Hakase/openssl-custom
1
0 Fork 0
Code Issues Pull Requests Releases Wiki Activity

Latest update.

Browse Source
This commit is contained in:
Hakase
2019-09-21 00:43:47 +09:00
parent 2e57f602ae
commit 62515c7d8d
1131 changed files with 47556 additions and 24957 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crypto/bn/asm/alpha-mont.pl
+1 -2
View File
@@ -22,8 +22,7 @@
# I.e. if you compare 1GHz 21264 and 2GHz Opteron, you'll observe ~2x
# difference.
$output=pop;
open STDOUT,">$output";
$output=pop and open STDOUT,">$output";
# int bn_mul_mont(
$rp="a0"; # BN_ULONG *rp,
crypto/bn/asm/armv4-gf2m.pl
+7 -5
View File
@@ -39,9 +39,10 @@
#
# http://conradoplg.cryptoland.net/files/2010/12/mocrysen13.pdf
$flavour = shift;
if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour && $flavour ne "void") {
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
@@ -49,9 +50,10 @@ if ($flavour && $flavour ne "void") {
( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
die "can't locate arm-xlate.pl";
open STDOUT,"| \"$^X\" $xlate $flavour $output";
open STDOUT,"| \"$^X\" $xlate $flavour \"$output\""
or die "can't call $xlate: $1";
} else {
open STDOUT,">$output";
$output and open STDOUT,">$output";
}
$code=<<___;
crypto/bn/asm/armv4-mont.pl
+7 -5
View File
@@ -54,9 +54,10 @@
# integer-only on Cortex-A8, ~10-210% on Cortex-A15, ~70-450% on
# Snapdragon S4.
$flavour = shift;
if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
my $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
my $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour && $flavour ne "void") {
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
@@ -64,9 +65,10 @@ if ($flavour && $flavour ne "void") {
( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
die "can't locate arm-xlate.pl";
open STDOUT,"| \"$^X\" $xlate $flavour $output";
open STDOUT,"| \"$^X\" $xlate $flavour \"$output\""
or die "can't call $xlate: $1";
} else {
open STDOUT,">$output";
$output and open STDOUT,">$output";
}
$num="r0"; # starts as num argument, but holds &tp[num-1]
crypto/bn/asm/armv8-mont.pl
+6 -3
View File
@@ -40,15 +40,18 @@
# 50-70% improvement for RSA4096 sign. RSA2048 sign is ~25% faster
# on Cortex-A57 and ~60-100% faster on others.
$flavour = shift;
$output = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
my $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
my $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
die "can't locate arm-xlate.pl";
open OUT,"| \"$^X\" $xlate $flavour $output";
open OUT,"| \"$^X\" $xlate $flavour \"$output\""
or die "can't call $xlate: $1";
*STDOUT=*OUT;
($lo0,$hi0,$aj,$m0,$alo,$ahi,
crypto/bn/asm/bn-586.pl
+1 -2
View File
@@ -11,8 +11,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
&asm_init($ARGV[0]);
crypto/bn/asm/c64xplus-gf2m.pl
+1 -2
View File
@@ -23,8 +23,7 @@
# totally unfair, because this module utilizes Galois Field Multiply
# instruction.
while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
($rp,$a1,$a0,$b1,$b0)=("A4","B4","A6","B6","A8"); # argument vector
crypto/bn/asm/co-586.pl
+1 -2
View File
@@ -10,8 +10,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
&asm_init($ARGV[0]);
crypto/bn/asm/ia64-mont.pl
+2 -1
View File
@@ -67,7 +67,8 @@
# hereafter less for longer keys, while verify - by 74-13%.
# DSA performance improves by 115-30%.
$output=pop;
# $output is the last argument if it looks like a file (it has an extension)
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
if ($^O eq "hpux") {
$ADDP="addp4";
crypto/bn/asm/mips-mont.pl
+7 -4
View File
@@ -52,8 +52,12 @@
# ($t0,$t1,$t2,$t3,$t8,$t9)=map("\$$_",(12..15,24,25));
# ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("\$$_",(16..23));
# ($gp,$sp,$fp,$ra)=map("\$$_",(28..31));
#
$flavour = shift || "o32"; # supported flavours are o32,n32,64,nubi32,nubi64
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
# supported flavours are o32,n32,64,nubi32,nubi64, default is o32
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : "o32";
if ($flavour =~ /64|n32/i) {
$PTR_ADD="daddu"; # incidentally works even on n32
@@ -74,8 +78,7 @@ $SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0x00fff000 : 0x00ff0000;
#
######################################################################
while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
$output and open STDOUT,">$output";
if ($flavour =~ /64|n32/i) {
$LD="ld";
crypto/bn/asm/mips.pl
+6 -3
View File
@@ -54,9 +54,10 @@
# has to content with 40-85% improvement depending on benchmark and
# key length, more for longer keys.
$flavour = shift || "o32";
while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : "o32";
if ($flavour =~ /64|n32/i) {
$LD="ld";
@@ -91,6 +92,8 @@ if ($flavour =~ /64|n32/i) {
$code="#if !(defined (__mips_isa_rev) && (__mips_isa_rev >= 6))\n.set mips2\n#endif\n";
}
$output and open STDOUT,">$output";
# Below is N32/64 register layout used in the original module.
#
($zero,$at,$v0,$v1)=map("\$$_",(0..3));
crypto/bn/asm/parisc-mont.pl
+5 -3
View File
@@ -69,10 +69,12 @@
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
$flavour = shift;
$output = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
open STDOUT,">$output";
$output and open STDOUT,">$output";
if ($flavour =~ /64/) {
$LEVEL ="2.0W";
crypto/bn/asm/ppc-mont.pl
+6 -2
View File
@@ -41,7 +41,10 @@
# builds. On low-end 32-bit processors performance improvement turned
# to be marginal...
$flavour = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour =~ /32/) {
$BITS= 32;
@@ -94,7 +97,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
die "can't locate ppc-xlate.pl";
open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
open STDOUT,"| $^X $xlate $flavour \"$output\""
or die "can't call $xlate: $!";
$sp="r1";
$toc="r2";
crypto/bn/asm/ppc.pl
+6 -2
View File
@@ -103,7 +103,10 @@
# Performance increase of ~60%
# Based on submission from Suresh N. Chari of IBM
$flavour = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour =~ /32/) {
$BITS= 32;
@@ -159,7 +162,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
die "can't locate ppc-xlate.pl";
open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
open STDOUT,"| $^X $xlate $flavour \"$output\""
or die "can't call $xlate: $!";
$data=<<EOF;
#--------------------------------------------------------------------
crypto/bn/asm/ppc64-mont.pl
+6 -2
View File
@@ -80,7 +80,10 @@
# ppc-mont.pl, but improvement coefficient is not as impressive
# for longer keys...
$flavour = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour =~ /32/) {
$SIZE_T=4;
@@ -108,7 +111,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
die "can't locate ppc-xlate.pl";
open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
open STDOUT,"| $^X $xlate $flavour \"$output\""
or die "can't call $xlate: $!";
$FRAME=64; # padded frame header
$TRANSFER=16*8;
crypto/bn/asm/rsaz-avx2.pl
+6 -4
View File
@@ -37,9 +37,10 @@
# (***) scalar AD*X code is faster than AVX2 and is preferred code
# path for Broadwell;
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
@@ -72,7 +73,8 @@ if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|based on LLVM) ([3-
$addx = ($ver>=3.03);
}
open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
or die "can't call $xlate: $!";
*STDOUT = *OUT;
if ($avx>1) {{{
crypto/bn/asm/rsaz-x86_64.pl
+6 -4
View File
@@ -52,9 +52,10 @@
# purposes;
# (**) MULX was attempted, but found to give only marginal improvement;
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
@@ -63,7 +64,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${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\""
or die "can't call $xlate: $!";
*STDOUT=*OUT;
if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
crypto/bn/asm/s390x-gf2m.pl
+5 -3
View File
@@ -32,7 +32,10 @@
# so that improvement coefficients can vary from one specific
# setup to another.
$flavour = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour =~ /3[12]/) {
$SIZE_T=4;
@@ -42,8 +45,7 @@ if ($flavour =~ /3[12]/) {
$g="g";
}
while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
$output and open STDOUT,">$output";
$stdframe=16*$SIZE_T+4*8;
crypto/bn/asm/s390x-mont.pl
+5 -3
View File
@@ -51,7 +51,10 @@
# On z990 it was measured to perform 2.6-2.2 times better than
# compiler-generated code, less for longer keys...
$flavour = shift;
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
if ($flavour =~ /3[12]/) {
$SIZE_T=4;
@@ -61,8 +64,7 @@ if ($flavour =~ /3[12]/) {
$g="g";
}
while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
$output and open STDOUT,">$output";
$stdframe=16*$SIZE_T+4*8;
crypto/bn/asm/sparct4-mont.pl
+1 -2
View File
@@ -83,8 +83,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "sparcv9_modes.pl";
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
$code.=<<___;
#include "sparc_arch.h"
crypto/bn/asm/sparcv9-gf2m.pl
+1 -2
View File
@@ -25,8 +25,7 @@
# ~100-230% faster than gcc-generated code and ~35-90% faster than
# the pure SPARCv9 code path.
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
$locals=16*8;
crypto/bn/asm/sparcv9-mont.pl
+1 -2
View File
@@ -49,8 +49,7 @@
# module still have hidden potential [see TODO list there], which is
# estimated to be larger than 20%...
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
# int bn_mul_mont(
$rp="%i0"; # BN_ULONG *rp,
crypto/bn/asm/sparcv9a-mont.pl
+4 -2
View File
@@ -62,8 +62,10 @@
# key length, more for longer keys] on USI&II cores and 30-80% - on
# USIII&IV.
$output = pop;
open STDOUT,">$output";
# $output is the last argument if it looks like a file (it has an extension)
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$output and open STDOUT,">$output";
$fname="bn_mul_mont_fpu";
crypto/bn/asm/via-mont.pl
+1 -2
View File
@@ -88,8 +88,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
&asm_init($ARGV[0]);
crypto/bn/asm/vis3-mont.pl
+1 -2
View File
@@ -25,8 +25,7 @@
# for reference purposes, because T4 has dedicated Montgomery
# multiplication and squaring *instructions* that deliver even more.
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
$frame = "STACK_FRAME";
$bias = "STACK_BIAS";
crypto/bn/asm/x86-gf2m.pl
+1 -2
View File
@@ -43,8 +43,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
&asm_init($ARGV[0],$x86only = $ARGV[$#ARGV] eq "386");
crypto/bn/asm/x86-mont.pl
+1 -2
View File
@@ -37,8 +37,7 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
push(@INC,"${dir}","${dir}../../perlasm");
require "x86asm.pl";
$output = pop;
open STDOUT,">$output";
$output = pop and open STDOUT,">$output";
&asm_init($ARGV[0]);
crypto/bn/asm/x86_64-gf2m.pl
+6 -4
View File
@@ -27,9 +27,10 @@
# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
# all CPU time is burnt in it...
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
@@ -38,7 +39,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${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\""
or die "can't call $xlate: $!";
*STDOUT=*OUT;
($lo,$hi)=("%rax","%rdx"); $a=$lo;
crypto/bn/asm/x86_64-mont.pl
+6 -4
View File
@@ -46,9 +46,10 @@
#
# Add MULX/ADOX/ADCX code path.
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
@@ -57,7 +58,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${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\""
or die "can't call $xlate: $!";
*STDOUT=*OUT;
if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
crypto/bn/asm/x86_64-mont5.pl
+6 -4
View File
@@ -31,9 +31,10 @@
# the np argument is not just modulus value, but one interleaved
# with 0. This is to optimize post-condition...
$flavour = shift;
$output = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
# $output is the last argument if it looks like a file (it has an extension)
# $flavour is the first argument if it doesn't look like a file
$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
@@ -42,7 +43,8 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${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\""
or die "can't call $xlate: $!";
*STDOUT=*OUT;
if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`