2 # Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved.
4 # Licensed under the OpenSSL license (the "License"). You may not use
5 # this file except in compliance with the License. You can obtain a copy
6 # in the file LICENSE in the source distribution or at
7 # https://www.openssl.org/source/license.html
10 # ====================================================================
11 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12 # project. The module is, however, dual licensed under OpenSSL and
13 # CRYPTOGAMS licenses depending on where you obtain it. For further
14 # details see http://www.openssl.org/~appro/cryptogams/.
15 # ====================================================================
17 # Multi-buffer SHA1 procedure processes n buffers in parallel by
18 # placing buffer data to designated lane of SIMD register. n is
19 # naturally limited to 4 on pre-AVX2 processors and to 8 on
20 # AVX2-capable processors such as Haswell.
22 # this +aesni(i) sha1 aesni-sha1 gain(iv)
23 # -------------------------------------------------------------------
24 # Westmere(ii) 10.7/n +1.28=3.96(n=4) 5.30 6.66 +68%
25 # Atom(ii) 18.1/n +3.93=8.46(n=4) 9.37 12.8 +51%
26 # Sandy Bridge (8.16 +5.15=13.3)/n 4.99 5.98 +80%
27 # Ivy Bridge (8.08 +5.14=13.2)/n 4.60 5.54 +68%
28 # Haswell(iii) (8.96 +5.00=14.0)/n 3.57 4.55 +160%
29 # Skylake (8.70 +5.00=13.7)/n 3.64 4.20 +145%
30 # Bulldozer (9.76 +5.76=15.5)/n 5.95 6.37 +64%
32 # (i) multi-block CBC encrypt with 128-bit key;
33 # (ii) (HASH+AES)/n does not apply to Westmere for n>3 and Atom,
34 # because of lower AES-NI instruction throughput;
35 # (iii) "this" is for n=8, when we gather twice as much data, result
36 # for n=4 is 8.00+4.44=12.4;
37 # (iv) presented improvement coefficients are asymptotic limits and
38 # in real-life application are somewhat lower, e.g. for 2KB
39 # fragments they range from 30% to 100% (on Haswell);
43 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
45 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
47 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
48 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
49 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
50 die "can't locate x86_64-xlate.pl";
54 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
55 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
56 $avx = ($1>=2.19) + ($1>=2.22);
59 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
60 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
61 $avx = ($1>=2.09) + ($1>=2.10);
64 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
65 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
66 $avx = ($1>=10) + ($1>=11);
69 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
70 $avx = ($2>=3.0) + ($2>3.0);
73 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
76 # void sha1_multi_block (
77 # struct { unsigned int A[8];
81 # unsigned int E[8]; } *ctx,
82 # struct { void *ptr; int blocks; } inp[8],
83 # int num); /* 1 or 2 */
85 $ctx="%rdi"; # 1st arg
86 $inp="%rsi"; # 2nd arg
88 @ptr=map("%r$_",(8..11));
91 @V=($A,$B,$C,$D,$E)=map("%xmm$_",(0..4));
92 ($t0,$t1,$t2,$t3,$tx)=map("%xmm$_",(5..9));
93 @Xi=map("%xmm$_",(10..14));
97 # Atom-specific optimization aiming to eliminate pshufb with high
98 # registers [and thus get rid of 48 cycles accumulated penalty]
99 @Xi=map("%xmm$_",(0..4));
100 ($tx,$t0,$t1,$t2,$t3)=map("%xmm$_",(5..9));
101 @V=($A,$B,$C,$D,$E)=map("%xmm$_",(10..14));
109 $off %= 16; $off *= $REG_SZ;
110 $off<256 ? "$off-128(%rax)" : "$off-256-128(%rbx)";
114 my ($i,$a,$b,$c,$d,$e)=@_;
118 # Loads are performed 2+3/4 iterations in advance. 3/4 means that out
119 # of 4 words you would expect to be loaded per given iteration one is
120 # spilled to next iteration. In other words indices in four input
121 # streams are distributed as following:
123 # $i==0: 0,0,0,0,1,1,1,1,2,2,2,
127 # $i==13: 14,15,15,15,
130 # Then at $i==15 Xupdate is applied one iteration in advance...
131 $code.=<<___ if ($i==0);
132 movd (@ptr[0]),@Xi[0]
133 lea `16*4`(@ptr[0]),@ptr[0]
134 movd (@ptr[1]),@Xi[2] # borrow @Xi[2]
135 lea `16*4`(@ptr[1]),@ptr[1]
136 movd (@ptr[2]),@Xi[3] # borrow @Xi[3]
137 lea `16*4`(@ptr[2]),@ptr[2]
138 movd (@ptr[3]),@Xi[4] # borrow @Xi[4]
139 lea `16*4`(@ptr[3]),@ptr[3]
140 punpckldq @Xi[3],@Xi[0]
141 movd `4*$j-16*4`(@ptr[0]),@Xi[1]
142 punpckldq @Xi[4],@Xi[2]
143 movd `4*$j-16*4`(@ptr[1]),$t3
144 punpckldq @Xi[2],@Xi[0]
145 movd `4*$j-16*4`(@ptr[2]),$t2
148 $code.=<<___ if ($i<14); # just load input
149 movd `4*$j-16*4`(@ptr[3]),$t1
152 paddd $K,$e # e+=K_00_19
162 movdqa @Xi[0],`&Xi_off($i)`
163 paddd @Xi[0],$e # e+=X[i]
164 movd `4*$k-16*4`(@ptr[0]),@Xi[2]
166 pxor $t1,$t0 # Ch(b,c,d)
169 por $t3,$t2 # rol(a,5)
170 movd `4*$k-16*4`(@ptr[1]),$t3
172 paddd $t0,$e # e+=Ch(b,c,d)
175 paddd $t2,$e # e+=rol(a,5)
177 movd `4*$k-16*4`(@ptr[2]),$t2
178 por $t1,$b # b=rol(b,30)
180 $code.=<<___ if ($i==14); # just load input
181 movd `4*$j-16*4`(@ptr[3]),$t1
184 paddd $K,$e # e+=K_00_19
189 prefetcht0 63(@ptr[0])
195 movdqa @Xi[0],`&Xi_off($i)`
196 paddd @Xi[0],$e # e+=X[i]
198 pxor $t1,$t0 # Ch(b,c,d)
200 prefetcht0 63(@ptr[1])
202 por $t3,$t2 # rol(a,5)
204 paddd $t0,$e # e+=Ch(b,c,d)
205 prefetcht0 63(@ptr[2])
208 paddd $t2,$e # e+=rol(a,5)
210 prefetcht0 63(@ptr[3])
211 por $t1,$b # b=rol(b,30)
213 $code.=<<___ if ($i>=13 && $i<15);
214 movdqa `&Xi_off($j+2)`,@Xi[3] # preload "X[2]"
216 $code.=<<___ if ($i>=15); # apply Xupdate
217 pxor @Xi[-2],@Xi[1] # "X[13]"
218 movdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
221 pxor `&Xi_off($j+8)`,@Xi[1]
222 paddd $K,$e # e+=K_00_19
234 movdqa @Xi[0],`&Xi_off($i)`
235 paddd @Xi[0],$e # e+=X[i]
237 pxor $t1,$t0 # Ch(b,c,d)
240 por $t3,$t2 # rol(a,5)
242 paddd $t0,$e # e+=Ch(b,c,d)
245 paddd $t2,$e # e+=rol(a,5)
246 por $tx,@Xi[1] # rol \$1,@Xi[1]
247 por $t1,$b # b=rol(b,30)
249 push(@Xi,shift(@Xi));
253 my ($i,$a,$b,$c,$d,$e)=@_;
256 $code.=<<___ if ($i<79);
257 pxor @Xi[-2],@Xi[1] # "X[13]"
258 movdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
262 pxor `&Xi_off($j+8)`,@Xi[1]
263 paddd $K,$e # e+=K_20_39
269 $code.=<<___ if ($i<72);
270 movdqa @Xi[0],`&Xi_off($i)`
272 $code.=<<___ if ($i<79);
273 paddd @Xi[0],$e # e+=X[i]
276 pxor $c,$t0 # Parity(b,c,d)
281 por $t3,$t2 # rol(a,5)
283 paddd $t0,$e # e+=Parity(b,c,d)
287 paddd $t2,$e # e+=rol(a,5)
288 por $tx,@Xi[1] # rol(@Xi[1],1)
289 por $t1,$b # b=rol(b,30)
291 $code.=<<___ if ($i==79);
293 paddd $K,$e # e+=K_20_39
299 paddd @Xi[0],$e # e+=X[i]
302 pxor $c,$t0 # Parity(b,c,d)
305 por $t3,$t2 # rol(a,5)
306 paddd $t0,$e # e+=Parity(b,c,d)
309 paddd $t2,$e # e+=rol(a,5)
310 por $t1,$b # b=rol(b,30)
312 push(@Xi,shift(@Xi));
316 my ($i,$a,$b,$c,$d,$e)=@_;
320 pxor @Xi[-2],@Xi[1] # "X[13]"
321 movdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
325 pxor `&Xi_off($j+8)`,@Xi[1]
327 paddd $K,$e # e+=K_40_59
338 movdqa @Xi[0],`&Xi_off($i)`
339 paddd @Xi[0],$e # e+=X[i]
340 por $t3,$t2 # rol(a,5)
347 paddd $t0,$e # e+=Maj(b,d,c)
350 paddd $t2,$e # e+=rol(a,5)
351 por $tx,@Xi[1] # rol(@X[1],1)
352 por $t1,$b # b=rol(b,30)
354 push(@Xi,shift(@Xi));
360 .extern OPENSSL_ia32cap_P
362 .globl sha1_multi_block
363 .type sha1_multi_block,\@function,3
367 mov OPENSSL_ia32cap_P+4(%rip),%rcx
368 bt \$61,%rcx # check SHA bit
371 $code.=<<___ if ($avx);
377 .cfi_def_cfa_register %rax
383 $code.=<<___ if ($win64);
386 movaps %xmm7,0x10(%rsp)
387 movaps %xmm8,0x20(%rsp)
388 movaps %xmm9,0x30(%rsp)
389 movaps %xmm10,-0x78(%rax)
390 movaps %xmm11,-0x68(%rax)
391 movaps %xmm12,-0x58(%rax)
392 movaps %xmm13,-0x48(%rax)
393 movaps %xmm14,-0x38(%rax)
394 movaps %xmm15,-0x28(%rax)
397 sub \$`$REG_SZ*18`,%rsp
399 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
400 .cfi_cfa_expression %rsp+`$REG_SZ*17`,deref,+8
402 lea K_XX_XX(%rip),$Tbl
403 lea `$REG_SZ*16`(%rsp),%rbx
406 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
409 for($i=0;$i<4;$i++) {
411 mov `16*$i+0`($inp),@ptr[$i] # input pointer
412 mov `16*$i+8`($inp),%ecx # number of blocks
414 cmovg %ecx,$num # find maximum
416 mov %ecx,`4*$i`(%rbx) # initialize counters
417 cmovle $Tbl,@ptr[$i] # cancel input
424 movdqu 0x00($ctx),$A # load context
430 movdqa 0x60($Tbl),$tx # pbswap_mask
431 movdqa -0x20($Tbl),$K # K_00_19
437 for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
438 $code.=" movdqa 0x00($Tbl),$K\n"; # K_20_39
439 for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
440 $code.=" movdqa 0x20($Tbl),$K\n"; # K_40_59
441 for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
442 $code.=" movdqa 0x40($Tbl),$K\n"; # K_60_79
443 for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
445 movdqa (%rbx),@Xi[0] # pull counters
447 cmp 4*0(%rbx),%ecx # examinte counters
449 cmovge $Tbl,@ptr[0] # cancel input
454 pcmpgtd $t2,@Xi[1] # mask value
457 paddd @Xi[1],@Xi[0] # counters--
460 movdqu 0x00($ctx),$t0
462 movdqu 0x20($ctx),$t1
465 movdqu 0x40($ctx),$t2
468 movdqu 0x60($ctx),$t3
471 movdqu 0x80($ctx),$tx
481 movdqa @Xi[0],(%rbx) # save counters
482 movdqa 0x60($Tbl),$tx # pbswap_mask
483 movdqa -0x20($Tbl),$K # K_00_19
487 mov `$REG_SZ*17+8`(%rsp),$num
488 lea $REG_SZ($ctx),$ctx
489 lea `16*$REG_SZ/4`($inp),$inp
494 mov `$REG_SZ*17`(%rsp),%rax # original %rsp
497 $code.=<<___ if ($win64);
498 movaps -0xb8(%rax),%xmm6
499 movaps -0xa8(%rax),%xmm7
500 movaps -0x98(%rax),%xmm8
501 movaps -0x88(%rax),%xmm9
502 movaps -0x78(%rax),%xmm10
503 movaps -0x68(%rax),%xmm11
504 movaps -0x58(%rax),%xmm12
505 movaps -0x48(%rax),%xmm13
506 movaps -0x38(%rax),%xmm14
507 movaps -0x28(%rax),%xmm15
515 .cfi_def_cfa_register %rsp
519 .size sha1_multi_block,.-sha1_multi_block
522 my ($ABCD0,$E0,$E0_,$BSWAP,$ABCD1,$E1,$E1_)=map("%xmm$_",(0..3,8..10));
523 my @MSG0=map("%xmm$_",(4..7));
524 my @MSG1=map("%xmm$_",(11..14));
527 .type sha1_multi_block_shaext,\@function,3
529 sha1_multi_block_shaext:
533 .cfi_def_cfa_register %rax
539 $code.=<<___ if ($win64);
542 movaps %xmm7,0x10(%rsp)
543 movaps %xmm8,0x20(%rsp)
544 movaps %xmm9,0x30(%rsp)
545 movaps %xmm10,-0x78(%rax)
546 movaps %xmm11,-0x68(%rax)
547 movaps %xmm12,-0x58(%rax)
548 movaps %xmm13,-0x48(%rax)
549 movaps %xmm14,-0x38(%rax)
550 movaps %xmm15,-0x28(%rax)
553 sub \$`$REG_SZ*18`,%rsp
554 shl \$1,$num # we process pair at a time
556 lea 0x40($ctx),$ctx # size optimization
557 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
559 lea `$REG_SZ*16`(%rsp),%rbx
560 movdqa K_XX_XX+0x80(%rip),$BSWAP # byte-n-word swap
563 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
566 for($i=0;$i<2;$i++) {
568 mov `16*$i+0`($inp),@ptr[$i] # input pointer
569 mov `16*$i+8`($inp),%ecx # number of blocks
571 cmovg %ecx,$num # find maximum
573 mov %ecx,`4*$i`(%rbx) # initialize counters
574 cmovle %rsp,@ptr[$i] # cancel input
581 movq 0x00-0x40($ctx),$ABCD0 # a1.a0
582 movq 0x20-0x40($ctx),@MSG0[0]# b1.b0
583 movq 0x40-0x40($ctx),@MSG0[1]# c1.c0
584 movq 0x60-0x40($ctx),@MSG0[2]# d1.d0
585 movq 0x80-0x40($ctx),@MSG0[3]# e1.e0
587 punpckldq @MSG0[0],$ABCD0 # b1.a1.b0.a0
588 punpckldq @MSG0[2],@MSG0[1] # d1.c1.d0.c0
591 punpcklqdq @MSG0[1],$ABCD0 # d0.c0.b0.a0
592 punpckhqdq @MSG0[1],$ABCD1 # d1.c1.b1.a1
594 pshufd \$0b00111111,@MSG0[3],$E0
595 pshufd \$0b01111111,@MSG0[3],$E1
596 pshufd \$0b00011011,$ABCD0,$ABCD0
597 pshufd \$0b00011011,$ABCD1,$ABCD1
602 movdqu 0x00(@ptr[0]),@MSG0[0]
603 movdqu 0x00(@ptr[1]),@MSG1[0]
604 movdqu 0x10(@ptr[0]),@MSG0[1]
605 movdqu 0x10(@ptr[1]),@MSG1[1]
606 movdqu 0x20(@ptr[0]),@MSG0[2]
607 pshufb $BSWAP,@MSG0[0]
608 movdqu 0x20(@ptr[1]),@MSG1[2]
609 pshufb $BSWAP,@MSG1[0]
610 movdqu 0x30(@ptr[0]),@MSG0[3]
611 lea 0x40(@ptr[0]),@ptr[0]
612 pshufb $BSWAP,@MSG0[1]
613 movdqu 0x30(@ptr[1]),@MSG1[3]
614 lea 0x40(@ptr[1]),@ptr[1]
615 pshufb $BSWAP,@MSG1[1]
617 movdqa $E0,0x50(%rsp) # offload
619 movdqa $E1,0x70(%rsp)
621 movdqa $ABCD0,0x40(%rsp) # offload
623 movdqa $ABCD1,0x60(%rsp)
625 sha1rnds4 \$0,$E0,$ABCD0 # 0-3
626 sha1nexte @MSG0[1],$E0_
627 sha1rnds4 \$0,$E1,$ABCD1 # 0-3
628 sha1nexte @MSG1[1],$E1_
629 pshufb $BSWAP,@MSG0[2]
630 prefetcht0 127(@ptr[0])
631 sha1msg1 @MSG0[1],@MSG0[0]
632 pshufb $BSWAP,@MSG1[2]
633 prefetcht0 127(@ptr[1])
634 sha1msg1 @MSG1[1],@MSG1[0]
636 pshufb $BSWAP,@MSG0[3]
638 pshufb $BSWAP,@MSG1[3]
640 sha1rnds4 \$0,$E0_,$ABCD0 # 4-7
641 sha1nexte @MSG0[2],$E0
642 sha1rnds4 \$0,$E1_,$ABCD1 # 4-7
643 sha1nexte @MSG1[2],$E1
644 pxor @MSG0[2],@MSG0[0]
645 sha1msg1 @MSG0[2],@MSG0[1]
646 pxor @MSG1[2],@MSG1[0]
647 sha1msg1 @MSG1[2],@MSG1[1]
649 for($i=2;$i<20-4;$i++) {
653 sha1rnds4 \$`int($i/5)`,$E0,$ABCD0 # 8-11
654 sha1nexte @MSG0[3],$E0_
655 sha1rnds4 \$`int($i/5)`,$E1,$ABCD1 # 8-11
656 sha1nexte @MSG1[3],$E1_
657 sha1msg2 @MSG0[3],@MSG0[0]
658 sha1msg2 @MSG1[3],@MSG1[0]
659 pxor @MSG0[3],@MSG0[1]
660 sha1msg1 @MSG0[3],@MSG0[2]
661 pxor @MSG1[3],@MSG1[1]
662 sha1msg1 @MSG1[3],@MSG1[2]
664 ($E0,$E0_)=($E0_,$E0); ($E1,$E1_)=($E1_,$E1);
665 push(@MSG0,shift(@MSG0)); push(@MSG1,shift(@MSG1));
670 sha1rnds4 \$3,$E0,$ABCD0 # 64-67
671 sha1nexte @MSG0[3],$E0_
672 sha1rnds4 \$3,$E1,$ABCD1 # 64-67
673 sha1nexte @MSG1[3],$E1_
674 sha1msg2 @MSG0[3],@MSG0[0]
675 sha1msg2 @MSG1[3],@MSG1[0]
676 pxor @MSG0[3],@MSG0[1]
677 pxor @MSG1[3],@MSG1[1]
680 pxor @MSG0[2],@MSG0[2] # zero
681 cmp 4*0(%rbx),%ecx # examine counters
682 cmovge %rsp,@ptr[0] # cancel input
686 sha1rnds4 \$3,$E0_,$ABCD0 # 68-71
687 sha1nexte @MSG0[0],$E0
688 sha1rnds4 \$3,$E1_,$ABCD1 # 68-71
689 sha1nexte @MSG1[0],$E1
690 sha1msg2 @MSG0[0],@MSG0[1]
691 sha1msg2 @MSG1[0],@MSG1[1]
695 movq (%rbx),@MSG0[0] # pull counters
699 sha1rnds4 \$3,$E0,$ABCD0 # 72-75
700 sha1nexte @MSG0[1],$E0_
701 sha1rnds4 \$3,$E1,$ABCD1 # 72-75
702 sha1nexte @MSG1[1],$E1_
704 pshufd \$0x00,@MSG0[0],@MSG1[2]
705 pshufd \$0x55,@MSG0[0],@MSG1[3]
706 movdqa @MSG0[0],@MSG0[1]
707 pcmpgtd @MSG0[2],@MSG1[2]
708 pcmpgtd @MSG0[2],@MSG1[3]
712 sha1rnds4 \$3,$E0_,$ABCD0 # 76-79
713 sha1nexte $MSG0[2],$E0
714 sha1rnds4 \$3,$E1_,$ABCD1 # 76-79
715 sha1nexte $MSG0[2],$E1
717 pcmpgtd @MSG0[2],@MSG0[1] # counter mask
722 paddd @MSG0[1],@MSG0[0] # counters--
724 paddd 0x40(%rsp),$ABCD0
726 paddd 0x60(%rsp),$ABCD1
729 movq @MSG0[0],(%rbx) # save counters
733 mov `$REG_SZ*17+8`(%rsp),$num
735 pshufd \$0b00011011,$ABCD0,$ABCD0
736 pshufd \$0b00011011,$ABCD1,$ABCD1
738 movdqa $ABCD0,@MSG0[0]
739 punpckldq $ABCD1,$ABCD0 # b1.b0.a1.a0
740 punpckhdq $ABCD1,@MSG0[0] # d1.d0.c1.c0
741 punpckhdq $E1,$E0 # e1.e0.xx.xx
742 movq $ABCD0,0x00-0x40($ctx) # a1.a0
744 movq @MSG0[0],0x40-0x40($ctx)# c1.c0
746 movq $ABCD0,0x20-0x40($ctx) # b1.b0
748 movq @MSG0[0],0x60-0x40($ctx)# d1.d0
749 movq $E0,0x80-0x40($ctx) # e1.e0
751 lea `$REG_SZ/2`($ctx),$ctx
752 lea `16*2`($inp),$inp
754 jnz .Loop_grande_shaext
757 #mov `$REG_SZ*17`(%rsp),%rax # original %rsp
759 $code.=<<___ if ($win64);
760 movaps -0xb8(%rax),%xmm6
761 movaps -0xa8(%rax),%xmm7
762 movaps -0x98(%rax),%xmm8
763 movaps -0x88(%rax),%xmm9
764 movaps -0x78(%rax),%xmm10
765 movaps -0x68(%rax),%xmm11
766 movaps -0x58(%rax),%xmm12
767 movaps -0x48(%rax),%xmm13
768 movaps -0x38(%rax),%xmm14
769 movaps -0x28(%rax),%xmm15
777 .cfi_def_cfa_register %rsp
781 .size sha1_multi_block_shaext,.-sha1_multi_block_shaext
787 my ($i,$a,$b,$c,$d,$e)=@_;
790 my $vpack = $REG_SZ==16 ? "vpunpckldq" : "vinserti128";
791 my $ptr_n = $REG_SZ==16 ? @ptr[1] : @ptr[4];
793 $code.=<<___ if ($i==0 && $REG_SZ==16);
794 vmovd (@ptr[0]),@Xi[0]
795 lea `16*4`(@ptr[0]),@ptr[0]
796 vmovd (@ptr[1]),@Xi[2] # borrow Xi[2]
797 lea `16*4`(@ptr[1]),@ptr[1]
798 vpinsrd \$1,(@ptr[2]),@Xi[0],@Xi[0]
799 lea `16*4`(@ptr[2]),@ptr[2]
800 vpinsrd \$1,(@ptr[3]),@Xi[2],@Xi[2]
801 lea `16*4`(@ptr[3]),@ptr[3]
802 vmovd `4*$j-16*4`(@ptr[0]),@Xi[1]
803 vpunpckldq @Xi[2],@Xi[0],@Xi[0]
804 vmovd `4*$j-16*4`($ptr_n),$t3
805 vpshufb $tx,@Xi[0],@Xi[0]
807 $code.=<<___ if ($i<15 && $REG_SZ==16); # just load input
808 vpinsrd \$1,`4*$j-16*4`(@ptr[2]),@Xi[1],@Xi[1]
809 vpinsrd \$1,`4*$j-16*4`(@ptr[3]),$t3,$t3
811 $code.=<<___ if ($i==0 && $REG_SZ==32);
812 vmovd (@ptr[0]),@Xi[0]
813 lea `16*4`(@ptr[0]),@ptr[0]
814 vmovd (@ptr[4]),@Xi[2] # borrow Xi[2]
815 lea `16*4`(@ptr[4]),@ptr[4]
817 lea `16*4`(@ptr[1]),@ptr[1]
819 lea `16*4`(@ptr[5]),@ptr[5]
820 vpinsrd \$1,(@ptr[2]),@Xi[0],@Xi[0]
821 lea `16*4`(@ptr[2]),@ptr[2]
822 vpinsrd \$1,(@ptr[6]),@Xi[2],@Xi[2]
823 lea `16*4`(@ptr[6]),@ptr[6]
824 vpinsrd \$1,(@ptr[3]),$t2,$t2
825 lea `16*4`(@ptr[3]),@ptr[3]
826 vpunpckldq $t2,@Xi[0],@Xi[0]
827 vpinsrd \$1,(@ptr[7]),$t1,$t1
828 lea `16*4`(@ptr[7]),@ptr[7]
829 vpunpckldq $t1,@Xi[2],@Xi[2]
830 vmovd `4*$j-16*4`(@ptr[0]),@Xi[1]
831 vinserti128 @Xi[2],@Xi[0],@Xi[0]
832 vmovd `4*$j-16*4`($ptr_n),$t3
833 vpshufb $tx,@Xi[0],@Xi[0]
835 $code.=<<___ if ($i<15 && $REG_SZ==32); # just load input
836 vmovd `4*$j-16*4`(@ptr[1]),$t2
837 vmovd `4*$j-16*4`(@ptr[5]),$t1
838 vpinsrd \$1,`4*$j-16*4`(@ptr[2]),@Xi[1],@Xi[1]
839 vpinsrd \$1,`4*$j-16*4`(@ptr[6]),$t3,$t3
840 vpinsrd \$1,`4*$j-16*4`(@ptr[3]),$t2,$t2
841 vpunpckldq $t2,@Xi[1],@Xi[1]
842 vpinsrd \$1,`4*$j-16*4`(@ptr[7]),$t1,$t1
843 vpunpckldq $t1,$t3,$t3
845 $code.=<<___ if ($i<14);
846 vpaddd $K,$e,$e # e+=K_00_19
851 vmovdqa @Xi[0],`&Xi_off($i)`
852 vpaddd @Xi[0],$e,$e # e+=X[i]
853 $vpack $t3,@Xi[1],@Xi[1]
855 vpxor $t1,$t0,$t0 # Ch(b,c,d)
856 vmovd `4*$k-16*4`(@ptr[0]),@Xi[2]
859 vpor $t3,$t2,$t2 # rol(a,5)
860 vmovd `4*$k-16*4`($ptr_n),$t3
861 vpaddd $t0,$e,$e # e+=Ch(b,c,d)
864 vpaddd $t2,$e,$e # e+=rol(a,5)
865 vpshufb $tx,@Xi[1],@Xi[1]
866 vpor $t1,$b,$b # b=rol(b,30)
868 $code.=<<___ if ($i==14);
869 vpaddd $K,$e,$e # e+=K_00_19
870 prefetcht0 63(@ptr[0])
875 vmovdqa @Xi[0],`&Xi_off($i)`
876 vpaddd @Xi[0],$e,$e # e+=X[i]
877 $vpack $t3,@Xi[1],@Xi[1]
879 prefetcht0 63(@ptr[1])
880 vpxor $t1,$t0,$t0 # Ch(b,c,d)
883 vpor $t3,$t2,$t2 # rol(a,5)
884 prefetcht0 63(@ptr[2])
885 vpaddd $t0,$e,$e # e+=Ch(b,c,d)
888 vpaddd $t2,$e,$e # e+=rol(a,5)
889 prefetcht0 63(@ptr[3])
890 vpshufb $tx,@Xi[1],@Xi[1]
891 vpor $t1,$b,$b # b=rol(b,30)
893 $code.=<<___ if ($i>=13 && $i<15);
894 vmovdqa `&Xi_off($j+2)`,@Xi[3] # preload "X[2]"
896 $code.=<<___ if ($i>=15); # apply Xupdate
897 vpxor @Xi[-2],@Xi[1],@Xi[1] # "X[13]"
898 vmovdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
900 vpaddd $K,$e,$e # e+=K_00_19
903 `"prefetcht0 63(@ptr[4])" if ($i==15 && $REG_SZ==32)`
906 vmovdqa @Xi[0],`&Xi_off($i)`
907 vpaddd @Xi[0],$e,$e # e+=X[i]
908 vpxor `&Xi_off($j+8)`,@Xi[1],@Xi[1]
910 vpxor $t1,$t0,$t0 # Ch(b,c,d)
911 vpxor @Xi[3],@Xi[1],@Xi[1]
912 `"prefetcht0 63(@ptr[5])" if ($i==15 && $REG_SZ==32)`
915 vpor $t3,$t2,$t2 # rol(a,5)
916 vpaddd $t0,$e,$e # e+=Ch(b,c,d)
917 `"prefetcht0 63(@ptr[6])" if ($i==15 && $REG_SZ==32)`
918 vpsrld \$31,@Xi[1],$tx
919 vpaddd @Xi[1],@Xi[1],@Xi[1]
922 `"prefetcht0 63(@ptr[7])" if ($i==15 && $REG_SZ==32)`
923 vpaddd $t2,$e,$e # e+=rol(a,5)
924 vpor $tx,@Xi[1],@Xi[1] # rol \$1,@Xi[1]
925 vpor $t1,$b,$b # b=rol(b,30)
927 push(@Xi,shift(@Xi));
931 my ($i,$a,$b,$c,$d,$e)=@_;
934 $code.=<<___ if ($i<79);
935 vpxor @Xi[-2],@Xi[1],@Xi[1] # "X[13]"
936 vmovdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
939 vpaddd $K,$e,$e # e+=K_20_39
942 $code.=<<___ if ($i<72);
943 vmovdqa @Xi[0],`&Xi_off($i)`
945 $code.=<<___ if ($i<79);
946 vpaddd @Xi[0],$e,$e # e+=X[i]
947 vpxor `&Xi_off($j+8)`,@Xi[1],@Xi[1]
949 vpxor $c,$t0,$t0 # Parity(b,c,d)
950 vpxor @Xi[3],@Xi[1],@Xi[1]
953 vpor $t3,$t2,$t2 # rol(a,5)
954 vpaddd $t0,$e,$e # e+=Parity(b,c,d)
955 vpsrld \$31,@Xi[1],$tx
956 vpaddd @Xi[1],@Xi[1],@Xi[1]
959 vpaddd $t2,$e,$e # e+=rol(a,5)
960 vpor $tx,@Xi[1],@Xi[1] # rol(@Xi[1],1)
961 vpor $t1,$b,$b # b=rol(b,30)
963 $code.=<<___ if ($i==79);
965 vpaddd $K,$e,$e # e+=K_20_39
969 vpaddd @Xi[0],$e,$e # e+=X[i]
970 vpxor $c,$t0,$t0 # Parity(b,c,d)
973 vpor $t3,$t2,$t2 # rol(a,5)
974 vpaddd $t0,$e,$e # e+=Parity(b,c,d)
977 vpaddd $t2,$e,$e # e+=rol(a,5)
978 vpor $t1,$b,$b # b=rol(b,30)
980 push(@Xi,shift(@Xi));
984 my ($i,$a,$b,$c,$d,$e)=@_;
988 vpxor @Xi[-2],@Xi[1],@Xi[1] # "X[13]"
989 vmovdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
991 vpaddd $K,$e,$e # e+=K_40_59
994 vpxor `&Xi_off($j+8)`,@Xi[1],@Xi[1]
999 vpxor @Xi[3],@Xi[1],@Xi[1]
1001 vmovdqu @Xi[0],`&Xi_off($i)`
1002 vpaddd @Xi[0],$e,$e # e+=X[i]
1003 vpor $t3,$t2,$t2 # rol(a,5)
1004 vpsrld \$31,@Xi[1],$tx
1006 vpaddd @Xi[1],@Xi[1],@Xi[1]
1009 vpaddd $t0,$e,$e # e+=Maj(b,d,c)
1012 vpaddd $t2,$e,$e # e+=rol(a,5)
1013 vpor $tx,@Xi[1],@Xi[1] # rol(@X[1],1)
1014 vpor $t1,$b,$b # b=rol(b,30)
1016 push(@Xi,shift(@Xi));
1020 .type sha1_multi_block_avx,\@function,3
1022 sha1_multi_block_avx:
1026 $code.=<<___ if ($avx>1);
1038 .cfi_def_cfa_register %rax
1044 $code.=<<___ if ($win64);
1045 lea -0xa8(%rsp),%rsp
1047 movaps %xmm7,0x10(%rsp)
1048 movaps %xmm8,0x20(%rsp)
1049 movaps %xmm9,0x30(%rsp)
1050 movaps %xmm10,-0x78(%rax)
1051 movaps %xmm11,-0x68(%rax)
1052 movaps %xmm12,-0x58(%rax)
1053 movaps %xmm13,-0x48(%rax)
1054 movaps %xmm14,-0x38(%rax)
1055 movaps %xmm15,-0x28(%rax)
1058 sub \$`$REG_SZ*18`, %rsp
1060 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
1061 .cfi_cfa_expression %rsp+`$REG_SZ*17`,deref,+8
1063 lea K_XX_XX(%rip),$Tbl
1064 lea `$REG_SZ*16`(%rsp),%rbx
1068 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
1071 for($i=0;$i<4;$i++) {
1073 mov `16*$i+0`($inp),@ptr[$i] # input pointer
1074 mov `16*$i+8`($inp),%ecx # number of blocks
1076 cmovg %ecx,$num # find maximum
1078 mov %ecx,`4*$i`(%rbx) # initialize counters
1079 cmovle $Tbl,@ptr[$i] # cancel input
1086 vmovdqu 0x00($ctx),$A # load context
1088 vmovdqu 0x20($ctx),$B
1089 vmovdqu 0x40($ctx),$C
1090 vmovdqu 0x60($ctx),$D
1091 vmovdqu 0x80($ctx),$E
1092 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1098 $code.=" vmovdqa -0x20($Tbl),$K\n"; # K_00_19
1099 for($i=0;$i<20;$i++) { &BODY_00_19_avx($i,@V); unshift(@V,pop(@V)); }
1100 $code.=" vmovdqa 0x00($Tbl),$K\n"; # K_20_39
1101 for(;$i<40;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1102 $code.=" vmovdqa 0x20($Tbl),$K\n"; # K_40_59
1103 for(;$i<60;$i++) { &BODY_40_59_avx($i,@V); unshift(@V,pop(@V)); }
1104 $code.=" vmovdqa 0x40($Tbl),$K\n"; # K_60_79
1105 for(;$i<80;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1109 for($i=0;$i<4;$i++) {
1111 cmp `4*$i`(%rbx),%ecx # examine counters
1112 cmovge $Tbl,@ptr[$i] # cancel input
1116 vmovdqu (%rbx),$t0 # pull counters
1119 vpcmpgtd $t2,$t1,$t1 # mask value
1120 vpaddd $t1,$t0,$t0 # counters--
1124 vpaddd 0x00($ctx),$A,$A
1126 vpaddd 0x20($ctx),$B,$B
1128 vpaddd 0x40($ctx),$C,$C
1130 vpaddd 0x60($ctx),$D,$D
1131 vpaddd 0x80($ctx),$E,$E
1132 vmovdqu $A,0x00($ctx)
1133 vmovdqu $B,0x20($ctx)
1134 vmovdqu $C,0x40($ctx)
1135 vmovdqu $D,0x60($ctx)
1136 vmovdqu $E,0x80($ctx)
1138 vmovdqu $t0,(%rbx) # save counters
1139 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1143 mov `$REG_SZ*17+8`(%rsp),$num
1144 lea $REG_SZ($ctx),$ctx
1145 lea `16*$REG_SZ/4`($inp),$inp
1147 jnz .Loop_grande_avx
1150 mov `$REG_SZ*17`(%rsp),%rax # original %rsp
1154 $code.=<<___ if ($win64);
1155 movaps -0xb8(%rax),%xmm6
1156 movaps -0xa8(%rax),%xmm7
1157 movaps -0x98(%rax),%xmm8
1158 movaps -0x88(%rax),%xmm9
1159 movaps -0x78(%rax),%xmm10
1160 movaps -0x68(%rax),%xmm11
1161 movaps -0x58(%rax),%xmm12
1162 movaps -0x48(%rax),%xmm13
1163 movaps -0x38(%rax),%xmm14
1164 movaps -0x28(%rax),%xmm15
1172 .cfi_def_cfa_register %rsp
1176 .size sha1_multi_block_avx,.-sha1_multi_block_avx
1180 $code =~ s/\`([^\`]*)\`/eval $1/gem;
1184 @ptr=map("%r$_",(12..15,8..11));
1186 @V=($A,$B,$C,$D,$E)=map("%ymm$_",(0..4));
1187 ($t0,$t1,$t2,$t3,$tx)=map("%ymm$_",(5..9));
1188 @Xi=map("%ymm$_",(10..14));
1192 .type sha1_multi_block_avx2,\@function,3
1194 sha1_multi_block_avx2:
1198 .cfi_def_cfa_register %rax
1212 $code.=<<___ if ($win64);
1213 lea -0xa8(%rsp),%rsp
1215 movaps %xmm7,0x10(%rsp)
1216 movaps %xmm8,0x20(%rsp)
1217 movaps %xmm9,0x30(%rsp)
1218 movaps %xmm10,0x40(%rsp)
1219 movaps %xmm11,0x50(%rsp)
1220 movaps %xmm12,-0x78(%rax)
1221 movaps %xmm13,-0x68(%rax)
1222 movaps %xmm14,-0x58(%rax)
1223 movaps %xmm15,-0x48(%rax)
1226 sub \$`$REG_SZ*18`, %rsp
1228 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
1229 .cfi_cfa_expression %rsp+`$REG_SZ*17`,deref,+8
1231 lea K_XX_XX(%rip),$Tbl
1236 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
1238 lea `$REG_SZ*16`(%rsp),%rbx
1240 for($i=0;$i<8;$i++) {
1242 mov `16*$i+0`($inp),@ptr[$i] # input pointer
1243 mov `16*$i+8`($inp),%ecx # number of blocks
1245 cmovg %ecx,$num # find maximum
1247 mov %ecx,`4*$i`(%rbx) # initialize counters
1248 cmovle $Tbl,@ptr[$i] # cancel input
1252 vmovdqu 0x00($ctx),$A # load context
1254 vmovdqu 0x20($ctx),$B
1255 lea 256+128(%rsp),%rbx
1256 vmovdqu 0x40($ctx),$C
1257 vmovdqu 0x60($ctx),$D
1258 vmovdqu 0x80($ctx),$E
1259 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1265 $code.=" vmovdqa -0x20($Tbl),$K\n"; # K_00_19
1266 for($i=0;$i<20;$i++) { &BODY_00_19_avx($i,@V); unshift(@V,pop(@V)); }
1267 $code.=" vmovdqa 0x00($Tbl),$K\n"; # K_20_39
1268 for(;$i<40;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1269 $code.=" vmovdqa 0x20($Tbl),$K\n"; # K_40_59
1270 for(;$i<60;$i++) { &BODY_40_59_avx($i,@V); unshift(@V,pop(@V)); }
1271 $code.=" vmovdqa 0x40($Tbl),$K\n"; # K_60_79
1272 for(;$i<80;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1275 lea `$REG_SZ*16`(%rsp),%rbx
1277 for($i=0;$i<8;$i++) {
1279 cmp `4*$i`(%rbx),%ecx # examine counters
1280 cmovge $Tbl,@ptr[$i] # cancel input
1284 vmovdqu (%rbx),$t0 # pull counters
1287 vpcmpgtd $t2,$t1,$t1 # mask value
1288 vpaddd $t1,$t0,$t0 # counters--
1292 vpaddd 0x00($ctx),$A,$A
1294 vpaddd 0x20($ctx),$B,$B
1296 vpaddd 0x40($ctx),$C,$C
1298 vpaddd 0x60($ctx),$D,$D
1299 vpaddd 0x80($ctx),$E,$E
1300 vmovdqu $A,0x00($ctx)
1301 vmovdqu $B,0x20($ctx)
1302 vmovdqu $C,0x40($ctx)
1303 vmovdqu $D,0x60($ctx)
1304 vmovdqu $E,0x80($ctx)
1306 vmovdqu $t0,(%rbx) # save counters
1307 lea 256+128(%rsp),%rbx
1308 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1312 #mov `$REG_SZ*17+8`(%rsp),$num
1313 #lea $REG_SZ($ctx),$ctx
1314 #lea `16*$REG_SZ/4`($inp),$inp
1316 #jnz .Loop_grande_avx2
1319 mov `$REG_SZ*17`(%rsp),%rax # original %rsp
1323 $code.=<<___ if ($win64);
1324 movaps -0xd8(%rax),%xmm6
1325 movaps -0xc8(%rax),%xmm7
1326 movaps -0xb8(%rax),%xmm8
1327 movaps -0xa8(%rax),%xmm9
1328 movaps -0x98(%rax),%xmm10
1329 movaps -0x88(%rax),%xmm11
1330 movaps -0x78(%rax),%xmm12
1331 movaps -0x68(%rax),%xmm13
1332 movaps -0x58(%rax),%xmm14
1333 movaps -0x48(%rax),%xmm15
1349 .cfi_def_cfa_register %rsp
1353 .size sha1_multi_block_avx2,.-sha1_multi_block_avx2
1359 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1360 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1362 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1363 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1364 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1365 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1366 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1367 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1368 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap
1369 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap
1370 .byte 0xf,0xe,0xd,0xc,0xb,0xa,0x9,0x8,0x7,0x6,0x5,0x4,0x3,0x2,0x1,0x0
1371 .asciz "SHA1 multi-block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1375 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1376 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1383 .extern __imp_RtlVirtualUnwind
1384 .type se_handler,\@abi-omnipotent
1398 mov 120($context),%rax # pull context->Rax
1399 mov 248($context),%rbx # pull context->Rip
1401 mov 8($disp),%rsi # disp->ImageBase
1402 mov 56($disp),%r11 # disp->HandlerData
1404 mov 0(%r11),%r10d # HandlerData[0]
1405 lea (%rsi,%r10),%r10 # end of prologue label
1406 cmp %r10,%rbx # context->Rip<.Lbody
1409 mov 152($context),%rax # pull context->Rsp
1411 mov 4(%r11),%r10d # HandlerData[1]
1412 lea (%rsi,%r10),%r10 # epilogue label
1413 cmp %r10,%rbx # context->Rip>=.Lepilogue
1416 mov `16*17`(%rax),%rax # pull saved stack pointer
1420 mov %rbx,144($context) # restore context->Rbx
1421 mov %rbp,160($context) # restore context->Rbp
1423 lea -24-10*16(%rax),%rsi
1424 lea 512($context),%rdi # &context.Xmm6
1426 .long 0xa548f3fc # cld; rep movsq
1431 mov %rax,152($context) # restore context->Rsp
1432 mov %rsi,168($context) # restore context->Rsi
1433 mov %rdi,176($context) # restore context->Rdi
1435 mov 40($disp),%rdi # disp->ContextRecord
1436 mov $context,%rsi # context
1437 mov \$154,%ecx # sizeof(CONTEXT)
1438 .long 0xa548f3fc # cld; rep movsq
1441 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1442 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1443 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1444 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1445 mov 40(%rsi),%r10 # disp->ContextRecord
1446 lea 56(%rsi),%r11 # &disp->HandlerData
1447 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1448 mov %r10,32(%rsp) # arg5
1449 mov %r11,40(%rsp) # arg6
1450 mov %r12,48(%rsp) # arg7
1451 mov %rcx,56(%rsp) # arg8, (NULL)
1452 call *__imp_RtlVirtualUnwind(%rip)
1454 mov \$1,%eax # ExceptionContinueSearch
1466 .size se_handler,.-se_handler
1468 $code.=<<___ if ($avx>1);
1469 .type avx2_handler,\@abi-omnipotent
1483 mov 120($context),%rax # pull context->Rax
1484 mov 248($context),%rbx # pull context->Rip
1486 mov 8($disp),%rsi # disp->ImageBase
1487 mov 56($disp),%r11 # disp->HandlerData
1489 mov 0(%r11),%r10d # HandlerData[0]
1490 lea (%rsi,%r10),%r10 # end of prologue label
1491 cmp %r10,%rbx # context->Rip<body label
1494 mov 152($context),%rax # pull context->Rsp
1496 mov 4(%r11),%r10d # HandlerData[1]
1497 lea (%rsi,%r10),%r10 # epilogue label
1498 cmp %r10,%rbx # context->Rip>=epilogue label
1501 mov `32*17`($context),%rax # pull saved stack pointer
1509 mov %rbx,144($context) # restore context->Rbx
1510 mov %rbp,160($context) # restore context->Rbp
1511 mov %r12,216($context) # restore cotnext->R12
1512 mov %r13,224($context) # restore cotnext->R13
1513 mov %r14,232($context) # restore cotnext->R14
1514 mov %r15,240($context) # restore cotnext->R15
1516 lea -56-10*16(%rax),%rsi
1517 lea 512($context),%rdi # &context.Xmm6
1519 .long 0xa548f3fc # cld; rep movsq
1522 .size avx2_handler,.-avx2_handler
1527 .rva .LSEH_begin_sha1_multi_block
1528 .rva .LSEH_end_sha1_multi_block
1529 .rva .LSEH_info_sha1_multi_block
1530 .rva .LSEH_begin_sha1_multi_block_shaext
1531 .rva .LSEH_end_sha1_multi_block_shaext
1532 .rva .LSEH_info_sha1_multi_block_shaext
1534 $code.=<<___ if ($avx);
1535 .rva .LSEH_begin_sha1_multi_block_avx
1536 .rva .LSEH_end_sha1_multi_block_avx
1537 .rva .LSEH_info_sha1_multi_block_avx
1539 $code.=<<___ if ($avx>1);
1540 .rva .LSEH_begin_sha1_multi_block_avx2
1541 .rva .LSEH_end_sha1_multi_block_avx2
1542 .rva .LSEH_info_sha1_multi_block_avx2
1547 .LSEH_info_sha1_multi_block:
1550 .rva .Lbody,.Lepilogue # HandlerData[]
1551 .LSEH_info_sha1_multi_block_shaext:
1554 .rva .Lbody_shaext,.Lepilogue_shaext # HandlerData[]
1556 $code.=<<___ if ($avx);
1557 .LSEH_info_sha1_multi_block_avx:
1560 .rva .Lbody_avx,.Lepilogue_avx # HandlerData[]
1562 $code.=<<___ if ($avx>1);
1563 .LSEH_info_sha1_multi_block_avx2:
1566 .rva .Lbody_avx2,.Lepilogue_avx2 # HandlerData[]
1569 ####################################################################
1572 local *opcode=shift;
1576 $rex|=0x04 if ($dst>=8);
1577 $rex|=0x01 if ($src>=8);
1578 unshift @opcode,$rex|0x40 if ($rex);
1582 if (@_[0] =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1583 my @opcode=(0x0f,0x3a,0xcc);
1584 rex(\@opcode,$3,$2);
1585 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1587 push @opcode,$c=~/^0/?oct($c):$c;
1588 return ".byte\t".join(',',@opcode);
1590 return "sha1rnds4\t".@_[0];
1597 "sha1nexte" => 0xc8,
1599 "sha1msg2" => 0xca );
1601 if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1602 my @opcode=(0x0f,0x38);
1603 rex(\@opcode,$2,$1);
1604 push @opcode,$opcodelet{$instr};
1605 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
1606 return ".byte\t".join(',',@opcode);
1608 return $instr."\t".@_[0];
1612 foreach (split("\n",$code)) {
1613 s/\`([^\`]*)\`/eval($1)/ge;
1615 s/\b(sha1rnds4)\s+(.*)/sha1rnds4($2)/geo or
1616 s/\b(sha1[^\s]*)\s+(.*)/sha1op38($1,$2)/geo or
1618 s/\b(vmov[dq])\b(.+)%ymm([0-9]+)/$1$2%xmm$3/go or
1619 s/\b(vmovdqu)\b(.+)%x%ymm([0-9]+)/$1$2%xmm$3/go or
1620 s/\b(vpinsr[qd])\b(.+)%ymm([0-9]+),%ymm([0-9]+)/$1$2%xmm$3,%xmm$4/go or
1621 s/\b(vpextr[qd])\b(.+)%ymm([0-9]+)/$1$2%xmm$3/go or
1622 s/\b(vinserti128)\b(\s+)%ymm/$1$2\$1,%xmm/go or
1623 s/\b(vpbroadcast[qd]\s+)%ymm([0-9]+)/$1%xmm$2/go;