3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
10 # sha1_block procedure for x86_64.
12 # It was brought to my attention that on EM64T compiler-generated code
13 # was far behind 32-bit assembler implementation. This is unlike on
14 # Opteron where compiler-generated code was only 15% behind 32-bit
15 # assembler, which originally made it hard to motivate the effort.
16 # There was suggestion to mechanically translate 32-bit code, but I
17 # dismissed it, reasoning that x86_64 offers enough register bank
18 # capacity to fully utilize SHA-1 parallelism. Therefore this fresh
19 # implementation:-) However! While 64-bit code does perform better
20 # on Opteron, I failed to beat 32-bit assembler on EM64T core. Well,
21 # x86_64 does offer larger *addressable* bank, but out-of-order core
22 # reaches for even more registers through dynamic aliasing, and EM64T
23 # core must have managed to run-time optimize even 32-bit code just as
24 # good as 64-bit one. Performance improvement is summarized in the
27 # gcc 3.4 32-bit asm cycles/byte
28 # Opteron +45% +20% 6.8
29 # Xeon P4 +65% +0% 9.9
34 # The code was revised to minimize code size and to maximize
35 # "distance" between instructions producing input to 'lea'
36 # instruction and the 'lea' instruction itself, which is essential
37 # for Intel Atom core.
41 # Add SSSE3, Supplemental[!] SSE3, implementation. The idea behind it
42 # is to offload message schedule denoted by Wt in NIST specification,
43 # or Xupdate in OpenSSL source, to SIMD unit. See sha1-586.pl module
44 # for background and implementation details. The only difference from
45 # 32-bit code is that 64-bit code doesn't have to spill @X[] elements
46 # to free temporary registers.
50 # Add AVX code path. See sha1-586.pl for further information.
54 # Add AVX2+BMI code path. Initial attempt (utilizing BMI instructions
55 # and loading pair of consecutive blocks to 256-bit %ymm registers)
56 # did not provide impressive performance improvement till a crucial
57 # hint regarding the number of Xupdate iterations to pre-compute in
58 # advance was provided by Ilya Albrekht of Intel Corp.
60 ######################################################################
61 # Current performance is summarized in following table. Numbers are
62 # CPU clock cycles spent to process single byte (less is better).
67 # Core2 6.70 6.05/+11% -
68 # Westmere 7.08 5.44/+30% -
69 # Sandy Bridge 7.93 6.16/+28% 4.99/+59%
70 # Ivy Bridge 6.30 4.63/+36% 4.60/+37%
71 # Haswell 5.98 4.36/+37% 3.57/+67%
72 # Bulldozer 10.9 5.95/+82%
73 # VIA Nano 10.2 7.46/+37%
78 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
80 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
82 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
83 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
84 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
85 die "can't locate x86_64-xlate.pl";
87 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
88 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
89 $avx = ($1>=2.19) + ($1>=2.22);
92 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
93 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
94 $avx = ($1>=2.09) + ($1>=2.10);
97 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
98 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
99 $avx = ($1>=10) + ($1>=11);
102 open OUT,"| \"$^X\" $xlate $flavour $output";
105 $ctx="%rdi"; # 1st arg
106 $inp="%rsi"; # 2nd arg
107 $num="%rdx"; # 3rd arg
109 # reassign arguments in order to produce more compact code
127 my ($i,$a,$b,$c,$d,$e)=@_;
129 $code.=<<___ if ($i==0);
130 mov `4*$i`($inp),$xi[0]
132 mov $xi[0],`4*$i`(%rsp)
134 $code.=<<___ if ($i<15);
136 mov `4*$j`($inp),$xi[1]
141 lea 0x5a827999($xi[0],$e),$e
143 mov $xi[1],`4*$j`(%rsp)
149 $code.=<<___ if ($i>=15);
150 mov `4*($j%16)`(%rsp),$xi[1]
153 xor `4*(($j+2)%16)`(%rsp),$xi[1]
156 xor `4*(($j+8)%16)`(%rsp),$xi[1]
158 lea 0x5a827999($xi[0],$e),$e
159 xor `4*(($j+13)%16)`(%rsp),$xi[1]
164 mov $xi[1],`4*($j%16)`(%rsp)
167 unshift(@xi,pop(@xi));
171 my ($i,$a,$b,$c,$d,$e)=@_;
173 my $K=($i<40)?0x6ed9eba1:0xca62c1d6;
174 $code.=<<___ if ($i<79);
175 mov `4*($j%16)`(%rsp),$xi[1]
178 xor `4*(($j+2)%16)`(%rsp),$xi[1]
182 xor `4*(($j+8)%16)`(%rsp),$xi[1]
185 xor `4*(($j+13)%16)`(%rsp),$xi[1]
190 $code.=<<___ if ($i<76);
191 mov $xi[1],`4*($j%16)`(%rsp)
193 $code.=<<___ if ($i==79);
204 unshift(@xi,pop(@xi));
208 my ($i,$a,$b,$c,$d,$e)=@_;
211 mov `4*($j%16)`(%rsp),$xi[1]
214 xor `4*(($j+2)%16)`(%rsp),$xi[1]
217 xor `4*(($j+8)%16)`(%rsp),$xi[1]
219 lea 0x8f1bbcdc($xi[0],$e),$e
221 xor `4*(($j+13)%16)`(%rsp),$xi[1]
227 mov $xi[1],`4*($j%16)`(%rsp)
230 unshift(@xi,pop(@xi));
235 .extern OPENSSL_ia32cap_P
237 .globl sha1_block_data_order
238 .type sha1_block_data_order,\@function,3
240 sha1_block_data_order:
241 mov OPENSSL_ia32cap_P+0(%rip),%r9d
242 mov OPENSSL_ia32cap_P+4(%rip),%r8d
243 mov OPENSSL_ia32cap_P+8(%rip),%r10d
244 test \$`1<<9`,%r8d # check SSSE3 bit
247 $code.=<<___ if ($avx>1);
248 and \$`1<<3|1<<5|1<<8`,%r10d # check AVX2+BMI1+BMI2
249 cmp \$`1<<3|1<<5|1<<8`,%r10d
252 $code.=<<___ if ($avx);
253 and \$`1<<28`,%r8d # mask AVX bit
254 and \$`1<<30`,%r9d # mask "Intel CPU" bit
256 cmp \$`1<<28|1<<30`,%r8d
269 mov %rdi,$ctx # reassigned argument
271 mov %rsi,$inp # reassigned argument
273 mov %rdx,$num # reassigned argument
274 mov %r11,`16*4`(%rsp)
287 for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
288 for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
289 for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
290 for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
304 lea `16*4`($inp),$inp
307 mov `16*4`(%rsp),%rsi
315 .size sha1_block_data_order,.-sha1_block_data_order
319 my @X=map("%xmm$_",(4..7,0..3));
320 my @Tx=map("%xmm$_",(8..10));
322 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
323 my @T=("%esi","%edi");
328 my $_rol=sub { &rol(@_) };
329 my $_ror=sub { &ror(@_) };
335 jmp .Lalign32_$sn # see "Decoded ICache" in manual
343 .type sha1_block_data_order_ssse3,\@function,3
345 sha1_block_data_order_ssse3:
350 lea `-64-($win64?6*16:0)`(%rsp),%rsp
352 $code.=<<___ if ($win64);
353 movaps %xmm6,64+0(%rsp)
354 movaps %xmm7,64+16(%rsp)
355 movaps %xmm8,64+32(%rsp)
356 movaps %xmm9,64+48(%rsp)
357 movaps %xmm10,64+64(%rsp)
358 movaps %xmm11,64+80(%rsp)
362 mov %rdi,$ctx # reassigned argument
363 mov %rsi,$inp # reassigned argument
364 mov %rdx,$num # reassigned argument
368 lea K_XX_XX+64(%rip),$K_XX_XX
370 mov 0($ctx),$A # load context
374 mov $B,@T[0] # magic seed
380 movdqa 64($K_XX_XX),@X[2] # pbswap mask
381 movdqa -64($K_XX_XX),@Tx[1] # K_00_19
382 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
383 movdqu 16($inp),@X[-3&7]
384 movdqu 32($inp),@X[-2&7]
385 movdqu 48($inp),@X[-1&7]
386 pshufb @X[2],@X[-4&7] # byte swap
388 pshufb @X[2],@X[-3&7]
389 pshufb @X[2],@X[-2&7]
390 pshufb @X[2],@X[-1&7]
391 paddd @Tx[1],@X[-4&7] # add K_00_19
392 paddd @Tx[1],@X[-3&7]
393 paddd @Tx[1],@X[-2&7]
394 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU
395 psubd @Tx[1],@X[-4&7] # restore X[]
396 movdqa @X[-3&7],16(%rsp)
397 psubd @Tx[1],@X[-3&7]
398 movdqa @X[-2&7],32(%rsp)
399 psubd @Tx[1],@X[-2&7]
403 sub AUTOLOAD() # thunk [simplified] 32-bit style perlasm
404 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
406 $arg = "\$$arg" if ($arg*1 eq $arg);
407 $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
410 sub Xupdate_ssse3_16_31() # recall that $Xi starts wtih 4
413 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
416 &movdqa (@X[0],@X[-3&7]);
419 &movdqa (@Tx[0],@X[-1&7]);
420 &palignr(@X[0],@X[-4&7],8); # compose "X[-14]" in "X[0]"
424 &paddd (@Tx[1],@X[-1&7]);
427 &psrldq (@Tx[0],4); # "X[-3]", 3 dwords
430 &pxor (@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
434 &pxor (@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
440 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
443 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
447 &movdqa (@Tx[2],@X[0]);
448 &movdqa (@Tx[0],@X[0]);
454 &pslldq (@Tx[2],12); # "X[0]"<<96, extract one dword
455 &paddd (@X[0],@X[0]);
464 &movdqa (@Tx[1],@Tx[2]);
469 &por (@X[0],@Tx[0]); # "X[0]"<<<=1
476 &pxor (@X[0],@Tx[2]);
479 &movdqa (@Tx[2],eval(2*16*(($Xi)/5)-64)."($K_XX_XX)"); # K_XX_XX
483 &pxor (@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2
485 foreach (@insns) { eval; } # remaining instructions [if any]
487 $Xi++; push(@X,shift(@X)); # "rotate" X[]
488 push(@Tx,shift(@Tx));
491 sub Xupdate_ssse3_32_79()
494 my @insns = (&$body,&$body,&$body,&$body); # 32 to 44 instructions
497 &movdqa (@Tx[0],@X[-1&7]) if ($Xi==8);
498 eval(shift(@insns)); # body_20_39
499 &pxor (@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
500 &palignr(@Tx[0],@X[-2&7],8); # compose "X[-6]"
503 eval(shift(@insns)); # rol
505 &pxor (@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
507 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
509 &movdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
510 } else { # ... or load next one
511 &movdqa (@Tx[2],eval(2*16*($Xi/5)-64)."($K_XX_XX)");
513 &paddd (@Tx[1],@X[-1&7]);
514 eval(shift(@insns)); # ror
517 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-6]"
518 eval(shift(@insns)); # body_20_39
521 eval(shift(@insns)); # rol
523 &movdqa (@Tx[0],@X[0]);
524 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
527 eval(shift(@insns)); # ror
531 eval(shift(@insns)); # body_20_39
535 eval(shift(@insns)); # rol
538 eval(shift(@insns)); # ror
541 &por (@X[0],@Tx[0]); # "X[0]"<<<=2
542 eval(shift(@insns)); # body_20_39
544 &movdqa (@Tx[1],@X[0]) if ($Xi<19);
546 eval(shift(@insns)); # rol
549 eval(shift(@insns)); # rol
552 foreach (@insns) { eval; } # remaining instructions
554 $Xi++; push(@X,shift(@X)); # "rotate" X[]
555 push(@Tx,shift(@Tx));
558 sub Xuplast_ssse3_80()
561 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
565 &paddd (@Tx[1],@X[-1&7]);
571 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
573 foreach (@insns) { eval; } # remaining instructions
576 &je (".Ldone_ssse3");
578 unshift(@Tx,pop(@Tx));
580 &movdqa (@X[2],"64($K_XX_XX)"); # pbswap mask
581 &movdqa (@Tx[1],"-64($K_XX_XX)"); # K_00_19
582 &movdqu (@X[-4&7],"0($inp)"); # load input
583 &movdqu (@X[-3&7],"16($inp)");
584 &movdqu (@X[-2&7],"32($inp)");
585 &movdqu (@X[-1&7],"48($inp)");
586 &pshufb (@X[-4&7],@X[2]); # byte swap
595 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
600 &pshufb (@X[($Xi-3)&7],@X[2]);
604 &paddd (@X[($Xi-4)&7],@Tx[1]);
609 &movdqa (eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]); # X[]+K xfer to IALU
612 &psubd (@X[($Xi-4)&7],@Tx[1]);
614 foreach (@insns) { eval; }
621 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
624 foreach (@insns) { eval; }
627 sub body_00_19 () { # ((c^d)&b)^d
628 # on start @T[0]=(c^d)&b
629 return &body_20_39() if ($rx==19); $rx++;
631 '($a,$b,$c,$d,$e)=@V;'.
632 '&$_ror ($b,$j?7:2)', # $b>>>2
634 '&mov (@T[1],$a)', # $b for next round
636 '&add ($e,eval(4*($j&15))."(%rsp)")', # X[]+K xfer
637 '&xor ($b,$c)', # $c^$d for next round
641 '&and (@T[1],$b)', # ($b&($c^$d)) for next round
643 '&xor ($b,$c)', # restore $b
644 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
648 sub body_20_39 () { # b^d^c
650 return &body_40_59() if ($rx==39); $rx++;
652 '($a,$b,$c,$d,$e)=@V;'.
653 '&add ($e,eval(4*($j&15))."(%rsp)")', # X[]+K xfer
654 '&xor (@T[0],$d) if($j==19);'.
655 '&xor (@T[0],$c) if($j> 19)', # ($b^$d^$c)
656 '&mov (@T[1],$a)', # $b for next round
660 '&xor (@T[1],$c) if ($j< 79)', # $b^$d for next round
662 '&$_ror ($b,7)', # $b>>>2
663 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
667 sub body_40_59 () { # ((b^c)&(c^d))^c
668 # on entry @T[0]=(b^c), (c^=d)
671 '($a,$b,$c,$d,$e)=@V;'.
672 '&add ($e,eval(4*($j&15))."(%rsp)")', # X[]+K xfer
673 '&and (@T[0],$c) if ($j>=40)', # (b^c)&(c^d)
674 '&xor ($c,$d) if ($j>=40)', # restore $c
676 '&$_ror ($b,7)', # $b>>>2
677 '&mov (@T[1],$a)', # $b for next round
682 '&xor (@T[1],$c) if ($j==59);'.
683 '&xor (@T[1],$b) if ($j< 59)', # b^c for next round
685 '&xor ($b,$c) if ($j< 59)', # c^d for next round
686 '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
693 &Xupdate_ssse3_16_31(\&body_00_19);
694 &Xupdate_ssse3_16_31(\&body_00_19);
695 &Xupdate_ssse3_16_31(\&body_00_19);
696 &Xupdate_ssse3_16_31(\&body_00_19);
697 &Xupdate_ssse3_32_79(\&body_00_19);
698 &Xupdate_ssse3_32_79(\&body_20_39);
699 &Xupdate_ssse3_32_79(\&body_20_39);
700 &Xupdate_ssse3_32_79(\&body_20_39);
701 &Xupdate_ssse3_32_79(\&body_20_39);
702 &Xupdate_ssse3_32_79(\&body_20_39);
703 &Xupdate_ssse3_32_79(\&body_40_59);
704 &Xupdate_ssse3_32_79(\&body_40_59);
705 &Xupdate_ssse3_32_79(\&body_40_59);
706 &Xupdate_ssse3_32_79(\&body_40_59);
707 &Xupdate_ssse3_32_79(\&body_40_59);
708 &Xupdate_ssse3_32_79(\&body_20_39);
709 &Xuplast_ssse3_80(\&body_20_39); # can jump to "done"
711 $saved_j=$j; @saved_V=@V;
713 &Xloop_ssse3(\&body_20_39);
714 &Xloop_ssse3(\&body_20_39);
715 &Xloop_ssse3(\&body_20_39);
718 add 0($ctx),$A # update context
725 mov @T[0],$B # magic seed
737 $j=$saved_j; @V=@saved_V;
739 &Xtail_ssse3(\&body_20_39);
740 &Xtail_ssse3(\&body_20_39);
741 &Xtail_ssse3(\&body_20_39);
744 add 0($ctx),$A # update context
755 $code.=<<___ if ($win64);
756 movaps 64+0(%rsp),%xmm6
757 movaps 64+16(%rsp),%xmm7
758 movaps 64+32(%rsp),%xmm8
759 movaps 64+48(%rsp),%xmm9
760 movaps 64+64(%rsp),%xmm10
761 movaps 64+80(%rsp),%xmm11
764 lea `64+($win64?6*16:0)`(%rsp),%rsi
771 .size sha1_block_data_order_ssse3,.-sha1_block_data_order_ssse3
775 $Xi=4; # reset variables
776 @X=map("%xmm$_",(4..7,0..3));
777 @Tx=map("%xmm$_",(8..10));
781 my $done_avx_label=".Ldone_avx";
783 my $_rol=sub { &shld(@_[0],@_) };
784 my $_ror=sub { &shrd(@_[0],@_) };
787 .type sha1_block_data_order_avx,\@function,3
789 sha1_block_data_order_avx:
794 lea `-64-($win64?6*16:0)`(%rsp),%rsp
796 $code.=<<___ if ($win64);
797 movaps %xmm6,64+0(%rsp)
798 movaps %xmm7,64+16(%rsp)
799 movaps %xmm8,64+32(%rsp)
800 movaps %xmm9,64+48(%rsp)
801 movaps %xmm10,64+64(%rsp)
802 movaps %xmm11,64+80(%rsp)
806 mov %rdi,$ctx # reassigned argument
807 mov %rsi,$inp # reassigned argument
808 mov %rdx,$num # reassigned argument
813 lea K_XX_XX+64(%rip),$K_XX_XX
815 mov 0($ctx),$A # load context
819 mov $B,@T[0] # magic seed
825 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
826 vmovdqa -64($K_XX_XX),$Kx # K_00_19
827 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
828 vmovdqu 16($inp),@X[-3&7]
829 vmovdqu 32($inp),@X[-2&7]
830 vmovdqu 48($inp),@X[-1&7]
831 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap
833 vpshufb @X[2],@X[-3&7],@X[-3&7]
834 vpshufb @X[2],@X[-2&7],@X[-2&7]
835 vpshufb @X[2],@X[-1&7],@X[-1&7]
836 vpaddd $Kx,@X[-4&7],@X[0] # add K_00_19
837 vpaddd $Kx,@X[-3&7],@X[1]
838 vpaddd $Kx,@X[-2&7],@X[2]
839 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
840 vmovdqa @X[1],16(%rsp)
841 vmovdqa @X[2],32(%rsp)
845 sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
848 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
853 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
857 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
860 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
863 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
867 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
873 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
876 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
880 &vpsrld (@Tx[0],@X[0],31);
886 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword
887 &vpaddd (@X[0],@X[0],@X[0]);
893 &vpsrld (@Tx[1],@Tx[2],30);
894 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
900 &vpslld (@Tx[2],@Tx[2],2);
901 &vpxor (@X[0],@X[0],@Tx[1]);
907 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
910 &vmovdqa ($Kx,eval(2*16*(($Xi)/5)-64)."($K_XX_XX)") if ($Xi%5==0); # K_XX_XX
915 foreach (@insns) { eval; } # remaining instructions [if any]
917 $Xi++; push(@X,shift(@X)); # "rotate" X[]
920 sub Xupdate_avx_32_79()
923 my @insns = (&$body,&$body,&$body,&$body); # 32 to 44 instructions
926 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]"
927 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
928 eval(shift(@insns)); # body_20_39
931 eval(shift(@insns)); # rol
933 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
935 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
936 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
937 &vmovdqa ($Kx,eval(2*16*($Xi/5)-64)."($K_XX_XX)") if ($Xi%5==0);
938 eval(shift(@insns)); # ror
941 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]"
942 eval(shift(@insns)); # body_20_39
945 eval(shift(@insns)); # rol
947 &vpsrld (@Tx[0],@X[0],30);
948 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
951 eval(shift(@insns)); # ror
954 &vpslld (@X[0],@X[0],2);
955 eval(shift(@insns)); # body_20_39
958 eval(shift(@insns)); # rol
961 eval(shift(@insns)); # ror
964 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
965 eval(shift(@insns)); # body_20_39
968 eval(shift(@insns)); # rol
971 eval(shift(@insns)); # rol
974 foreach (@insns) { eval; } # remaining instructions
976 $Xi++; push(@X,shift(@X)); # "rotate" X[]
982 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
986 &vpaddd (@Tx[1],$Kx,@X[-1&7]);
992 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
994 foreach (@insns) { eval; } # remaining instructions
997 &je ($done_avx_label);
999 &vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask
1000 &vmovdqa($Kx,"-64($K_XX_XX)"); # K_00_19
1001 &vmovdqu(@X[-4&7],"0($inp)"); # load input
1002 &vmovdqu(@X[-3&7],"16($inp)");
1003 &vmovdqu(@X[-2&7],"32($inp)");
1004 &vmovdqu(@X[-1&7],"48($inp)");
1005 &vpshufb(@X[-4&7],@X[-4&7],@X[2]); # byte swap
1014 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
1015 my ($a,$b,$c,$d,$e);
1017 eval(shift(@insns));
1018 eval(shift(@insns));
1019 &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]);
1020 eval(shift(@insns));
1021 eval(shift(@insns));
1022 &vpaddd (@X[$Xi&7],@X[($Xi-4)&7],$Kx);
1023 eval(shift(@insns));
1024 eval(shift(@insns));
1025 eval(shift(@insns));
1026 eval(shift(@insns));
1027 &vmovdqa(eval(16*$Xi)."(%rsp)",@X[$Xi&7]); # X[]+K xfer to IALU
1028 eval(shift(@insns));
1029 eval(shift(@insns));
1031 foreach (@insns) { eval; }
1038 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
1039 my ($a,$b,$c,$d,$e);
1041 foreach (@insns) { eval; }
1048 &Xupdate_avx_16_31(\&body_00_19);
1049 &Xupdate_avx_16_31(\&body_00_19);
1050 &Xupdate_avx_16_31(\&body_00_19);
1051 &Xupdate_avx_16_31(\&body_00_19);
1052 &Xupdate_avx_32_79(\&body_00_19);
1053 &Xupdate_avx_32_79(\&body_20_39);
1054 &Xupdate_avx_32_79(\&body_20_39);
1055 &Xupdate_avx_32_79(\&body_20_39);
1056 &Xupdate_avx_32_79(\&body_20_39);
1057 &Xupdate_avx_32_79(\&body_20_39);
1058 &Xupdate_avx_32_79(\&body_40_59);
1059 &Xupdate_avx_32_79(\&body_40_59);
1060 &Xupdate_avx_32_79(\&body_40_59);
1061 &Xupdate_avx_32_79(\&body_40_59);
1062 &Xupdate_avx_32_79(\&body_40_59);
1063 &Xupdate_avx_32_79(\&body_20_39);
1064 &Xuplast_avx_80(\&body_20_39); # can jump to "done"
1066 $saved_j=$j; @saved_V=@V;
1068 &Xloop_avx(\&body_20_39);
1069 &Xloop_avx(\&body_20_39);
1070 &Xloop_avx(\&body_20_39);
1073 add 0($ctx),$A # update context
1080 mov @T[0],$B # magic seed
1092 $j=$saved_j; @V=@saved_V;
1094 &Xtail_avx(\&body_20_39);
1095 &Xtail_avx(\&body_20_39);
1096 &Xtail_avx(\&body_20_39);
1101 add 0($ctx),$A # update context
1112 $code.=<<___ if ($win64);
1113 movaps 64+0(%rsp),%xmm6
1114 movaps 64+16(%rsp),%xmm7
1115 movaps 64+32(%rsp),%xmm8
1116 movaps 64+48(%rsp),%xmm9
1117 movaps 64+64(%rsp),%xmm10
1118 movaps 64+80(%rsp),%xmm11
1121 lea `64+($win64?6*16:0)`(%rsp),%rsi
1128 .size sha1_block_data_order_avx,.-sha1_block_data_order_avx
1133 $Xi=4; # reset variables
1134 @X=map("%ymm$_",(4..7,0..3));
1135 @Tx=map("%ymm$_",(8..10));
1139 my @ROTX=("%eax","%ebp","%ebx","%ecx","%edx","%esi");
1140 my ($a5,$t0)=("%r12d","%edi");
1142 my ($A,$F,$B,$C,$D,$E)=@ROTX;
1147 .type sha1_block_data_order_avx2,\@function,3
1149 sha1_block_data_order_avx2:
1158 $code.=<<___ if ($win64);
1159 lea -6*16(%rsp),%rsp
1160 movaps %xmm6,-6*16(%r14)
1161 movaps %xmm7,-5*16(%r14)
1162 movaps %xmm8,-4*16(%r14)
1163 movaps %xmm9,-3*16(%r14)
1164 movaps %xmm10,-2*16(%r14)
1165 movaps %xmm11,-1*16(%r14)
1169 mov %rdi,$ctx # reassigned argument
1170 mov %rsi,$inp # reassigned argument
1171 mov %rdx,$num # reassigned argument
1179 lea K_XX_XX+64(%rip),$K_XX_XX
1181 mov 0($ctx),$A # load context
1183 cmovae $inp,$frame # next or same block
1188 vmovdqu 64($K_XX_XX),@X[2] # pbswap mask
1190 vmovdqu ($inp),%xmm0
1191 vmovdqu 16($inp),%xmm1
1192 vmovdqu 32($inp),%xmm2
1193 vmovdqu 48($inp),%xmm3
1195 vinserti128 \$1,($frame),@X[-4&7],@X[-4&7]
1196 vinserti128 \$1,16($frame),@X[-3&7],@X[-3&7]
1197 vpshufb @X[2],@X[-4&7],@X[-4&7]
1198 vinserti128 \$1,32($frame),@X[-2&7],@X[-2&7]
1199 vpshufb @X[2],@X[-3&7],@X[-3&7]
1200 vinserti128 \$1,48($frame),@X[-1&7],@X[-1&7]
1201 vpshufb @X[2],@X[-2&7],@X[-2&7]
1202 vmovdqu -64($K_XX_XX),$Kx # K_00_19
1203 vpshufb @X[2],@X[-1&7],@X[-1&7]
1205 vpaddd $Kx,@X[-4&7],@X[0] # add K_00_19
1206 vpaddd $Kx,@X[-3&7],@X[1]
1207 vmovdqu @X[0],0(%rsp) # X[]+K xfer to IALU
1208 vpaddd $Kx,@X[-2&7],@X[2]
1209 vmovdqu @X[1],32(%rsp)
1210 vpaddd $Kx,@X[-1&7],@X[3]
1211 vmovdqu @X[2],64(%rsp)
1212 vmovdqu @X[3],96(%rsp)
1214 for (;$Xi<8;$Xi++) { # Xupdate_avx2_16_31
1217 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
1218 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
1219 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
1220 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
1221 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
1222 &vpsrld (@Tx[0],@X[0],31);
1223 &vmovdqu($Kx,eval(2*16*(($Xi)/5)-64)."($K_XX_XX)") if ($Xi%5==0); # K_XX_XX
1224 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword
1225 &vpaddd (@X[0],@X[0],@X[0]);
1226 &vpsrld (@Tx[1],@Tx[2],30);
1227 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
1228 &vpslld (@Tx[2],@Tx[2],2);
1229 &vpxor (@X[0],@X[0],@Tx[1]);
1230 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
1231 &vpaddd (@Tx[1],@X[0],$Kx);
1232 &vmovdqu("32*$Xi(%rsp)",@Tx[1]); # X[]+K xfer to IALU
1234 push(@X,shift(@X)); # "rotate" X[]
1237 lea 128(%rsp),$frame
1246 sub bodyx_00_19 () { # 8 instructions, 3 cycles critical path
1247 # at start $f=(b&c)^(~b&d), $b>>>=2
1248 return &bodyx_20_39() if ($rx==19); $rx++;
1250 '($a,$f,$b,$c,$d,$e)=@ROTX;'.
1252 '&add ($e,((32*($j/4)+4*($j%4))%256-128)."($frame)");'. # e+=X[i]+K
1253 '&lea ($frame,"256($frame)") if ($j%32==31);',
1254 '&andn ($t0,$a,$c)', # ~b&d for next round
1256 '&add ($e,$f)', # e+=(b&c)^(~b&d)
1257 '&rorx ($a5,$a,27)', # a<<<5
1258 '&rorx ($f,$a,2)', # b>>>2 for next round
1259 '&and ($a,$b)', # b&c for next round
1261 '&add ($e,$a5)', # e+=a<<<5
1262 '&xor ($a,$t0);'. # f=(b&c)^(~b&d) for next round
1264 'unshift(@ROTX,pop(@ROTX)); $j++;'
1268 sub bodyx_20_39 () { # 7 instructions, 2 cycles critical path
1269 # on entry $f=b^c^d, $b>>>=2
1270 return &bodyx_40_59() if ($rx==39); $rx++;
1272 '($a,$f,$b,$c,$d,$e)=@ROTX;'.
1274 '&add ($e,((32*($j/4)+4*($j%4))%256-128)."($frame)");'. # e+=X[i]+K
1275 '&lea ($frame,"256($frame)") if ($j%32==31);',
1277 '&lea ($e,"($e,$f)")', # e+=b^c^d
1278 '&rorx ($a5,$a,27)', # a<<<5
1279 '&rorx ($f,$a,2) if ($j<79)', # b>>>2 in next round
1280 '&xor ($a,$b) if ($j<79)', # b^c for next round
1282 '&add ($e,$a5)', # e+=a<<<5
1283 '&xor ($a,$c) if ($j<79);'. # f=b^c^d for next round
1285 'unshift(@ROTX,pop(@ROTX)); $j++;'
1289 sub bodyx_40_59 () { # 10 instructions, 3 cycles critical path
1290 # on entry $f=((b^c)&(c^d)), $b>>>=2
1293 '($a,$f,$b,$c,$d,$e)=@ROTX;'.
1295 '&add ($e,((32*($j/4)+4*($j%4))%256-128)."($frame)");'. # e+=X[i]+K
1296 '&lea ($frame,"256($frame)") if ($j%32==31);',
1297 '&xor ($f,$c) if ($j>39)', # (b^c)&(c^d)^c
1298 '&mov ($t0,$b) if ($j<59)', # count on zero latency
1299 '&xor ($t0,$c) if ($j<59)', # c^d for next round
1301 '&lea ($e,"($e,$f)")', # e+=(b^c)&(c^d)^c
1302 '&rorx ($a5,$a,27)', # a<<<5
1303 '&rorx ($f,$a,2)', # b>>>2 in next round
1304 '&xor ($a,$b)', # b^c for next round
1306 '&add ($e,$a5)', # e+=a<<<5
1307 '&and ($a,$t0) if ($j< 59);'. # f=(b^c)&(c^d) for next round
1308 '&xor ($a,$c) if ($j==59);'. # f=b^c^d for next round
1310 'unshift(@ROTX,pop(@ROTX)); $j++;'
1314 sub Xupdate_avx2_16_31() # recall that $Xi starts wtih 4
1317 my @insns = (&$body,&$body,&$body,&$body,&$body); # 35 instructions
1318 my ($a,$b,$c,$d,$e);
1320 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
1321 eval(shift(@insns));
1322 eval(shift(@insns));
1323 eval(shift(@insns));
1324 eval(shift(@insns));
1326 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
1327 eval(shift(@insns));
1328 eval(shift(@insns));
1329 eval(shift(@insns));
1331 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
1332 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
1333 eval(shift(@insns));
1334 eval(shift(@insns));
1336 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
1337 eval(shift(@insns));
1338 eval(shift(@insns));
1339 eval(shift(@insns));
1340 eval(shift(@insns));
1342 &vpsrld (@Tx[0],@X[0],31);
1343 &vmovdqu($Kx,eval(2*16*(($Xi)/5)-64)."($K_XX_XX)") if ($Xi%5==0); # K_XX_XX
1344 eval(shift(@insns));
1345 eval(shift(@insns));
1346 eval(shift(@insns));
1348 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword
1349 &vpaddd (@X[0],@X[0],@X[0]);
1350 eval(shift(@insns));
1351 eval(shift(@insns));
1353 &vpsrld (@Tx[1],@Tx[2],30);
1354 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
1355 eval(shift(@insns));
1356 eval(shift(@insns));
1358 &vpslld (@Tx[2],@Tx[2],2);
1359 &vpxor (@X[0],@X[0],@Tx[1]);
1360 eval(shift(@insns));
1361 eval(shift(@insns));
1363 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
1364 eval(shift(@insns));
1365 eval(shift(@insns));
1366 eval(shift(@insns));
1368 &vpaddd (@Tx[1],@X[0],$Kx);
1369 eval(shift(@insns));
1370 eval(shift(@insns));
1371 eval(shift(@insns));
1372 &vmovdqu(eval(32*($Xi))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
1374 foreach (@insns) { eval; } # remaining instructions [if any]
1377 push(@X,shift(@X)); # "rotate" X[]
1380 sub Xupdate_avx2_32_79()
1383 my @insns = (&$body,&$body,&$body,&$body,&$body); # 35 to 50 instructions
1384 my ($a,$b,$c,$d,$e);
1386 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]"
1387 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
1388 eval(shift(@insns));
1389 eval(shift(@insns));
1391 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
1392 &vmovdqu($Kx,eval(2*16*($Xi/5)-64)."($K_XX_XX)") if ($Xi%5==0);
1393 eval(shift(@insns));
1394 eval(shift(@insns));
1395 eval(shift(@insns));
1397 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]"
1398 eval(shift(@insns));
1399 eval(shift(@insns));
1400 eval(shift(@insns));
1402 &vpsrld (@Tx[0],@X[0],30);
1403 &vpslld (@X[0],@X[0],2);
1404 eval(shift(@insns));
1405 eval(shift(@insns));
1406 eval(shift(@insns));
1408 #&vpslld (@X[0],@X[0],2);
1409 eval(shift(@insns));
1410 eval(shift(@insns));
1411 eval(shift(@insns));
1413 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
1414 eval(shift(@insns));
1415 eval(shift(@insns));
1416 eval(shift(@insns));
1417 eval(shift(@insns));
1419 &vpaddd (@Tx[1],@X[0],$Kx);
1420 eval(shift(@insns));
1421 eval(shift(@insns));
1422 eval(shift(@insns));
1423 eval(shift(@insns));
1425 &vmovdqu("32*$Xi(%rsp)",@Tx[1]); # X[]+K xfer to IALU
1427 foreach (@insns) { eval; } # remaining instructions
1430 push(@X,shift(@X)); # "rotate" X[]
1436 my @insns = (&$body,&$body,&$body,&$body,&$body); # 32 instructions
1437 my ($a,$b,$c,$d,$e);
1439 foreach (@insns) { eval; }
1443 &Xupdate_avx2_32_79(\&bodyx_00_19);
1444 &Xupdate_avx2_32_79(\&bodyx_00_19);
1445 &Xupdate_avx2_32_79(\&bodyx_00_19);
1446 &Xupdate_avx2_32_79(\&bodyx_00_19);
1448 &Xupdate_avx2_32_79(\&bodyx_20_39);
1449 &Xupdate_avx2_32_79(\&bodyx_20_39);
1450 &Xupdate_avx2_32_79(\&bodyx_20_39);
1451 &Xupdate_avx2_32_79(\&bodyx_20_39);
1454 &Xupdate_avx2_32_79(\&bodyx_40_59);
1455 &Xupdate_avx2_32_79(\&bodyx_40_59);
1456 &Xupdate_avx2_32_79(\&bodyx_40_59);
1457 &Xupdate_avx2_32_79(\&bodyx_40_59);
1459 &Xloop_avx2(\&bodyx_20_39);
1460 &Xloop_avx2(\&bodyx_20_39);
1461 &Xloop_avx2(\&bodyx_20_39);
1462 &Xloop_avx2(\&bodyx_20_39);
1465 lea 128($inp),$frame
1466 lea 128($inp),%rdi # borrow $t0
1468 cmovae $inp,$frame # next or previous block
1470 # output is d-e-[a]-f-b-c => A=d,F=e,C=f,D=b,E=c
1471 add 0($ctx),@ROTX[0] # update context
1472 add 4($ctx),@ROTX[1]
1473 add 8($ctx),@ROTX[3]
1474 mov @ROTX[0],0($ctx)
1475 add 12($ctx),@ROTX[4]
1476 mov @ROTX[1],4($ctx)
1477 mov @ROTX[0],$A # A=d
1478 add 16($ctx),@ROTX[5]
1480 mov @ROTX[3],8($ctx)
1481 mov @ROTX[4],$D # D=b
1482 #xchg @ROTX[5],$F # F=c, C=f
1483 mov @ROTX[4],12($ctx)
1484 mov @ROTX[1],$F # F=e
1485 mov @ROTX[5],16($ctx)
1487 mov @ROTX[5],$E # E=c
1489 #xchg $F,$E # E=c, F=e
1495 $Xi=4; # reset variables
1496 @X=map("%ymm$_",(4..7,0..3));
1499 vmovdqu 64($K_XX_XX),@X[2] # pbswap mask
1500 cmp $num,%rdi # borrowed $t0
1503 vmovdqu -64(%rdi),%xmm0 # low part of @X[-4&7]
1504 vmovdqu -48(%rdi),%xmm1
1505 vmovdqu -32(%rdi),%xmm2
1506 vmovdqu -16(%rdi),%xmm3
1507 vinserti128 \$1,0($frame),@X[-4&7],@X[-4&7]
1508 vinserti128 \$1,16($frame),@X[-3&7],@X[-3&7]
1509 vinserti128 \$1,32($frame),@X[-2&7],@X[-2&7]
1510 vinserti128 \$1,48($frame),@X[-1&7],@X[-1&7]
1515 lea 128+16(%rsp),$frame
1522 $rx=$j=0; @ROTX=($A,$F,$B,$C,$D,$E);
1524 &Xloop_avx2 (\&bodyx_00_19);
1525 &Xloop_avx2 (\&bodyx_00_19);
1526 &Xloop_avx2 (\&bodyx_00_19);
1527 &Xloop_avx2 (\&bodyx_00_19);
1529 &Xloop_avx2 (\&bodyx_20_39);
1530 &vmovdqu ($Kx,"-64($K_XX_XX)"); # K_00_19
1531 &vpshufb (@X[-4&7],@X[-4&7],@X[2]); # byte swap
1532 &Xloop_avx2 (\&bodyx_20_39);
1533 &vpshufb (@X[-3&7],@X[-3&7],@X[2]);
1534 &vpaddd (@Tx[0],@X[-4&7],$Kx); # add K_00_19
1535 &Xloop_avx2 (\&bodyx_20_39);
1536 &vmovdqu ("0(%rsp)",@Tx[0]);
1537 &vpshufb (@X[-2&7],@X[-2&7],@X[2]);
1538 &vpaddd (@Tx[1],@X[-3&7],$Kx);
1539 &Xloop_avx2 (\&bodyx_20_39);
1540 &vmovdqu ("32(%rsp)",@Tx[1]);
1541 &vpshufb (@X[-1&7],@X[-1&7],@X[2]);
1542 &vpaddd (@X[2],@X[-2&7],$Kx);
1544 &Xloop_avx2 (\&bodyx_40_59);
1546 &vmovdqu ("64(%rsp)",@X[2]);
1547 &vpaddd (@X[3],@X[-1&7],$Kx);
1548 &Xloop_avx2 (\&bodyx_40_59);
1549 &vmovdqu ("96(%rsp)",@X[3]);
1550 &Xloop_avx2 (\&bodyx_40_59);
1551 &Xupdate_avx2_16_31(\&bodyx_40_59);
1553 &Xupdate_avx2_16_31(\&bodyx_20_39);
1554 &Xupdate_avx2_16_31(\&bodyx_20_39);
1555 &Xupdate_avx2_16_31(\&bodyx_20_39);
1556 &Xloop_avx2 (\&bodyx_20_39);
1559 lea 128(%rsp),$frame
1561 # output is d-e-[a]-f-b-c => A=d,F=e,C=f,D=b,E=c
1562 add 0($ctx),@ROTX[0] # update context
1563 add 4($ctx),@ROTX[1]
1564 add 8($ctx),@ROTX[3]
1565 mov @ROTX[0],0($ctx)
1566 add 12($ctx),@ROTX[4]
1567 mov @ROTX[1],4($ctx)
1568 mov @ROTX[0],$A # A=d
1569 add 16($ctx),@ROTX[5]
1571 mov @ROTX[3],8($ctx)
1572 mov @ROTX[4],$D # D=b
1573 #xchg @ROTX[5],$F # F=c, C=f
1574 mov @ROTX[4],12($ctx)
1575 mov @ROTX[1],$F # F=e
1576 mov @ROTX[5],16($ctx)
1578 mov @ROTX[5],$E # E=c
1580 #xchg $F,$E # E=c, F=e
1588 $code.=<<___ if ($win64);
1589 movaps -6*16(%r14),%xmm6
1590 movaps -5*16(%r14),%xmm7
1591 movaps -4*16(%r14),%xmm8
1592 movaps -3*16(%r14),%xmm9
1593 movaps -2*16(%r14),%xmm10
1594 movaps -1*16(%r14),%xmm11
1606 .size sha1_block_data_order_avx2,.-sha1_block_data_order_avx2
1613 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1614 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1615 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1616 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1617 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1618 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1619 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1620 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1621 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask
1622 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask
1626 .asciz "SHA1 block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1630 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1631 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1639 .extern __imp_RtlVirtualUnwind
1640 .type se_handler,\@abi-omnipotent
1654 mov 120($context),%rax # pull context->Rax
1655 mov 248($context),%rbx # pull context->Rip
1657 lea .Lprologue(%rip),%r10
1658 cmp %r10,%rbx # context->Rip<.Lprologue
1659 jb .Lcommon_seh_tail
1661 mov 152($context),%rax # pull context->Rsp
1663 lea .Lepilogue(%rip),%r10
1664 cmp %r10,%rbx # context->Rip>=.Lepilogue
1665 jae .Lcommon_seh_tail
1667 mov `16*4`(%rax),%rax # pull saved stack pointer
1674 mov %rbx,144($context) # restore context->Rbx
1675 mov %rbp,160($context) # restore context->Rbp
1676 mov %r12,216($context) # restore context->R12
1677 mov %r13,224($context) # restore context->R13
1679 jmp .Lcommon_seh_tail
1680 .size se_handler,.-se_handler
1682 .type ssse3_handler,\@abi-omnipotent
1696 mov 120($context),%rax # pull context->Rax
1697 mov 248($context),%rbx # pull context->Rip
1699 mov 8($disp),%rsi # disp->ImageBase
1700 mov 56($disp),%r11 # disp->HandlerData
1702 mov 0(%r11),%r10d # HandlerData[0]
1703 lea (%rsi,%r10),%r10 # prologue label
1704 cmp %r10,%rbx # context->Rip<prologue label
1705 jb .Lcommon_seh_tail
1707 mov 152($context),%rax # pull context->Rsp
1709 mov 4(%r11),%r10d # HandlerData[1]
1710 lea (%rsi,%r10),%r10 # epilogue label
1711 cmp %r10,%rbx # context->Rip>=epilogue label
1712 jae .Lcommon_seh_tail
1715 lea 512($context),%rdi # &context.Xmm6
1717 .long 0xa548f3fc # cld; rep movsq
1718 lea `24+64+6*16`(%rax),%rax # adjust stack pointer
1723 mov %rbx,144($context) # restore context->Rbx
1724 mov %rbp,160($context) # restore context->Rbp
1725 mov %r12,216($context) # restore cotnext->R12
1730 mov %rax,152($context) # restore context->Rsp
1731 mov %rsi,168($context) # restore context->Rsi
1732 mov %rdi,176($context) # restore context->Rdi
1734 mov 40($disp),%rdi # disp->ContextRecord
1735 mov $context,%rsi # context
1736 mov \$154,%ecx # sizeof(CONTEXT)
1737 .long 0xa548f3fc # cld; rep movsq
1740 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1741 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1742 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1743 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1744 mov 40(%rsi),%r10 # disp->ContextRecord
1745 lea 56(%rsi),%r11 # &disp->HandlerData
1746 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1747 mov %r10,32(%rsp) # arg5
1748 mov %r11,40(%rsp) # arg6
1749 mov %r12,48(%rsp) # arg7
1750 mov %rcx,56(%rsp) # arg8, (NULL)
1751 call *__imp_RtlVirtualUnwind(%rip)
1753 mov \$1,%eax # ExceptionContinueSearch
1765 .size ssse3_handler,.-ssse3_handler
1769 .rva .LSEH_begin_sha1_block_data_order
1770 .rva .LSEH_end_sha1_block_data_order
1771 .rva .LSEH_info_sha1_block_data_order
1772 .rva .LSEH_begin_sha1_block_data_order_ssse3
1773 .rva .LSEH_end_sha1_block_data_order_ssse3
1774 .rva .LSEH_info_sha1_block_data_order_ssse3
1776 $code.=<<___ if ($avx);
1777 .rva .LSEH_begin_sha1_block_data_order_avx
1778 .rva .LSEH_end_sha1_block_data_order_avx
1779 .rva .LSEH_info_sha1_block_data_order_avx
1781 $code.=<<___ if ($avx>1);
1782 .rva .LSEH_begin_sha1_block_data_order_avx2
1783 .rva .LSEH_end_sha1_block_data_order_avx2
1784 .rva .LSEH_info_sha1_block_data_order_avx2
1789 .LSEH_info_sha1_block_data_order:
1792 .LSEH_info_sha1_block_data_order_ssse3:
1795 .rva .Lprologue_ssse3,.Lepilogue_ssse3 # HandlerData[]
1797 $code.=<<___ if ($avx);
1798 .LSEH_info_sha1_block_data_order_avx:
1801 .rva .Lprologue_avx,.Lepilogue_avx # HandlerData[]
1803 $code.=<<___ if ($avx>1);
1804 .LSEH_info_sha1_block_data_order_avx2:
1807 .rva .Lprologue_avx2,.Lepilogue_avx2 # HandlerData[]
1811 ####################################################################
1813 $code =~ s/\`([^\`]*)\`/eval $1/gem;