sha1-586.pl revision 9dc0df1bac950d6e491f9a7c7e4888f2b301cb15
# It was noted that Intel IA-32 C compiler generates code which
# performs ~30% *faster* on P4 CPU than original *hand-coded*
# SHA1 assembler implementation. To address this problem (and
# prove that humans are still better than machines:-), the
# original code was overhauled, which resulted in following
# performance changes:
#
# compared with original compared with Intel cc
# assembler impl. generated code
# Pentium -16% +48%
# P4 +85%(!) +45%
#
# As you can see Pentium came out as looser:-( Yet I reckoned that
# improvement on P4 outweights the loss and incorporate this
# re-tuned code to 0.9.7 and later.
# ----------------------------------------------------------------
# Those who for any particular reason absolutely must score on
# Pentium can replace this module with one from 0.9.6 distribution.
# This "offer" shall be revoked the moment programming interface to
# this module is changed, in which case this paragraph should be
# removed.
# ----------------------------------------------------------------
# <appro@fy.chalmers.se>
$normal=0;
require "x86asm.pl";
$A="eax";
$B="ecx";
$C="ebx";
$D="edx";
$E="edi";
$T="esi";
$tmp1="ebp";
@K=(0x5a827999,0x6ed9eba1,0x8f1bbcdc,0xca62c1d6);
&sha1_block_data("sha1_block_asm_data_order");
&asm_finish();
sub Nn
{
local($p)=@_;
local(%n)=($A,$T,$B,$A,$C,$B,$D,$C,$E,$D,$T,$E);
return($n{$p});
}
sub Np
{
local($p)=@_;
local(%n)=($A,$T,$B,$A,$C,$B,$D,$C,$E,$D,$T,$E);
local(%n)=($A,$B,$B,$C,$C,$D,$D,$E,$E,$T,$T,$A);
return($n{$p});
}
sub Na
{
local($n)=@_;
return( (($n )&0x0f),
(($n+ 2)&0x0f),
(($n+ 8)&0x0f),
(($n+13)&0x0f),
(($n+ 1)&0x0f));
}
sub X_expand
{
local($in)=@_;
&comment("First, load the words onto the stack in network byte order");
for ($i=0; $i<16; $i+=2)
{
&bswap($A);
&bswap($B);
}
&comment("We now have the X array on the stack");
&comment("starting at sp-4");
}
# Rules of engagement
# F is always trashable at the start, the running total.
# E becomes the next F so it can be trashed after it has been 'accumulated'
# F becomes A in the next round. We don't need to access it much.
# During the X update part, the result ends up in $X[$n0].
sub BODY_00_15
{
local($pos,$K,$X,$n,$a,$b,$c,$d,$e,$f)=@_;
&comment("00_15 $n");
&mov($f,$c); # f to hold F_00_19(b,c,d)
&xor($f,$d);
&and($f,$b);
# is loaded with xi
&xor($f,$d); # f holds F_00_19(b,c,d)
}
sub BODY_16_19
{
local($pos,$K,$X,$n,$a,$b,$c,$d,$e,$f)=@_;
&comment("16_19 $n");
&mov($e,$a); # e becomes volatile
&add($f,$e); # f+=ROTATE(a,5)
}
sub BODY_20_39
{
local($pos,$K,$X,$n,$a,$b,$c,$d,$e,$f)=@_;
&comment("20_39 $n");
&mov($e,$a); # e becomes volatile
&add($f,$e); # f+=ROTATE(a,5)
}
sub BODY_40_59
{
local($pos,$K,$X,$n,$a,$b,$c,$d,$e,$f)=@_;
&comment("40_59 $n");
&mov($e,$b); # e becomes volatile and is used
# to calculate F_40_59(b,c,d)
&and($e,$c);
&mov($e,$a);
&add($f,$e); # f+=ROTATE(a,5)
}
sub BODY_60_79
{
&BODY_20_39(@_);
}
sub sha1_block_host
{
# parameter 1 is the MD5_CTX structure.
# A 0
# B 4
# C 8
# D 12
# E 16
&push("esi");
&push("ebp");
&push("ebx");
&push("edi");
&comment("First we need to setup the X array");
for ($i=0; $i<16; $i+=2)
{
}
&function_end_B($name);
}
sub sha1_block_data
{
local($name)=@_;
# parameter 1 is the MD5_CTX structure.
# A 0
# B 4
# C 8
# D 12
# E 16
&push("esi");
&push("ebp");
&push("ebx");
&push("edi");
&comment("First we need to setup the X array");
&X_expand("esi");
&comment("");
&comment("Start processing");
# odd start
$X="esp";
&comment("End processing");
&comment("");
# D is the tmp value
# E -> A
# T -> B
# A -> C
# B -> D
# C -> E
# D -> T
&add($D,$B);
&add($B,$T);
&mov($T, $A);
&add($A,$E);
&add($E,$C);
&add($C,$T);
&pop("edi");
&pop("ebx");
&pop("ebp");
&pop("esi");
&ret();
# keep a note of shortcut label so it can be used outside
# block.
&function_end_B($name);
# Putting this here avoids problems with MASM in debugging mode
}