1 ;*****************************************************************************
2 ;* deblock-a.asm: x86 deblocking
3 ;*****************************************************************************
4 ;* Copyright (C) 2005-2016 x264 project
6 ;* Authors: Loren Merritt <lorenm@u.washington.edu>
7 ;* Fiona Glaser <fiona@x264.com>
8 ;* Oskar Arvidsson <oskar@irock.se>
10 ;* This program is free software; you can redistribute it and/or modify
11 ;* it under the terms of the GNU General Public License as published by
12 ;* the Free Software Foundation; either version 2 of the License, or
13 ;* (at your option) any later version.
15 ;* This program is distributed in the hope that it will be useful,
16 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
17 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 ;* GNU General Public License for more details.
20 ;* You should have received a copy of the GNU General Public License
21 ;* along with this program; if not, write to the Free Software
22 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 ;* This program is also available under a commercial proprietary license.
25 ;* For more information, contact us at licensing@x264.com.
26 ;*****************************************************************************
29 %include "x86util.asm"
33 load_bytes_shuf: times 2 db 3,4,5,6,11,12,13,14,4,5,6,7,12,13,14,15
34 insert_top_shuf: dd 0,1,4,5,7,2,3,6
35 transpose_shuf: db 0,4,8,12,1,5,9,13,2,6,10,14,3,7,11,15
50 ; out: %4 = |%1-%2|-%3
59 ; out: %4 = |%1-%2|<%3
65 psubw %5, %3 ; |%1-%2|-%3
66 pcmpgtw %4, %5 ; 0 > |%1-%2|-%3
90 ; in: %1=p1, %2=p0, %3=q0, %4=q1
91 ; %5=alpha, %6=beta, %7-%9=tmp
94 ABS_SUB %2, %3, %5, %8, %7 ; |p0-q0| - alpha
95 ABS_SUB %1, %2, %6, %9, %7 ; |p1-p0| - beta
97 ABS_SUB %3, %4, %6, %9, %7 ; |q1-q0| - beta
103 ; in: %1=p0, %2=q0, %3=p1, %4=q1, %5=mask, %6=tmp, %7=tmp
104 ; out: %1=p0', m2=q0'
105 %macro DEBLOCK_P0_Q0 7
114 mova %6, [pw_pixel_max]
123 ; in: %1=x2, %2=x1, %3=p0, %4=q0 %5=mask&tc, %6=tmp
125 pavgw %6, %3, %4 ; (p0+q0+1)>>1
135 %macro LUMA_DEBLOCK_ONE 3
136 DIFF_LT m5, %1, bm, m4, m6
143 LUMA_Q1 m5, %2, m1, m2, m4, m6
146 %macro LUMA_H_STORE 2
158 movhps [%1+r1*2-4], m2
160 movhps [%1+r1*4-4], m3
164 %macro DEBLOCK_LUMA 0
165 ;-----------------------------------------------------------------------------
166 ; void deblock_v_luma( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
167 ;-----------------------------------------------------------------------------
168 cglobal deblock_v_luma, 5,5,8,0-5*mmsize
170 %define ms1 [rsp+mmsize]
171 %define ms2 [rsp+mmsize*2]
172 %define am [rsp+mmsize*3]
173 %define bm [rsp+mmsize*4]
175 LOAD_AB m4, m5, r2d, r3d
189 LOAD_MASK m0, m1, m2, m3, am, bm, m7, m4, m6
194 LUMA_DEBLOCK_ONE m1, m0, ms1
198 LUMA_DEBLOCK_ONE m2, m3, ms2
208 DEBLOCK_P0_Q0 m1, m2, m0, m3, m5, m7, m6
219 cglobal deblock_h_luma, 5,6,8,0-7*mmsize
221 %define ms1 [rsp+mmsize]
222 %define ms2 [rsp+mmsize*2]
223 %define p1m [rsp+mmsize*3]
224 %define p2m [rsp+mmsize*4]
225 %define am [rsp+mmsize*5]
226 %define bm [rsp+mmsize*6]
228 LOAD_AB m4, m5, r2d, r3d
241 movq m2, [r0-8] ; y q2 q1 q0
248 TRANSPOSE4x4W 2, 5, 0, 1, 4
251 TRANSPOSE4x4W 2, 3, 6, 7, 4
253 movu m5, [r0-8] ; y q2 q1 q0 p0 p1 p2 x
257 TRANSPOSE4x4W 5, 0, 2, 3, 6
264 TRANSPOSE4x4W 4, 1, 3, 7, 6
269 SBUTTERFLY qdq, 0, 1, 7
270 SBUTTERFLY qdq, 2, 3, 7
274 LOAD_MASK m0, m1, m2, m3, am, bm, m7, m4, m6
278 LUMA_DEBLOCK_ONE m1, m0, ms1
282 LUMA_DEBLOCK_ONE m2, m3, ms2
292 DEBLOCK_P0_Q0 m1, m2, m0, m3, m5, m7, m6
295 TRANSPOSE4x4W 0, 1, 2, 3, 4
299 lea r0, [r0+r1*(mmsize/2)]
300 lea r2, [r2+r1*(mmsize/2)]
307 ; in: m0=p1, m1=p0, m2=q0, m3=q1, m8=p2, m9=q2
308 ; m12=alpha, m13=beta
309 ; out: m0=p1', m3=q1', m1=p0', m2=q0'
310 ; clobbers: m4, m5, m6, m7, m10, m11, m14
311 %macro DEBLOCK_LUMA_INTER_SSE2 0
312 LOAD_MASK m0, m1, m2, m3, m12, m13, m7, m4, m6
314 DIFF_LT m8, m1, m13, m10, m4
315 DIFF_LT m9, m2, m13, m11, m4
325 LUMA_Q1 m8, m0, m1, m2, m5, m4
329 LUMA_Q1 m9, m3, m1, m2, m5, m4
337 DEBLOCK_P0_Q0 m1, m2, m0, m3, m4, m5, m6
343 %macro DEBLOCK_LUMA_64 0
344 cglobal deblock_v_luma, 5,5,15
355 LOAD_AB m12, m13, r2d, r3d
368 DEBLOCK_LUMA_INTER_SSE2
380 cglobal deblock_h_luma, 5,7,15
382 LOAD_AB m12, m13, r2d, r3d
390 movu m8, [r0-8] ; y q2 q1 q0 p0 p1 p2 x
399 TRANSPOSE4x4W 8, 0, 2, 9, 10
400 TRANSPOSE4x4W 5, 1, 3, 7, 10
403 SBUTTERFLY qdq, 0, 1, 10
404 SBUTTERFLY qdq, 2, 3, 10
407 DEBLOCK_LUMA_INTER_SSE2
409 TRANSPOSE4x4W 0, 1, 2, 3, 4
433 ; in: t0-t2: tmp registers
434 ; %1=p0 %2=p1 %3=p2 %4=p3 %5=q0 %6=q1 %7=mask0
435 ; %8=mask1p %9=2 %10=p0' %11=p1' %12=p2'
436 %macro LUMA_INTRA_P012 12 ; p0..p3 in memory
451 paddw t0, %9 ; (p2 + p1 + p0 + q0 + 2)
452 paddw t2, t0 ; (2*p3 + 3*p2 + p1 + p0 + q0 + 4)
471 psrlw t1, 2 ; (2*p1 + p0 + q1 + 2)/4
472 psrlw t0, 3 ; (p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4)>>3
484 %macro LUMA_INTRA_INIT 1
491 CAT_XDEFINE t, i, [rsp+mmsize*(i-4)]
497 ; in: %1-%3=tmp, %4=p2, %5=q2
498 %macro LUMA_INTRA_INTER 5
499 LOAD_AB t0, t1, r2d, r3d
501 LOAD_MASK m0, m1, m2, m3, %1, t1, t0, t2, t3
510 paddw t3, [pw_2] ; alpha/4+2
511 DIFF_LT m1, m2, t3, t2, t0 ; t2 = |p0-q0| < alpha/4+2
515 DIFF_LT t3, m2, t1, t2, t0 ; t2 = |q2-q0| < beta
519 DIFF_LT t3, m1, t1, t2, t0 ; t2 = |p2-p0| < beta
524 %macro LUMA_H_INTRA_LOAD 0
530 TRANSPOSE4x4W 4, 5, 0, 1, 2
538 TRANSPOSE4x4W 2, 3, 4, 5, 6
550 TRANSPOSE8x8W 4, 5, 0, 1, 2, 3, 6, 7, t4, t5
558 ; in: %1=q3 %2=q2' %3=q1' %4=q0' %5=p0' %6=p1' %7=p2' %8=p3 %9=tmp
559 %macro LUMA_H_INTRA_STORE 9
561 TRANSPOSE4x4W %1, %2, %3, %4, %9
564 movq [r0+r1*2-8], m%3
567 TRANSPOSE4x4W %5, %6, %7, %1, %9
573 TRANSPOSE2x4x4W %1, %2, %3, %4, %9
576 movq [r0+r1*2-8], m%3
579 movhps [r4+r1-8], m%2
580 movhps [r4+r1*2-8], m%3
581 movhps [r4+r5-8], m%4
587 TRANSPOSE2x4x4W %5, %6, %7, %1, %9
594 movhps [r4+r1*2], m%7
600 ;-----------------------------------------------------------------------------
601 ; void deblock_v_luma_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
602 ;-----------------------------------------------------------------------------
603 %macro DEBLOCK_LUMA_INTRA_64 0
604 cglobal deblock_v_luma_intra, 4,7,16
618 lea r5, [r1*3] ; 3*stride
620 add r4, r0 ; pix-4*stride
623 LOAD_AB aa, bb, r2d, r3d
632 LOAD_MASK p1, p0, q0, q1, aa, bb, m3, t0, t1
635 paddw t2, m0 ; alpha/4+2
636 DIFF_LT p0, q0, t2, m6, t0 ; m6 = |p0-q0| < alpha/4+2
637 DIFF_LT p2, p0, bb, t1, t0 ; m7 = |p2-p0| < beta
638 DIFF_LT q2, q0, bb, m7, t0 ; t1 = |q2-q0| < beta
642 LUMA_INTRA_P012 p0, p1, p2, [r4], q0, q1, m3, m6, m0, [r4+r5], [r4+2*r1], [r4+r1]
643 LUMA_INTRA_P012 q0, q1, q2, [r0+r5], p0, p1, m3, m7, m0, [r0], [r0+r1], [r0+2*r1]
650 ;-----------------------------------------------------------------------------
651 ; void deblock_h_luma_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
652 ;-----------------------------------------------------------------------------
653 cglobal deblock_h_luma_intra, 4,7,16
666 %assign pad 24-(stack_offset&15)
670 lea r5, [r1*3] ; 3*stride
671 add r4, r0 ; pix+4*stride
683 TRANSPOSE8x8W 5, 8, 9, 10, 11, 12, 13, 4, 1
685 LOAD_AB m1, m2, r2d, r3d
686 LOAD_MASK q1, q0, p0, p1, m1, m2, m3, t0, t1
688 paddw m1, m0 ; alpha/4+2
689 DIFF_LT p0, q0, m1, m6, t0 ; m6 = |p0-q0| < alpha/4+2
690 DIFF_LT q2, q0, m2, t1, t0 ; t1 = |q2-q0| < beta
691 DIFF_LT p0, p2, m2, m7, t0 ; m7 = |p2-p0| < beta
697 LUMA_INTRA_P012 q0, q1, q2, q3, p0, p1, m3, m6, m0, m5, m1, q2
698 LUMA_INTRA_P012 p0, p1, p2, p3, q0, q1, m3, m7, m0, p0, m6, p2
701 LUMA_H_INTRA_STORE 7, 8, 1, 5, 11, 6, 13, 4, 14
712 DEBLOCK_LUMA_INTRA_64
714 DEBLOCK_LUMA_INTRA_64
718 %macro DEBLOCK_LUMA_INTRA 0
719 ;-----------------------------------------------------------------------------
720 ; void deblock_v_luma_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
721 ;-----------------------------------------------------------------------------
722 cglobal deblock_v_luma_intra, 4,7,8,0-3*mmsize
730 mova m0, [r4+r1*2] ; p1
731 mova m1, [r4+r5] ; p0
733 mova m3, [r0+r1] ; q1
734 LUMA_INTRA_INTER t4, t5, t6, [r4+r1], [r0+r1*2]
735 LUMA_INTRA_P012 m1, m0, t3, [r4], m2, m3, t5, t4, [pw_2], [r4+r5], [r4+2*r1], [r4+r1]
736 mova t3, [r0+r1*2] ; q2
737 LUMA_INTRA_P012 m2, m3, t3, [r0+r5], m1, m0, t5, t6, [pw_2], [r0], [r0+r1], [r0+2*r1]
744 ;-----------------------------------------------------------------------------
745 ; void deblock_h_luma_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
746 ;-----------------------------------------------------------------------------
747 cglobal deblock_h_luma_intra, 4,7,8,0-8*mmsize
754 lea r5, [r1*3] ; 3*stride
755 add r4, r0 ; pix+4*stride
760 LUMA_INTRA_INTER t8, t9, t10, t5, t6
762 LUMA_INTRA_P012 m1, m0, t3, t4, m2, m3, t9, t8, [pw_2], t8, t5, t11
764 LUMA_INTRA_P012 m2, m3, t3, t7, m1, m0, t9, t10, [pw_2], m4, t6, m5
772 LUMA_H_INTRA_STORE 2, 0, 1, 3, 4, 6, 5, t7, 7
774 lea r0, [r0+r1*(mmsize/2)]
778 lea r4, [r4+r1*(mmsize/2)]
796 %endif ; HIGH_BIT_DEPTH
798 %if HIGH_BIT_DEPTH == 0
799 ; expands to [base],...,[base+7*stride]
800 %define PASS8ROWS(base, base3, stride, stride3) \
801 [base], [base+stride], [base+stride*2], [base3], \
802 [base3+stride], [base3+stride*2], [base3+stride3], [base3+stride*4]
804 %define PASS8ROWS(base, base3, stride, stride3, offset) \
805 PASS8ROWS(base+offset, base3+offset, stride, stride3)
807 ; in: 4 rows of 8 bytes in m0..m3
808 ; out: 8 rows of 4 bytes in %1..%8
809 %macro TRANSPOSE8x4B_STORE 8
838 ; in: 8 rows of 4 bytes in %9..%10
839 ; out: 8 rows of 4 bytes in %1..%8
851 ; in: 4 rows of 4 words in %1..%4
852 ; out: 4 rows of 4 word in m0..m3
854 %macro TRANSPOSE4x4W_LOAD 4-8
861 TRANSPOSE4x4W 0, 1, 2, 3, 4
877 ; in: 2 rows of 4 words in m1..m2
878 ; out: 4 rows of 2 words in %1..%4
880 %macro TRANSPOSE4x2W_STORE 4-8
896 ; in: 4/8 rows of 4 words in %1..%8
897 ; out: 4 rows of 4/8 word in m0..m3
898 ; clobbers: m4, m5, m6, m7
899 %macro TRANSPOSE4x8W_LOAD 8
901 TRANSPOSE4x4W_LOAD %1, %2, %3, %4
923 punpckhqdq m1, m0, m4
924 punpckhqdq m3, m2, m6
930 ; in: 2 rows of 4/8 words in m1..m2
931 ; out: 4/8 rows of 2 words in %1..%8
933 %macro TRANSPOSE8x2W_STORE 8
935 TRANSPOSE4x2W_STORE %1, %2, %3, %4
961 ; in: 8 rows of 8 (only the middle 6 pels are used) in %1..%8
962 ; out: 6 rows of 8 in [%9+0*16] .. [%9+5*16]
963 %macro TRANSPOSE6x8_MEM 9
980 punpcklbw m0, m2 ; __ AG BH CI DJ EK FL __
981 punpcklbw m1, m3 ; __ MS NT OU PV QW RX __
984 punpcklbw m2, m3 ; __ Ye Zf ag bh ci dj __
987 punpcklbw m3, m4 ; __ kq lr ms nt ou pv __
989 SBUTTERFLY wd, 0, 1, 4 ; __ __ AG MS BH NT CI OU
990 ; DJ PV EK QW FL RX __ __
991 SBUTTERFLY wd, 2, 3, 4 ; __ __ Ye kq Zf lr ag ms
992 ; bh nt ci ou dj pv __ __
993 SBUTTERFLY dq, 0, 2, 4 ; __ __ __ __ AG MS Ye kq
994 ; BH NT Zf lr CI FL OU RX
995 SBUTTERFLY dq, 1, 3, 4 ; DJ PV bh nt EK QW Zf lr
996 ; FL RX dj pv __ __ __ __
1001 movhps [%9+0x40], m1
1011 SBUTTERFLY bw, 0, 1, 7
1012 SBUTTERFLY bw, 2, 3, 7
1013 SBUTTERFLY bw, 4, 5, 7
1015 SBUTTERFLY3 bw, m6, %8, m7
1016 SBUTTERFLY wd, 0, 2, 3
1017 SBUTTERFLY wd, 4, 6, 3
1020 SBUTTERFLY3 wd, m1, [%9+0x10], m3
1021 SBUTTERFLY wd, 5, 7, 0
1022 SBUTTERFLY dq, 1, 5, 0
1023 SBUTTERFLY dq, 2, 6, 0
1031 RESET_MM_PERMUTATION
1035 ; in: 8 rows of 8 in %1..%8
1036 ; out: 8 rows of 8 in %9..%16
1037 %macro TRANSPOSE8x8_MEM 16
1038 RESET_MM_PERMUTATION
1052 SBUTTERFLY wd, 0, 1, 4
1053 SBUTTERFLY wd, 2, 3, 4
1054 SBUTTERFLY dq, 0, 2, 4
1055 SBUTTERFLY dq, 1, 3, 4
1072 SBUTTERFLY bw, 0, 1, 7
1073 SBUTTERFLY bw, 2, 3, 7
1074 SBUTTERFLY bw, 4, 5, 7
1075 SBUTTERFLY3 bw, m6, %8, m7
1077 SBUTTERFLY wd, 0, 2, 5
1078 SBUTTERFLY wd, 4, 6, 5
1079 SBUTTERFLY wd, 1, 3, 5
1082 SBUTTERFLY wd, 6, 7, 5
1083 SBUTTERFLY dq, 0, 4, 5
1084 SBUTTERFLY dq, 1, 6, 5
1089 SBUTTERFLY3 dq, m2, %11, m0
1090 SBUTTERFLY dq, 3, 7, 4
1096 RESET_MM_PERMUTATION
1099 ; out: %4 = |%1-%2|>%3
1102 %if avx_enabled == 0
1115 ; out: %4 = |%1-%2|>%3
1132 ; in: m0=p1 m1=p0 m2=q0 m3=q1 %1=alpha %2=beta
1133 ; out: m5=beta-1, m7=mask, %3=alpha-1
1135 %macro LOAD_MASK 2-3
1151 psubusb m4, m6 ; alpha - 1
1152 psubusb m5, m6 ; beta - 1
1156 DIFF_GT m1, m2, m4, m7, m6 ; |p0-q0| > alpha-1
1157 DIFF_GT m0, m1, m5, m4, m6 ; |p1-p0| > beta-1
1159 DIFF_GT m3, m2, m5, m4, m6 ; |q1-q0| > beta-1
1165 ; in: m0=p1 m1=p0 m2=q0 m3=q1 m7=(tc&mask)
1166 ; out: m1=p0' m2=q0'
1168 %macro DEBLOCK_P0_Q0 0
1169 pxor m5, m1, m2 ; p0^q0
1170 pand m5, [pb_1] ; (p0^q0)&1
1173 pavgb m3, m0 ; (p1 - q1 + 256)>>1
1174 pavgb m3, [pb_3] ; (((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2
1176 pavgb m4, m2 ; (q0 - p0 + 256)>>1
1178 paddusb m3, m4 ; d+128+33
1191 ; %1=p1 %2=q2 %3=[q2] %4=[q1] %5=tc0 %6=tmp
1192 ; out: [q1] = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
1193 ; clobbers: q2, tmp, tc0
1196 pavgb %2, %6 ; avg(p2,avg(p0,q0))
1198 pand %6, [pb_1] ; (p2^avg(p0,q0))&1
1199 psubusb %2, %6 ; (p2+((p0+q0+1)>>1))>>1
1208 ;-----------------------------------------------------------------------------
1209 ; void deblock_v_luma( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1210 ;-----------------------------------------------------------------------------
1211 %macro DEBLOCK_LUMA 0
1212 cglobal deblock_v_luma, 5,5,10
1216 add r4, r0 ; pix-3*stride
1218 mova m0, [r4+r1] ; p1
1219 mova m1, [r4+2*r1] ; p0
1221 mova m3, [r0+r1] ; q1
1225 pshufb m8, [pb_unpackbd1]
1226 pblendvb m9, m7, m6, m8
1229 punpcklbw m8, m8 ; tc = 4x tc0[3], 4x tc0[2], 4x tc0[1], 4x tc0[0]
1237 DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
1239 psubb m7, m8, m6 ; tc++
1241 LUMA_Q1 m0, m3, [r4], [r4+r1], m6, m4
1243 mova m4, [r0+2*r1] ; q2
1244 DIFF_GT2 m2, m4, m5, m6, m3 ; |q2-q0| > beta-1
1249 LUMA_Q1 m3, m4, [r0+2*r1], [r0+r1], m8, m6
1256 ;-----------------------------------------------------------------------------
1257 ; void deblock_h_luma( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1258 ;-----------------------------------------------------------------------------
1265 cglobal deblock_h_luma, 5,9,0,0x60+16*WIN64
1269 %xdefine pix_tmp rsp+0x30*WIN64 ; shadow space + r4
1271 ; transpose 6x16 -> tmp space
1272 TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r1, r8), pix_tmp
1275 TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r1, r8), pix_tmp+8
1278 ; alpha, beta, tc0 are still in r2d, r3d, r4
1279 ; don't backup r6, r5, r7, r8 because deblock_v_luma_sse2 doesn't use them
1281 lea r0, [pix_tmp+0x30]
1288 ; transpose 16x4 -> original space (only the middle 4 rows were changed by the filter)
1292 mova m0, [pix_tmp+0x10]
1293 mova m1, [pix_tmp+0x20]
1294 mova m2, [pix_tmp+0x30]
1295 mova m3, [pix_tmp+0x40]
1296 SBUTTERFLY bw, 0, 1, 4
1297 SBUTTERFLY bw, 2, 3, 4
1298 SBUTTERFLY wd, 0, 2, 4
1299 SBUTTERFLY wd, 1, 3, 4
1300 STORE_8x4B PASS8ROWS(r6, r5, r7, r8), m1, m3
1305 STORE_8x4B PASS8ROWS(r6, r5, r7, r8), m0, m2
1307 movq m0, [pix_tmp+0x18]
1308 movq m1, [pix_tmp+0x28]
1309 movq m2, [pix_tmp+0x38]
1310 movq m3, [pix_tmp+0x48]
1311 TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r7, r8)
1317 movq m0, [pix_tmp+0x10]
1318 movq m1, [pix_tmp+0x20]
1319 movq m2, [pix_tmp+0x30]
1320 movq m3, [pix_tmp+0x40]
1321 TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r7, r8)
1334 %macro DEBLOCK_LUMA 2
1335 ;-----------------------------------------------------------------------------
1336 ; void deblock_v8_luma( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1337 ;-----------------------------------------------------------------------------
1338 cglobal deblock_%1_luma, 5,5,8,2*%2
1341 add r4, r0 ; pix-3*stride
1343 mova m0, [r4+r1] ; p1
1344 mova m1, [r4+2*r1] ; p0
1346 mova m3, [r0+r1] ; q1
1352 pshufb m4, [pb_unpackbd1]
1353 mova [esp+%2], m4 ; tc
1354 pblendvb m4, m7, m6, m4
1357 punpcklbw m4, m4 ; tc = 4x tc0[3], 4x tc0[2], 4x tc0[1], 4x tc0[0]
1358 mova [esp+%2], m4 ; tc
1363 mova [esp], m4 ; mask
1366 DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
1368 pand m4, [esp+%2] ; tc
1371 LUMA_Q1 m0, m3, [r4], [r4+r1], m6, m4
1373 mova m4, [r0+2*r1] ; q2
1374 DIFF_GT2 m2, m4, m5, m6, m3 ; |q2-q0| > beta-1
1375 mova m5, [esp] ; mask
1377 mova m5, [esp+%2] ; tc
1381 LUMA_Q1 m3, m4, [r0+2*r1], [r0+r1], m5, m6
1388 ;-----------------------------------------------------------------------------
1389 ; void deblock_h_luma( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1390 ;-----------------------------------------------------------------------------
1396 cglobal deblock_h_luma, 1,5,8,0x60+12
1401 %define pix_tmp esp+12
1402 ; esp is intentionally misaligned to make it aligned after pushing the arguments for deblock_%1_luma.
1404 ; transpose 6x16 -> tmp space
1405 TRANSPOSE6x8_MEM PASS8ROWS(r0, r1, r3, r4), pix_tmp
1408 TRANSPOSE6x8_MEM PASS8ROWS(r0, r1, r3, r4), pix_tmp+8
1411 lea r0, [pix_tmp+0x30]
1417 call deblock_%1_luma
1419 add dword [esp ], 8 ; pix_tmp+0x38
1420 add dword [esp+16], 2 ; tc0+2
1421 call deblock_%1_luma
1425 ; transpose 16x4 -> original space (only the middle 4 rows were changed by the filter)
1431 mova m0, [pix_tmp+0x10]
1432 mova m1, [pix_tmp+0x20]
1433 mova m2, [pix_tmp+0x30]
1434 mova m3, [pix_tmp+0x40]
1435 SBUTTERFLY bw, 0, 1, 4
1436 SBUTTERFLY bw, 2, 3, 4
1437 SBUTTERFLY wd, 0, 2, 4
1438 SBUTTERFLY wd, 1, 3, 4
1439 STORE_8x4B PASS8ROWS(r0, r1, r3, r4), m0, m2
1442 STORE_8x4B PASS8ROWS(r0, r1, r3, r4), m1, m3
1444 movq m0, [pix_tmp+0x10]
1445 movq m1, [pix_tmp+0x20]
1446 movq m2, [pix_tmp+0x30]
1447 movq m3, [pix_tmp+0x40]
1448 TRANSPOSE8x4B_STORE PASS8ROWS(r0, r1, r3, r4)
1452 movq m0, [pix_tmp+0x18]
1453 movq m1, [pix_tmp+0x28]
1454 movq m2, [pix_tmp+0x38]
1455 movq m3, [pix_tmp+0x48]
1456 TRANSPOSE8x4B_STORE PASS8ROWS(r0, r1, r3, r4)
1460 %endmacro ; DEBLOCK_LUMA
1473 %macro LUMA_INTRA_P012 4 ; p0..p3 in memory
1483 pavgb t0, t1 ; ((p2+p1+1)/2 + (p0+q0+1)/2 + 1)/2
1501 psubb t0, t2 ; p1' = (p2+p1+p0+q0+2)/4;
1513 psubb t3, t2 ; p2+2*p1+2*p0+2*q0+q1
1517 pavgb t1, t5 ; (((p2+q1)/2 + p1+1)/2 + (p0+q0+1)/2 + 1)/2
1522 psubb t1, t3 ; p0'a = (p2+2*p1+2*p0+2*q0+q1+4)/8
1528 pavgb t2, p1 ; p0'b = (2*p1+p0+q0+2)/4
1536 mova %1, t1 ; store p0
1541 pavgb t1, t0 ; (p3+p2+1)/2 + (p2+p1+p0+q0+2)/4
1543 paddb t2, t4 ; 2*p3+3*p2+p1+p0+q0
1548 psubb t1, t2 ; p2' = (2*p3+3*p2+p1+p0+q0+4)/8
1556 mova %2, t0 ; store p1
1557 mova %3, t1 ; store p2
1560 %macro LUMA_INTRA_SWAP_PQ 0
1566 %define mask1p mask1q
1569 %macro DEBLOCK_LUMA_INTRA 1
1586 %define mask1q [rsp]
1588 %define mask1q [rsp-24]
1593 %define spill(x) [esp+16*x]
1595 %define q2 [r0+2*r1]
1598 %define mask0 spill(2)
1599 %define mask1p spill(3)
1600 %define mask1q spill(4)
1601 %define mpb_0 [pb_0]
1602 %define mpb_1 [pb_1]
1605 ;-----------------------------------------------------------------------------
1606 ; void deblock_v_luma_intra( uint8_t *pix, intptr_t stride, int alpha, int beta )
1607 ;-----------------------------------------------------------------------------
1608 cglobal deblock_%1_luma_intra, 4,6,16,0-(1-ARCH_X86_64)*0x50-WIN64*0x10
1610 lea r5, [r1*3] ; 3*stride
1612 add r4, r0 ; pix-4*stride
1620 LOAD_MASK r2d, r3d, t5 ; m5=beta-1, t5=alpha-1, m7=mask0
1621 SWAP 7, 12 ; m12=mask0
1623 pavgb t5, mpb_1 ; alpha/4+1
1625 movdqa q2, [r0+2*r1]
1626 DIFF_GT2 p0, q0, t5, t0, t3 ; t0 = |p0-q0| > alpha/4+1
1627 DIFF_GT2 p0, p2, m5, t2, t5, 1 ; mask1 = |p2-p0| > beta-1
1628 DIFF_GT2 q0, q2, m5, t4, t5, 1 ; t4 = |q2-q0| > beta-1
1635 LOAD_MASK r2d, r3d, t5 ; m5=beta-1, t5=alpha-1, m7=mask0
1639 pavgb m4, [pb_1] ; alpha/4+1
1640 DIFF_GT2 p0, q0, m4, m6, m7 ; m6 = |p0-q0| > alpha/4+1
1642 DIFF_GT2 p0, p2, m5, m4, m7, 1 ; m4 = |p2-p0| > beta-1
1645 DIFF_GT2 q0, q2, m5, m4, m7, 1 ; m4 = |q2-q0| > beta-1
1649 LUMA_INTRA_P012 [r4+r5], [r4+2*r1], [r4+r1], [r4]
1651 LUMA_INTRA_P012 [r0], [r0+r1], [r0+2*r1], [r0+r5]
1661 ;-----------------------------------------------------------------------------
1662 ; void deblock_h_luma_intra( uint8_t *pix, intptr_t stride, int alpha, int beta )
1663 ;-----------------------------------------------------------------------------
1664 cglobal deblock_h_luma_intra, 4,9,0,0x80
1669 %define pix_tmp rsp+0x20 ; shadow space
1674 ; transpose 8x16 -> tmp space
1675 TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r1, r8), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30)
1678 TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r1, r8), PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30)
1681 lea r0, [pix_tmp+0x40]
1683 call deblock_v_luma_intra
1685 ; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
1687 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r6, r5, r7, r8)
1692 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r6, r5, r7, r8)
1695 cglobal deblock_h_luma_intra, 2,4,8,0x80
1701 ; transpose 8x16 -> tmp space
1702 TRANSPOSE8x8_MEM PASS8ROWS(r0, r2, r1, r3), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30)
1705 TRANSPOSE8x8_MEM PASS8ROWS(r0, r2, r1, r3), PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30)
1707 lea r0, [pix_tmp+0x40]
1712 call deblock_%1_luma_intra
1714 add dword [rsp], 8 ; pix_tmp+8
1715 call deblock_%1_luma_intra
1724 ; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
1725 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r0, r2, r1, r3)
1728 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r0, r2, r1, r3)
1730 %endif ; ARCH_X86_64
1731 %endmacro ; DEBLOCK_LUMA_INTRA
1734 DEBLOCK_LUMA_INTRA v
1736 DEBLOCK_LUMA_INTRA v
1737 %if ARCH_X86_64 == 0
1739 DEBLOCK_LUMA_INTRA v8
1741 %endif ; !HIGH_BIT_DEPTH
1744 ; in: %1=p0, %2=q0, %3=p1, %4=q1, %5=mask, %6=tmp, %7=tmp
1745 ; out: %1=p0', %2=q0'
1746 %macro CHROMA_DEBLOCK_P0_Q0_INTRA 7
1766 %macro CHROMA_H_LOAD 0-1
1767 movq m0, [r0-8] ; p1 p1 p0 p0
1768 movq m2, [r0] ; q0 q0 q1 q1
1774 punpckldq m0, m5 ; p1
1775 punpckhdq m1, m5 ; p0
1776 punpckldq m2, m7 ; q0
1777 punpckhdq m3, m7 ; q1
1779 movq m4, [r0+r1*2-8]
1783 punpckldq m0, m5 ; p1 ... p0 ...
1784 punpckldq m2, m7 ; q0 ... q1 ...
1787 punpckhqdq m1, m0, m4 ; p0
1788 punpcklqdq m0, m4 ; p1
1789 punpckhqdq m3, m2, m6 ; q1
1790 punpcklqdq m2, m6 ; q0
1794 %macro CHROMA_V_LOAD 1
1796 mova m1, [r0+r1] ; p0
1798 mova m3, [%1+r1] ; q1
1801 ; clobbers: m1, m2, m3
1802 %macro CHROMA_H_STORE 0-1
1803 SBUTTERFLY dq, 1, 2, 3
1809 movq [r0+r1*2-4], m2
1810 movhps [r0+r1-4], m1
1811 movhps [r0+%1-4], m2
1815 %macro CHROMA_V_STORE 0
1820 %macro DEBLOCK_CHROMA 0
1821 cglobal deblock_inter_body
1822 LOAD_AB m4, m5, r2d, r3d
1823 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
1828 DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
1831 ;-----------------------------------------------------------------------------
1832 ; void deblock_v_chroma( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1833 ;-----------------------------------------------------------------------------
1834 cglobal deblock_v_chroma, 5,7,8
1842 call deblock_inter_body
1851 ;-----------------------------------------------------------------------------
1852 ; void deblock_h_chroma( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1853 ;-----------------------------------------------------------------------------
1854 cglobal deblock_h_chroma, 5,7,8
1862 call deblock_inter_body
1864 lea r0, [r0+r1*(mmsize/4)]
1871 cglobal deblock_intra_body
1872 LOAD_AB m4, m5, r2d, r3d
1873 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
1874 CHROMA_DEBLOCK_P0_Q0_INTRA m1, m2, m0, m3, m7, m5, m6
1877 ;-----------------------------------------------------------------------------
1878 ; void deblock_v_chroma_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
1879 ;-----------------------------------------------------------------------------
1880 cglobal deblock_v_chroma_intra, 4,6,8
1890 call deblock_intra_body
1898 ;-----------------------------------------------------------------------------
1899 ; void deblock_h_chroma_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
1900 ;-----------------------------------------------------------------------------
1901 cglobal deblock_h_chroma_intra, 4,6,8
1909 call deblock_intra_body
1911 lea r0, [r0+r1*(mmsize/4)]
1916 ;-----------------------------------------------------------------------------
1917 ; void deblock_h_chroma_intra_mbaff( uint16_t *pix, intptr_t stride, int alpha, int beta )
1918 ;-----------------------------------------------------------------------------
1919 cglobal deblock_h_chroma_intra_mbaff, 4,6,8
1928 LOAD_AB m4, m5, r2d, r3d
1929 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
1930 CHROMA_DEBLOCK_P0_Q0_INTRA m1, m2, m0, m3, m7, m5, m6
1933 lea r0, [r0+r1*(mmsize/4)]
1939 ;-----------------------------------------------------------------------------
1940 ; void deblock_h_chroma_mbaff( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1941 ;-----------------------------------------------------------------------------
1942 cglobal deblock_h_chroma_mbaff, 5,7,8
1950 LOAD_AB m4, m5, r2d, r3d
1951 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
1957 DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
1960 lea r0, [r0+r1*(mmsize/4)]
1967 ;-----------------------------------------------------------------------------
1968 ; void deblock_h_chroma_422_intra( uint16_t *pix, intptr_t stride, int alpha, int beta )
1969 ;-----------------------------------------------------------------------------
1970 cglobal deblock_h_chroma_422_intra, 4,6,8
1978 call deblock_intra_body
1980 lea r0, [r0+r1*(mmsize/4)]
1985 ;-----------------------------------------------------------------------------
1986 ; void deblock_h_chroma_422( uint16_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
1987 ;-----------------------------------------------------------------------------
1988 cglobal deblock_h_chroma_422, 5,7,8
1994 LOAD_AB m4, m5, r2m, r3d
1995 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
2002 DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
2004 lea r0, [r0+r1*(mmsize/4)]
2010 add r4, r2 ; increment once every 2 iterations
2015 %endmacro ; DEBLOCK_CHROMA
2017 %if ARCH_X86_64 == 0
2025 %endif ; HIGH_BIT_DEPTH
2027 %if HIGH_BIT_DEPTH == 0
2028 %macro CHROMA_V_START 0
2038 %macro CHROMA_H_START 0
2045 %macro CHROMA_V_LOOP 1
2057 %macro CHROMA_H_LOOP 1
2072 %macro DEBLOCK_CHROMA 0
2073 cglobal chroma_inter_body
2082 ;-----------------------------------------------------------------------------
2083 ; void deblock_v_chroma( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
2084 ;-----------------------------------------------------------------------------
2085 cglobal deblock_v_chroma, 5,6,8
2091 call chroma_inter_body
2097 ;-----------------------------------------------------------------------------
2098 ; void deblock_h_chroma( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
2099 ;-----------------------------------------------------------------------------
2100 cglobal deblock_h_chroma, 5,7,8
2106 TRANSPOSE4x8W_LOAD PASS8ROWS(t5, r0, r1, t6)
2107 call chroma_inter_body
2108 TRANSPOSE8x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
2111 %endmacro ; DEBLOCK_CHROMA
2117 %if ARCH_X86_64 == 0
2122 ;-----------------------------------------------------------------------------
2123 ; void deblock_h_chroma_mbaff( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 )
2124 ;-----------------------------------------------------------------------------
2125 %macro DEBLOCK_H_CHROMA_420_MBAFF 0
2126 cglobal deblock_h_chroma_mbaff, 5,7,8
2128 TRANSPOSE4x4W_LOAD PASS8ROWS(t5, r0, r1, t6)
2134 TRANSPOSE4x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
2139 DEBLOCK_H_CHROMA_420_MBAFF
2140 %if ARCH_X86_64 == 0
2142 DEBLOCK_H_CHROMA_420_MBAFF
2145 %macro DEBLOCK_H_CHROMA_422 0
2146 cglobal deblock_h_chroma_422, 5,8,8
2150 %define cntr dword r0m
2155 TRANSPOSE4x8W_LOAD PASS8ROWS(t5, r0, r1, t6)
2163 pshufw m6, m6, q0000
2167 TRANSPOSE8x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
2168 lea r0, [r0+r1*(mmsize/2)]
2169 lea t5, [t5+r1*(mmsize/2)]
2177 DEBLOCK_H_CHROMA_422
2179 DEBLOCK_H_CHROMA_422
2181 DEBLOCK_H_CHROMA_422
2183 ; in: %1=p0 %2=p1 %3=q1
2184 ; out: p0 = (p0 + q1 + 2*p1 + 2) >> 2
2185 %macro CHROMA_INTRA_P0 3
2187 pand m4, [pb_1] ; m4 = (p0^q1)&1
2190 pavgb %1, %2 ; dst = avg(p1, avg(p0,q1) - ((p0^q1)&1))
2196 %macro DEBLOCK_CHROMA_INTRA_BODY 0
2197 cglobal chroma_intra_body
2201 CHROMA_INTRA_P0 m1, m0, m3
2202 CHROMA_INTRA_P0 m2, m3, m0
2212 %macro DEBLOCK_CHROMA_INTRA 0
2213 ;-----------------------------------------------------------------------------
2214 ; void deblock_v_chroma_intra( uint8_t *pix, intptr_t stride, int alpha, int beta )
2215 ;-----------------------------------------------------------------------------
2216 cglobal deblock_v_chroma_intra, 4,5,8
2222 call chroma_intra_body
2228 ;-----------------------------------------------------------------------------
2229 ; void deblock_h_chroma_intra( uint8_t *pix, intptr_t stride, int alpha, int beta )
2230 ;-----------------------------------------------------------------------------
2231 cglobal deblock_h_chroma_intra, 4,6,8
2237 TRANSPOSE4x8W_LOAD PASS8ROWS(t5, r0, r1, t6)
2238 call chroma_intra_body
2239 TRANSPOSE8x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
2243 cglobal deblock_h_chroma_422_intra, 4,7,8
2247 TRANSPOSE4x8W_LOAD PASS8ROWS(t5, r0, r1, t6)
2248 call chroma_intra_body
2249 TRANSPOSE8x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
2250 lea r0, [r0+r1*(mmsize/2)]
2251 lea t5, [t5+r1*(mmsize/2)]
2255 %endmacro ; DEBLOCK_CHROMA_INTRA
2258 DEBLOCK_CHROMA_INTRA_BODY
2259 DEBLOCK_CHROMA_INTRA
2261 DEBLOCK_CHROMA_INTRA_BODY
2262 DEBLOCK_CHROMA_INTRA
2264 DEBLOCK_CHROMA_INTRA_BODY
2265 %if ARCH_X86_64 == 0
2266 DEBLOCK_CHROMA_INTRA
2269 ;-----------------------------------------------------------------------------
2270 ; void deblock_h_chroma_intra_mbaff( uint8_t *pix, intptr_t stride, int alpha, int beta )
2271 ;-----------------------------------------------------------------------------
2273 cglobal deblock_h_chroma_intra_mbaff, 4,6,8
2275 TRANSPOSE4x4W_LOAD PASS8ROWS(t5, r0, r1, t6)
2276 call chroma_intra_body
2277 TRANSPOSE4x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
2279 %endif ; !HIGH_BIT_DEPTH
2283 ;-----------------------------------------------------------------------------
2284 ; static void deblock_strength( uint8_t nnz[48], int8_t ref[2][40], int16_t mv[2][40][2],
2285 ; uint8_t bs[2][4][4], int mvy_limit, int bframe )
2286 ;-----------------------------------------------------------------------------
2288 %define scan8start (4+1*8)
2289 %define nnz r0+scan8start
2290 %define ref r1+scan8start
2291 %define mv r2+scan8start*4
2295 %macro LOAD_BYTES_MMX 1
2300 punpckldq m2, [%1+8*1-1]
2301 punpckldq m0, [%1+8*1]
2302 punpckldq m3, [%1+8*3-1]
2303 punpckldq m1, [%1+8*3]
2306 %macro DEBLOCK_STRENGTH_REFS_MMX 0
2317 punpckldq m2, m0 ; row -1, row 0
2318 punpckldq m3, m1 ; row 1, row 2
2327 %macro DEBLOCK_STRENGTH_MVS_MMX 2
2340 %macro DEBLOCK_STRENGTH_NNZ_MMX 1
2347 pminub m4, m6 ; mv ? 1 : 0
2349 paddb m2, m2 ; nnz ? 2 : 0
2355 %macro LOAD_BYTES_XMM 1
2356 movu m2, [%1-4] ; FIXME could be aligned if we changed nnz's allocation
2359 shufps m2, m1, q3131 ; cur nnz, all rows
2361 shufps m0, m1, q3131 ; left neighbors
2363 movd m3, [%1-8] ; could be palignr if nnz was aligned
2364 por m1, m3 ; top neighbors
2368 cglobal deblock_strength, 6,6
2369 ; Prepare mv comparison register
2382 DEBLOCK_STRENGTH_REFS_MMX
2385 DEBLOCK_STRENGTH_MVS_MMX bs0, 4
2386 DEBLOCK_STRENGTH_MVS_MMX bs1, 4*8
2399 DEBLOCK_STRENGTH_NNZ_MMX bs0
2400 ; Transpose column output
2401 SBUTTERFLY bw, 2, 3, 4
2402 SBUTTERFLY bw, 2, 3, 4
2407 punpckldq m2, m0 ; row -1, row 0
2408 punpckldq m3, m1 ; row 1, row 2
2409 DEBLOCK_STRENGTH_NNZ_MMX bs1
2414 %macro DEBLOCK_STRENGTH_XMM 0
2415 cglobal deblock_strength, 6,6,7
2416 ; Prepare mv comparison register
2436 palignr m3, m0, [mv+4*8*0-16], 12
2437 palignr m2, m1, [mv+4*8*1-16], 12
2444 palignr m3, m2, [mv+4*8*2-16], 12
2446 palignr m3, m1, [mv+4*8*3-16], 12
2450 movu m0, [mv-4+4*8*0]
2451 movu m1, [mv-4+4*8*1]
2452 movu m2, [mv-4+4*8*2]
2453 movu m3, [mv-4+4*8*3]
2454 psubw m0, [mv+4*8*0]
2455 psubw m1, [mv+4*8*1]
2456 psubw m2, [mv+4*8*2]
2457 psubw m3, [mv+4*8*3]
2468 mova m0, [mv+4*8*-1]
2469 mova m1, [mv+4*8* 0]
2470 mova m2, [mv+4*8* 1]
2471 mova m3, [mv+4*8* 2]
2475 psubw m3, [mv+4*8* 3]
2496 pminub m4, m6 ; mv ? 1 : 0
2498 paddb m0, m0 ; nnz ? 2 : 0
2503 pshufb m4, [transpose_shuf]
2516 DEBLOCK_STRENGTH_XMM
2518 DEBLOCK_STRENGTH_XMM
2520 DEBLOCK_STRENGTH_XMM
2522 %macro LOAD_BYTES_YMM 1
2523 movu m0, [%1-4] ; ___E FGHI ___J KLMN ___O PQRS ___T UVWX
2524 pshufb m0, [load_bytes_shuf] ; EFGH JKLM FGHI KLMN OPQR TUVW PQRS UVWX
2525 mova m2, [insert_top_shuf]
2526 vpermq m1, m0, q3131 ; FGHI KLMN PQRS UVWX x2
2527 vpermd m0, m2, m0 ; EFGH JKLM OPQR TUVW ____ FGHI KLMN PQRS
2528 vpbroadcastd m2, [%1-8] ; ABCD ....
2529 vpblendd m0, m0, m2, 00010000b ; EFGH JKLM OPQR TUVW ABCD FGHI KLMN PQRS
2533 cglobal deblock_strength, 6,6,7
2534 ; Prepare mv comparison register
2538 vpbroadcastw m6, xm6
2539 pxor m5, m5 ; bs0,bs1
2548 movu xm0, [mv-4+4*8*0]
2549 vinserti128 m0, m0, [mv+4*8*-1], 1
2550 vbroadcasti128 m2, [mv+4*8* 0]
2551 vinserti128 m1, m2, [mv-4+4*8*1], 0
2552 vbroadcasti128 m3, [mv+4*8* 1]
2556 vinserti128 m2, m3, [mv-4+4*8*2], 0
2557 vbroadcasti128 m4, [mv+4*8* 2]
2558 vinserti128 m3, m4, [mv-4+4*8*3], 0
2560 vbroadcasti128 m4, [mv+4*8* 3]
2581 pminub m5, m6 ; mv ? 1 : 0
2582 paddb m0, m0 ; nnz ? 2 : 0
2584 vextracti128 [bs1], m5, 1
2585 pshufb xm5, [transpose_shuf]