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
366 mov OPENSSL_ia32cap_P+4(%rip),%rcx
367 bt \$61,%rcx # check SHA bit
370 $code.=<<___ if ($avx);
379 $code.=<<___ if ($win64);
382 movaps %xmm7,0x10(%rsp)
383 movaps %xmm8,0x20(%rsp)
384 movaps %xmm9,0x30(%rsp)
385 movaps %xmm10,-0x78(%rax)
386 movaps %xmm11,-0x68(%rax)
387 movaps %xmm12,-0x58(%rax)
388 movaps %xmm13,-0x48(%rax)
389 movaps %xmm14,-0x38(%rax)
390 movaps %xmm15,-0x28(%rax)
393 sub \$`$REG_SZ*18`,%rsp
395 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
397 lea K_XX_XX(%rip),$Tbl
398 lea `$REG_SZ*16`(%rsp),%rbx
401 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
404 for($i=0;$i<4;$i++) {
406 mov `16*$i+0`($inp),@ptr[$i] # input pointer
407 mov `16*$i+8`($inp),%ecx # number of blocks
409 cmovg %ecx,$num # find maximum
411 mov %ecx,`4*$i`(%rbx) # initialize counters
412 cmovle $Tbl,@ptr[$i] # cancel input
419 movdqu 0x00($ctx),$A # load context
425 movdqa 0x60($Tbl),$tx # pbswap_mask
426 movdqa -0x20($Tbl),$K # K_00_19
432 for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
433 $code.=" movdqa 0x00($Tbl),$K\n"; # K_20_39
434 for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
435 $code.=" movdqa 0x20($Tbl),$K\n"; # K_40_59
436 for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
437 $code.=" movdqa 0x40($Tbl),$K\n"; # K_60_79
438 for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
440 movdqa (%rbx),@Xi[0] # pull counters
442 cmp 4*0(%rbx),%ecx # examinte counters
444 cmovge $Tbl,@ptr[0] # cancel input
449 pcmpgtd $t2,@Xi[1] # mask value
452 paddd @Xi[1],@Xi[0] # counters--
455 movdqu 0x00($ctx),$t0
457 movdqu 0x20($ctx),$t1
460 movdqu 0x40($ctx),$t2
463 movdqu 0x60($ctx),$t3
466 movdqu 0x80($ctx),$tx
476 movdqa @Xi[0],(%rbx) # save counters
477 movdqa 0x60($Tbl),$tx # pbswap_mask
478 movdqa -0x20($Tbl),$K # K_00_19
482 mov `$REG_SZ*17+8`(%rsp),$num
483 lea $REG_SZ($ctx),$ctx
484 lea `16*$REG_SZ/4`($inp),$inp
489 mov `$REG_SZ*17`(%rsp),%rax # original %rsp
491 $code.=<<___ if ($win64);
492 movaps -0xb8(%rax),%xmm6
493 movaps -0xa8(%rax),%xmm7
494 movaps -0x98(%rax),%xmm8
495 movaps -0x88(%rax),%xmm9
496 movaps -0x78(%rax),%xmm10
497 movaps -0x68(%rax),%xmm11
498 movaps -0x58(%rax),%xmm12
499 movaps -0x48(%rax),%xmm13
500 movaps -0x38(%rax),%xmm14
501 movaps -0x28(%rax),%xmm15
509 .size sha1_multi_block,.-sha1_multi_block
512 my ($ABCD0,$E0,$E0_,$BSWAP,$ABCD1,$E1,$E1_)=map("%xmm$_",(0..3,8..10));
513 my @MSG0=map("%xmm$_",(4..7));
514 my @MSG1=map("%xmm$_",(11..14));
517 .type sha1_multi_block_shaext,\@function,3
519 sha1_multi_block_shaext:
525 $code.=<<___ if ($win64);
528 movaps %xmm7,0x10(%rsp)
529 movaps %xmm8,0x20(%rsp)
530 movaps %xmm9,0x30(%rsp)
531 movaps %xmm10,-0x78(%rax)
532 movaps %xmm11,-0x68(%rax)
533 movaps %xmm12,-0x58(%rax)
534 movaps %xmm13,-0x48(%rax)
535 movaps %xmm14,-0x38(%rax)
536 movaps %xmm15,-0x28(%rax)
539 sub \$`$REG_SZ*18`,%rsp
540 shl \$1,$num # we process pair at a time
542 lea 0x40($ctx),$ctx # size optimization
543 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
545 lea `$REG_SZ*16`(%rsp),%rbx
546 movdqa K_XX_XX+0x80(%rip),$BSWAP # byte-n-word swap
549 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
552 for($i=0;$i<2;$i++) {
554 mov `16*$i+0`($inp),@ptr[$i] # input pointer
555 mov `16*$i+8`($inp),%ecx # number of blocks
557 cmovg %ecx,$num # find maximum
559 mov %ecx,`4*$i`(%rbx) # initialize counters
560 cmovle %rsp,@ptr[$i] # cancel input
567 movq 0x00-0x40($ctx),$ABCD0 # a1.a0
568 movq 0x20-0x40($ctx),@MSG0[0]# b1.b0
569 movq 0x40-0x40($ctx),@MSG0[1]# c1.c0
570 movq 0x60-0x40($ctx),@MSG0[2]# d1.d0
571 movq 0x80-0x40($ctx),@MSG0[3]# e1.e0
573 punpckldq @MSG0[0],$ABCD0 # b1.a1.b0.a0
574 punpckldq @MSG0[2],@MSG0[1] # d1.c1.d0.c0
577 punpcklqdq @MSG0[1],$ABCD0 # d0.c0.b0.a0
578 punpckhqdq @MSG0[1],$ABCD1 # d1.c1.b1.a1
580 pshufd \$0b00111111,@MSG0[3],$E0
581 pshufd \$0b01111111,@MSG0[3],$E1
582 pshufd \$0b00011011,$ABCD0,$ABCD0
583 pshufd \$0b00011011,$ABCD1,$ABCD1
588 movdqu 0x00(@ptr[0]),@MSG0[0]
589 movdqu 0x00(@ptr[1]),@MSG1[0]
590 movdqu 0x10(@ptr[0]),@MSG0[1]
591 movdqu 0x10(@ptr[1]),@MSG1[1]
592 movdqu 0x20(@ptr[0]),@MSG0[2]
593 pshufb $BSWAP,@MSG0[0]
594 movdqu 0x20(@ptr[1]),@MSG1[2]
595 pshufb $BSWAP,@MSG1[0]
596 movdqu 0x30(@ptr[0]),@MSG0[3]
597 lea 0x40(@ptr[0]),@ptr[0]
598 pshufb $BSWAP,@MSG0[1]
599 movdqu 0x30(@ptr[1]),@MSG1[3]
600 lea 0x40(@ptr[1]),@ptr[1]
601 pshufb $BSWAP,@MSG1[1]
603 movdqa $E0,0x50(%rsp) # offload
605 movdqa $E1,0x70(%rsp)
607 movdqa $ABCD0,0x40(%rsp) # offload
609 movdqa $ABCD1,0x60(%rsp)
611 sha1rnds4 \$0,$E0,$ABCD0 # 0-3
612 sha1nexte @MSG0[1],$E0_
613 sha1rnds4 \$0,$E1,$ABCD1 # 0-3
614 sha1nexte @MSG1[1],$E1_
615 pshufb $BSWAP,@MSG0[2]
616 prefetcht0 127(@ptr[0])
617 sha1msg1 @MSG0[1],@MSG0[0]
618 pshufb $BSWAP,@MSG1[2]
619 prefetcht0 127(@ptr[1])
620 sha1msg1 @MSG1[1],@MSG1[0]
622 pshufb $BSWAP,@MSG0[3]
624 pshufb $BSWAP,@MSG1[3]
626 sha1rnds4 \$0,$E0_,$ABCD0 # 4-7
627 sha1nexte @MSG0[2],$E0
628 sha1rnds4 \$0,$E1_,$ABCD1 # 4-7
629 sha1nexte @MSG1[2],$E1
630 pxor @MSG0[2],@MSG0[0]
631 sha1msg1 @MSG0[2],@MSG0[1]
632 pxor @MSG1[2],@MSG1[0]
633 sha1msg1 @MSG1[2],@MSG1[1]
635 for($i=2;$i<20-4;$i++) {
639 sha1rnds4 \$`int($i/5)`,$E0,$ABCD0 # 8-11
640 sha1nexte @MSG0[3],$E0_
641 sha1rnds4 \$`int($i/5)`,$E1,$ABCD1 # 8-11
642 sha1nexte @MSG1[3],$E1_
643 sha1msg2 @MSG0[3],@MSG0[0]
644 sha1msg2 @MSG1[3],@MSG1[0]
645 pxor @MSG0[3],@MSG0[1]
646 sha1msg1 @MSG0[3],@MSG0[2]
647 pxor @MSG1[3],@MSG1[1]
648 sha1msg1 @MSG1[3],@MSG1[2]
650 ($E0,$E0_)=($E0_,$E0); ($E1,$E1_)=($E1_,$E1);
651 push(@MSG0,shift(@MSG0)); push(@MSG1,shift(@MSG1));
656 sha1rnds4 \$3,$E0,$ABCD0 # 64-67
657 sha1nexte @MSG0[3],$E0_
658 sha1rnds4 \$3,$E1,$ABCD1 # 64-67
659 sha1nexte @MSG1[3],$E1_
660 sha1msg2 @MSG0[3],@MSG0[0]
661 sha1msg2 @MSG1[3],@MSG1[0]
662 pxor @MSG0[3],@MSG0[1]
663 pxor @MSG1[3],@MSG1[1]
666 pxor @MSG0[2],@MSG0[2] # zero
667 cmp 4*0(%rbx),%ecx # examine counters
668 cmovge %rsp,@ptr[0] # cancel input
672 sha1rnds4 \$3,$E0_,$ABCD0 # 68-71
673 sha1nexte @MSG0[0],$E0
674 sha1rnds4 \$3,$E1_,$ABCD1 # 68-71
675 sha1nexte @MSG1[0],$E1
676 sha1msg2 @MSG0[0],@MSG0[1]
677 sha1msg2 @MSG1[0],@MSG1[1]
681 movq (%rbx),@MSG0[0] # pull counters
685 sha1rnds4 \$3,$E0,$ABCD0 # 72-75
686 sha1nexte @MSG0[1],$E0_
687 sha1rnds4 \$3,$E1,$ABCD1 # 72-75
688 sha1nexte @MSG1[1],$E1_
690 pshufd \$0x00,@MSG0[0],@MSG1[2]
691 pshufd \$0x55,@MSG0[0],@MSG1[3]
692 movdqa @MSG0[0],@MSG0[1]
693 pcmpgtd @MSG0[2],@MSG1[2]
694 pcmpgtd @MSG0[2],@MSG1[3]
698 sha1rnds4 \$3,$E0_,$ABCD0 # 76-79
699 sha1nexte $MSG0[2],$E0
700 sha1rnds4 \$3,$E1_,$ABCD1 # 76-79
701 sha1nexte $MSG0[2],$E1
703 pcmpgtd @MSG0[2],@MSG0[1] # counter mask
708 paddd @MSG0[1],@MSG0[0] # counters--
710 paddd 0x40(%rsp),$ABCD0
712 paddd 0x60(%rsp),$ABCD1
715 movq @MSG0[0],(%rbx) # save counters
719 mov `$REG_SZ*17+8`(%rsp),$num
721 pshufd \$0b00011011,$ABCD0,$ABCD0
722 pshufd \$0b00011011,$ABCD1,$ABCD1
724 movdqa $ABCD0,@MSG0[0]
725 punpckldq $ABCD1,$ABCD0 # b1.b0.a1.a0
726 punpckhdq $ABCD1,@MSG0[0] # d1.d0.c1.c0
727 punpckhdq $E1,$E0 # e1.e0.xx.xx
728 movq $ABCD0,0x00-0x40($ctx) # a1.a0
730 movq @MSG0[0],0x40-0x40($ctx)# c1.c0
732 movq $ABCD0,0x20-0x40($ctx) # b1.b0
734 movq @MSG0[0],0x60-0x40($ctx)# d1.d0
735 movq $E0,0x80-0x40($ctx) # e1.e0
737 lea `$REG_SZ/2`($ctx),$ctx
738 lea `16*2`($inp),$inp
740 jnz .Loop_grande_shaext
743 #mov `$REG_SZ*17`(%rsp),%rax # original %rsp
745 $code.=<<___ if ($win64);
746 movaps -0xb8(%rax),%xmm6
747 movaps -0xa8(%rax),%xmm7
748 movaps -0x98(%rax),%xmm8
749 movaps -0x88(%rax),%xmm9
750 movaps -0x78(%rax),%xmm10
751 movaps -0x68(%rax),%xmm11
752 movaps -0x58(%rax),%xmm12
753 movaps -0x48(%rax),%xmm13
754 movaps -0x38(%rax),%xmm14
755 movaps -0x28(%rax),%xmm15
763 .size sha1_multi_block_shaext,.-sha1_multi_block_shaext
769 my ($i,$a,$b,$c,$d,$e)=@_;
772 my $vpack = $REG_SZ==16 ? "vpunpckldq" : "vinserti128";
773 my $ptr_n = $REG_SZ==16 ? @ptr[1] : @ptr[4];
775 $code.=<<___ if ($i==0 && $REG_SZ==16);
776 vmovd (@ptr[0]),@Xi[0]
777 lea `16*4`(@ptr[0]),@ptr[0]
778 vmovd (@ptr[1]),@Xi[2] # borrow Xi[2]
779 lea `16*4`(@ptr[1]),@ptr[1]
780 vpinsrd \$1,(@ptr[2]),@Xi[0],@Xi[0]
781 lea `16*4`(@ptr[2]),@ptr[2]
782 vpinsrd \$1,(@ptr[3]),@Xi[2],@Xi[2]
783 lea `16*4`(@ptr[3]),@ptr[3]
784 vmovd `4*$j-16*4`(@ptr[0]),@Xi[1]
785 vpunpckldq @Xi[2],@Xi[0],@Xi[0]
786 vmovd `4*$j-16*4`($ptr_n),$t3
787 vpshufb $tx,@Xi[0],@Xi[0]
789 $code.=<<___ if ($i<15 && $REG_SZ==16); # just load input
790 vpinsrd \$1,`4*$j-16*4`(@ptr[2]),@Xi[1],@Xi[1]
791 vpinsrd \$1,`4*$j-16*4`(@ptr[3]),$t3,$t3
793 $code.=<<___ if ($i==0 && $REG_SZ==32);
794 vmovd (@ptr[0]),@Xi[0]
795 lea `16*4`(@ptr[0]),@ptr[0]
796 vmovd (@ptr[4]),@Xi[2] # borrow Xi[2]
797 lea `16*4`(@ptr[4]),@ptr[4]
799 lea `16*4`(@ptr[1]),@ptr[1]
801 lea `16*4`(@ptr[5]),@ptr[5]
802 vpinsrd \$1,(@ptr[2]),@Xi[0],@Xi[0]
803 lea `16*4`(@ptr[2]),@ptr[2]
804 vpinsrd \$1,(@ptr[6]),@Xi[2],@Xi[2]
805 lea `16*4`(@ptr[6]),@ptr[6]
806 vpinsrd \$1,(@ptr[3]),$t2,$t2
807 lea `16*4`(@ptr[3]),@ptr[3]
808 vpunpckldq $t2,@Xi[0],@Xi[0]
809 vpinsrd \$1,(@ptr[7]),$t1,$t1
810 lea `16*4`(@ptr[7]),@ptr[7]
811 vpunpckldq $t1,@Xi[2],@Xi[2]
812 vmovd `4*$j-16*4`(@ptr[0]),@Xi[1]
813 vinserti128 @Xi[2],@Xi[0],@Xi[0]
814 vmovd `4*$j-16*4`($ptr_n),$t3
815 vpshufb $tx,@Xi[0],@Xi[0]
817 $code.=<<___ if ($i<15 && $REG_SZ==32); # just load input
818 vmovd `4*$j-16*4`(@ptr[1]),$t2
819 vmovd `4*$j-16*4`(@ptr[5]),$t1
820 vpinsrd \$1,`4*$j-16*4`(@ptr[2]),@Xi[1],@Xi[1]
821 vpinsrd \$1,`4*$j-16*4`(@ptr[6]),$t3,$t3
822 vpinsrd \$1,`4*$j-16*4`(@ptr[3]),$t2,$t2
823 vpunpckldq $t2,@Xi[1],@Xi[1]
824 vpinsrd \$1,`4*$j-16*4`(@ptr[7]),$t1,$t1
825 vpunpckldq $t1,$t3,$t3
827 $code.=<<___ if ($i<14);
828 vpaddd $K,$e,$e # e+=K_00_19
833 vmovdqa @Xi[0],`&Xi_off($i)`
834 vpaddd @Xi[0],$e,$e # e+=X[i]
835 $vpack $t3,@Xi[1],@Xi[1]
837 vpxor $t1,$t0,$t0 # Ch(b,c,d)
838 vmovd `4*$k-16*4`(@ptr[0]),@Xi[2]
841 vpor $t3,$t2,$t2 # rol(a,5)
842 vmovd `4*$k-16*4`($ptr_n),$t3
843 vpaddd $t0,$e,$e # e+=Ch(b,c,d)
846 vpaddd $t2,$e,$e # e+=rol(a,5)
847 vpshufb $tx,@Xi[1],@Xi[1]
848 vpor $t1,$b,$b # b=rol(b,30)
850 $code.=<<___ if ($i==14);
851 vpaddd $K,$e,$e # e+=K_00_19
852 prefetcht0 63(@ptr[0])
857 vmovdqa @Xi[0],`&Xi_off($i)`
858 vpaddd @Xi[0],$e,$e # e+=X[i]
859 $vpack $t3,@Xi[1],@Xi[1]
861 prefetcht0 63(@ptr[1])
862 vpxor $t1,$t0,$t0 # Ch(b,c,d)
865 vpor $t3,$t2,$t2 # rol(a,5)
866 prefetcht0 63(@ptr[2])
867 vpaddd $t0,$e,$e # e+=Ch(b,c,d)
870 vpaddd $t2,$e,$e # e+=rol(a,5)
871 prefetcht0 63(@ptr[3])
872 vpshufb $tx,@Xi[1],@Xi[1]
873 vpor $t1,$b,$b # b=rol(b,30)
875 $code.=<<___ if ($i>=13 && $i<15);
876 vmovdqa `&Xi_off($j+2)`,@Xi[3] # preload "X[2]"
878 $code.=<<___ if ($i>=15); # apply Xupdate
879 vpxor @Xi[-2],@Xi[1],@Xi[1] # "X[13]"
880 vmovdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
882 vpaddd $K,$e,$e # e+=K_00_19
885 `"prefetcht0 63(@ptr[4])" if ($i==15 && $REG_SZ==32)`
888 vmovdqa @Xi[0],`&Xi_off($i)`
889 vpaddd @Xi[0],$e,$e # e+=X[i]
890 vpxor `&Xi_off($j+8)`,@Xi[1],@Xi[1]
892 vpxor $t1,$t0,$t0 # Ch(b,c,d)
893 vpxor @Xi[3],@Xi[1],@Xi[1]
894 `"prefetcht0 63(@ptr[5])" if ($i==15 && $REG_SZ==32)`
897 vpor $t3,$t2,$t2 # rol(a,5)
898 vpaddd $t0,$e,$e # e+=Ch(b,c,d)
899 `"prefetcht0 63(@ptr[6])" if ($i==15 && $REG_SZ==32)`
900 vpsrld \$31,@Xi[1],$tx
901 vpaddd @Xi[1],@Xi[1],@Xi[1]
904 `"prefetcht0 63(@ptr[7])" if ($i==15 && $REG_SZ==32)`
905 vpaddd $t2,$e,$e # e+=rol(a,5)
906 vpor $tx,@Xi[1],@Xi[1] # rol \$1,@Xi[1]
907 vpor $t1,$b,$b # b=rol(b,30)
909 push(@Xi,shift(@Xi));
913 my ($i,$a,$b,$c,$d,$e)=@_;
916 $code.=<<___ if ($i<79);
917 vpxor @Xi[-2],@Xi[1],@Xi[1] # "X[13]"
918 vmovdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
921 vpaddd $K,$e,$e # e+=K_20_39
924 $code.=<<___ if ($i<72);
925 vmovdqa @Xi[0],`&Xi_off($i)`
927 $code.=<<___ if ($i<79);
928 vpaddd @Xi[0],$e,$e # e+=X[i]
929 vpxor `&Xi_off($j+8)`,@Xi[1],@Xi[1]
931 vpxor $c,$t0,$t0 # Parity(b,c,d)
932 vpxor @Xi[3],@Xi[1],@Xi[1]
935 vpor $t3,$t2,$t2 # rol(a,5)
936 vpaddd $t0,$e,$e # e+=Parity(b,c,d)
937 vpsrld \$31,@Xi[1],$tx
938 vpaddd @Xi[1],@Xi[1],@Xi[1]
941 vpaddd $t2,$e,$e # e+=rol(a,5)
942 vpor $tx,@Xi[1],@Xi[1] # rol(@Xi[1],1)
943 vpor $t1,$b,$b # b=rol(b,30)
945 $code.=<<___ if ($i==79);
947 vpaddd $K,$e,$e # e+=K_20_39
951 vpaddd @Xi[0],$e,$e # e+=X[i]
952 vpxor $c,$t0,$t0 # Parity(b,c,d)
955 vpor $t3,$t2,$t2 # rol(a,5)
956 vpaddd $t0,$e,$e # e+=Parity(b,c,d)
959 vpaddd $t2,$e,$e # e+=rol(a,5)
960 vpor $t1,$b,$b # b=rol(b,30)
962 push(@Xi,shift(@Xi));
966 my ($i,$a,$b,$c,$d,$e)=@_;
970 vpxor @Xi[-2],@Xi[1],@Xi[1] # "X[13]"
971 vmovdqa `&Xi_off($j+2)`,@Xi[3] # "X[2]"
973 vpaddd $K,$e,$e # e+=K_40_59
976 vpxor `&Xi_off($j+8)`,@Xi[1],@Xi[1]
981 vpxor @Xi[3],@Xi[1],@Xi[1]
983 vmovdqu @Xi[0],`&Xi_off($i)`
984 vpaddd @Xi[0],$e,$e # e+=X[i]
985 vpor $t3,$t2,$t2 # rol(a,5)
986 vpsrld \$31,@Xi[1],$tx
988 vpaddd @Xi[1],@Xi[1],@Xi[1]
991 vpaddd $t0,$e,$e # e+=Maj(b,d,c)
994 vpaddd $t2,$e,$e # e+=rol(a,5)
995 vpor $tx,@Xi[1],@Xi[1] # rol(@X[1],1)
996 vpor $t1,$b,$b # b=rol(b,30)
998 push(@Xi,shift(@Xi));
1002 .type sha1_multi_block_avx,\@function,3
1004 sha1_multi_block_avx:
1007 $code.=<<___ if ($avx>1);
1022 $code.=<<___ if ($win64);
1023 lea -0xa8(%rsp),%rsp
1025 movaps %xmm7,0x10(%rsp)
1026 movaps %xmm8,0x20(%rsp)
1027 movaps %xmm9,0x30(%rsp)
1028 movaps %xmm10,-0x78(%rax)
1029 movaps %xmm11,-0x68(%rax)
1030 movaps %xmm12,-0x58(%rax)
1031 movaps %xmm13,-0x48(%rax)
1032 movaps %xmm14,-0x38(%rax)
1033 movaps %xmm15,-0x28(%rax)
1036 sub \$`$REG_SZ*18`, %rsp
1038 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
1040 lea K_XX_XX(%rip),$Tbl
1041 lea `$REG_SZ*16`(%rsp),%rbx
1045 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
1048 for($i=0;$i<4;$i++) {
1050 mov `16*$i+0`($inp),@ptr[$i] # input pointer
1051 mov `16*$i+8`($inp),%ecx # number of blocks
1053 cmovg %ecx,$num # find maximum
1055 mov %ecx,`4*$i`(%rbx) # initialize counters
1056 cmovle $Tbl,@ptr[$i] # cancel input
1063 vmovdqu 0x00($ctx),$A # load context
1065 vmovdqu 0x20($ctx),$B
1066 vmovdqu 0x40($ctx),$C
1067 vmovdqu 0x60($ctx),$D
1068 vmovdqu 0x80($ctx),$E
1069 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1075 $code.=" vmovdqa -0x20($Tbl),$K\n"; # K_00_19
1076 for($i=0;$i<20;$i++) { &BODY_00_19_avx($i,@V); unshift(@V,pop(@V)); }
1077 $code.=" vmovdqa 0x00($Tbl),$K\n"; # K_20_39
1078 for(;$i<40;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1079 $code.=" vmovdqa 0x20($Tbl),$K\n"; # K_40_59
1080 for(;$i<60;$i++) { &BODY_40_59_avx($i,@V); unshift(@V,pop(@V)); }
1081 $code.=" vmovdqa 0x40($Tbl),$K\n"; # K_60_79
1082 for(;$i<80;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1086 for($i=0;$i<4;$i++) {
1088 cmp `4*$i`(%rbx),%ecx # examine counters
1089 cmovge $Tbl,@ptr[$i] # cancel input
1093 vmovdqu (%rbx),$t0 # pull counters
1096 vpcmpgtd $t2,$t1,$t1 # mask value
1097 vpaddd $t1,$t0,$t0 # counters--
1101 vpaddd 0x00($ctx),$A,$A
1103 vpaddd 0x20($ctx),$B,$B
1105 vpaddd 0x40($ctx),$C,$C
1107 vpaddd 0x60($ctx),$D,$D
1108 vpaddd 0x80($ctx),$E,$E
1109 vmovdqu $A,0x00($ctx)
1110 vmovdqu $B,0x20($ctx)
1111 vmovdqu $C,0x40($ctx)
1112 vmovdqu $D,0x60($ctx)
1113 vmovdqu $E,0x80($ctx)
1115 vmovdqu $t0,(%rbx) # save counters
1116 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1120 mov `$REG_SZ*17+8`(%rsp),$num
1121 lea $REG_SZ($ctx),$ctx
1122 lea `16*$REG_SZ/4`($inp),$inp
1124 jnz .Loop_grande_avx
1127 mov `$REG_SZ*17`(%rsp),%rax # original %rsp
1130 $code.=<<___ if ($win64);
1131 movaps -0xb8(%rax),%xmm6
1132 movaps -0xa8(%rax),%xmm7
1133 movaps -0x98(%rax),%xmm8
1134 movaps -0x88(%rax),%xmm9
1135 movaps -0x78(%rax),%xmm10
1136 movaps -0x68(%rax),%xmm11
1137 movaps -0x58(%rax),%xmm12
1138 movaps -0x48(%rax),%xmm13
1139 movaps -0x38(%rax),%xmm14
1140 movaps -0x28(%rax),%xmm15
1148 .size sha1_multi_block_avx,.-sha1_multi_block_avx
1152 $code =~ s/\`([^\`]*)\`/eval $1/gem;
1156 @ptr=map("%r$_",(12..15,8..11));
1158 @V=($A,$B,$C,$D,$E)=map("%ymm$_",(0..4));
1159 ($t0,$t1,$t2,$t3,$tx)=map("%ymm$_",(5..9));
1160 @Xi=map("%ymm$_",(10..14));
1164 .type sha1_multi_block_avx2,\@function,3
1166 sha1_multi_block_avx2:
1176 $code.=<<___ if ($win64);
1177 lea -0xa8(%rsp),%rsp
1179 movaps %xmm7,0x10(%rsp)
1180 movaps %xmm8,0x20(%rsp)
1181 movaps %xmm9,0x30(%rsp)
1182 movaps %xmm10,0x40(%rsp)
1183 movaps %xmm11,0x50(%rsp)
1184 movaps %xmm12,-0x78(%rax)
1185 movaps %xmm13,-0x68(%rax)
1186 movaps %xmm14,-0x58(%rax)
1187 movaps %xmm15,-0x48(%rax)
1190 sub \$`$REG_SZ*18`, %rsp
1192 mov %rax,`$REG_SZ*17`(%rsp) # original %rsp
1194 lea K_XX_XX(%rip),$Tbl
1199 mov $num,`$REG_SZ*17+8`(%rsp) # original $num
1201 lea `$REG_SZ*16`(%rsp),%rbx
1203 for($i=0;$i<8;$i++) {
1205 mov `16*$i+0`($inp),@ptr[$i] # input pointer
1206 mov `16*$i+8`($inp),%ecx # number of blocks
1208 cmovg %ecx,$num # find maximum
1210 mov %ecx,`4*$i`(%rbx) # initialize counters
1211 cmovle $Tbl,@ptr[$i] # cancel input
1215 vmovdqu 0x00($ctx),$A # load context
1217 vmovdqu 0x20($ctx),$B
1218 lea 256+128(%rsp),%rbx
1219 vmovdqu 0x40($ctx),$C
1220 vmovdqu 0x60($ctx),$D
1221 vmovdqu 0x80($ctx),$E
1222 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1228 $code.=" vmovdqa -0x20($Tbl),$K\n"; # K_00_19
1229 for($i=0;$i<20;$i++) { &BODY_00_19_avx($i,@V); unshift(@V,pop(@V)); }
1230 $code.=" vmovdqa 0x00($Tbl),$K\n"; # K_20_39
1231 for(;$i<40;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1232 $code.=" vmovdqa 0x20($Tbl),$K\n"; # K_40_59
1233 for(;$i<60;$i++) { &BODY_40_59_avx($i,@V); unshift(@V,pop(@V)); }
1234 $code.=" vmovdqa 0x40($Tbl),$K\n"; # K_60_79
1235 for(;$i<80;$i++) { &BODY_20_39_avx($i,@V); unshift(@V,pop(@V)); }
1238 lea `$REG_SZ*16`(%rsp),%rbx
1240 for($i=0;$i<8;$i++) {
1242 cmp `4*$i`(%rbx),%ecx # examine counters
1243 cmovge $Tbl,@ptr[$i] # cancel input
1247 vmovdqu (%rbx),$t0 # pull counters
1250 vpcmpgtd $t2,$t1,$t1 # mask value
1251 vpaddd $t1,$t0,$t0 # counters--
1255 vpaddd 0x00($ctx),$A,$A
1257 vpaddd 0x20($ctx),$B,$B
1259 vpaddd 0x40($ctx),$C,$C
1261 vpaddd 0x60($ctx),$D,$D
1262 vpaddd 0x80($ctx),$E,$E
1263 vmovdqu $A,0x00($ctx)
1264 vmovdqu $B,0x20($ctx)
1265 vmovdqu $C,0x40($ctx)
1266 vmovdqu $D,0x60($ctx)
1267 vmovdqu $E,0x80($ctx)
1269 vmovdqu $t0,(%rbx) # save counters
1270 lea 256+128(%rsp),%rbx
1271 vmovdqu 0x60($Tbl),$tx # pbswap_mask
1275 #mov `$REG_SZ*17+8`(%rsp),$num
1276 #lea $REG_SZ($ctx),$ctx
1277 #lea `16*$REG_SZ/4`($inp),$inp
1279 #jnz .Loop_grande_avx2
1282 mov `$REG_SZ*17`(%rsp),%rax # original %rsp
1285 $code.=<<___ if ($win64);
1286 movaps -0xd8(%rax),%xmm6
1287 movaps -0xc8(%rax),%xmm7
1288 movaps -0xb8(%rax),%xmm8
1289 movaps -0xa8(%rax),%xmm9
1290 movaps -0x98(%rax),%xmm10
1291 movaps -0x88(%rax),%xmm11
1292 movaps -0x78(%rax),%xmm12
1293 movaps -0x68(%rax),%xmm13
1294 movaps -0x58(%rax),%xmm14
1295 movaps -0x48(%rax),%xmm15
1307 .size sha1_multi_block_avx2,.-sha1_multi_block_avx2
1313 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1314 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1316 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1317 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1318 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1319 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1320 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1321 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1322 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap
1323 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap
1324 .byte 0xf,0xe,0xd,0xc,0xb,0xa,0x9,0x8,0x7,0x6,0x5,0x4,0x3,0x2,0x1,0x0
1325 .asciz "SHA1 multi-block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1329 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1330 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1337 .extern __imp_RtlVirtualUnwind
1338 .type se_handler,\@abi-omnipotent
1352 mov 120($context),%rax # pull context->Rax
1353 mov 248($context),%rbx # pull context->Rip
1355 mov 8($disp),%rsi # disp->ImageBase
1356 mov 56($disp),%r11 # disp->HandlerData
1358 mov 0(%r11),%r10d # HandlerData[0]
1359 lea (%rsi,%r10),%r10 # end of prologue label
1360 cmp %r10,%rbx # context->Rip<.Lbody
1363 mov 152($context),%rax # pull context->Rsp
1365 mov 4(%r11),%r10d # HandlerData[1]
1366 lea (%rsi,%r10),%r10 # epilogue label
1367 cmp %r10,%rbx # context->Rip>=.Lepilogue
1370 mov `16*17`(%rax),%rax # pull saved stack pointer
1374 mov %rbx,144($context) # restore context->Rbx
1375 mov %rbp,160($context) # restore context->Rbp
1377 lea -24-10*16(%rax),%rsi
1378 lea 512($context),%rdi # &context.Xmm6
1380 .long 0xa548f3fc # cld; rep movsq
1385 mov %rax,152($context) # restore context->Rsp
1386 mov %rsi,168($context) # restore context->Rsi
1387 mov %rdi,176($context) # restore context->Rdi
1389 mov 40($disp),%rdi # disp->ContextRecord
1390 mov $context,%rsi # context
1391 mov \$154,%ecx # sizeof(CONTEXT)
1392 .long 0xa548f3fc # cld; rep movsq
1395 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1396 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1397 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1398 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1399 mov 40(%rsi),%r10 # disp->ContextRecord
1400 lea 56(%rsi),%r11 # &disp->HandlerData
1401 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1402 mov %r10,32(%rsp) # arg5
1403 mov %r11,40(%rsp) # arg6
1404 mov %r12,48(%rsp) # arg7
1405 mov %rcx,56(%rsp) # arg8, (NULL)
1406 call *__imp_RtlVirtualUnwind(%rip)
1408 mov \$1,%eax # ExceptionContinueSearch
1420 .size se_handler,.-se_handler
1422 $code.=<<___ if ($avx>1);
1423 .type avx2_handler,\@abi-omnipotent
1437 mov 120($context),%rax # pull context->Rax
1438 mov 248($context),%rbx # pull context->Rip
1440 mov 8($disp),%rsi # disp->ImageBase
1441 mov 56($disp),%r11 # disp->HandlerData
1443 mov 0(%r11),%r10d # HandlerData[0]
1444 lea (%rsi,%r10),%r10 # end of prologue label
1445 cmp %r10,%rbx # context->Rip<body label
1448 mov 152($context),%rax # pull context->Rsp
1450 mov 4(%r11),%r10d # HandlerData[1]
1451 lea (%rsi,%r10),%r10 # epilogue label
1452 cmp %r10,%rbx # context->Rip>=epilogue label
1455 mov `32*17`($context),%rax # pull saved stack pointer
1463 mov %rbx,144($context) # restore context->Rbx
1464 mov %rbp,160($context) # restore context->Rbp
1465 mov %r12,216($context) # restore cotnext->R12
1466 mov %r13,224($context) # restore cotnext->R13
1467 mov %r14,232($context) # restore cotnext->R14
1468 mov %r15,240($context) # restore cotnext->R15
1470 lea -56-10*16(%rax),%rsi
1471 lea 512($context),%rdi # &context.Xmm6
1473 .long 0xa548f3fc # cld; rep movsq
1476 .size avx2_handler,.-avx2_handler
1481 .rva .LSEH_begin_sha1_multi_block
1482 .rva .LSEH_end_sha1_multi_block
1483 .rva .LSEH_info_sha1_multi_block
1484 .rva .LSEH_begin_sha1_multi_block_shaext
1485 .rva .LSEH_end_sha1_multi_block_shaext
1486 .rva .LSEH_info_sha1_multi_block_shaext
1488 $code.=<<___ if ($avx);
1489 .rva .LSEH_begin_sha1_multi_block_avx
1490 .rva .LSEH_end_sha1_multi_block_avx
1491 .rva .LSEH_info_sha1_multi_block_avx
1493 $code.=<<___ if ($avx>1);
1494 .rva .LSEH_begin_sha1_multi_block_avx2
1495 .rva .LSEH_end_sha1_multi_block_avx2
1496 .rva .LSEH_info_sha1_multi_block_avx2
1501 .LSEH_info_sha1_multi_block:
1504 .rva .Lbody,.Lepilogue # HandlerData[]
1505 .LSEH_info_sha1_multi_block_shaext:
1508 .rva .Lbody_shaext,.Lepilogue_shaext # HandlerData[]
1510 $code.=<<___ if ($avx);
1511 .LSEH_info_sha1_multi_block_avx:
1514 .rva .Lbody_avx,.Lepilogue_avx # HandlerData[]
1516 $code.=<<___ if ($avx>1);
1517 .LSEH_info_sha1_multi_block_avx2:
1520 .rva .Lbody_avx2,.Lepilogue_avx2 # HandlerData[]
1523 ####################################################################
1526 local *opcode=shift;
1530 $rex|=0x04 if ($dst>=8);
1531 $rex|=0x01 if ($src>=8);
1532 unshift @opcode,$rex|0x40 if ($rex);
1536 if (@_[0] =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1537 my @opcode=(0x0f,0x3a,0xcc);
1538 rex(\@opcode,$3,$2);
1539 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
1541 push @opcode,$c=~/^0/?oct($c):$c;
1542 return ".byte\t".join(',',@opcode);
1544 return "sha1rnds4\t".@_[0];
1551 "sha1nexte" => 0xc8,
1553 "sha1msg2" => 0xca );
1555 if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
1556 my @opcode=(0x0f,0x38);
1557 rex(\@opcode,$2,$1);
1558 push @opcode,$opcodelet{$instr};
1559 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
1560 return ".byte\t".join(',',@opcode);
1562 return $instr."\t".@_[0];
1566 foreach (split("\n",$code)) {
1567 s/\`([^\`]*)\`/eval($1)/ge;
1569 s/\b(sha1rnds4)\s+(.*)/sha1rnds4($2)/geo or
1570 s/\b(sha1[^\s]*)\s+(.*)/sha1op38($1,$2)/geo or
1572 s/\b(vmov[dq])\b(.+)%ymm([0-9]+)/$1$2%xmm$3/go or
1573 s/\b(vmovdqu)\b(.+)%x%ymm([0-9]+)/$1$2%xmm$3/go or
1574 s/\b(vpinsr[qd])\b(.+)%ymm([0-9]+),%ymm([0-9]+)/$1$2%xmm$3,%xmm$4/go or
1575 s/\b(vpextr[qd])\b(.+)%ymm([0-9]+)/$1$2%xmm$3/go or
1576 s/\b(vinserti128)\b(\s+)%ymm/$1$2\$1,%xmm/go or
1577 s/\b(vpbroadcast[qd]\s+)%ymm([0-9]+)/$1%xmm$2/go;