sort H.264 mmx dsp functions into their own file
[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 "movq %%mm5, %%mm6 \n\t"\
166 "pand %%mm4, %%mm6 \n\t"\
167 "paddusb %%mm6, %%mm0 \n\t"\
168 "pavgb %%mm1, %%mm3 \n\t"\
169 "pandn %%mm5, %%mm4 \n\t"\
170 "paddusb %%mm4, %%mm3 \n\t"\
171 /* p0 += clip(delta, -tc0, tc0)
172 * q0 -= clip(delta, -tc0, tc0) */\
173 "movq %%mm0, %%mm4 \n\t"\
174 "psubusb %%mm3, %%mm0 \n\t"\
175 "psubusb %%mm4, %%mm3 \n\t"\
176 "pminub %%mm7, %%mm0 \n\t"\
177 "pminub %%mm7, %%mm3 \n\t"\
178 "paddusb %%mm0, %%mm1 \n\t"\
179 "paddusb %%mm3, %%mm2 \n\t"\
180 "psubusb %%mm3, %%mm1 \n\t"\
181 "psubusb %%mm0, %%mm2 \n\t"
182
183 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
184 // out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
185 // clobbers: q2, tmp, tc0
186 #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
187 "movq %%mm1, "#tmp" \n\t"\
188 "pavgb %%mm2, "#tmp" \n\t"\
189 "pavgb "#tmp", "#q2" \n\t" /* avg(p2,avg(p0,q0)) */\
190 "pxor "q2addr", "#tmp" \n\t"\
191 "pand %8, "#tmp" \n\t" /* (p2^avg(p0,q0))&1 */\
192 "psubusb "#tmp", "#q2" \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
193 "movq "#p1", "#tmp" \n\t"\
194 "psubusb "#tc0", "#tmp" \n\t"\
195 "paddusb "#p1", "#tc0" \n\t"\
196 "pmaxub "#tmp", "#q2" \n\t"\
197 "pminub "#tc0", "#q2" \n\t"\
198 "movq "#q2", "q1addr" \n\t"
199
200 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
201 {
202 uint64_t tmp0;
203 uint64_t tc = (uint8_t)tc0[1]*0x01010000 | (uint8_t)tc0[0]*0x0101;
204 // with luma, tc0=0 doesn't mean no filtering, so we need a separate input mask
205 uint32_t mask[2] = { (tc0[0]>=0)*0xffffffff, (tc0[1]>=0)*0xffffffff };
206
207 asm volatile(
208 "movq (%1,%3), %%mm0 \n\t" //p1
209 "movq (%1,%3,2), %%mm1 \n\t" //p0
210 "movq (%2), %%mm2 \n\t" //q0
211 "movq (%2,%3), %%mm3 \n\t" //q1
212 H264_DEBLOCK_MASK(%6, %7)
213 "pand %5, %%mm7 \n\t"
214 "movq %%mm7, %0 \n\t"
215
216 /* filter p1 */
217 "movq (%1), %%mm3 \n\t" //p2
218 DIFF_GT_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
219 "pandn %%mm7, %%mm6 \n\t"
220 "pcmpeqb %%mm7, %%mm6 \n\t"
221 "pand %%mm7, %%mm6 \n\t" // mask & |p2-p0|<beta
222 "pshufw $80, %4, %%mm4 \n\t"
223 "pand %%mm7, %%mm4 \n\t" // mask & tc0
224 "movq %8, %%mm7 \n\t"
225 "pand %%mm6, %%mm7 \n\t" // mask & |p2-p0|<beta & 1
226 "pand %%mm4, %%mm6 \n\t" // mask & |p2-p0|<beta & tc0
227 "paddb %%mm4, %%mm7 \n\t" // tc++
228 H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
229
230 /* filter q1 */
231 "movq (%2,%3,2), %%mm4 \n\t" //q2
232 DIFF_GT_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
233 "pandn %0, %%mm6 \n\t"
234 "pcmpeqb %0, %%mm6 \n\t"
235 "pand %0, %%mm6 \n\t"
236 "pshufw $80, %4, %%mm5 \n\t"
237 "pand %%mm6, %%mm5 \n\t"
238 "pand %8, %%mm6 \n\t"
239 "paddb %%mm6, %%mm7 \n\t"
240 "movq (%2,%3), %%mm3 \n\t"
241 H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
242
243 /* filter p0, q0 */
244 H264_DEBLOCK_P0_Q0(%8, %9)
245 "movq %%mm1, (%1,%3,2) \n\t"
246 "movq %%mm2, (%2) \n\t"
247
248 : "=m"(tmp0)
249 : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
250 "m"(tc), "m"(*(uint64_t*)mask), "m"(alpha1), "m"(beta1),
251 "m"(mm_bone), "m"(ff_pb_3F)
252 );
253 }
254
255 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
256 {
257 if((tc0[0] & tc0[1]) >= 0)
258 h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
259 if((tc0[2] & tc0[3]) >= 0)
260 h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
261 }
262 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
263 {
264 //FIXME: could cut some load/stores by merging transpose with filter
265 // also, it only needs to transpose 6x8
266 uint8_t trans[8*8];
267 int i;
268 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
269 if((tc0[0] & tc0[1]) < 0)
270 continue;
271 transpose4x4(trans, pix-4, 8, stride);
272 transpose4x4(trans +4*8, pix, 8, stride);
273 transpose4x4(trans+4, pix-4+4*stride, 8, stride);
274 transpose4x4(trans+4+4*8, pix +4*stride, 8, stride);
275 h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
276 transpose4x4(pix-2, trans +2*8, stride, 8);
277 transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
278 }
279 }
280
281 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
282 {
283 asm volatile(
284 "movq (%0), %%mm0 \n\t" //p1
285 "movq (%0,%2), %%mm1 \n\t" //p0
286 "movq (%1), %%mm2 \n\t" //q0
287 "movq (%1,%2), %%mm3 \n\t" //q1
288 H264_DEBLOCK_MASK(%4, %5)
289 "movd %3, %%mm6 \n\t"
290 "punpcklbw %%mm6, %%mm6 \n\t"
291 "pand %%mm6, %%mm7 \n\t" // mm7 = tc&mask
292 H264_DEBLOCK_P0_Q0(%6, %7)
293 "movq %%mm1, (%0,%2) \n\t"
294 "movq %%mm2, (%1) \n\t"
295
296 :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
297 "r"(*(uint32_t*)tc0),
298 "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
299 );
300 }
301
302 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
303 {
304 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
305 }
306
307 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
308 {
309 //FIXME: could cut some load/stores by merging transpose with filter
310 uint8_t trans[8*4];
311 transpose4x4(trans, pix-2, 8, stride);
312 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
313 h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
314 transpose4x4(pix-2, trans, stride, 8);
315 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
316 }
317
318 // p0 = (p0 + q1 + 2*p1 + 2) >> 2
319 #define H264_FILTER_CHROMA4(p0, p1, q1, one) \
320 "movq "#p0", %%mm4 \n\t"\
321 "pxor "#q1", %%mm4 \n\t"\
322 "pand "#one", %%mm4 \n\t" /* mm4 = (p0^q1)&1 */\
323 "pavgb "#q1", "#p0" \n\t"\
324 "psubusb %%mm4, "#p0" \n\t"\
325 "pavgb "#p1", "#p0" \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
326
327 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
328 {
329 asm volatile(
330 "movq (%0), %%mm0 \n\t"
331 "movq (%0,%2), %%mm1 \n\t"
332 "movq (%1), %%mm2 \n\t"
333 "movq (%1,%2), %%mm3 \n\t"
334 H264_DEBLOCK_MASK(%3, %4)
335 "movq %%mm1, %%mm5 \n\t"
336 "movq %%mm2, %%mm6 \n\t"
337 H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
338 H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
339 "psubb %%mm5, %%mm1 \n\t"
340 "psubb %%mm6, %%mm2 \n\t"
341 "pand %%mm7, %%mm1 \n\t"
342 "pand %%mm7, %%mm2 \n\t"
343 "paddb %%mm5, %%mm1 \n\t"
344 "paddb %%mm6, %%mm2 \n\t"
345 "movq %%mm1, (%0,%2) \n\t"
346 "movq %%mm2, (%1) \n\t"
347 :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
348 "m"(alpha1), "m"(beta1), "m"(mm_bone)
349 );
350 }
351
352 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
353 {
354 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
355 }
356
357 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
358 {
359 //FIXME: could cut some load/stores by merging transpose with filter
360 uint8_t trans[8*4];
361 transpose4x4(trans, pix-2, 8, stride);
362 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
363 h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
364 transpose4x4(pix-2, trans, stride, 8);
365 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
366 }
367
368
369 /***********************************/
370 /* motion compensation */
371
372 #define QPEL_H264V(A,B,C,D,E,F,OP)\
373 "movd (%0), "#F" \n\t"\
374 "movq "#C", %%mm6 \n\t"\
375 "paddw "#D", %%mm6 \n\t"\
376 "psllw $2, %%mm6 \n\t"\
377 "psubw "#B", %%mm6 \n\t"\
378 "psubw "#E", %%mm6 \n\t"\
379 "pmullw %4, %%mm6 \n\t"\
380 "add %2, %0 \n\t"\
381 "punpcklbw %%mm7, "#F" \n\t"\
382 "paddw %5, "#A" \n\t"\
383 "paddw "#F", "#A" \n\t"\
384 "paddw "#A", %%mm6 \n\t"\
385 "psraw $5, %%mm6 \n\t"\
386 "packuswb %%mm6, %%mm6 \n\t"\
387 OP(%%mm6, (%1), A, d)\
388 "add %3, %1 \n\t"
389
390 #define QPEL_H264HV(A,B,C,D,E,F,OF)\
391 "movd (%0), "#F" \n\t"\
392 "movq "#C", %%mm6 \n\t"\
393 "paddw "#D", %%mm6 \n\t"\
394 "psllw $2, %%mm6 \n\t"\
395 "psubw "#B", %%mm6 \n\t"\
396 "psubw "#E", %%mm6 \n\t"\
397 "pmullw %3, %%mm6 \n\t"\
398 "add %2, %0 \n\t"\
399 "punpcklbw %%mm7, "#F" \n\t"\
400 "paddw "#F", "#A" \n\t"\
401 "paddw "#A", %%mm6 \n\t"\
402 "movq %%mm6, "#OF"(%1) \n\t"
403
404 #define QPEL_H264(OPNAME, OP, MMX)\
405 static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
406 int h=4;\
407 \
408 asm volatile(\
409 "pxor %%mm7, %%mm7 \n\t"\
410 "movq %5, %%mm4 \n\t"\
411 "movq %6, %%mm5 \n\t"\
412 "1: \n\t"\
413 "movd -1(%0), %%mm1 \n\t"\
414 "movd (%0), %%mm2 \n\t"\
415 "movd 1(%0), %%mm3 \n\t"\
416 "movd 2(%0), %%mm0 \n\t"\
417 "punpcklbw %%mm7, %%mm1 \n\t"\
418 "punpcklbw %%mm7, %%mm2 \n\t"\
419 "punpcklbw %%mm7, %%mm3 \n\t"\
420 "punpcklbw %%mm7, %%mm0 \n\t"\
421 "paddw %%mm0, %%mm1 \n\t"\
422 "paddw %%mm3, %%mm2 \n\t"\
423 "movd -2(%0), %%mm0 \n\t"\
424 "movd 3(%0), %%mm3 \n\t"\
425 "punpcklbw %%mm7, %%mm0 \n\t"\
426 "punpcklbw %%mm7, %%mm3 \n\t"\
427 "paddw %%mm3, %%mm0 \n\t"\
428 "psllw $2, %%mm2 \n\t"\
429 "psubw %%mm1, %%mm2 \n\t"\
430 "pmullw %%mm4, %%mm2 \n\t"\
431 "paddw %%mm5, %%mm0 \n\t"\
432 "paddw %%mm2, %%mm0 \n\t"\
433 "psraw $5, %%mm0 \n\t"\
434 "packuswb %%mm0, %%mm0 \n\t"\
435 OP(%%mm0, (%1),%%mm6, d)\
436 "add %3, %0 \n\t"\
437 "add %4, %1 \n\t"\
438 "decl %2 \n\t"\
439 " jnz 1b \n\t"\
440 : "+a"(src), "+c"(dst), "+m"(h)\
441 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
442 : "memory"\
443 );\
444 }\
445 static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
446 src -= 2*srcStride;\
447 asm volatile(\
448 "pxor %%mm7, %%mm7 \n\t"\
449 "movd (%0), %%mm0 \n\t"\
450 "add %2, %0 \n\t"\
451 "movd (%0), %%mm1 \n\t"\
452 "add %2, %0 \n\t"\
453 "movd (%0), %%mm2 \n\t"\
454 "add %2, %0 \n\t"\
455 "movd (%0), %%mm3 \n\t"\
456 "add %2, %0 \n\t"\
457 "movd (%0), %%mm4 \n\t"\
458 "add %2, %0 \n\t"\
459 "punpcklbw %%mm7, %%mm0 \n\t"\
460 "punpcklbw %%mm7, %%mm1 \n\t"\
461 "punpcklbw %%mm7, %%mm2 \n\t"\
462 "punpcklbw %%mm7, %%mm3 \n\t"\
463 "punpcklbw %%mm7, %%mm4 \n\t"\
464 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
465 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
466 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
467 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
468 \
469 : "+a"(src), "+c"(dst)\
470 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
471 : "memory"\
472 );\
473 }\
474 static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
475 int h=4;\
476 int w=3;\
477 src -= 2*srcStride+2;\
478 while(w--){\
479 asm volatile(\
480 "pxor %%mm7, %%mm7 \n\t"\
481 "movd (%0), %%mm0 \n\t"\
482 "add %2, %0 \n\t"\
483 "movd (%0), %%mm1 \n\t"\
484 "add %2, %0 \n\t"\
485 "movd (%0), %%mm2 \n\t"\
486 "add %2, %0 \n\t"\
487 "movd (%0), %%mm3 \n\t"\
488 "add %2, %0 \n\t"\
489 "movd (%0), %%mm4 \n\t"\
490 "add %2, %0 \n\t"\
491 "punpcklbw %%mm7, %%mm0 \n\t"\
492 "punpcklbw %%mm7, %%mm1 \n\t"\
493 "punpcklbw %%mm7, %%mm2 \n\t"\
494 "punpcklbw %%mm7, %%mm3 \n\t"\
495 "punpcklbw %%mm7, %%mm4 \n\t"\
496 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
497 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
498 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
499 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
500 \
501 : "+a"(src)\
502 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
503 : "memory"\
504 );\
505 tmp += 4;\
506 src += 4 - 9*srcStride;\
507 }\
508 tmp -= 3*4;\
509 asm volatile(\
510 "movq %4, %%mm6 \n\t"\
511 "1: \n\t"\
512 "movq (%0), %%mm0 \n\t"\
513 "paddw 10(%0), %%mm0 \n\t"\
514 "movq 2(%0), %%mm1 \n\t"\
515 "paddw 8(%0), %%mm1 \n\t"\
516 "movq 4(%0), %%mm2 \n\t"\
517 "paddw 6(%0), %%mm2 \n\t"\
518 "psubw %%mm1, %%mm0 \n\t"/*a-b (abccba)*/\
519 "psraw $2, %%mm0 \n\t"/*(a-b)/4 */\
520 "psubw %%mm1, %%mm0 \n\t"/*(a-b)/4-b */\
521 "paddsw %%mm2, %%mm0 \n\t"\
522 "psraw $2, %%mm0 \n\t"/*((a-b)/4-b)/4 */\
523 "paddw %%mm6, %%mm2 \n\t"\
524 "paddw %%mm2, %%mm0 \n\t"\
525 "psraw $6, %%mm0 \n\t"\
526 "packuswb %%mm0, %%mm0 \n\t"\
527 OP(%%mm0, (%1),%%mm7, d)\
528 "add $24, %0 \n\t"\
529 "add %3, %1 \n\t"\
530 "decl %2 \n\t"\
531 " jnz 1b \n\t"\
532 : "+a"(tmp), "+c"(dst), "+m"(h)\
533 : "S"((long)dstStride), "m"(ff_pw_32)\
534 : "memory"\
535 );\
536 }\
537 \
538 static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
539 int h=8;\
540 asm volatile(\
541 "pxor %%mm7, %%mm7 \n\t"\
542 "movq %5, %%mm6 \n\t"\
543 "1: \n\t"\
544 "movq (%0), %%mm0 \n\t"\
545 "movq 1(%0), %%mm2 \n\t"\
546 "movq %%mm0, %%mm1 \n\t"\
547 "movq %%mm2, %%mm3 \n\t"\
548 "punpcklbw %%mm7, %%mm0 \n\t"\
549 "punpckhbw %%mm7, %%mm1 \n\t"\
550 "punpcklbw %%mm7, %%mm2 \n\t"\
551 "punpckhbw %%mm7, %%mm3 \n\t"\
552 "paddw %%mm2, %%mm0 \n\t"\
553 "paddw %%mm3, %%mm1 \n\t"\
554 "psllw $2, %%mm0 \n\t"\
555 "psllw $2, %%mm1 \n\t"\
556 "movq -1(%0), %%mm2 \n\t"\
557 "movq 2(%0), %%mm4 \n\t"\
558 "movq %%mm2, %%mm3 \n\t"\
559 "movq %%mm4, %%mm5 \n\t"\
560 "punpcklbw %%mm7, %%mm2 \n\t"\
561 "punpckhbw %%mm7, %%mm3 \n\t"\
562 "punpcklbw %%mm7, %%mm4 \n\t"\
563 "punpckhbw %%mm7, %%mm5 \n\t"\
564 "paddw %%mm4, %%mm2 \n\t"\
565 "paddw %%mm3, %%mm5 \n\t"\
566 "psubw %%mm2, %%mm0 \n\t"\
567 "psubw %%mm5, %%mm1 \n\t"\
568 "pmullw %%mm6, %%mm0 \n\t"\
569 "pmullw %%mm6, %%mm1 \n\t"\
570 "movd -2(%0), %%mm2 \n\t"\
571 "movd 7(%0), %%mm5 \n\t"\
572 "punpcklbw %%mm7, %%mm2 \n\t"\
573 "punpcklbw %%mm7, %%mm5 \n\t"\
574 "paddw %%mm3, %%mm2 \n\t"\
575 "paddw %%mm5, %%mm4 \n\t"\
576 "movq %6, %%mm5 \n\t"\
577 "paddw %%mm5, %%mm2 \n\t"\
578 "paddw %%mm5, %%mm4 \n\t"\
579 "paddw %%mm2, %%mm0 \n\t"\
580 "paddw %%mm4, %%mm1 \n\t"\
581 "psraw $5, %%mm0 \n\t"\
582 "psraw $5, %%mm1 \n\t"\
583 "packuswb %%mm1, %%mm0 \n\t"\
584 OP(%%mm0, (%1),%%mm5, q)\
585 "add %3, %0 \n\t"\
586 "add %4, %1 \n\t"\
587 "decl %2 \n\t"\
588 " jnz 1b \n\t"\
589 : "+a"(src), "+c"(dst), "+m"(h)\
590 : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
591 : "memory"\
592 );\
593 }\
594 \
595 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
596 int h= 2;\
597 src -= 2*srcStride;\
598 \
599 while(h--){\
600 asm volatile(\
601 "pxor %%mm7, %%mm7 \n\t"\
602 "movd (%0), %%mm0 \n\t"\
603 "add %2, %0 \n\t"\
604 "movd (%0), %%mm1 \n\t"\
605 "add %2, %0 \n\t"\
606 "movd (%0), %%mm2 \n\t"\
607 "add %2, %0 \n\t"\
608 "movd (%0), %%mm3 \n\t"\
609 "add %2, %0 \n\t"\
610 "movd (%0), %%mm4 \n\t"\
611 "add %2, %0 \n\t"\
612 "punpcklbw %%mm7, %%mm0 \n\t"\
613 "punpcklbw %%mm7, %%mm1 \n\t"\
614 "punpcklbw %%mm7, %%mm2 \n\t"\
615 "punpcklbw %%mm7, %%mm3 \n\t"\
616 "punpcklbw %%mm7, %%mm4 \n\t"\
617 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
618 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
619 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
620 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
621 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
622 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
623 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
624 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
625 \
626 : "+a"(src), "+c"(dst)\
627 : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
628 : "memory"\
629 );\
630 src += 4-13*srcStride;\
631 dst += 4-8*dstStride;\
632 }\
633 }\
634 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
635 int h=8;\
636 int w=4;\
637 src -= 2*srcStride+2;\
638 while(w--){\
639 asm volatile(\
640 "pxor %%mm7, %%mm7 \n\t"\
641 "movd (%0), %%mm0 \n\t"\
642 "add %2, %0 \n\t"\
643 "movd (%0), %%mm1 \n\t"\
644 "add %2, %0 \n\t"\
645 "movd (%0), %%mm2 \n\t"\
646 "add %2, %0 \n\t"\
647 "movd (%0), %%mm3 \n\t"\
648 "add %2, %0 \n\t"\
649 "movd (%0), %%mm4 \n\t"\
650 "add %2, %0 \n\t"\
651 "punpcklbw %%mm7, %%mm0 \n\t"\
652 "punpcklbw %%mm7, %%mm1 \n\t"\
653 "punpcklbw %%mm7, %%mm2 \n\t"\
654 "punpcklbw %%mm7, %%mm3 \n\t"\
655 "punpcklbw %%mm7, %%mm4 \n\t"\
656 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*4)\
657 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*4)\
658 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*4)\
659 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*4)\
660 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*8*4)\
661 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*8*4)\
662 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*8*4)\
663 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*8*4)\
664 \
665 : "+a"(src)\
666 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
667 : "memory"\
668 );\
669 tmp += 4;\
670 src += 4 - 13*srcStride;\
671 }\
672 tmp -= 4*4;\
673 asm volatile(\
674 "movq %4, %%mm6 \n\t"\
675 "1: \n\t"\
676 "movq (%0), %%mm0 \n\t"\
677 "movq 8(%0), %%mm3 \n\t"\
678 "movq 2(%0), %%mm1 \n\t"\
679 "movq 10(%0), %%mm4 \n\t"\
680 "paddw %%mm4, %%mm0 \n\t"\
681 "paddw %%mm3, %%mm1 \n\t"\
682 "paddw 18(%0), %%mm3 \n\t"\
683 "paddw 16(%0), %%mm4 \n\t"\
684 "movq 4(%0), %%mm2 \n\t"\
685 "movq 12(%0), %%mm5 \n\t"\
686 "paddw 6(%0), %%mm2 \n\t"\
687 "paddw 14(%0), %%mm5 \n\t"\
688 "psubw %%mm1, %%mm0 \n\t"\
689 "psubw %%mm4, %%mm3 \n\t"\
690 "psraw $2, %%mm0 \n\t"\
691 "psraw $2, %%mm3 \n\t"\
692 "psubw %%mm1, %%mm0 \n\t"\
693 "psubw %%mm4, %%mm3 \n\t"\
694 "paddsw %%mm2, %%mm0 \n\t"\
695 "paddsw %%mm5, %%mm3 \n\t"\
696 "psraw $2, %%mm0 \n\t"\
697 "psraw $2, %%mm3 \n\t"\
698 "paddw %%mm6, %%mm2 \n\t"\
699 "paddw %%mm6, %%mm5 \n\t"\
700 "paddw %%mm2, %%mm0 \n\t"\
701 "paddw %%mm5, %%mm3 \n\t"\
702 "psraw $6, %%mm0 \n\t"\
703 "psraw $6, %%mm3 \n\t"\
704 "packuswb %%mm3, %%mm0 \n\t"\
705 OP(%%mm0, (%1),%%mm7, q)\
706 "add $32, %0 \n\t"\
707 "add %3, %1 \n\t"\
708 "decl %2 \n\t"\
709 " jnz 1b \n\t"\
710 : "+a"(tmp), "+c"(dst), "+m"(h)\
711 : "S"((long)dstStride), "m"(ff_pw_32)\
712 : "memory"\
713 );\
714 }\
715 static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
716 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
717 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
718 src += 8*srcStride;\
719 dst += 8*dstStride;\
720 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
721 OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
722 }\
723 \
724 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
725 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
726 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
727 src += 8*srcStride;\
728 dst += 8*dstStride;\
729 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
730 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
731 }\
732 \
733 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
734 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride);\
735 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp , src+8, dstStride, tmpStride, srcStride);\
736 src += 8*srcStride;\
737 dst += 8*dstStride;\
738 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride);\
739 OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp , src+8, dstStride, tmpStride, srcStride);\
740 }\
741
742 #define H264_MC(OPNAME, SIZE, MMX) \
743 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
744 OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
745 }\
746 \
747 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
748 uint64_t temp[SIZE*SIZE/8];\
749 uint8_t * const half= (uint8_t*)temp;\
750 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
751 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
752 }\
753 \
754 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
755 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
756 }\
757 \
758 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
759 uint64_t temp[SIZE*SIZE/8];\
760 uint8_t * const half= (uint8_t*)temp;\
761 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\
762 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+1, half, stride, stride, SIZE);\
763 }\
764 \
765 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## 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 ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
769 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
770 }\
771 \
772 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
773 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
774 }\
775 \
776 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## 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 ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
780 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
781 }\
782 \
783 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
784 uint64_t temp[SIZE*SIZE/4];\
785 uint8_t * const halfH= (uint8_t*)temp;\
786 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
787 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
788 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
789 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
790 }\
791 \
792 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
793 uint64_t temp[SIZE*SIZE/4];\
794 uint8_t * const halfH= (uint8_t*)temp;\
795 uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
796 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
797 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
798 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\
799 }\
800 \
801 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## 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 + stride, 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);\
808 }\
809 \
810 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## 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 + stride, 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);\
817 }\
818 \
819 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
820 uint64_t temp[SIZE*(SIZE+8)/4];\
821 int16_t * const tmp= (int16_t*)temp;\
822 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
823 }\
824 \
825 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
826 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
827 uint8_t * const halfH= (uint8_t*)temp;\
828 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
829 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
830 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\
831 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
832 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
833 }\
834 \
835 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
836 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
837 uint8_t * const halfH= (uint8_t*)temp;\
838 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
839 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
840 put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\
841 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
842 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\
843 }\
844 \
845 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
846 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
847 uint8_t * const halfV= (uint8_t*)temp;\
848 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
849 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
850 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
851 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
852 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
853 }\
854 \
855 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
856 uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\
857 uint8_t * const halfV= (uint8_t*)temp;\
858 uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
859 int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
860 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
861 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
862 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\
863 }\
864
865
866 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
867 #define AVG_3DNOW_OP(a,b,temp, size) \
868 "mov" #size " " #b ", " #temp " \n\t"\
869 "pavgusb " #temp ", " #a " \n\t"\
870 "mov" #size " " #a ", " #b " \n\t"
871 #define AVG_MMX2_OP(a,b,temp, size) \
872 "mov" #size " " #b ", " #temp " \n\t"\
873 "pavgb " #temp ", " #a " \n\t"\
874 "mov" #size " " #a ", " #b " \n\t"
875
876 QPEL_H264(put_, PUT_OP, 3dnow)
877 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
878 QPEL_H264(put_, PUT_OP, mmx2)
879 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
880
881 H264_MC(put_, 4, 3dnow)
882 H264_MC(put_, 8, 3dnow)
883 H264_MC(put_, 16,3dnow)
884 H264_MC(avg_, 4, 3dnow)
885 H264_MC(avg_, 8, 3dnow)
886 H264_MC(avg_, 16,3dnow)
887 H264_MC(put_, 4, mmx2)
888 H264_MC(put_, 8, mmx2)
889 H264_MC(put_, 16,mmx2)
890 H264_MC(avg_, 4, mmx2)
891 H264_MC(avg_, 8, mmx2)
892 H264_MC(avg_, 16,mmx2)
893
894
895 #define H264_CHROMA_OP(S,D)
896 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
897 #include "dsputil_h264_template_mmx.c"
898 #undef H264_CHROMA_OP
899 #undef H264_CHROMA_MC8_TMPL
900
901 #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
902 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
903 #include "dsputil_h264_template_mmx.c"
904 #undef H264_CHROMA_OP
905 #undef H264_CHROMA_MC8_TMPL
906
907 #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
908 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
909 #include "dsputil_h264_template_mmx.c"
910 #undef H264_CHROMA_OP
911 #undef H264_CHROMA_MC8_TMPL
912