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 ;******************************************************************************
23 %include "libavutil/x86/x86util.asm"
30 pb_4: times 16 db 0x04
31 pb_10: times 16 db 0x10
32 pb_40: times 16 db 0x40
33 pb_81: times 16 db 0x81
34 pb_f8: times 16 db 0xf8
35 pb_fe: times 16 db 0xfe
36 pb_ff: times 16 db 0xff
41 ; with mix functions, two 8-bit thresholds are stored in a 16-bit storage,
42 ; the following mask is used to splat both in the same register
43 mask_mix: times 8 db 0
46 mask_mix84: times 8 db 0xff
48 mask_mix48: times 8 db 0x00
70 %macro ABSSUB 4 ; dst, src1 (RO), src2 (RO), tmp
84 %macro CMP_GT 2-3 ; src/dst, cmp, pb_80
91 ; %1 = abs(%2-%3) > %4
92 %macro ABSSUB_GT 5-6 [pb_80]; dst, src1, src2, cmp, tmp, [pb_80]
93 ABSSUB %1, %2, %3, %5 ; dst = abs(src1-src2)
94 CMP_GT %1, %4, %6 ; dst > cmp
97 %macro MASK_APPLY 4 ; %1=new_data/dst %2=old_data %3=mask %4=tmp
98 pand %1, %3 ; new &= mask
99 pandn %4, %3, %2 ; tmp = ~mask & old
100 por %1, %4 ; new&mask | old&~mask
105 punpck%1bw %2, %3, %4
112 %macro FILTER_SUBx2_ADDx2 11 ; %1=dst %2=h/l %3=cache %4=stack_off %5=sub1 %6=sub2 %7=add1
113 ; %8=add2 %9=rshift, [unpack], [unpack_is_mem_on_x86_32]
114 psubw %3, [rsp+%4+%5*32]
115 psubw %3, [rsp+%4+%6*32]
116 paddw %3, [rsp+%4+%7*32]
119 punpck%2bw %1, %10, m0
121 UNPACK %2, %1, %10, m0
123 mova [rsp+%4+%8*32], %1
126 paddw %3, [rsp+%4+%8*32]
131 ; FIXME interleave l/h better (for instruction pairing)
132 %macro FILTER_INIT 9 ; tmp1, tmp2, cacheL, cacheH, dstp, stack_off, filterid, mask, source
133 FILTER%7_INIT %1, l, %3, %6 + 0
134 FILTER%7_INIT %2, h, %4, %6 + 16
136 MASK_APPLY %1, %9, %8, %2
141 %macro FILTER_UPDATE 12-16 "", "", "", 0 ; tmp1, tmp2, cacheL, cacheH, dstp, stack_off, -, -, +, +, rshift,
142 ; mask, [source], [unpack + src], [unpack_is_mem_on_x86_32]
143 ; FIXME interleave this properly with the subx2/addx2
145 %if %16 == 0 || ARCH_X86_64
149 FILTER_SUBx2_ADDx2 %1, l, %3, %6 + 0, %7, %8, %9, %10, %11, %14, %16
150 FILTER_SUBx2_ADDx2 %2, h, %4, %6 + 16, %7, %8, %9, %10, %11, %14, %16
153 MASK_APPLY %1, %13, %12, %2
155 MASK_APPLY %1, %5, %12, %2
160 %macro SRSHIFT3B_2X 4 ; reg1, reg2, [pb_10], tmp
172 %macro EXTRACT_POS_NEG 3 ; i8, neg, pos
175 pcmpgtb %3, %1 ; i8 < 0 mask
176 psubb %2, %1 ; neg values (only the originally - will be kept)
177 pand %2, %3 ; negative values of i8 (but stored as +)
178 pandn %3, %1 ; positive values of i8
182 %macro SIGN_ADD 4 ; dst, u8, i8, tmp1
183 EXTRACT_POS_NEG %3, %4, %1
184 paddusb %1, %2 ; add the positives
185 psubusb %1, %4 ; sub the negatives
189 %macro SIGN_SUB 4 ; dst, u8, i8, tmp1
190 EXTRACT_POS_NEG %3, %1, %4
191 paddusb %1, %2 ; add the negatives
192 psubusb %1, %4 ; sub the positives
195 %macro FILTER6_INIT 4 ; %1=dst %2=h/l %3=cache, %4=stack_off
196 UNPACK %2, %1, rp3, m0 ; p3: B->W
197 mova [rsp+%4+0*32], %1
198 paddw %3, %1, %1 ; p3*2
200 punpck%2bw %1, m1, m0 ; p2: B->W
201 mova [rsp+%4+1*32], %1
202 paddw %3, %1 ; p3*3 + p2
203 paddw %3, %1 ; p3*3 + p2*2
204 UNPACK %2, %1, rp1, m0 ; p1: B->W
205 mova [rsp+%4+2*32], %1
206 paddw %3, %1 ; p3*3 + p2*2 + p1
207 UNPACK %2, %1, rp0, m0 ; p0: B->W
208 mova [rsp+%4+3*32], %1
209 paddw %3, %1 ; p3*3 + p2*2 + p1 + p0
210 UNPACK %2, %1, rq0, m0 ; q0: B->W
211 mova [rsp+%4+4*32], %1
212 paddw %3, %1 ; p3*3 + p2*2 + p1 + p0 + q0
213 paddw %3, [pw_4] ; p3*3 + p2*2 + p1 + p0 + q0 + 4
214 psraw %1, %3, 3 ; (p3*3 + p2*2 + p1 + p0 + q0 + 4) >> 3
217 %macro FILTER14_INIT 4 ; %1=dst %2=h/l %3=cache, %4=stack_off
218 punpck%2bw %1, m2, m0 ; p7: B->W
219 mova [rsp+%4+ 8*32], %1
220 psllw %3, %1, 3 ; p7*8
222 punpck%2bw %1, m3, m0 ; p6: B->W
223 mova [rsp+%4+ 9*32], %1
224 paddw %3, %1 ; p7*7 + p6
225 paddw %3, %1 ; p7*7 + p6*2
226 UNPACK %2, %1, rp5, m0 ; p5: B->W
227 mova [rsp+%4+10*32], %1
228 paddw %3, %1 ; p7*7 + p6*2 + p5
229 UNPACK %2, %1, rp4, m0 ; p4: B->W
230 mova [rsp+%4+11*32], %1
231 paddw %3, %1 ; p7*7 + p6*2 + p5 + p4
232 paddw %3, [rsp+%4+ 0*32] ; p7*7 + p6*2 + p5 + p4 + p3
233 paddw %3, [rsp+%4+ 1*32] ; p7*7 + p6*2 + p5 + .. + p2
234 paddw %3, [rsp+%4+ 2*32] ; p7*7 + p6*2 + p5 + .. + p1
235 paddw %3, [rsp+%4+ 3*32] ; p7*7 + p6*2 + p5 + .. + p0
236 paddw %3, [rsp+%4+ 4*32] ; p7*7 + p6*2 + p5 + .. + p0 + q0
237 paddw %3, [pw_8] ; p7*7 + p6*2 + p5 + .. + p0 + q0 + 8
238 psraw %1, %3, 4 ; (p7*7 + p6*2 + p5 + .. + p0 + q0 + 8) >> 4
241 %macro TRANSPOSE16x16B 17
243 SBUTTERFLY bw, %1, %2, %16
244 SBUTTERFLY bw, %3, %4, %16
245 SBUTTERFLY bw, %5, %6, %16
246 SBUTTERFLY bw, %7, %8, %16
247 SBUTTERFLY bw, %9, %10, %16
248 SBUTTERFLY bw, %11, %12, %16
249 SBUTTERFLY bw, %13, %14, %16
252 SBUTTERFLY bw, %15, %16, %14
253 SBUTTERFLY wd, %1, %3, %14
254 SBUTTERFLY wd, %2, %4, %14
255 SBUTTERFLY wd, %5, %7, %14
256 SBUTTERFLY wd, %6, %8, %14
257 SBUTTERFLY wd, %9, %11, %14
258 SBUTTERFLY wd, %10, %12, %14
259 SBUTTERFLY wd, %13, %15, %14
262 SBUTTERFLY wd, %14, %16, %12
263 SBUTTERFLY dq, %1, %5, %12
264 SBUTTERFLY dq, %2, %6, %12
265 SBUTTERFLY dq, %3, %7, %12
266 SBUTTERFLY dq, %4, %8, %12
267 SBUTTERFLY dq, %9, %13, %12
268 SBUTTERFLY dq, %10, %14, %12
269 SBUTTERFLY dq, %11, %15, %12
272 SBUTTERFLY dq, %12, %16, %8
273 SBUTTERFLY qdq, %1, %9, %8
274 SBUTTERFLY qdq, %2, %10, %8
275 SBUTTERFLY qdq, %3, %11, %8
276 SBUTTERFLY qdq, %4, %12, %8
277 SBUTTERFLY qdq, %5, %13, %8
278 SBUTTERFLY qdq, %6, %14, %8
279 SBUTTERFLY qdq, %7, %15, %8
282 SBUTTERFLY qdq, %8, %16, %1
292 ; transpose 16 half lines (high part) to 8 full centered lines
293 %macro TRANSPOSE16x8B 16
302 SBUTTERFLY wd, %1, %3, %2
303 SBUTTERFLY wd, %5, %7, %2
304 SBUTTERFLY wd, %9, %11, %2
305 SBUTTERFLY wd, %13, %15, %2
306 SBUTTERFLY dq, %1, %5, %2
307 SBUTTERFLY dq, %3, %7, %2
308 SBUTTERFLY dq, %9, %13, %2
309 SBUTTERFLY dq, %11, %15, %2
310 SBUTTERFLY qdq, %1, %9, %2
311 SBUTTERFLY qdq, %3, %11, %2
312 SBUTTERFLY qdq, %5, %13, %2
313 SBUTTERFLY qdq, %7, %15, %2
324 %macro DEFINE_REAL_P7_TO_Q7 0-1 0
325 %define P7 dstq + 4*mstrideq + %1
326 %define P6 dstq + mstride3q + %1
327 %define P5 dstq + 2*mstrideq + %1
328 %define P4 dstq + mstrideq + %1
330 %define P2 dstq + strideq + %1
331 %define P1 dstq + 2* strideq + %1
332 %define P0 dstq + stride3q + %1
333 %define Q0 dstq + 4* strideq + %1
334 %define Q1 dst2q + mstride3q + %1
335 %define Q2 dst2q + 2*mstrideq + %1
336 %define Q3 dst2q + mstrideq + %1
337 %define Q4 dst2q + %1
338 %define Q5 dst2q + strideq + %1
339 %define Q6 dst2q + 2* strideq + %1
340 %define Q7 dst2q + stride3q + %1
343 ; ..............AB -> AAAAAAAABBBBBBBB
344 %macro SPLATB_MIX 1-2 [mask_mix]
354 %macro LOOPFILTER 5 ; %1=v/h %2=size1 %3+%4=stack, %5=32bit stack only
356 cglobal vp9_loop_filter_%1_%2_16, 5, 9, 16, %3 + %4, dst, stride, E, I, H, mstride, dst2, stride3, mstride3
359 cglobal vp9_loop_filter_%1_%2_16, 4, 8, 16, %3 + %4, dst, stride, E, I, mstride, dst2, stride3, mstride3
361 cglobal vp9_loop_filter_%1_%2_16, 2, 6, 16, %3 + %4 + %5, dst, stride, mstride, dst2, stride3, mstride3
368 mov mstrideq, strideq
371 lea stride3q, [strideq*3]
372 lea mstride3q, [mstrideq*3]
377 lea dstq, [dstq + 4*strideq - 4]
380 lea dstq, [dstq + 4*strideq - 8] ; go from top center (h pos) to center left (v pos)
382 lea dst2q, [dstq + 8*strideq]
384 lea dstq, [dstq + 4*mstrideq]
385 lea dst2q, [dstq + 8*strideq]
425 TRANSPOSE16x16B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, [rsp]
431 TRANSPOSE16x8B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
454 SPLATB_REG m2, I, m0 ; I I I I ...
455 SPLATB_REG m3, E, m0 ; E E E E ...
479 ; In case of horizontal, P3..Q3 are already present in some registers due
480 ; to the previous transpose, so we just swap registers.
504 ABSSUB_GT m5, rp3, rp2, m2, m7, m0 ; m5 = abs(p3-p2) <= I
505 ABSSUB_GT m1, rp2, rp1, m2, m7, m0 ; m1 = abs(p2-p1) <= I
507 ABSSUB_GT m1, rp1, rp0, m2, m7, m0 ; m1 = abs(p1-p0) <= I
509 ABSSUB_GT m1, rq0, rq1, m2, m7, m0 ; m1 = abs(q1-q0) <= I
511 ABSSUB_GT m1, rq1, rq2, m2, m7, m0 ; m1 = abs(q2-q1) <= I
513 ABSSUB_GT m1, rq2, rq3, m2, m7, m0 ; m1 = abs(q3-q2) <= I
515 ABSSUB m1, rp0, rq0, m7 ; abs(p0-q0)
516 paddusb m1, m1 ; abs(p0-q0) * 2
517 ABSSUB m2, rp1, rq1, m7 ; abs(p1-q1)
518 pand m2, [pb_fe] ; drop lsb so shift can work
519 psrlq m2, 1 ; abs(p1-q1)/2
520 paddusb m1, m2 ; abs(p0-q0)*2 + abs(p1-q1)/2
523 por m1, m5 ; fm final value
527 ; (m3: fm, m8..15: p3 p2 p1 p0 q0 q1 q2 q3)
528 ; calc flat8in (if not 44_16) and hev masks
530 mova m6, [pb_81] ; [1 1 1 1 ...] ^ 0x80
531 ABSSUB_GT m2, rp3, rp0, m6, m5 ; abs(p3 - p0) <= 1
538 ABSSUB_GT m1, rp2, rp0, m6, m5, rb80 ; abs(p2 - p0) <= 1
540 ABSSUB m4, rp1, rp0, m5 ; abs(p1 - p0)
545 SPLATB_REG m7, H, m0 ; H H H H ...
552 pcmpgtb m0, m4, m7 ; abs(p1 - p0) > H (1/2 hev condition)
553 CMP_GT m4, m6 ; abs(p1 - p0) <= 1
554 por m2, m4 ; (flat8in)
555 ABSSUB m4, rq1, rq0, m1 ; abs(q1 - q0)
557 pcmpgtb m5, m4, m7 ; abs(q1 - q0) > H (2/2 hev condition)
558 por m0, m5 ; hev final value
559 CMP_GT m4, m6 ; abs(q1 - q0) <= 1
560 por m2, m4 ; (flat8in)
561 ABSSUB_GT m1, rq2, rq0, m6, m5, rb80 ; abs(q2 - q0) <= 1
563 ABSSUB_GT m1, rq3, rq0, m6, m5, rb80 ; abs(q3 - q0) <= 1
564 por m2, m1 ; flat8in final value
566 %if %2 == 84 || %2 == 48
567 pand m2, [mask_mix%2]
574 ABSSUB m4, rp1, rp0, m1 ; abs(p1 - p0)
576 pcmpgtb m0, m4, m7 ; abs(p1 - p0) > H (1/2 hev condition)
577 ABSSUB m4, rq1, rq0, m1 ; abs(q1 - q0)
579 pcmpgtb m5, m4, m7 ; abs(q1 - q0) > H (2/2 hev condition)
580 por m0, m5 ; hev final value
584 ; (m0: hev, m2: flat8in, m3: fm, m6: pb_81, m9..15: p2 p1 p0 q0 q1 q2 q3)
595 ABSSUB_GT m1, rp7, rp0, m6, m5 ; abs(p7 - p0) <= 1
596 ABSSUB_GT m7, rp6, rp0, m6, m5 ; abs(p6 - p0) <= 1
607 ABSSUB_GT m7, rp5, rp0, m6, m5 ; abs(p5 - p0) <= 1
609 ABSSUB_GT m7, rp4, rp0, m6, m5 ; abs(p4 - p0) <= 1
620 ABSSUB_GT m7, rq4, rq0, m6, m5 ; abs(q4 - q0) <= 1
622 ABSSUB_GT m7, rq5, rq0, m6, m5 ; abs(q5 - q0) <= 1
633 ABSSUB_GT m7, rq6, rq0, m6, m5 ; abs(q4 - q0) <= 1
635 ABSSUB_GT m7, rq7, rq0, m6, m5 ; abs(q5 - q0) <= 1
636 por m1, m7 ; flat8out final value
641 ; if (out && in) filter_14()
642 ; else if (in) filter_6()
643 ; else if (hev) filter_2()
648 ; f6: fm & ~f14 & in => fm & ~(out & in) & in => fm & ~out & in
649 ; f2: fm & ~f14 & ~f6 & hev => fm & ~(out & in) & ~(~out & in) & hev => fm & ~in & hev
650 ; f4: fm & ~f14 & ~f6 & ~f2 => fm & ~(out & in) & ~(~out & in) & ~(~in & hev) => fm & ~in & ~hev
652 ; (m0: hev, [m1: flat8out], [m2: flat8in], m3: fm, m8..15: p5 p4 p1 p0 q0 q1 q6 q7)
655 mova m6, [pb_80] ; already in m6 if 44_16
656 SCRATCH 2, 15, rsp+%3+%4
658 SCRATCH 1, 8, rsp+%3+%4+16
661 pxor m2, m6, rq0 ; q0 ^ 0x80
662 pxor m4, m6, rp0 ; p0 ^ 0x80
663 psubsb m2, m4 ; (signed) q0 - p0
664 pxor m4, m6, rp1 ; p1 ^ 0x80
665 pxor m5, m6, rq1 ; q1 ^ 0x80
666 psubsb m4, m5 ; (signed) p1 - q1
667 paddsb m4, m2 ; (q0 - p0) + (p1 - q1)
668 paddsb m4, m2 ; 2*(q0 - p0) + (p1 - q1)
669 paddsb m4, m2 ; 3*(q0 - p0) + (p1 - q1)
670 paddsb m6, m4, [pb_4] ; m6: f1 = clip(f + 4, 127)
671 paddsb m4, [pb_3] ; m4: f2 = clip(f + 3, 127)
673 mova m14, [pb_10] ; will be reused in filter4()
678 SRSHIFT3B_2X m6, m4, rb10, m7 ; f1 and f2 sign byte shift by 3
679 SIGN_SUB m7, rq0, m6, m5 ; m7 = q0 - f1
680 SIGN_ADD m1, rp0, m4, m5 ; m1 = p0 + f2
683 pandn m6, m15, m3 ; ~mask(in) & mask(fm)
688 pand m6, m0 ; (~mask(in) & mask(fm)) & mask(hev)
692 MASK_APPLY m7, rq0, m6, m5 ; m7 = filter2(q0) & mask / we write it in filter4()
693 MASK_APPLY m1, rp0, m6, m5 ; m1 = filter2(p0) & mask / we write it in filter4()
695 ; (m0: hev, m1: p0', m2: q0-p0, m3: fm, m7: q0', [m8: flat8out], m10..13: p1 p0 q0 q1, m14: pb_10, [m15: flat8in], )
698 paddsb m2, m4 ; 2 * (q0 - p0)
699 paddsb m2, m4 ; 3 * (q0 - p0)
700 paddsb m6, m2, [pb_4] ; m6: f1 = clip(f + 4, 127)
701 paddsb m2, [pb_3] ; m2: f2 = clip(f + 3, 127)
702 SRSHIFT3B_2X m6, m2, rb10, m4 ; f1 and f2 sign byte shift by 3
705 pandn m5, m15, m3 ; ~mask(in) & mask(fm)
710 pandn m0, m5 ; ~mask(hev) & (~mask(in) & mask(fm))
714 SIGN_SUB m5, rq0, m6, m4 ; q0 - f1
715 MASK_APPLY m5, m7, m0, m4 ; filter4(q0) & mask
717 SIGN_ADD m7, rp0, m2, m4 ; p0 + f2
718 MASK_APPLY m7, m1, m0, m4 ; filter4(p0) & mask
721 pxor m1, m1 ; f=(f1+1)>>1
724 SIGN_ADD m1, rp1, m6, m2 ; p1 + f
725 SIGN_SUB m4, rq1, m6, m2 ; q1 - f
726 MASK_APPLY m1, rp1, m0, m2 ; m1 = filter4(p1)
727 MASK_APPLY m4, rq1, m0, m2 ; m4 = filter4(q1)
732 UNSCRATCH 2, 15, rsp+%3+%4
735 ; ([m1: flat8out], m2: flat8in, m3: fm, m10..13: p1 p0 q0 q1)
742 pand m2, m3 ; mask(fm) & mask(in)
744 pandn m3, m8, m2 ; ~mask(out) & (mask(fm) & mask(in))
746 mova m3, [rsp+%3+%4+16]
760 FILTER_INIT m4, m5, m6, m7, [P2], %4, 6, m3, m1 ; [p2]
762 FILTER_UPDATE m4, m5, m6, m7, [P1], %4, 0, 1, 2, 5, 3, m3, "", rq1, "", 1 ; [p1] -p3 -p2 +p1 +q1
763 FILTER_UPDATE m4, m5, m6, m7, [P0], %4, 0, 2, 3, 6, 3, m3, "", m1 ; [p0] -p3 -p1 +p0 +q2
764 FILTER_UPDATE m4, m5, m6, m7, [Q0], %4, 0, 3, 4, 7, 3, m3, "", rq3, "", 1 ; [q0] -p3 -p0 +q0 +q3
765 FILTER_UPDATE m4, m5, m6, m7, [Q1], %4, 1, 4, 5, 7, 3, m3, "" ; [q1] -p2 -q0 +q1 +q3
766 FILTER_UPDATE m4, m5, m6, m7, [Q2], %4, 2, 5, 6, 7, 3, m3, m1 ; [q2] -p1 -q1 +q2 +q3
770 UNSCRATCH 1, 8, rsp+%3+%4+16
773 ; (m0: 0, [m1: flat8out], m2: fm & flat8in, m8..15: q2 q3 p1 p0 q0 q1 p3 p2)
776 ; m2 m3 m8 m9 m14 m15 m10 m11 m12 m13
778 ; q2 q3 p3 p2 p1 p0 q0 q1
779 ; p6 -7 p7 p6 p5 p4 . . . . .
780 ; p5 -6 -p7 -p6 +p5 +q1 . . . .
781 ; p4 -5 -p7 -p5 +p4 +q2 . . . q2
782 ; p3 -4 -p7 -p4 +p3 +q3 . . . q3
783 ; p2 -3 -p7 -p3 +p2 +q4 . . . q4
784 ; p1 -2 -p7 -p2 +p1 +q5 . . . q5
785 ; p0 -1 -p7 -p1 +p0 +q6 . . . q6
786 ; q0 +0 -p7 -p0 +q0 +q7 . . . q7
787 ; q1 +1 -p6 -q0 +q1 +q7 q1 . . .
788 ; q2 +2 -p5 -q1 +q2 +q7 . q2 . .
789 ; q3 +3 -p4 -q2 +q3 +q7 . q3 . .
790 ; q4 +4 -p3 -q3 +q4 +q7 . q4 . .
791 ; q5 +5 -p2 -q4 +q5 +q7 . q5 . .
792 ; q6 +6 -p1 -q5 +q6 +q7 . q6 . .
795 pand m1, m2 ; mask(out) & (mask(fm) & mask(in))
827 FILTER_INIT m4, m5, m6, m7, [P6], %4, 14, m1, m3 ; [p6]
828 FILTER_UPDATE m4, m5, m6, m7, [P5], %4, 8, 9, 10, 5, 4, m1, rp5s ; [p5] -p7 -p6 +p5 +q1
829 FILTER_UPDATE m4, m5, m6, m7, [P4], %4, 8, 10, 11, 6, 4, m1, rp4s ; [p4] -p7 -p5 +p4 +q2
830 FILTER_UPDATE m4, m5, m6, m7, [P3], %4, 8, 11, 0, 7, 4, m1, rp3s ; [p3] -p7 -p4 +p3 +q3
831 FILTER_UPDATE m4, m5, m6, m7, [P2], %4, 8, 0, 1, 12, 4, m1, "", rq4, [Q4], 1 ; [p2] -p7 -p3 +p2 +q4
832 FILTER_UPDATE m4, m5, m6, m7, [P1], %4, 8, 1, 2, 13, 4, m1, "", rq5, [Q5], 1 ; [p1] -p7 -p2 +p1 +q5
833 FILTER_UPDATE m4, m5, m6, m7, [P0], %4, 8, 2, 3, 14, 4, m1, "", rq6, [Q6], 1 ; [p0] -p7 -p1 +p0 +q6
834 FILTER_UPDATE m4, m5, m6, m7, [Q0], %4, 8, 3, 4, 15, 4, m1, "", rq7, [Q7], 1 ; [q0] -p7 -p0 +q0 +q7
835 FILTER_UPDATE m4, m5, m6, m7, [Q1], %4, 9, 4, 5, 15, 4, m1, "" ; [q1] -p6 -q0 +q1 +q7
836 FILTER_UPDATE m4, m5, m6, m7, [Q2], %4, 10, 5, 6, 15, 4, m1, "" ; [q2] -p5 -q1 +q2 +q7
837 FILTER_UPDATE m4, m5, m6, m7, [Q3], %4, 11, 6, 7, 15, 4, m1, "" ; [q3] -p4 -q2 +q3 +q7
838 FILTER_UPDATE m4, m5, m6, m7, [Q4], %4, 0, 7, 12, 15, 4, m1, rq4s ; [q4] -p3 -q3 +q4 +q7
839 FILTER_UPDATE m4, m5, m6, m7, [Q5], %4, 1, 12, 13, 15, 4, m1, rq5s ; [q5] -p2 -q4 +q5 +q7
840 FILTER_UPDATE m4, m5, m6, m7, [Q6], %4, 2, 13, 14, 15, 4, m1, rq6s ; [q6] -p1 -q5 +q6 +q7
861 TRANSPOSE16x16B 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, [rsp]
884 DEFINE_REAL_P7_TO_Q7 2
885 SBUTTERFLY bw, 0, 1, 4
886 SBUTTERFLY bw, 2, 3, 4
887 SBUTTERFLY wd, 0, 2, 4
888 SBUTTERFLY wd, 1, 3, 4
918 ; the following code do a transpose of 8 full lines to 16 half
919 ; lines (high part). It is inlined to avoid the need of a staging area
929 SBUTTERFLY bw, 0, 1, 8
930 SBUTTERFLY bw, 2, 3, 8
931 SBUTTERFLY bw, 4, 5, 8
932 SBUTTERFLY bw, 6, 7, 8
933 SBUTTERFLY wd, 0, 2, 8
934 SBUTTERFLY wd, 1, 3, 8
935 SBUTTERFLY wd, 4, 6, 8
936 SBUTTERFLY wd, 5, 7, 8
937 SBUTTERFLY dq, 0, 4, 8
938 SBUTTERFLY dq, 1, 5, 8
939 SBUTTERFLY dq, 2, 6, 8
940 SBUTTERFLY dq, 3, 7, 8
965 LOOPFILTER v, %1, %2, 0, %3
967 LOOPFILTER h, %1, %2, 256, %3
971 %macro LPF_16_VH_ALL_OPTS 2-3 0
972 LPF_16_VH %1, %2, %3, sse2
973 LPF_16_VH %1, %2, %3, ssse3
974 LPF_16_VH %1, %2, %3, avx
977 LPF_16_VH_ALL_OPTS 16, 512, 32
978 LPF_16_VH_ALL_OPTS 44, 0, 0
979 LPF_16_VH_ALL_OPTS 48, 256, 16
980 LPF_16_VH_ALL_OPTS 84, 256, 16
981 LPF_16_VH_ALL_OPTS 88, 256, 16