3 # ====================================================================
4 # Written by Andy Polyakov <appro@fy.chalmers.se> 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15 # assembler, which originally made it hard to motivate the effort.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.
52 ######################################################################
53 # Current performance is summarized in following table. Numbers are
54 # CPU clock cycles spent to process single byte (less is better).
60 # Atom 11.0 9.5/+15% -
61 # Westmere 7.1 5.5/+29% -
62 # Sandy Bridge 7.9 6.2/+28% 5.1/+54%
63 # Ivy Bridge 6.4 4.7/+35% 4.6/+37%
64 # Bulldozer 10.9 6.0/+82%
65 # VIA Nano 10.2 7.4/+38%
69 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
71 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
73 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
74 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
75 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
76 die "can't locate x86_64-xlate.pl";
78 $avx=1 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
79 =~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
81 $avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
82 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
84 $avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
85 `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
88 open OUT,"| \"$^X\" $xlate $flavour $output";
91 $ctx="%rdi"; # 1st arg
92 $inp="%rsi"; # 2nd arg
93 $num="%rdx"; # 3rd arg
95 # reassign arguments in order to produce more compact code
113 my ($i,$a,$b,$c,$d,$e)=@_;
115 $code.=<<___ if ($i==0);
116 mov `4*$i`($inp),$xi[0]
118 mov $xi[0],`4*$i`(%rsp)
120 $code.=<<___ if ($i<15);
122 mov `4*$j`($inp),$xi[1]
127 lea 0x5a827999($xi[0],$e),$e
129 mov $xi[1],`4*$j`(%rsp)
135 $code.=<<___ if ($i>=15);
136 mov `4*($j%16)`(%rsp),$xi[1]
139 xor `4*(($j+2)%16)`(%rsp),$xi[1]
142 xor `4*(($j+8)%16)`(%rsp),$xi[1]
144 lea 0x5a827999($xi[0],$e),$e
145 xor `4*(($j+13)%16)`(%rsp),$xi[1]
150 mov $xi[1],`4*($j%16)`(%rsp)
153 unshift(@xi,pop(@xi));
157 my ($i,$a,$b,$c,$d,$e)=@_;
159 my $K=($i<40)?0x6ed9eba1:0xca62c1d6;
160 $code.=<<___ if ($i<79);
161 mov `4*($j%16)`(%rsp),$xi[1]
164 xor `4*(($j+2)%16)`(%rsp),$xi[1]
168 xor `4*(($j+8)%16)`(%rsp),$xi[1]
171 xor `4*(($j+13)%16)`(%rsp),$xi[1]
176 $code.=<<___ if ($i<76);
177 mov $xi[1],`4*($j%16)`(%rsp)
179 $code.=<<___ if ($i==79);
190 unshift(@xi,pop(@xi));
194 my ($i,$a,$b,$c,$d,$e)=@_;
197 mov `4*($j%16)`(%rsp),$xi[1]
200 xor `4*(($j+2)%16)`(%rsp),$xi[1]
203 xor `4*(($j+8)%16)`(%rsp),$xi[1]
205 lea 0x8f1bbcdc($xi[0],$e),$e
207 xor `4*(($j+13)%16)`(%rsp),$xi[1]
213 mov $xi[1],`4*($j%16)`(%rsp)
216 unshift(@xi,pop(@xi));
221 .extern OPENSSL_ia32cap_P
223 .globl sha1_block_data_order
224 .type sha1_block_data_order,\@function,3
226 sha1_block_data_order:
227 mov OPENSSL_ia32cap_P+0(%rip),%r9d
228 mov OPENSSL_ia32cap_P+4(%rip),%r8d
229 test \$`1<<9`,%r8d # check SSSE3 bit
232 $code.=<<___ if ($avx);
233 and \$`1<<28`,%r8d # mask AVX bit
234 and \$`1<<30`,%r9d # mask "Intel CPU" bit
236 cmp \$`1<<28|1<<30`,%r8d
249 mov %rdi,$ctx # reassigned argument
251 mov %rsi,$inp # reassigned argument
253 mov %rdx,$num # reassigned argument
254 mov %r11,`16*4`(%rsp)
267 for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
268 for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
269 for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
270 for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
284 lea `16*4`($inp),$inp
287 mov `16*4`(%rsp),%rsi
295 .size sha1_block_data_order,.-sha1_block_data_order
299 my @X=map("%xmm$_",(4..7,0..3));
300 my @Tx=map("%xmm$_",(8..10));
301 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
302 my @T=("%esi","%edi");
306 my $_rol=sub { &rol(@_) };
307 my $_ror=sub { &ror(@_) };
310 .type sha1_block_data_order_ssse3,\@function,3
312 sha1_block_data_order_ssse3:
317 lea `-64-($win64?5*16:0)`(%rsp),%rsp
319 $code.=<<___ if ($win64);
320 movaps %xmm6,64+0(%rsp)
321 movaps %xmm7,64+16(%rsp)
322 movaps %xmm8,64+32(%rsp)
323 movaps %xmm9,64+48(%rsp)
324 movaps %xmm10,64+64(%rsp)
328 mov %rdi,$ctx # reassigned argument
329 mov %rsi,$inp # reassigned argument
330 mov %rdx,$num # reassigned argument
334 lea K_XX_XX(%rip),$K_XX_XX
336 mov 0($ctx),$A # load context
340 mov $B,@T[0] # magic seed
343 movdqa 64($K_XX_XX),@X[2] # pbswap mask
344 movdqa 0($K_XX_XX),@Tx[1] # K_00_19
345 movdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
346 movdqu 16($inp),@X[-3&7]
347 movdqu 32($inp),@X[-2&7]
348 movdqu 48($inp),@X[-1&7]
349 pshufb @X[2],@X[-4&7] # byte swap
351 pshufb @X[2],@X[-3&7]
352 pshufb @X[2],@X[-2&7]
353 pshufb @X[2],@X[-1&7]
354 paddd @Tx[1],@X[-4&7] # add K_00_19
355 paddd @Tx[1],@X[-3&7]
356 paddd @Tx[1],@X[-2&7]
357 movdqa @X[-4&7],0(%rsp) # X[]+K xfer to IALU
358 psubd @Tx[1],@X[-4&7] # restore X[]
359 movdqa @X[-3&7],16(%rsp)
360 psubd @Tx[1],@X[-3&7]
361 movdqa @X[-2&7],32(%rsp)
362 psubd @Tx[1],@X[-2&7]
366 sub AUTOLOAD() # thunk [simplified] 32-bit style perlasm
367 { my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
369 $arg = "\$$arg" if ($arg*1 eq $arg);
370 $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
373 sub Xupdate_ssse3_16_31() # recall that $Xi starts wtih 4
376 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
379 &movdqa (@X[0],@X[-3&7]);
382 &movdqa (@Tx[0],@X[-1&7]);
383 &palignr(@X[0],@X[-4&7],8); # compose "X[-14]" in "X[0]"
387 &paddd (@Tx[1],@X[-1&7]);
390 &psrldq (@Tx[0],4); # "X[-3]", 3 dwords
393 &pxor (@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
397 &pxor (@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
403 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
406 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
410 &movdqa (@Tx[2],@X[0]);
411 &movdqa (@Tx[0],@X[0]);
417 &pslldq (@Tx[2],12); # "X[0]"<<96, extract one dword
418 &paddd (@X[0],@X[0]);
427 &movdqa (@Tx[1],@Tx[2]);
432 &por (@X[0],@Tx[0]); # "X[0]"<<<=1
439 &pxor (@X[0],@Tx[2]);
442 &movdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX
446 &pxor (@X[0],@Tx[1]); # "X[0]"^=("X[0]">>96)<<<2
448 foreach (@insns) { eval; } # remaining instructions [if any]
450 $Xi++; push(@X,shift(@X)); # "rotate" X[]
451 push(@Tx,shift(@Tx));
454 sub Xupdate_ssse3_32_79()
457 my @insns = (&$body,&$body,&$body,&$body); # 32 to 44 instructions
460 &movdqa (@Tx[0],@X[-1&7]) if ($Xi==8);
461 eval(shift(@insns)); # body_20_39
462 &pxor (@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
463 &palignr(@Tx[0],@X[-2&7],8); # compose "X[-6]"
466 eval(shift(@insns)); # rol
468 &pxor (@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
470 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
472 &movdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
473 } else { # ... or load next one
474 &movdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
476 &paddd (@Tx[1],@X[-1&7]);
477 eval(shift(@insns)); # ror
480 &pxor (@X[0],@Tx[0]); # "X[0]"^="X[-6]"
481 eval(shift(@insns)); # body_20_39
484 eval(shift(@insns)); # rol
486 &movdqa (@Tx[0],@X[0]);
487 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
490 eval(shift(@insns)); # ror
494 eval(shift(@insns)); # body_20_39
498 eval(shift(@insns)); # rol
501 eval(shift(@insns)); # ror
504 &por (@X[0],@Tx[0]); # "X[0]"<<<=2
505 eval(shift(@insns)); # body_20_39
507 &movdqa (@Tx[1],@X[0]) if ($Xi<19);
509 eval(shift(@insns)); # rol
512 eval(shift(@insns)); # rol
515 foreach (@insns) { eval; } # remaining instructions
517 $Xi++; push(@X,shift(@X)); # "rotate" X[]
518 push(@Tx,shift(@Tx));
521 sub Xuplast_ssse3_80()
524 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
528 &paddd (@Tx[1],@X[-1&7]);
534 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
536 foreach (@insns) { eval; } # remaining instructions
539 &je (".Ldone_ssse3");
541 unshift(@Tx,pop(@Tx));
543 &movdqa (@X[2],"64($K_XX_XX)"); # pbswap mask
544 &movdqa (@Tx[1],"0($K_XX_XX)"); # K_00_19
545 &movdqu (@X[-4&7],"0($inp)"); # load input
546 &movdqu (@X[-3&7],"16($inp)");
547 &movdqu (@X[-2&7],"32($inp)");
548 &movdqu (@X[-1&7],"48($inp)");
549 &pshufb (@X[-4&7],@X[2]); # byte swap
558 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
563 &pshufb (@X[($Xi-3)&7],@X[2]);
566 &paddd (@X[($Xi-4)&7],@Tx[1]);
571 &movdqa (eval(16*$Xi)."(%rsp)",@X[($Xi-4)&7]); # X[]+K xfer to IALU
574 &psubd (@X[($Xi-4)&7],@Tx[1]);
576 foreach (@insns) { eval; }
583 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
586 foreach (@insns) { eval; }
591 '($a,$b,$c,$d,$e)=@V;'.
592 '&add ($e,eval(4*($j&15))."(%rsp)");', # X[]+K xfer
594 '&mov (@T[1],$a);', # $b in next round
596 '&and (@T[0],$c);', # ($b&($c^$d))
597 '&xor ($c,$d);', # restore $c
600 '&$_ror ($b,$j?7:2);', # $b>>>2
601 '&add ($e,@T[0]);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
607 '($a,$b,$c,$d,$e)=@V;'.
608 '&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer
609 '&xor (@T[0],$d);', # ($b^$d)
610 '&mov (@T[1],$a);', # $b in next round
612 '&xor (@T[0],$c);', # ($b^$d^$c)
614 '&$_ror ($b,7);', # $b>>>2
615 '&add ($e,@T[0]);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));'
621 '($a,$b,$c,$d,$e)=@V;'.
624 '&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer
625 '&and (@T[0],$T[1]);',
626 '&$_ror ($b,7);', # $b>>>2
628 '&mov (@T[1],$a);', # $b in next round
631 '&mov (@T[0],$b);', # copy of $c in next round
632 '&add ($e,$a);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));'
639 &Xupdate_ssse3_16_31(\&body_00_19);
640 &Xupdate_ssse3_16_31(\&body_00_19);
641 &Xupdate_ssse3_16_31(\&body_00_19);
642 &Xupdate_ssse3_16_31(\&body_00_19);
643 &Xupdate_ssse3_32_79(\&body_00_19);
644 &Xupdate_ssse3_32_79(\&body_20_39);
645 &Xupdate_ssse3_32_79(\&body_20_39);
646 &Xupdate_ssse3_32_79(\&body_20_39);
647 &Xupdate_ssse3_32_79(\&body_20_39);
648 &Xupdate_ssse3_32_79(\&body_20_39);
649 &mov (@T[1],@V[2]); # copy of $c in next round
650 &Xupdate_ssse3_32_79(\&body_40_59);
651 &Xupdate_ssse3_32_79(\&body_40_59);
652 &Xupdate_ssse3_32_79(\&body_40_59);
653 &Xupdate_ssse3_32_79(\&body_40_59);
654 &Xupdate_ssse3_32_79(\&body_40_59);
655 &Xupdate_ssse3_32_79(\&body_20_39);
656 &Xuplast_ssse3_80(\&body_20_39); # can jump to "done"
658 $saved_j=$j; @saved_V=@V;
660 &Xloop_ssse3(\&body_20_39);
661 &Xloop_ssse3(\&body_20_39);
662 &Xloop_ssse3(\&body_20_39);
665 add 0($ctx),$A # update context
672 mov @T[0],$B # magic seed
681 $j=$saved_j; @V=@saved_V;
683 &Xtail_ssse3(\&body_20_39);
684 &Xtail_ssse3(\&body_20_39);
685 &Xtail_ssse3(\&body_20_39);
688 add 0($ctx),$A # update context
699 $code.=<<___ if ($win64);
700 movaps 64+0(%rsp),%xmm6
701 movaps 64+16(%rsp),%xmm7
702 movaps 64+32(%rsp),%xmm8
703 movaps 64+48(%rsp),%xmm9
704 movaps 64+64(%rsp),%xmm10
707 lea `64+($win64?5*16:0)`(%rsp),%rsi
714 .size sha1_block_data_order_ssse3,.-sha1_block_data_order_ssse3
719 my @X=map("%xmm$_",(4..7,0..3));
720 my @Tx=map("%xmm$_",(8..10));
721 my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
722 my @T=("%esi","%edi");
726 my $_rol=sub { &shld(@_[0],@_) };
727 my $_ror=sub { &shrd(@_[0],@_) };
730 .type sha1_block_data_order_avx,\@function,3
732 sha1_block_data_order_avx:
737 lea `-64-($win64?5*16:0)`(%rsp),%rsp
739 $code.=<<___ if ($win64);
740 movaps %xmm6,64+0(%rsp)
741 movaps %xmm7,64+16(%rsp)
742 movaps %xmm8,64+32(%rsp)
743 movaps %xmm9,64+48(%rsp)
744 movaps %xmm10,64+64(%rsp)
748 mov %rdi,$ctx # reassigned argument
749 mov %rsi,$inp # reassigned argument
750 mov %rdx,$num # reassigned argument
755 lea K_XX_XX(%rip),$K_XX_XX
757 mov 0($ctx),$A # load context
761 mov $B,@T[0] # magic seed
764 vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
765 vmovdqa 0($K_XX_XX),@Tx[1] # K_00_19
766 vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
767 vmovdqu 16($inp),@X[-3&7]
768 vmovdqu 32($inp),@X[-2&7]
769 vmovdqu 48($inp),@X[-1&7]
770 vpshufb @X[2],@X[-4&7],@X[-4&7] # byte swap
772 vpshufb @X[2],@X[-3&7],@X[-3&7]
773 vpshufb @X[2],@X[-2&7],@X[-2&7]
774 vpshufb @X[2],@X[-1&7],@X[-1&7]
775 vpaddd @Tx[1],@X[-4&7],@X[0] # add K_00_19
776 vpaddd @Tx[1],@X[-3&7],@X[1]
777 vpaddd @Tx[1],@X[-2&7],@X[2]
778 vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
779 vmovdqa @X[1],16(%rsp)
780 vmovdqa @X[2],32(%rsp)
784 sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
787 my @insns = (&$body,&$body,&$body,&$body); # 40 instructions
792 &vpalignr(@X[0],@X[-3&7],@X[-4&7],8); # compose "X[-14]" in "X[0]"
796 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
799 &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
802 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
806 &vpxor (@Tx[0],@Tx[0],@X[-2&7]); # "X[-3]"^"X[-8]"
812 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-3]"^"X[-8]"
815 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
819 &vpsrld (@Tx[0],@X[0],31);
825 &vpslldq(@Tx[2],@X[0],12); # "X[0]"<<96, extract one dword
826 &vpaddd (@X[0],@X[0],@X[0]);
832 &vpsrld (@Tx[1],@Tx[2],30);
833 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=1
839 &vpslld (@Tx[2],@Tx[2],2);
840 &vpxor (@X[0],@X[0],@Tx[1]);
846 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
849 &vmovdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX
854 foreach (@insns) { eval; } # remaining instructions [if any]
856 $Xi++; push(@X,shift(@X)); # "rotate" X[]
857 push(@Tx,shift(@Tx));
860 sub Xupdate_avx_32_79()
863 my @insns = (&$body,&$body,&$body,&$body); # 32 to 44 instructions
866 &vpalignr(@Tx[0],@X[-1&7],@X[-2&7],8); # compose "X[-6]"
867 &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"="X[-32]"^"X[-16]"
868 eval(shift(@insns)); # body_20_39
871 eval(shift(@insns)); # rol
873 &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
875 eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
877 &vmovdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX...
878 } else { # ... or load next one
879 &vmovdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
881 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
882 eval(shift(@insns)); # ror
885 &vpxor (@X[0],@X[0],@Tx[0]); # "X[0]"^="X[-6]"
886 eval(shift(@insns)); # body_20_39
889 eval(shift(@insns)); # rol
891 &vpsrld (@Tx[0],@X[0],30);
892 &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer to IALU
895 eval(shift(@insns)); # ror
898 &vpslld (@X[0],@X[0],2);
899 eval(shift(@insns)); # body_20_39
902 eval(shift(@insns)); # rol
905 eval(shift(@insns)); # ror
908 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
909 eval(shift(@insns)); # body_20_39
911 &vmovdqa (@Tx[1],@X[0]) if ($Xi<19);
913 eval(shift(@insns)); # rol
916 eval(shift(@insns)); # rol
919 foreach (@insns) { eval; } # remaining instructions
921 $Xi++; push(@X,shift(@X)); # "rotate" X[]
922 push(@Tx,shift(@Tx));
928 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
932 &vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
938 &movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
940 foreach (@insns) { eval; } # remaining instructions
945 unshift(@Tx,pop(@Tx));
947 &vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask
948 &vmovdqa(@Tx[1],"0($K_XX_XX)"); # K_00_19
949 &vmovdqu(@X[-4&7],"0($inp)"); # load input
950 &vmovdqu(@X[-3&7],"16($inp)");
951 &vmovdqu(@X[-2&7],"32($inp)");
952 &vmovdqu(@X[-1&7],"48($inp)");
953 &vpshufb(@X[-4&7],@X[-4&7],@X[2]); # byte swap
962 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
967 &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]);
970 &vpaddd (@X[$Xi&7],@X[($Xi-4)&7],@Tx[1]);
975 &vmovdqa(eval(16*$Xi)."(%rsp)",@X[$Xi&7]); # X[]+K xfer to IALU
979 foreach (@insns) { eval; }
986 my @insns = (&$body,&$body,&$body,&$body); # 32 instructions
989 foreach (@insns) { eval; }
996 &Xupdate_avx_16_31(\&body_00_19);
997 &Xupdate_avx_16_31(\&body_00_19);
998 &Xupdate_avx_16_31(\&body_00_19);
999 &Xupdate_avx_16_31(\&body_00_19);
1000 &Xupdate_avx_32_79(\&body_00_19);
1001 &Xupdate_avx_32_79(\&body_20_39);
1002 &Xupdate_avx_32_79(\&body_20_39);
1003 &Xupdate_avx_32_79(\&body_20_39);
1004 &Xupdate_avx_32_79(\&body_20_39);
1005 &Xupdate_avx_32_79(\&body_20_39);
1006 &mov (@T[1],@V[2]); # copy of $c in next round
1007 &Xupdate_avx_32_79(\&body_40_59);
1008 &Xupdate_avx_32_79(\&body_40_59);
1009 &Xupdate_avx_32_79(\&body_40_59);
1010 &Xupdate_avx_32_79(\&body_40_59);
1011 &Xupdate_avx_32_79(\&body_40_59);
1012 &Xupdate_avx_32_79(\&body_20_39);
1013 &Xuplast_avx_80(\&body_20_39); # can jump to "done"
1015 $saved_j=$j; @saved_V=@V;
1017 &Xloop_avx(\&body_20_39);
1018 &Xloop_avx(\&body_20_39);
1019 &Xloop_avx(\&body_20_39);
1022 add 0($ctx),$A # update context
1029 mov @T[0],$B # magic seed
1038 $j=$saved_j; @V=@saved_V;
1040 &Xtail_avx(\&body_20_39);
1041 &Xtail_avx(\&body_20_39);
1042 &Xtail_avx(\&body_20_39);
1047 add 0($ctx),$A # update context
1058 $code.=<<___ if ($win64);
1059 movaps 64+0(%rsp),%xmm6
1060 movaps 64+16(%rsp),%xmm7
1061 movaps 64+32(%rsp),%xmm8
1062 movaps 64+48(%rsp),%xmm9
1063 movaps 64+64(%rsp),%xmm10
1066 lea `64+($win64?5*16:0)`(%rsp),%rsi
1073 .size sha1_block_data_order_avx,.-sha1_block_data_order_avx
1079 .long 0x5a827999,0x5a827999,0x5a827999,0x5a827999 # K_00_19
1080 .long 0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1 # K_20_39
1081 .long 0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc # K_40_59
1082 .long 0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6 # K_60_79
1083 .long 0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f # pbswap mask
1087 .asciz "SHA1 block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
1091 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1092 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1100 .extern __imp_RtlVirtualUnwind
1101 .type se_handler,\@abi-omnipotent
1115 mov 120($context),%rax # pull context->Rax
1116 mov 248($context),%rbx # pull context->Rip
1118 lea .Lprologue(%rip),%r10
1119 cmp %r10,%rbx # context->Rip<.Lprologue
1120 jb .Lcommon_seh_tail
1122 mov 152($context),%rax # pull context->Rsp
1124 lea .Lepilogue(%rip),%r10
1125 cmp %r10,%rbx # context->Rip>=.Lepilogue
1126 jae .Lcommon_seh_tail
1128 mov `16*4`(%rax),%rax # pull saved stack pointer
1135 mov %rbx,144($context) # restore context->Rbx
1136 mov %rbp,160($context) # restore context->Rbp
1137 mov %r12,216($context) # restore context->R12
1138 mov %r13,224($context) # restore context->R13
1140 jmp .Lcommon_seh_tail
1141 .size se_handler,.-se_handler
1143 .type ssse3_handler,\@abi-omnipotent
1157 mov 120($context),%rax # pull context->Rax
1158 mov 248($context),%rbx # pull context->Rip
1160 mov 8($disp),%rsi # disp->ImageBase
1161 mov 56($disp),%r11 # disp->HandlerData
1163 mov 0(%r11),%r10d # HandlerData[0]
1164 lea (%rsi,%r10),%r10 # prologue label
1165 cmp %r10,%rbx # context->Rip<prologue label
1166 jb .Lcommon_seh_tail
1168 mov 152($context),%rax # pull context->Rsp
1170 mov 4(%r11),%r10d # HandlerData[1]
1171 lea (%rsi,%r10),%r10 # epilogue label
1172 cmp %r10,%rbx # context->Rip>=epilogue label
1173 jae .Lcommon_seh_tail
1176 lea 512($context),%rdi # &context.Xmm6
1178 .long 0xa548f3fc # cld; rep movsq
1179 lea `24+64+5*16`(%rax),%rax # adjust stack pointer
1184 mov %rbx,144($context) # restore context->Rbx
1185 mov %rbp,160($context) # restore context->Rbp
1186 mov %r12,216($context) # restore cotnext->R12
1191 mov %rax,152($context) # restore context->Rsp
1192 mov %rsi,168($context) # restore context->Rsi
1193 mov %rdi,176($context) # restore context->Rdi
1195 mov 40($disp),%rdi # disp->ContextRecord
1196 mov $context,%rsi # context
1197 mov \$154,%ecx # sizeof(CONTEXT)
1198 .long 0xa548f3fc # cld; rep movsq
1201 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1202 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1203 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1204 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1205 mov 40(%rsi),%r10 # disp->ContextRecord
1206 lea 56(%rsi),%r11 # &disp->HandlerData
1207 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1208 mov %r10,32(%rsp) # arg5
1209 mov %r11,40(%rsp) # arg6
1210 mov %r12,48(%rsp) # arg7
1211 mov %rcx,56(%rsp) # arg8, (NULL)
1212 call *__imp_RtlVirtualUnwind(%rip)
1214 mov \$1,%eax # ExceptionContinueSearch
1226 .size ssse3_handler,.-ssse3_handler
1230 .rva .LSEH_begin_sha1_block_data_order
1231 .rva .LSEH_end_sha1_block_data_order
1232 .rva .LSEH_info_sha1_block_data_order
1233 .rva .LSEH_begin_sha1_block_data_order_ssse3
1234 .rva .LSEH_end_sha1_block_data_order_ssse3
1235 .rva .LSEH_info_sha1_block_data_order_ssse3
1237 $code.=<<___ if ($avx);
1238 .rva .LSEH_begin_sha1_block_data_order_avx
1239 .rva .LSEH_end_sha1_block_data_order_avx
1240 .rva .LSEH_info_sha1_block_data_order_avx
1245 .LSEH_info_sha1_block_data_order:
1248 .LSEH_info_sha1_block_data_order_ssse3:
1251 .rva .Lprologue_ssse3,.Lepilogue_ssse3 # HandlerData[]
1253 $code.=<<___ if ($avx);
1254 .LSEH_info_sha1_block_data_order_avx:
1257 .rva .Lprologue_avx,.Lepilogue_avx # HandlerData[]
1261 ####################################################################
1263 $code =~ s/\`([^\`]*)\`/eval $1/gem;