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

aesni-sha1-x86_64.pl: update from master.

Browse Source
This commit is contained in:
Andy Polyakov
2013-05-19 22:18:52 +02:00
parent aacfb2a2c6
commit 1f2d194acf
1 changed files with 92 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

151
crypto/aes/asm/aesni-sha1-x86_64.pl
View File

@@ -21,16 +21,22 @@
# subroutine: # subroutine:
# #
# AES-128-CBC +SHA1 stitch gain # AES-128-CBC +SHA1 stitch gain
# Westmere 3.77[+5.6] 9.37 6.65 +41% # Westmere 3.77[+5.5] 9.26 6.58 +41%
# Sandy Bridge 5.05[+5.2(6.3)] 10.25(11.35) 6.16(7.08) +67%(+60%) # Sandy Bridge 5.05[+5.0(6.2)] 10.06(11.21) 6.09(7.05) +65%(+59%)
# Ivy Bridge 5.05[+4.6] 9.65 5.54 +74%
# Bulldozer 5.77[+6.0] 11.72 6.37 +84%
# #
# AES-192-CBC # AES-192-CBC
# Westmere 4.51 10.11 6.97 +45% # Westmere 4.51 10.00 6.87 +46%
# Sandy Bridge 6.05 11.25(12.35) 6.34(7.27) +77%(+70%) # Sandy Bridge 6.05 11.06(12.21) 6.11(7.20) +81%(+70%)
# Ivy Bridge 6.05 10.65 6.07 +75%
# Bulldozer 6.89 12.84 6.96 +84%
# #
# AES-256-CBC # AES-256-CBC
# Westmere 5.25 10.85 7.25 +50% # Westmere 5.25 10.74 7.19 +49%
# Sandy Bridge 7.05 12.25(13.35) 7.06(7.70) +74%(+73%) # Sandy Bridge 7.05 12.06(13.21) 7.12(7.68) +69%(+72%)
# Ivy Bridge 7.05 11.65 7.12 +64%
# Bulldozer 8.00 13.95 8.25 +69%
# #
# (*) There are two code paths: SSSE3 and AVX. See sha1-568.pl for # (*) There are two code paths: SSSE3 and AVX. See sha1-568.pl for
# background information. Above numbers in parentheses are SSSE3 # background information. Above numbers in parentheses are SSSE3
@@ -47,6 +53,8 @@
# AES-128-CBC AES-192-CBC AES-256-CBC # AES-128-CBC AES-192-CBC AES-256-CBC
# Westmere 1.31 1.55 1.80 # Westmere 1.31 1.55 1.80
# Sandy Bridge 0.93 1.06 1.22 # Sandy Bridge 0.93 1.06 1.22
# Ivy Bridge 0.92 1.06 1.21
# Bulldozer 0.76 0.90 1.04
$flavour = shift; $flavour = shift;
$output = shift; $output = shift;
@@ -112,7 +120,7 @@ my @X=map("%xmm$_",(4..7,0..3));
my @Tx=map("%xmm$_",(8..10)); my @Tx=map("%xmm$_",(8..10));
my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
my @T=("%esi","%edi"); my @T=("%esi","%edi");
my $j=0; my $jj=0; my $r=0; my $sn=0; my $j=0; my $jj=0; my $r=0; my $sn=0; my $rx=0;
my $K_XX_XX="%r11"; my $K_XX_XX="%r11";
my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13)); my ($iv,$in,$rndkey0)=map("%xmm$_",(11..13));
my @rndkey=("%xmm14","%xmm15"); my @rndkey=("%xmm14","%xmm15");
@@ -180,6 +188,9 @@ $code.=<<___;
mov 12($ctx),$D mov 12($ctx),$D
mov $B,@T[0] # magic seed mov $B,@T[0] # magic seed
mov 16($ctx),$E mov 16($ctx),$E
mov $C,@T[1]
xor $D,@T[1]
and @T[1],@T[0]
movdqa 64($K_XX_XX),@X[2] # pbswap mask movdqa 64($K_XX_XX),@X[2] # pbswap mask
movdqa 0($K_XX_XX),@Tx[1] # K_00_19 movdqa 0($K_XX_XX),@Tx[1] # K_00_19
@@ -465,21 +476,27 @@ sub Xtail_ssse3()
foreach (@insns) { eval; } foreach (@insns) { eval; }
} }
sub body_00_19 () { sub body_00_19 () { # ((c^d)&b)^d
# on start @T[0]=(c^d)&b
return &body_20_39() if ($rx==19); $rx++;
use integer; use integer;
my ($k,$n); my ($k,$n);
my @r=( my @r=(
'($a,$b,$c,$d,$e)=@V;'. '($a,$b,$c,$d,$e)=@V;'.
'&add ($e,eval(4*($j&15))."(%rsp)");', # X[]+K xfer
'&xor ($c,$d);',
'&mov (@T[1],$a);', # $b in next round
'&$_rol ($a,5);',
'&and (@T[0],$c);', # ($b&($c^$d))
'&xor ($c,$d);', # restore $c
'&xor (@T[0],$d);',
'&add ($e,$a);',
'&$_ror ($b,$j?7:2);', # $b>>>2 '&$_ror ($b,$j?7:2);', # $b>>>2
'&add ($e,@T[0]);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));' '&xor (@T[0],$d);',
'&mov (@T[1],$a);', # $b for next round
'&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
'&xor ($b,$c);', # $c^$d for next round
'&$_rol ($a,5);',
'&add ($e,@T[0]);',
'&and (@T[1],$b);', # ($b&($c^$d)) for next round
'&xor ($b,$c);', # restore $b
'&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
); );
$n = scalar(@r); $n = scalar(@r);
$k = (($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds $k = (($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
@@ -488,48 +505,60 @@ sub body_00_19 () {
return @r; return @r;
} }
sub body_20_39 () { sub body_20_39 () { # b^d^c
# on entry @T[0]=b^d
return &body_40_59() if ($rx==39); $rx++;
use integer; use integer;
my ($k,$n); my ($k,$n);
my @r=( my @r=(
'($a,$b,$c,$d,$e)=@V;'. '($a,$b,$c,$d,$e)=@V;'.
'&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
'&xor (@T[0],$d);', # ($b^$d) '&xor (@T[0],$d) if($j==19);'.
'&mov (@T[1],$a);', # $b in next round '&xor (@T[0],$c) if($j> 19);', # ($b^$d^$c)
'&mov (@T[1],$a);', # $b for next round
'&$_rol ($a,5);', '&$_rol ($a,5);',
'&xor (@T[0],$c);', # ($b^$d^$c) '&add ($e,@T[0]);',
'&add ($e,$a);', '&xor (@T[1],$c) if ($j< 79);', # $b^$d for next round
'&$_ror ($b,7);', # $b>>>2 '&$_ror ($b,7);', # $b>>>2
'&add ($e,@T[0]);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));' '&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
); );
$n = scalar(@r); $n = scalar(@r);
$k = (($jj+1)*8/20)*20*$n/8; # 8 aesencs per these 20 rounds $k = (($jj+1)*8/20)*20*$n/8; # 8 aesencs per these 20 rounds
@r[$k%$n].='&$aesenc();' if ($jj==$k/$n); @r[$k%$n].='&$aesenc();' if ($jj==$k/$n && $rx!=20);
$jj++; $jj++;
return @r; return @r;
} }
sub body_40_59 () { sub body_40_59 () { # ((b^c)&(c^d))^c
# on entry @T[0]=(b^c), (c^=d)
$rx++;
use integer; use integer;
my ($k,$n); my ($k,$n);
my @r=( my @r=(
'($a,$b,$c,$d,$e)=@V;'. '($a,$b,$c,$d,$e)=@V;'.
'&mov (@T[1],$c);', '&add ($e,eval(4*($j&15))."(%rsp)");',# X[]+K xfer
'&xor ($c,$d);', '&and (@T[0],$c) if ($j>=40);', # (b^c)&(c^d)
'&add ($e,eval(4*($j++&15))."(%rsp)");', # X[]+K xfer '&xor ($c,$d) if ($j>=40);', # restore $c
'&and (@T[1],$d);',
'&and (@T[0],$c);', # ($b&($c^$d))
'&$_ror ($b,7);', # $b>>>2 '&$_ror ($b,7);', # $b>>>2
'&add ($e,@T[1]);', '&mov (@T[1],$a);', # $b for next round
'&mov (@T[1],$a);', # $b in next round '&xor (@T[0],$c);',
'&$_rol ($a,5);', '&$_rol ($a,5);',
'&add ($e,@T[0]);', '&add ($e,@T[0]);',
'&xor ($c,$d);', # restore $c '&xor (@T[1],$c) if ($j==59);'.
'&add ($e,$a);' .'unshift(@V,pop(@V)); unshift(@T,pop(@T));' '&xor (@T[1],$b) if ($j< 59);', # b^c for next round
'&xor ($b,$c) if ($j< 59);', # c^d for next round
'&add ($e,$a);' .'$j++; unshift(@V,pop(@V)); unshift(@T,pop(@T));'
); );
$n = scalar(@r); $n = scalar(@r);
$k=(($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds $k=(($jj+1)*12/20)*20*$n/12; # 12 aesencs per these 20 rounds
@r[$k%$n].='&$aesenc();' if ($jj==$k/$n); @r[$k%$n].='&$aesenc();' if ($jj==$k/$n && $rx!=40);
$jj++; $jj++;
return @r; return @r;
} }
@@ -575,8 +604,11 @@ $code.=<<___;
mov @T[0],4($ctx) mov @T[0],4($ctx)
mov @T[0],$B # magic seed mov @T[0],$B # magic seed
mov $C,8($ctx) mov $C,8($ctx)
mov $C,@T[1]
mov $D,12($ctx) mov $D,12($ctx)
xor $D,@T[1]
mov $E,16($ctx) mov $E,16($ctx)
and @T[1],@T[0]
jmp .Loop_ssse3 jmp .Loop_ssse3
.align 16 .align 16
@@ -631,7 +663,7 @@ $code.=<<___;
.size aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3 .size aesni_cbc_sha1_enc_ssse3,.-aesni_cbc_sha1_enc_ssse3
___ ___
$j=$jj=$r=$sn=0; $j=$jj=$r=$sn=$rx=0;
if ($avx) { if ($avx) {
my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10"); my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
@@ -639,6 +671,7 @@ my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9"
my $Xi=4; my $Xi=4;
my @X=map("%xmm$_",(4..7,0..3)); my @X=map("%xmm$_",(4..7,0..3));
my @Tx=map("%xmm$_",(8..10)); my @Tx=map("%xmm$_",(8..10));
my $Kx=$rndkey0;
my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization my @V=($A,$B,$C,$D,$E)=("%eax","%ebx","%ecx","%edx","%ebp"); # size optimization
my @T=("%esi","%edi"); my @T=("%esi","%edi");
@@ -700,9 +733,12 @@ $code.=<<___;
mov 12($ctx),$D mov 12($ctx),$D
mov $B,@T[0] # magic seed mov $B,@T[0] # magic seed
mov 16($ctx),$E mov 16($ctx),$E
mov $C,@T[1]
xor $D,@T[1]
and @T[1],@T[0]
vmovdqa 64($K_XX_XX),@X[2] # pbswap mask vmovdqa 64($K_XX_XX),@X[2] # pbswap mask
vmovdqa 0($K_XX_XX),@Tx[1] # K_00_19 vmovdqa 0($K_XX_XX),$Kx # K_00_19
vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3] vmovdqu 0($inp),@X[-4&7] # load input to %xmm[0-3]
vmovdqu 16($inp),@X[-3&7] vmovdqu 16($inp),@X[-3&7]
vmovdqu 32($inp),@X[-2&7] vmovdqu 32($inp),@X[-2&7]
@@ -712,13 +748,13 @@ $code.=<<___;
vpshufb @X[2],@X[-3&7],@X[-3&7] vpshufb @X[2],@X[-3&7],@X[-3&7]
vpshufb @X[2],@X[-2&7],@X[-2&7] vpshufb @X[2],@X[-2&7],@X[-2&7]
vpshufb @X[2],@X[-1&7],@X[-1&7] vpshufb @X[2],@X[-1&7],@X[-1&7]
vpaddd @Tx[1],@X[-4&7],@X[0] # add K_00_19 vpaddd $Kx,@X[-4&7],@X[0] # add K_00_19
vpaddd @Tx[1],@X[-3&7],@X[1] vpaddd $Kx,@X[-3&7],@X[1]
vpaddd @Tx[1],@X[-2&7],@X[2] vpaddd $Kx,@X[-2&7],@X[2]
vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU vmovdqa @X[0],0(%rsp) # X[]+K xfer to IALU
vmovdqa @X[1],16(%rsp) vmovdqa @X[1],16(%rsp)
vmovdqa @X[2],32(%rsp) vmovdqa @X[2],32(%rsp)
vmovups -112($key),$rndkey0 # $key[0] vmovups -112($key),$rndkey[1] # $key[0]
vmovups 16-112($key),$rndkey[0] # forward reference vmovups 16-112($key),$rndkey[0] # forward reference
jmp .Loop_avx jmp .Loop_avx
___ ___
@@ -729,7 +765,7 @@ my $aesenc=sub {
if ($k==0) { if ($k==0) {
$code.=<<___; $code.=<<___;
vmovups `16*$n`($in0),$in # load input vmovups `16*$n`($in0),$in # load input
vxorps $rndkey0,$in,$in vxorps $rndkey[1],$in,$in
___ ___
$code.=<<___ if ($n); $code.=<<___ if ($n);
vmovups $iv,`16*($n-1)`($out,$in0) # write output vmovups $iv,`16*($n-1)`($out,$in0) # write output
@@ -755,6 +791,7 @@ ___
vmovups `32+16*($k+3)-112`($key),$rndkey[0] vmovups `32+16*($k+3)-112`($key),$rndkey[0]
.Lvaesenclast$sn: .Lvaesenclast$sn:
vaesenclast $rndkey[0],$iv,$iv vaesenclast $rndkey[0],$iv,$iv
vmovups -112($key),$rndkey[0]
vmovups 16-112($key),$rndkey[1] # forward reference vmovups 16-112($key),$rndkey[1] # forward reference
___ ___
} else { } else {
@@ -778,10 +815,10 @@ sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
&vpaddd (@Tx[1],@Tx[1],@X[-1&7]); &vpaddd (@Tx[1],$Kx,@X[-1&7]);
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
&vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords &vpsrldq(@Tx[0],@X[-1&7],4); # "X[-3]", 3 dwords
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
&vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]" &vpxor (@X[0],@X[0],@X[-4&7]); # "X[0]"^="X[-16]"
@@ -831,7 +868,7 @@ sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
&vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2 &vpxor (@X[0],@X[0],@Tx[2]); # "X[0]"^=("X[0]">>96)<<<2
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
&vmovdqa (@Tx[2],eval(16*(($Xi)/5))."($K_XX_XX)"); # K_XX_XX &vmovdqa ($Kx,eval(16*(($Xi)/5))."($K_XX_XX)") if ($Xi%5==0); # K_XX_XX
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
@@ -839,7 +876,6 @@ sub Xupdate_avx_16_31() # recall that $Xi starts wtih 4
foreach (@insns) { eval; } # remaining instructions [if any] foreach (@insns) { eval; } # remaining instructions [if any]
$Xi++; push(@X,shift(@X)); # "rotate" X[] $Xi++; push(@X,shift(@X)); # "rotate" X[]
push(@Tx,shift(@Tx));
} }
sub Xupdate_avx_32_79() sub Xupdate_avx_32_79()
@@ -858,12 +894,8 @@ sub Xupdate_avx_32_79()
&vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]" &vpxor (@X[0],@X[0],@X[-7&7]); # "X[0]"^="X[-28]"
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/); eval(shift(@insns)) if (@insns[0] !~ /&ro[rl]/);
if ($Xi%5) { &vpaddd (@Tx[1],$Kx,@X[-1&7]);
&vmovdqa (@Tx[2],@Tx[1]);# "perpetuate" K_XX_XX... &vmovdqa ($Kx,eval(16*($Xi/5))."($K_XX_XX)") if ($Xi%5==0);
} else { # ... or load next one
&vmovdqa (@Tx[2],eval(16*($Xi/5))."($K_XX_XX)");
}
&vpaddd (@Tx[1],@Tx[1],@X[-1&7]);
eval(shift(@insns)); # ror eval(shift(@insns)); # ror
eval(shift(@insns)); eval(shift(@insns));
@@ -893,7 +925,6 @@ sub Xupdate_avx_32_79()
&vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2 &vpor (@X[0],@X[0],@Tx[0]); # "X[0]"<<<=2
eval(shift(@insns)); # body_20_39 eval(shift(@insns)); # body_20_39
eval(shift(@insns)); eval(shift(@insns));
&vmovdqa (@Tx[1],@X[0]) if ($Xi<19);
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); # rol eval(shift(@insns)); # rol
eval(shift(@insns)); eval(shift(@insns));
@@ -904,7 +935,6 @@ sub Xupdate_avx_32_79()
foreach (@insns) { eval; } # remaining instructions foreach (@insns) { eval; } # remaining instructions
$Xi++; push(@X,shift(@X)); # "rotate" X[] $Xi++; push(@X,shift(@X)); # "rotate" X[]
push(@Tx,shift(@Tx));
} }
sub Xuplast_avx_80() sub Xuplast_avx_80()
@@ -914,13 +944,13 @@ sub Xuplast_avx_80()
my ($a,$b,$c,$d,$e); my ($a,$b,$c,$d,$e);
eval(shift(@insns)); eval(shift(@insns));
&vpaddd (@Tx[1],@Tx[1],@X[-1&7]); &vpaddd (@Tx[1],$Kx,@X[-1&7]);
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
&movdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU &vmovdqa (eval(16*(($Xi-1)&3))."(%rsp)",@Tx[1]); # X[]+K xfer IALU
foreach (@insns) { eval; } # remaining instructions foreach (@insns) { eval; } # remaining instructions
@@ -930,7 +960,7 @@ sub Xuplast_avx_80()
unshift(@Tx,pop(@Tx)); unshift(@Tx,pop(@Tx));
&vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask &vmovdqa(@X[2],"64($K_XX_XX)"); # pbswap mask
&vmovdqa(@Tx[1],"0($K_XX_XX)"); # K_00_19 &vmovdqa($Kx,"0($K_XX_XX)"); # K_00_19
&vmovdqu(@X[-4&7],"0($inp)"); # load input &vmovdqu(@X[-4&7],"0($inp)"); # load input
&vmovdqu(@X[-3&7],"16($inp)"); &vmovdqu(@X[-3&7],"16($inp)");
&vmovdqu(@X[-2&7],"32($inp)"); &vmovdqu(@X[-2&7],"32($inp)");
@@ -952,7 +982,7 @@ sub Xloop_avx()
&vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]); &vpshufb(@X[($Xi-3)&7],@X[($Xi-3)&7],@X[2]);
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
&vpaddd (@X[$Xi&7],@X[($Xi-4)&7],@Tx[1]); &vpaddd (@X[$Xi&7],@X[($Xi-4)&7],$Kx);
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
eval(shift(@insns)); eval(shift(@insns));
@@ -1016,8 +1046,11 @@ $code.=<<___;
mov @T[0],4($ctx) mov @T[0],4($ctx)
mov @T[0],$B # magic seed mov @T[0],$B # magic seed
mov $C,8($ctx) mov $C,8($ctx)
mov $C,@T[1]
mov $D,12($ctx) mov $D,12($ctx)
xor $D,@T[1]
mov $E,16($ctx) mov $E,16($ctx)
and @T[1],@T[0]
jmp .Loop_avx jmp .Loop_avx
.align 16 .align 16