
929 for(y=0; y<h; y+=2){
930 for(x=0; x<w; x+=4){
931 asm volatile(
932 "movd %0, %%mm0 \n\t"
933 "movd %1, %%mm1 \n\t"
934 "punpcklbw %%mm7, %%mm0 \n\t"
935 "punpcklbw %%mm7, %%mm1 \n\t"
936 "pmullw %%mm4, %%mm0 \n\t"
937 "pmullw %%mm4, %%mm1 \n\t"
938 "paddw %%mm5, %%mm0 \n\t"
939 "paddw %%mm5, %%mm1 \n\t"
940 "psraw %%mm6, %%mm0 \n\t"
941 "psraw %%mm6, %%mm1 \n\t"
942 "packuswb %%mm7, %%mm0 \n\t"
943 "packuswb %%mm7, %%mm1 \n\t"
944 "movd %%mm0, %0 \n\t"
945 "movd %%mm1, %1 \n\t"
946 : "+m"(*(uint32_t*)(dst+x)),
947 "+m"(*(uint32_t*)(dst+x+stride))
948 );
949 }
950 dst += 2*stride;
951 }
952 }
953
954 static inline void ff_h264_biweight_WxH_mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offsetd, int offsets, int w, int h)
955 {
956 int x, y;
957 int offset = ((offsets + offsetd + 1) | 1) << log2_denom;
958 asm volatile(
959 "movd %0, %%mm3 \n\t"
960 "movd %1, %%mm4 \n\t"
961 "movd %2, %%mm5 \n\t"
962 "movd %3, %%mm6 \n\t"
963 "pshufw $0, %%mm3, %%mm3 \n\t"
964 "pshufw $0, %%mm4, %%mm4 \n\t"
965 "pshufw $0, %%mm5, %%mm5 \n\t"
966 "pxor %%mm7, %%mm7 \n\t"
967 :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
968 );
969 for(y=0; y<h; y++){
970 for(x=0; x<w; x+=4){
971 asm volatile(
972 "movd %0, %%mm0 \n\t"
973 "movd %1, %%mm1 \n\t"
974 "punpcklbw %%mm7, %%mm0 \n\t"
975 "punpcklbw %%mm7, %%mm1 \n\t"
976 "pmullw %%mm3, %%mm0 \n\t"
977 "pmullw %%mm4, %%mm1 \n\t"
978 "paddw %%mm5, %%mm0 \n\t"
979 "paddw %%mm1, %%mm0 \n\t"
980 "psraw %%mm6, %%mm0 \n\t"
981 "packuswb %%mm0, %%mm0 \n\t"
982 "movd %%mm0, %0 \n\t"
983 : "+m"(*(uint32_t*)(dst+x))
984 : "m"(*(uint32_t*)(src+x))
985 );
986 }
987 src += stride;
988 dst += stride;
989 }
990 }
991
992 #define H264_WEIGHT(W,H) \
993 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 offsetd, int offsets){ \
994 ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offsetd, offsets, W, H); \
995 } \
996 static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
997 ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
998 }
999
1000 H264_WEIGHT(16,16)
1001 H264_WEIGHT(16, 8)
1002 H264_WEIGHT( 8,16)
1003 H264_WEIGHT( 8, 8)
1004 H264_WEIGHT( 8, 4)
1005 H264_WEIGHT( 4, 8)
1006 H264_WEIGHT( 4, 4)
1007 H264_WEIGHT( 4, 2)
1008