PPC assembly pack: jumbo update from master.

Add Vector Permutation AES and little-endian support.
This commit is contained in:
Andy Polyakov 2014-02-01 21:48:31 +01:00
parent 5572bc4e2f
commit 50f1b47c7f
9 changed files with 1834 additions and 108 deletions

View File

@ -139,8 +139,8 @@ my $s390x_asm="s390xcap.o s390xcpuid.o:bn-s390x.o s390x-mont.o s390x-gf2m.o::aes
my $armv4_asm="armcap.o armv4cpuid.o:bn_asm.o armv4-mont.o armv4-gf2m.o::aes_cbc.o aes-armv4.o bsaes-armv7.o:::sha1-armv4-large.o sha256-armv4.o sha512-armv4.o:::::::ghash-armv4.o::void";
my $parisc11_asm="pariscid.o:bn_asm.o parisc-mont.o::aes_core.o aes_cbc.o aes-parisc.o:::sha1-parisc.o sha256-parisc.o sha512-parisc.o::rc4-parisc.o:::::ghash-parisc.o::32";
my $parisc20_asm="pariscid.o:pa-risc2W.o parisc-mont.o::aes_core.o aes_cbc.o aes-parisc.o:::sha1-parisc.o sha256-parisc.o sha512-parisc.o::rc4-parisc.o:::::ghash-parisc.o::64";
my $ppc32_asm="ppccpuid.o ppccap.o:bn-ppc.o ppc-mont.o ppc64-mont.o::aes_core.o aes_cbc.o aes-ppc.o:::sha1-ppc.o sha256-ppc.o::::::::";
my $ppc64_asm="ppccpuid.o ppccap.o:bn-ppc.o ppc-mont.o ppc64-mont.o::aes_core.o aes_cbc.o aes-ppc.o:::sha1-ppc.o sha256-ppc.o sha512-ppc.o::::::::";
my $ppc64_asm="ppccpuid.o ppccap.o:bn-ppc.o ppc-mont.o ppc64-mont.o::aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o:::sha1-ppc.o sha256-ppc.o sha512-ppc.o::::::::";
my $ppc32_asm=$ppc64_asm;
my $no_asm=":::::::::::::::void";
# As for $BSDthreads. Idea is to maintain "collective" set of flags,
@ -365,6 +365,7 @@ my %table=(
####
"linux-generic64","gcc:-DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_INT DES_UNROLL BF_PTR:${no_asm}:dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
"linux-ppc64", "gcc:-m64 -DB_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL:${ppc64_asm}:linux64:dlfcn:linux-shared:-fPIC:-m64:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR):::64",
"linux-ppc64le","gcc:-m64 -DL_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL:$ppc64_asm:linux64le:dlfcn:linux-shared:-fPIC:-m64:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR):::",
"linux-ia64", "gcc:-DL_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG RC4_CHUNK DES_UNROLL DES_INT:${ia64_asm}:dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
"linux-ia64-icc","icc:-DL_ENDIAN -DTERMIO -O2 -Wall::-D_REENTRANT::-ldl -no_cpprt:SIXTY_FOUR_BIT_LONG RC4_CHUNK DES_RISC1 DES_INT:${ia64_asm}:dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
"linux-x86_64", "gcc:-m64 -DL_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG RC4_CHUNK DES_INT DES_UNROLL:${x86_64_asm}:elf:dlfcn:linux-shared:-fPIC:-m64:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR):::64",

61
TABLE
View File

@ -902,10 +902,10 @@ $bn_ops = BN_LLONG RC4_CHAR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
$cast_obj =
$rc4_obj =
$rmd160_obj =
@ -935,10 +935,10 @@ $bn_ops = BN_LLONG RC4_CHAR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
$cast_obj =
$rc4_obj =
$rmd160_obj =
@ -1001,7 +1001,7 @@ $bn_ops = SIXTY_FOUR_BIT_LONG RC4_CHAR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
@ -1034,7 +1034,7 @@ $bn_ops = SIXTY_FOUR_BIT_LONG RC4_CHAR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
@ -1463,10 +1463,10 @@ $bn_ops = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
$cast_obj =
$rc4_obj =
$rmd160_obj =
@ -1496,7 +1496,7 @@ $bn_ops = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
@ -2189,10 +2189,10 @@ $bn_ops = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
$cast_obj =
$rc4_obj =
$rmd160_obj =
@ -4400,10 +4400,10 @@ $bn_ops = BN_LLONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
$cast_obj =
$rc4_obj =
$rmd160_obj =
@ -4433,7 +4433,7 @@ $bn_ops = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
@ -4455,6 +4455,39 @@ $ranlib =
$arflags =
$multilib = 64
*** linux-ppc64le
$cc = gcc
$cflags = -m64 -DL_ENDIAN -DTERMIO -O3 -Wall
$unistd =
$thread_cflag = -D_REENTRANT
$sys_id =
$lflags = -ldl
$bn_ops = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL
$cpuid_obj = ppccpuid.o ppccap.o
$bn_obj = bn-ppc.o ppc-mont.o ppc64-mont.o
$des_obj =
$aes_obj = aes_core.o aes_cbc.o aes-ppc.o vpaes-ppc.o
$bf_obj =
$md5_obj =
$sha1_obj = sha1-ppc.o sha256-ppc.o sha512-ppc.o
$cast_obj =
$rc4_obj =
$rmd160_obj =
$rc5_obj =
$wp_obj =
$cmll_obj =
$modes_obj =
$engines_obj =
$perlasm_scheme = linux64le
$dso_scheme = dlfcn
$shared_target= linux-shared
$shared_cflag = -fPIC
$shared_ldflag = -m64
$shared_extension = .so.$(SHLIB_MAJOR).$(SHLIB_MINOR)
$ranlib =
$arflags =
$multilib =
*** linux-sparcv8
$cc = gcc
$cflags = -mv8 -DB_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -Wall -DBN_DIV2W

View File

@ -75,6 +75,8 @@ aest4-sparcv9.s: asm/aest4-sparcv9.pl
aes-ppc.s: asm/aes-ppc.pl
$(PERL) asm/aes-ppc.pl $(PERLASM_SCHEME) $@
vpaes-ppc.s: asm/vpaes-ppc.pl
$(PERL) asm/vpaes-ppc.pl $(PERLASM_SCHEME) $@
aes-parisc.s: asm/aes-parisc.pl
$(PERL) asm/aes-parisc.pl $(PERLASM_SCHEME) $@

View File

@ -45,6 +45,8 @@ if ($flavour =~ /64/) {
$PUSH ="stw";
} else { die "nonsense $flavour"; }
$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
@ -361,17 +363,61 @@ $code.=<<___;
bne Lenc_unaligned
Lenc_unaligned_ok:
___
$code.=<<___ if (!$LITTLE_ENDIAN);
lwz $s0,0($inp)
lwz $s1,4($inp)
lwz $s2,8($inp)
lwz $s3,12($inp)
___
$code.=<<___ if ($LITTLE_ENDIAN);
lwz $t0,0($inp)
lwz $t1,4($inp)
lwz $t2,8($inp)
lwz $t3,12($inp)
rotlwi $s0,$t0,8
rotlwi $s1,$t1,8
rotlwi $s2,$t2,8
rotlwi $s3,$t3,8
rlwimi $s0,$t0,24,0,7
rlwimi $s1,$t1,24,0,7
rlwimi $s2,$t2,24,0,7
rlwimi $s3,$t3,24,0,7
rlwimi $s0,$t0,24,16,23
rlwimi $s1,$t1,24,16,23
rlwimi $s2,$t2,24,16,23
rlwimi $s3,$t3,24,16,23
___
$code.=<<___;
bl LAES_Te
bl Lppc_AES_encrypt_compact
$POP $out,`$FRAME-$SIZE_T*19`($sp)
___
$code.=<<___ if ($LITTLE_ENDIAN);
rotlwi $t0,$s0,8
rotlwi $t1,$s1,8
rotlwi $t2,$s2,8
rotlwi $t3,$s3,8
rlwimi $t0,$s0,24,0,7
rlwimi $t1,$s1,24,0,7
rlwimi $t2,$s2,24,0,7
rlwimi $t3,$s3,24,0,7
rlwimi $t0,$s0,24,16,23
rlwimi $t1,$s1,24,16,23
rlwimi $t2,$s2,24,16,23
rlwimi $t3,$s3,24,16,23
stw $t0,0($out)
stw $t1,4($out)
stw $t2,8($out)
stw $t3,12($out)
___
$code.=<<___ if (!$LITTLE_ENDIAN);
stw $s0,0($out)
stw $s1,4($out)
stw $s2,8($out)
stw $s3,12($out)
___
$code.=<<___;
b Lenc_done
Lenc_unaligned:
@ -795,17 +841,61 @@ Lenc_compact_done:
bne Ldec_unaligned
Ldec_unaligned_ok:
___
$code.=<<___ if (!$LITTLE_ENDIAN);
lwz $s0,0($inp)
lwz $s1,4($inp)
lwz $s2,8($inp)
lwz $s3,12($inp)
___
$code.=<<___ if ($LITTLE_ENDIAN);
lwz $t0,0($inp)
lwz $t1,4($inp)
lwz $t2,8($inp)
lwz $t3,12($inp)
rotlwi $s0,$t0,8
rotlwi $s1,$t1,8
rotlwi $s2,$t2,8
rotlwi $s3,$t3,8
rlwimi $s0,$t0,24,0,7
rlwimi $s1,$t1,24,0,7
rlwimi $s2,$t2,24,0,7
rlwimi $s3,$t3,24,0,7
rlwimi $s0,$t0,24,16,23
rlwimi $s1,$t1,24,16,23
rlwimi $s2,$t2,24,16,23
rlwimi $s3,$t3,24,16,23
___
$code.=<<___;
bl LAES_Td
bl Lppc_AES_decrypt_compact
$POP $out,`$FRAME-$SIZE_T*19`($sp)
___
$code.=<<___ if ($LITTLE_ENDIAN);
rotlwi $t0,$s0,8
rotlwi $t1,$s1,8
rotlwi $t2,$s2,8
rotlwi $t3,$s3,8
rlwimi $t0,$s0,24,0,7
rlwimi $t1,$s1,24,0,7
rlwimi $t2,$s2,24,0,7
rlwimi $t3,$s3,24,0,7
rlwimi $t0,$s0,24,16,23
rlwimi $t1,$s1,24,16,23
rlwimi $t2,$s2,24,16,23
rlwimi $t3,$s3,24,16,23
stw $t0,0($out)
stw $t1,4($out)
stw $t2,8($out)
stw $t3,12($out)
___
$code.=<<___ if (!$LITTLE_ENDIAN);
stw $s0,0($out)
stw $s1,4($out)
stw $s2,8($out)
stw $s3,12($out)
___
$code.=<<___;
b Ldec_done
Ldec_unaligned:

1507
crypto/aes/asm/vpaes-ppc.pl Normal file

File diff suppressed because it is too large Load Diff

View File

@ -157,6 +157,11 @@ void AES_xts_decrypt(const char *inp,char *out,size_t len,
const unsigned char iv[16]);
#endif
#if defined(VPAES_ASM) && (defined(__powerpc__) || defined(__ppc__) || defined(_ARCH_PPC))
extern unsigned int OPENSSL_ppccap_P;
#define VPAES_CAPABLE (OPENSSL_ppccap_P&(1<<1))
#endif
#if defined(AES_ASM) && !defined(I386_ONLY) && ( \
((defined(__i386) || defined(__i386__) || \
defined(_M_IX86)) && defined(OPENSSL_IA32_SSE2))|| \

View File

@ -27,7 +27,8 @@ my $globl = sub {
/osx/ && do { $name = "_$name";
last;
};
/linux.*32/ && do { $ret .= ".globl $name\n";
/linux.*(32|64le)/
&& do { $ret .= ".globl $name\n";
$ret .= ".type $name,\@function";
last;
};
@ -49,7 +50,9 @@ my $globl = sub {
$ret;
};
my $text = sub {
($flavour =~ /aix/) ? ".csect" : ".text";
my $ret = ($flavour =~ /aix/) ? ".csect\t.text[PR],7" : ".text";
$ret = ".abiversion 2\n".$ret if ($flavour =~ /linux.*64le/);
$ret;
};
my $machine = sub {
my $junk = shift;
@ -64,8 +67,8 @@ my $size = sub {
if ($flavour =~ /linux/)
{ shift;
my $name = shift; $name =~ s|^[\.\_]||;
my $ret = ".size $name,.-".($flavour=~/64/?".":"").$name;
$ret .= "\n.size .$name,.-.$name" if ($flavour=~/64/);
my $ret = ".size $name,.-".($flavour=~/64$/?".":"").$name;
$ret .= "\n.size .$name,.-.$name" if ($flavour=~/64$/);
$ret;
}
else
@ -79,6 +82,25 @@ my $asciz = sub {
else
{ ""; }
};
my $quad = sub {
shift;
my @ret;
my ($hi,$lo);
for (@_) {
if (/^0x([0-9a-f]*?)([0-9a-f]{1,8})$/io)
{ $hi=$1?"0x$1":"0"; $lo="0x$2"; }
elsif (/^([0-9]+)$/o)
{ $hi=$1>>32; $lo=$1&0xffffffff; } # error-prone with 32-bit perl
else
{ $hi=undef; $lo=$_; }
if (defined($hi))
{ push(@ret,$flavour=~/le$/o?".long\t$lo,$hi":".long\t$hi,$lo"); }
else
{ push(@ret,".quad $lo"); }
}
join("\n",@ret);
};
################################################################
# simplified mnemonics not handled by at least one assembler
@ -124,6 +146,10 @@ my $extrdi = sub {
$b = ($b+$n)&63; $n = 64-$n;
" rldicl $ra,$rs,$b,$n";
};
my $vmr = sub {
my ($f,$vx,$vy) = @_;
" vor $vx,$vy,$vy";
};
while($line=<>) {
@ -140,7 +166,10 @@ while($line=<>) {
{
$line =~ s|(^[\.\w]+)\:\s*||;
my $label = $1;
printf "%s:",($GLOBALS{$label} or $label) if ($label);
if ($label) {
printf "%s:",($GLOBALS{$label} or $label);
printf "\n.localentry\t$GLOBALS{$label},0" if ($GLOBALS{$label} && $flavour =~ /linux.*64le/);
}
}
{
@ -149,7 +178,7 @@ while($line=<>) {
my $mnemonic = $2;
my $f = $3;
my $opcode = eval("\$$mnemonic");
$line =~ s|\bc?[rf]([0-9]+)\b|$1|g if ($c ne "." and $flavour !~ /osx/);
$line =~ s/\b(c?[rf]|v|vs)([0-9]+)\b/$2/g if ($c ne "." and $flavour !~ /osx/);
if (ref($opcode) eq 'CODE') { $line = &$opcode($f,split(',',$line)); }
elsif ($mnemonic) { $line = $c.$mnemonic.$f."\t".$line; }
}

View File

@ -9,8 +9,7 @@
# I let hardware handle unaligned input(*), except on page boundaries
# (see below for details). Otherwise straightforward implementation
# with X vector in register bank. The module is big-endian [which is
# not big deal as there're no little-endian targets left around].
# with X vector in register bank.
#
# (*) this means that this module is inappropriate for PPC403? Does
# anybody know if pre-POWER3 can sustain unaligned load?
@ -38,6 +37,10 @@ if ($flavour =~ /64/) {
$PUSH ="stw";
} else { die "nonsense $flavour"; }
# Define endianess based on flavour
# i.e.: linux64le
$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
@ -68,14 +71,28 @@ $T ="r12";
@X=("r16","r17","r18","r19","r20","r21","r22","r23",
"r24","r25","r26","r27","r28","r29","r30","r31");
sub loadbe {
my ($dst, $src, $temp_reg) = @_;
$code.=<<___ if (!$LITTLE_ENDIAN);
lwz $dst,$src
___
$code.=<<___ if ($LITTLE_ENDIAN);
lwz $temp_reg,$src
rotlwi $dst,$temp_reg,8
rlwimi $dst,$temp_reg,24,0,7
rlwimi $dst,$temp_reg,24,16,23
___
}
sub BODY_00_19 {
my ($i,$a,$b,$c,$d,$e,$f)=@_;
my $j=$i+1;
$code.=<<___ if ($i==0);
lwz @X[$i],`$i*4`($inp)
___
# Since the last value of $f is discarded, we can use
# it as a temp reg to swap byte-order when needed.
loadbe("@X[$i]","`$i*4`($inp)",$f) if ($i==0);
loadbe("@X[$j]","`$j*4`($inp)",$f) if ($i<15);
$code.=<<___ if ($i<15);
lwz @X[$j],`$j*4`($inp)
add $f,$K,$e
rotlwi $e,$a,5
add $f,$f,@X[$i]

View File

@ -9,8 +9,7 @@
# I let hardware handle unaligned input, except on page boundaries
# (see below for details). Otherwise straightforward implementation
# with X vector in register bank. The module is big-endian [which is
# not big deal as there're no little-endian targets left around].
# with X vector in register bank.
# sha256 | sha512
# -m64 -m32 | -m64 -m32
@ -56,6 +55,8 @@ if ($flavour =~ /64/) {
$PUSH="stw";
} else { die "nonsense $flavour"; }
$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
@ -229,7 +230,7 @@ ___
} else {
for ($i=16;$i<32;$i++) {
$code.=<<___;
lwz r$i,`4*($i-16)`($ctx)
lwz r$i,`$LITTLE_ENDIAN^(4*($i-16))`($ctx)
___
}
}
@ -353,15 +354,32 @@ Lsha2_block_private:
$LD $t1,0($Tbl)
___
for($i=0;$i<16;$i++) {
$code.=<<___ if ($SZ==4);
$code.=<<___ if ($SZ==4 && !$LITTLE_ENDIAN);
lwz @X[$i],`$i*$SZ`($inp)
___
$code.=<<___ if ($SZ==4 && $LITTLE_ENDIAN);
lwz $a0,`$i*$SZ`($inp)
rotlwi @X[$i],$a0,8
rlwimi @X[$i],$a0,24,0,7
rlwimi @X[$i],$a0,24,16,23
___
# 64-bit loads are split to 2x32-bit ones, as CPU can't handle
# unaligned 64-bit loads, only 32-bit ones...
$code.=<<___ if ($SZ==8);
$code.=<<___ if ($SZ==8 && !$LITTLE_ENDIAN);
lwz $t0,`$i*$SZ`($inp)
lwz @X[$i],`$i*$SZ+4`($inp)
insrdi @X[$i],$t0,32,0
___
$code.=<<___ if ($SZ==8 && $LITTLE_ENDIAN);
lwz $a0,`$i*$SZ`($inp)
lwz $a1,`$i*$SZ+4`($inp)
rotlwi $t0,$a0,8
rotlwi @X[$i],$a1,8
rlwimi $t0,$a0,24,0,7
rlwimi @X[$i],$a1,24,0,7
rlwimi $t0,$a0,24,16,23
rlwimi @X[$i],$a1,24,16,23
insrdi @X[$i],$t0,32,0
___
&ROUND_00_15($i,@V);
unshift(@V,pop(@V));
@ -445,9 +463,9 @@ my ($i, $ahi,$alo,$bhi,$blo,$chi,$clo,$dhi,$dlo,
$ehi,$elo,$fhi,$flo,$ghi,$glo,$hhi,$hlo)=@_;
$code.=<<___;
lwz $t2,`$SZ*($i%16)+4`($Tbl)
lwz $t2,`$SZ*($i%16)+($LITTLE_ENDIAN^4)`($Tbl)
xor $a0,$flo,$glo
lwz $t3,`$SZ*($i%16)+0`($Tbl)
lwz $t3,`$SZ*($i%16)+($LITTLE_ENDIAN^0)`($Tbl)
xor $a1,$fhi,$ghi
addc $hlo,$hlo,$t0 ; h+=x[i]
stw $t0,`$XOFF+0+$SZ*($i%16)`($sp) ; save x[i]
@ -510,10 +528,20 @@ $code.=<<___ if ($i>=15);
lwz $t0,`$XOFF+0+$SZ*(($i+2)%16)`($sp)
lwz $t1,`$XOFF+4+$SZ*(($i+2)%16)`($sp)
___
$code.=<<___ if ($i<15);
$code.=<<___ if ($i<15 && !$LITTLE_ENDIAN);
lwz $t1,`$SZ*($i+1)+0`($inp)
lwz $t0,`$SZ*($i+1)+4`($inp)
___
$code.=<<___ if ($i<15 && $LITTLE_ENDIAN);
lwz $a2,`$SZ*($i+1)+0`($inp)
lwz $a3,`$SZ*($i+1)+4`($inp)
rotlwi $t1,$a2,8
rotlwi $t0,$a3,8
rlwimi $t1,$a2,24,0,7
rlwimi $t0,$a3,24,0,7
rlwimi $t1,$a2,24,16,23
rlwimi $t0,$a3,24,16,23
___
$code.=<<___;
xor $s0,$s0,$t2 ; Sigma0(a)
xor $s1,$s1,$t3
@ -579,11 +607,25 @@ ___
$code.=<<___;
.align 4
Lsha2_block_private:
___
$code.=<<___ if (!$LITTLE_ENDIAN);
lwz $t1,0($inp)
xor $a2,@V[3],@V[5] ; B^C, magic seed
lwz $t0,4($inp)
xor $a3,@V[2],@V[4]
___
$code.=<<___ if ($LITTLE_ENDIAN);
lwz $a1,0($inp)
xor $a2,@V[3],@V[5] ; B^C, magic seed
lwz $a0,4($inp)
xor $a3,@V[2],@V[4]
rotlwi $t1,$a1,8
rotlwi $t0,$a0,8
rlwimi $t1,$a1,24,0,7
rlwimi $t0,$a0,24,0,7
rlwimi $t1,$a1,24,16,23
rlwimi $t0,$a0,24,16,23
___
for($i=0;$i<16;$i++) {
&ROUND_00_15_ppc32($i,@V);
unshift(@V,pop(@V)); unshift(@V,pop(@V));
@ -609,54 +651,54 @@ $code.=<<___;
$POP $num,`$FRAME-$SIZE_T*24`($sp) ; end pointer
subi $Tbl,$Tbl,`($rounds-16)*$SZ` ; rewind Tbl
lwz $t0,0($ctx)
lwz $t1,4($ctx)
lwz $t2,8($ctx)
lwz $t3,12($ctx)
lwz $a0,16($ctx)
lwz $a1,20($ctx)
lwz $a2,24($ctx)
lwz $t0,`$LITTLE_ENDIAN^0`($ctx)
lwz $t1,`$LITTLE_ENDIAN^4`($ctx)
lwz $t2,`$LITTLE_ENDIAN^8`($ctx)
lwz $t3,`$LITTLE_ENDIAN^12`($ctx)
lwz $a0,`$LITTLE_ENDIAN^16`($ctx)
lwz $a1,`$LITTLE_ENDIAN^20`($ctx)
lwz $a2,`$LITTLE_ENDIAN^24`($ctx)
addc @V[1],@V[1],$t1
lwz $a3,28($ctx)
lwz $a3,`$LITTLE_ENDIAN^28`($ctx)
adde @V[0],@V[0],$t0
lwz $t0,32($ctx)
lwz $t0,`$LITTLE_ENDIAN^32`($ctx)
addc @V[3],@V[3],$t3
lwz $t1,36($ctx)
lwz $t1,`$LITTLE_ENDIAN^36`($ctx)
adde @V[2],@V[2],$t2
lwz $t2,40($ctx)
lwz $t2,`$LITTLE_ENDIAN^40`($ctx)
addc @V[5],@V[5],$a1
lwz $t3,44($ctx)
lwz $t3,`$LITTLE_ENDIAN^44`($ctx)
adde @V[4],@V[4],$a0
lwz $a0,48($ctx)
lwz $a0,`$LITTLE_ENDIAN^48`($ctx)
addc @V[7],@V[7],$a3
lwz $a1,52($ctx)
lwz $a1,`$LITTLE_ENDIAN^52`($ctx)
adde @V[6],@V[6],$a2
lwz $a2,56($ctx)
lwz $a2,`$LITTLE_ENDIAN^56`($ctx)
addc @V[9],@V[9],$t1
lwz $a3,60($ctx)
lwz $a3,`$LITTLE_ENDIAN^60`($ctx)
adde @V[8],@V[8],$t0
stw @V[0],0($ctx)
stw @V[1],4($ctx)
stw @V[0],`$LITTLE_ENDIAN^0`($ctx)
stw @V[1],`$LITTLE_ENDIAN^4`($ctx)
addc @V[11],@V[11],$t3
stw @V[2],8($ctx)
stw @V[3],12($ctx)
stw @V[2],`$LITTLE_ENDIAN^8`($ctx)
stw @V[3],`$LITTLE_ENDIAN^12`($ctx)
adde @V[10],@V[10],$t2
stw @V[4],16($ctx)
stw @V[5],20($ctx)
stw @V[4],`$LITTLE_ENDIAN^16`($ctx)
stw @V[5],`$LITTLE_ENDIAN^20`($ctx)
addc @V[13],@V[13],$a1
stw @V[6],24($ctx)
stw @V[7],28($ctx)
stw @V[6],`$LITTLE_ENDIAN^24`($ctx)
stw @V[7],`$LITTLE_ENDIAN^28`($ctx)
adde @V[12],@V[12],$a0
stw @V[8],32($ctx)
stw @V[9],36($ctx)
stw @V[8],`$LITTLE_ENDIAN^32`($ctx)
stw @V[9],`$LITTLE_ENDIAN^36`($ctx)
addc @V[15],@V[15],$a3
stw @V[10],40($ctx)
stw @V[11],44($ctx)
stw @V[10],`$LITTLE_ENDIAN^40`($ctx)
stw @V[11],`$LITTLE_ENDIAN^44`($ctx)
adde @V[14],@V[14],$a2
stw @V[12],48($ctx)
stw @V[13],52($ctx)
stw @V[14],56($ctx)
stw @V[15],60($ctx)
stw @V[12],`$LITTLE_ENDIAN^48`($ctx)
stw @V[13],`$LITTLE_ENDIAN^52`($ctx)
stw @V[14],`$LITTLE_ENDIAN^56`($ctx)
stw @V[15],`$LITTLE_ENDIAN^60`($ctx)
addi $inp,$inp,`16*$SZ` ; advance inp
$PUSH $inp,`$FRAME-$SIZE_T*23`($sp)
@ -685,46 +727,46 @@ LPICmeup:
.space `64-9*4`
___
$code.=<<___ if ($SZ==8);
.long 0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd
.long 0xb5c0fbcf,0xec4d3b2f,0xe9b5dba5,0x8189dbbc
.long 0x3956c25b,0xf348b538,0x59f111f1,0xb605d019
.long 0x923f82a4,0xaf194f9b,0xab1c5ed5,0xda6d8118
.long 0xd807aa98,0xa3030242,0x12835b01,0x45706fbe
.long 0x243185be,0x4ee4b28c,0x550c7dc3,0xd5ffb4e2
.long 0x72be5d74,0xf27b896f,0x80deb1fe,0x3b1696b1
.long 0x9bdc06a7,0x25c71235,0xc19bf174,0xcf692694
.long 0xe49b69c1,0x9ef14ad2,0xefbe4786,0x384f25e3
.long 0x0fc19dc6,0x8b8cd5b5,0x240ca1cc,0x77ac9c65
.long 0x2de92c6f,0x592b0275,0x4a7484aa,0x6ea6e483
.long 0x5cb0a9dc,0xbd41fbd4,0x76f988da,0x831153b5
.long 0x983e5152,0xee66dfab,0xa831c66d,0x2db43210
.long 0xb00327c8,0x98fb213f,0xbf597fc7,0xbeef0ee4
.long 0xc6e00bf3,0x3da88fc2,0xd5a79147,0x930aa725
.long 0x06ca6351,0xe003826f,0x14292967,0x0a0e6e70
.long 0x27b70a85,0x46d22ffc,0x2e1b2138,0x5c26c926
.long 0x4d2c6dfc,0x5ac42aed,0x53380d13,0x9d95b3df
.long 0x650a7354,0x8baf63de,0x766a0abb,0x3c77b2a8
.long 0x81c2c92e,0x47edaee6,0x92722c85,0x1482353b
.long 0xa2bfe8a1,0x4cf10364,0xa81a664b,0xbc423001
.long 0xc24b8b70,0xd0f89791,0xc76c51a3,0x0654be30
.long 0xd192e819,0xd6ef5218,0xd6990624,0x5565a910
.long 0xf40e3585,0x5771202a,0x106aa070,0x32bbd1b8
.long 0x19a4c116,0xb8d2d0c8,0x1e376c08,0x5141ab53
.long 0x2748774c,0xdf8eeb99,0x34b0bcb5,0xe19b48a8
.long 0x391c0cb3,0xc5c95a63,0x4ed8aa4a,0xe3418acb
.long 0x5b9cca4f,0x7763e373,0x682e6ff3,0xd6b2b8a3
.long 0x748f82ee,0x5defb2fc,0x78a5636f,0x43172f60
.long 0x84c87814,0xa1f0ab72,0x8cc70208,0x1a6439ec
.long 0x90befffa,0x23631e28,0xa4506ceb,0xde82bde9
.long 0xbef9a3f7,0xb2c67915,0xc67178f2,0xe372532b
.long 0xca273ece,0xea26619c,0xd186b8c7,0x21c0c207
.long 0xeada7dd6,0xcde0eb1e,0xf57d4f7f,0xee6ed178
.long 0x06f067aa,0x72176fba,0x0a637dc5,0xa2c898a6
.long 0x113f9804,0xbef90dae,0x1b710b35,0x131c471b
.long 0x28db77f5,0x23047d84,0x32caab7b,0x40c72493
.long 0x3c9ebe0a,0x15c9bebc,0x431d67c4,0x9c100d4c
.long 0x4cc5d4be,0xcb3e42b6,0x597f299c,0xfc657e2a
.long 0x5fcb6fab,0x3ad6faec,0x6c44198c,0x4a475817
.quad 0x428a2f98d728ae22,0x7137449123ef65cd
.quad 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
.quad 0x3956c25bf348b538,0x59f111f1b605d019
.quad 0x923f82a4af194f9b,0xab1c5ed5da6d8118
.quad 0xd807aa98a3030242,0x12835b0145706fbe
.quad 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
.quad 0x72be5d74f27b896f,0x80deb1fe3b1696b1
.quad 0x9bdc06a725c71235,0xc19bf174cf692694
.quad 0xe49b69c19ef14ad2,0xefbe4786384f25e3
.quad 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
.quad 0x2de92c6f592b0275,0x4a7484aa6ea6e483
.quad 0x5cb0a9dcbd41fbd4,0x76f988da831153b5
.quad 0x983e5152ee66dfab,0xa831c66d2db43210
.quad 0xb00327c898fb213f,0xbf597fc7beef0ee4
.quad 0xc6e00bf33da88fc2,0xd5a79147930aa725
.quad 0x06ca6351e003826f,0x142929670a0e6e70
.quad 0x27b70a8546d22ffc,0x2e1b21385c26c926
.quad 0x4d2c6dfc5ac42aed,0x53380d139d95b3df
.quad 0x650a73548baf63de,0x766a0abb3c77b2a8
.quad 0x81c2c92e47edaee6,0x92722c851482353b
.quad 0xa2bfe8a14cf10364,0xa81a664bbc423001
.quad 0xc24b8b70d0f89791,0xc76c51a30654be30
.quad 0xd192e819d6ef5218,0xd69906245565a910
.quad 0xf40e35855771202a,0x106aa07032bbd1b8
.quad 0x19a4c116b8d2d0c8,0x1e376c085141ab53
.quad 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
.quad 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
.quad 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
.quad 0x748f82ee5defb2fc,0x78a5636f43172f60
.quad 0x84c87814a1f0ab72,0x8cc702081a6439ec
.quad 0x90befffa23631e28,0xa4506cebde82bde9
.quad 0xbef9a3f7b2c67915,0xc67178f2e372532b
.quad 0xca273eceea26619c,0xd186b8c721c0c207
.quad 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
.quad 0x06f067aa72176fba,0x0a637dc5a2c898a6
.quad 0x113f9804bef90dae,0x1b710b35131c471b
.quad 0x28db77f523047d84,0x32caab7b40c72493
.quad 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
.quad 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
.quad 0x5fcb6fab3ad6faec,0x6c44198c4a475817
___
$code.=<<___ if ($SZ==4);
.long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5