1 ;******************************************************************************
2 ;* VP9 loop filter SIMD optimizations
4 ;* Copyright (C) 2013-2014 Clément Bœsch <u pkh me>
6 ;* This file is part of FFmpeg.
8 ;* FFmpeg is free software; you can redistribute it and/or
9 ;* modify it under the terms of the GNU Lesser General Public
10 ;* License as published by the Free Software Foundation; either
11 ;* version 2.1 of the License, or (at your option) any later version.
13 ;* FFmpeg is distributed in the hope that it will be useful,
14 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
15 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 ;* Lesser General Public License for more details.
18 ;* You should have received a copy of the GNU Lesser General Public
19 ;* License along with FFmpeg; if not, write to the Free Software
20 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 ;******************************************************************************
25 %include "libavutil/x86/x86util.asm"
32 pb_4: times 16 db 0x04
33 pb_10: times 16 db 0x10
34 pb_40: times 16 db 0x40
35 pb_81: times 16 db 0x81
36 pb_f8: times 16 db 0xf8
37 pb_fe: times 16 db 0xfe
42 ; with mix functions, two 8-bit thresholds are stored in a 16-bit storage,
43 ; the following mask is used to splat both in the same register
44 mask_mix: times 8 db 0
47 mask_mix84: times 8 db 0xff
49 mask_mix48: times 8 db 0x00
55 %macro ABSSUB 4 ; dst, src1 (RO), src2 (RO), tmp
62 %macro CMP_LTE 3-4 ; src/dst, cmp, tmp, pb_80
66 pcmpgtb %3, %2, %1 ; cmp > src?
67 pcmpeqb %1, %2 ; cmp == src? XXX: avoid this with a -1/+1 well placed?
68 por %1, %3 ; cmp >= src?
71 ; %1 = abs(%2-%3) <= %4
72 %macro ABSSUB_CMP 6-7 [pb_80]; dst, src1, src2, cmp, tmp1, tmp2, [pb_80]
73 ABSSUB %1, %2, %3, %6 ; dst = abs(src1-src2)
74 CMP_LTE %1, %4, %6, %7 ; dst <= cmp
77 %macro MASK_APPLY 4 ; %1=new_data/dst %2=old_data %3=mask %4=tmp
78 pand %1, %3 ; new &= mask
79 pandn %4, %3, %2 ; tmp = ~mask & old
80 por %1, %4 ; new&mask | old&~mask
83 %macro FILTER_SUBx2_ADDx2 8 ; %1=dst %2=h/l %3=cache %4=sub1 %5=sub2 %6=add1 %7=add2 %8=rshift
95 %macro FILTER_INIT 8 ; tmp1, tmp2, cacheL, cacheH, dstp, filterid, mask, source
96 FILTER%6_INIT %1, l, %3
97 FILTER%6_INIT %2, h, %4
99 MASK_APPLY %1, %8, %7, %2
103 %macro FILTER_UPDATE 11-14 ; tmp1, tmp2, cacheL, cacheH, dstp, -, -, +, +, rshift, mask, [source], [preload reg + value]
104 %if %0 == 13 ; no source + preload
106 %elif %0 == 14 ; source + preload
109 FILTER_SUBx2_ADDx2 %1, l, %3, %6, %7, %8, %9, %10
110 FILTER_SUBx2_ADDx2 %2, h, %4, %6, %7, %8, %9, %10
112 %if %0 == 12 || %0 == 14
113 MASK_APPLY %1, %12, %11, %2
115 MASK_APPLY %1, %5, %11, %2
120 %macro SRSHIFT3B_2X 4 ; reg1, reg2, [pb_10], tmp
132 %macro EXTRACT_POS_NEG 3 ; i8, neg, pos
135 pcmpgtb %3, %1 ; i8 < 0 mask
136 psubb %2, %1 ; neg values (only the originally - will be kept)
137 pand %2, %3 ; negative values of i8 (but stored as +)
138 pandn %3, %1 ; positive values of i8
142 %macro SIGN_ADD 5 ; dst, u8, i8, tmp1, tmp2
143 EXTRACT_POS_NEG %3, %4, %5
144 psubusb %1, %2, %4 ; sub the negatives
145 paddusb %1, %5 ; add the positives
149 %macro SIGN_SUB 5 ; dst, u8, i8, tmp1, tmp2
150 EXTRACT_POS_NEG %3, %4, %5
151 psubusb %1, %2, %5 ; sub the positives
152 paddusb %1, %4 ; add the negatives
155 %macro FILTER6_INIT 3 ; %1=dst %2=h/l %3=cache
156 punpck%2bw %1, m14, m0 ; p3: B->W
157 paddw %3, %1, %1 ; p3*2
159 punpck%2bw %1, m15, m0 ; p2: B->W
160 paddw %3, %1 ; p3*3 + p2
161 paddw %3, %1 ; p3*3 + p2*2
162 punpck%2bw %1, m10, m0 ; p1: B->W
163 paddw %3, %1 ; p3*3 + p2*2 + p1
164 punpck%2bw %1, m11, m0 ; p0: B->W
165 paddw %3, %1 ; p3*3 + p2*2 + p1 + p0
166 punpck%2bw %1, m12, m0 ; q0: B->W
167 paddw %3, %1 ; p3*3 + p2*2 + p1 + p0 + q0
168 paddw %3, [pw_4] ; p3*3 + p2*2 + p1 + p0 + q0 + 4
169 psraw %1, %3, 3 ; (p3*3 + p2*2 + p1 + p0 + q0 + 4) >> 3
172 %macro FILTER14_INIT 3 ; %1=dst %2=h/l %3=cache
173 punpck%2bw %1, m2, m0 ; p7: B->W
174 psllw %3, %1, 3 ; p7*8
176 punpck%2bw %1, m3, m0 ; p6: B->W
177 paddw %3, %1 ; p7*7 + p6
178 paddw %3, %1 ; p7*7 + p6*2
179 punpck%2bw %1, m8, m0 ; p5: B->W
180 paddw %3, %1 ; p7*7 + p6*2 + p5
181 punpck%2bw %1, m9, m0 ; p4: B->W
182 paddw %3, %1 ; p7*7 + p6*2 + p5 + p4
183 punpck%2bw %1, m14, m0 ; p3: B->W
184 paddw %3, %1 ; p7*7 + p6*2 + p5 + p4 + p3
185 punpck%2bw %1, m15, m0 ; p2: B->W
186 paddw %3, %1 ; p7*7 + p6*2 + p5 + .. + p2
187 punpck%2bw %1, m10, m0 ; p1: B->W
188 paddw %3, %1 ; p7*7 + p6*2 + p5 + .. + p1
189 punpck%2bw %1, m11, m0 ; p0: B->W
190 paddw %3, %1 ; p7*7 + p6*2 + p5 + .. + p0
191 punpck%2bw %1, m12, m0 ; q0: B->W
192 paddw %3, %1 ; p7*7 + p6*2 + p5 + .. + p0 + q0
193 paddw %3, [pw_8] ; p7*7 + p6*2 + p5 + .. + p0 + q0 + 8
194 psraw %1, %3, 4 ; (p7*7 + p6*2 + p5 + .. + p0 + q0 + 8) >> 4
197 %macro TRANSPOSE16x16B 17
199 SBUTTERFLY bw, %1, %2, %16
200 SBUTTERFLY bw, %3, %4, %16
201 SBUTTERFLY bw, %5, %6, %16
202 SBUTTERFLY bw, %7, %8, %16
203 SBUTTERFLY bw, %9, %10, %16
204 SBUTTERFLY bw, %11, %12, %16
205 SBUTTERFLY bw, %13, %14, %16
208 SBUTTERFLY bw, %15, %16, %14
209 SBUTTERFLY wd, %1, %3, %14
210 SBUTTERFLY wd, %2, %4, %14
211 SBUTTERFLY wd, %5, %7, %14
212 SBUTTERFLY wd, %6, %8, %14
213 SBUTTERFLY wd, %9, %11, %14
214 SBUTTERFLY wd, %10, %12, %14
215 SBUTTERFLY wd, %13, %15, %14
218 SBUTTERFLY wd, %14, %16, %12
219 SBUTTERFLY dq, %1, %5, %12
220 SBUTTERFLY dq, %2, %6, %12
221 SBUTTERFLY dq, %3, %7, %12
222 SBUTTERFLY dq, %4, %8, %12
223 SBUTTERFLY dq, %9, %13, %12
224 SBUTTERFLY dq, %10, %14, %12
225 SBUTTERFLY dq, %11, %15, %12
228 SBUTTERFLY dq, %12, %16, %8
229 SBUTTERFLY qdq, %1, %9, %8
230 SBUTTERFLY qdq, %2, %10, %8
231 SBUTTERFLY qdq, %3, %11, %8
232 SBUTTERFLY qdq, %4, %12, %8
233 SBUTTERFLY qdq, %5, %13, %8
234 SBUTTERFLY qdq, %6, %14, %8
235 SBUTTERFLY qdq, %7, %15, %8
238 SBUTTERFLY qdq, %8, %16, %1
248 ; transpose 16 half lines (high part) to 8 full centered lines
249 %macro TRANSPOSE16x8B 16
258 SBUTTERFLY wd, %1, %3, %2
259 SBUTTERFLY wd, %5, %7, %2
260 SBUTTERFLY wd, %9, %11, %2
261 SBUTTERFLY wd, %13, %15, %2
262 SBUTTERFLY dq, %1, %5, %2
263 SBUTTERFLY dq, %3, %7, %2
264 SBUTTERFLY dq, %9, %13, %2
265 SBUTTERFLY dq, %11, %15, %2
266 SBUTTERFLY qdq, %1, %9, %2
267 SBUTTERFLY qdq, %3, %11, %2
268 SBUTTERFLY qdq, %5, %13, %2
269 SBUTTERFLY qdq, %7, %15, %2
280 %macro DEFINE_REAL_P7_TO_Q7 0-1 0
281 %define P7 dst1q + 2*mstrideq + %1
282 %define P6 dst1q + mstrideq + %1
283 %define P5 dst1q + %1
284 %define P4 dst1q + strideq + %1
285 %define P3 dstq + 4*mstrideq + %1
286 %define P2 dstq + mstride3q + %1
287 %define P1 dstq + 2*mstrideq + %1
288 %define P0 dstq + mstrideq + %1
290 %define Q1 dstq + strideq + %1
291 %define Q2 dstq + 2*strideq + %1
292 %define Q3 dstq + stride3q + %1
293 %define Q4 dstq + 4*strideq + %1
294 %define Q5 dst2q + mstrideq + %1
295 %define Q6 dst2q + %1
296 %define Q7 dst2q + strideq + %1
299 ; ..............AB -> AAAAAAAABBBBBBBB
300 %macro SPLATB_MIX 1-2 [mask_mix]
311 %macro LOOPFILTER 2 ; %1=v/h %2=size1
312 lea mstrideq, [strideq]
315 lea stride3q, [strideq+2*strideq]
316 mov mstride3q, stride3q
322 lea dstq, [dstq + 8*strideq - 4]
325 lea dstq, [dstq + 8*strideq - 8] ; go from top center (h pos) to center left (v pos)
329 lea dst1q, [dstq + 2*mstride3q] ; dst1q = &dst[stride * -6]
330 lea dst2q, [dstq + 2* stride3q] ; dst2q = &dst[stride * +6]
369 TRANSPOSE16x16B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, [rsp]
375 TRANSPOSE16x8B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
398 SPLATB_REG m2, I, m0 ; I I I I ...
399 SPLATB_REG m3, E, m0 ; E E E E ...
422 ; In case of horizontal, P3..Q3 are already present in some registers due
423 ; to the previous transpose, so we just swap registers.
429 ABSSUB_CMP m5, m8, m9, m2, m6, m7, m0 ; m5 = abs(p3-p2) <= I
430 ABSSUB_CMP m1, m9, m10, m2, m6, m7, m0 ; m1 = abs(p2-p1) <= I
432 ABSSUB_CMP m1, m10, m11, m2, m6, m7, m0 ; m1 = abs(p1-p0) <= I
434 ABSSUB_CMP m1, m12, m13, m2, m6, m7, m0 ; m1 = abs(q1-q0) <= I
436 ABSSUB_CMP m1, m13, m14, m2, m6, m7, m0 ; m1 = abs(q2-q1) <= I
438 ABSSUB_CMP m1, m14, m15, m2, m6, m7, m0 ; m1 = abs(q3-q2) <= I
440 ABSSUB m1, m11, m12, m7 ; abs(p0-q0)
441 paddusb m1, m1 ; abs(p0-q0) * 2
442 ABSSUB m2, m10, m13, m7 ; abs(p1-q1)
443 pand m2, [pb_fe] ; drop lsb so shift can work
444 psrlq m2, 1 ; abs(p1-q1)/2
445 paddusb m1, m2 ; abs(p0-q0)*2 + abs(p1-q1)/2
447 pcmpgtb m4, m3, m1 ; E > X?
448 pcmpeqb m3, m1 ; E == X?
450 pand m3, m5 ; fm final value
452 ; (m3: fm, m8..15: p3 p2 p1 p0 q0 q1 q2 q3)
453 ; calc flat8in (if not 44_16) and hev masks
454 mova m6, [pb_81] ; [1 1 1 1 ...] ^ 0x80
456 ABSSUB_CMP m2, m8, m11, m6, m4, m5 ; abs(p3 - p0) <= 1
458 ABSSUB_CMP m1, m9, m11, m6, m4, m5, m8 ; abs(p2 - p0) <= 1
460 ABSSUB m4, m10, m11, m5 ; abs(p1 - p0)
465 SPLATB_REG m7, H, m0 ; H H H H ...
472 pcmpgtb m0, m4, m7 ; abs(p1 - p0) > H (1/2 hev condition)
473 CMP_LTE m4, m6, m5 ; abs(p1 - p0) <= 1
474 pand m2, m4 ; (flat8in)
475 ABSSUB m4, m13, m12, m1 ; abs(q1 - q0)
477 pcmpgtb m5, m4, m7 ; abs(q1 - q0) > H (2/2 hev condition)
478 por m0, m5 ; hev final value
479 CMP_LTE m4, m6, m5 ; abs(q1 - q0) <= 1
480 pand m2, m4 ; (flat8in)
481 ABSSUB_CMP m1, m14, m12, m6, m4, m5, m8 ; abs(q2 - q0) <= 1
483 ABSSUB_CMP m1, m15, m12, m6, m4, m5, m8 ; abs(q3 - q0) <= 1
484 pand m2, m1 ; flat8in final value
485 %if %2 == 84 || %2 == 48
486 pand m2, [mask_mix%2]
493 ABSSUB m4, m10, m11, m1 ; abs(p1 - p0)
495 pcmpgtb m0, m4, m7 ; abs(p1 - p0) > H (1/2 hev condition)
496 ABSSUB m4, m13, m12, m1 ; abs(q1 - q0)
498 pcmpgtb m5, m4, m7 ; abs(q1 - q0) > H (2/2 hev condition)
499 por m0, m5 ; hev final value
503 ; (m0: hev, m2: flat8in, m3: fm, m6: pb_81, m9..15: p2 p1 p0 q0 q1 q2 q3)
507 ABSSUB_CMP m1, m8, m11, m6, m4, m5 ; abs(p7 - p0) <= 1
508 ABSSUB_CMP m7, m9, m11, m6, m4, m5 ; abs(p6 - p0) <= 1
512 ABSSUB_CMP m7, m8, m11, m6, m4, m5 ; abs(p5 - p0) <= 1
514 ABSSUB_CMP m7, m9, m11, m6, m4, m5 ; abs(p4 - p0) <= 1
518 ABSSUB_CMP m7, m14, m12, m6, m4, m5 ; abs(q4 - q0) <= 1
520 ABSSUB_CMP m7, m15, m12, m6, m4, m5 ; abs(q5 - q0) <= 1
524 ABSSUB_CMP m7, m14, m12, m6, m4, m5 ; abs(q4 - q0) <= 1
526 ABSSUB_CMP m7, m15, m12, m6, m4, m5 ; abs(q5 - q0) <= 1
527 pand m1, m7 ; flat8out final value
531 ; if (out && in) filter_14()
532 ; else if (in) filter_6()
533 ; else if (hev) filter_2()
538 ; f6: fm & ~f14 & in => fm & ~(out & in) & in => fm & ~out & in
539 ; f2: fm & ~f14 & ~f6 & hev => fm & ~(out & in) & ~(~out & in) & hev => fm & ~in & hev
540 ; f4: fm & ~f14 & ~f6 & ~f2 => fm & ~(out & in) & ~(~out & in) & ~(~in & hev) => fm & ~in & ~hev
542 ; (m0: hev, [m1: flat8out], [m2: flat8in], m3: fm, m8..15: p5 p4 p1 p0 q0 q1 q6 q7)
545 mova m6, [pb_80] ; already in m6 if 44_16
547 pxor m15, m12, m6 ; q0 ^ 0x80
548 pxor m14, m11, m6 ; p0 ^ 0x80
549 psubsb m15, m14 ; (signed) q0 - p0
550 pxor m4, m10, m6 ; p1 ^ 0x80
551 pxor m5, m13, m6 ; q1 ^ 0x80
552 psubsb m4, m5 ; (signed) p1 - q1
553 paddsb m4, m15 ; (q0 - p0) + (p1 - q1)
554 paddsb m4, m15 ; 2*(q0 - p0) + (p1 - q1)
555 paddsb m4, m15 ; 3*(q0 - p0) + (p1 - q1)
556 paddsb m6, m4, [pb_4] ; m6: f1 = clip(f + 4, 127)
557 paddsb m4, [pb_3] ; m4: f2 = clip(f + 3, 127)
558 mova m14, [pb_10] ; will be reused in filter4()
559 SRSHIFT3B_2X m6, m4, m14, m7 ; f1 and f2 sign byte shift by 3
560 SIGN_SUB m7, m12, m6, m5, m9 ; m7 = q0 - f1
561 SIGN_ADD m8, m11, m4, m5, m9 ; m8 = p0 + f2
563 pandn m6, m2, m3 ; ~mask(in) & mask(fm)
564 pand m6, m0 ; (~mask(in) & mask(fm)) & mask(hev)
568 MASK_APPLY m7, m12, m6, m5 ; m7 = filter2(q0) & mask / we write it in filter4()
569 MASK_APPLY m8, m11, m6, m5 ; m8 = filter2(p0) & mask / we write it in filter4()
571 ; (m0: hev, [m1: flat8out], [m2: flat8in], m3: fm, m7..m8: q0' p0', m10..13: p1 p0 q0 q1, m14: pb_10, m15: q0-p0)
574 paddsb m15, m4 ; 2 * (q0 - p0)
575 paddsb m15, m4 ; 3 * (q0 - p0)
576 paddsb m6, m15, [pb_4] ; m6: f1 = clip(f + 4, 127)
577 paddsb m15, [pb_3] ; m15: f2 = clip(f + 3, 127)
578 SRSHIFT3B_2X m6, m15, m14, m9 ; f1 and f2 sign byte shift by 3
582 pandn m5, m2, m3 ; ~mask(in) & mask(fm)
583 pandn m0, m5 ; ~mask(hev) & (~mask(in) & mask(fm))
589 SIGN_SUB q0tmp, m12, m6, m4, m14 ; q0 - f1
590 MASK_APPLY q0tmp, m7, m0, m5 ; filter4(q0) & mask
592 SIGN_ADD p0tmp, m11, m15, m4, m14 ; p0 + f2
593 MASK_APPLY p0tmp, m8, m0, m5 ; filter4(p0) & mask
596 pxor m8, m8 ; f=(f1+1)>>1
599 SIGN_ADD m7, m10, m6, m8, m9 ; p1 + f
600 SIGN_SUB m4, m13, m6, m8, m9 ; q1 - f
601 MASK_APPLY m7, m10, m0, m14 ; m7 = filter4(p1)
602 MASK_APPLY m4, m13, m0, m14 ; m4 = filter4(q1)
606 ; ([m1: flat8out], m2: flat8in, m3: fm, m10..13: p1 p0 q0 q1)
613 pand m2, m3 ; mask(fm) & mask(in)
614 pandn m3, m1, m2 ; ~mask(out) & (mask(fm) & mask(in))
620 FILTER_INIT m4, m5, m6, m7, [P2], 6, m3, m15 ; [p2]
621 FILTER_UPDATE m6, m7, m4, m5, [P1], m14, m15, m10, m13, 3, m3 ; [p1] -p3 -p2 +p1 +q1
622 FILTER_UPDATE m4, m5, m6, m7, [P0], m14, m10, m11, m8, 3, m3 ; [p0] -p3 -p1 +p0 +q2
623 FILTER_UPDATE m6, m7, m4, m5, [Q0], m14, m11, m12, m9, 3, m3 ; [q0] -p3 -p0 +q0 +q3
624 FILTER_UPDATE m4, m5, m6, m7, [Q1], m15, m12, m13, m9, 3, m3 ; [q1] -p2 -q0 +q1 +q3
625 FILTER_UPDATE m6, m7, m4, m5, [Q2], m10, m13, m8, m9, 3, m3, m8 ; [q2] -p1 -q1 +q2 +q3
628 ; (m0: 0, [m1: flat8out], m2: fm & flat8in, m8..15: q2 q3 p1 p0 q0 q1 p3 p2)
631 ; m2 m3 m8 m9 m14 m15 m10 m11 m12 m13
633 ; q2 q3 p3 p2 p1 p0 q0 q1
634 ; p6 -7 p7 p6 p5 p4 . . . . .
635 ; p5 -6 -p7 -p6 +p5 +q1 . . . .
636 ; p4 -5 -p7 -p5 +p4 +q2 . . . q2
637 ; p3 -4 -p7 -p4 +p3 +q3 . . . q3
638 ; p2 -3 -p7 -p3 +p2 +q4 . . . q4
639 ; p1 -2 -p7 -p2 +p1 +q5 . . . q5
640 ; p0 -1 -p7 -p1 +p0 +q6 . . . q6
641 ; q0 +0 -p7 -p0 +q0 +q7 . . . q7
642 ; q1 +1 -p6 -q0 +q1 +q7 q1 . . .
643 ; q2 +2 -p5 -q1 +q2 +q7 . q2 . .
644 ; q3 +3 -p4 -q2 +q3 +q7 . q3 . .
645 ; q4 +4 -p3 -q3 +q4 +q7 . q4 . .
646 ; q5 +5 -p2 -q4 +q5 +q7 . q5 . .
647 ; q6 +6 -p1 -q5 +q6 +q7 . q6 . .
650 pand m1, m2 ; mask(out) & (mask(fm) & mask(in))
655 FILTER_INIT m4, m5, m6, m7, [P6], 14, m1, m3
656 FILTER_UPDATE m6, m7, m4, m5, [P5], m2, m3, m8, m13, 4, m1, m8 ; [p5] -p7 -p6 +p5 +q1
657 FILTER_UPDATE m4, m5, m6, m7, [P4], m2, m8, m9, m13, 4, m1, m9, m13, [Q2] ; [p4] -p7 -p5 +p4 +q2
658 FILTER_UPDATE m6, m7, m4, m5, [P3], m2, m9, m14, m13, 4, m1, m14, m13, [Q3] ; [p3] -p7 -p4 +p3 +q3
659 FILTER_UPDATE m4, m5, m6, m7, [P2], m2, m14, m15, m13, 4, m1, m13, [Q4] ; [p2] -p7 -p3 +p2 +q4
660 FILTER_UPDATE m6, m7, m4, m5, [P1], m2, m15, m10, m13, 4, m1, m13, [Q5] ; [p1] -p7 -p2 +p1 +q5
661 FILTER_UPDATE m4, m5, m6, m7, [P0], m2, m10, m11, m13, 4, m1, m13, [Q6] ; [p0] -p7 -p1 +p0 +q6
662 FILTER_UPDATE m6, m7, m4, m5, [Q0], m2, m11, m12, m13, 4, m1, m13, [Q7] ; [q0] -p7 -p0 +q0 +q7
663 FILTER_UPDATE m4, m5, m6, m7, [Q1], m3, m12, m2, m13, 4, m1, m2, [Q1] ; [q1] -p6 -q0 +q1 +q7
664 FILTER_UPDATE m6, m7, m4, m5, [Q2], m8, m2, m3, m13, 4, m1, m3, [Q2] ; [q2] -p5 -q1 +q2 +q7
665 FILTER_UPDATE m4, m5, m6, m7, [Q3], m9, m3, m8, m13, 4, m1, m8, m8, [Q3] ; [q3] -p4 -q2 +q3 +q7
666 FILTER_UPDATE m6, m7, m4, m5, [Q4], m14, m8, m9, m13, 4, m1, m9, m9, [Q4] ; [q4] -p3 -q3 +q4 +q7
667 FILTER_UPDATE m4, m5, m6, m7, [Q5], m15, m9, m14, m13, 4, m1, m14, m14, [Q5] ; [q5] -p2 -q4 +q5 +q7
668 FILTER_UPDATE m6, m7, m4, m5, [Q6], m10, m14, m15, m13, 4, m1, m15, m15, [Q6] ; [q6] -p1 -q5 +q6 +q7
689 TRANSPOSE16x16B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, [rsp]
710 DEFINE_REAL_P7_TO_Q7 2
711 SBUTTERFLY bw, 0, 1, 8
712 SBUTTERFLY bw, 2, 3, 8
713 SBUTTERFLY wd, 0, 2, 8
714 SBUTTERFLY wd, 1, 3, 8
715 SBUTTERFLY dq, 0, 4, 8
716 SBUTTERFLY dq, 1, 5, 8
717 SBUTTERFLY dq, 2, 6, 8
718 SBUTTERFLY dq, 3, 7, 8
744 ; the following code do a transpose of 8 full lines to 16 half
745 ; lines (high part). It is inlined to avoid the need of a staging area
755 SBUTTERFLY bw, 0, 1, 8
756 SBUTTERFLY bw, 2, 3, 8
757 SBUTTERFLY bw, 4, 5, 8
758 SBUTTERFLY bw, 6, 7, 8
759 SBUTTERFLY wd, 0, 2, 8
760 SBUTTERFLY wd, 1, 3, 8
761 SBUTTERFLY wd, 4, 6, 8
762 SBUTTERFLY wd, 5, 7, 8
763 SBUTTERFLY dq, 0, 4, 8
764 SBUTTERFLY dq, 1, 5, 8
765 SBUTTERFLY dq, 2, 6, 8
766 SBUTTERFLY dq, 3, 7, 8
799 cglobal vp9_loop_filter_v_%1_16, 5,10,16, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3
801 cglobal vp9_loop_filter_h_%1_16, 5,10,16, 256, dst, stride, E, I, H, mstride, dst1, dst2, stride3, mstride3
805 %macro LPF_16_VH_ALL_OPTS 1
811 LPF_16_VH_ALL_OPTS 16
812 LPF_16_VH_ALL_OPTS 44
813 LPF_16_VH_ALL_OPTS 48
814 LPF_16_VH_ALL_OPTS 84
815 LPF_16_VH_ALL_OPTS 88