Quelle  sha512p8-ppc.pl

#!/usr/bin/env perl

# ====================================================================
# Written by Andy Polyakov, @dot-asm, initially for use in the OpenSSL
# project. The module is dual licensed under OpenSSL and CRYPTOGAMS
# licenses depending on where you obtain it. For further details see
# https://github.com/dot-asm/cryptogams/.
# ====================================================================

# SHA256/512 for PowerISA v2.07.
#
# Accurate performance measurements are problematic, because it's
# always virtualized setup with possibly throttled processor.
# Relative comparison is therefore more informative. This module is
# ~60% faster than integer-only sha512-ppc.pl. To anchor to something
# else, SHA256 is 24% slower than sha1-ppc.pl and 2.5x slower than
# hardware-assisted aes-128-cbc encrypt. SHA512 is 20% faster than
# sha1-ppc.pl and 1.6x slower than aes-128-cbc. Another interesting
# result is degree of computational resources' utilization. POWER8 is
# "massively multi-threaded chip" and difference between single- and
# maximum multi-process benchmark results tells that utilization is
# whooping 94%. For sha512-ppc.pl we get [not unimpressive] 84% and
# for sha1-ppc.pl - 73%. 100% means that multi-process result equals
# to single-process one, given that all threads end up on the same
# physical core.
#
######################################################################
# Believed-to-be-accurate results in cycles per processed byte [on
# little-endian system]. Numbers in square brackets are for 64-bit
# build of sha512-ppc.pl, presented for reference.
#
# POWER8 POWER9
# SHA256 9.7 [15.8] 11.2 [12.5]
# SHA512 6.1 [10.3] 7.0 [7.9]

$flavour=shift;
$output =shift;

if ($flavour =~ /64/) {
 $SIZE_T=8;
 $LRSAVE=2*$SIZE_T;
 $STU="stdu";
 $POP="ld";
 $PUSH="std";
} elsif ($flavour =~ /32/) {
 $SIZE_T=4;
 $LRSAVE=$SIZE_T;
 $STU="stwu";
 $POP="lwz";
 $PUSH="stw";
} else { die "nonsense $flavour"; }

$LENDIAN=($flavour=~/le/);

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $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";

open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!";

if ($output =~ /512/) {
 $bits=512;
 $SZ=8;
 $sz="d";
 $rounds=80;
} else {
 $bits=256;
 $SZ=4;
 $sz="w";
 $rounds=64;
}

$func="sha${bits}_block_p8";
$LOCALS=8*$SIZE_T+8*16;
$FRAME=$LOCALS+9*16+6*$SIZE_T;

$sp ="r1";
$toc="r2";
$ctx="r3";
$inp="r4";
$num="r5";
$Tbl="r6";
$idx="r7";
$lrsave="r8";
$offload="r11";
$vrsave="r12";
@I = ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70) = (0,map("r$_",(10,26..31)));

@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("v$_",(0..7));
@X=map("v$_",(8..19,24..27));
($Ki,$Func,$Sigma,$lemask)=map("v$_",(28..31));

sub ROUND {
my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
my $j=($i+1)%16;
my $k=($i+2)%8;

$code.=<<___  if ($i<15 && ($i%(16/$SZ))==(16/$SZ-1));
 lvx_u  @X[$i+1],0,$inp  ; load X[i] in advance
 addi  $inp,$inp,16
___
$code.=<<___  if ($i<16 && ($i%(16/$SZ)));
 vsldoi  @X[$i],@X[$i-1],@X[$i-1],$SZ
___
$code.=<<___  if ($LENDIAN && $i<16 && ($i%(16/$SZ))==0);
 vperm  @X[$i],@X[$i],@X[$i],$lemask
___
$code.=<<___  if ($i>=15);
 vshasigma${sz} $Sigma,@X[($j+1)%16],0,0
 vaddu${sz}m @X[$j],@X[$j],$Sigma
 vshasigma${sz} $Sigma,@X[($j+14)%16],0,15
 vaddu${sz}m @X[$j],@X[$j],$Sigma
 vaddu${sz}m @X[$j],@X[$j],@X[($j+9)%16]
___
$code.=<<___;
 vaddu${sz}m $h,$h,@X[$i%16]  ; h+=X[i]
 vsel  $Func,$g,$f,$e  ; Ch(e,f,g)
 vaddu${sz}m $g,$g,$Ki  ; future h+=K[i]
 vaddu${sz}m $h,$h,$Func  ; h+=Ch(e,f,g)
 vshasigma${sz} $Sigma,$e,1,15  ; Sigma1(e)
 vaddu${sz}m $h,$h,$Sigma  ; h+=Sigma1(e)
 vxor  $Func,$a,$b
 vsel  $Func,$b,$c,$Func ; Maj(a,b,c)
 vaddu${sz}m $d,$d,$h  ; d+=h
 vshasigma${sz} $Sigma,$a,1,0  ; Sigma0(a)
 vaddu${sz}m $Sigma,$Sigma,$Func ; Sigma0(a)+Maj(a,b,c)
 vaddu${sz}m $h,$h,$Sigma  ; h+=Sigma0(a)+Maj(a,b,c)
 lvx  $Ki,@I[$k],$idx  ; load next K[i]
___
$code.=<<___  if ($k == 7);
 addi  $idx,$idx,0x80
___
}

$code=<<___;
.machine "any"
.text

.globl $func
.align 6
$func:
 $STU  $sp,-$FRAME($sp)
 mflr  $lrsave
 li  r10,`$LOCALS+15`
 li  r11,`$LOCALS+31`
 stvx  v24,r10,$sp  # ABI says so
 addi  r10,r10,32
 mfspr  $vrsave,256
 stvx  v25,r11,$sp
 addi  r11,r11,32
 stvx  v26,r10,$sp
 addi  r10,r10,32
 stvx  v27,r11,$sp
 addi  r11,r11,32
 stvx  v28,r10,$sp
 addi  r10,r10,32
 stvx  v29,r11,$sp
 addi  r11,r11,32
 stvx  v30,r10,$sp
 stvx  v31,r11,$sp
 li  r11,-4096+255  # 0xfffff0ff
 stw  $vrsave,`$FRAME-6*$SIZE_T-4`($sp) # save vrsave
 li  $x10,0x10
 $PUSH  r26,`$FRAME-6*$SIZE_T`($sp)
 li  $x20,0x20
 $PUSH  r27,`$FRAME-5*$SIZE_T`($sp)
 li  $x30,0x30
 $PUSH  r28,`$FRAME-4*$SIZE_T`($sp)
 li  $x40,0x40
 $PUSH  r29,`$FRAME-3*$SIZE_T`($sp)
 li  $x50,0x50
 $PUSH  r30,`$FRAME-2*$SIZE_T`($sp)
 li  $x60,0x60
 $PUSH  r31,`$FRAME-1*$SIZE_T`($sp)
 li  $x70,0x70
 $PUSH  $lrsave,`$FRAME+$LRSAVE`($sp)
 mtspr  256,r11

 bl  LPICmeup
 addi  $offload,$sp,`8*$SIZE_T+15`
___
$code.=<<___  if ($LENDIAN);
 li  $idx,8
 lvsl  $lemask,0,$idx
 vspltisb $Ki,0x0f
 vxor  $lemask,$lemask,$Ki
___
$code.=<<___  if ($SZ==4);
 lvx_4w  $A,$x00,$ctx
 lvx_4w  $E,$x10,$ctx
 vsldoi  $B,$A,$A,4  # unpack
 vsldoi  $C,$A,$A,8
 vsldoi  $D,$A,$A,12
 vsldoi  $F,$E,$E,4
 vsldoi  $G,$E,$E,8
 vsldoi  $H,$E,$E,12
___
$code.=<<___  if ($SZ==8);
 lvx_u  $A,$x00,$ctx
 lvx_u  $C,$x10,$ctx
 lvx_u  $E,$x20,$ctx
 vsldoi  $B,$A,$A,8  # unpack
 lvx_u  $G,$x30,$ctx
 vsldoi  $D,$C,$C,8
 vsldoi  $F,$E,$E,8
 vsldoi  $H,$G,$G,8
___
$code.=<<___;
 li  r0,`($rounds-16)/16` # inner loop counter
 b  Loop
.align 5
Loop:
 lvx  $Ki,$x00,$Tbl
 lvx_u  @X[0],0,$inp
 addi  $inp,$inp,16
 mr  $idx,$Tbl  # copy $Tbl
 stvx  $A,$x00,$offload # offload $A-$H
 stvx  $B,$x10,$offload
 stvx  $C,$x20,$offload
 stvx  $D,$x30,$offload
 stvx  $E,$x40,$offload
 stvx  $F,$x50,$offload
 stvx  $G,$x60,$offload
 stvx  $H,$x70,$offload
 vaddu${sz}m $H,$H,$Ki  # h+K[i]
 lvx  $Ki,$x10,$Tbl
___
for ($i=0;$i<16;$i++) { &ROUND($i,@V); unshift(@V,pop(@V)); }
$code.=<<___;
 mtctr  r0
 b  L16_xx
.align 5
L16_xx:
___
for (;$i<32;$i++) { &ROUND($i,@V); unshift(@V,pop(@V)); }
$code.=<<___;
 bdnz  L16_xx

 lvx  @X[2],$x00,$offload
 subic.  $num,$num,1
 lvx  @X[3],$x10,$offload
 vaddu${sz}m $A,$A,@X[2]
 lvx  @X[4],$x20,$offload
 vaddu${sz}m $B,$B,@X[3]
 lvx  @X[5],$x30,$offload
 vaddu${sz}m $C,$C,@X[4]
 lvx  @X[6],$x40,$offload
 vaddu${sz}m $D,$D,@X[5]
 lvx  @X[7],$x50,$offload
 vaddu${sz}m $E,$E,@X[6]
 lvx  @X[8],$x60,$offload
 vaddu${sz}m $F,$F,@X[7]
 lvx  @X[9],$x70,$offload
 vaddu${sz}m $G,$G,@X[8]
 vaddu${sz}m $H,$H,@X[9]
 bne  Loop
___
$code.=<<___  if ($SZ==4);
 lvx  @X[0],$x20,$idx
 vperm  $A,$A,$B,$Ki  # pack the answer
 lvx  @X[1],$x30,$idx
 vperm  $E,$E,$F,$Ki
 vperm  $A,$A,$C,@X[0]
 vperm  $E,$E,$G,@X[0]
 vperm  $A,$A,$D,@X[1]
 vperm  $E,$E,$H,@X[1]
 stvx_4w  $A,$x00,$ctx
 stvx_4w  $E,$x10,$ctx
___
$code.=<<___  if ($SZ==8);
 vperm  $A,$A,$B,$Ki  # pack the answer
 vperm  $C,$C,$D,$Ki
 vperm  $E,$E,$F,$Ki
 vperm  $G,$G,$H,$Ki
 stvx_u  $A,$x00,$ctx
 stvx_u  $C,$x10,$ctx
 stvx_u  $E,$x20,$ctx
 stvx_u  $G,$x30,$ctx
___
$code.=<<___;
 addi  $offload,$sp,`$LOCALS+15`
 mtlr  $lrsave
 mtspr  256,$vrsave
 lvx  v24,$x00,$offload # ABI says so
 lvx  v25,$x10,$offload
 lvx  v26,$x20,$offload
 lvx  v27,$x30,$offload
 lvx  v28,$x40,$offload
 lvx  v29,$x50,$offload
 lvx  v30,$x60,$offload
 lvx  v31,$x70,$offload
 $POP  r26,`$FRAME-6*$SIZE_T`($sp)
 $POP  r27,`$FRAME-5*$SIZE_T`($sp)
 $POP  r28,`$FRAME-4*$SIZE_T`($sp)
 $POP  r29,`$FRAME-3*$SIZE_T`($sp)
 $POP  r30,`$FRAME-2*$SIZE_T`($sp)
 $POP  r31,`$FRAME-1*$SIZE_T`($sp)
 addi  $sp,$sp,$FRAME
 blr
 .long  0
 .byte  0,12,4,1,0x80,6,3,0
 .long  0
.size $func,.-$func
___

# Ugly hack here, because PPC assembler syntax seem to vary too
# much from platforms to platform...
$code.=<<___;
.align 6
LPICmeup:
 mflr r0
 bcl 20,31,\$+4
 mflr $Tbl ; vvvvvv "distance" between . and 1st data entry
 addi $Tbl,$Tbl,`64-8`
 mtlr r0
 blr
 .long 0
 .byte 0,12,0x14,0,0,0,0,0
 .space `64-9*4`
___

if ($SZ==8) {
    local *table = sub {
 foreach(@_) { $code.=".quad $_,$_\n"; }
    };
    table(
 "0x428a2f98d728ae22","0x7137449123ef65cd",
 "0xb5c0fbcfec4d3b2f","0xe9b5dba58189dbbc",
 "0x3956c25bf348b538","0x59f111f1b605d019",
 "0x923f82a4af194f9b","0xab1c5ed5da6d8118",
 "0xd807aa98a3030242","0x12835b0145706fbe",
 "0x243185be4ee4b28c","0x550c7dc3d5ffb4e2",
 "0x72be5d74f27b896f","0x80deb1fe3b1696b1",
 "0x9bdc06a725c71235","0xc19bf174cf692694",
 "0xe49b69c19ef14ad2","0xefbe4786384f25e3",
 "0x0fc19dc68b8cd5b5","0x240ca1cc77ac9c65",
 "0x2de92c6f592b0275","0x4a7484aa6ea6e483",
 "0x5cb0a9dcbd41fbd4","0x76f988da831153b5",
 "0x983e5152ee66dfab","0xa831c66d2db43210",
 "0xb00327c898fb213f","0xbf597fc7beef0ee4",
 "0xc6e00bf33da88fc2","0xd5a79147930aa725",
 "0x06ca6351e003826f","0x142929670a0e6e70",
 "0x27b70a8546d22ffc","0x2e1b21385c26c926",
 "0x4d2c6dfc5ac42aed","0x53380d139d95b3df",
 "0x650a73548baf63de","0x766a0abb3c77b2a8",
 "0x81c2c92e47edaee6","0x92722c851482353b",
 "0xa2bfe8a14cf10364","0xa81a664bbc423001",
 "0xc24b8b70d0f89791","0xc76c51a30654be30",
 "0xd192e819d6ef5218","0xd69906245565a910",
 "0xf40e35855771202a","0x106aa07032bbd1b8",
 "0x19a4c116b8d2d0c8","0x1e376c085141ab53",
 "0x2748774cdf8eeb99","0x34b0bcb5e19b48a8",
 "0x391c0cb3c5c95a63","0x4ed8aa4ae3418acb",
 "0x5b9cca4f7763e373","0x682e6ff3d6b2b8a3",
 "0x748f82ee5defb2fc","0x78a5636f43172f60",
 "0x84c87814a1f0ab72","0x8cc702081a6439ec",
 "0x90befffa23631e28","0xa4506cebde82bde9",
 "0xbef9a3f7b2c67915","0xc67178f2e372532b",
 "0xca273eceea26619c","0xd186b8c721c0c207",
 "0xeada7dd6cde0eb1e","0xf57d4f7fee6ed178",
 "0x06f067aa72176fba","0x0a637dc5a2c898a6",
 "0x113f9804bef90dae","0x1b710b35131c471b",
 "0x28db77f523047d84","0x32caab7b40c72493",
 "0x3c9ebe0a15c9bebc","0x431d67c49c100d4c",
 "0x4cc5d4becb3e42b6","0x597f299cfc657e2a",
 "0x5fcb6fab3ad6faec","0x6c44198c4a475817","0");
$code.=<<___ if (!$LENDIAN);
.quad 0x0001020304050607,0x1011121314151617
___
$code.=<<___ if ($LENDIAN); # quad-swapped
.quad 0x1011121314151617,0x0001020304050607
___
} else {
    local *table = sub {
 foreach(@_) { $code.=".long $_,$_,$_,$_\n"; }
    };
    table(
 "0x428a2f98","0x71374491","0xb5c0fbcf","0xe9b5dba5",
 "0x3956c25b","0x59f111f1","0x923f82a4","0xab1c5ed5",
 "0xd807aa98","0x12835b01","0x243185be","0x550c7dc3",
 "0x72be5d74","0x80deb1fe","0x9bdc06a7","0xc19bf174",
 "0xe49b69c1","0xefbe4786","0x0fc19dc6","0x240ca1cc",
 "0x2de92c6f","0x4a7484aa","0x5cb0a9dc","0x76f988da",
 "0x983e5152","0xa831c66d","0xb00327c8","0xbf597fc7",
 "0xc6e00bf3","0xd5a79147","0x06ca6351","0x14292967",
 "0x27b70a85","0x2e1b2138","0x4d2c6dfc","0x53380d13",
 "0x650a7354","0x766a0abb","0x81c2c92e","0x92722c85",
 "0xa2bfe8a1","0xa81a664b","0xc24b8b70","0xc76c51a3",
 "0xd192e819","0xd6990624","0xf40e3585","0x106aa070",
 "0x19a4c116","0x1e376c08","0x2748774c","0x34b0bcb5",
 "0x391c0cb3","0x4ed8aa4a","0x5b9cca4f","0x682e6ff3",
 "0x748f82ee","0x78a5636f","0x84c87814","0x8cc70208",
 "0x90befffa","0xa4506ceb","0xbef9a3f7","0xc67178f2","0");
$code.=<<___ if (!$LENDIAN);
.long 0x00010203,0x10111213,0x10111213,0x10111213
.long 0x00010203,0x04050607,0x10111213,0x10111213
.long 0x00010203,0x04050607,0x08090a0b,0x10111213
___
$code.=<<___ if ($LENDIAN); # word-swapped
.long 0x10111213,0x10111213,0x10111213,0x00010203
.long 0x10111213,0x10111213,0x04050607,0x00010203
.long 0x10111213,0x08090a0b,0x04050607,0x00010203
___
}
$code.=<<___;
.asciz "SHA${bits} for PowerISA 2.07, CRYPTOGAMS by <appro\@openssl.org>"
.align 2
___

$code =~ s/\`([^\`]*)\`/eval $1/gem;
print $code;
close STDOUT;