diff options
Diffstat (limited to 'crypto/rc4/asm/rc4-md5-x86_64.pl')
| -rwxr-xr-x | crypto/rc4/asm/rc4-md5-x86_64.pl | 37 |
1 files changed, 33 insertions, 4 deletions
diff --git a/crypto/rc4/asm/rc4-md5-x86_64.pl b/crypto/rc4/asm/rc4-md5-x86_64.pl index 272fa91e1a1e..74e5191051eb 100755 --- a/crypto/rc4/asm/rc4-md5-x86_64.pl +++ b/crypto/rc4/asm/rc4-md5-x86_64.pl @@ -1,4 +1,11 @@ -#!/usr/bin/env perl +#! /usr/bin/env perl +# Copyright 2011-2016 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@openssl.org> for the OpenSSL @@ -29,16 +36,22 @@ # Core2 6.5 5.8 12.3 7.7 +60% # Westmere 4.3 5.2 9.5 7.0 +36% # Sandy Bridge 4.2 5.5 9.7 6.8 +43% +# Ivy Bridge 4.1 5.2 9.3 6.0 +54% +# Haswell 4.0 5.0 9.0 5.7 +60% +# Skylake 6.3(**) 5.0 11.3 5.3 +110% # Atom 9.3 6.5 15.8 11.1 +42% +# VIA Nano 6.3 5.4 11.7 8.6 +37% +# Bulldozer 4.5 5.4 9.9 7.7 +29% # # (*) rc4-x86_64.pl delivers 5.3 on Opteron, so real improvement # is +53%... +# (**) unidentified anomaly; my ($rc4,$md5)=(1,1); # what to generate? my $D="#" if (!$md5); # if set to "#", MD5 is stitched into RC4(), # but its result is discarded. Idea here is # to be able to use 'openssl speed rc4' for - # benchmarking the stitched subroutine... + # benchmarking the stitched subroutine... my $flavour = shift; my $output = shift; @@ -51,7 +64,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate; ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or die "can't locate x86_64-xlate.pl"; -open OUT,"| \"$^X\" $xlate $flavour $output"; +open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; *STDOUT=*OUT; my ($dat,$in0,$out,$ctx,$inp,$len, $func,$nargs); @@ -111,15 +124,23 @@ $code.=<<___; .globl $func .type $func,\@function,$nargs $func: +.cfi_startproc cmp \$0,$len je .Labort push %rbx +.cfi_push %rbx push %rbp +.cfi_push %rbp push %r12 +.cfi_push %r12 push %r13 +.cfi_push %r13 push %r14 +.cfi_push %r14 push %r15 +.cfi_push %r15 sub \$40,%rsp +.cfi_adjust_cfa_offset 40 .Lbody: ___ if ($rc4) { @@ -406,7 +427,7 @@ $code.=<<___ if ($rc4 && (!$md5 || $D)); and \$63,$len # remaining bytes jnz .Loop1 jmp .Ldone - + .align 16 .Loop1: add $TX[0]#b,$YY#b @@ -431,15 +452,23 @@ $code.=<<___; #rc4# movl $YY#d,-4($dat) mov 40(%rsp),%r15 +.cfi_restore %r15 mov 48(%rsp),%r14 +.cfi_restore %r14 mov 56(%rsp),%r13 +.cfi_restore %r13 mov 64(%rsp),%r12 +.cfi_restore %r12 mov 72(%rsp),%rbp +.cfi_restore %rbp mov 80(%rsp),%rbx +.cfi_restore %rbx lea 88(%rsp),%rsp +.cfi_adjust_cfa_offset -88 .Lepilogue: .Labort: ret +.cfi_endproc .size $func,.-$func ___ |