-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License"). You may not use
+# this file except in compliance with the License. You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
#
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# table]. I stick to value of 2 for two reasons: 1. smaller table
# minimizes cache trashing and thus mitigates the hazard of side-
# channel leakage similar to AES cache-timing one; 2. performance
-# gap among different µ-archs is smaller.
+# gap among different µ-archs is smaller.
#
# Performance table lists rounded amounts of CPU cycles spent by
# whirlpool_block_mmx routine on single 64 byte input block, i.e.
# multiplying 64 by CPU clock frequency and dividing by relevant
# value from the given table:
#
-# $SCALE=2/8 icc8 gcc3
+# $SCALE=2/8 icc8 gcc3
# Intel P4 3200/4600 4600(*) 6400
# Intel PIII 2900/3000 4900 5400
# AMD K[78] 2500/1800 9900 8200(**)
#
# (*) I've sketched even non-MMX assembler, but for the record
# I've failed to beat the Intel compiler on P4, without using
-* MMX that is...
+# MMX that is...
# (**) ... on AMD on the other hand non-MMX assembler was observed
# to perform significantly better, but I figured this MMX
# implementation is even faster anyway, so why bother? As for
# non-MMX implementation would actually pay off, but till
# opposite is proved "unlikely" is assumed.
-push(@INC,"../../perlasm");
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
-&asm_init($ARGV[0],"wp-mmx.pl");
+$output=pop;
+open STDOUT,">$output";
+
+&asm_init($ARGV[0]);
sub L() { &data_byte(@_); }
sub LL()
for($i=0;$i<8;$i++) { &movq(@mm[$i],&QWP($i*8,"esi")); } # L=H
&set_label("outerloop");
- for($i=0;$i<8;$i++) { &movq(&QWP($i*8,"esp"),@mm[$i]); } # K=H
+ for($i=0;$i<8;$i++) { &movq(&QWP($i*8,"esp"),@mm[$i]); } # K=L
for($i=0;$i<8;$i++) { &pxor(@mm[$i],&QWP($i*8,"edi")); } # L^=inp
- for($i=0;$i<8;$i++) { &movq(&QWP(64+$i*8,"esp"),@mm[$i]); } #S=L
+ for($i=0;$i<8;$i++) { &movq(&QWP(64+$i*8,"esp"),@mm[$i]); } # S=L
&xor ("esi","esi");
&mov (&DWP(12,"ebx"),"esi"); # zero round counter
&set_label("round",16);
- &movq (@mm[0],&DWP(2048*$SCALE,$tbl,"esi",8)); # rc[r]
+ &movq (@mm[0],&QWP(2048*$SCALE,$tbl,"esi",8)); # rc[r]
&mov ("eax",&DWP(0,"esp"));
&mov ("ebx",&DWP(4,"esp"));
+ &movz ("ecx",&LB("eax"));
+ &movz ("edx",&HB("eax"));
for($i=0;$i<8;$i++) {
- my $func = ($i==0)? movq : pxor;
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
+ my $func = ($i==0)? \&movq : \&pxor;
+ &shr ("eax",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
- &shr ("eax",16);
+ &movz ("edx",&HB("eax"));
&pxor (@mm[0],&QWP(&row(0),$tbl,"esi",8));
&$func (@mm[1],&QWP(&row(1),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
&mov ("eax",&DWP(($i+1)*8,"esp"));
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
+ &movz ("edx",&HB("ebx"));
&$func (@mm[2],&QWP(&row(2),$tbl,"esi",8));
&$func (@mm[3],&QWP(&row(3),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
+ &shr ("ebx",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
- &shr ("ebx",16);
+ &movz ("edx",&HB("ebx"));
&$func (@mm[4],&QWP(&row(4),$tbl,"esi",8));
&$func (@mm[5],&QWP(&row(5),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
&mov ("ebx",&DWP(($i+1)*8+4,"esp"));
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
+ &movz ("edx",&HB("eax"));
&$func (@mm[6],&QWP(&row(6),$tbl,"esi",8));
&$func (@mm[7],&QWP(&row(7),$tbl,"edi",8));
push(@mm,shift(@mm));
for($i=0;$i<8;$i++) { &movq(&QWP($i*8,"esp"),@mm[$i]); } # K=L
for($i=0;$i<8;$i++) {
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
+ &shr ("eax",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
- &shr ("eax",16);
+ &movz ("edx",&HB("eax"));
&pxor (@mm[0],&QWP(&row(0),$tbl,"esi",8));
&pxor (@mm[1],&QWP(&row(1),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("eax"));
- &movb (&LB("edx"),&HB("eax"));
&mov ("eax",&DWP(64+($i+1)*8,"esp")) if ($i<7);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
+ &movz ("edx",&HB("ebx"));
&pxor (@mm[2],&QWP(&row(2),$tbl,"esi",8));
&pxor (@mm[3],&QWP(&row(3),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
+ &shr ("ebx",16);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("ebx"));
&scale ("edi","edx");
- &shr ("ebx",16);
+ &movz ("edx",&HB("ebx"));
&pxor (@mm[4],&QWP(&row(4),$tbl,"esi",8));
&pxor (@mm[5],&QWP(&row(5),$tbl,"edi",8));
- &movb (&LB("ecx"),&LB("ebx"));
- &movb (&LB("edx"),&HB("ebx"));
&mov ("ebx",&DWP(64+($i+1)*8+4,"esp")) if ($i<7);
&scale ("esi","ecx");
+ &movz ("ecx",&LB("eax"));
&scale ("edi","edx");
+ &movz ("edx",&HB("eax"));
&pxor (@mm[6],&QWP(&row(6),$tbl,"esi",8));
&pxor (@mm[7],&QWP(&row(7),$tbl,"edi",8));
push(@mm,shift(@mm));
&L(0xfb,0xee,0x7c,0x66,0xdd,0x17,0x47,0x9e);
&L(0xca,0x2d,0xbf,0x07,0xad,0x5a,0x83,0x33);
-&function_end_B("whrilpool_block_mmx");
-&asm_finish();
+&function_end_B("whirlpool_block_mmx");
+&asm_finish();
+
+close STDOUT;
projects / openssl.git / blobdiff