-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the Apache License 2.0 (the "License"). You may not use
+# this file except in compliance with the License. You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
#
# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# project. Rights for redistribution and usage in source and binary
-# forms are granted according to the OpenSSL license.
+# forms are granted according to the License.
# ====================================================================
#
# sha256/512_block procedure for x86_64.
# level parallelism, on a given CPU implementation in this case.
#
# Special note on Intel EM64T. While Opteron CPU exhibits perfect
-# perfromance ratio of 1.5 between 64- and 32-bit flavors [see above],
+# performance ratio of 1.5 between 64- and 32-bit flavors [see above],
# [currently available] EM64T CPUs apparently are far from it. On the
# contrary, 64-bit version, sha512_block, is ~30% *slower* than 32-bit
# sha256_block:-( This is presumably because 64-bit shifts/rotates
#
# Optimization including one of Pavel Semjanov's ideas, alternative
# Maj, resulted in >=5% improvement on most CPUs, +20% SHA256 and
-# unfortunately -10% SHA512 on P4 [which nobody should care about
+# unfortunately -2% SHA512 on P4 [which nobody should care about
# that much].
#
# June 2012.
# significant 128-bit halves and data from second to most significant.
# The data is then processed with same SIMD instruction sequence as
# for AVX, but with %ymm as operands. Side effect is increased stack
-# frame, 448 additional bytes in SHA256 and 1152 in SHA512.
+# frame, 448 additional bytes in SHA256 and 1152 in SHA512, and 1.2KB
+# code size increase.
+#
+# March 2014.
+#
+# Add support for Intel SHA Extensions.
######################################################################
# Current performance in cycles per processed byte (less is better):
#
# SHA256 SSSE3 AVX/XOP(*) SHA512 AVX/XOP(*)
#
-# AMD K8 15.1 - - 9.70 -
-# P4 17.5 - - 33.4 -
-# Core 2 15.5 13.8(+12%) - 10.3 -
-# Westmere 15.1 12.7(+19%) - 9.72 -
+# AMD K8 14.9 - - 9.57 -
+# P4 17.3 - - 30.8 -
+# Core 2 15.6 13.8(+13%) - 9.97 -
+# Westmere 14.8 12.3(+19%) - 9.58 -
# Sandy Bridge 17.4 14.2(+23%) 11.6(+50%(**)) 11.2 8.10(+38%(**))
# Ivy Bridge 12.6 10.5(+20%) 10.3(+22%) 8.17 7.22(+13%)
-# Bulldozer 21.5 13.7(+57%) 13.7(+57%(***)) 13.5 8.58(+57%)
+# Haswell 12.2 9.28(+31%) 7.80(+56%) 7.66 5.40(+42%)
+# Skylake 11.4 9.03(+26%) 7.70(+48%) 7.25 5.20(+40%)
+# Bulldozer 21.1 13.6(+54%) 13.6(+54%(***)) 13.5 8.58(+57%)
+# Ryzen 11.0 9.02(+22%) 2.05(+440%) 7.05 5.67(+20%)
# VIA Nano 23.0 16.5(+39%) - 14.7 -
-# Atom 23.0 18.7(+23%) - 14.7 -
+# Atom 23.0 18.9(+22%) - 14.7 -
+# Silvermont 27.4 20.6(+33%) - 17.5 -
+# Knights L 27.4 21.0(+30%) 19.6(+40%) 17.5 12.8(+37%)
+# Goldmont 18.9 14.3(+32%) 4.16(+350%) 12.0 -
#
-# (*) whichever best applicable;
+# (*) whichever best applicable, including SHAEXT;
# (**) switch from ror to shrd stands for fair share of improvement;
# (***) execution time is fully determined by remaining integer-only
# part, body_00_15; reducing the amount of SIMD instructions
# below certain limit makes no difference/sense; to conserve
# space SHA256 XOP code path is therefore omitted;
-$flavour = shift;
-$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/ || $output =~ /\.asm$/);
$avx = ($1>=10) + ($1>=11);
}
-open OUT,"| \"$^X\" $xlate $flavour $output";
+if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
+ $avx = ($2>=3.0) + ($2>3.0);
+}
+
+$shaext=1; ### set to zero if compiling for 1.0.1
+$avx=1 if (!$shaext && $avx);
+
+open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
+ or die "can't call $xlate: $!";
*STDOUT=*OUT;
if ($output =~ /512/) {
$_ctx="16*$SZ+0*8(%rsp)";
$_inp="16*$SZ+1*8(%rsp)";
$_end="16*$SZ+2*8(%rsp)";
-$_rsp="16*$SZ+3*8(%rsp)";
+$_rsp="`16*$SZ+3*8`(%rsp)";
$framesz="16*$SZ+4*8";
ror \$`$Sigma1[2]-$Sigma1[1]`,$a0
mov $f,$a2
- ror \$`$Sigma0[2]-$Sigma0[1]`,$a1
xor $e,$a0
+ ror \$`$Sigma0[2]-$Sigma0[1]`,$a1
xor $g,$a2 # f^g
mov $T1,`$SZ*($i&0xf)`(%rsp)
add ($Tbl),$T1 # T1+=K[round]
xor $a,$a1
- ror \$$Sigma1[0],$a0 # Sigma1(e)
xor $b,$a2 # a^b, b^c in next round
+ ror \$$Sigma1[0],$a0 # Sigma1(e)
mov $b,$h
- ror \$$Sigma0[0],$a1 # Sigma0(a)
and $a2,$a3
+ ror \$$Sigma0[0],$a1 # Sigma0(a)
add $a0,$T1 # T1+=Sigma1(e)
xor $a3,$h # h=Maj(a,b,c)=Ch(a^b,c,b)
add $T1,$d # d+=T1
add $T1,$h # h+=T1
-___
-$code.=<<___ if ($i>=15);
- mov `$SZ*(($i+2)&0xf)`(%rsp),$a0
-___
-$code.=<<___;
+
lea $STRIDE($Tbl),$Tbl # round++
+___
+$code.=<<___ if ($i<15);
add $a1,$h # h+=Sigma0(a)
-
___
($a2,$a3) = ($a3,$a2);
}
{ my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
$code.=<<___;
- #mov `$SZ*(($i+1)&0xf)`(%rsp),$a0
- mov `$SZ*(($i+14)&0xf)`(%rsp),$a1
+ mov `$SZ*(($i+1)&0xf)`(%rsp),$a0
+ mov `$SZ*(($i+14)&0xf)`(%rsp),$a2
mov $a0,$T1
ror \$`$sigma0[1]-$sigma0[0]`,$a0
- mov $a1,$a2
- ror \$`$sigma1[1]-$sigma1[0]`,$a1
+ add $a1,$a # modulo-scheduled h+=Sigma0(a)
+ mov $a2,$a1
+ ror \$`$sigma1[1]-$sigma1[0]`,$a2
xor $T1,$a0
shr \$$sigma0[2],$T1
ror \$$sigma0[0],$a0
- xor $a2,$a1
- shr \$$sigma1[2],$a2
+ xor $a1,$a2
+ shr \$$sigma1[2],$a1
+ ror \$$sigma1[0],$a2
xor $a0,$T1 # sigma0(X[(i+1)&0xf])
- ror \$$sigma1[0],$a1
+ xor $a1,$a2 # sigma1(X[(i+14)&0xf])
add `$SZ*(($i+9)&0xf)`(%rsp),$T1
- xor $a2,$a1 # sigma1(X[(i+14)&0xf])
add `$SZ*($i&0xf)`(%rsp),$T1
mov $e,$a0
- add $a1,$T1
+ add $a2,$T1
mov $a,$a1
___
&ROUND_00_15(@_);
.type $func,\@function,3
.align 16
$func:
+.cfi_startproc
___
$code.=<<___ if ($SZ==4 || $avx);
lea OPENSSL_ia32cap_P(%rip),%r11
mov 4(%r11),%r10d
mov 8(%r11),%r11d
___
+$code.=<<___ if ($SZ==4 && $shaext);
+ test \$`1<<29`,%r11d # check for SHA
+ jnz _shaext_shortcut
+___
$code.=<<___ if ($avx && $SZ==8);
test \$`1<<11`,%r10d # check for XOP
jnz .Lxop_shortcut
jnz .Lssse3_shortcut
___
$code.=<<___;
+ mov %rsp,%rax # copy %rsp
+.cfi_def_cfa_register %rax
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
- mov %rsp,%r11 # copy %rsp
+.cfi_push %r15
shl \$4,%rdx # num*16
sub \$$framesz,%rsp
lea ($inp,%rdx,$SZ),%rdx # inp+num*16*$SZ
mov $ctx,$_ctx # save ctx, 1st arg
mov $inp,$_inp # save inp, 2nd arh
mov %rdx,$_end # save end pointer, "3rd" arg
- mov %r11,$_rsp # save copy of %rsp
+ mov %rax,$_rsp # save copy of %rsp
+.cfi_cfa_expression $_rsp,deref,+8
.Lprologue:
mov $SZ*0($ctx),$A
jnz .Lrounds_16_xx
mov $_ctx,$ctx
+ add $a1,$A # modulo-scheduled h+=Sigma0(a)
lea 16*$SZ($inp),$inp
add $SZ*0($ctx),$A
jb .Lloop
mov $_rsp,%rsi
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+.cfi_def_cfa %rsi,8
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lepilogue:
ret
+.cfi_endproc
.size $func,.-$func
___
######################################################################
# SIMD code paths
#
+if ($SZ==4 && $shaext) {{{
+######################################################################
+# Intel SHA Extensions implementation of SHA256 update function.
+#
+my ($ctx,$inp,$num,$Tbl)=("%rdi","%rsi","%rdx","%rcx");
+
+my ($Wi,$ABEF,$CDGH,$TMP,$BSWAP,$ABEF_SAVE,$CDGH_SAVE)=map("%xmm$_",(0..2,7..10));
+my @MSG=map("%xmm$_",(3..6));
+
+$code.=<<___;
+.type sha256_block_data_order_shaext,\@function,3
+.align 64
+sha256_block_data_order_shaext:
+_shaext_shortcut:
+.cfi_startproc
+___
+$code.=<<___ if ($win64);
+ lea `-8-5*16`(%rsp),%rsp
+ movaps %xmm6,-8-5*16(%rax)
+ movaps %xmm7,-8-4*16(%rax)
+ movaps %xmm8,-8-3*16(%rax)
+ movaps %xmm9,-8-2*16(%rax)
+ movaps %xmm10,-8-1*16(%rax)
+.Lprologue_shaext:
+___
+$code.=<<___;
+ lea K256+0x80(%rip),$Tbl
+ movdqu ($ctx),$ABEF # DCBA
+ movdqu 16($ctx),$CDGH # HGFE
+ movdqa 0x200-0x80($Tbl),$TMP # byte swap mask
+
+ pshufd \$0x1b,$ABEF,$Wi # ABCD
+ pshufd \$0xb1,$ABEF,$ABEF # CDAB
+ pshufd \$0x1b,$CDGH,$CDGH # EFGH
+ movdqa $TMP,$BSWAP # offload
+ palignr \$8,$CDGH,$ABEF # ABEF
+ punpcklqdq $Wi,$CDGH # CDGH
+ jmp .Loop_shaext
+
+.align 16
+.Loop_shaext:
+ movdqu ($inp),@MSG[0]
+ movdqu 0x10($inp),@MSG[1]
+ movdqu 0x20($inp),@MSG[2]
+ pshufb $TMP,@MSG[0]
+ movdqu 0x30($inp),@MSG[3]
+
+ movdqa 0*32-0x80($Tbl),$Wi
+ paddd @MSG[0],$Wi
+ pshufb $TMP,@MSG[1]
+ movdqa $CDGH,$CDGH_SAVE # offload
+ sha256rnds2 $ABEF,$CDGH # 0-3
+ pshufd \$0x0e,$Wi,$Wi
+ nop
+ movdqa $ABEF,$ABEF_SAVE # offload
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 1*32-0x80($Tbl),$Wi
+ paddd @MSG[1],$Wi
+ pshufb $TMP,@MSG[2]
+ sha256rnds2 $ABEF,$CDGH # 4-7
+ pshufd \$0x0e,$Wi,$Wi
+ lea 0x40($inp),$inp
+ sha256msg1 @MSG[1],@MSG[0]
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 2*32-0x80($Tbl),$Wi
+ paddd @MSG[2],$Wi
+ pshufb $TMP,@MSG[3]
+ sha256rnds2 $ABEF,$CDGH # 8-11
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[3],$TMP
+ palignr \$4,@MSG[2],$TMP
+ nop
+ paddd $TMP,@MSG[0]
+ sha256msg1 @MSG[2],@MSG[1]
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 3*32-0x80($Tbl),$Wi
+ paddd @MSG[3],$Wi
+ sha256msg2 @MSG[3],@MSG[0]
+ sha256rnds2 $ABEF,$CDGH # 12-15
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[0],$TMP
+ palignr \$4,@MSG[3],$TMP
+ nop
+ paddd $TMP,@MSG[1]
+ sha256msg1 @MSG[3],@MSG[2]
+ sha256rnds2 $CDGH,$ABEF
+___
+for($i=4;$i<16-3;$i++) {
+$code.=<<___;
+ movdqa $i*32-0x80($Tbl),$Wi
+ paddd @MSG[0],$Wi
+ sha256msg2 @MSG[0],@MSG[1]
+ sha256rnds2 $ABEF,$CDGH # 16-19...
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[1],$TMP
+ palignr \$4,@MSG[0],$TMP
+ nop
+ paddd $TMP,@MSG[2]
+ sha256msg1 @MSG[0],@MSG[3]
+ sha256rnds2 $CDGH,$ABEF
+___
+ push(@MSG,shift(@MSG));
+}
+$code.=<<___;
+ movdqa 13*32-0x80($Tbl),$Wi
+ paddd @MSG[0],$Wi
+ sha256msg2 @MSG[0],@MSG[1]
+ sha256rnds2 $ABEF,$CDGH # 52-55
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[1],$TMP
+ palignr \$4,@MSG[0],$TMP
+ sha256rnds2 $CDGH,$ABEF
+ paddd $TMP,@MSG[2]
+
+ movdqa 14*32-0x80($Tbl),$Wi
+ paddd @MSG[1],$Wi
+ sha256rnds2 $ABEF,$CDGH # 56-59
+ pshufd \$0x0e,$Wi,$Wi
+ sha256msg2 @MSG[1],@MSG[2]
+ movdqa $BSWAP,$TMP
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 15*32-0x80($Tbl),$Wi
+ paddd @MSG[2],$Wi
+ nop
+ sha256rnds2 $ABEF,$CDGH # 60-63
+ pshufd \$0x0e,$Wi,$Wi
+ dec $num
+ nop
+ sha256rnds2 $CDGH,$ABEF
+
+ paddd $CDGH_SAVE,$CDGH
+ paddd $ABEF_SAVE,$ABEF
+ jnz .Loop_shaext
+
+ pshufd \$0xb1,$CDGH,$CDGH # DCHG
+ pshufd \$0x1b,$ABEF,$TMP # FEBA
+ pshufd \$0xb1,$ABEF,$ABEF # BAFE
+ punpckhqdq $CDGH,$ABEF # DCBA
+ palignr \$8,$TMP,$CDGH # HGFE
+
+ movdqu $ABEF,($ctx)
+ movdqu $CDGH,16($ctx)
+___
+$code.=<<___ if ($win64);
+ movaps -8-5*16(%rax),%xmm6
+ movaps -8-4*16(%rax),%xmm7
+ movaps -8-3*16(%rax),%xmm8
+ movaps -8-2*16(%rax),%xmm9
+ movaps -8-1*16(%rax),%xmm10
+ mov %rax,%rsp
+.Lepilogue_shaext:
+___
+$code.=<<___;
+ ret
+.cfi_endproc
+.size sha256_block_data_order_shaext,.-sha256_block_data_order_shaext
+___
+}}}
{{{
my $a4=$T1;
'&mov ($a,$a1)',
'&mov ($a4,$f)',
- '&xor ($a0,$e)',
'&ror ($a1,$Sigma0[2]-$Sigma0[1])',
+ '&xor ($a0,$e)',
'&xor ($a4,$g)', # f^g
'&ror ($a0,$Sigma1[1]-$Sigma1[0])',
'&add ($h,$SZ*($i&15)."(%rsp)")', # h+=X[i]+K[i]
'&mov ($a2,$a)',
- '&ror ($a1,$Sigma0[1]-$Sigma0[0])',
'&xor ($a4,$g)', # Ch(e,f,g)=((f^g)&e)^g
+ '&ror ($a1,$Sigma0[1]-$Sigma0[0])',
'&xor ($a2,$b)', # a^b, b^c in next round
- '&ror ($a0,$Sigma1[0])', # Sigma1(e)
'&add ($h,$a4)', # h+=Ch(e,f,g)
+ '&ror ($a0,$Sigma1[0])', # Sigma1(e)
'&and ($a3,$a2)', # (b^c)&(a^b)
'&xor ($a1,$a)',
'&add ($h,$a0)', # h+=Sigma1(e)
'&xor ($a3,$b)', # Maj(a,b,c)=Ch(a^b,c,b)
- '&add ($d,$h)', # d+=h
'&ror ($a1,$Sigma0[0])', # Sigma0(a)
+ '&add ($d,$h)', # d+=h
'&add ($h,$a3)', # h+=Maj(a,b,c)
'&mov ($a0,$d)',
.type ${func}_ssse3,\@function,3
.align 64
${func}_ssse3:
+.cfi_startproc
.Lssse3_shortcut:
+ mov %rsp,%rax # copy %rsp
+.cfi_def_cfa_register %rax
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
- mov %rsp,%r11 # copy %rsp
+.cfi_push %r15
shl \$4,%rdx # num*16
sub \$`$framesz+$win64*16*4`,%rsp
lea ($inp,%rdx,$SZ),%rdx # inp+num*16*$SZ
mov $ctx,$_ctx # save ctx, 1st arg
mov $inp,$_inp # save inp, 2nd arh
mov %rdx,$_end # save end pointer, "3rd" arg
- mov %r11,$_rsp # save copy of %rsp
+ mov %rax,$_rsp # save copy of %rsp
+.cfi_cfa_expression $_rsp,deref,+8
___
$code.=<<___ if ($win64);
movaps %xmm6,16*$SZ+32(%rsp)
movdqu 0x00($inp),@X[0]
movdqu 0x10($inp),@X[1]
movdqu 0x20($inp),@X[2]
- movdqu 0x30($inp),@X[3]
pshufb $t3,@X[0]
+ movdqu 0x30($inp),@X[3]
lea $TABLE(%rip),$Tbl
pshufb $t3,@X[1]
movdqa 0x00($Tbl),$t0
- pshufb $t3,@X[2]
movdqa 0x20($Tbl),$t1
+ pshufb $t3,@X[2]
paddd @X[0],$t0
movdqa 0x40($Tbl),$t2
pshufb $t3,@X[3]
.align 16
.Lssse3_00_47:
- sub \$-16*2*$SZ,$Tbl # size optimization
+ sub \$`-16*2*$SZ`,$Tbl # size optimization
___
sub Xupdate_256_SSSE3 () {
(
eval(shift(@insns));
eval(shift(@insns));
}
- } else { # squeeze extra 3% on Westmere and Atom
+ } else { # squeeze extra 4% on Westmere and 19% on Atom
eval(shift(@insns)); #@
- eval(shift(@insns));
&movdqa ($t0,@X[1]);
eval(shift(@insns));
+ eval(shift(@insns));
&movdqa ($t3,@X[3]);
+ eval(shift(@insns)); #@
+ eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns)); #@
eval(shift(@insns));
&palignr ($t0,@X[0],$SZ); # X[1..4]
- eval(shift(@insns)); #@
eval(shift(@insns));
- &palignr ($t3,@X[2],$SZ); # X[9..12]
eval(shift(@insns));
+ &palignr ($t3,@X[2],$SZ); # X[9..12]
eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns)); #@
- eval(shift(@insns));
&movdqa ($t1,$t0);
eval(shift(@insns));
+ eval(shift(@insns));
&movdqa ($t2,$t0);
eval(shift(@insns)); #@
eval(shift(@insns));
- eval(shift(@insns));
&psrld ($t0,$sigma0[2]);
eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns));
&paddd (@X[0],$t3); # X[0..3] += X[9..12]
- eval(shift(@insns));
eval(shift(@insns)); #@
eval(shift(@insns));
&psrld ($t2,$sigma0[0]);
eval(shift(@insns));
eval(shift(@insns));
- eval(shift(@insns)); #@
- eval(shift(@insns));
&pshufd ($t3,@X[3],0b11111010); # X[4..15]
eval(shift(@insns));
+ eval(shift(@insns)); #@
&pslld ($t1,8*$SZ-$sigma0[1]);
eval(shift(@insns));
+ eval(shift(@insns));
&pxor ($t0,$t2);
eval(shift(@insns)); #@
eval(shift(@insns));
- &psrld ($t2,$sigma0[1]-$sigma0[0]);
+ eval(shift(@insns));
eval(shift(@insns)); #@
+ &psrld ($t2,$sigma0[1]-$sigma0[0]);
eval(shift(@insns));
&pxor ($t0,$t1);
eval(shift(@insns));
eval(shift(@insns));
&pslld ($t1,$sigma0[1]-$sigma0[0]);
eval(shift(@insns));
+ eval(shift(@insns));
&pxor ($t0,$t2);
eval(shift(@insns));
eval(shift(@insns)); #@
- eval(shift(@insns));
&movdqa ($t2,$t3);
eval(shift(@insns));
- eval(shift(@insns)); #@
eval(shift(@insns));
&pxor ($t0,$t1); # sigma0(X[1..4])
+ eval(shift(@insns)); #@
eval(shift(@insns));
eval(shift(@insns));
&psrld ($t3,$sigma1[2]);
eval(shift(@insns));
eval(shift(@insns));
&paddd (@X[0],$t0); # X[0..3] += sigma0(X[1..4])
- eval(shift(@insns));
eval(shift(@insns)); #@
eval(shift(@insns));
- eval(shift(@insns));
&psrlq ($t2,$sigma1[0]);
eval(shift(@insns));
- eval(shift(@insns)); #@
eval(shift(@insns));
eval(shift(@insns));
&pxor ($t3,$t2);
+ eval(shift(@insns)); #@
+ eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns)); #@
&psrlq ($t2,$sigma1[1]-$sigma1[0]);
eval(shift(@insns));
- eval(shift(@insns)); #@
eval(shift(@insns));
&pxor ($t3,$t2);
- eval(shift(@insns));
+ eval(shift(@insns)); #@
eval(shift(@insns));
eval(shift(@insns));
#&pshufb ($t3,$t4); # sigma1(X[14..15])
&pshufd ($t3,$t3,0b10000000);
eval(shift(@insns));
- eval(shift(@insns)); #@
+ eval(shift(@insns));
eval(shift(@insns));
&psrldq ($t3,8);
eval(shift(@insns));
eval(shift(@insns)); #@
+ eval(shift(@insns));
+ eval(shift(@insns));
+ eval(shift(@insns)); #@
&paddd (@X[0],$t3); # X[0..1] += sigma1(X[14..15])
eval(shift(@insns));
- &pshufd ($t3,@X[0],0b01010000); # X[16..17]
eval(shift(@insns));
eval(shift(@insns));
+ &pshufd ($t3,@X[0],0b01010000); # X[16..17]
eval(shift(@insns));
+ eval(shift(@insns)); #@
eval(shift(@insns));
&movdqa ($t2,$t3);
eval(shift(@insns));
- eval(shift(@insns)); #@
eval(shift(@insns));
&psrld ($t3,$sigma1[2]);
eval(shift(@insns));
- &psrlq ($t2,$sigma1[0]);
- eval(shift(@insns));
eval(shift(@insns)); #@
+ &psrlq ($t2,$sigma1[0]);
eval(shift(@insns));
eval(shift(@insns));
&pxor ($t3,$t2);
+ eval(shift(@insns)); #@
+ eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns)); #@
eval(shift(@insns));
&psrlq ($t2,$sigma1[1]-$sigma1[0]);
- eval(shift(@insns)); #@
eval(shift(@insns));
eval(shift(@insns));
eval(shift(@insns));
&pxor ($t3,$t2);
eval(shift(@insns));
eval(shift(@insns));
- &movdqa ($t2,16*2*$j."($Tbl)");
eval(shift(@insns)); #@
- eval(shift(@insns));
#&pshufb ($t3,$t5);
&pshufd ($t3,$t3,0b00001000);
eval(shift(@insns));
+ eval(shift(@insns));
+ &movdqa ($t2,16*2*$j."($Tbl)");
eval(shift(@insns)); #@
eval(shift(@insns));
&pslldq ($t3,8);
eval(shift(@insns));
eval(shift(@insns));
- &paddd (@X[0],$t3); # X[2..3] += sigma1(X[16..17])
eval(shift(@insns));
+ &paddd (@X[0],$t3); # X[2..3] += sigma1(X[16..17])
+ eval(shift(@insns)); #@
eval(shift(@insns));
eval(shift(@insns));
}
jb .Lloop_ssse3
mov $_rsp,%rsi
+.cfi_def_cfa %rsi,8
___
$code.=<<___ if ($win64);
movaps 16*$SZ+32(%rsp),%xmm6
movaps 16*$SZ+80(%rsp),%xmm9
___
$code.=<<___;
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lepilogue_ssse3:
ret
+.cfi_endproc
.size ${func}_ssse3,.-${func}_ssse3
___
}
.type ${func}_xop,\@function,3
.align 64
${func}_xop:
+.cfi_startproc
.Lxop_shortcut:
+ mov %rsp,%rax # copy %rsp
+.cfi_def_cfa_register %rax
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
- mov %rsp,%r11 # copy %rsp
+.cfi_push %r15
shl \$4,%rdx # num*16
sub \$`$framesz+$win64*16*($SZ==4?4:6)`,%rsp
lea ($inp,%rdx,$SZ),%rdx # inp+num*16*$SZ
mov $ctx,$_ctx # save ctx, 1st arg
mov $inp,$_inp # save inp, 2nd arh
mov %rdx,$_end # save end pointer, "3rd" arg
- mov %r11,$_rsp # save copy of %rsp
+ mov %rax,$_rsp # save copy of %rsp
+.cfi_cfa_expression $_rsp,deref,+8
___
$code.=<<___ if ($win64);
movaps %xmm6,16*$SZ+32(%rsp)
$code.=<<___;
.Lprologue_xop:
- vzeroall
+ vzeroupper
mov $SZ*0($ctx),$A
mov $SZ*1($ctx),$B
mov $SZ*2($ctx),$C
.align 16
.Lxop_00_47:
- sub \$-16*2*$SZ,$Tbl # size optimization
+ sub \$`-16*2*$SZ`,$Tbl # size optimization
___
sub XOP_256_00_47 () {
my $j = shift;
.align 16
.Lxop_00_47:
- add \$16*2*$SZ,$Tbl
+ add \$`16*2*$SZ`,$Tbl
___
sub XOP_512_00_47 () {
my $j = shift;
jb .Lloop_xop
mov $_rsp,%rsi
- vzeroall
+.cfi_def_cfa %rsi,8
+ vzeroupper
___
$code.=<<___ if ($win64);
movaps 16*$SZ+32(%rsp),%xmm6
movaps 16*$SZ+112(%rsp),%xmm11
___
$code.=<<___;
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lepilogue_xop:
ret
+.cfi_endproc
.size ${func}_xop,.-${func}_xop
___
}
.type ${func}_avx,\@function,3
.align 64
${func}_avx:
+.cfi_startproc
.Lavx_shortcut:
+ mov %rsp,%rax # copy %rsp
+.cfi_def_cfa_register %rax
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
- mov %rsp,%r11 # copy %rsp
+.cfi_push %r15
shl \$4,%rdx # num*16
sub \$`$framesz+$win64*16*($SZ==4?4:6)`,%rsp
lea ($inp,%rdx,$SZ),%rdx # inp+num*16*$SZ
mov $ctx,$_ctx # save ctx, 1st arg
mov $inp,$_inp # save inp, 2nd arh
mov %rdx,$_end # save end pointer, "3rd" arg
- mov %r11,$_rsp # save copy of %rsp
+ mov %rax,$_rsp # save copy of %rsp
+.cfi_cfa_expression $_rsp,deref,+8
___
$code.=<<___ if ($win64);
movaps %xmm6,16*$SZ+32(%rsp)
$code.=<<___;
.Lprologue_avx:
- vzeroall
+ vzeroupper
mov $SZ*0($ctx),$A
mov $SZ*1($ctx),$B
mov $SZ*2($ctx),$C
.align 16
.Lavx_00_47:
- sub \$-16*2*$SZ,$Tbl # size optimization
+ sub \$`-16*2*$SZ`,$Tbl # size optimization
___
sub Xupdate_256_AVX () {
(
.align 16
.Lavx_00_47:
- add \$16*2*$SZ,$Tbl
+ add \$`16*2*$SZ`,$Tbl
___
sub Xupdate_512_AVX () {
(
jb .Lloop_avx
mov $_rsp,%rsi
- vzeroall
+.cfi_def_cfa %rsi,8
+ vzeroupper
___
$code.=<<___ if ($win64);
movaps 16*$SZ+32(%rsp),%xmm6
movaps 16*$SZ+112(%rsp),%xmm11
___
$code.=<<___;
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lepilogue_avx:
ret
+.cfi_endproc
.size ${func}_avx,.-${func}_avx
___
######################################################################
# AVX2+BMI code path
#
-my $a5=$SZ==4?"%esi":"%rsi"; # zap $inp
+my $a5=$SZ==4?"%esi":"%rsi"; # zap $inp
my $PUSH8=8*2*$SZ;
use integer;
.type ${func}_avx2,\@function,3
.align 64
${func}_avx2:
+.cfi_startproc
.Lavx2_shortcut:
+ mov %rsp,%rax # copy %rsp
+.cfi_def_cfa_register %rax
push %rbx
+.cfi_push %rbx
push %rbp
+.cfi_push %rbp
push %r12
+.cfi_push %r12
push %r13
+.cfi_push %r13
push %r14
+.cfi_push %r14
push %r15
- mov %rsp,%r11 # copy %rsp
+.cfi_push %r15
sub \$`2*$SZ*$rounds+4*8+$win64*16*($SZ==4?4:6)`,%rsp
shl \$4,%rdx # num*16
and \$-256*$SZ,%rsp # align stack frame
mov $ctx,$_ctx # save ctx, 1st arg
mov $inp,$_inp # save inp, 2nd arh
mov %rdx,$_end # save end pointer, "3rd" arg
- mov %r11,$_rsp # save copy of %rsp
+ mov %rax,$_rsp # save copy of %rsp
+.cfi_cfa_expression $_rsp,deref,+8
___
$code.=<<___ if ($win64);
movaps %xmm6,16*$SZ+32(%rsp)
$code.=<<___;
.Lprologue_avx2:
- vzeroall
+ vzeroupper
sub \$-16*$SZ,$inp # inp++, size optimization
mov $SZ*0($ctx),$A
mov $inp,%r12 # borrow $T1
vmovdqa $t0,0x00(%rsp)
xor $a1,$a1
vmovdqa $t1,0x20(%rsp)
+___
+$code.=<<___ if (!$win64);
+# temporarily use %rdi as frame pointer
+ mov $_rsp,%rdi
+.cfi_def_cfa %rdi,8
+___
+$code.=<<___;
lea -$PUSH8(%rsp),%rsp
+___
+$code.=<<___ if (!$win64);
+# the frame info is at $_rsp, but the stack is moving...
+# so a second frame pointer is saved at -8(%rsp)
+# that is in the red zone
+ mov %rdi,-8(%rsp)
+.cfi_cfa_expression %rsp-8,deref,+8
+___
+$code.=<<___;
mov $B,$a3
vmovdqa $t2,0x00(%rsp)
xor $C,$a3 # magic
my @insns = (&$body,&$body,&$body,&$body); # 96 instructions
my $base = "+2*$PUSH8(%rsp)";
- &lea ("%rsp","-$PUSH8(%rsp)") if (($j%2)==0);
+ if (($j%2)==0) {
+ &lea ("%rsp","-$PUSH8(%rsp)");
+$code.=<<___ if (!$win64);
+.cfi_cfa_expression %rsp+`$PUSH8-8`,deref,+8
+# copy secondary frame pointer to new location again at -8(%rsp)
+ pushq $PUSH8-8(%rsp)
+.cfi_cfa_expression %rsp,deref,+8
+ lea 8(%rsp),%rsp
+.cfi_cfa_expression %rsp-8,deref,+8
+___
+ }
+
foreach (Xupdate_256_AVX()) { # 29 instructions
eval;
eval(shift(@insns));
vmovdqa $t2,0x40(%rsp)
vpaddq 0x40($Tbl),@X[6],$t2
vmovdqa $t3,0x60(%rsp)
+___
+$code.=<<___ if (!$win64);
+# temporarily use %rdi as frame pointer
+ mov $_rsp,%rdi
+.cfi_def_cfa %rdi,8
+___
+$code.=<<___;
lea -$PUSH8(%rsp),%rsp
+___
+$code.=<<___ if (!$win64);
+# the frame info is at $_rsp, but the stack is moving...
+# so a second frame pointer is saved at -8(%rsp)
+# that is in the red zone
+ mov %rdi,-8(%rsp)
+.cfi_cfa_expression %rsp-8,deref,+8
+___
+$code.=<<___;
vpaddq 0x60($Tbl),@X[7],$t3
vmovdqa $t0,0x00(%rsp)
xor $a1,$a1
my @insns = (&$body,&$body); # 48 instructions
my $base = "+2*$PUSH8(%rsp)";
- &lea ("%rsp","-$PUSH8(%rsp)") if (($j%4)==0);
+ if (($j%4)==0) {
+ &lea ("%rsp","-$PUSH8(%rsp)");
+$code.=<<___ if (!$win64);
+.cfi_cfa_expression %rsp+`$PUSH8-8`,deref,+8
+# copy secondary frame pointer to new location again at -8(%rsp)
+ pushq $PUSH8-8(%rsp)
+.cfi_cfa_expression %rsp,deref,+8
+ lea 8(%rsp),%rsp
+.cfi_cfa_expression %rsp-8,deref,+8
+___
+ }
+
foreach (Xupdate_512_AVX()) { # 23 instructions
eval;
if ($_ !~ /\;$/) {
add $a1,$A
#mov `2*$SZ*$rounds+8`(%rsp),$inp # $_inp
lea `2*$SZ*($rounds-8)`(%rsp),%rsp
+# restore frame pointer to original location at $_rsp
+.cfi_cfa_expression $_rsp,deref,+8
add $SZ*0($ctx),$A
add $SZ*1($ctx),$B
jbe .Loop_avx2
lea (%rsp),$Tbl
+# temporarily use $Tbl as index to $_rsp
+# this avoids the need to save a secondary frame pointer at -8(%rsp)
+.cfi_cfa_expression $Tbl+`16*$SZ+3*8`,deref,+8
.Ldone_avx2:
- lea ($Tbl),%rsp
- mov $_rsp,%rsi
- vzeroall
+ mov `16*$SZ+3*8`($Tbl),%rsi
+.cfi_def_cfa %rsi,8
+ vzeroupper
___
$code.=<<___ if ($win64);
- movaps 16*$SZ+32(%rsp),%xmm6
- movaps 16*$SZ+48(%rsp),%xmm7
- movaps 16*$SZ+64(%rsp),%xmm8
- movaps 16*$SZ+80(%rsp),%xmm9
+ movaps 16*$SZ+32($Tbl),%xmm6
+ movaps 16*$SZ+48($Tbl),%xmm7
+ movaps 16*$SZ+64($Tbl),%xmm8
+ movaps 16*$SZ+80($Tbl),%xmm9
___
$code.=<<___ if ($win64 && $SZ>4);
- movaps 16*$SZ+96(%rsp),%xmm10
- movaps 16*$SZ+112(%rsp),%xmm11
+ movaps 16*$SZ+96($Tbl),%xmm10
+ movaps 16*$SZ+112($Tbl),%xmm11
___
$code.=<<___;
- mov (%rsi),%r15
- mov 8(%rsi),%r14
- mov 16(%rsi),%r13
- mov 24(%rsi),%r12
- mov 32(%rsi),%rbp
- mov 40(%rsi),%rbx
- lea 48(%rsi),%rsp
+ mov -48(%rsi),%r15
+.cfi_restore %r15
+ mov -40(%rsi),%r14
+.cfi_restore %r14
+ mov -32(%rsi),%r13
+.cfi_restore %r13
+ mov -24(%rsi),%r12
+.cfi_restore %r12
+ mov -16(%rsi),%rbp
+.cfi_restore %rbp
+ mov -8(%rsi),%rbx
+.cfi_restore %rbx
+ lea (%rsi),%rsp
+.cfi_def_cfa_register %rsp
.Lepilogue_avx2:
ret
+.cfi_endproc
.size ${func}_avx2,.-${func}_avx2
___
}}
$code.=<<___;
mov %rax,%rsi # put aside Rsp
mov 16*$SZ+3*8(%rax),%rax # pull $_rsp
- lea 48(%rax),%rax
mov -8(%rax),%rbx
mov -16(%rax),%rbp
pop %rsi
ret
.size se_handler,.-se_handler
+___
+
+$code.=<<___ if ($SZ==4 && $shaext);
+.type shaext_handler,\@abi-omnipotent
+.align 16
+shaext_handler:
+ push %rsi
+ push %rdi
+ push %rbx
+ push %rbp
+ push %r12
+ push %r13
+ push %r14
+ push %r15
+ pushfq
+ sub \$64,%rsp
+
+ mov 120($context),%rax # pull context->Rax
+ mov 248($context),%rbx # pull context->Rip
+
+ lea .Lprologue_shaext(%rip),%r10
+ cmp %r10,%rbx # context->Rip<.Lprologue
+ jb .Lin_prologue
+
+ lea .Lepilogue_shaext(%rip),%r10
+ cmp %r10,%rbx # context->Rip>=.Lepilogue
+ jae .Lin_prologue
+
+ lea -8-5*16(%rax),%rsi
+ lea 512($context),%rdi # &context.Xmm6
+ mov \$10,%ecx
+ .long 0xa548f3fc # cld; rep movsq
+
+ jmp .Lin_prologue
+.size shaext_handler,.-shaext_handler
+___
+$code.=<<___;
.section .pdata
.align 4
.rva .LSEH_begin_$func
.rva .LSEH_end_$func
.rva .LSEH_info_$func
___
+$code.=<<___ if ($SZ==4 && $shaext);
+ .rva .LSEH_begin_${func}_shaext
+ .rva .LSEH_end_${func}_shaext
+ .rva .LSEH_info_${func}_shaext
+___
$code.=<<___ if ($SZ==4);
.rva .LSEH_begin_${func}_ssse3
.rva .LSEH_end_${func}_ssse3
.rva se_handler
.rva .Lprologue,.Lepilogue # HandlerData[]
___
+$code.=<<___ if ($SZ==4 && $shaext);
+.LSEH_info_${func}_shaext:
+ .byte 9,0,0,0
+ .rva shaext_handler
+___
$code.=<<___ if ($SZ==4);
.LSEH_info_${func}_ssse3:
.byte 9,0,0,0
___
}
-$code =~ s/\`([^\`]*)\`/eval $1/gem;
-print $code;
+sub sha256op38 {
+ my $instr = shift;
+ my %opcodelet = (
+ "sha256rnds2" => 0xcb,
+ "sha256msg1" => 0xcc,
+ "sha256msg2" => 0xcd );
+
+ if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-7]),\s*%xmm([0-7])/) {
+ my @opcode=(0x0f,0x38);
+ push @opcode,$opcodelet{$instr};
+ push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
+ return ".byte\t".join(',',@opcode);
+ } else {
+ return $instr."\t".@_[0];
+ }
+}
+
+foreach (split("\n",$code)) {
+ s/\`([^\`]*)\`/eval $1/geo;
+
+ s/\b(sha256[^\s]*)\s+(.*)/sha256op38($1,$2)/geo;
+
+ print $_,"\n";
+}
close STDOUT;
projects / openssl.git / blobdiff