From d7ce91a2b7d322ffce68e9eea917fa481e578d29 Mon Sep 17 00:00:00 2001 From: cvs2svn Date: Sat, 9 Sep 2006 00:11:04 +0000 Subject: This commit was manufactured by cvs2git to create tag 'OPENBSD_4_0_BASE'. --- src/lib/libcrypto/rc4/asm/rc4-586.pl | 229 -------------------------------- src/lib/libcrypto/rc4/asm/rc4-x86_64.pl | 150 --------------------- 2 files changed, 379 deletions(-) delete mode 100644 src/lib/libcrypto/rc4/asm/rc4-586.pl delete mode 100755 src/lib/libcrypto/rc4/asm/rc4-x86_64.pl (limited to 'src/lib/libcrypto/rc4/asm') diff --git a/src/lib/libcrypto/rc4/asm/rc4-586.pl b/src/lib/libcrypto/rc4/asm/rc4-586.pl deleted file mode 100644 index d6e98f0811..0000000000 --- a/src/lib/libcrypto/rc4/asm/rc4-586.pl +++ /dev/null @@ -1,229 +0,0 @@ -#!/usr/local/bin/perl - -# At some point it became apparent that the original SSLeay RC4 -# assembler implementation performs suboptimaly on latest IA-32 -# microarchitectures. After re-tuning performance has changed as -# following: -# -# Pentium +0% -# Pentium III +17% -# AMD +52%(*) -# P4 +180%(**) -# -# (*) This number is actually a trade-off:-) It's possible to -# achieve +72%, but at the cost of -48% off PIII performance. -# In other words code performing further 13% faster on AMD -# would perform almost 2 times slower on Intel PIII... -# For reference! This code delivers ~80% of rc4-amd64.pl -# performance on the same Opteron machine. -# (**) This number requires compressed key schedule set up by -# RC4_set_key and therefore doesn't apply to 0.9.7 [option for -# compressed key schedule is implemented in 0.9.8 and later, -# see commentary section in rc4_skey.c for further details]. -# -# - -push(@INC,"perlasm","../../perlasm"); -require "x86asm.pl"; - -&asm_init($ARGV[0],"rc4-586.pl"); - -$x="eax"; -$y="ebx"; -$tx="ecx"; -$ty="edx"; -$in="esi"; -$out="edi"; -$d="ebp"; - -&RC4("RC4"); - -&asm_finish(); - -sub RC4_loop - { - local($n,$p,$char)=@_; - - &comment("Round $n"); - - if ($char) - { - if ($p >= 0) - { - &mov($ty, &swtmp(2)); - &cmp($ty, $in); - &jbe(&label("finished")); - &inc($in); - } - else - { - &add($ty, 8); - &inc($in); - &cmp($ty, $in); - &jb(&label("finished")); - &mov(&swtmp(2), $ty); - } - } - # Moved out - # &mov( $tx, &DWP(0,$d,$x,4)) if $p < 0; - - &add( &LB($y), &LB($tx)); - &mov( $ty, &DWP(0,$d,$y,4)); - # XXX - &mov( &DWP(0,$d,$x,4),$ty); - &add( $ty, $tx); - &mov( &DWP(0,$d,$y,4),$tx); - &and( $ty, 0xff); - &inc( &LB($x)); # NEXT ROUND - &mov( $tx, &DWP(0,$d,$x,4)) if $p < 1; # NEXT ROUND - &mov( $ty, &DWP(0,$d,$ty,4)); - - if (!$char) - { - #moved up into last round - if ($p >= 1) - { - &add( $out, 8) - } - &movb( &BP($n,"esp","",0), &LB($ty)); - } - else - { - # Note in+=8 has occured - &movb( &HB($ty), &BP(-1,$in,"",0)); - # XXX - &xorb(&LB($ty), &HB($ty)); - # XXX - &movb(&BP($n,$out,"",0),&LB($ty)); - } - } - - -sub RC4 - { - local($name)=@_; - - &function_begin_B($name,""); - - &mov($ty,&wparam(1)); # len - &cmp($ty,0); - &jne(&label("proceed")); - &ret(); - &set_label("proceed"); - - &comment(""); - - &push("ebp"); - &push("ebx"); - &push("esi"); - &xor( $x, $x); # avoid partial register stalls - &push("edi"); - &xor( $y, $y); # avoid partial register stalls - &mov( $d, &wparam(0)); # key - &mov( $in, &wparam(2)); - - &movb( &LB($x), &BP(0,$d,"",1)); - &movb( &LB($y), &BP(4,$d,"",1)); - - &mov( $out, &wparam(3)); - &inc( &LB($x)); - - &stack_push(3); # 3 temp variables - &add( $d, 8); - - # detect compressed schedule, see commentary section in rc4_skey.c... - # in 0.9.7 context ~50 bytes below RC4_CHAR label remain redundant, - # as compressed key schedule is set up in 0.9.8 and later. - &cmp(&DWP(256,$d),-1); - &je(&label("RC4_CHAR")); - - &lea( $ty, &DWP(-8,$ty,$in)); - - # check for 0 length input - - &mov( &swtmp(2), $ty); # this is now address to exit at - &mov( $tx, &DWP(0,$d,$x,4)); - - &cmp( $ty, $in); - &jb( &label("end")); # less than 8 bytes - - &set_label("start"); - - # filling DELAY SLOT - &add( $in, 8); - - &RC4_loop(0,-1,0); - &RC4_loop(1,0,0); - &RC4_loop(2,0,0); - &RC4_loop(3,0,0); - &RC4_loop(4,0,0); - &RC4_loop(5,0,0); - &RC4_loop(6,0,0); - &RC4_loop(7,1,0); - - &comment("apply the cipher text"); - # xor the cipher data with input - - #&add( $out, 8); #moved up into last round - - &mov( $tx, &swtmp(0)); - &mov( $ty, &DWP(-8,$in,"",0)); - &xor( $tx, $ty); - &mov( $ty, &DWP(-4,$in,"",0)); - &mov( &DWP(-8,$out,"",0), $tx); - &mov( $tx, &swtmp(1)); - &xor( $tx, $ty); - &mov( $ty, &swtmp(2)); # load end ptr; - &mov( &DWP(-4,$out,"",0), $tx); - &mov( $tx, &DWP(0,$d,$x,4)); - &cmp($in, $ty); - &jbe(&label("start")); - - &set_label("end"); - - # There is quite a bit of extra crap in RC4_loop() for this - # first round - &RC4_loop(0,-1,1); - &RC4_loop(1,0,1); - &RC4_loop(2,0,1); - &RC4_loop(3,0,1); - &RC4_loop(4,0,1); - &RC4_loop(5,0,1); - &RC4_loop(6,1,1); - - &jmp(&label("finished")); - - &align(16); - # this is essentially Intel P4 specific codepath, see rc4_skey.c, - # and is engaged in 0.9.8 and later context... - &set_label("RC4_CHAR"); - - &lea ($ty,&DWP(0,$in,$ty)); - &mov (&swtmp(2),$ty); - - # strangely enough unrolled loop performs over 20% slower... - &set_label("RC4_CHAR_loop"); - &movz ($tx,&BP(0,$d,$x)); - &add (&LB($y),&LB($tx)); - &movz ($ty,&BP(0,$d,$y)); - &movb (&BP(0,$d,$y),&LB($tx)); - &movb (&BP(0,$d,$x),&LB($ty)); - &add (&LB($ty),&LB($tx)); - &movz ($ty,&BP(0,$d,$ty)); - &xorb (&LB($ty),&BP(0,$in)); - &movb (&BP(0,$out),&LB($ty)); - &inc (&LB($x)); - &inc ($in); - &inc ($out); - &cmp ($in,&swtmp(2)); - &jb (&label("RC4_CHAR_loop")); - - &set_label("finished"); - &dec( $x); - &stack_pop(3); - &movb( &BP(-4,$d,"",0),&LB($y)); - &movb( &BP(-8,$d,"",0),&LB($x)); - - &function_end($name); - } - diff --git a/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl b/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl deleted file mode 100755 index b628daca70..0000000000 --- a/src/lib/libcrypto/rc4/asm/rc4-x86_64.pl +++ /dev/null @@ -1,150 +0,0 @@ -#!/usr/bin/env perl -# -# ==================================================================== -# Written by Andy Polyakov for the OpenSSL -# project. Rights for redistribution and usage in source and binary -# forms are granted according to the OpenSSL license. -# ==================================================================== -# -# Unlike 0.9.7f this code expects RC4_CHAR back in config line! See -# commentary section in corresponding script in development branch -# for background information about this option carousel. For those -# who don't have energy to figure out these gory details, here is -# basis in form of performance matrix relative to the original -# 0.9.7e C code-base: -# -# 0.9.7e 0.9.7f this -# AMD64 1x 3.3x 2.4x -# EM64T 1x 0.8x 1.5x -# -# In other words idea is to trade -25% AMD64 performance to compensate -# for deterioration and gain +90% on EM64T core. Development branch -# maintains best performance for either target, i.e. 3.3x for AMD64 -# and 1.5x for EM64T. - -$output=shift; - -open STDOUT,">$output" || die "can't open $output: $!"; - -$dat="%rdi"; # arg1 -$len="%rsi"; # arg2 -$inp="%rdx"; # arg3 -$out="%rcx"; # arg4 - -@XX=("%r8","%r10"); -@TX=("%r9","%r11"); -$YY="%r12"; -$TY="%r13"; - -$code=<<___;; -.text - -.globl RC4 -.type RC4,\@function -.align 16 -RC4: or $len,$len - jne .Lentry - repret -.Lentry: - push %r12 - push %r13 - - add \$2,$dat - movzb -2($dat),$XX[0]#d - movzb -1($dat),$YY#d - - add \$1,$XX[0]#b - movzb ($dat,$XX[0]),$TX[0]#d - test \$-8,$len - jz .Lcloop1 - push %rbx -.align 16 # incidentally aligned already -.Lcloop8: - mov ($inp),%eax - mov 4($inp),%ebx -___ -# unroll 2x4-wise, because 64-bit rotates kill Intel P4... -for ($i=0;$i<4;$i++) { -$code.=<<___; - add $TX[0]#b,$YY#b - lea 1($XX[0]),$XX[1] - movzb ($dat,$YY),$TY#d - movzb $XX[1]#b,$XX[1]#d - movzb ($dat,$XX[1]),$TX[1]#d - movb $TX[0]#b,($dat,$YY) - cmp $XX[1],$YY - movb $TY#b,($dat,$XX[0]) - jne .Lcmov$i # Intel cmov is sloooow... - mov $TX[0],$TX[1] -.Lcmov$i: - add $TX[0]#b,$TY#b - xor ($dat,$TY),%al - ror \$8,%eax -___ -push(@TX,shift(@TX)); push(@XX,shift(@XX)); # "rotate" registers -} -for ($i=4;$i<8;$i++) { -$code.=<<___; - add $TX[0]#b,$YY#b - lea 1($XX[0]),$XX[1] - movzb ($dat,$YY),$TY#d - movzb $XX[1]#b,$XX[1]#d - movzb ($dat,$XX[1]),$TX[1]#d - movb $TX[0]#b,($dat,$YY) - cmp $XX[1],$YY - movb $TY#b,($dat,$XX[0]) - jne .Lcmov$i # Intel cmov is sloooow... - mov $TX[0],$TX[1] -.Lcmov$i: - add $TX[0]#b,$TY#b - xor ($dat,$TY),%bl - ror \$8,%ebx -___ -push(@TX,shift(@TX)); push(@XX,shift(@XX)); # "rotate" registers -} -$code.=<<___; - lea -8($len),$len - mov %eax,($out) - lea 8($inp),$inp - mov %ebx,4($out) - lea 8($out),$out - - test \$-8,$len - jnz .Lcloop8 - pop %rbx - cmp \$0,$len - jne .Lcloop1 -.Lexit: - sub \$1,$XX[0]#b - movb $XX[0]#b,-2($dat) - movb $YY#b,-1($dat) - - pop %r13 - pop %r12 - repret - -.align 16 -.Lcloop1: - add $TX[0]#b,$YY#b - movzb ($dat,$YY),$TY#d - movb $TX[0]#b,($dat,$YY) - movb $TY#b,($dat,$XX[0]) - add $TX[0]#b,$TY#b - add \$1,$XX[0]#b - movzb ($dat,$TY),$TY#d - movzb ($dat,$XX[0]),$TX[0]#d - xorb ($inp),$TY#b - lea 1($inp),$inp - movb $TY#b,($out) - lea 1($out),$out - sub \$1,$len - jnz .Lcloop1 - jmp .Lexit -.size RC4,.-RC4 -___ - -$code =~ s/#([bwd])/$1/gm; - -$code =~ s/repret/.byte\t0xF3,0xC3/gm; - -print $code; -- cgit v1.2.3-55-g6feb