2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
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.
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.
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
20 /***********************************/
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"
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"
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"
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 )
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 */\
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 */
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"
68 void ff_h264_idct_add_mmx2(uint8_t *dst, int16_t *block, int stride)
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"
79 /* mm1=s02+s13 mm2=s02-s13 mm4=d02+d13 mm0=d02-d13 */
80 IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
83 /* in: 1,4,0,2 out: 1,2,3,0 */
84 TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
86 "paddw %%mm6, %%mm3 \n\t"
88 /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */
89 IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
91 "pxor %%mm7, %%mm7 \n\t"
95 STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
97 STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
99 STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
101 STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
108 /***********************************/
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"
121 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
122 // out: mm5=beta-1, mm7=mask
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"
137 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
138 // out: mm1=p0' mm2=q0'
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"
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"
199 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
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 };
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"
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)
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)
243 H264_DEBLOCK_P0_Q0(%8, %9)
244 "movq %%mm1, (%1,%3,2) \n\t"
245 "movq %%mm2, (%2) \n\t"
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)
254 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
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);
261 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
263 //FIXME: could cut some load/stores by merging transpose with filter
264 // also, it only needs to transpose 6x8
267 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
268 if((tc0[0] & tc0[1]) < 0)
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);
280 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
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"
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)
301 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
303 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
306 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
308 //FIXME: could cut some load/stores by merging transpose with filter
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);
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)) */\
326 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
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)
351 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
353 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
356 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
358 //FIXME: could cut some load/stores by merging transpose with filter
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);
368 /***********************************/
369 /* motion compensation */
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"\
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)\
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"\
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"
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){\
408 "pxor %%mm7, %%mm7 \n\t"\
409 "movq %5, %%mm4 \n\t"\
410 "movq %6, %%mm5 \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)\
439 : "+a"(src), "+c"(dst), "+m"(h)\
440 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
444 static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
447 "pxor %%mm7, %%mm7 \n\t"\
448 "movd (%0), %%mm0 \n\t"\
450 "movd (%0), %%mm1 \n\t"\
452 "movd (%0), %%mm2 \n\t"\
454 "movd (%0), %%mm3 \n\t"\
456 "movd (%0), %%mm4 \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)\
468 : "+a"(src), "+c"(dst)\
469 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
473 static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
476 src -= 2*srcStride+2;\
479 "pxor %%mm7, %%mm7 \n\t"\
480 "movd (%0), %%mm0 \n\t"\
482 "movd (%0), %%mm1 \n\t"\
484 "movd (%0), %%mm2 \n\t"\
486 "movd (%0), %%mm3 \n\t"\
488 "movd (%0), %%mm4 \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)\
501 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
505 src += 4 - 9*srcStride;\
509 "movq %4, %%mm6 \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+c)/4 */\
522 "paddw %%mm6, %%mm2 \n\t"\
523 "paddw %%mm2, %%mm0 \n\t"/*(a-5*b+20*c)/16 +32 */\
524 "psraw $6, %%mm0 \n\t"\
525 "packuswb %%mm0, %%mm0 \n\t"\
526 OP(%%mm0, (%1),%%mm7, d)\
531 : "+a"(tmp), "+c"(dst), "+m"(h)\
532 : "S"((long)dstStride), "m"(ff_pw_32)\
537 static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
540 "pxor %%mm7, %%mm7 \n\t"\
541 "movq %5, %%mm6 \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)\
588 : "+a"(src), "+c"(dst), "+m"(h)\
589 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
594 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
600 "pxor %%mm7, %%mm7 \n\t"\
601 "movd (%0), %%mm0 \n\t"\
603 "movd (%0), %%mm1 \n\t"\
605 "movd (%0), %%mm2 \n\t"\
607 "movd (%0), %%mm3 \n\t"\
609 "movd (%0), %%mm4 \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)\
625 : "+a"(src), "+c"(dst)\
626 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
629 src += 4-13*srcStride;\
630 dst += 4-8*dstStride;\
633 static 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){\
635 int w = (size+8)>>2;\
636 src -= 2*srcStride+2;\
639 "pxor %%mm7, %%mm7 \n\t"\
640 "movd (%0), %%mm0 \n\t"\
642 "movd (%0), %%mm1 \n\t"\
644 "movd (%0), %%mm2 \n\t"\
646 "movd (%0), %%mm3 \n\t"\
648 "movd (%0), %%mm4 \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*48)\
656 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
657 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
658 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
659 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
660 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
661 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
662 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
664 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
669 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 8*48)\
670 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 9*48)\
671 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
672 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
673 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
674 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
675 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
676 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
678 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
683 src += 4 - (size+5)*srcStride;\
690 "movq %4, %%mm6 \n\t"\
692 "movq (%0), %%mm0 \n\t"\
693 "movq 8(%0), %%mm3 \n\t"\
694 "movq 2(%0), %%mm1 \n\t"\
695 "movq 10(%0), %%mm4 \n\t"\
696 "paddw %%mm4, %%mm0 \n\t"\
697 "paddw %%mm3, %%mm1 \n\t"\
698 "paddw 18(%0), %%mm3 \n\t"\
699 "paddw 16(%0), %%mm4 \n\t"\
700 "movq 4(%0), %%mm2 \n\t"\
701 "movq 12(%0), %%mm5 \n\t"\
702 "paddw 6(%0), %%mm2 \n\t"\
703 "paddw 14(%0), %%mm5 \n\t"\
704 "psubw %%mm1, %%mm0 \n\t"\
705 "psubw %%mm4, %%mm3 \n\t"\
706 "psraw $2, %%mm0 \n\t"\
707 "psraw $2, %%mm3 \n\t"\
708 "psubw %%mm1, %%mm0 \n\t"\
709 "psubw %%mm4, %%mm3 \n\t"\
710 "paddsw %%mm2, %%mm0 \n\t"\
711 "paddsw %%mm5, %%mm3 \n\t"\
712 "psraw $2, %%mm0 \n\t"\
713 "psraw $2, %%mm3 \n\t"\
714 "paddw %%mm6, %%mm2 \n\t"\
715 "paddw %%mm6, %%mm5 \n\t"\
716 "paddw %%mm2, %%mm0 \n\t"\
717 "paddw %%mm5, %%mm3 \n\t"\
718 "psraw $6, %%mm0 \n\t"\
719 "psraw $6, %%mm3 \n\t"\
720 "packuswb %%mm3, %%mm0 \n\t"\
721 OP(%%mm0, (%1),%%mm7, q)\
726 : "+a"(tmp), "+c"(dst), "+m"(h)\
727 : "S"((long)dstStride), "m"(ff_pw_32)\
731 dst += 8 - size*dstStride;\
734 static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
735 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
736 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
739 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
740 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
743 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
744 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
745 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
748 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
749 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
752 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
753 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
756 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
757 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
760 #define H264_MC(OPNAME, SIZE, MMX) \
761 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
762 OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
765 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
766 uint64_t temp[SIZE*SIZE/8];\
767 uint8_t * const half= (uint8_t*)temp;\
768 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
769 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
772 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
773 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
776 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
777 uint64_t temp[SIZE*SIZE/8];\
778 uint8_t * const half= (uint8_t*)temp;\
779 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
780 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+1, half, stride, stride, SIZE);\
783 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
784 uint64_t temp[SIZE*SIZE/8];\
785 uint8_t * const half= (uint8_t*)temp;\
786 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
787 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
790 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
791 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
794 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
795 uint64_t temp[SIZE*SIZE/8];\
796 uint8_t * const half= (uint8_t*)temp;\
797 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
798 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
801 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
802 uint64_t temp[SIZE*SIZE/4];\
803 uint8_t * const halfH= (uint8_t*)temp;\
804 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
805 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
806 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
807 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
810 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
811 uint64_t temp[SIZE*SIZE/4];\
812 uint8_t * const halfH= (uint8_t*)temp;\
813 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
814 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
815 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
816 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
819 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
820 uint64_t temp[SIZE*SIZE/4];\
821 uint8_t * const halfH= (uint8_t*)temp;\
822 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
823 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
824 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
825 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
828 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
829 uint64_t temp[SIZE*SIZE/4];\
830 uint8_t * const halfH= (uint8_t*)temp;\
831 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
832 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
833 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
834 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
837 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
838 uint64_t temp[SIZE*(SIZE<8?12:24)/4];\
839 int16_t * const tmp= (int16_t*)temp;\
840 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
843 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
844 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/4];\
845 uint8_t * const halfH= (uint8_t*)temp;\
846 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
847 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
848 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
849 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
850 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
853 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
854 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/4];\
855 uint8_t * const halfH= (uint8_t*)temp;\
856 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
857 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
858 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
859 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
860 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
863 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
864 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/4];\
865 uint8_t * const halfV= (uint8_t*)temp;\
866 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
867 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
868 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
869 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
870 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
873 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
874 uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/4];\
875 uint8_t * const halfV= (uint8_t*)temp;\
876 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
877 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
878 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
879 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
880 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
884 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
885 #define AVG_3DNOW_OP(a,b,temp, size) \
886 "mov" #size " " #b ", " #temp " \n\t"\
887 "pavgusb " #temp ", " #a " \n\t"\
888 "mov" #size " " #a ", " #b " \n\t"
889 #define AVG_MMX2_OP(a,b,temp, size) \
890 "mov" #size " " #b ", " #temp " \n\t"\
891 "pavgb " #temp ", " #a " \n\t"\
892 "mov" #size " " #a ", " #b " \n\t"
894 QPEL_H264(put_, PUT_OP, 3dnow)
895 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
896 QPEL_H264(put_, PUT_OP, mmx2)
897 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
899 H264_MC(put_, 4, 3dnow)
900 H264_MC(put_, 8, 3dnow)
901 H264_MC(put_, 16,3dnow)
902 H264_MC(avg_, 4, 3dnow)
903 H264_MC(avg_, 8, 3dnow)
904 H264_MC(avg_, 16,3dnow)
905 H264_MC(put_, 4, mmx2)
906 H264_MC(put_, 8, mmx2)
907 H264_MC(put_, 16,mmx2)
908 H264_MC(avg_, 4, mmx2)
909 H264_MC(avg_, 8, mmx2)
910 H264_MC(avg_, 16,mmx2)
913 #define H264_CHROMA_OP(S,D)
914 #define H264_CHROMA_OP4(S,D,T)
915 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
916 #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
917 #define H264_CHROMA_MC8_MV0 put_pixels8_mmx
918 #include "dsputil_h264_template_mmx.c"
919 #undef H264_CHROMA_OP
920 #undef H264_CHROMA_OP4
921 #undef H264_CHROMA_MC8_TMPL
922 #undef H264_CHROMA_MC4_TMPL
923 #undef H264_CHROMA_MC8_MV0
925 #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
926 #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
927 "pavgb " #T ", " #D " \n\t"
928 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
929 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
930 #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
931 #include "dsputil_h264_template_mmx.c"
932 #undef H264_CHROMA_OP
933 #undef H264_CHROMA_OP4
934 #undef H264_CHROMA_MC8_TMPL
935 #undef H264_CHROMA_MC4_TMPL
936 #undef H264_CHROMA_MC8_MV0
938 #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
939 #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
940 "pavgusb " #T ", " #D " \n\t"
941 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
942 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
943 #define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
944 #include "dsputil_h264_template_mmx.c"
945 #undef H264_CHROMA_OP
946 #undef H264_CHROMA_OP4
947 #undef H264_CHROMA_MC8_TMPL
948 #undef H264_CHROMA_MC4_TMPL
949 #undef H264_CHROMA_MC8_MV0
951 /***********************************/
952 /* weighted prediction */
954 static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
957 offset <<= log2_denom;
958 offset += (1 << log2_denom) >> 1;
960 "movd %0, %%mm4 \n\t"
961 "movd %1, %%mm5 \n\t"
962 "movd %2, %%mm6 \n\t"
963 "pshufw $0, %%mm4, %%mm4 \n\t"
964 "pshufw $0, %%mm5, %%mm5 \n\t"
965 "pxor %%mm7, %%mm7 \n\t"
966 :: "g"(weight), "g"(offset), "g"(log2_denom)
971 "movd %0, %%mm0 \n\t"
972 "movd %1, %%mm1 \n\t"
973 "punpcklbw %%mm7, %%mm0 \n\t"
974 "punpcklbw %%mm7, %%mm1 \n\t"
975 "pmullw %%mm4, %%mm0 \n\t"
976 "pmullw %%mm4, %%mm1 \n\t"
977 "paddsw %%mm5, %%mm0 \n\t"
978 "paddsw %%mm5, %%mm1 \n\t"
979 "psraw %%mm6, %%mm0 \n\t"
980 "psraw %%mm6, %%mm1 \n\t"
981 "packuswb %%mm7, %%mm0 \n\t"
982 "packuswb %%mm7, %%mm1 \n\t"
983 "movd %%mm0, %0 \n\t"
984 "movd %%mm1, %1 \n\t"
985 : "+m"(*(uint32_t*)(dst+x)),
986 "+m"(*(uint32_t*)(dst+x+stride))
993 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)
996 offset = ((offset + 1) | 1) << log2_denom;
998 "movd %0, %%mm3 \n\t"
999 "movd %1, %%mm4 \n\t"
1000 "movd %2, %%mm5 \n\t"
1001 "movd %3, %%mm6 \n\t"
1002 "pshufw $0, %%mm3, %%mm3 \n\t"
1003 "pshufw $0, %%mm4, %%mm4 \n\t"
1004 "pshufw $0, %%mm5, %%mm5 \n\t"
1005 "pxor %%mm7, %%mm7 \n\t"
1006 :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
1009 for(x=0; x<w; x+=4){
1011 "movd %0, %%mm0 \n\t"
1012 "movd %1, %%mm1 \n\t"
1013 "punpcklbw %%mm7, %%mm0 \n\t"
1014 "punpcklbw %%mm7, %%mm1 \n\t"
1015 "pmullw %%mm3, %%mm0 \n\t"
1016 "pmullw %%mm4, %%mm1 \n\t"
1017 "paddsw %%mm1, %%mm0 \n\t"
1018 "paddsw %%mm5, %%mm0 \n\t"
1019 "psraw %%mm6, %%mm0 \n\t"
1020 "packuswb %%mm0, %%mm0 \n\t"
1021 "movd %%mm0, %0 \n\t"
1022 : "+m"(*(uint32_t*)(dst+x))
1023 : "m"(*(uint32_t*)(src+x))
1031 #define H264_WEIGHT(W,H) \
1032 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){ \
1033 ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
1035 static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
1036 ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \