?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
clarify *_offsets.asm differences
[libvpx.git] / vp8 / common / x86 / loopfilter_mmx.asm
blobc6c215c3c6fcaef13c4328534d58fec0927aa16c
1 ;
2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 ;
4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree.
9 ;
10
11
12 %include "vpx_ports/x86_abi_support.asm"
13
14
15 ;void vp8_loop_filter_horizontal_edge_mmx
16 ;(
17 ; unsigned char *src_ptr,
18 ; int src_pixel_step,
19 ; const char *flimit,
20 ; const char *limit,
21 ; const char *thresh,
22 ; int count
23 ;)
24 global sym(vp8_loop_filter_horizontal_edge_mmx)
25 sym(vp8_loop_filter_horizontal_edge_mmx):
26 push rbp
27 mov rbp, rsp
28 SHADOW_ARGS_TO_STACK 6
29 GET_GOT rbx
30 push rsi
31 push rdi
32 ; end prolog
33
34 ALIGN_STACK 16, rax
35 sub rsp, 32 ; reserve 32 bytes
36 %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
37 %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
38
39 mov rsi, arg(0) ;src_ptr
40 movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
41
42 movsxd rcx, dword ptr arg(5) ;count
43 next8_h:
44 mov rdx, arg(3) ;limit
45 movq mm7, [rdx]
46 mov rdi, rsi ; rdi points to row +1 for indirect addressing
47 add rdi, rax
48
49 ; calculate breakout conditions
50 movq mm2, [rdi+2*rax] ; q3
51 movq mm1, [rsi+2*rax] ; q2
52 movq mm6, mm1 ; q2
53 psubusb mm1, mm2 ; q2-=q3
54 psubusb mm2, mm6 ; q3-=q2
55 por mm1, mm2 ; abs(q3-q2)
56 psubusb mm1, mm7 ;
57
58
59 movq mm4, [rsi+rax] ; q1
60 movq mm3, mm4 ; q1
61 psubusb mm4, mm6 ; q1-=q2
62 psubusb mm6, mm3 ; q2-=q1
63 por mm4, mm6 ; abs(q2-q1)
64
65 psubusb mm4, mm7
66 por mm1, mm4
67
68 movq mm4, [rsi] ; q0
69 movq mm0, mm4 ; q0
70 psubusb mm4, mm3 ; q0-=q1
71 psubusb mm3, mm0 ; q1-=q0
72 por mm4, mm3 ; abs(q0-q1)
73 movq t0, mm4 ; save to t0
74 psubusb mm4, mm7
75 por mm1, mm4
76
77
78 neg rax ; negate pitch to deal with above border
79
80 movq mm2, [rsi+4*rax] ; p3
81 movq mm4, [rdi+4*rax] ; p2
82 movq mm5, mm4 ; p2
83 psubusb mm4, mm2 ; p2-=p3
84 psubusb mm2, mm5 ; p3-=p2
85 por mm4, mm2 ; abs(p3 - p2)
86 psubusb mm4, mm7
87 por mm1, mm4
88
89
90 movq mm4, [rsi+2*rax] ; p1
91 movq mm3, mm4 ; p1
92 psubusb mm4, mm5 ; p1-=p2
93 psubusb mm5, mm3 ; p2-=p1
94 por mm4, mm5 ; abs(p2 - p1)
95 psubusb mm4, mm7
96 por mm1, mm4
97
98 movq mm2, mm3 ; p1
99
100 movq mm4, [rsi+rax] ; p0
101 movq mm5, mm4 ; p0
102 psubusb mm4, mm3 ; p0-=p1
103 psubusb mm3, mm5 ; p1-=p0
104 por mm4, mm3 ; abs(p1 - p0)
105 movq t1, mm4 ; save to t1
106 psubusb mm4, mm7
107 por mm1, mm4
108
109 movq mm3, [rdi] ; q1
110 movq mm4, mm3 ; q1
111 psubusb mm3, mm2 ; q1-=p1
112 psubusb mm2, mm4 ; p1-=q1
113 por mm2, mm3 ; abs(p1-q1)
114 pand mm2, [GLOBAL(tfe)] ; set lsb of each byte to zero
115 psrlw mm2, 1 ; abs(p1-q1)/2
116
117 movq mm6, mm5 ; p0
118 movq mm3, [rsi] ; q0
119 psubusb mm5, mm3 ; p0-=q0
120 psubusb mm3, mm6 ; q0-=p0
121 por mm5, mm3 ; abs(p0 - q0)
122 paddusb mm5, mm5 ; abs(p0-q0)*2
123 paddusb mm5, mm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2
124
125 mov rdx, arg(2) ;flimit ; get flimit
126 movq mm2, [rdx] ; flimit mm2
127 paddb mm2, mm2 ; flimit*2 (less than 255)
128 paddb mm7, mm2 ; flimit * 2 + limit (less than 255)
129
130 psubusb mm5, mm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > flimit * 2 + limit
131 por mm1, mm5
132 pxor mm5, mm5
133 pcmpeqb mm1, mm5 ; mask mm1
134
135 ; calculate high edge variance
136 mov rdx, arg(4) ;thresh ; get thresh
137 movq mm7, [rdx] ;
138 movq mm4, t0 ; get abs (q1 - q0)
139 psubusb mm4, mm7
140 movq mm3, t1 ; get abs (p1 - p0)
141 psubusb mm3, mm7
142 paddb mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
143
144 pcmpeqb mm4, mm5
145
146 pcmpeqb mm5, mm5
147 pxor mm4, mm5
148
149
150 ; start work on filters
151 movq mm2, [rsi+2*rax] ; p1
152 movq mm7, [rdi] ; q1
153 pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
154 pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
155 psubsb mm2, mm7 ; p1 - q1
156 pand mm2, mm4 ; high var mask (hvm)(p1 - q1)
157 pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
158 pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
159 movq mm3, mm0 ; q0
160 psubsb mm0, mm6 ; q0 - p0
161 paddsb mm2, mm0 ; 1 * (q0 - p0) + hvm(p1 - q1)
162 paddsb mm2, mm0 ; 2 * (q0 - p0) + hvm(p1 - q1)
163 paddsb mm2, mm0 ; 3 * (q0 - p0) + hvm(p1 - q1)
164 pand mm1, mm2 ; mask filter values we don't care about
165 movq mm2, mm1
166 paddsb mm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4
167 paddsb mm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3
168
169 pxor mm0, mm0 ;
170 pxor mm5, mm5
171 punpcklbw mm0, mm2 ;
172 punpckhbw mm5, mm2 ;
173 psraw mm0, 11 ;
174 psraw mm5, 11
175 packsswb mm0, mm5
176 movq mm2, mm0 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
177
178 pxor mm0, mm0 ; 0
179 movq mm5, mm1 ; abcdefgh
180 punpcklbw mm0, mm1 ; e0f0g0h0
181 psraw mm0, 11 ; sign extended shift right by 3
182 pxor mm1, mm1 ; 0
183 punpckhbw mm1, mm5 ; a0b0c0d0
184 psraw mm1, 11 ; sign extended shift right by 3
185 movq mm5, mm0 ; save results
186
187 packsswb mm0, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
188 paddsw mm5, [GLOBAL(ones)]
189 paddsw mm1, [GLOBAL(ones)]
190 psraw mm5, 1 ; partial shifted one more time for 2nd tap
191 psraw mm1, 1 ; partial shifted one more time for 2nd tap
192 packsswb mm5, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
193 pandn mm4, mm5 ; high edge variance additive
194
195 paddsb mm6, mm2 ; p0+= p0 add
196 pxor mm6, [GLOBAL(t80)] ; unoffset
197 movq [rsi+rax], mm6 ; write back
198
199 movq mm6, [rsi+2*rax] ; p1
200 pxor mm6, [GLOBAL(t80)] ; reoffset
201 paddsb mm6, mm4 ; p1+= p1 add
202 pxor mm6, [GLOBAL(t80)] ; unoffset
203 movq [rsi+2*rax], mm6 ; write back
204
205 psubsb mm3, mm0 ; q0-= q0 add
206 pxor mm3, [GLOBAL(t80)] ; unoffset
207 movq [rsi], mm3 ; write back
208
209 psubsb mm7, mm4 ; q1-= q1 add
210 pxor mm7, [GLOBAL(t80)] ; unoffset
211 movq [rdi], mm7 ; write back
212
213 add rsi,8
214 neg rax
215 dec rcx
216 jnz next8_h
217
218 add rsp, 32
219 pop rsp
220 ; begin epilog
221 pop rdi
222 pop rsi
223 RESTORE_GOT
224 UNSHADOW_ARGS
225 pop rbp
226 ret
227
228
229 ;void vp8_loop_filter_vertical_edge_mmx
230 ;(
231 ; unsigned char *src_ptr,
232 ; int src_pixel_step,
233 ; const char *flimit,
234 ; const char *limit,
235 ; const char *thresh,
236 ; int count
237 ;)
238 global sym(vp8_loop_filter_vertical_edge_mmx)
239 sym(vp8_loop_filter_vertical_edge_mmx):
240 push rbp
241 mov rbp, rsp
242 SHADOW_ARGS_TO_STACK 6
243 GET_GOT rbx
244 push rsi
245 push rdi
246 ; end prolog
247
248 ALIGN_STACK 16, rax
249 sub rsp, 64 ; reserve 64 bytes
250 %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
251 %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
252 %define srct [rsp + 32] ;__declspec(align(16)) char srct[32];
253
254 mov rsi, arg(0) ;src_ptr
255 movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
256
257 lea rsi, [rsi + rax*4 - 4]
258
259 movsxd rcx, dword ptr arg(5) ;count
260 next8_v:
261 mov rdi, rsi ; rdi points to row +1 for indirect addressing
262 add rdi, rax
263
264
265 ;transpose
266 movq mm6, [rsi+2*rax] ; 67 66 65 64 63 62 61 60
267 movq mm7, mm6 ; 77 76 75 74 73 72 71 70
268
269 punpckhbw mm7, [rdi+2*rax] ; 77 67 76 66 75 65 74 64
270 punpcklbw mm6, [rdi+2*rax] ; 73 63 72 62 71 61 70 60
271
272 movq mm4, [rsi] ; 47 46 45 44 43 42 41 40
273 movq mm5, mm4 ; 47 46 45 44 43 42 41 40
274
275 punpckhbw mm5, [rsi+rax] ; 57 47 56 46 55 45 54 44
276 punpcklbw mm4, [rsi+rax] ; 53 43 52 42 51 41 50 40
277
278 movq mm3, mm5 ; 57 47 56 46 55 45 54 44
279 punpckhwd mm5, mm7 ; 77 67 57 47 76 66 56 46
280
281 punpcklwd mm3, mm7 ; 75 65 55 45 74 64 54 44
282 movq mm2, mm4 ; 53 43 52 42 51 41 50 40
283
284 punpckhwd mm4, mm6 ; 73 63 53 43 72 62 52 42
285 punpcklwd mm2, mm6 ; 71 61 51 41 70 60 50 40
286
287 neg rax
288 movq mm6, [rsi+rax*2] ; 27 26 25 24 23 22 21 20
289
290 movq mm1, mm6 ; 27 26 25 24 23 22 21 20
291 punpckhbw mm6, [rsi+rax] ; 37 27 36 36 35 25 34 24
292
293 punpcklbw mm1, [rsi+rax] ; 33 23 32 22 31 21 30 20
294 movq mm7, [rsi+rax*4]; ; 07 06 05 04 03 02 01 00
295
296 punpckhbw mm7, [rdi+rax*4] ; 17 07 16 06 15 05 14 04
297 movq mm0, mm7 ; 17 07 16 06 15 05 14 04
298
299 punpckhwd mm7, mm6 ; 37 27 17 07 36 26 16 06
300 punpcklwd mm0, mm6 ; 35 25 15 05 34 24 14 04
301
302 movq mm6, mm7 ; 37 27 17 07 36 26 16 06
303 punpckhdq mm7, mm5 ; 77 67 57 47 37 27 17 07 = q3
304
305 punpckldq mm6, mm5 ; 76 66 56 46 36 26 16 06 = q2
306
307 movq mm5, mm6 ; 76 66 56 46 36 26 16 06
308 psubusb mm5, mm7 ; q2-q3
309
310 psubusb mm7, mm6 ; q3-q2
311 por mm7, mm5; ; mm7=abs (q3-q2)
312
313 movq mm5, mm0 ; 35 25 15 05 34 24 14 04
314 punpckhdq mm5, mm3 ; 75 65 55 45 35 25 15 05 = q1
315
316 punpckldq mm0, mm3 ; 74 64 54 44 34 24 15 04 = q0
317 movq mm3, mm5 ; 75 65 55 45 35 25 15 05 = q1
318
319 psubusb mm3, mm6 ; q1-q2
320 psubusb mm6, mm5 ; q2-q1
321
322 por mm6, mm3 ; mm6=abs(q2-q1)
323 lea rdx, srct
324
325 movq [rdx+24], mm5 ; save q1
326 movq [rdx+16], mm0 ; save q0
327
328 movq mm3, [rsi+rax*4] ; 07 06 05 04 03 02 01 00
329 punpcklbw mm3, [rdi+rax*4] ; 13 03 12 02 11 01 10 00
330
331 movq mm0, mm3 ; 13 03 12 02 11 01 10 00
332 punpcklwd mm0, mm1 ; 31 21 11 01 30 20 10 00
333
334 punpckhwd mm3, mm1 ; 33 23 13 03 32 22 12 02
335 movq mm1, mm0 ; 31 21 11 01 30 20 10 00
336
337 punpckldq mm0, mm2 ; 70 60 50 40 30 20 10 00 =p3
338 punpckhdq mm1, mm2 ; 71 61 51 41 31 21 11 01 =p2
339
340 movq mm2, mm1 ; 71 61 51 41 31 21 11 01 =p2
341 psubusb mm2, mm0 ; p2-p3
342
343 psubusb mm0, mm1 ; p3-p2
344 por mm0, mm2 ; mm0=abs(p3-p2)
345
346 movq mm2, mm3 ; 33 23 13 03 32 22 12 02
347 punpckldq mm2, mm4 ; 72 62 52 42 32 22 12 02 = p1
348
349 punpckhdq mm3, mm4 ; 73 63 53 43 33 23 13 03 = p0
350 movq [rdx+8], mm3 ; save p0
351
352 movq [rdx], mm2 ; save p1
353 movq mm5, mm2 ; mm5 = p1
354
355 psubusb mm2, mm1 ; p1-p2
356 psubusb mm1, mm5 ; p2-p1
357
358 por mm1, mm2 ; mm1=abs(p2-p1)
359 mov rdx, arg(3) ;limit
360
361 movq mm4, [rdx] ; mm4 = limit
362 psubusb mm7, mm4
363
364 psubusb mm0, mm4
365 psubusb mm1, mm4
366
367 psubusb mm6, mm4
368 por mm7, mm6
369
370 por mm0, mm1
371 por mm0, mm7 ; abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit
372
373 movq mm1, mm5 ; p1
374
375 movq mm7, mm3 ; mm3=mm7=p0
376 psubusb mm7, mm5 ; p0 - p1
377
378 psubusb mm5, mm3 ; p1 - p0
379 por mm5, mm7 ; abs(p1-p0)
380
381 movq t0, mm5 ; save abs(p1-p0)
382 lea rdx, srct
383
384 psubusb mm5, mm4
385 por mm0, mm5 ; mm0=mask
386
387 movq mm5, [rdx+16] ; mm5=q0
388 movq mm7, [rdx+24] ; mm7=q1
389
390 movq mm6, mm5 ; mm6=q0
391 movq mm2, mm7 ; q1
392 psubusb mm5, mm7 ; q0-q1
393
394 psubusb mm7, mm6 ; q1-q0
395 por mm7, mm5 ; abs(q1-q0)
396
397 movq t1, mm7 ; save abs(q1-q0)
398 psubusb mm7, mm4
399
400 por mm0, mm7 ; mask
401
402 movq mm5, mm2 ; q1
403 psubusb mm5, mm1 ; q1-=p1
404 psubusb mm1, mm2 ; p1-=q1
405 por mm5, mm1 ; abs(p1-q1)
406 pand mm5, [GLOBAL(tfe)] ; set lsb of each byte to zero
407 psrlw mm5, 1 ; abs(p1-q1)/2
408
409 mov rdx, arg(2) ;flimit ;
410
411 movq mm2, [rdx] ;flimit mm2
412 movq mm1, mm3 ; mm1=mm3=p0
413
414 movq mm7, mm6 ; mm7=mm6=q0
415 psubusb mm1, mm7 ; p0-q0
416
417 psubusb mm7, mm3 ; q0-p0
418 por mm1, mm7 ; abs(q0-p0)
419 paddusb mm1, mm1 ; abs(q0-p0)*2
420 paddusb mm1, mm5 ; abs (p0 - q0) *2 + abs(p1-q1)/2
421
422 paddb mm2, mm2 ; flimit*2 (less than 255)
423 paddb mm4, mm2 ; flimit * 2 + limit (less than 255)
424
425 psubusb mm1, mm4 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > flimit * 2 + limit
426 por mm1, mm0; ; mask
427
428 pxor mm0, mm0
429 pcmpeqb mm1, mm0
430
431 ; calculate high edge variance
432 mov rdx, arg(4) ;thresh ; get thresh
433 movq mm7, [rdx]
434 ;
435 movq mm4, t0 ; get abs (q1 - q0)
436 psubusb mm4, mm7
437
438 movq mm3, t1 ; get abs (p1 - p0)
439 psubusb mm3, mm7
440
441 por mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
442 pcmpeqb mm4, mm0
443
444 pcmpeqb mm0, mm0
445 pxor mm4, mm0
446
447
448
449 ; start work on filters
450 lea rdx, srct
451
452 movq mm2, [rdx] ; p1
453 movq mm7, [rdx+24] ; q1
454
455 movq mm6, [rdx+8] ; p0
456 movq mm0, [rdx+16] ; q0
457
458 pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
459 pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
460
461 psubsb mm2, mm7 ; p1 - q1
462 pand mm2, mm4 ; high var mask (hvm)(p1 - q1)
463
464 pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
465 pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
466
467 movq mm3, mm0 ; q0
468 psubsb mm0, mm6 ; q0 - p0
469
470 paddsb mm2, mm0 ; 1 * (q0 - p0) + hvm(p1 - q1)
471 paddsb mm2, mm0 ; 2 * (q0 - p0) + hvm(p1 - q1)
472
473 paddsb mm2, mm0 ; 3 * (q0 - p0) + hvm(p1 - q1)
474 pand mm1, mm2 ; mask filter values we don't care about
475
476 movq mm2, mm1
477 paddsb mm1, [GLOBAL(t4)] ; 3* (q0 - p0) + hvm(p1 - q1) + 4
478
479 paddsb mm2, [GLOBAL(t3)] ; 3* (q0 - p0) + hvm(p1 - q1) + 3
480 pxor mm0, mm0 ;
481
482 pxor mm5, mm5
483 punpcklbw mm0, mm2 ;
484
485 punpckhbw mm5, mm2 ;
486 psraw mm0, 11 ;
487
488 psraw mm5, 11
489 packsswb mm0, mm5
490
491 movq mm2, mm0 ; (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
492
493 pxor mm0, mm0 ; 0
494 movq mm5, mm1 ; abcdefgh
495
496 punpcklbw mm0, mm1 ; e0f0g0h0
497 psraw mm0, 11 ; sign extended shift right by 3
498
499 pxor mm1, mm1 ; 0
500 punpckhbw mm1, mm5 ; a0b0c0d0
501
502 psraw mm1, 11 ; sign extended shift right by 3
503 movq mm5, mm0 ; save results
504
505 packsswb mm0, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
506 paddsw mm5, [GLOBAL(ones)]
507
508 paddsw mm1, [GLOBAL(ones)]
509 psraw mm5, 1 ; partial shifted one more time for 2nd tap
510
511 psraw mm1, 1 ; partial shifted one more time for 2nd tap
512 packsswb mm5, mm1 ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
513
514 pandn mm4, mm5 ; high edge variance additive
515
516 paddsb mm6, mm2 ; p0+= p0 add
517 pxor mm6, [GLOBAL(t80)] ; unoffset
518
519 ; mm6=p0 ;
520 movq mm1, [rdx] ; p1
521 pxor mm1, [GLOBAL(t80)] ; reoffset
522
523 paddsb mm1, mm4 ; p1+= p1 add
524 pxor mm1, [GLOBAL(t80)] ; unoffset
525 ; mm6 = p0 mm1 = p1
526
527 psubsb mm3, mm0 ; q0-= q0 add
528 pxor mm3, [GLOBAL(t80)] ; unoffset
529
530 ; mm3 = q0
531 psubsb mm7, mm4 ; q1-= q1 add
532 pxor mm7, [GLOBAL(t80)] ; unoffset
533 ; mm7 = q1
534
535 ; tranpose and write back
536 ; mm1 = 72 62 52 42 32 22 12 02
537 ; mm6 = 73 63 53 43 33 23 13 03
538 ; mm3 = 74 64 54 44 34 24 14 04
539 ; mm7 = 75 65 55 45 35 25 15 05
540
541 movq mm2, mm1 ; 72 62 52 42 32 22 12 02
542 punpcklbw mm2, mm6 ; 33 32 23 22 13 12 03 02
543
544 movq mm4, mm3 ; 74 64 54 44 34 24 14 04
545 punpckhbw mm1, mm6 ; 73 72 63 62 53 52 43 42
546
547 punpcklbw mm4, mm7 ; 35 34 25 24 15 14 05 04
548 punpckhbw mm3, mm7 ; 75 74 65 64 55 54 45 44
549
550 movq mm6, mm2 ; 33 32 23 22 13 12 03 02
551 punpcklwd mm2, mm4 ; 15 14 13 12 05 04 03 02
552
553 punpckhwd mm6, mm4 ; 35 34 33 32 25 24 23 22
554 movq mm5, mm1 ; 73 72 63 62 53 52 43 42
555
556 punpcklwd mm1, mm3 ; 55 54 53 52 45 44 43 42
557 punpckhwd mm5, mm3 ; 75 74 73 72 65 64 63 62
558
559
560 ; mm2 = 15 14 13 12 05 04 03 02
561 ; mm6 = 35 34 33 32 25 24 23 22
562 ; mm5 = 55 54 53 52 45 44 43 42
563 ; mm1 = 75 74 73 72 65 64 63 62
564
565
566
567 movd [rsi+rax*4+2], mm2
568 psrlq mm2, 32
569
570 movd [rdi+rax*4+2], mm2
571 movd [rsi+rax*2+2], mm6
572
573 psrlq mm6, 32
574 movd [rsi+rax+2],mm6
575
576 movd [rsi+2], mm1
577 psrlq mm1, 32
578
579 movd [rdi+2], mm1
580 neg rax
581
582 movd [rdi+rax+2],mm5
583 psrlq mm5, 32
584
585 movd [rdi+rax*2+2], mm5
586
587 lea rsi, [rsi+rax*8]
588 dec rcx
589 jnz next8_v
590
591 add rsp, 64
592 pop rsp
593 ; begin epilog
594 pop rdi
595 pop rsi
596 RESTORE_GOT
597 UNSHADOW_ARGS
598 pop rbp
599 ret
600
601
602 ;void vp8_mbloop_filter_horizontal_edge_mmx
603 ;(
604 ; unsigned char *src_ptr,
605 ; int src_pixel_step,
606 ; const char *flimit,
607 ; const char *limit,
608 ; const char *thresh,
609 ; int count
610 ;)
611 global sym(vp8_mbloop_filter_horizontal_edge_mmx)
612 sym(vp8_mbloop_filter_horizontal_edge_mmx):
613 push rbp
614 mov rbp, rsp
615 SHADOW_ARGS_TO_STACK 6
616 GET_GOT rbx
617 push rsi
618 push rdi
619 ; end prolog
620
621 ALIGN_STACK 16, rax
622 sub rsp, 32 ; reserve 32 bytes
623 %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
624 %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
625
626 mov rsi, arg(0) ;src_ptr
627 movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
628
629 movsxd rcx, dword ptr arg(5) ;count
630 next8_mbh:
631 mov rdx, arg(3) ;limit
632 movq mm7, [rdx]
633 mov rdi, rsi ; rdi points to row +1 for indirect addressing
634 add rdi, rax
635
636 ; calculate breakout conditions
637 movq mm2, [rdi+2*rax] ; q3
638
639 movq mm1, [rsi+2*rax] ; q2
640 movq mm6, mm1 ; q2
641 psubusb mm1, mm2 ; q2-=q3
642 psubusb mm2, mm6 ; q3-=q2
643 por mm1, mm2 ; abs(q3-q2)
644 psubusb mm1, mm7
645
646
647 ; mm1 = abs(q3-q2), mm6 =q2, mm7 = limit
648 movq mm4, [rsi+rax] ; q1
649 movq mm3, mm4 ; q1
650 psubusb mm4, mm6 ; q1-=q2
651 psubusb mm6, mm3 ; q2-=q1
652 por mm4, mm6 ; abs(q2-q1)
653 psubusb mm4, mm7
654 por mm1, mm4
655
656
657 ; mm1 = mask, mm3=q1, mm7 = limit
658
659 movq mm4, [rsi] ; q0
660 movq mm0, mm4 ; q0
661 psubusb mm4, mm3 ; q0-=q1
662 psubusb mm3, mm0 ; q1-=q0
663 por mm4, mm3 ; abs(q0-q1)
664 movq t0, mm4 ; save to t0
665 psubusb mm4, mm7
666 por mm1, mm4
667
668
669 ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1)
670
671 neg rax ; negate pitch to deal with above border
672
673 movq mm2, [rsi+4*rax] ; p3
674 movq mm4, [rdi+4*rax] ; p2
675 movq mm5, mm4 ; p2
676 psubusb mm4, mm2 ; p2-=p3
677 psubusb mm2, mm5 ; p3-=p2
678 por mm4, mm2 ; abs(p3 - p2)
679 psubusb mm4, mm7
680 por mm1, mm4
681 ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1)
682
683 movq mm4, [rsi+2*rax] ; p1
684 movq mm3, mm4 ; p1
685 psubusb mm4, mm5 ; p1-=p2
686 psubusb mm5, mm3 ; p2-=p1
687 por mm4, mm5 ; abs(p2 - p1)
688 psubusb mm4, mm7
689 por mm1, mm4
690
691 movq mm2, mm3 ; p1
692
693
694 ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1)
695
696 movq mm4, [rsi+rax] ; p0
697 movq mm5, mm4 ; p0
698 psubusb mm4, mm3 ; p0-=p1
699 psubusb mm3, mm5 ; p1-=p0
700 por mm4, mm3 ; abs(p1 - p0)
701 movq t1, mm4 ; save to t1
702 psubusb mm4, mm7
703 por mm1, mm4
704 ; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1) t1 = abs(p1-p0)
705 ; mm5 = p0
706 movq mm3, [rdi] ; q1
707 movq mm4, mm3 ; q1
708 psubusb mm3, mm2 ; q1-=p1
709 psubusb mm2, mm4 ; p1-=q1
710 por mm2, mm3 ; abs(p1-q1)
711 pand mm2, [GLOBAL(tfe)] ; set lsb of each byte to zero
712 psrlw mm2, 1 ; abs(p1-q1)/2
713
714 movq mm6, mm5 ; p0
715 movq mm3, mm0 ; q0
716 psubusb mm5, mm3 ; p0-=q0
717 psubusb mm3, mm6 ; q0-=p0
718 por mm5, mm3 ; abs(p0 - q0)
719 paddusb mm5, mm5 ; abs(p0-q0)*2
720 paddusb mm5, mm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2
721
722 mov rdx, arg(2) ;flimit ; get flimit
723 movq mm2, [rdx] ; flimit mm2
724 paddb mm2, mm2 ; flimit*2 (less than 255)
725 paddb mm7, mm2 ; flimit * 2 + limit (less than 255)
726
727 psubusb mm5, mm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > flimit * 2 + limit
728 por mm1, mm5
729 pxor mm5, mm5
730 pcmpeqb mm1, mm5 ; mask mm1
731
732 ; mm1 = mask, mm0=q0, mm7 = flimit, t0 = abs(q0-q1) t1 = abs(p1-p0)
733 ; mm6 = p0,
734
735 ; calculate high edge variance
736 mov rdx, arg(4) ;thresh ; get thresh
737 movq mm7, [rdx] ;
738 movq mm4, t0 ; get abs (q1 - q0)
739 psubusb mm4, mm7
740 movq mm3, t1 ; get abs (p1 - p0)
741 psubusb mm3, mm7
742 paddb mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
743
744 pcmpeqb mm4, mm5
745
746 pcmpeqb mm5, mm5
747 pxor mm4, mm5
748
749
750
751 ; mm1 = mask, mm0=q0, mm7 = thresh, t0 = abs(q0-q1) t1 = abs(p1-p0)
752 ; mm6 = p0, mm4=hev
753 ; start work on filters
754 movq mm2, [rsi+2*rax] ; p1
755 movq mm7, [rdi] ; q1
756 pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
757 pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
758 psubsb mm2, mm7 ; p1 - q1
759
760 pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
761 pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
762 movq mm3, mm0 ; q0
763 psubsb mm0, mm6 ; q0 - p0
764 paddsb mm2, mm0 ; 1 * (q0 - p0) + (p1 - q1)
765 paddsb mm2, mm0 ; 2 * (q0 - p0)
766 paddsb mm2, mm0 ; 3 * (q0 - p0) + (p1 - q1)
767 pand mm1, mm2 ; mask filter values we don't care about
768
769
770 ; mm1 = vp8_filter, mm4=hev, mm6=ps0, mm3=qs0
771 movq mm2, mm1 ; vp8_filter
772 pand mm2, mm4; ; Filter2 = vp8_filter & hev
773
774 movq mm5, mm2 ;
775 paddsb mm5, [GLOBAL(t3)];
776
777 pxor mm0, mm0 ; 0
778 pxor mm7, mm7 ; 0
779
780 punpcklbw mm0, mm5 ; e0f0g0h0
781 psraw mm0, 11 ; sign extended shift right by 3
782 punpckhbw mm7, mm5 ; a0b0c0d0
783 psraw mm7, 11 ; sign extended shift right by 3
784 packsswb mm0, mm7 ; Filter2 >>=3;
785
786 movq mm5, mm0 ; Filter2
787
788 paddsb mm2, [GLOBAL(t4)] ; vp8_signed_char_clamp(Filter2 + 4)
789 pxor mm0, mm0 ; 0
790 pxor mm7, mm7 ; 0
791
792 punpcklbw mm0, mm2 ; e0f0g0h0
793 psraw mm0, 11 ; sign extended shift right by 3
794 punpckhbw mm7, mm2 ; a0b0c0d0
795 psraw mm7, 11 ; sign extended shift right by 3
796 packsswb mm0, mm7 ; Filter2 >>=3;
797
798 ; mm0= filter2 mm1 = vp8_filter, mm3 =qs0 mm5=s mm4 =hev mm6=ps0
799 psubsb mm3, mm0 ; qs0 =qs0 - filter1
800 paddsb mm6, mm5 ; ps0 =ps0 + Fitler2
801
802 ; mm1=vp8_filter, mm3=qs0, mm4 =hev mm6=ps0
803 ; vp8_filter &= ~hev;
804 ; Filter2 = vp8_filter;
805 pandn mm4, mm1 ; vp8_filter&=~hev
806
807
808 ; mm3=qs0, mm4=filter2, mm6=ps0
809
810 ; u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7);
811 ; s = vp8_signed_char_clamp(qs0 - u);
812 ; *oq0 = s^0x80;
813 ; s = vp8_signed_char_clamp(ps0 + u);
814 ; *op0 = s^0x80;
815 pxor mm0, mm0
816
817 pxor mm1, mm1
818 pxor mm2, mm2
819 punpcklbw mm1, mm4
820 punpckhbw mm2, mm4
821 pmulhw mm1, [GLOBAL(s27)]
822 pmulhw mm2, [GLOBAL(s27)]
823 paddw mm1, [GLOBAL(s63)]
824 paddw mm2, [GLOBAL(s63)]
825 psraw mm1, 7
826 psraw mm2, 7
827 packsswb mm1, mm2
828
829 psubsb mm3, mm1
830 paddsb mm6, mm1
831
832 pxor mm3, [GLOBAL(t80)]
833 pxor mm6, [GLOBAL(t80)]
834 movq [rsi+rax], mm6
835 movq [rsi], mm3
836
837 ; roughly 2/7th difference across boundary
838 ; u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7);
839 ; s = vp8_signed_char_clamp(qs1 - u);
840 ; *oq1 = s^0x80;
841 ; s = vp8_signed_char_clamp(ps1 + u);
842 ; *op1 = s^0x80;
843 pxor mm1, mm1
844 pxor mm2, mm2
845 punpcklbw mm1, mm4
846 punpckhbw mm2, mm4
847 pmulhw mm1, [GLOBAL(s18)]
848 pmulhw mm2, [GLOBAL(s18)]
849 paddw mm1, [GLOBAL(s63)]
850 paddw mm2, [GLOBAL(s63)]
851 psraw mm1, 7
852 psraw mm2, 7
853 packsswb mm1, mm2
854
855 movq mm3, [rdi]
856 movq mm6, [rsi+rax*2] ; p1
857
858 pxor mm3, [GLOBAL(t80)]
859 pxor mm6, [GLOBAL(t80)]
860
861 paddsb mm6, mm1
862 psubsb mm3, mm1
863
864 pxor mm6, [GLOBAL(t80)]
865 pxor mm3, [GLOBAL(t80)]
866 movq [rdi], mm3
867 movq [rsi+rax*2], mm6
868
869 ; roughly 1/7th difference across boundary
870 ; u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7);
871 ; s = vp8_signed_char_clamp(qs2 - u);
872 ; *oq2 = s^0x80;
873 ; s = vp8_signed_char_clamp(ps2 + u);
874 ; *op2 = s^0x80;
875 pxor mm1, mm1
876 pxor mm2, mm2
877 punpcklbw mm1, mm4
878 punpckhbw mm2, mm4
879 pmulhw mm1, [GLOBAL(s9)]
880 pmulhw mm2, [GLOBAL(s9)]
881 paddw mm1, [GLOBAL(s63)]
882 paddw mm2, [GLOBAL(s63)]
883 psraw mm1, 7
884 psraw mm2, 7
885 packsswb mm1, mm2
886
887
888 movq mm6, [rdi+rax*4]
889 neg rax
890 movq mm3, [rdi+rax ]
891
892 pxor mm6, [GLOBAL(t80)]
893 pxor mm3, [GLOBAL(t80)]
894
895 paddsb mm6, mm1
896 psubsb mm3, mm1
897
898 pxor mm6, [GLOBAL(t80)]
899 pxor mm3, [GLOBAL(t80)]
900 movq [rdi+rax ], mm3
901 neg rax
902 movq [rdi+rax*4], mm6
903
904 ;EARLY_BREAK_OUT:
905 neg rax
906 add rsi,8
907 dec rcx
908 jnz next8_mbh
909
910 add rsp, 32
911 pop rsp
912 ; begin epilog
913 pop rdi
914 pop rsi
915 RESTORE_GOT
916 UNSHADOW_ARGS
917 pop rbp
918 ret
919
920
921 ;void vp8_mbloop_filter_vertical_edge_mmx
922 ;(
923 ; unsigned char *src_ptr,
924 ; int src_pixel_step,
925 ; const char *flimit,
926 ; const char *limit,
927 ; const char *thresh,
928 ; int count
929 ;)
930 global sym(vp8_mbloop_filter_vertical_edge_mmx)
931 sym(vp8_mbloop_filter_vertical_edge_mmx):
932 push rbp
933 mov rbp, rsp
934 SHADOW_ARGS_TO_STACK 6
935 GET_GOT rbx
936 push rsi
937 push rdi
938 ; end prolog
939
940 ALIGN_STACK 16, rax
941 sub rsp, 96 ; reserve 96 bytes
942 %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
943 %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
944 %define srct [rsp + 32] ;__declspec(align(16)) char srct[64];
945
946 mov rsi, arg(0) ;src_ptr
947 movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
948
949 lea rsi, [rsi + rax*4 - 4]
950
951 movsxd rcx, dword ptr arg(5) ;count
952 next8_mbv:
953 lea rdi, [rsi + rax] ; rdi points to row +1 for indirect addressing
954
955 ;transpose
956 movq mm0, [rdi+2*rax] ; 77 76 75 74 73 72 71 70
957 movq mm6, [rsi+2*rax] ; 67 66 65 64 63 62 61 60
958
959 movq mm7, mm6 ; 77 76 75 74 73 72 71 70
960 punpckhbw mm7, mm0 ; 77 67 76 66 75 65 74 64
961
962 punpcklbw mm6, mm0 ; 73 63 72 62 71 61 70 60
963 movq mm0, [rsi+rax] ; 57 56 55 54 53 52 51 50
964
965 movq mm4, [rsi] ; 47 46 45 44 43 42 41 40
966 movq mm5, mm4 ; 47 46 45 44 43 42 41 40
967
968 punpckhbw mm5, mm0 ; 57 47 56 46 55 45 54 44
969 punpcklbw mm4, mm0 ; 53 43 52 42 51 41 50 40
970
971 movq mm3, mm5 ; 57 47 56 46 55 45 54 44
972 punpckhwd mm5, mm7 ; 77 67 57 47 76 66 56 46
973
974 punpcklwd mm3, mm7 ; 75 65 55 45 74 64 54 44
975 movq mm2, mm4 ; 53 43 52 42 51 41 50 40
976
977 punpckhwd mm4, mm6 ; 73 63 53 43 72 62 52 42
978 punpcklwd mm2, mm6 ; 71 61 51 41 70 60 50 40
979
980 neg rax
981
982 movq mm7, [rsi+rax] ; 37 36 35 34 33 32 31 30
983 movq mm6, [rsi+rax*2] ; 27 26 25 24 23 22 21 20
984
985 movq mm1, mm6 ; 27 26 25 24 23 22 21 20
986 punpckhbw mm6, mm7 ; 37 27 36 36 35 25 34 24
987
988 punpcklbw mm1, mm7 ; 33 23 32 22 31 21 30 20
989
990 movq mm7, [rsi+rax*4]; ; 07 06 05 04 03 02 01 00
991 punpckhbw mm7, [rdi+rax*4] ; 17 07 16 06 15 05 14 04
992
993 movq mm0, mm7 ; 17 07 16 06 15 05 14 04
994 punpckhwd mm7, mm6 ; 37 27 17 07 36 26 16 06
995
996 punpcklwd mm0, mm6 ; 35 25 15 05 34 24 14 04
997 movq mm6, mm7 ; 37 27 17 07 36 26 16 06
998
999 punpckhdq mm7, mm5 ; 77 67 57 47 37 27 17 07 = q3
1000 punpckldq mm6, mm5 ; 76 66 56 46 36 26 16 06 = q2
1001
1002 lea rdx, srct
1003 movq mm5, mm6 ; 76 66 56 46 36 26 16 06
1004
1005 movq [rdx+56], mm7
1006 psubusb mm5, mm7 ; q2-q3
1007
1008
1009 movq [rdx+48], mm6
1010 psubusb mm7, mm6 ; q3-q2
1011
1012 por mm7, mm5; ; mm7=abs (q3-q2)
1013 movq mm5, mm0 ; 35 25 15 05 34 24 14 04
1014
1015 punpckhdq mm5, mm3 ; 75 65 55 45 35 25 15 05 = q1
1016 punpckldq mm0, mm3 ; 74 64 54 44 34 24 15 04 = q0
1017
1018 movq mm3, mm5 ; 75 65 55 45 35 25 15 05 = q1
1019 psubusb mm3, mm6 ; q1-q2
1020
1021 psubusb mm6, mm5 ; q2-q1
1022 por mm6, mm3 ; mm6=abs(q2-q1)
1023
1024 movq [rdx+40], mm5 ; save q1
1025 movq [rdx+32], mm0 ; save q0
1026
1027 movq mm3, [rsi+rax*4] ; 07 06 05 04 03 02 01 00
1028 punpcklbw mm3, [rdi+rax*4] ; 13 03 12 02 11 01 10 00
1029
1030 movq mm0, mm3 ; 13 03 12 02 11 01 10 00
1031 punpcklwd mm0, mm1 ; 31 21 11 01 30 20 10 00
1032
1033 punpckhwd mm3, mm1 ; 33 23 13 03 32 22 12 02
1034 movq mm1, mm0 ; 31 21 11 01 30 20 10 00
1035
1036 punpckldq mm0, mm2 ; 70 60 50 40 30 20 10 00 =p3
1037 punpckhdq mm1, mm2 ; 71 61 51 41 31 21 11 01 =p2
1038
1039 movq [rdx], mm0 ; save p3
1040 movq [rdx+8], mm1 ; save p2
1041
1042 movq mm2, mm1 ; 71 61 51 41 31 21 11 01 =p2
1043 psubusb mm2, mm0 ; p2-p3
1044
1045 psubusb mm0, mm1 ; p3-p2
1046 por mm0, mm2 ; mm0=abs(p3-p2)
1047
1048 movq mm2, mm3 ; 33 23 13 03 32 22 12 02
1049 punpckldq mm2, mm4 ; 72 62 52 42 32 22 12 02 = p1
1050
1051 punpckhdq mm3, mm4 ; 73 63 53 43 33 23 13 03 = p0
1052 movq [rdx+24], mm3 ; save p0
1053
1054 movq [rdx+16], mm2 ; save p1
1055 movq mm5, mm2 ; mm5 = p1
1056
1057 psubusb mm2, mm1 ; p1-p2
1058 psubusb mm1, mm5 ; p2-p1
1059
1060 por mm1, mm2 ; mm1=abs(p2-p1)
1061 mov rdx, arg(3) ;limit
1062
1063 movq mm4, [rdx] ; mm4 = limit
1064 psubusb mm7, mm4 ; abs(q3-q2) > limit
1065
1066 psubusb mm0, mm4 ; abs(p3-p2) > limit
1067 psubusb mm1, mm4 ; abs(p2-p1) > limit
1068
1069 psubusb mm6, mm4 ; abs(q2-q1) > limit
1070 por mm7, mm6 ; or
1071
1072 por mm0, mm1 ;
1073 por mm0, mm7 ; abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit
1074
1075 movq mm1, mm5 ; p1
1076
1077 movq mm7, mm3 ; mm3=mm7=p0
1078 psubusb mm7, mm5 ; p0 - p1
1079
1080 psubusb mm5, mm3 ; p1 - p0
1081 por mm5, mm7 ; abs(p1-p0)
1082
1083 movq t0, mm5 ; save abs(p1-p0)
1084 lea rdx, srct
1085
1086 psubusb mm5, mm4 ; mm5 = abs(p1-p0) > limit
1087 por mm0, mm5 ; mm0=mask
1088
1089 movq mm5, [rdx+32] ; mm5=q0
1090 movq mm7, [rdx+40] ; mm7=q1
1091
1092 movq mm6, mm5 ; mm6=q0
1093 movq mm2, mm7 ; q1
1094 psubusb mm5, mm7 ; q0-q1
1095
1096 psubusb mm7, mm6 ; q1-q0
1097 por mm7, mm5 ; abs(q1-q0)
1098
1099 movq t1, mm7 ; save abs(q1-q0)
1100 psubusb mm7, mm4 ; mm7=abs(q1-q0)> limit
1101
1102 por mm0, mm7 ; mask
1103
1104 movq mm5, mm2 ; q1
1105 psubusb mm5, mm1 ; q1-=p1
1106 psubusb mm1, mm2 ; p1-=q1
1107 por mm5, mm1 ; abs(p1-q1)
1108 pand mm5, [GLOBAL(tfe)] ; set lsb of each byte to zero
1109 psrlw mm5, 1 ; abs(p1-q1)/2
1110
1111 mov rdx, arg(2) ;flimit ;
1112
1113 movq mm2, [rdx] ;flimit mm2
1114 movq mm1, mm3 ; mm1=mm3=p0
1115
1116 movq mm7, mm6 ; mm7=mm6=q0
1117 psubusb mm1, mm7 ; p0-q0
1118
1119 psubusb mm7, mm3 ; q0-p0
1120 por mm1, mm7 ; abs(q0-p0)
1121 paddusb mm1, mm1 ; abs(q0-p0)*2
1122 paddusb mm1, mm5 ; abs (p0 - q0) *2 + abs(p1-q1)/2
1123
1124 paddb mm2, mm2 ; flimit*2 (less than 255)
1125 paddb mm4, mm2 ; flimit * 2 + limit (less than 255)
1126
1127 psubusb mm1, mm4 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > flimit * 2 + limit
1128 por mm1, mm0; ; mask
1129
1130 pxor mm0, mm0
1131 pcmpeqb mm1, mm0
1132
1133 ; calculate high edge variance
1134 mov rdx, arg(4) ;thresh ; get thresh
1135 movq mm7, [rdx]
1136 ;
1137 movq mm4, t0 ; get abs (q1 - q0)
1138 psubusb mm4, mm7 ; abs(q1 - q0) > thresh
1139
1140 movq mm3, t1 ; get abs (p1 - p0)
1141 psubusb mm3, mm7 ; abs(p1 - p0)> thresh
1142
1143 por mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
1144 pcmpeqb mm4, mm0
1145
1146 pcmpeqb mm0, mm0
1147 pxor mm4, mm0
1148
1149
1150
1151
1152 ; start work on filters
1153 lea rdx, srct
1154
1155 ; start work on filters
1156 movq mm2, [rdx+16] ; p1
1157 movq mm7, [rdx+40] ; q1
1158 pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
1159 pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
1160 psubsb mm2, mm7 ; p1 - q1
1161
1162 movq mm6, [rdx+24] ; p0
1163 movq mm0, [rdx+32] ; q0
1164 pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
1165 pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
1166
1167 movq mm3, mm0 ; q0
1168 psubsb mm0, mm6 ; q0 - p0
1169 paddsb mm2, mm0 ; 1 * (q0 - p0) + (p1 - q1)
1170 paddsb mm2, mm0 ; 2 * (q0 - p0)
1171 paddsb mm2, mm0 ; 3 * (q0 - p0) + (p1 - q1)
1172 pand mm1, mm2 ; mask filter values we don't care about
1173
1174 ; mm1 = vp8_filter, mm4=hev, mm6=ps0, mm3=qs0
1175 movq mm2, mm1 ; vp8_filter
1176 pand mm2, mm4; ; Filter2 = vp8_filter & hev
1177
1178 movq mm5, mm2 ;
1179 paddsb mm5, [GLOBAL(t3)];
1180
1181 pxor mm0, mm0 ; 0
1182 pxor mm7, mm7 ; 0
1183
1184 punpcklbw mm0, mm5 ; e0f0g0h0
1185 psraw mm0, 11 ; sign extended shift right by 3
1186 punpckhbw mm7, mm5 ; a0b0c0d0
1187 psraw mm7, 11 ; sign extended shift right by 3
1188 packsswb mm0, mm7 ; Filter2 >>=3;
1189
1190 movq mm5, mm0 ; Filter2
1191
1192 paddsb mm2, [GLOBAL(t4)] ; vp8_signed_char_clamp(Filter2 + 4)
1193 pxor mm0, mm0 ; 0
1194 pxor mm7, mm7 ; 0
1195
1196 punpcklbw mm0, mm2 ; e0f0g0h0
1197 psraw mm0, 11 ; sign extended shift right by 3
1198 punpckhbw mm7, mm2 ; a0b0c0d0
1199 psraw mm7, 11 ; sign extended shift right by 3
1200 packsswb mm0, mm7 ; Filter2 >>=3;
1201
1202 ; mm0= filter2 mm1 = vp8_filter, mm3 =qs0 mm5=s mm4 =hev mm6=ps0
1203 psubsb mm3, mm0 ; qs0 =qs0 - filter1
1204 paddsb mm6, mm5 ; ps0 =ps0 + Fitler2
1205
1206 ; mm1=vp8_filter, mm3=qs0, mm4 =hev mm6=ps0
1207 ; vp8_filter &= ~hev;
1208 ; Filter2 = vp8_filter;
1209 pandn mm4, mm1 ; vp8_filter&=~hev
1210
1211
1212 ; mm3=qs0, mm4=filter2, mm6=ps0
1213
1214 ; u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7);
1215 ; s = vp8_signed_char_clamp(qs0 - u);
1216 ; *oq0 = s^0x80;
1217 ; s = vp8_signed_char_clamp(ps0 + u);
1218 ; *op0 = s^0x80;
1219 pxor mm0, mm0
1220
1221 pxor mm1, mm1
1222 pxor mm2, mm2
1223 punpcklbw mm1, mm4
1224 punpckhbw mm2, mm4
1225 pmulhw mm1, [GLOBAL(s27)]
1226 pmulhw mm2, [GLOBAL(s27)]
1227 paddw mm1, [GLOBAL(s63)]
1228 paddw mm2, [GLOBAL(s63)]
1229 psraw mm1, 7
1230 psraw mm2, 7
1231 packsswb mm1, mm2
1232
1233 psubsb mm3, mm1
1234 paddsb mm6, mm1
1235
1236 pxor mm3, [GLOBAL(t80)]
1237 pxor mm6, [GLOBAL(t80)]
1238 movq [rdx+24], mm6
1239 movq [rdx+32], mm3
1240
1241 ; roughly 2/7th difference across boundary
1242 ; u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7);
1243 ; s = vp8_signed_char_clamp(qs1 - u);
1244 ; *oq1 = s^0x80;
1245 ; s = vp8_signed_char_clamp(ps1 + u);
1246 ; *op1 = s^0x80;
1247 pxor mm1, mm1
1248 pxor mm2, mm2
1249 punpcklbw mm1, mm4
1250 punpckhbw mm2, mm4
1251 pmulhw mm1, [GLOBAL(s18)]
1252 pmulhw mm2, [GLOBAL(s18)]
1253 paddw mm1, [GLOBAL(s63)]
1254 paddw mm2, [GLOBAL(s63)]
1255 psraw mm1, 7
1256 psraw mm2, 7
1257 packsswb mm1, mm2
1258
1259 movq mm3, [rdx + 40]
1260 movq mm6, [rdx + 16] ; p1
1261 pxor mm3, [GLOBAL(t80)]
1262 pxor mm6, [GLOBAL(t80)]
1263
1264 paddsb mm6, mm1
1265 psubsb mm3, mm1
1266
1267 pxor mm6, [GLOBAL(t80)]
1268 pxor mm3, [GLOBAL(t80)]
1269 movq [rdx + 40], mm3
1270 movq [rdx + 16], mm6
1271
1272 ; roughly 1/7th difference across boundary
1273 ; u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7);
1274 ; s = vp8_signed_char_clamp(qs2 - u);
1275 ; *oq2 = s^0x80;
1276 ; s = vp8_signed_char_clamp(ps2 + u);
1277 ; *op2 = s^0x80;
1278 pxor mm1, mm1
1279 pxor mm2, mm2
1280 punpcklbw mm1, mm4
1281 punpckhbw mm2, mm4
1282 pmulhw mm1, [GLOBAL(s9)]
1283 pmulhw mm2, [GLOBAL(s9)]
1284 paddw mm1, [GLOBAL(s63)]
1285 paddw mm2, [GLOBAL(s63)]
1286 psraw mm1, 7
1287 psraw mm2, 7
1288 packsswb mm1, mm2
1289
1290 movq mm6, [rdx+ 8]
1291 movq mm3, [rdx+48]
1292
1293 pxor mm6, [GLOBAL(t80)]
1294 pxor mm3, [GLOBAL(t80)]
1295
1296 paddsb mm6, mm1
1297 psubsb mm3, mm1
1298
1299 pxor mm6, [GLOBAL(t80)] ; mm6 = 71 61 51 41 31 21 11 01
1300 pxor mm3, [GLOBAL(t80)] ; mm3 = 76 66 56 46 36 26 15 06
1301
1302 ; tranpose and write back
1303 movq mm0, [rdx] ; mm0 = 70 60 50 40 30 20 10 00
1304 movq mm1, mm0 ; mm0 = 70 60 50 40 30 20 10 00
1305
1306 punpcklbw mm0, mm6 ; mm0 = 31 30 21 20 11 10 01 00
1307 punpckhbw mm1, mm6 ; mm3 = 71 70 61 60 51 50 41 40
1308
1309 movq mm2, [rdx+16] ; mm2 = 72 62 52 42 32 22 12 02
1310 movq mm6, mm2 ; mm3 = 72 62 52 42 32 22 12 02
1311
1312 punpcklbw mm2, [rdx+24] ; mm2 = 33 32 23 22 13 12 03 02
1313 punpckhbw mm6, [rdx+24] ; mm3 = 73 72 63 62 53 52 43 42
1314
1315 movq mm5, mm0 ; mm5 = 31 30 21 20 11 10 01 00
1316 punpcklwd mm0, mm2 ; mm0 = 13 12 11 10 03 02 01 00
1317
1318 punpckhwd mm5, mm2 ; mm5 = 33 32 31 30 23 22 21 20
1319 movq mm4, mm1 ; mm4 = 71 70 61 60 51 50 41 40
1320
1321 punpcklwd mm1, mm6 ; mm1 = 53 52 51 50 43 42 41 40
1322 punpckhwd mm4, mm6 ; mm4 = 73 72 71 70 63 62 61 60
1323
1324 movq mm2, [rdx+32] ; mm2 = 74 64 54 44 34 24 14 04
1325 punpcklbw mm2, [rdx+40] ; mm2 = 35 34 25 24 15 14 05 04
1326
1327 movq mm6, mm3 ; mm6 = 76 66 56 46 36 26 15 06
1328 punpcklbw mm6, [rdx+56] ; mm6 = 37 36 27 26 17 16 07 06
1329
1330 movq mm7, mm2 ; mm7 = 35 34 25 24 15 14 05 04
1331 punpcklwd mm2, mm6 ; mm2 = 17 16 15 14 07 06 05 04
1332
1333 punpckhwd mm7, mm6 ; mm7 = 37 36 35 34 27 26 25 24
1334 movq mm6, mm0 ; mm6 = 13 12 11 10 03 02 01 00
1335
1336 punpckldq mm0, mm2 ; mm0 = 07 06 05 04 03 02 01 00
1337 punpckhdq mm6, mm2 ; mm6 = 17 16 15 14 13 12 11 10
1338
1339 movq [rsi+rax*4], mm0 ; write out
1340 movq [rdi+rax*4], mm6 ; write out
1341
1342 movq mm0, mm5 ; mm0 = 33 32 31 30 23 22 21 20
1343 punpckldq mm0, mm7 ; mm0 = 27 26 25 24 23 22 20 20
1344
1345 punpckhdq mm5, mm7 ; mm5 = 37 36 35 34 33 32 31 30
1346 movq [rsi+rax*2], mm0 ; write out
1347
1348 movq [rdi+rax*2], mm5 ; write out
1349 movq mm2, [rdx+32] ; mm2 = 74 64 54 44 34 24 14 04
1350
1351 punpckhbw mm2, [rdx+40] ; mm2 = 75 74 65 64 54 54 45 44
1352 punpckhbw mm3, [rdx+56] ; mm3 = 77 76 67 66 57 56 47 46
1353
1354 movq mm5, mm2 ; mm5 = 75 74 65 64 54 54 45 44
1355 punpcklwd mm2, mm3 ; mm2 = 57 56 55 54 47 46 45 44
1356
1357 punpckhwd mm5, mm3 ; mm5 = 77 76 75 74 67 66 65 64
1358 movq mm0, mm1 ; mm0= 53 52 51 50 43 42 41 40
1359
1360 movq mm3, mm4 ; mm4 = 73 72 71 70 63 62 61 60
1361 punpckldq mm0, mm2 ; mm0 = 47 46 45 44 43 42 41 40
1362
1363 punpckhdq mm1, mm2 ; mm1 = 57 56 55 54 53 52 51 50
1364 movq [rsi], mm0 ; write out
1365
1366 movq [rdi], mm1 ; write out
1367 neg rax
1368
1369 punpckldq mm3, mm5 ; mm3 = 67 66 65 64 63 62 61 60
1370 punpckhdq mm4, mm5 ; mm4 = 77 76 75 74 73 72 71 60
1371
1372 movq [rsi+rax*2], mm3
1373 movq [rdi+rax*2], mm4
1374
1375 lea rsi, [rsi+rax*8]
1376 dec rcx
1377
1378 jnz next8_mbv
1379
1380 add rsp, 96
1381 pop rsp
1382 ; begin epilog
1383 pop rdi
1384 pop rsi
1385 RESTORE_GOT
1386 UNSHADOW_ARGS
1387 pop rbp
1388 ret
1389
1390
1391 ;void vp8_loop_filter_simple_horizontal_edge_mmx
1392 ;(
1393 ; unsigned char *src_ptr,
1394 ; int src_pixel_step,
1395 ; const char *flimit,
1396 ; const char *limit,
1397 ; const char *thresh,
1398 ; int count
1399 ;)
1400 global sym(vp8_loop_filter_simple_horizontal_edge_mmx)
1401 sym(vp8_loop_filter_simple_horizontal_edge_mmx):
1402 push rbp
1403 mov rbp, rsp
1404 SHADOW_ARGS_TO_STACK 6
1405 GET_GOT rbx
1406 push rsi
1407 push rdi
1408 ; end prolog
1409
1410 mov rsi, arg(0) ;src_ptr
1411 movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
1412
1413 movsxd rcx, dword ptr arg(5) ;count
1414 nexts8_h:
1415 mov rdx, arg(3) ;limit
1416 movq mm7, [rdx]
1417 mov rdx, arg(2) ;flimit ; get flimit
1418 movq mm3, [rdx] ;
1419 paddb mm3, mm3 ; flimit*2 (less than 255)
1420 paddb mm3, mm7 ; flimit * 2 + limit (less than 255)
1421
1422 mov rdi, rsi ; rdi points to row +1 for indirect addressing
1423 add rdi, rax
1424 neg rax
1425
1426 ; calculate mask
1427 movq mm1, [rsi+2*rax] ; p1
1428 movq mm0, [rdi] ; q1
1429 movq mm2, mm1
1430 movq mm7, mm0
1431 movq mm4, mm0
1432 psubusb mm0, mm1 ; q1-=p1
1433 psubusb mm1, mm4 ; p1-=q1
1434 por mm1, mm0 ; abs(p1-q1)
1435 pand mm1, [GLOBAL(tfe)] ; set lsb of each byte to zero
1436 psrlw mm1, 1 ; abs(p1-q1)/2
1437
1438 movq mm5, [rsi+rax] ; p0
1439 movq mm4, [rsi] ; q0
1440 movq mm0, mm4 ; q0
1441 movq mm6, mm5 ; p0
1442 psubusb mm5, mm4 ; p0-=q0
1443 psubusb mm4, mm6 ; q0-=p0
1444 por mm5, mm4 ; abs(p0 - q0)
1445 paddusb mm5, mm5 ; abs(p0-q0)*2
1446 paddusb mm5, mm1 ; abs (p0 - q0) *2 + abs(p1-q1)/2
1447
1448 psubusb mm5, mm3 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > flimit * 2 + limit
1449 pxor mm3, mm3
1450 pcmpeqb mm5, mm3
1451
1452 ; start work on filters
1453 pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
1454 pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
1455 psubsb mm2, mm7 ; p1 - q1
1456
1457 pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
1458 pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
1459 movq mm3, mm0 ; q0
1460 psubsb mm0, mm6 ; q0 - p0
1461 paddsb mm2, mm0 ; p1 - q1 + 1 * (q0 - p0)
1462 paddsb mm2, mm0 ; p1 - q1 + 2 * (q0 - p0)
1463 paddsb mm2, mm0 ; p1 - q1 + 3 * (q0 - p0)
1464 pand mm5, mm2 ; mask filter values we don't care about
1465
1466 ; do + 4 side
1467 paddsb mm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4
1468
1469 movq mm0, mm5 ; get a copy of filters
1470 psllw mm0, 8 ; shift left 8
1471 psraw mm0, 3 ; arithmetic shift right 11
1472 psrlw mm0, 8
1473 movq mm1, mm5 ; get a copy of filters
1474 psraw mm1, 11 ; arithmetic shift right 11
1475 psllw mm1, 8 ; shift left 8 to put it back
1476
1477 por mm0, mm1 ; put the two together to get result
1478
1479 psubsb mm3, mm0 ; q0-= q0 add
1480 pxor mm3, [GLOBAL(t80)] ; unoffset
1481 movq [rsi], mm3 ; write back
1482
1483
1484 ; now do +3 side
1485 psubsb mm5, [GLOBAL(t1s)] ; +3 instead of +4
1486
1487 movq mm0, mm5 ; get a copy of filters
1488 psllw mm0, 8 ; shift left 8
1489 psraw mm0, 3 ; arithmetic shift right 11
1490 psrlw mm0, 8
1491 psraw mm5, 11 ; arithmetic shift right 11
1492 psllw mm5, 8 ; shift left 8 to put it back
1493 por mm0, mm5 ; put the two together to get result
1494
1495
1496 paddsb mm6, mm0 ; p0+= p0 add
1497 pxor mm6, [GLOBAL(t80)] ; unoffset
1498 movq [rsi+rax], mm6 ; write back
1499
1500 add rsi,8
1501 neg rax
1502 dec rcx
1503 jnz nexts8_h
1504
1505 ; begin epilog
1506 pop rdi
1507 pop rsi
1508 RESTORE_GOT
1509 UNSHADOW_ARGS
1510 pop rbp
1511 ret
1512
1513
1514 ;void vp8_loop_filter_simple_vertical_edge_mmx
1515 ;(
1516 ; unsigned char *src_ptr,
1517 ; int src_pixel_step,
1518 ; const char *flimit,
1519 ; const char *limit,
1520 ; const char *thresh,
1521 ; int count
1522 ;)
1523 global sym(vp8_loop_filter_simple_vertical_edge_mmx)
1524 sym(vp8_loop_filter_simple_vertical_edge_mmx):
1525 push rbp
1526 mov rbp, rsp
1527 SHADOW_ARGS_TO_STACK 6
1528 GET_GOT rbx
1529 push rsi
1530 push rdi
1531 ; end prolog
1532
1533 ALIGN_STACK 16, rax
1534 sub rsp, 32 ; reserve 32 bytes
1535 %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
1536 %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
1537
1538 mov rsi, arg(0) ;src_ptr
1539 movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
1540
1541 lea rsi, [rsi + rax*4- 2]; ;
1542 movsxd rcx, dword ptr arg(5) ;count
1543 nexts8_v:
1544
1545 lea rdi, [rsi + rax];
1546 movd mm0, [rdi + rax * 2] ; xx xx xx xx 73 72 71 70
1547
1548 movd mm6, [rsi + rax * 2] ; xx xx xx xx 63 62 61 60
1549 punpcklbw mm6, mm0 ; 73 63 72 62 71 61 70 60
1550
1551 movd mm0, [rsi + rax] ; xx xx xx xx 53 52 51 50
1552 movd mm4, [rsi] ; xx xx xx xx 43 42 41 40
1553
1554 punpcklbw mm4, mm0 ; 53 43 52 42 51 41 50 40
1555 movq mm5, mm4 ; 53 43 52 42 51 41 50 40
1556
1557 punpcklwd mm4, mm6 ; 71 61 51 41 70 60 50 40
1558 punpckhwd mm5, mm6 ; 73 63 53 43 72 62 52 42
1559
1560 neg rax
1561
1562 movd mm7, [rsi + rax] ; xx xx xx xx 33 32 31 30
1563 movd mm6, [rsi + rax * 2] ; xx xx xx xx 23 22 21 20
1564
1565 punpcklbw mm6, mm7 ; 33 23 32 22 31 21 30 20
1566 movd mm1, [rdi + rax * 4] ; xx xx xx xx 13 12 11 10
1567
1568 movd mm0, [rsi + rax * 4] ; xx xx xx xx 03 02 01 00
1569 punpcklbw mm0, mm1 ; 13 03 12 02 11 01 10 00
1570
1571 movq mm2, mm0 ; 13 03 12 02 11 01 10 00
1572 punpcklwd mm0, mm6 ; 31 21 11 01 30 20 10 00
1573
1574 punpckhwd mm2, mm6 ; 33 23 13 03 32 22 12 02
1575 movq mm1, mm0 ; 13 03 12 02 11 01 10 00
1576
1577 punpckldq mm0, mm4 ; 70 60 50 40 30 20 10 00 = p1
1578 movq mm3, mm2 ; 33 23 13 03 32 22 12 02
1579
1580 punpckhdq mm1, mm4 ; 71 61 51 41 31 21 11 01 = p0
1581 punpckldq mm2, mm5 ; 72 62 52 42 32 22 12 02 = q0
1582
1583 punpckhdq mm3, mm5 ; 73 63 53 43 33 23 13 03 = q1
1584
1585
1586 ; calculate mask
1587 movq mm6, mm0 ; p1
1588 movq mm7, mm3 ; q1
1589 psubusb mm7, mm6 ; q1-=p1
1590 psubusb mm6, mm3 ; p1-=q1
1591 por mm6, mm7 ; abs(p1-q1)
1592 pand mm6, [GLOBAL(tfe)] ; set lsb of each byte to zero
1593 psrlw mm6, 1 ; abs(p1-q1)/2
1594
1595 movq mm5, mm1 ; p0
1596 movq mm4, mm2 ; q0
1597
1598 psubusb mm5, mm2 ; p0-=q0
1599 psubusb mm4, mm1 ; q0-=p0
1600
1601 por mm5, mm4 ; abs(p0 - q0)
1602 paddusb mm5, mm5 ; abs(p0-q0)*2
1603 paddusb mm5, mm6 ; abs (p0 - q0) *2 + abs(p1-q1)/2
1604
1605 mov rdx, arg(2) ;flimit ; get flimit
1606 movq mm7, [rdx]
1607 mov rdx, arg(3) ; get limit
1608 movq mm6, [rdx]
1609 paddb mm7, mm7 ; flimit*2 (less than 255)
1610 paddb mm7, mm6 ; flimit * 2 + limit (less than 255)
1611
1612 psubusb mm5, mm7 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > flimit * 2 + limit
1613 pxor mm7, mm7
1614 pcmpeqb mm5, mm7 ; mm5 = mask
1615
1616 ; start work on filters
1617 movq t0, mm0
1618 movq t1, mm3
1619
1620 pxor mm0, [GLOBAL(t80)] ; p1 offset to convert to signed values
1621 pxor mm3, [GLOBAL(t80)] ; q1 offset to convert to signed values
1622
1623 psubsb mm0, mm3 ; p1 - q1
1624 movq mm6, mm1 ; p0
1625
1626 movq mm7, mm2 ; q0
1627 pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
1628
1629 pxor mm7, [GLOBAL(t80)] ; offset to convert to signed values
1630 movq mm3, mm7 ; offseted ; q0
1631
1632 psubsb mm7, mm6 ; q0 - p0
1633 paddsb mm0, mm7 ; p1 - q1 + 1 * (q0 - p0)
1634
1635 paddsb mm0, mm7 ; p1 - q1 + 2 * (q0 - p0)
1636 paddsb mm0, mm7 ; p1 - q1 + 3 * (q0 - p0)
1637
1638 pand mm5, mm0 ; mask filter values we don't care about
1639
1640 paddsb mm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4
1641
1642 movq mm0, mm5 ; get a copy of filters
1643 psllw mm0, 8 ; shift left 8
1644 psraw mm0, 3 ; arithmetic shift right 11
1645 psrlw mm0, 8
1646
1647 movq mm7, mm5 ; get a copy of filters
1648 psraw mm7, 11 ; arithmetic shift right 11
1649 psllw mm7, 8 ; shift left 8 to put it back
1650
1651 por mm0, mm7 ; put the two together to get result
1652
1653 psubsb mm3, mm0 ; q0-= q0sz add
1654 pxor mm3, [GLOBAL(t80)] ; unoffset
1655
1656 ; now do +3 side
1657 psubsb mm5, [GLOBAL(t1s)] ; +3 instead of +4
1658
1659 movq mm0, mm5 ; get a copy of filters
1660 psllw mm0, 8 ; shift left 8
1661 psraw mm0, 3 ; arithmetic shift right 11
1662 psrlw mm0, 8
1663
1664 psraw mm5, 11 ; arithmetic shift right 11
1665 psllw mm5, 8 ; shift left 8 to put it back
1666 por mm0, mm5 ; put the two together to get result
1667
1668 paddsb mm6, mm0 ; p0+= p0 add
1669 pxor mm6, [GLOBAL(t80)] ; unoffset
1670
1671
1672 movq mm0, t0
1673 movq mm4, t1
1674
1675 ; mm0 = 70 60 50 40 30 20 10 00
1676 ; mm6 = 71 61 51 41 31 21 11 01
1677 ; mm3 = 72 62 52 42 32 22 12 02
1678 ; mm4 = 73 63 53 43 33 23 13 03
1679 ; transpose back to write out
1680
1681 movq mm1, mm0 ;
1682 punpcklbw mm0, mm6 ; 31 30 21 20 11 10 01 00
1683
1684 punpckhbw mm1, mm6 ; 71 70 61 60 51 50 41 40
1685 movq mm2, mm3 ;
1686
1687 punpcklbw mm2, mm4 ; 33 32 23 22 13 12 03 02
1688 movq mm5, mm1 ; 71 70 61 60 51 50 41 40
1689
1690 punpckhbw mm3, mm4 ; 73 72 63 62 53 52 43 42
1691 movq mm6, mm0 ; 31 30 21 20 11 10 01 00
1692
1693 punpcklwd mm0, mm2 ; 13 12 11 10 03 02 01 00
1694 punpckhwd mm6, mm2 ; 33 32 31 30 23 22 21 20
1695
1696 movd [rsi+rax*4], mm0 ; write 03 02 01 00
1697 punpcklwd mm1, mm3 ; 53 52 51 50 43 42 41 40
1698
1699 psrlq mm0, 32 ; xx xx xx xx 13 12 11 10
1700 punpckhwd mm5, mm3 ; 73 72 71 70 63 62 61 60
1701
1702 movd [rdi+rax*4], mm0 ; write 13 12 11 10
1703 movd [rsi+rax*2], mm6 ; write 23 22 21 20
1704
1705 psrlq mm6, 32 ; 33 32 31 30
1706 movd [rsi], mm1 ; write 43 42 41 40
1707
1708 movd [rsi + rax], mm6 ; write 33 32 31 30
1709 neg rax
1710
1711 movd [rsi + rax*2], mm5 ; write 63 62 61 60
1712 psrlq mm1, 32 ; 53 52 51 50
1713
1714 movd [rdi], mm1 ; write out 53 52 51 50
1715 psrlq mm5, 32 ; 73 72 71 70
1716
1717 movd [rdi + rax*2], mm5 ; write 73 72 71 70
1718
1719 lea rsi, [rsi+rax*8] ; next 8
1720
1721 dec rcx
1722 jnz nexts8_v
1723
1724 add rsp, 32
1725 pop rsp
1726 ; begin epilog
1727 pop rdi
1728 pop rsi
1729 RESTORE_GOT
1730 UNSHADOW_ARGS
1731 pop rbp
1732 ret
1733
1734
1735
1736 ;void fast_loop_filter_vertical_edges_mmx(unsigned char *y_ptr,
1737 ; int y_stride,
1738 ; loop_filter_info *lfi)
1739 ;{
1740 ;
1741 ;
1742 ; vp8_loop_filter_simple_vertical_edge_mmx(y_ptr+4, y_stride, lfi->flim,lfi->lim,lfi->thr,2);
1743 ; vp8_loop_filter_simple_vertical_edge_mmx(y_ptr+8, y_stride, lfi->flim,lfi->lim,lfi->thr,2);
1744 ; vp8_loop_filter_simple_vertical_edge_mmx(y_ptr+12, y_stride, lfi->flim,lfi->lim,lfi->thr,2);
1745 ;}
1746
1747 SECTION_RODATA
1748 align 16
1749 tfe:
1750 times 8 db 0xfe
1751 align 16
1752 t80:
1753 times 8 db 0x80
1754 align 16
1755 t1s:
1756 times 8 db 0x01
1757 align 16
1758 t3:
1759 times 8 db 0x03
1760 align 16
1761 t4:
1762 times 8 db 0x04
1763 align 16
1764 ones:
1765 times 4 dw 0x0001
1766 align 16
1767 s27:
1768 times 4 dw 0x1b00
1769 align 16
1770 s18:
1771 times 4 dw 0x1200
1772 align 16
1773 s9:
1774 times 4 dw 0x0900
1775 align 16
1776 s63:
1777 times 4 dw 0x003f
mirror for libvpx