generic-library/openssl
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

MIPS assembler pack update from HEAD.

Browse Source
This commit is contained in:
Andy Polyakov 2011-11-14 20:55:24 +00:00
parent cf96d71c22
commit b66723b23e
5 changed files with 5431 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1611
crypto/aes/asm/aes-mips.pl Normal file
View File

File diff suppressed because it is too large Load Diff

426
crypto/bn/asm/mips-mont.pl Normal file
View File

@ -0,0 +1,426 @@
#!/usr/bin/env perl
#
# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
# This module doesn't present direct interest for OpenSSL, because it
# doesn't provide better performance for longer keys, at least not on
# in-order-execution cores. While 512-bit RSA sign operations can be
# 65% faster in 64-bit mode, 1024-bit ones are only 15% faster, and
# 4096-bit ones are up to 15% slower. In 32-bit mode it varies from
# 16% improvement for 512-bit RSA sign to -33% for 4096-bit RSA
# verify:-( All comparisons are against bn_mul_mont-free assembler.
# The module might be of interest to embedded system developers, as
# the code is smaller than 1KB, yet offers >3x improvement on MIPS64
# and 75-30% [less for longer keys] on MIPS32 over compiler-generated
# code.
######################################################################
# There is a number of MIPS ABI in use, O32 and N32/64 are most
# widely used. Then there is a new contender: NUBI. It appears that if
# one picks the latter, it's possible to arrange code in ABI neutral
# manner. Therefore let's stick to NUBI register layout:
#
($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
#
# The return value is placed in $a0. Following coding rules facilitate
# interoperability:
#
# - never ever touch $tp, "thread pointer", former $gp;
# - copy return value to $t0, former $v0 [or to $a0 if you're adapting
# old code];
# - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
#
# For reference here is register layout for N32/64 MIPS ABIs:
#
# ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
# ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
# ($t0,$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));
#
$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
if ($flavour =~ /64|n32/i) {
$PTR_ADD="dadd"; # incidentally works even on n32
$PTR_SUB="dsub"; # incidentally works even on n32
$REG_S="sd";
$REG_L="ld";
$SZREG=8;
} else {
$PTR_ADD="add";
$PTR_SUB="sub";
$REG_S="sw";
$REG_L="lw";
$SZREG=4;
}
$SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0x00fff000 : 0x00ff0000;
#
# <appro@openssl.org>
#
######################################################################
while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
open STDOUT,">$output";
if ($flavour =~ /64|n32/i) {
$LD="ld";
$ST="sd";
$MULTU="dmultu";
$ADDU="daddu";
$SUBU="dsubu";
$BNSZ=8;
} else {
$LD="lw";
$ST="sw";
$MULTU="multu";
$ADDU="addu";
$SUBU="subu";
$BNSZ=4;
}
# int bn_mul_mont(
$rp=$a0; # BN_ULONG *rp,
$ap=$a1; # const BN_ULONG *ap,
$bp=$a2; # const BN_ULONG *bp,
$np=$a3; # const BN_ULONG *np,
$n0=$a4; # const BN_ULONG *n0,
$num=$a5; # int num);
$lo0=$a6;
$hi0=$a7;
$lo1=$t1;
$hi1=$t2;
$aj=$s0;
$bi=$s1;
$nj=$s2;
$tp=$s3;
$alo=$s4;
$ahi=$s5;
$nlo=$s6;
$nhi=$s7;
$tj=$s8;
$i=$s9;
$j=$s10;
$m1=$s11;
$FRAMESIZE=14;
$code=<<___;
.text
.set noat
.set noreorder
.align 5
.globl bn_mul_mont
.ent bn_mul_mont
bn_mul_mont:
___
$code.=<<___ if ($flavour =~ /o32/i);
lw $n0,16($sp)
lw $num,20($sp)
___
$code.=<<___;
slt $at,$num,4
bnez $at,1f
li $t0,0
slt $at,$num,17 # on in-order CPU
bnezl $at,bn_mul_mont_internal
nop
1: jr $ra
li $a0,0
.end bn_mul_mont
.align 5
.ent bn_mul_mont_internal
bn_mul_mont_internal:
.frame $fp,$FRAMESIZE*$SZREG,$ra
.mask 0x40000000|$SAVED_REGS_MASK,-$SZREG
$PTR_SUB $sp,$FRAMESIZE*$SZREG
$REG_S $fp,($FRAMESIZE-1)*$SZREG($sp)
$REG_S $s11,($FRAMESIZE-2)*$SZREG($sp)
$REG_S $s10,($FRAMESIZE-3)*$SZREG($sp)
$REG_S $s9,($FRAMESIZE-4)*$SZREG($sp)
$REG_S $s8,($FRAMESIZE-5)*$SZREG($sp)
$REG_S $s7,($FRAMESIZE-6)*$SZREG($sp)
$REG_S $s6,($FRAMESIZE-7)*$SZREG($sp)
$REG_S $s5,($FRAMESIZE-8)*$SZREG($sp)
$REG_S $s4,($FRAMESIZE-9)*$SZREG($sp)
___
$code.=<<___ if ($flavour =~ /nubi/i);
$REG_S $s3,($FRAMESIZE-10)*$SZREG($sp)
$REG_S $s2,($FRAMESIZE-11)*$SZREG($sp)
$REG_S $s1,($FRAMESIZE-12)*$SZREG($sp)
$REG_S $s0,($FRAMESIZE-13)*$SZREG($sp)
___
$code.=<<___;
move $fp,$sp
.set reorder
$LD $n0,0($n0)
$LD $bi,0($bp) # bp[0]
$LD $aj,0($ap) # ap[0]
$LD $nj,0($np) # np[0]
$PTR_SUB $sp,2*$BNSZ # place for two extra words
sll $num,`log($BNSZ)/log(2)`
li $at,-4096
$PTR_SUB $sp,$num
and $sp,$at
$MULTU $aj,$bi
$LD $alo,$BNSZ($ap)
$LD $nlo,$BNSZ($np)
mflo $lo0
mfhi $hi0
$MULTU $lo0,$n0
mflo $m1
$MULTU $alo,$bi
mflo $alo
mfhi $ahi
$MULTU $nj,$m1
mflo $lo1
mfhi $hi1
$MULTU $nlo,$m1
$ADDU $lo1,$lo0
sltu $at,$lo1,$lo0
$ADDU $hi1,$at
mflo $nlo
mfhi $nhi
move $tp,$sp
li $j,2*$BNSZ
.align 4
.L1st:
.set noreorder
$PTR_ADD $aj,$ap,$j
$PTR_ADD $nj,$np,$j
$LD $aj,($aj)
$LD $nj,($nj)
$MULTU $aj,$bi
$ADDU $lo0,$alo,$hi0
$ADDU $lo1,$nlo,$hi1
sltu $at,$lo0,$hi0
sltu $t0,$lo1,$hi1
$ADDU $hi0,$ahi,$at
$ADDU $hi1,$nhi,$t0
mflo $alo
mfhi $ahi
$ADDU $lo1,$lo0
sltu $at,$lo1,$lo0
$MULTU $nj,$m1
$ADDU $hi1,$at
addu $j,$BNSZ
$ST $lo1,($tp)
sltu $t0,$j,$num
mflo $nlo
mfhi $nhi
bnez $t0,.L1st
$PTR_ADD $tp,$BNSZ
.set reorder
$ADDU $lo0,$alo,$hi0
sltu $at,$lo0,$hi0
$ADDU $hi0,$ahi,$at
$ADDU $lo1,$nlo,$hi1
sltu $t0,$lo1,$hi1
$ADDU $hi1,$nhi,$t0
$ADDU $lo1,$lo0
sltu $at,$lo1,$lo0
$ADDU $hi1,$at
$ST $lo1,($tp)
$ADDU $hi1,$hi0
sltu $at,$hi1,$hi0
$ST $hi1,$BNSZ($tp)
$ST $at,2*$BNSZ($tp)
li $i,$BNSZ
.align 4
.Louter:
$PTR_ADD $bi,$bp,$i
$LD $bi,($bi)
$LD $aj,($ap)
$LD $alo,$BNSZ($ap)
$LD $tj,($sp)
$MULTU $aj,$bi
$LD $nj,($np)
$LD $nlo,$BNSZ($np)
mflo $lo0
mfhi $hi0
$ADDU $lo0,$tj
$MULTU $lo0,$n0
sltu $at,$lo0,$tj
$ADDU $hi0,$at
mflo $m1
$MULTU $alo,$bi
mflo $alo
mfhi $ahi
$MULTU $nj,$m1
mflo $lo1
mfhi $hi1
$MULTU $nlo,$m1
$ADDU $lo1,$lo0
sltu $at,$lo1,$lo0
$ADDU $hi1,$at
mflo $nlo
mfhi $nhi
move $tp,$sp
li $j,2*$BNSZ
$LD $tj,$BNSZ($tp)
.align 4
.Linner:
.set noreorder
$PTR_ADD $aj,$ap,$j
$PTR_ADD $nj,$np,$j
$LD $aj,($aj)
$LD $nj,($nj)
$MULTU $aj,$bi
$ADDU $lo0,$alo,$hi0
$ADDU $lo1,$nlo,$hi1
sltu $at,$lo0,$hi0
sltu $t0,$lo1,$hi1
$ADDU $hi0,$ahi,$at
$ADDU $hi1,$nhi,$t0
mflo $alo
mfhi $ahi
$ADDU $lo0,$tj
addu $j,$BNSZ
$MULTU $nj,$m1
sltu $at,$lo0,$tj
$ADDU $lo1,$lo0
$ADDU $hi0,$at
sltu $t0,$lo1,$lo0
$LD $tj,2*$BNSZ($tp)
$ADDU $hi1,$t0
sltu $at,$j,$num
mflo $nlo
mfhi $nhi
$ST $lo1,($tp)
bnez $at,.Linner
$PTR_ADD $tp,$BNSZ
.set reorder
$ADDU $lo0,$alo,$hi0
sltu $at,$lo0,$hi0
$ADDU $hi0,$ahi,$at
$ADDU $lo0,$tj
sltu $t0,$lo0,$tj
$ADDU $hi0,$t0
$LD $tj,2*$BNSZ($tp)
$ADDU $lo1,$nlo,$hi1
sltu $at,$lo1,$hi1
$ADDU $hi1,$nhi,$at
$ADDU $lo1,$lo0
sltu $t0,$lo1,$lo0
$ADDU $hi1,$t0
$ST $lo1,($tp)
$ADDU $lo1,$hi1,$hi0
sltu $hi1,$lo1,$hi0
$ADDU $lo1,$tj
sltu $at,$lo1,$tj
$ADDU $hi1,$at
$ST $lo1,$BNSZ($tp)
$ST $hi1,2*$BNSZ($tp)
addu $i,$BNSZ
sltu $t0,$i,$num
bnez $t0,.Louter
.set noreorder
$PTR_ADD $tj,$sp,$num # &tp[num]
move $tp,$sp
move $ap,$sp
li $hi0,0 # clear borrow bit
.align 4
.Lsub: $LD $lo0,($tp)
$LD $lo1,($np)
$PTR_ADD $tp,$BNSZ
$PTR_ADD $np,$BNSZ
$SUBU $lo1,$lo0,$lo1 # tp[i]-np[i]
sgtu $at,$lo1,$lo0
$SUBU $lo0,$lo1,$hi0
sgtu $hi0,$lo0,$lo1
$ST $lo0,($rp)
or $hi0,$at
sltu $at,$tp,$tj
bnez $at,.Lsub
$PTR_ADD $rp,$BNSZ
$SUBU $hi0,$hi1,$hi0 # handle upmost overflow bit
move $tp,$sp
$PTR_SUB $rp,$num # restore rp
not $hi1,$hi0
and $ap,$hi0,$sp
and $bp,$hi1,$rp
or $ap,$ap,$bp # ap=borrow?tp:rp
.align 4
.Lcopy: $LD $aj,($ap)
$PTR_ADD $ap,$BNSZ
$ST $zero,($tp)
$PTR_ADD $tp,$BNSZ
sltu $at,$tp,$tj
$ST $aj,($rp)
bnez $at,.Lcopy
$PTR_ADD $rp,$BNSZ
li $a0,1
li $t0,1
.set noreorder
move $sp,$fp
$REG_L $fp,($FRAMESIZE-1)*$SZREG($sp)
$REG_L $s11,($FRAMESIZE-2)*$SZREG($sp)
$REG_L $s10,($FRAMESIZE-3)*$SZREG($sp)
$REG_L $s9,($FRAMESIZE-4)*$SZREG($sp)
$REG_L $s8,($FRAMESIZE-5)*$SZREG($sp)
$REG_L $s7,($FRAMESIZE-6)*$SZREG($sp)
$REG_L $s6,($FRAMESIZE-7)*$SZREG($sp)
$REG_L $s5,($FRAMESIZE-8)*$SZREG($sp)
$REG_L $s4,($FRAMESIZE-9)*$SZREG($sp)
___
$code.=<<___ if ($flavour =~ /nubi/i);
$REG_L $s3,($FRAMESIZE-10)*$SZREG($sp)
$REG_L $s2,($FRAMESIZE-11)*$SZREG($sp)
$REG_L $s1,($FRAMESIZE-12)*$SZREG($sp)
$REG_L $s0,($FRAMESIZE-13)*$SZREG($sp)
___
$code.=<<___;
jr $ra
$PTR_ADD $sp,$FRAMESIZE*$SZREG
.end bn_mul_mont_internal
.rdata
.asciiz "Montgomery Multiplication for MIPS, CRYPTOGAMS by <appro\@openssl.org>"
___
$code =~ s/\`([^\`]*)\`/eval $1/gem;
print $code;
close STDOUT;

2585
crypto/bn/asm/mips.pl Normal file
View File

File diff suppressed because it is too large Load Diff

354
crypto/sha/asm/sha1-mips.pl Normal file
View File

@ -0,0 +1,354 @@
#!/usr/bin/env perl
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
# SHA1 block procedure for MIPS.
# Performance improvement is 30% on unaligned input. The "secret" is
# to deploy lwl/lwr pair to load unaligned input. One could have
# vectorized Xupdate on MIPSIII/IV, but the goal was to code MIPS32-
# compatible subroutine. There is room for minor optimization on
# little-endian platforms...
######################################################################
# There is a number of MIPS ABI in use, O32 and N32/64 are most
# widely used. Then there is a new contender: NUBI. It appears that if
# one picks the latter, it's possible to arrange code in ABI neutral
# manner. Therefore let's stick to NUBI register layout:
#
($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
#
# The return value is placed in $a0. Following coding rules facilitate
# interoperability:
#
# - never ever touch $tp, "thread pointer", former $gp;
# - copy return value to $t0, former $v0 [or to $a0 if you're adapting
# old code];
# - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
#
# For reference here is register layout for N32/64 MIPS ABIs:
#
# ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
# ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
# ($t0,$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));
#
$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
if ($flavour =~ /64|n32/i) {
$PTR_ADD="dadd"; # incidentally works even on n32
$PTR_SUB="dsub"; # incidentally works even on n32
$REG_S="sd";
$REG_L="ld";
$PTR_SLL="dsll"; # incidentally works even on n32
$SZREG=8;
} else {
$PTR_ADD="add";
$PTR_SUB="sub";
$REG_S="sw";
$REG_L="lw";
$PTR_SLL="sll";
$SZREG=4;
}
#
# <appro@openssl.org>
#
######################################################################
$big_endian=(`echo MIPSEL | $ENV{CC} -E -P -`=~/MIPSEL/)?1:0;
for (@ARGV) { $output=$_ if (/^\w[\w\-]*\.\w+$/); }
open STDOUT,">$output";
if (!defined($big_endian))
{ $big_endian=(unpack('L',pack('N',1))==1); }
# offsets of the Most and Least Significant Bytes
$MSB=$big_endian?0:3;
$LSB=3&~$MSB;
@X=map("\$$_",(8..23)); # a4-a7,s0-s11
$ctx=$a0;
$inp=$a1;
$num=$a2;
$A="\$1";
$B="\$2";
$C="\$3";
$D="\$7";
$E="\$24"; @V=($A,$B,$C,$D,$E);
$t0="\$25";
$t1=$num; # $num is offloaded to stack
$t2="\$30"; # fp
$K="\$31"; # ra
sub BODY_00_14 {
my ($i,$a,$b,$c,$d,$e)=@_;
my $j=$i+1;
$code.=<<___ if (!$big_endian);
srl $t0,@X[$i],24 # byte swap($i)
srl $t1,@X[$i],8
andi $t2,@X[$i],0xFF00
sll @X[$i],@X[$i],24
andi $t1,0xFF00
sll $t2,$t2,8
or @X[$i],$t0
or $t1,$t2
or @X[$i],$t1
___
$code.=<<___;
lwl @X[$j],$j*4+$MSB($inp)
sll $t0,$a,5 # $i
addu $e,$K
lwr @X[$j],$j*4+$LSB($inp)
srl $t1,$a,27
addu $e,$t0
xor $t0,$c,$d
addu $e,$t1
sll $t2,$b,30
and $t0,$b
srl $b,$b,2
xor $t0,$d
addu $e,@X[$i]
or $b,$t2
addu $e,$t0
___
}
sub BODY_15_19 {
my ($i,$a,$b,$c,$d,$e)=@_;
my $j=$i+1;
$code.=<<___ if (!$big_endian && $i==15);
srl $t0,@X[$i],24 # byte swap($i)
srl $t1,@X[$i],8
andi $t2,@X[$i],0xFF00
sll @X[$i],@X[$i],24
andi $t1,0xFF00
sll $t2,$t2,8
or @X[$i],$t0
or @X[$i],$t1
or @X[$i],$t2
___
$code.=<<___;
xor @X[$j%16],@X[($j+2)%16]
sll $t0,$a,5 # $i
addu $e,$K
srl $t1,$a,27
addu $e,$t0
xor @X[$j%16],@X[($j+8)%16]
xor $t0,$c,$d
addu $e,$t1
xor @X[$j%16],@X[($j+13)%16]
sll $t2,$b,30
and $t0,$b
srl $t1,@X[$j%16],31
addu @X[$j%16],@X[$j%16]
srl $b,$b,2
xor $t0,$d
or @X[$j%16],$t1
addu $e,@X[$i%16]
or $b,$t2
addu $e,$t0
___
}
sub BODY_20_39 {
my ($i,$a,$b,$c,$d,$e)=@_;
my $j=$i+1;
$code.=<<___ if ($i<79);
xor @X[$j%16],@X[($j+2)%16]
sll $t0,$a,5 # $i
addu $e,$K
srl $t1,$a,27
addu $e,$t0
xor @X[$j%16],@X[($j+8)%16]
xor $t0,$c,$d
addu $e,$t1
xor @X[$j%16],@X[($j+13)%16]
sll $t2,$b,30
xor $t0,$b
srl $t1,@X[$j%16],31
addu @X[$j%16],@X[$j%16]
srl $b,$b,2
addu $e,@X[$i%16]
or @X[$j%16],$t1
or $b,$t2
addu $e,$t0
___
$code.=<<___ if ($i==79);
lw @X[0],0($ctx)
sll $t0,$a,5 # $i
addu $e,$K
lw @X[1],4($ctx)
srl $t1,$a,27
addu $e,$t0
lw @X[2],8($ctx)
xor $t0,$c,$d
addu $e,$t1
lw @X[3],12($ctx)
sll $t2,$b,30
xor $t0,$b
lw @X[4],16($ctx)
srl $b,$b,2
addu $e,@X[$i%16]
or $b,$t2
addu $e,$t0
___
}
sub BODY_40_59 {
my ($i,$a,$b,$c,$d,$e)=@_;
my $j=$i+1;
$code.=<<___ if ($i<79);
xor @X[$j%16],@X[($j+2)%16]
sll $t0,$a,5 # $i
addu $e,$K
srl $t1,$a,27
addu $e,$t0
xor @X[$j%16],@X[($j+8)%16]
and $t0,$c,$d
addu $e,$t1
xor @X[$j%16],@X[($j+13)%16]
sll $t2,$b,30
addu $e,$t0
srl $t1,@X[$j%16],31
xor $t0,$c,$d
addu @X[$j%16],@X[$j%16]
and $t0,$b
srl $b,$b,2
or @X[$j%16],$t1
addu $e,@X[$i%16]
or $b,$t2
addu $e,$t0
___
}
$FRAMESIZE=16; # large enough to accomodate NUBI saved registers
$SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0xc0fff008 : 0xc0ff0000;
$code=<<___;
#ifdef OPENSSL_FIPSCANISTER
# include <openssl/fipssyms.h>
#endif
.text
.set noat
.set noreorder
.align 5
.globl sha1_block_data_order
.ent sha1_block_data_order
sha1_block_data_order:
.frame $sp,$FRAMESIZE*$SZREG,$ra
.mask $SAVED_REGS_MASK,-$SZREG
.set noreorder
$PTR_SUB $sp,$FRAMESIZE*$SZREG
$REG_S $ra,($FRAMESIZE-1)*$SZREG($sp)
$REG_S $fp,($FRAMESIZE-2)*$SZREG($sp)
$REG_S $s11,($FRAMESIZE-3)*$SZREG($sp)
$REG_S $s10,($FRAMESIZE-4)*$SZREG($sp)
$REG_S $s9,($FRAMESIZE-5)*$SZREG($sp)
$REG_S $s8,($FRAMESIZE-6)*$SZREG($sp)
$REG_S $s7,($FRAMESIZE-7)*$SZREG($sp)
$REG_S $s6,($FRAMESIZE-8)*$SZREG($sp)
$REG_S $s5,($FRAMESIZE-9)*$SZREG($sp)
$REG_S $s4,($FRAMESIZE-10)*$SZREG($sp)
___
$code.=<<___ if ($flavour =~ /nubi/i); # optimize non-nubi prologue
$REG_S $s3,($FRAMESIZE-11)*$SZREG($sp)
$REG_S $s2,($FRAMESIZE-12)*$SZREG($sp)
$REG_S $s1,($FRAMESIZE-13)*$SZREG($sp)
$REG_S $s0,($FRAMESIZE-14)*$SZREG($sp)
$REG_S $gp,($FRAMESIZE-15)*$SZREG($sp)
___
$code.=<<___;
$PTR_SLL $num,6
$PTR_ADD $num,$inp
$REG_S $num,0($sp)
lw $A,0($ctx)
lw $B,4($ctx)
lw $C,8($ctx)
lw $D,12($ctx)
b .Loop
lw $E,16($ctx)
.align 4
.Loop:
.set reorder
lwl @X[0],$MSB($inp)
lui $K,0x5a82
lwr @X[0],$LSB($inp)
ori $K,0x7999 # K_00_19
___
for ($i=0;$i<15;$i++) { &BODY_00_14($i,@V); unshift(@V,pop(@V)); }
for (;$i<20;$i++) { &BODY_15_19($i,@V); unshift(@V,pop(@V)); }
$code.=<<___;
lui $K,0x6ed9
ori $K,0xeba1 # K_20_39
___
for (;$i<40;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
$code.=<<___;
lui $K,0x8f1b
ori $K,0xbcdc # K_40_59
___
for (;$i<60;$i++) { &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
$code.=<<___;
lui $K,0xca62
ori $K,0xc1d6 # K_60_79
___
for (;$i<80;$i++) { &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
$code.=<<___;
$PTR_ADD $inp,64
$REG_L $num,0($sp)
addu $A,$X[0]
addu $B,$X[1]
sw $A,0($ctx)
addu $C,$X[2]
addu $D,$X[3]
sw $B,4($ctx)
addu $E,$X[4]
sw $C,8($ctx)
sw $D,12($ctx)
sw $E,16($ctx)
.set noreorder
bne $inp,$num,.Loop
nop
.set noreorder
$REG_L $ra,($FRAMESIZE-1)*$SZREG($sp)
$REG_L $fp,($FRAMESIZE-2)*$SZREG($sp)
$REG_L $s11,($FRAMESIZE-3)*$SZREG($sp)
$REG_L $s10,($FRAMESIZE-4)*$SZREG($sp)
$REG_L $s9,($FRAMESIZE-5)*$SZREG($sp)
$REG_L $s8,($FRAMESIZE-6)*$SZREG($sp)
$REG_L $s7,($FRAMESIZE-7)*$SZREG($sp)
$REG_L $s6,($FRAMESIZE-8)*$SZREG($sp)
$REG_L $s5,($FRAMESIZE-9)*$SZREG($sp)
$REG_L $s4,($FRAMESIZE-10)*$SZREG($sp)
___
$code.=<<___ if ($flavour =~ /nubi/i);
$REG_L $s3,($FRAMESIZE-11)*$SZREG($sp)
$REG_L $s2,($FRAMESIZE-12)*$SZREG($sp)
$REG_L $s1,($FRAMESIZE-13)*$SZREG($sp)
$REG_L $s0,($FRAMESIZE-14)*$SZREG($sp)
$REG_L $gp,($FRAMESIZE-15)*$SZREG($sp)
___
$code.=<<___;
jr $ra
$PTR_ADD $sp,$FRAMESIZE*$SZREG
.end sha1_block_data_order
.rdata
.asciiz "SHA1 for MIPS, CRYPTOGAMS by <appro\@openssl.org>"
___
print $code;
close STDOUT;

455
crypto/sha/asm/sha512-mips.pl Normal file
View File

@ -0,0 +1,455 @@
#!/usr/bin/env perl
# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
# project. The module is, however, dual licensed under OpenSSL and
# CRYPTOGAMS licenses depending on where you obtain it. For further
# details see http://www.openssl.org/~appro/cryptogams/.
# ====================================================================
# SHA2 block procedures for MIPS.
# October 2010.
#
# SHA256 performance improvement on MIPS R5000 CPU is ~27% over gcc-
# generated code in o32 build and ~55% in n32/64 build. SHA512 [which
# for now can only be compiled for MIPS64 ISA] improvement is modest
# ~17%, but it comes for free, because it's same instruction sequence.
# Improvement coefficients are for aligned input.
######################################################################
# There is a number of MIPS ABI in use, O32 and N32/64 are most
# widely used. Then there is a new contender: NUBI. It appears that if
# one picks the latter, it's possible to arrange code in ABI neutral
# manner. Therefore let's stick to NUBI register layout:
#
($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
#
# The return value is placed in $a0. Following coding rules facilitate
# interoperability:
#
# - never ever touch $tp, "thread pointer", former $gp [o32 can be
# excluded from the rule, because it's specified volatile];
# - copy return value to $t0, former $v0 [or to $a0 if you're adapting
# old code];
# - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
#
# For reference here is register layout for N32/64 MIPS ABIs:
#
# ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
# ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
# ($t0,$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));
#
$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
if ($flavour =~ /64|n32/i) {
$PTR_ADD="dadd"; # incidentally works even on n32
$PTR_SUB="dsub"; # incidentally works even on n32
$REG_S="sd";
$REG_L="ld";
$PTR_SLL="dsll"; # incidentally works even on n32
$SZREG=8;
} else {
$PTR_ADD="add";
$PTR_SUB="sub";
$REG_S="sw";
$REG_L="lw";
$PTR_SLL="sll";
$SZREG=4;
}
$pf = ($flavour =~ /nubi/i) ? $t0 : $t2;
#
# <appro@openssl.org>
#
######################################################################
$big_endian=(`echo MIPSEL | $ENV{CC} -E -P -`=~/MIPSEL/)?1:0;
for (@ARGV) { $output=$_ if (/^\w[\w\-]*\.\w+$/); }
open STDOUT,">$output";
if (!defined($big_endian)) { $big_endian=(unpack('L',pack('N',1))==1); }
if ($output =~ /512/) {
$label="512";
$SZ=8;
$LD="ld"; # load from memory
$ST="sd"; # store to memory
$SLL="dsll"; # shift left logical
$SRL="dsrl"; # shift right logical
$ADDU="daddu";
@Sigma0=(28,34,39);
@Sigma1=(14,18,41);
@sigma0=( 7, 1, 8); # right shift first
@sigma1=( 6,19,61); # right shift first
$lastK=0x817;
$rounds=80;
} else {
$label="256";
$SZ=4;
$LD="lw"; # load from memory
$ST="sw"; # store to memory
$SLL="sll"; # shift left logical
$SRL="srl"; # shift right logical
$ADDU="addu";
@Sigma0=( 2,13,22);
@Sigma1=( 6,11,25);
@sigma0=( 3, 7,18); # right shift first
@sigma1=(10,17,19); # right shift first
$lastK=0x8f2;
$rounds=64;
}
$MSB = $big_endian ? 0 : ($SZ-1);
$LSB = ($SZ-1)&~$MSB;
@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("\$$_",(1,2,3,7,24,25,30,31));
@X=map("\$$_",(8..23));
$ctx=$a0;
$inp=$a1;
$len=$a2; $Ktbl=$len;
sub BODY_00_15 {
my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
my ($T1,$tmp0,$tmp1,$tmp2)=(@X[4],@X[5],@X[6],@X[7]);
$code.=<<___ if ($i<15);
${LD}l @X[1],`($i+1)*$SZ+$MSB`($inp)
${LD}r @X[1],`($i+1)*$SZ+$LSB`($inp)
___
$code.=<<___ if (!$big_endian && $i<16 && $SZ==4);
srl $tmp0,@X[0],24 # byte swap($i)
srl $tmp1,@X[0],8
andi $tmp2,@X[0],0xFF00
sll @X[0],@X[0],24
andi $tmp1,0xFF00
sll $tmp2,$tmp2,8
or @X[0],$tmp0
or $tmp1,$tmp2
or @X[0],$tmp1
___
$code.=<<___ if (!$big_endian && $i<16 && $SZ==8);
ori $tmp0,$zero,0xFF
dsll $tmp2,$tmp0,32
or $tmp0,$tmp2 # 0x000000FF000000FF
and $tmp1,@X[0],$tmp0 # byte swap($i)
dsrl $tmp2,@X[0],24
dsll $tmp1,24
and $tmp2,$tmp0
dsll $tmp0,8 # 0x0000FF000000FF00
or $tmp1,$tmp2
and $tmp2,@X[0],$tmp0
dsrl @X[0],8
dsll $tmp2,8
and @X[0],$tmp0
or $tmp1,$tmp2
or @X[0],$tmp1
dsrl $tmp1,@X[0],32
dsll @X[0],32
or @X[0],$tmp1
___
$code.=<<___;
$ADDU $T1,$X[0],$h # $i
$SRL $h,$e,@Sigma1[0]
xor $tmp2,$f,$g
$SLL $tmp1,$e,`$SZ*8-@Sigma1[2]`
and $tmp2,$e
$SRL $tmp0,$e,@Sigma1[1]
xor $h,$tmp1
$SLL $tmp1,$e,`$SZ*8-@Sigma1[1]`
xor $h,$tmp0
$SRL $tmp0,$e,@Sigma1[2]
xor $h,$tmp1
$SLL $tmp1,$e,`$SZ*8-@Sigma1[0]`
xor $h,$tmp0
xor $tmp2,$g # Ch(e,f,g)
xor $tmp0,$tmp1,$h # Sigma1(e)
$SRL $h,$a,@Sigma0[0]
$ADDU $T1,$tmp2
$LD $tmp2,`$i*$SZ`($Ktbl) # K[$i]
$SLL $tmp1,$a,`$SZ*8-@Sigma0[2]`
$ADDU $T1,$tmp0
$SRL $tmp0,$a,@Sigma0[1]
xor $h,$tmp1
$SLL $tmp1,$a,`$SZ*8-@Sigma0[1]`
xor $h,$tmp0
$SRL $tmp0,$a,@Sigma0[2]
xor $h,$tmp1
$SLL $tmp1,$a,`$SZ*8-@Sigma0[0]`
xor $h,$tmp0
$ST @X[0],`($i%16)*$SZ`($sp) # offload to ring buffer
xor $h,$tmp1 # Sigma0(a)
or $tmp0,$a,$b
and $tmp1,$a,$b
and $tmp0,$c
or $tmp1,$tmp0 # Maj(a,b,c)
$ADDU $T1,$tmp2 # +=K[$i]
$ADDU $h,$tmp1
$ADDU $d,$T1
$ADDU $h,$T1
___
$code.=<<___ if ($i>=13);
$LD @X[3],`(($i+3)%16)*$SZ`($sp) # prefetch from ring buffer
___
}
sub BODY_16_XX {
my $i=@_[0];
my ($tmp0,$tmp1,$tmp2,$tmp3)=(@X[4],@X[5],@X[6],@X[7]);
$code.=<<___;
$SRL $tmp2,@X[1],@sigma0[0] # Xupdate($i)
$ADDU @X[0],@X[9] # +=X[i+9]
$SLL $tmp1,@X[1],`$SZ*8-@sigma0[2]`
$SRL $tmp0,@X[1],@sigma0[1]
xor $tmp2,$tmp1
$SLL $tmp1,`@sigma0[2]-@sigma0[1]`
xor $tmp2,$tmp0
$SRL $tmp0,@X[1],@sigma0[2]
xor $tmp2,$tmp1
$SRL $tmp3,@X[14],@sigma1[0]
xor $tmp2,$tmp0 # sigma0(X[i+1])
$SLL $tmp1,@X[14],`$SZ*8-@sigma1[2]`
$ADDU @X[0],$tmp2
$SRL $tmp0,@X[14],@sigma1[1]
xor $tmp3,$tmp1
$SLL $tmp1,`@sigma1[2]-@sigma1[1]`
xor $tmp3,$tmp0
$SRL $tmp0,@X[14],@sigma1[2]
xor $tmp3,$tmp1
xor $tmp3,$tmp0 # sigma1(X[i+14])
$ADDU @X[0],$tmp3
___
&BODY_00_15(@_);
}
$FRAMESIZE=16*$SZ+16*$SZREG;
$SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0xc0fff008 : 0xc0ff0000;
$code.=<<___;
#ifdef OPENSSL_FIPSCANISTER
# include <openssl/fipssyms.h>
#endif
.text
.set noat
#if !defined(__vxworks) || defined(__pic__)
.option pic2
#endif
.align 5
.globl sha${label}_block_data_order
.ent sha${label}_block_data_order
sha${label}_block_data_order:
.frame $sp,$FRAMESIZE,$ra
.mask $SAVED_REGS_MASK,-$SZREG
.set noreorder
___
$code.=<<___ if ($flavour =~ /o32/i); # o32 PIC-ification
.cpload $pf
___
$code.=<<___;
$PTR_SUB $sp,$FRAMESIZE
$REG_S $ra,$FRAMESIZE-1*$SZREG($sp)
$REG_S $fp,$FRAMESIZE-2*$SZREG($sp)
$REG_S $s11,$FRAMESIZE-3*$SZREG($sp)
$REG_S $s10,$FRAMESIZE-4*$SZREG($sp)
$REG_S $s9,$FRAMESIZE-5*$SZREG($sp)
$REG_S $s8,$FRAMESIZE-6*$SZREG($sp)
$REG_S $s7,$FRAMESIZE-7*$SZREG($sp)
$REG_S $s6,$FRAMESIZE-8*$SZREG($sp)
$REG_S $s5,$FRAMESIZE-9*$SZREG($sp)
$REG_S $s4,$FRAMESIZE-10*$SZREG($sp)
___
$code.=<<___ if ($flavour =~ /nubi/i); # optimize non-nubi prologue
$REG_S $s3,$FRAMESIZE-11*$SZREG($sp)
$REG_S $s2,$FRAMESIZE-12*$SZREG($sp)
$REG_S $s1,$FRAMESIZE-13*$SZREG($sp)
$REG_S $s0,$FRAMESIZE-14*$SZREG($sp)
$REG_S $gp,$FRAMESIZE-15*$SZREG($sp)
___
$code.=<<___;
$PTR_SLL @X[15],$len,`log(16*$SZ)/log(2)`
___
$code.=<<___ if ($flavour !~ /o32/i); # non-o32 PIC-ification
.cplocal $Ktbl
.cpsetup $pf,$zero,sha${label}_block_data_order
___
$code.=<<___;
.set reorder
la $Ktbl,K${label} # PIC-ified 'load address'
$LD $A,0*$SZ($ctx) # load context
$LD $B,1*$SZ($ctx)
$LD $C,2*$SZ($ctx)
$LD $D,3*$SZ($ctx)
$LD $E,4*$SZ($ctx)
$LD $F,5*$SZ($ctx)
$LD $G,6*$SZ($ctx)
$LD $H,7*$SZ($ctx)
$PTR_ADD @X[15],$inp # pointer to the end of input
$REG_S @X[15],16*$SZ($sp)
b .Loop
.align 5
.Loop:
${LD}l @X[0],$MSB($inp)
${LD}r @X[0],$LSB($inp)
___
for ($i=0;$i<16;$i++)
{ &BODY_00_15($i,@V); unshift(@V,pop(@V)); push(@X,shift(@X)); }
$code.=<<___;
b .L16_xx
.align 4
.L16_xx:
___
for (;$i<32;$i++)
{ &BODY_16_XX($i,@V); unshift(@V,pop(@V)); push(@X,shift(@X)); }
$code.=<<___;
and @X[6],0xfff
li @X[7],$lastK
.set noreorder
bne @X[6],@X[7],.L16_xx
$PTR_ADD $Ktbl,16*$SZ # Ktbl+=16
$REG_L @X[15],16*$SZ($sp) # restore pointer to the end of input
$LD @X[0],0*$SZ($ctx)
$LD @X[1],1*$SZ($ctx)
$LD @X[2],2*$SZ($ctx)
$PTR_ADD $inp,16*$SZ
$LD @X[3],3*$SZ($ctx)
$ADDU $A,@X[0]
$LD @X[4],4*$SZ($ctx)
$ADDU $B,@X[1]
$LD @X[5],5*$SZ($ctx)
$ADDU $C,@X[2]
$LD @X[6],6*$SZ($ctx)
$ADDU $D,@X[3]
$LD @X[7],7*$SZ($ctx)
$ADDU $E,@X[4]
$ST $A,0*$SZ($ctx)
$ADDU $F,@X[5]
$ST $B,1*$SZ($ctx)
$ADDU $G,@X[6]
$ST $C,2*$SZ($ctx)
$ADDU $H,@X[7]
$ST $D,3*$SZ($ctx)
$ST $E,4*$SZ($ctx)
$ST $F,5*$SZ($ctx)
$ST $G,6*$SZ($ctx)
$ST $H,7*$SZ($ctx)
bnel $inp,@X[15],.Loop
$PTR_SUB $Ktbl,`($rounds-16)*$SZ` # rewind $Ktbl
$REG_L $ra,$FRAMESIZE-1*$SZREG($sp)
$REG_L $fp,$FRAMESIZE-2*$SZREG($sp)
$REG_L $s11,$FRAMESIZE-3*$SZREG($sp)
$REG_L $s10,$FRAMESIZE-4*$SZREG($sp)
$REG_L $s9,$FRAMESIZE-5*$SZREG($sp)
$REG_L $s8,$FRAMESIZE-6*$SZREG($sp)
$REG_L $s7,$FRAMESIZE-7*$SZREG($sp)
$REG_L $s6,$FRAMESIZE-8*$SZREG($sp)
$REG_L $s5,$FRAMESIZE-9*$SZREG($sp)
$REG_L $s4,$FRAMESIZE-10*$SZREG($sp)
___
$code.=<<___ if ($flavour =~ /nubi/i);
$REG_L $s3,$FRAMESIZE-11*$SZREG($sp)
$REG_L $s2,$FRAMESIZE-12*$SZREG($sp)
$REG_L $s1,$FRAMESIZE-13*$SZREG($sp)
$REG_L $s0,$FRAMESIZE-14*$SZREG($sp)
$REG_L $gp,$FRAMESIZE-15*$SZREG($sp)
___
$code.=<<___;
jr $ra
$PTR_ADD $sp,$FRAMESIZE
.end sha${label}_block_data_order
.rdata
.align 5
K${label}:
___
if ($SZ==4) {
$code.=<<___;
.word 0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5
.word 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5
.word 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3
.word 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174
.word 0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc
.word 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da
.word 0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7
.word 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967
.word 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13
.word 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85
.word 0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3
.word 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070
.word 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5
.word 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3
.word 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208
.word 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
___
} else {
$code.=<<___;
.dword 0x428a2f98d728ae22, 0x7137449123ef65cd
.dword 0xb5c0fbcfec4d3b2f, 0xe9b5dba58189dbbc
.dword 0x3956c25bf348b538, 0x59f111f1b605d019
.dword 0x923f82a4af194f9b, 0xab1c5ed5da6d8118
.dword 0xd807aa98a3030242, 0x12835b0145706fbe
.dword 0x243185be4ee4b28c, 0x550c7dc3d5ffb4e2
.dword 0x72be5d74f27b896f, 0x80deb1fe3b1696b1
.dword 0x9bdc06a725c71235, 0xc19bf174cf692694
.dword 0xe49b69c19ef14ad2, 0xefbe4786384f25e3
.dword 0x0fc19dc68b8cd5b5, 0x240ca1cc77ac9c65
.dword 0x2de92c6f592b0275, 0x4a7484aa6ea6e483
.dword 0x5cb0a9dcbd41fbd4, 0x76f988da831153b5
.dword 0x983e5152ee66dfab, 0xa831c66d2db43210
.dword 0xb00327c898fb213f, 0xbf597fc7beef0ee4
.dword 0xc6e00bf33da88fc2, 0xd5a79147930aa725
.dword 0x06ca6351e003826f, 0x142929670a0e6e70
.dword 0x27b70a8546d22ffc, 0x2e1b21385c26c926
.dword 0x4d2c6dfc5ac42aed, 0x53380d139d95b3df
.dword 0x650a73548baf63de, 0x766a0abb3c77b2a8
.dword 0x81c2c92e47edaee6, 0x92722c851482353b
.dword 0xa2bfe8a14cf10364, 0xa81a664bbc423001
.dword 0xc24b8b70d0f89791, 0xc76c51a30654be30
.dword 0xd192e819d6ef5218, 0xd69906245565a910
.dword 0xf40e35855771202a, 0x106aa07032bbd1b8
.dword 0x19a4c116b8d2d0c8, 0x1e376c085141ab53
.dword 0x2748774cdf8eeb99, 0x34b0bcb5e19b48a8
.dword 0x391c0cb3c5c95a63, 0x4ed8aa4ae3418acb
.dword 0x5b9cca4f7763e373, 0x682e6ff3d6b2b8a3
.dword 0x748f82ee5defb2fc, 0x78a5636f43172f60
.dword 0x84c87814a1f0ab72, 0x8cc702081a6439ec
.dword 0x90befffa23631e28, 0xa4506cebde82bde9
.dword 0xbef9a3f7b2c67915, 0xc67178f2e372532b
.dword 0xca273eceea26619c, 0xd186b8c721c0c207
.dword 0xeada7dd6cde0eb1e, 0xf57d4f7fee6ed178
.dword 0x06f067aa72176fba, 0x0a637dc5a2c898a6
.dword 0x113f9804bef90dae, 0x1b710b35131c471b
.dword 0x28db77f523047d84, 0x32caab7b40c72493
.dword 0x3c9ebe0a15c9bebc, 0x431d67c49c100d4c
.dword 0x4cc5d4becb3e42b6, 0x597f299cfc657e2a
.dword 0x5fcb6fab3ad6faec, 0x6c44198c4a475817
___
}
$code.=<<___;
.asciiz "SHA${label} for MIPS, CRYPTOGAMS by <appro\@openssl.org>"
.align 5
___
$code =~ s/\`([^\`]*)\`/eval $1/gem;
print $code;
close STDOUT;