From 3562e78743202e43aec8727005182a2558117eca Mon Sep 17 00:00:00 2001 From: marha Date: Sun, 28 Jun 2009 22:07:26 +0000 Subject: Checked in the following released items: xkeyboard-config-1.4.tar.gz ttf-bitstream-vera-1.10.tar.gz font-alias-1.0.1.tar.gz font-sun-misc-1.0.0.tar.gz font-sun-misc-1.0.0.tar.gz font-sony-misc-1.0.0.tar.gz font-schumacher-misc-1.0.0.tar.gz font-mutt-misc-1.0.0.tar.gz font-misc-misc-1.0.0.tar.gz font-misc-meltho-1.0.0.tar.gz font-micro-misc-1.0.0.tar.gz font-jis-misc-1.0.0.tar.gz font-isas-misc-1.0.0.tar.gz font-dec-misc-1.0.0.tar.gz font-daewoo-misc-1.0.0.tar.gz font-cursor-misc-1.0.0.tar.gz font-arabic-misc-1.0.0.tar.gz font-winitzki-cyrillic-1.0.0.tar.gz font-misc-cyrillic-1.0.0.tar.gz font-cronyx-cyrillic-1.0.0.tar.gz font-screen-cyrillic-1.0.1.tar.gz font-xfree86-type1-1.0.1.tar.gz font-adobe-utopia-type1-1.0.1.tar.gz font-ibm-type1-1.0.0.tar.gz font-bitstream-type1-1.0.0.tar.gz font-bitstream-speedo-1.0.0.tar.gz font-bh-ttf-1.0.0.tar.gz font-bh-type1-1.0.0.tar.gz font-bitstream-100dpi-1.0.0.tar.gz font-bh-lucidatypewriter-100dpi-1.0.0.tar.gz font-bh-100dpi-1.0.0.tar.gz font-adobe-utopia-100dpi-1.0.1.tar.gz font-adobe-100dpi-1.0.0.tar.gz font-util-1.0.1.tar.gz font-bitstream-75dpi-1.0.0.tar.gz font-bh-lucidatypewriter-75dpi-1.0.0.tar.gz font-adobe-utopia-75dpi-1.0.1.tar.gz font-bh-75dpi-1.0.0.tar.gz bdftopcf-1.0.1.tar.gz font-adobe-75dpi-1.0.0.tar.gz mkfontscale-1.0.6.tar.gz openssl-0.9.8k.tar.gz bigreqsproto-1.0.2.tar.gz xtrans-1.2.2.tar.gz resourceproto-1.0.2.tar.gz inputproto-1.4.4.tar.gz compositeproto-0.4.tar.gz damageproto-1.1.0.tar.gz zlib-1.2.3.tar.gz xkbcomp-1.0.5.tar.gz freetype-2.3.9.tar.gz pthreads-w32-2-8-0-release.tar.gz pixman-0.12.0.tar.gz kbproto-1.0.3.tar.gz evieext-1.0.2.tar.gz fixesproto-4.0.tar.gz recordproto-1.13.2.tar.gz randrproto-1.2.2.tar.gz scrnsaverproto-1.1.0.tar.gz renderproto-0.9.3.tar.gz xcmiscproto-1.1.2.tar.gz fontsproto-2.0.2.tar.gz xextproto-7.0.3.tar.gz xproto-7.0.14.tar.gz libXdmcp-1.0.2.tar.gz libxkbfile-1.0.5.tar.gz libfontenc-1.0.4.tar.gz libXfont-1.3.4.tar.gz libX11-1.1.5.tar.gz libXau-1.0.4.tar.gz libxcb-1.1.tar.gz xorg-server-1.5.3.tar.gz --- openssl/crypto/sha/asm/sha1-x86_64.pl | 242 ++++++++++++++++++++++++++++++++++ 1 file changed, 242 insertions(+) create mode 100644 openssl/crypto/sha/asm/sha1-x86_64.pl (limited to 'openssl/crypto/sha/asm/sha1-x86_64.pl') diff --git a/openssl/crypto/sha/asm/sha1-x86_64.pl b/openssl/crypto/sha/asm/sha1-x86_64.pl new file mode 100644 index 000000000..f7ed67a72 --- /dev/null +++ b/openssl/crypto/sha/asm/sha1-x86_64.pl @@ -0,0 +1,242 @@ +#!/usr/bin/env perl +# +# ==================================================================== +# Written by Andy Polyakov for the OpenSSL +# project. The module is, however, dual licensed under OpenSSL and +# CRYPTOGAMS licenses depending on where you obtain it. For further +# details see http://www.openssl.org/~appro/cryptogams/. +# ==================================================================== +# +# sha1_block procedure for x86_64. +# +# It was brought to my attention that on EM64T compiler-generated code +# was far behind 32-bit assembler implementation. This is unlike on +# Opteron where compiler-generated code was only 15% behind 32-bit +# assembler, which originally made it hard to motivate the effort. +# There was suggestion to mechanically translate 32-bit code, but I +# dismissed it, reasoning that x86_64 offers enough register bank +# capacity to fully utilize SHA-1 parallelism. Therefore this fresh +# implementation:-) However! While 64-bit code does performs better +# on Opteron, I failed to beat 32-bit assembler on EM64T core. Well, +# x86_64 does offer larger *addressable* bank, but out-of-order core +# reaches for even more registers through dynamic aliasing, and EM64T +# core must have managed to run-time optimize even 32-bit code just as +# good as 64-bit one. Performance improvement is summarized in the +# following table: +# +# gcc 3.4 32-bit asm cycles/byte +# Opteron +45% +20% 6.8 +# Xeon P4 +65% +0% 9.9 +# Core2 +60% +10% 7.0 + +$output=shift; + +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or +( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or +die "can't locate x86_64-xlate.pl"; + +open STDOUT,"| $^X $xlate $output"; + +$ctx="%rdi"; # 1st arg +$inp="%rsi"; # 2nd arg +$num="%rdx"; # 3rd arg + +# reassign arguments in order to produce more compact code +$ctx="%r8"; +$inp="%r9"; +$num="%r10"; + +$xi="%eax"; +$t0="%ebx"; +$t1="%ecx"; +$A="%edx"; +$B="%esi"; +$C="%edi"; +$D="%ebp"; +$E="%r11d"; +$T="%r12d"; + +@V=($A,$B,$C,$D,$E,$T); + +sub PROLOGUE { +my $func=shift; +$code.=<<___; +.globl $func +.type $func,\@function,3 +.align 16 +$func: + push %rbx + push %rbp + push %r12 + mov %rsp,%rax + mov %rdi,$ctx # reassigned argument + sub \$`8+16*4`,%rsp + mov %rsi,$inp # reassigned argument + and \$-64,%rsp + mov %rdx,$num # reassigned argument + mov %rax,`16*4`(%rsp) + + mov 0($ctx),$A + mov 4($ctx),$B + mov 8($ctx),$C + mov 12($ctx),$D + mov 16($ctx),$E +___ +} + +sub EPILOGUE { +my $func=shift; +$code.=<<___; + mov `16*4`(%rsp),%rsp + pop %r12 + pop %rbp + pop %rbx + ret +.size $func,.-$func +___ +} + +sub BODY_00_19 { +my ($i,$a,$b,$c,$d,$e,$f,$host)=@_; +my $j=$i+1; +$code.=<<___ if ($i==0); + mov `4*$i`($inp),$xi + `"bswap $xi" if(!defined($host))` + mov $xi,`4*$i`(%rsp) +___ +$code.=<<___ if ($i<15); + lea 0x5a827999($xi,$e),$f + mov $c,$t0 + mov `4*$j`($inp),$xi + mov $a,$e + xor $d,$t0 + `"bswap $xi" if(!defined($host))` + rol \$5,$e + and $b,$t0 + mov $xi,`4*$j`(%rsp) + add $e,$f + xor $d,$t0 + rol \$30,$b + add $t0,$f +___ +$code.=<<___ if ($i>=15); + lea 0x5a827999($xi,$e),$f + mov `4*($j%16)`(%rsp),$xi + mov $c,$t0 + mov $a,$e + xor `4*(($j+2)%16)`(%rsp),$xi + xor $d,$t0 + rol \$5,$e + xor `4*(($j+8)%16)`(%rsp),$xi + and $b,$t0 + add $e,$f + xor `4*(($j+13)%16)`(%rsp),$xi + xor $d,$t0 + rol \$30,$b + add $t0,$f + rol \$1,$xi + mov $xi,`4*($j%16)`(%rsp) +___ +} + +sub BODY_20_39 { +my ($i,$a,$b,$c,$d,$e,$f)=@_; +my $j=$i+1; +my $K=($i<40)?0x6ed9eba1:0xca62c1d6; +$code.=<<___ if ($i<79); + lea $K($xi,$e),$f + mov `4*($j%16)`(%rsp),$xi + mov $c,$t0 + mov $a,$e + xor `4*(($j+2)%16)`(%rsp),$xi + xor $b,$t0 + rol \$5,$e + xor `4*(($j+8)%16)`(%rsp),$xi + xor $d,$t0 + add $e,$f + xor `4*(($j+13)%16)`(%rsp),$xi + rol \$30,$b + add $t0,$f + rol \$1,$xi +___ +$code.=<<___ if ($i<76); + mov $xi,`4*($j%16)`(%rsp) +___ +$code.=<<___ if ($i==79); + lea $K($xi,$e),$f + mov $c,$t0 + mov $a,$e + xor $b,$t0 + rol \$5,$e + xor $d,$t0 + add $e,$f + rol \$30,$b + add $t0,$f +___ +} + +sub BODY_40_59 { +my ($i,$a,$b,$c,$d,$e,$f)=@_; +my $j=$i+1; +$code.=<<___; + lea 0x8f1bbcdc($xi,$e),$f + mov `4*($j%16)`(%rsp),$xi + mov $b,$t0 + mov $b,$t1 + xor `4*(($j+2)%16)`(%rsp),$xi + mov $a,$e + and $c,$t0 + xor `4*(($j+8)%16)`(%rsp),$xi + or $c,$t1 + rol \$5,$e + xor `4*(($j+13)%16)`(%rsp),$xi + and $d,$t1 + add $e,$f + rol \$1,$xi + or $t1,$t0 + rol \$30,$b + mov $xi,`4*($j%16)`(%rsp) + add $t0,$f +___ +} + +$code=".text\n"; + +&PROLOGUE("sha1_block_data_order"); +$code.=".align 4\n.Lloop:\n"; +for($i=0;$i<20;$i++) { &BODY_00_19($i,@V); unshift(@V,pop(@V)); } +for(;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } +for(;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); } +for(;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); } +$code.=<<___; + add 0($ctx),$E + add 4($ctx),$T + add 8($ctx),$A + add 12($ctx),$B + add 16($ctx),$C + mov $E,0($ctx) + mov $T,4($ctx) + mov $A,8($ctx) + mov $B,12($ctx) + mov $C,16($ctx) + + xchg $E,$A # mov $E,$A + xchg $T,$B # mov $T,$B + xchg $E,$C # mov $A,$C + xchg $T,$D # mov $B,$D + # mov $C,$E + lea `16*4`($inp),$inp + sub \$1,$num + jnz .Lloop +___ +&EPILOGUE("sha1_block_data_order"); +$code.=<<___; +.asciz "SHA1 block transform for x86_64, CRYPTOGAMS by " +___ + +#################################################################### + +$code =~ s/\`([^\`]*)\`/eval $1/gem; +print $code; +close STDOUT; -- cgit v1.2.3