1 ;*****************************************************************************
2 ;* pixel.asm: h264 encoder library
3 ;*****************************************************************************
4 ;* Copyright (C) 2003-2008 x264 project
6 ;* Authors: Loren Merritt <lorenm@u.washington.edu>
7 ;* Laurent Aimar <fenrir@via.ecp.fr>
8 ;* Alex Izvorski <aizvorksi@gmail.com>
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.
23 ;*****************************************************************************
26 %include "x86util.asm"
30 ssim_c1: times 4 dd 416 ; .01*.01*255*255*64
31 ssim_c2: times 4 dd 235963 ; .03*.03*255*255*64*63
32 mask_ff: times 16 db 0xff
34 mask_ac4: dw 0,-1,-1,-1, 0,-1,-1,-1
35 mask_ac8: dw 0,-1,-1,-1,-1,-1,-1,-1
39 %macro HADDD 2 ; sum junk
53 pmaddwd %1, [pw_1 GLOBAL]
66 ;=============================================================================
68 ;=============================================================================
144 pinsrd m1, [r0+%1], 1
145 pinsrd m2, [r2+%2], 1
146 pinsrd m3, [r0+%3], 1
147 pinsrd m4, [r2+%4], 1
169 ;-----------------------------------------------------------------------------
170 ; int x264_pixel_ssd_16x16_mmx( uint8_t *, int, uint8_t *, int )
171 ;-----------------------------------------------------------------------------
173 cglobal x264_pixel_ssd_%1x%2_%3, 4,4
180 SSD_FULL 0, 0, mmsize, mmsize, i, 0
181 SSD_FULL r1, r3, r1+mmsize, r3+mmsize, 1, i<%2/2-1
183 SSD_FULL 0, 0, r1, r3, i, i<%2/2-1
185 SSD_HALF 0, 0, r1, r3, i, i<%2/2-1
209 cglobal x264_pixel_ssd_4x8_sse4, 4,4
210 SSD_QUARTER 0, 0, r1, r3, 0, 1
211 SSD_QUARTER 0, 0, r1, r3, 1, 0
216 cglobal x264_pixel_ssd_4x4_sse4, 4,4
217 SSD_QUARTER 0, 0, r1, r3, 0, 0
223 ;=============================================================================
225 ;=============================================================================
229 pxor m6, m6 ; sum squared
245 movd [r2], m5 ; return sum
250 sub eax, r1d ; sqr - (sum * sum >> shift)
288 ;-----------------------------------------------------------------------------
289 ; int x264_pixel_var_wxh_mmxext( uint8_t *, int, int * )
290 ;-----------------------------------------------------------------------------
292 cglobal x264_pixel_var_16x16_mmxext, 2,3
297 cglobal x264_pixel_var_8x8_mmxext, 2,3
303 cglobal x264_pixel_var_16x16_sse2, 2,3
308 cglobal x264_pixel_var_8x8_sse2, 2,3
330 ;=============================================================================
332 ;=============================================================================
334 ; phaddw is used only in 4x4 hadamard, because in 8x8 it's slower:
335 ; even on Penryn, phaddw has latency 3 while paddw and punpck* have 1.
336 ; 4x4 is special in that 4x4 transpose in xmmregs takes extra munging,
337 ; whereas phaddw-based transform doesn't care what order the coefs end up in.
346 %macro HADAMARD4_ROW_PHADD 5
354 %macro HADAMARD4_1D 4
355 SUMSUB_BADC %1, %2, %3, %4
356 SUMSUB_BADC %1, %3, %2, %4
359 %macro HADAMARD4x4_SUM 1 ; %1 = dest (row sum of one block)
361 HADAMARD4_1D m4, m5, m6, m7
362 TRANSPOSE4x4W 4, 5, 6, 7, %%n
363 HADAMARD4_1D m4, m5, m6, m7
364 ABS2 m4, m5, m3, m %+ %%n
365 ABS2 m6, m7, m3, m %+ %%n
372 ; in: r4=3*stride1, r5=3*stride2
373 ; in: %2 = horizontal offset
374 ; in: %3 = whether we need to increment pix1 and pix2
377 %macro SATD_4x4_MMX 3
378 LOAD_DIFF m4, m3, none, [r0+%2], [r2+%2]
379 LOAD_DIFF m5, m3, none, [r0+r1+%2], [r2+r3+%2]
380 LOAD_DIFF m6, m3, none, [r0+2*r1+%2], [r2+2*r3+%2]
381 LOAD_DIFF m7, m3, none, [r0+r4+%2], [r2+r5+%2]
389 %macro SATD_8x4_SSE2 1
390 HADAMARD4_1D m0, m1, m2, m3
391 %ifidn %1, ssse3_phadd
392 HADAMARD4_ROW_PHADD 0, 1, 2, 3, 4
394 TRANSPOSE2x4x4W 0, 1, 2, 3, 4
395 HADAMARD4_1D m0, m1, m2, m3
397 ABS4 m0, m1, m2, m3, m4, m5
404 %macro SATD_START_MMX 0
405 lea r4, [3*r1] ; 3*stride1
406 lea r5, [3*r3] ; 3*stride2
409 %macro SATD_END_MMX 0
410 pshufw m1, m0, 01001110b
412 pshufw m1, m0, 10110001b
419 ; FIXME avoid the spilling of regs to hold 3*stride.
420 ; for small blocks on x86_32, modify pixel pointer instead.
422 ;-----------------------------------------------------------------------------
423 ; int x264_pixel_satd_16x16_mmxext (uint8_t *, int, uint8_t *, int )
424 ;-----------------------------------------------------------------------------
426 cglobal x264_pixel_satd_16x4_internal_mmxext
427 SATD_4x4_MMX m2, 0, 0
428 SATD_4x4_MMX m1, 4, 0
430 SATD_4x4_MMX m2, 8, 0
432 SATD_4x4_MMX m1, 12, 0
437 cglobal x264_pixel_satd_8x8_internal_mmxext
438 SATD_4x4_MMX m2, 0, 0
439 SATD_4x4_MMX m1, 4, 1
442 x264_pixel_satd_8x4_internal_mmxext:
443 SATD_4x4_MMX m2, 0, 0
444 SATD_4x4_MMX m1, 4, 0
449 cglobal x264_pixel_satd_16x16_mmxext, 4,6
453 call x264_pixel_satd_16x4_internal_mmxext
457 call x264_pixel_satd_16x4_internal_mmxext
462 cglobal x264_pixel_satd_16x8_mmxext, 4,6
465 call x264_pixel_satd_16x4_internal_mmxext
468 call x264_pixel_satd_16x4_internal_mmxext
471 cglobal x264_pixel_satd_8x16_mmxext, 4,6
474 call x264_pixel_satd_8x8_internal_mmxext
477 call x264_pixel_satd_8x8_internal_mmxext
480 cglobal x264_pixel_satd_8x8_mmxext, 4,6
483 call x264_pixel_satd_8x8_internal_mmxext
486 cglobal x264_pixel_satd_8x4_mmxext, 4,6
489 call x264_pixel_satd_8x4_internal_mmxext
492 cglobal x264_pixel_satd_4x8_mmxext, 4,6
494 SATD_4x4_MMX m0, 0, 1
495 SATD_4x4_MMX m1, 0, 0
501 cglobal x264_pixel_satd_4x4_%1, 4,6
503 SATD_4x4_MMX m0, 0, 0
509 %macro SATD_START_SSE2 0
515 %macro SATD_END_SSE2 0
522 %macro BACKUP_POINTERS 0
529 %macro RESTORE_AND_INC_POINTERS 0
541 ;-----------------------------------------------------------------------------
542 ; int x264_pixel_satd_8x4_sse2 (uint8_t *, int, uint8_t *, int )
543 ;-----------------------------------------------------------------------------
546 cglobal x264_pixel_satd_8x8_internal_%1
547 LOAD_DIFF_8x4P m0, m1, m2, m3, m4, m5
551 x264_pixel_satd_8x4_internal_%1:
552 LOAD_DIFF_8x4P m0, m1, m2, m3, m4, m5
553 x264_pixel_satd_4x8_internal_%1:
554 SAVE_MM_PERMUTATION satd_4x8_internal
558 cglobal x264_pixel_satd_16x16_%1, 4,6
561 call x264_pixel_satd_8x8_internal_%1
564 call x264_pixel_satd_8x8_internal_%1
565 RESTORE_AND_INC_POINTERS
566 call x264_pixel_satd_8x8_internal_%1
569 call x264_pixel_satd_8x8_internal_%1
572 cglobal x264_pixel_satd_16x8_%1, 4,6
575 call x264_pixel_satd_8x8_internal_%1
576 RESTORE_AND_INC_POINTERS
577 call x264_pixel_satd_8x8_internal_%1
580 cglobal x264_pixel_satd_8x16_%1, 4,6
582 call x264_pixel_satd_8x8_internal_%1
585 call x264_pixel_satd_8x8_internal_%1
588 cglobal x264_pixel_satd_8x8_%1, 4,6
590 call x264_pixel_satd_8x8_internal_%1
593 cglobal x264_pixel_satd_8x4_%1, 4,6
595 call x264_pixel_satd_8x4_internal_%1
598 cglobal x264_pixel_satd_4x8_%1, 4,6
600 LOAD_MM_PERMUTATION satd_4x8_internal
603 LOAD_DIFF m0, m7, m6, [r0], [r2]
604 LOAD_DIFF m1, m7, m6, [r0+r1], [r2+r3]
605 LOAD_DIFF m2, m7, m6, [r0+2*r1], [r2+2*r3]
606 LOAD_DIFF m3, m7, m6, [r0+r4], [r2+r5]
609 LOAD_DIFF m4, m7, m6, [r0], [r2]
610 LOAD_DIFF m5, m7, m6, [r0+r1], [r2+r3]
613 LOAD_DIFF m4, m7, m6, [r0+2*r1], [r2+2*r3]
614 LOAD_DIFF m5, m7, m6, [r0+r4], [r2+r5]
618 call x264_pixel_satd_4x8_internal_%1
622 ;-----------------------------------------------------------------------------
623 ; int x264_pixel_sa8d_8x8_sse2( uint8_t *, int, uint8_t *, int )
624 ;-----------------------------------------------------------------------------
625 cglobal x264_pixel_sa8d_8x8_internal_%1
628 LOAD_DIFF_8x4P m0, m1, m2, m3, m8, m9, r0, r2
629 LOAD_DIFF_8x4P m4, m5, m6, m7, m8, m9, r10, r11
631 HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
632 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
633 HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
635 ABS4 m0, m1, m2, m3, m8, m9
636 ABS4 m4, m5, m6, m7, m8, m9
646 cglobal x264_pixel_sa8d_8x8_%1, 4,6
649 call x264_pixel_sa8d_8x8_internal_%1
656 cglobal x264_pixel_sa8d_16x16_%1, 4,6
659 call x264_pixel_sa8d_8x8_internal_%1 ; pix[0]
663 call x264_pixel_sa8d_8x8_internal_%1 ; pix[8]
667 call x264_pixel_sa8d_8x8_internal_%1 ; pix[8*stride+8]
671 call x264_pixel_sa8d_8x8_internal_%1 ; pix[8*stride]
680 cglobal x264_pixel_sa8d_8x8_internal_%1
681 LOAD_DIFF_8x4P m0, m1, m2, m3, m6, m7
685 LOAD_DIFF_8x4P m4, m5, m6, m7, m2, m2
688 HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
689 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, [esp+4], [esp+20]
690 HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
713 %endmacro ; SATDS_SSE2
715 %macro SA8D_16x16_32 1
717 cglobal x264_pixel_sa8d_8x8_%1, 4,7
723 call x264_pixel_sa8d_8x8_internal_%1
731 cglobal x264_pixel_sa8d_16x16_%1, 4,7
737 call x264_pixel_sa8d_8x8_internal_%1
741 call x264_pixel_sa8d_8x8_internal_%1
748 call x264_pixel_sa8d_8x8_internal_%1
754 mova [esp+48-mmsize], m0
755 call x264_pixel_sa8d_8x8_internal_%1
756 paddusw m0, [esp+48-mmsize]
778 %endif ; !ARCH_X86_64
779 %endmacro ; SA8D_16x16_32
783 ;=============================================================================
785 ;=============================================================================
787 %macro INTRA_SA8D_SSE2 1
790 ;-----------------------------------------------------------------------------
791 ; void x264_intra_sa8d_x3_8x8_core_sse2( uint8_t *fenc, int16_t edges[2][8], int *res )
792 ;-----------------------------------------------------------------------------
793 cglobal x264_intra_sa8d_x3_8x8_core_%1
796 movq m0, [r0+0*FENC_STRIDE]
797 movq m1, [r0+1*FENC_STRIDE]
798 movq m2, [r0+2*FENC_STRIDE]
799 movq m3, [r0+3*FENC_STRIDE]
800 movq m4, [r0+4*FENC_STRIDE]
801 movq m5, [r0+5*FENC_STRIDE]
802 movq m6, [r0+6*FENC_STRIDE]
803 movq m7, [r0+7*FENC_STRIDE]
812 HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
813 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
814 HADAMARD8_1D m0, m1, m2, m3, m4, m5, m6, m7
817 movzx edi, word [r1+0]
828 ABS4 m8, m9, m10, m11, m12, m13
839 ABS2 m10, m11, m13, m14
846 movdqa m14, m15 ; 7x8 sum
848 movdqa m8, [r1+0] ; left edge
854 ABS1 m9, m11 ; 1x8 sum
863 punpcklqdq m0, m4 ; transpose
864 movdqa m1, [r1+16] ; top edge
867 psrldq m2, 2 ; 8x7 sum
868 psubw m0, m1 ; 8x1 sum
873 movdqa m7, [pw_1 GLOBAL]
889 movq [r2], m3 ; i8x8_v, i8x8_h
891 movd [r2+8], m3 ; i8x8_dc
894 %endmacro ; INTRA_SA8D_SSE2
897 ; out: m0..m3 = hadamard coefs
902 movd m0, [r0+0*FENC_STRIDE]
903 movd m1, [r0+1*FENC_STRIDE]
904 movd m2, [r0+2*FENC_STRIDE]
905 movd m3, [r0+3*FENC_STRIDE]
910 HADAMARD4_1D m0, m1, m2, m3
911 TRANSPOSE4x4W 0, 1, 2, 3, 4
912 HADAMARD4_1D m0, m1, m2, m3
913 SAVE_MM_PERMUTATION load_hadamard
916 %macro SCALAR_SUMSUB 4
925 %macro SCALAR_HADAMARD_LEFT 5 ; y, 4x tmp
927 shl %1d, 5 ; log(FDEC_STRIDE)
929 movzx %2d, byte [r1+%1-1+0*FDEC_STRIDE]
930 movzx %3d, byte [r1+%1-1+1*FDEC_STRIDE]
931 movzx %4d, byte [r1+%1-1+2*FDEC_STRIDE]
932 movzx %5d, byte [r1+%1-1+3*FDEC_STRIDE]
936 SCALAR_SUMSUB %2d, %3d, %4d, %5d
937 SCALAR_SUMSUB %2d, %4d, %3d, %5d
938 mov [left_1d+2*%1+0], %2w
939 mov [left_1d+2*%1+2], %3w
940 mov [left_1d+2*%1+4], %4w
941 mov [left_1d+2*%1+6], %5w
944 %macro SCALAR_HADAMARD_TOP 5 ; x, 4x tmp
945 movzx %2d, byte [r1+%1-FDEC_STRIDE+0]
946 movzx %3d, byte [r1+%1-FDEC_STRIDE+1]
947 movzx %4d, byte [r1+%1-FDEC_STRIDE+2]
948 movzx %5d, byte [r1+%1-FDEC_STRIDE+3]
949 SCALAR_SUMSUB %2d, %3d, %4d, %5d
950 SCALAR_SUMSUB %2d, %4d, %3d, %5d
951 mov [top_1d+2*%1+0], %2w
952 mov [top_1d+2*%1+2], %3w
953 mov [top_1d+2*%1+4], %4w
954 mov [top_1d+2*%1+6], %5w
957 %macro SUM_MM_X3 8 ; 3x sum, 4x tmp, op
959 pshufw %4, %1, 01001110b
960 pshufw %5, %2, 01001110b
961 pshufw %6, %3, 01001110b
968 pshufw %4, %1, 01001110b
969 pshufw %5, %2, 01001110b
970 pshufw %6, %3, 01001110b
978 mov qword [sums+0], 0
979 mov qword [sums+8], 0
980 mov qword [sums+16], 0
1009 ; in: m0..m3 (4x4), m7 (3x4)
1010 ; out: m0 v, m4 h, m5 dc
1012 %macro SUM4x3 3 ; dc, left, top
1020 punpckldq m0, m2 ; transpose
1024 ABS2 m4, m5, m2, m3 ; 1x4 sum
1025 ABS1 m0, m1 ; 4x1 sum
1028 %macro INTRA_SATDS_MMX 1
1030 ;-----------------------------------------------------------------------------
1031 ; void x264_intra_satd_x3_4x4_mmxext( uint8_t *fenc, uint8_t *fdec, int *res )
1032 ;-----------------------------------------------------------------------------
1033 cglobal x264_intra_satd_x3_4x4_%1, 2,6
1035 ; stack is 16 byte aligned because abi says so
1036 %define top_1d rsp-8 ; size 8
1037 %define left_1d rsp-16 ; size 8
1041 ; stack is 16 byte aligned at least in gcc, and we've pushed 3 regs + return address, so it's still aligned
1043 %define top_1d esp+8
1050 SCALAR_HADAMARD_LEFT 0, r0, r3, r4, r5
1052 SCALAR_HADAMARD_TOP 0, r0, r3, r4, r5
1053 lea t0d, [t0d + r0d + 4]
1058 SUM4x3 t0d, [left_1d], [top_1d]
1062 psrlq m1, 16 ; 4x3 sum
1065 SUM_MM_X3 m0, m4, m5, m1, m2, m3, m6, pavgw
1069 movd [r2+0], m0 ; i4x4_v satd
1070 movd [r2+4], m4 ; i4x4_h satd
1071 movd [r2+8], m5 ; i4x4_dc satd
1091 ;-----------------------------------------------------------------------------
1092 ; void x264_intra_satd_x3_16x16_mmxext( uint8_t *fenc, uint8_t *fdec, int *res )
1093 ;-----------------------------------------------------------------------------
1094 cglobal x264_intra_satd_x3_16x16_%1, 0,7
1096 %assign stack_pad 88
1098 %assign stack_pad 88 + ((stack_offset+88+4)&15)
1100 ; not really needed on x86_64, just shuts up valgrind about storing data below the stack across a function call
1102 %define sums rsp+64 ; size 24
1103 %define top_1d rsp+32 ; size 32
1104 %define left_1d rsp ; size 32
1112 SCALAR_HADAMARD_LEFT t0, r3, r4, r5, r6
1114 SCALAR_HADAMARD_TOP t0, r3, r4, r5, r6
1131 SUM4x3 t2d, [left_1d+8*r3], [top_1d+8*r4]
1134 paddw m0, [sums+0] ; i16x16_v satd
1135 paddw m4, [sums+8] ; i16x16_h satd
1136 paddw m5, [sums+16] ; i16x16_dc satd
1145 add r0, 4*FENC_STRIDE-16
1156 SUM_MM_X3 m0, m1, m2, m3, m4, m5, m6, paddd
1162 movd [r2+8], m2 ; i16x16_dc satd
1163 movd [r2+4], m1 ; i16x16_h satd
1164 movd [r2+0], m0 ; i16x16_v satd
1168 ;-----------------------------------------------------------------------------
1169 ; void x264_intra_satd_x3_8x8c_mmxext( uint8_t *fenc, uint8_t *fdec, int *res )
1170 ;-----------------------------------------------------------------------------
1171 cglobal x264_intra_satd_x3_8x8c_%1, 0,6
1172 ; not really needed on x86_64, just shuts up valgrind about storing data below the stack across a function call
1174 %define sums rsp+48 ; size 24
1175 %define dc_1d rsp+32 ; size 16
1176 %define top_1d rsp+16 ; size 16
1177 %define left_1d rsp ; size 16
1184 SCALAR_HADAMARD_LEFT t0, t2, r3, r4, r5
1185 SCALAR_HADAMARD_TOP t0, t2, r3, r4, r5
1190 movzx t2d, word [left_1d+0]
1191 movzx r3d, word [top_1d+0]
1192 movzx r4d, word [left_1d+8]
1193 movzx r5d, word [top_1d+8]
1204 mov [dc_1d+ 0], t2d ; tl
1205 mov [dc_1d+ 4], r5d ; tr
1206 mov [dc_1d+ 8], r4d ; bl
1207 mov [dc_1d+12], r3d ; br
1220 SUM4x3 [r5+4*r4], [left_1d+8*r3], [top_1d+8*r4]
1223 paddw m0, [sums+16] ; i4x4_v satd
1224 paddw m4, [sums+8] ; i4x4_h satd
1225 paddw m5, [sums+0] ; i4x4_dc satd
1234 add r0, 4*FENC_STRIDE-8
1247 SUM_MM_X3 m0, m1, m2, m3, m4, m5, m6, paddd
1249 movd [r2+0], m0 ; i8x8c_dc satd
1250 movd [r2+4], m1 ; i8x8c_h satd
1251 movd [r2+8], m2 ; i8x8c_v satd
1254 %endmacro ; INTRA_SATDS_MMX
1257 %macro ABS_MOV_SSSE3 2
1261 %macro ABS_MOV_MMX 2
1267 %define ABS_MOV ABS_MOV_MMX
1269 ; in: r0=pix, r1=stride, r2=stride*3, r3=tmp, m6=mask_ac4, m7=0
1270 ; out: [tmp]=hadamard4, m0=satd
1271 cglobal x264_hadamard_ac_4x4_mmxext
1280 HADAMARD4_1D m0, m1, m2, m3
1281 TRANSPOSE4x4W 0, 1, 2, 3, 4
1282 HADAMARD4_1D m0, m1, m2, m3
1295 SAVE_MM_PERMUTATION x264_hadamard_ac_4x4_mmxext
1298 cglobal x264_hadamard_ac_2x2_mmxext
1303 HADAMARD4_1D m0, m1, m2, m3
1306 SAVE_MM_PERMUTATION x264_hadamard_ac_2x2_mmxext
1309 cglobal x264_hadamard_ac_8x8_mmxext
1310 mova m6, [mask_ac4 GLOBAL]
1312 call x264_hadamard_ac_4x4_mmxext
1316 call x264_hadamard_ac_4x4_mmxext
1320 call x264_hadamard_ac_4x4_mmxext
1324 call x264_hadamard_ac_4x4_mmxext
1327 mova [rsp+gprsize+8], m5 ; save satd
1328 call x264_hadamard_ac_2x2_mmxext
1335 call x264_hadamard_ac_2x2_mmxext
1342 call x264_hadamard_ac_2x2_mmxext
1349 mova [rsp+gprsize], m6 ; save sa8d
1351 SAVE_MM_PERMUTATION x264_hadamard_ac_8x8_mmxext
1354 %macro HADAMARD_AC_WXH_MMX 2
1355 cglobal x264_pixel_hadamard_ac_%1x%2_mmxext, 2,4
1356 %assign pad 16-gprsize-(stack_offset&15)
1361 call x264_hadamard_ac_8x8_mmxext
1366 call x264_hadamard_ac_8x8_mmxext
1371 lea r0, [r0+ysub*4+8]
1373 call x264_hadamard_ac_8x8_mmxext
1377 call x264_hadamard_ac_8x8_mmxext
1382 paddusw m0, [rsp+0x10]
1383 paddusw m1, [rsp+0x18]
1387 paddusw m1, [rsp+0x28]
1388 paddusw m2, [rsp+0x30]
1390 paddusw m1, [rsp+0x38]
1392 pand m3, [pw_1 GLOBAL]
1409 add rsp, 128+%1*%2/4+pad
1411 %endmacro ; HADAMARD_AC_WXH_MMX
1413 HADAMARD_AC_WXH_MMX 16, 16
1414 HADAMARD_AC_WXH_MMX 8, 16
1415 HADAMARD_AC_WXH_MMX 16, 8
1416 HADAMARD_AC_WXH_MMX 8, 8
1418 %macro HADAMARD_AC_SSE2 1
1420 ; in: r0=pix, r1=stride, r2=stride*3
1421 ; out: [esp+16]=sa8d, [esp+32]=satd, r0+=stride*4
1422 cglobal x264_hadamard_ac_8x8_%1
1428 %define spill0 [rsp+gprsize]
1429 %define spill1 [rsp+gprsize+16]
1430 %define spill2 [rsp+gprsize+32]
1442 HADAMARD4_1D m0, m1, m2, m3
1453 HADAMARD4_1D m4, m5, m6, m7
1456 TRANSPOSE8x8W 0,1,2,3,4,5,6,7,8
1458 TRANSPOSE8x8W 0,1,2,3,4,5,6,7,spill0,spill1
1460 HADAMARD4_1D m0, m1, m2, m3
1461 HADAMARD4_1D m4, m5, m6, m7
1470 pand m1, [mask_ac4 GLOBAL]
1486 mova [rsp+gprsize+32], m1 ; save satd
1493 SBUTTERFLY qdq, 0, 4, 7
1494 SBUTTERFLY qdq, 1, 5, 7
1495 SBUTTERFLY qdq, 2, 6, 7
1496 SUMSUB_BADC m0, m4, m1, m5
1500 pand m0, [mask_ac8 GLOBAL]
1507 SBUTTERFLY qdq, 3, 7, 4
1517 mova [rsp+gprsize+16], m0 ; save sa8d
1518 SAVE_MM_PERMUTATION x264_hadamard_ac_8x8_%1
1521 HADAMARD_AC_WXH_SSE2 16, 16, %1
1522 HADAMARD_AC_WXH_SSE2 8, 16, %1
1523 HADAMARD_AC_WXH_SSE2 16, 8, %1
1524 HADAMARD_AC_WXH_SSE2 8, 8, %1
1525 %endmacro ; HADAMARD_AC_SSE2
1527 ; struct { int satd, int sa8d; } x264_pixel_hadamard_ac_16x16( uint8_t *pix, int stride )
1528 %macro HADAMARD_AC_WXH_SSE2 3
1529 cglobal x264_pixel_hadamard_ac_%1x%2_%3, 2,3
1530 %assign pad 16-gprsize-(stack_offset&15)
1534 call x264_hadamard_ac_8x8_%3
1539 call x264_hadamard_ac_8x8_%3
1544 lea r0, [r0+ysub*4+8]
1546 call x264_hadamard_ac_8x8_%3
1550 call x264_hadamard_ac_8x8_%3
1555 paddusw m0, [rsp+0x30]
1556 paddusw m1, [rsp+0x40]
1559 paddusw m0, [rsp+0x50]
1560 paddusw m1, [rsp+0x60]
1561 paddusw m0, [rsp+0x70]
1562 paddusw m1, [rsp+0x80]
1569 shr edx, 2 - (%1*%2 >> 8)
1575 add rsp, 16+%1*%2/2+pad
1577 %endmacro ; HADAMARD_AC_WXH_SSE2
1582 cextern x264_pixel_sa8d_8x8_internal_mmxext
1583 SA8D_16x16_32 mmxext
1586 %define ABS1 ABS1_MMX
1587 %define ABS2 ABS2_MMX
1590 INTRA_SA8D_SSE2 sse2
1591 INTRA_SATDS_MMX mmxext
1592 HADAMARD_AC_SSE2 sse2
1593 %define ABS1 ABS1_SSSE3
1594 %define ABS2 ABS2_SSSE3
1595 %define ABS_MOV ABS_MOV_SSSE3
1596 SATD_W4 ssse3 ; mmx, but uses pabsw from ssse3.
1599 INTRA_SA8D_SSE2 ssse3
1600 INTRA_SATDS_MMX ssse3
1601 HADAMARD_AC_SSE2 ssse3
1602 SATDS_SSE2 ssse3_phadd
1606 ;=============================================================================
1608 ;=============================================================================
1610 ;-----------------------------------------------------------------------------
1611 ; void x264_pixel_ssim_4x4x2_core_sse2( const uint8_t *pix1, int stride1,
1612 ; const uint8_t *pix2, int stride2, int sums[2][4] )
1613 ;-----------------------------------------------------------------------------
1614 cglobal x264_pixel_ssim_4x4x2_core_sse2, 4,4
1639 movdqa m7, [pw_1 GLOBAL]
1667 ;-----------------------------------------------------------------------------
1668 ; float x264_pixel_ssim_end_sse2( int sum0[5][4], int sum1[5][4], int width )
1669 ;-----------------------------------------------------------------------------
1670 cglobal x264_pixel_ssim_end4_sse2, 3,3
1685 movdqa m5, [ssim_c1 GLOBAL]
1686 movdqa m6, [ssim_c2 GLOBAL]
1687 TRANSPOSE4x4D 0, 1, 2, 3, 4
1689 ; s1=m0, s2=m1, ss=m2, s12=m3
1692 pmaddwd m4, m0 ; s1*s2
1694 pmaddwd m0, m0 ; s1*s1 + s2*s2
1698 psubd m3, m4 ; covar*2
1704 cvtdq2ps m0, m0 ; (float)(s1*s1 + s2*s2 + ssim_c1)
1705 cvtdq2ps m4, m4 ; (float)(s1*s2*2 + ssim_c1)
1706 cvtdq2ps m3, m3 ; (float)(covar*2 + ssim_c2)
1707 cvtdq2ps m2, m2 ; (float)(vars + ssim_c2)
1713 je .skip ; faster only if this is the common case; remove branch if we use ssim on a macroblock level
1716 lea r3, [mask_ff + 16 GLOBAL]
1717 movdqu m1, [r3 + r2*4]
1719 movdqu m1, [mask_ff + r2*4 + 16 GLOBAL]
1735 ;=============================================================================
1736 ; Successive Elimination ADS
1737 ;=============================================================================
1739 %macro ADS_START 1 ; unroll_size
1764 %define ABS1 ABS1_MMX
1766 ;-----------------------------------------------------------------------------
1767 ; int x264_pixel_ads4_mmxext( int enc_dc[4], uint16_t *sums, int delta,
1768 ; uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
1769 ;-----------------------------------------------------------------------------
1770 cglobal x264_pixel_ads4_mmxext, 4,7
1774 pshufw mm6, mm6, 0xAA
1776 pshufw mm4, mm4, 0xAA
1786 movq mm3, [r1+r2+16]
1795 pshufw mm1, [r10+8], 0
1797 pshufw mm1, [ebp+stack_offset+28], 0
1805 cglobal x264_pixel_ads2_mmxext, 4,7
1809 pshufw mm6, mm6, 0xAA
1826 cglobal x264_pixel_ads1_mmxext, 4,7
1848 cglobal x264_pixel_ads4_%1, 4,7
1850 pshuflw xmm7, xmm4, 0
1851 pshuflw xmm6, xmm4, 0xAA
1852 pshufhw xmm5, xmm4, 0
1853 pshufhw xmm4, xmm4, 0xAA
1854 punpcklqdq xmm7, xmm7
1855 punpcklqdq xmm6, xmm6
1856 punpckhqdq xmm5, xmm5
1857 punpckhqdq xmm4, xmm4
1859 pshuflw xmm8, r6m, 0
1860 punpcklqdq xmm8, xmm8
1863 movdqu xmm11, [r1+r2]
1866 movdqu xmm1, [r1+16]
1873 movdqu xmm3, [r1+r2+16]
1892 movdqu xmm1, [r1+16]
1897 movdqu xmm2, [r1+r2]
1898 movdqu xmm3, [r1+r2+16]
1906 movd xmm1, [ebp+stack_offset+28]
1908 pshuflw xmm1, xmm1, 0
1909 punpcklqdq xmm1, xmm1
1917 cglobal x264_pixel_ads2_%1, 4,7
1920 pshuflw xmm7, xmm6, 0
1921 pshuflw xmm6, xmm6, 0xAA
1922 pshuflw xmm5, xmm5, 0
1923 punpcklqdq xmm7, xmm7
1924 punpcklqdq xmm6, xmm6
1925 punpcklqdq xmm5, xmm5
1929 movdqu xmm1, [r1+r2]
1943 cglobal x264_pixel_ads1_%1, 4,7
1946 pshuflw xmm7, xmm7, 0
1947 pshuflw xmm6, xmm6, 0
1948 punpcklqdq xmm7, xmm7
1949 punpcklqdq xmm6, xmm6
1953 movdqu xmm1, [r1+16]
1957 movdqu xmm3, [r3+16]
1972 %define ABS1 ABS1_SSSE3
1975 ; int x264_pixel_ads_mvs( int16_t *mvs, uint8_t *masks, int width )
1978 ; *(uint32_t*)(masks+width) = 0;
1979 ; for( i=0; i<width; i+=8 )
1981 ; uint64_t mask = *(uint64_t*)(masks+i);
1982 ; if( !mask ) continue;
1983 ; for( j=0; j<8; j++ )
1984 ; if( mask & (255<<j*8) )
1989 cglobal x264_pixel_ads_mvs
1997 ; clear last block in case width isn't divisible by 8. (assume divisible by 4, so clearing 4 bytes is enough.)
1998 mov dword [rsp+r5], 0
2011 test edi, 0xff<<(%1*8)
2032 ; no PROLOGUE, inherit from x264_pixel_ads1
2033 mov ebx, [ebp+stack_offset+20] ; mvs
2034 mov edi, [ebp+stack_offset+24] ; width
2035 mov dword [esp+edi], 0
2043 mov ebp, [esp+esi+4]
2044 mov edx, [esp+esi+8]
2051 test %2, 0xff<<(%1*8)