1 ;*****************************************************************************
2 ;* deblock-a.asm: x86 deblocking
3 ;*****************************************************************************
4 ;* Copyright (C) 2005-2011 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 transpose_shuf: db 0,4,8,12,1,5,9,13,2,6,10,14,3,7,11,15
47 ; out: %4 = |%1-%2|-%3
58 ; out: %4 = |%1-%2|<%3
66 psubw %5, %3 ; |%1-%2|-%3
67 pcmpgtw %4, %5 ; 0 > |%1-%2|-%3
85 pshuflw %1, %1, 01010000b
86 pshufd %1, %1, 01010000b
91 ; in: %1=p1, %2=p0, %3=q0, %4=q1
92 ; %5=alpha, %6=beta, %7-%9=tmp
95 ABS_SUB %2, %3, %5, %8, %7 ; |p0-q0| - alpha
96 ABS_SUB %1, %2, %6, %9, %7 ; |p1-p0| - beta
98 ABS_SUB %3, %4, %6, %9, %7 ; |q1-q0| - beta
104 ; in: %1=p0, %2=q0, %3=p1, %4=q1, %5=mask, %6=tmp, %7=tmp
105 ; out: %1=p0', m2=q0'
106 %macro DEBLOCK_P0_Q0 7
116 mova %6, [pw_pixel_max]
125 ; in: %1=x2, %2=x1, %3=p0, %4=q0 %5=mask&tc, %6=tmp
128 pavgw %6, %4 ; (p0+q0+1)>>1
138 %macro LUMA_DEBLOCK_ONE 3
139 DIFF_LT m5, %1, bm, m4, m6
146 LUMA_Q1 m5, %2, m1, m2, m4, m6
149 %macro LUMA_H_STORE 2
161 movhps [%1+r1*2-4], m2
163 movhps [%1+r1*4-4], m3
167 %macro DEBLOCK_LUMA 1
168 ;-----------------------------------------------------------------------------
169 ; void deblock_v_luma( uint16_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
170 ;-----------------------------------------------------------------------------
171 cglobal deblock_v_luma_%1, 5,5,8*(mmsize/16)
172 %assign pad 5*mmsize+12-(stack_offset&15)
174 %define ms1 [rsp+mmsize]
175 %define ms2 [rsp+mmsize*2]
176 %define am [rsp+mmsize*3]
177 %define bm [rsp+mmsize*4]
180 LOAD_AB m4, m5, r2, r3
194 LOAD_MASK m0, m1, m2, m3, am, bm, m7, m4, m6
199 LUMA_DEBLOCK_ONE m1, m0, ms1
203 LUMA_DEBLOCK_ONE m2, m3, ms2
213 DEBLOCK_P0_Q0 m1, m2, m0, m3, m5, m7, m6
225 cglobal deblock_h_luma_%1, 5,6,8*(mmsize/16)
226 %assign pad 7*mmsize+12-(stack_offset&15)
228 %define ms1 [rsp+mmsize]
229 %define ms2 [rsp+mmsize*2]
230 %define p1m [rsp+mmsize*3]
231 %define p2m [rsp+mmsize*4]
232 %define am [rsp+mmsize*5]
233 %define bm [rsp+mmsize*6]
236 LOAD_AB m4, m5, r2, r3
249 movq m2, [r0-8] ; y q2 q1 q0
256 TRANSPOSE4x4W 2, 5, 0, 1, 4
259 TRANSPOSE4x4W 2, 3, 6, 7, 4
261 movu m5, [r0-8] ; y q2 q1 q0 p0 p1 p2 x
265 TRANSPOSE4x4W 5, 0, 2, 3, 6
272 TRANSPOSE4x4W 4, 1, 3, 7, 6
277 SBUTTERFLY qdq, 0, 1, 7
278 SBUTTERFLY qdq, 2, 3, 7
282 LOAD_MASK m0, m1, m2, m3, am, bm, m7, m4, m6
286 LUMA_DEBLOCK_ONE m1, m0, ms1
290 LUMA_DEBLOCK_ONE m2, m3, ms2
300 DEBLOCK_P0_Q0 m1, m2, m0, m3, m5, m7, m6
303 TRANSPOSE4x4W 0, 1, 2, 3, 4
307 lea r0, [r0+r1*(mmsize/2)]
308 lea r2, [r2+r1*(mmsize/2)]
317 ; in: m0=p1, m1=p0, m2=q0, m3=q1, m8=p2, m9=q2
318 ; m12=alpha, m13=beta
319 ; out: m0=p1', m3=q1', m1=p0', m2=q0'
320 ; clobbers: m4, m5, m6, m7, m10, m11, m14
321 %macro DEBLOCK_LUMA_INTER_SSE2 0
322 LOAD_MASK m0, m1, m2, m3, m12, m13, m7, m4, m6
324 DIFF_LT m8, m1, m13, m10, m4
325 DIFF_LT m9, m2, m13, m11, m4
335 LUMA_Q1 m8, m0, m1, m2, m5, m4
339 LUMA_Q1 m9, m3, m1, m2, m5, m4
347 DEBLOCK_P0_Q0 m1, m2, m0, m3, m4, m5, m6
353 cglobal deblock_v_luma_sse2, 5,5,15
364 LOAD_AB m12, m13, r2, r3
377 DEBLOCK_LUMA_INTER_SSE2
389 cglobal deblock_h_luma_sse2, 5,7,15
391 LOAD_AB m12, m13, r2, r3
399 movu m8, [r0-8] ; y q2 q1 q0 p0 p1 p2 x
408 TRANSPOSE4x4W 8, 0, 2, 9, 10
409 TRANSPOSE4x4W 5, 1, 3, 7, 10
412 SBUTTERFLY qdq, 0, 1, 10
413 SBUTTERFLY qdq, 2, 3, 10
416 DEBLOCK_LUMA_INTER_SSE2
418 TRANSPOSE4x4W 0, 1, 2, 3, 4
436 ; in: t0-t2: tmp registers
437 ; %1=p0 %2=p1 %3=p2 %4=p3 %5=q0 %6=q1 %7=mask0
438 ; %8=mask1p %9=2 %10=p0' %11=p1' %12=p2'
439 %macro LUMA_INTRA_P012 12 ; p0..p3 in memory
448 paddw t0, %9 ; (p2 + p1 + p0 + q0 + 2)
449 paddw t2, t0 ; (2*p3 + 3*p2 + p1 + p0 + q0 + 4)
470 psrlw t1, 2 ; (2*p1 + p0 + q1 + 2)/4
471 psrlw t0, 3 ; (p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4)>>3
483 %macro LUMA_INTRA_INIT 1
484 %xdefine pad %1*mmsize+((gprsize*3) % mmsize)-(stack_offset&15)
491 CAT_XDEFINE t, i, [rsp+mmsize*(i-4)]
498 ; in: %1-%3=tmp, %4=p2, %5=q2
499 %macro LUMA_INTRA_INTER 5
500 LOAD_AB t0, t1, r2d, r3d
502 LOAD_MASK m0, m1, m2, m3, %1, t1, t0, t2, t3
506 paddw t3, [pw_2] ; alpha/4+2
507 DIFF_LT m1, m2, t3, t2, t0 ; t2 = |p0-q0| < alpha/4+2
511 DIFF_LT t3, m2, t1, t2, t0 ; t2 = |q2-q0| < beta
515 DIFF_LT t3, m1, t1, t2, t0 ; t2 = |p2-p0| < beta
520 %macro LUMA_H_INTRA_LOAD 0
526 TRANSPOSE4x4W 4, 5, 0, 1, 2
534 TRANSPOSE4x4W 2, 3, 4, 5, 6
546 TRANSPOSE8x8W 4, 5, 0, 1, 2, 3, 6, 7, t4, t5
554 ; in: %1=q3 %2=q2' %3=q1' %4=q0' %5=p0' %6=p1' %7=p2' %8=p3 %9=tmp
555 %macro LUMA_H_INTRA_STORE 9
557 TRANSPOSE4x4W %1, %2, %3, %4, %9
560 movq [r0+r1*2-8], m%3
563 TRANSPOSE4x4W %5, %6, %7, %1, %9
569 TRANSPOSE2x4x4W %1, %2, %3, %4, %9
572 movq [r0+r1*2-8], m%3
575 movhps [r4+r1-8], m%2
576 movhps [r4+r1*2-8], m%3
577 movhps [r4+r5-8], m%4
583 TRANSPOSE2x4x4W %5, %6, %7, %1, %9
590 movhps [r4+r1*2], m%7
597 ;-----------------------------------------------------------------------------
598 ; void deblock_v_luma_intra( uint16_t *pix, int stride, int alpha, int beta )
599 ;-----------------------------------------------------------------------------
600 cglobal deblock_v_luma_intra_sse2, 4,7,16
614 lea r5, [r1*3] ; 3*stride
616 add r4, r0 ; pix-4*stride
619 LOAD_AB aa, bb, r2d, r3d
628 LOAD_MASK p1, p0, q0, q1, aa, bb, m3, t0, t1
631 paddw t2, m0 ; alpha/4+2
632 DIFF_LT p0, q0, t2, m6, t0 ; m6 = |p0-q0| < alpha/4+2
633 DIFF_LT p2, p0, bb, t1, t0 ; m7 = |p2-p0| < beta
634 DIFF_LT q2, q0, bb, m7, t0 ; t1 = |q2-q0| < beta
638 LUMA_INTRA_P012 p0, p1, p2, [r4], q0, q1, m3, m6, m0, [r4+r5], [r4+2*r1], [r4+r1]
639 LUMA_INTRA_P012 q0, q1, q2, [r0+r5], p0, p1, m3, m7, m0, [r0], [r0+r1], [r0+2*r1]
646 ;-----------------------------------------------------------------------------
647 ; void deblock_h_luma_intra( uint16_t *pix, int stride, int alpha, int beta )
648 ;-----------------------------------------------------------------------------
649 cglobal deblock_h_luma_intra_sse2, 4,7,16
662 %assign pad 24-(stack_offset&15)
666 lea r5, [r1*3] ; 3*stride
667 add r4, r0 ; pix+4*stride
679 TRANSPOSE8x8W 5, 8, 9, 10, 11, 12, 13, 4, 1
681 LOAD_AB m1, m2, r2d, r3d
682 LOAD_MASK q1, q0, p0, p1, m1, m2, m3, t0, t1
684 paddw m1, m0 ; alpha/4+2
685 DIFF_LT p0, q0, m1, m6, t0 ; m6 = |p0-q0| < alpha/4+2
686 DIFF_LT q2, q0, m2, t1, t0 ; t1 = |q2-q0| < beta
687 DIFF_LT p0, p2, m2, m7, t0 ; m7 = |p2-p0| < beta
693 LUMA_INTRA_P012 q0, q1, q2, q3, p0, p1, m3, m6, m0, m5, m1, q2
694 LUMA_INTRA_P012 p0, p1, p2, p3, q0, q1, m3, m7, m0, p0, m6, p2
697 LUMA_H_INTRA_STORE 7, 8, 1, 5, 11, 6, 13, 4, 14
707 %macro DEBLOCK_LUMA_INTRA 1
708 ;-----------------------------------------------------------------------------
709 ; void deblock_v_luma_intra( uint16_t *pix, int stride, int alpha, int beta )
710 ;-----------------------------------------------------------------------------
711 cglobal deblock_v_luma_intra_%1, 4,7,8*(mmsize/16)
719 mova m0, [r4+r1*2] ; p1
720 mova m1, [r4+r5] ; p0
722 mova m3, [r0+r1] ; q1
723 LUMA_INTRA_INTER t4, t5, t6, [r4+r1], [r0+r1*2]
724 LUMA_INTRA_P012 m1, m0, t3, [r4], m2, m3, t5, t4, [pw_2], [r4+r5], [r4+2*r1], [r4+r1]
725 mova t3, [r0+r1*2] ; q2
726 LUMA_INTRA_P012 m2, m3, t3, [r0+r5], m1, m0, t5, t6, [pw_2], [r0], [r0+r1], [r0+2*r1]
734 ;-----------------------------------------------------------------------------
735 ; void deblock_h_luma_intra( uint16_t *pix, int stride, int alpha, int beta )
736 ;-----------------------------------------------------------------------------
737 cglobal deblock_h_luma_intra_%1, 4,7,8*(mmsize/16)
744 lea r5, [r1*3] ; 3*stride
745 add r4, r0 ; pix+4*stride
750 LUMA_INTRA_INTER t8, t9, t10, t5, t6
752 LUMA_INTRA_P012 m1, m0, t3, t4, m2, m3, t9, t8, [pw_2], t8, t5, t11
754 LUMA_INTRA_P012 m2, m3, t3, t7, m1, m0, t9, t10, [pw_2], m4, t6, m5
762 LUMA_H_INTRA_STORE 2, 0, 1, 3, 4, 6, 5, t7, 7
764 lea r0, [r0+r1*(mmsize/2)]
768 lea r4, [r4+r1*(mmsize/2)]
779 DEBLOCK_LUMA_INTRA mmxext
782 DEBLOCK_LUMA_INTRA sse2
784 %endif ; HIGH_BIT_DEPTH
786 %ifndef HIGH_BIT_DEPTH
787 ; expands to [base],...,[base+7*stride]
788 %define PASS8ROWS(base, base3, stride, stride3) \
789 [base], [base+stride], [base+stride*2], [base3], \
790 [base3+stride], [base3+stride*2], [base3+stride3], [base3+stride*4]
792 %define PASS8ROWS(base, base3, stride, stride3, offset) \
793 PASS8ROWS(base+offset, base3+offset, stride, stride3)
795 ; in: 8 rows of 4 bytes in %4..%11
796 ; out: 4 rows of 8 bytes in m0..m3
797 %macro TRANSPOSE4x8_LOAD 11
826 ; in: 4 rows of 8 bytes in m0..m3
827 ; out: 8 rows of 4 bytes in %1..%8
828 %macro TRANSPOSE8x4B_STORE 8
862 %macro TRANSPOSE4x8B_LOAD 8
863 TRANSPOSE4x8_LOAD bw, wd, dq, %1, %2, %3, %4, %5, %6, %7, %8
866 %macro TRANSPOSE4x8W_LOAD 8
868 TRANSPOSE4x8_LOAD wd, dq, qdq, %1, %2, %3, %4, %5, %6, %7, %8
875 TRANSPOSE4x4W 0, 1, 2, 3, 4
879 %macro TRANSPOSE8x2W_STORE 8
914 ; in: 8 rows of 8 (only the middle 6 pels are used) in %1..%8
915 ; out: 6 rows of 8 in [%9+0*16] .. [%9+5*16]
916 %macro TRANSPOSE6x8_MEM 9
925 SBUTTERFLY bw, 0, 1, 7
926 SBUTTERFLY bw, 2, 3, 7
927 SBUTTERFLY bw, 4, 5, 7
929 SBUTTERFLY3 bw, m6, %8, m7
930 SBUTTERFLY wd, 0, 2, 3
931 SBUTTERFLY wd, 4, 6, 3
934 SBUTTERFLY3 wd, m1, [%9+0x10], m3
935 SBUTTERFLY wd, 5, 7, 0
936 SBUTTERFLY dq, 1, 5, 0
937 SBUTTERFLY dq, 2, 6, 0
947 ; in: 8 rows of 8 in %1..%8
948 ; out: 8 rows of 8 in %9..%16
949 %macro TRANSPOSE8x8_MEM 16
958 SBUTTERFLY bw, 0, 1, 7
959 SBUTTERFLY bw, 2, 3, 7
960 SBUTTERFLY bw, 4, 5, 7
961 SBUTTERFLY3 bw, m6, %8, m7
963 SBUTTERFLY wd, 0, 2, 5
964 SBUTTERFLY wd, 4, 6, 5
965 SBUTTERFLY wd, 1, 3, 5
968 SBUTTERFLY wd, 6, 7, 5
969 SBUTTERFLY dq, 0, 4, 5
970 SBUTTERFLY dq, 1, 6, 5
975 SBUTTERFLY3 dq, m2, %11, m0
976 SBUTTERFLY dq, 3, 7, 4
984 ; out: %4 = |%1-%2|>%3
995 ; out: %4 = |%1-%2|>%3
1007 ; in: m0=p1 m1=p0 m2=q0 m3=q1 %1=alpha-1 %2=beta-1
1008 ; out: m5=beta-1, m7=mask, %3=alpha-1
1010 %macro LOAD_MASK 2-3
1015 packuswb m4, m4 ; 16x alpha-1
1016 packuswb m5, m5 ; 16x beta-1
1020 DIFF_GT m1, m2, m4, m7, m6 ; |p0-q0| > alpha-1
1021 DIFF_GT m0, m1, m5, m4, m6 ; |p1-p0| > beta-1
1023 DIFF_GT m3, m2, m5, m4, m6 ; |q1-q0| > beta-1
1029 ; in: m0=p1 m1=p0 m2=q0 m3=q1 m7=(tc&mask)
1030 ; out: m1=p0' m2=q0'
1032 %macro DEBLOCK_P0_Q0 0
1035 pand m5, [pb_1] ; (p0^q0)&1
1038 pavgb m3, m0 ; (p1 - q1 + 256)>>1
1039 pavgb m3, [pb_3] ; (((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2
1041 pavgb m4, m2 ; (q0 - p0 + 256)>>1
1043 paddusb m3, m4 ; d+128+33
1056 ; %1=p1 %2=q2 %3=[q2] %4=[q1] %5=tc0 %6=tmp
1057 ; out: [q1] = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
1058 ; clobbers: q2, tmp, tc0
1062 pavgb %2, %6 ; avg(p2,avg(p0,q0))
1064 pand %6, [pb_1] ; (p2^avg(p0,q0))&1
1065 psubusb %2, %6 ; (p2+((p0+q0+1)>>1))>>1
1075 ;-----------------------------------------------------------------------------
1076 ; void deblock_v_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1077 ;-----------------------------------------------------------------------------
1079 cglobal deblock_v_luma_sse2, 5,5,10
1085 add r4, r0 ; pix-3*stride
1087 mova m0, [r4+r1] ; p1
1088 mova m1, [r4+2*r1] ; p0
1090 mova m3, [r0+r1] ; q1
1094 punpcklbw m8, m8 ; tc = 4x tc0[3], 4x tc0[2], 4x tc0[1], 4x tc0[0]
1100 movdqa m3, [r4] ; p2
1101 DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
1106 LUMA_Q1 m0, m3, [r4], [r4+r1], m6, m4
1108 movdqa m4, [r0+2*r1] ; q2
1109 DIFF_GT2 m2, m4, m5, m6, m3 ; |q2-q0| > beta-1
1114 LUMA_Q1 m3, m4, [r0+2*r1], [r0+r1], m8, m6
1121 ;-----------------------------------------------------------------------------
1122 ; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1123 ;-----------------------------------------------------------------------------
1125 cglobal deblock_h_luma_sse2, 5,7
1127 lea r11, [r10+r10*2]
1132 %define pix_tmp rsp+0x30
1138 ; transpose 6x16 -> tmp space
1139 TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r10, r11), pix_tmp
1142 TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r10, r11), pix_tmp+8
1145 ; alpha, beta, tc0 are still in r2d, r3d, r4
1146 ; don't backup r6, r5, r10, r11 because deblock_v_luma_sse2 doesn't use them
1147 lea r0, [pix_tmp+0x30]
1152 call deblock_v_luma_sse2
1154 ; transpose 16x4 -> original space (only the middle 4 rows were changed by the filter)
1157 movq m0, [pix_tmp+0x18]
1158 movq m1, [pix_tmp+0x28]
1159 movq m2, [pix_tmp+0x38]
1160 movq m3, [pix_tmp+0x48]
1161 TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r10, r11)
1167 movq m0, [pix_tmp+0x10]
1168 movq m1, [pix_tmp+0x20]
1169 movq m2, [pix_tmp+0x30]
1170 movq m3, [pix_tmp+0x40]
1171 TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r10, r11)
1182 %macro DEBLOCK_LUMA 3
1183 ;-----------------------------------------------------------------------------
1184 ; void deblock_v8_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1185 ;-----------------------------------------------------------------------------
1186 cglobal deblock_%2_luma_%1, 5,5
1191 add r4, r0 ; pix-3*stride
1192 %assign pad 2*%3+12-(stack_offset&15)
1195 mova m0, [r4+r1] ; p1
1196 mova m1, [r4+2*r1] ; p0
1198 mova m3, [r0+r1] ; q1
1204 punpcklbw m4, m4 ; tc = 4x tc0[3], 4x tc0[2], 4x tc0[1], 4x tc0[0]
1205 mova [esp+%3], m4 ; tc
1209 mova [esp], m4 ; mask
1212 DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
1214 pand m4, [esp+%3] ; tc
1218 LUMA_Q1 m0, m3, [r4], [r4+r1], m6, m4
1220 mova m4, [r0+2*r1] ; q2
1221 DIFF_GT2 m2, m4, m5, m6, m3 ; |q2-q0| > beta-1
1222 mova m5, [esp] ; mask
1224 mova m5, [esp+%3] ; tc
1228 LUMA_Q1 m3, m4, [r0+2*r1], [r0+r1], m5, m6
1236 ;-----------------------------------------------------------------------------
1237 ; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1238 ;-----------------------------------------------------------------------------
1240 cglobal deblock_h_luma_%1, 0,5
1246 %assign pad 0x78-(stack_offset&15)
1248 %define pix_tmp esp+12
1250 ; transpose 6x16 -> tmp space
1251 TRANSPOSE6x8_MEM PASS8ROWS(r0, r1, r3, r4), pix_tmp
1254 TRANSPOSE6x8_MEM PASS8ROWS(r0, r1, r3, r4), pix_tmp+8
1257 lea r0, [pix_tmp+0x30]
1263 call deblock_%2_luma_%1
1265 add dword [esp ], 8 ; pix_tmp+0x38
1266 add dword [esp+16], 2 ; tc0+2
1267 call deblock_%2_luma_%1
1271 ; transpose 16x4 -> original space (only the middle 4 rows were changed by the filter)
1276 movq m0, [pix_tmp+0x10]
1277 movq m1, [pix_tmp+0x20]
1278 movq m2, [pix_tmp+0x30]
1279 movq m3, [pix_tmp+0x40]
1280 TRANSPOSE8x4B_STORE PASS8ROWS(r0, r1, r3, r4)
1284 movq m0, [pix_tmp+0x18]
1285 movq m1, [pix_tmp+0x28]
1286 movq m2, [pix_tmp+0x38]
1287 movq m3, [pix_tmp+0x48]
1288 TRANSPOSE8x4B_STORE PASS8ROWS(r0, r1, r3, r4)
1292 %endmacro ; DEBLOCK_LUMA
1295 DEBLOCK_LUMA mmxext, v8, 8
1297 DEBLOCK_LUMA sse2, v, 16
1303 %macro LUMA_INTRA_P012 4 ; p0..p3 in memory
1308 pavgb t0, t1 ; ((p2+p1+1)/2 + (p0+q0+1)/2 + 1)/2
1321 psubb t0, t2 ; p1' = (p2+p1+p0+q0+2)/4;
1328 psubb t3, t2 ; p2+2*p1+2*p0+2*q0+q1
1332 pavgb t1, t5 ; (((p2+q1)/2 + p1+1)/2 + (p0+q0+1)/2 + 1)/2
1337 psubb t1, t3 ; p0'a = (p2+2*p1+2*p0+2*q0+q1+4)/8
1345 pavgb t2, p1 ; p0'b = (2*p1+p0+q0+2)/4
1353 mova %1, t1 ; store p0
1359 pavgb t1, t0 ; (p3+p2+1)/2 + (p2+p1+p0+q0+2)/4
1361 paddb t2, t4 ; 2*p3+3*p2+p1+p0+q0
1366 psubb t1, t2 ; p2' = (2*p3+3*p2+p1+p0+q0+4)/8
1374 mova %2, t0 ; store p1
1375 mova %3, t1 ; store p2
1378 %macro LUMA_INTRA_SWAP_PQ 0
1384 %define mask1p mask1q
1387 %macro DEBLOCK_LUMA_INTRA 2
1403 %define mask1q [rsp-24]
1407 %define spill(x) [esp+16*x+((stack_offset+4)&15)]
1409 %define q2 [r0+2*r1]
1412 %define mask0 spill(2)
1413 %define mask1p spill(3)
1414 %define mask1q spill(4)
1415 %define mpb_0 [pb_0]
1416 %define mpb_1 [pb_1]
1419 ;-----------------------------------------------------------------------------
1420 ; void deblock_v_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
1421 ;-----------------------------------------------------------------------------
1422 cglobal deblock_%2_luma_intra_%1, 4,6,16
1427 lea r5, [r1*3] ; 3*stride
1433 add r4, r0 ; pix-4*stride
1441 LOAD_MASK r2d, r3d, t5 ; m5=beta-1, t5=alpha-1, m7=mask0
1442 SWAP 7, 12 ; m12=mask0
1444 pavgb t5, mpb_1 ; alpha/4+1
1446 movdqa q2, [r0+2*r1]
1447 DIFF_GT2 p0, q0, t5, t0, t3 ; t0 = |p0-q0| > alpha/4+1
1448 DIFF_GT2 p0, p2, m5, t2, t5 ; mask1 = |p2-p0| > beta-1
1449 DIFF_GT2 q0, q2, m5, t4, t5 ; t4 = |q2-q0| > beta-1
1456 LOAD_MASK r2d, r3d, t5 ; m5=beta-1, t5=alpha-1, m7=mask0
1460 pavgb m4, [pb_1] ; alpha/4+1
1461 DIFF_GT2 p0, q0, m4, m6, m7 ; m6 = |p0-q0| > alpha/4+1
1463 DIFF_GT2 p0, p2, m5, m4, m7 ; m4 = |p2-p0| > beta-1
1466 DIFF_GT2 q0, q2, m5, m4, m7 ; m4 = |q2-q0| > beta-1
1470 LUMA_INTRA_P012 [r4+r5], [r4+2*r1], [r4+r1], [r4]
1472 LUMA_INTRA_P012 [r0], [r0+r1], [r0+2*r1], [r0+r5]
1481 ;-----------------------------------------------------------------------------
1482 ; void deblock_h_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
1483 ;-----------------------------------------------------------------------------
1484 cglobal deblock_h_luma_intra_%1, 4,7
1492 ; transpose 8x16 -> tmp space
1493 TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r10, r11), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30)
1496 TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r10, r11), PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30)
1498 lea r0, [pix_tmp+0x40]
1500 call deblock_v_luma_intra_%1
1502 ; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
1504 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r6, r5, r10, r11)
1509 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r6, r5, r10, r11)
1513 cglobal deblock_h_luma_intra_%1, 2,4
1517 %assign pad 0x8c-(stack_offset&15)
1521 ; transpose 8x16 -> tmp space
1522 TRANSPOSE8x8_MEM PASS8ROWS(r0, r2, r1, r3), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30)
1525 TRANSPOSE8x8_MEM PASS8ROWS(r0, r2, r1, r3), PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30)
1527 lea r0, [pix_tmp+0x40]
1532 call deblock_%2_luma_intra_%1
1534 add dword [rsp], 8 ; pix_tmp+8
1535 call deblock_%2_luma_intra_%1
1544 ; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
1545 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r0, r2, r1, r3)
1548 TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r0, r2, r1, r3)
1551 %endif ; ARCH_X86_64
1552 %endmacro ; DEBLOCK_LUMA_INTRA
1555 DEBLOCK_LUMA_INTRA sse2, v
1558 DEBLOCK_LUMA_INTRA mmxext, v8
1560 %endif ; !HIGH_BIT_DEPTH
1562 %ifdef HIGH_BIT_DEPTH
1563 ; in: %1=p0, %2=q0, %3=p1, %4=q1, %5=mask, %6=tmp, %7=tmp
1564 ; out: %1=p0', %2=q0'
1565 %macro CHROMA_DEBLOCK_P0_Q0_INTRA 7
1586 %macro CHROMA_H_LOAD 0-1
1587 movq m0, [r0-8] ; p1 p1 p0 p0
1588 movq m2, [r0] ; q0 q0 q1 q1
1594 punpckldq m0, m5 ; p1
1595 punpckhdq m1, m5 ; p0
1596 punpckldq m2, m7 ; q0
1597 punpckhdq m3, m7 ; q1
1599 movq m4, [r0+r1*2-8]
1603 punpckldq m0, m5 ; p1 ... p0 ...
1604 punpckldq m2, m7 ; q0 ... q1 ...
1609 punpcklqdq m0, m4 ; p1
1610 punpckhqdq m1, m4 ; p0
1611 punpcklqdq m2, m6 ; q0
1612 punpckhqdq m3, m6 ; q1
1616 %macro CHROMA_V_LOAD 1
1618 mova m1, [r0+r1] ; p0
1620 mova m3, [%1+r1] ; q1
1623 ; clobbers: m1, m2, m3
1624 %macro CHROMA_H_STORE 0-1
1625 SBUTTERFLY dq, 1, 2, 3
1631 movq [r0+r1*2-4], m2
1632 movhps [r0+r1-4], m1
1633 movhps [r0+%1-4], m2
1637 %macro CHROMA_V_STORE 0
1642 %macro DEBLOCK_CHROMA 1
1643 ;-----------------------------------------------------------------------------
1644 ; void deblock_v_chroma( uint16_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1645 ;-----------------------------------------------------------------------------
1646 cglobal deblock_v_chroma_%1, 7,7,8*(mmsize/16)
1654 call deblock_inter_body_%1
1663 ;-----------------------------------------------------------------------------
1664 ; void deblock_h_chroma( uint16_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1665 ;-----------------------------------------------------------------------------
1666 cglobal deblock_h_chroma_%1, 5,7,8*(mmsize/16)
1674 call deblock_inter_body_%1
1676 lea r0, [r0+r1*(mmsize/4)]
1682 deblock_inter_body_%1:
1683 RESET_MM_PERMUTATION
1684 LOAD_AB m4, m5, r2, r3
1685 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
1690 DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
1693 ;-----------------------------------------------------------------------------
1694 ; void deblock_v_chroma_intra( uint16_t *pix, int stride, int alpha, int beta )
1695 ;-----------------------------------------------------------------------------
1696 cglobal deblock_v_chroma_intra_%1, 4,6,8*(mmsize/16)
1706 call deblock_intra_body_%1
1714 ;-----------------------------------------------------------------------------
1715 ; void deblock_h_chroma_intra( uint16_t *pix, int stride, int alpha, int beta )
1716 ;-----------------------------------------------------------------------------
1717 cglobal deblock_h_chroma_intra_%1, 4,6,8*(mmsize/16)
1725 call deblock_intra_body_%1
1727 lea r0, [r0+r1*(mmsize/4)]
1732 deblock_intra_body_%1:
1733 RESET_MM_PERMUTATION
1734 LOAD_AB m4, m5, r2, r3
1735 LOAD_MASK m0, m1, m2, m3, m4, m5, m7, m6, m4
1736 CHROMA_DEBLOCK_P0_Q0_INTRA m1, m2, m0, m3, m7, m5, m6
1742 DEBLOCK_CHROMA mmxext
1746 %endif ; HIGH_BIT_DEPTH
1748 %ifndef HIGH_BIT_DEPTH
1749 %macro CHROMA_V_START 0
1761 %macro CHROMA_H_START 0
1774 %macro CHROMA_V_LOOP 1
1786 %macro CHROMA_H_LOOP 1
1801 %macro DEBLOCK_CHROMA 1
1802 ;-----------------------------------------------------------------------------
1803 ; void deblock_v_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1804 ;-----------------------------------------------------------------------------
1805 cglobal deblock_v_chroma_%1, 5,6,8
1811 call chroma_inter_body_%1
1817 ;-----------------------------------------------------------------------------
1818 ; void deblock_h_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
1819 ;-----------------------------------------------------------------------------
1820 cglobal deblock_h_chroma_%1, 5,7,8
1822 TRANSPOSE4x8W_LOAD PASS8ROWS(t5, r0, r1, t6)
1823 call chroma_inter_body_%1
1824 TRANSPOSE8x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
1829 RESET_MM_PERMUTATION
1830 chroma_inter_body_%1:
1838 %endmacro ; DEBLOCK_CHROMA
1844 DEBLOCK_CHROMA mmxext
1848 ; in: %1=p0 %2=p1 %3=q1
1849 ; out: p0 = (p0 + q1 + 2*p1 + 2) >> 2
1850 %macro CHROMA_INTRA_P0 3
1853 pand m4, [pb_1] ; m4 = (p0^q1)&1
1856 pavgb %1, %2 ; dst = avg(p1, avg(p0,q1) - ((p0^q1)&1))
1862 %macro DEBLOCK_CHROMA_INTRA 1
1863 ;-----------------------------------------------------------------------------
1864 ; void deblock_v_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
1865 ;-----------------------------------------------------------------------------
1866 cglobal deblock_v_chroma_intra_%1, 4,5,8
1872 call chroma_intra_body_%1
1878 ;-----------------------------------------------------------------------------
1879 ; void deblock_h_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
1880 ;-----------------------------------------------------------------------------
1881 cglobal deblock_h_chroma_intra_%1, 4,6,8
1883 TRANSPOSE4x8W_LOAD PASS8ROWS(t5, r0, r1, t6)
1884 call chroma_intra_body_%1
1885 TRANSPOSE8x2W_STORE PASS8ROWS(t5, r0, r1, t6, 2)
1890 RESET_MM_PERMUTATION
1891 chroma_intra_body_%1:
1895 CHROMA_INTRA_P0 m1, m0, m3
1896 CHROMA_INTRA_P0 m2, m3, m0
1904 %endmacro ; DEBLOCK_CHROMA_INTRA
1907 DEBLOCK_CHROMA_INTRA sse2
1910 DEBLOCK_CHROMA_INTRA mmxext
1912 %endif ; !HIGH_BIT_DEPTH
1916 ;-----------------------------------------------------------------------------
1917 ; static void deblock_strength( uint8_t nnz[48], int8_t ref[2][40], int16_t mv[2][40][2],
1918 ; uint8_t bs[2][4][4], int mvy_limit, int bframe )
1919 ;-----------------------------------------------------------------------------
1921 %define scan8start (4+1*8)
1922 %define nnz r0+scan8start
1923 %define ref r1+scan8start
1924 %define mv r2+scan8start*4
1928 %macro LOAD_BYTES_MMX 1
1933 punpckldq m2, [%1+8*1-1]
1934 punpckldq m0, [%1+8*1]
1935 punpckldq m3, [%1+8*3-1]
1936 punpckldq m1, [%1+8*3]
1939 %macro DEBLOCK_STRENGTH_REFS_MMX 0
1950 punpckldq m2, m0 ; row -1, row 0
1951 punpckldq m3, m1 ; row 1, row 2
1960 %macro DEBLOCK_STRENGTH_MVS_MMX 2
1973 %macro DEBLOCK_STRENGTH_NNZ_MMX 1
1980 pminub m4, m6 ; mv ? 1 : 0
1982 paddb m2, m2 ; nnz ? 2 : 0
1988 %macro LOAD_BYTES_XMM 1
1989 movu m0, [%1-4] ; FIXME could be aligned if we changed nnz's allocation
1993 shufps m2, m1, 0xdd ; cur nnz, all rows
1995 shufps m0, m1, 0xdd ; left neighbors
1997 movd m3, [%1-8] ; could be palignr if nnz was aligned
1999 por m1, m3 ; top neighbors
2003 cglobal deblock_strength_mmxext, 6,6
2004 ; Prepare mv comparison register
2017 DEBLOCK_STRENGTH_REFS_MMX
2020 DEBLOCK_STRENGTH_MVS_MMX bs0, 4
2021 DEBLOCK_STRENGTH_MVS_MMX bs1, 4*8
2034 DEBLOCK_STRENGTH_NNZ_MMX bs0
2035 ; Transpose column output
2036 SBUTTERFLY bw, 2, 3, 4
2037 SBUTTERFLY bw, 2, 3, 4
2042 punpckldq m2, m0 ; row -1, row 0
2043 punpckldq m3, m1 ; row 1, row 2
2044 DEBLOCK_STRENGTH_NNZ_MMX bs1
2049 %macro DEBLOCK_STRENGTH_XMM 1
2050 cglobal deblock_strength_%1, 6,6,8
2051 ; Prepare mv comparison register
2073 palignr m3, [mv+4*8*0-16], 12
2074 palignr m2, [mv+4*8*1-16], 12
2083 palignr m3, [mv+4*8*2-16], 12
2084 palignr m7, [mv+4*8*3-16], 12
2089 movu m0, [mv-4+4*8*0]
2090 movu m1, [mv-4+4*8*1]
2091 movu m2, [mv-4+4*8*2]
2092 movu m3, [mv-4+4*8*3]
2093 psubw m0, [mv+4*8*0]
2094 psubw m1, [mv+4*8*1]
2095 psubw m2, [mv+4*8*2]
2096 psubw m3, [mv+4*8*3]
2100 ABSB2 m0, m2, m1, m3
2106 mova m0, [mv+4*8*-1]
2107 mova m1, [mv+4*8* 0]
2108 mova m2, [mv+4*8* 1]
2109 mova m3, [mv+4*8* 2]
2113 psubw m3, [mv+4*8* 3]
2116 ABSB2 m0, m2, m1, m3
2133 pminub m4, m6 ; mv ? 1 : 0
2135 paddb m0, m0 ; nnz ? 2 : 0
2140 pshufb m4, [transpose_shuf]
2153 DEBLOCK_STRENGTH_XMM sse2
2154 %define ABSB2 ABSB2_SSSE3
2155 DEBLOCK_STRENGTH_XMM ssse3