# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
-# project. Rights for redistribution and usage in source and binary
-# forms are granted according to the OpenSSL license.
+# 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/.
# ====================================================================
-# I let hardware handle unaligned input, except on page boundaries
+# 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].
+#
+# (*) this means that this module is inappropriate for PPC403? Does
+# anybody know if pre-POWER3 can sustain unaligned load?
-# gcc-4.0.0 -m64 -m32
-# --------------------------
-# sha1 +76% +59%
+# -m64 -m32
+# ----------------------------------
+# PPC970,gcc-4.0.0 +76% +59%
+# Power6,xlc-7 +68% +33%
-$output = shift;
+$flavour = shift;
-if ($output =~ /64\.s/) {
+if ($flavour =~ /64/) {
$SIZE_T =8;
- $RZONE =288;
+ $LRSAVE =2*$SIZE_T;
$UCMP ="cmpld";
$STU ="stdu";
$POP ="ld";
$PUSH ="std";
-} elsif ($output =~ /32\.s/) {
+} elsif ($flavour =~ /32/) {
$SIZE_T =4;
- $RZONE =224;
+ $LRSAVE =$SIZE_T;
$UCMP ="cmplw";
$STU ="stwu";
$POP ="lwz";
$PUSH ="stw";
-} else { die "nonsense $output"; }
+} else { die "nonsense $flavour"; }
-( defined shift || open STDOUT,"| $^X ../perlasm/ppc-xlate.pl $output" ) ||
- die "can't call ../perlasm/ppc-xlate.pl: $!";
+# Define endianess based on flavour
+# i.e.: linux64le
+$LITTLE_ENDIAN=0;
+if ($flavour =~ /le$/) {
+ die "little-endian is 64-bit only: $flavour" if ($SIZE_T == 4);
+ $LITTLE_ENDIAN=1;
+}
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
+die "can't locate ppc-xlate.pl";
-$FRAME=24*$SIZE_T;
+open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
+
+$FRAME=24*$SIZE_T+64;
+$LOCALS=6*$SIZE_T;
$K ="r0";
$sp ="r1";
@X=("r16","r17","r18","r19","r20","r21","r22","r23",
"r24","r25","r26","r27","r28","r29","r30","r31");
+sub loadbe {
+my ($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]
}
$code=<<___;
-.machine any
+.machine "any"
.text
-.globl .sha1_block_asm_data_order
+.globl .sha1_block_data_order
.align 4
-.sha1_block_asm_data_order:
+.sha1_block_data_order:
+ $STU $sp,-$FRAME($sp)
mflr r0
- $STU $sp,`-($FRAME+64+$RZONE)`($sp)
- $PUSH r0,`$FRAME-$SIZE_T*18`($sp)
$PUSH r15,`$FRAME-$SIZE_T*17`($sp)
$PUSH r16,`$FRAME-$SIZE_T*16`($sp)
$PUSH r17,`$FRAME-$SIZE_T*15`($sp)
$PUSH r29,`$FRAME-$SIZE_T*3`($sp)
$PUSH r30,`$FRAME-$SIZE_T*2`($sp)
$PUSH r31,`$FRAME-$SIZE_T*1`($sp)
+ $PUSH r0,`$FRAME+$LRSAVE`($sp)
lwz $A,0($ctx)
lwz $B,4($ctx)
lwz $C,8($ctx)
Laligned:
mtctr $num
bl Lsha1_block_private
-Ldone:
- $POP r0,`$FRAME-$SIZE_T*18`($sp)
- $POP r15,`$FRAME-$SIZE_T*17`($sp)
- $POP r16,`$FRAME-$SIZE_T*16`($sp)
- $POP r17,`$FRAME-$SIZE_T*15`($sp)
- $POP r18,`$FRAME-$SIZE_T*14`($sp)
- $POP r19,`$FRAME-$SIZE_T*13`($sp)
- $POP r20,`$FRAME-$SIZE_T*12`($sp)
- $POP r21,`$FRAME-$SIZE_T*11`($sp)
- $POP r22,`$FRAME-$SIZE_T*10`($sp)
- $POP r23,`$FRAME-$SIZE_T*9`($sp)
- $POP r24,`$FRAME-$SIZE_T*8`($sp)
- $POP r25,`$FRAME-$SIZE_T*7`($sp)
- $POP r26,`$FRAME-$SIZE_T*6`($sp)
- $POP r27,`$FRAME-$SIZE_T*5`($sp)
- $POP r28,`$FRAME-$SIZE_T*4`($sp)
- $POP r29,`$FRAME-$SIZE_T*3`($sp)
- $POP r30,`$FRAME-$SIZE_T*2`($sp)
- $POP r31,`$FRAME-$SIZE_T*1`($sp)
- mtlr r0
- addi $sp,$sp,`$FRAME+64+$RZONE`
- blr
-___
+ b Ldone
-# PowerPC specification allows an implementation to be ill-behaved
-# upon unaligned access which crosses page boundary. "Better safe
-# than sorry" principle makes me treat it specially. But I don't
-# look for particular offending word, but rather for 64-byte input
-# block which crosses the boundary. Once found that block is aligned
-# and hashed separately...
-$code.=<<___;
+; PowerPC specification allows an implementation to be ill-behaved
+; upon unaligned access which crosses page boundary. "Better safe
+; than sorry" principle makes me treat it specially. But I don't
+; look for particular offending word, but rather for 64-byte input
+; block which crosses the boundary. Once found that block is aligned
+; and hashed separately...
.align 4
Lunaligned:
- li $t1,4096
- subf $t1,$inp,$t1
+ subfic $t1,$inp,4096
andi. $t1,$t1,4095 ; distance to closest page boundary
srwi. $t1,$t1,6 ; t1/=64
beq Lcross_page
$UCMP $num,$t1
ble- Laligned ; didn't cross the page boundary
mtctr $t1
- subf $num,$t1,$num
+ subfc $num,$t1,$num
bl Lsha1_block_private
Lcross_page:
li $t1,16
mtctr $t1
- addi r20,$sp,$FRAME ; spot below the frame
+ addi r20,$sp,$LOCALS ; spot within the frame
Lmemcpy:
lbz r16,0($inp)
lbz r17,1($inp)
addi r20,r20,4
bdnz Lmemcpy
- $PUSH $inp,`$FRAME-$SIZE_T*19`($sp)
+ $PUSH $inp,`$FRAME-$SIZE_T*18`($sp)
li $t1,1
- addi $inp,$sp,$FRAME
+ addi $inp,$sp,$LOCALS
mtctr $t1
bl Lsha1_block_private
- $POP $inp,`$FRAME-$SIZE_T*19`($sp)
+ $POP $inp,`$FRAME-$SIZE_T*18`($sp)
addic. $num,$num,-1
bne- Lunaligned
- b Ldone
+
+Ldone:
+ $POP r0,`$FRAME+$LRSAVE`($sp)
+ $POP r15,`$FRAME-$SIZE_T*17`($sp)
+ $POP r16,`$FRAME-$SIZE_T*16`($sp)
+ $POP r17,`$FRAME-$SIZE_T*15`($sp)
+ $POP r18,`$FRAME-$SIZE_T*14`($sp)
+ $POP r19,`$FRAME-$SIZE_T*13`($sp)
+ $POP r20,`$FRAME-$SIZE_T*12`($sp)
+ $POP r21,`$FRAME-$SIZE_T*11`($sp)
+ $POP r22,`$FRAME-$SIZE_T*10`($sp)
+ $POP r23,`$FRAME-$SIZE_T*9`($sp)
+ $POP r24,`$FRAME-$SIZE_T*8`($sp)
+ $POP r25,`$FRAME-$SIZE_T*7`($sp)
+ $POP r26,`$FRAME-$SIZE_T*6`($sp)
+ $POP r27,`$FRAME-$SIZE_T*5`($sp)
+ $POP r28,`$FRAME-$SIZE_T*4`($sp)
+ $POP r29,`$FRAME-$SIZE_T*3`($sp)
+ $POP r30,`$FRAME-$SIZE_T*2`($sp)
+ $POP r31,`$FRAME-$SIZE_T*1`($sp)
+ mtlr r0
+ addi $sp,$sp,$FRAME
+ blr
+ .long 0
+ .byte 0,12,4,1,0x80,18,3,0
+ .long 0
___
# This is private block function, which uses tailored calling
addi $inp,$inp,`16*4`
bdnz- Lsha1_block_private
blr
+ .long 0
+ .byte 0,12,0x14,0,0,0,0,0
+.size .sha1_block_data_order,.-.sha1_block_data_order
+___
+$code.=<<___;
+.asciz "SHA1 block transform for PPC, CRYPTOGAMS by <appro\@fy.chalmers.se>"
___
$code =~ s/\`([^\`]*)\`/eval $1/gem;
projects / openssl.git / blobdiff