# AES-128/-192/-256+SHA256 this(**)gain
# Sandy Bridge 5.05/6.05/7.05+11.6 13.0 +28%/36%/43%
# Ivy Bridge 5.05/6.05/7.05+10.3 11.6 +32%/41%/50%
+# Haswell 4.43/5.29/6.19+7.80 8.79 +39%/49%/59%
# Bulldozer 5.77/6.89/8.00+13.7 13.7 +42%/50%/58%
#
# (*) there are XOP, AVX1 and AVX2 code pathes, meaning that
cmp \$0,`$win64?"%rcx":"%rdi"`
je .Lprobe
mov 0(%r11),%eax
- mov 4(%r11),%r10d
- mov 8(%r11),%r11d
+ mov 4(%r11),%r10
+
+ bt \$61,%r10 # check for SHA
+ jc ${func}_shaext
+
+ mov %r10,%r11
+ shr \$32,%r11
test \$`1<<11`,%r10d # check for XOP
jnz ${func}_xop
sub \$-16*$SZ,%r13 # inp++, size optimization
mov $SZ*0(%r15),$A
- xor %r12,%r12 # borrow $a0
+ lea (%rsi,%r13),%r12 # borrow $a0
mov $SZ*1(%r15),$B
cmp $len,%r13 # $_end
mov $SZ*2(%r15),$C
- sete %r12b
+ cmove %rsp,%r12 # next block or random data
mov $SZ*3(%r15),$D
mov $SZ*4(%r15),$E
mov $SZ*5(%r15),$F
jmp .Loop_avx2
.align 16
.Loop_avx2:
- shl \$`log(16*$SZ)/log(2)`,%r12
vmovdqa $TABLE+`$SZ*2*$rounds`(%rip),$t3
- neg %r12
vmovdqu -16*$SZ+0(%rsi,%r13),%xmm0
- add %rsi,%r12 # next or same input block
vmovdqu -16*$SZ+16(%rsi,%r13),%xmm1
vmovdqu -16*$SZ+32(%rsi,%r13),%xmm2
vmovdqu -16*$SZ+48(%rsi,%r13),%xmm3
- vinserti128 \$1,(%r12,%r13),@X[0],@X[0]
- vinserti128 \$1,16(%r12,%r13),@X[1],@X[1]
+ vinserti128 \$1,(%r12),@X[0],@X[0]
+ vinserti128 \$1,16(%r12),@X[1],@X[1]
vpshufb $t3,@X[0],@X[0]
- vinserti128 \$1,32(%r12,%r13),@X[2],@X[2]
+ vinserti128 \$1,32(%r12),@X[2],@X[2]
vpshufb $t3,@X[1],@X[1]
- vinserti128 \$1,48(%r12,%r13),@X[3],@X[3]
+ vinserti128 \$1,48(%r12),@X[3],@X[3]
lea $TABLE(%rip),$Tbl
vpshufb $t3,@X[2],@X[2]
add $SZ*4(%r15),$E
add $SZ*5(%r15),$F
add $SZ*6(%r15),$G
- xor %r12,%r12
+ lea (%rsi,%r13),%r12
add $SZ*7(%r15),$H
cmp $_end,%r13
mov $A,$SZ*0(%r15)
+ cmove %rsp,%r12 # next block or stale data
mov $B,$SZ*1(%r15)
mov $C,$SZ*2(%r15)
mov $D,$SZ*3(%r15)
mov $G,$SZ*6(%r15)
mov $H,$SZ*7(%r15)
- sete %r12b
jbe .Loop_avx2
lea (%rsp),$Tbl
.size ${func}_avx2,.-${func}_avx2
___
}}
+}}
+{{
+my ($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
+
+my ($rounds,$Tbl)=("%r11d","%rbx");
+
+my ($iv,$in,$rndkey0)=map("%xmm$_",(6,14,15));
+my @rndkey=("%xmm4","%xmm5");
+my $r=0;
+my $sn=0;
+
+my ($Wi,$ABEF,$CDGH,$TMP,$BSWAP,$ABEF_SAVE,$CDGH_SAVE)=map("%xmm$_",(0..3,7..9));
+my @MSG=map("%xmm$_",(10..13));
+
+my $aesenc=sub {
+ use integer;
+ my ($n,$k)=($r/10,$r%10);
+ if ($k==0) {
+ $code.=<<___;
+ movups `16*$n`($in0),$in # load input
+ xorps $rndkey0,$in
+___
+ $code.=<<___ if ($n);
+ movups $iv,`16*($n-1)`($out,$in0) # write output
+___
+ $code.=<<___;
+ xorps $in,$iv
+ movups `32+16*$k-112`($key),$rndkey[1]
+ aesenc $rndkey[0],$iv
+___
+ } elsif ($k==9) {
+ $sn++;
+ $code.=<<___;
+ cmp \$11,$rounds
+ jb .Laesenclast$sn
+ movups `32+16*($k+0)-112`($key),$rndkey[1]
+ aesenc $rndkey[0],$iv
+ movups `32+16*($k+1)-112`($key),$rndkey[0]
+ aesenc $rndkey[1],$iv
+ je .Laesenclast$sn
+ movups `32+16*($k+2)-112`($key),$rndkey[1]
+ aesenc $rndkey[0],$iv
+ movups `32+16*($k+3)-112`($key),$rndkey[0]
+ aesenc $rndkey[1],$iv
+.Laesenclast$sn:
+ aesenclast $rndkey[0],$iv
+ movups 16-112($key),$rndkey[1] # forward reference
+ nop
+___
+ } else {
+ $code.=<<___;
+ movups `32+16*$k-112`($key),$rndkey[1]
+ aesenc $rndkey[0],$iv
+___
+ }
+ $r++; unshift(@rndkey,pop(@rndkey));
+};
+
+$code.=<<___;
+.type ${func}_shaext,\@function,6
+.align 32
+${func}_shaext:
+ mov %rsp,%rax
+ mov `($win64?56:8)`(%rsp),$inp # load 7th argument
+ push %rbx
+___
+$code.=<<___ if ($win64);
+ lea `-4*16`(%rsp),%rsp
+ movaps %xmm6,-8-10*16(%rax)
+ movaps %xmm7,-8-9*16(%rax)
+ movaps %xmm8,-8-8*16(%rax)
+ movaps %xmm9,-8-7*16(%rax)
+ movaps %xmm10,-8-6*16(%rax)
+ movaps %xmm11,-8-5*16(%rax)
+ movaps %xmm12,-8-4*16(%rax)
+ movaps %xmm13,-8-3*16(%rax)
+ movaps %xmm14,-8-2*16(%rax)
+ movaps %xmm15,-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
+
+ mov 240($key),$rounds
+ sub $in0,$out
+ movups ($key),$rndkey0 # $key[0]
+ movups 16($key),$rndkey[0] # forward reference
+ lea 112($key),$key # size optimization
+
+ pshufd \$0x1b,$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
+ movdqa $ABEF,$ABEF_SAVE # offload
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 0-3
+ pshufd \$0x0e,$Wi,$Wi
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 1*32-0x80($Tbl),$Wi
+ paddd @MSG[1],$Wi
+ pshufb $TMP,@MSG[2]
+ lea 0x40($inp),$inp
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 4-7
+ pshufd \$0x0e,$Wi,$Wi
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 2*32-0x80($Tbl),$Wi
+ paddd @MSG[2],$Wi
+ pshufb $TMP,@MSG[3]
+ sha256msg1 @MSG[1],@MSG[0]
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 8-11
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[3],$TMP
+ palignr \$4,@MSG[2],$TMP
+ paddd $TMP,@MSG[0]
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 3*32-0x80($Tbl),$Wi
+ paddd @MSG[3],$Wi
+ sha256msg2 @MSG[3],@MSG[0]
+ sha256msg1 @MSG[2],@MSG[1]
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 12-15
+ pshufd \$0x0e,$Wi,$Wi
+___
+ &$aesenc();
+$code.=<<___;
+ movdqa @MSG[0],$TMP
+ palignr \$4,@MSG[3],$TMP
+ paddd $TMP,@MSG[1]
+ sha256rnds2 $CDGH,$ABEF
+___
+for($i=4;$i<16-3;$i++) {
+ &$aesenc() if (($r%10)==0);
+$code.=<<___;
+ movdqa $i*32-0x80($Tbl),$Wi
+ paddd @MSG[0],$Wi
+ sha256msg2 @MSG[0],@MSG[1]
+ sha256msg1 @MSG[3],@MSG[2]
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 16-19...
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[1],$TMP
+ palignr \$4,@MSG[0],$TMP
+ paddd $TMP,@MSG[2]
+___
+ &$aesenc();
+ &$aesenc() if ($r==19);
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+___
+ push(@MSG,shift(@MSG));
+}
+$code.=<<___;
+ movdqa 13*32-0x80($Tbl),$Wi
+ paddd @MSG[0],$Wi
+ sha256msg2 @MSG[0],@MSG[1]
+ sha256msg1 @MSG[3],@MSG[2]
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 52-55
+ pshufd \$0x0e,$Wi,$Wi
+ movdqa @MSG[1],$TMP
+ palignr \$4,@MSG[0],$TMP
+ paddd $TMP,@MSG[2]
+___
+ &$aesenc();
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 14*32-0x80($Tbl),$Wi
+ paddd @MSG[1],$Wi
+ sha256msg2 @MSG[1],@MSG[2]
+ movdqa $BSWAP,$TMP
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 56-59
+ pshufd \$0x0e,$Wi,$Wi
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+
+ movdqa 15*32-0x80($Tbl),$Wi
+ paddd @MSG[2],$Wi
+___
+ &$aesenc();
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $ABEF,$CDGH # 60-63
+ pshufd \$0x0e,$Wi,$Wi
+___
+ &$aesenc();
+$code.=<<___;
+ sha256rnds2 $CDGH,$ABEF
+ #pxor $CDGH,$rndkey0 # black magic
+___
+ while ($r<40) { &$aesenc(); } # remaining aesenc's
+$code.=<<___;
+ #xorps $CDGH,$rndkey0 # black magic
+ paddd $CDGH_SAVE,$CDGH
+ paddd $ABEF_SAVE,$ABEF
+
+ dec $len
+ movups $iv,48($out,$in0) # write output
+ lea 64($in0),$in0
+ jnz .Loop_shaext
+
+ pshufd \$0xb1,$CDGH,$CDGH # DCHG
+ pshufd \$0x1b,$ABEF,$TMP # FEBA
+ pshufd \$0xb1,$ABEF,$ABEF # BAFE
+ punpckhqdq $CDGH,$ABEF # DCBA
+ palignr \$8,$TMP,$CDGH # HGFE
+
+ movups $iv,($ivp) # write IV
+ movdqu $ABEF,($ctx)
+ movdqu $CDGH,16($ctx)
+___
+$code.=<<___ if ($win64);
+ movaps -8-10*16(%rax),%xmm6
+ movaps -8-9*16(%rax),%xmm7
+ movaps -8-8*16(%rax),%xmm8
+ movaps -8-7*16(%rax),%xmm9
+ movaps -8-6*16(%rax),%xmm10
+ movaps -8-5*16(%rax),%xmm11
+ movaps -8-4*16(%rax),%xmm12
+ movaps -8-3*16(%rax),%xmm13
+ movaps -8-2*16(%rax),%xmm14
+ movaps -8-1*16(%rax),%xmm15
+.Lepilogue_shaext:
+___
+$code.=<<___;
+ mov -8(%rax),%rbx
+ mov %rax,%rsp
+ ret
+.size ${func}_shaext,.-${func}_shaext
+___
}}}}}
# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
mov %r14,232($context) # restore context->R14
mov %r15,240($context) # restore context->R15
- lea .Lepilogue(%rip),%r10
- cmp %r10,%rbx
- jb .Lin_prologue # non-AVX code
-
lea 16*$SZ+8*8(%rsi),%rsi # Xmm6- save area
lea 512($context),%rdi # &context.Xmm6
mov \$20,%ecx
___
}
+####################################################################
+sub rex {
+ local *opcode=shift;
+ my ($dst,$src)=@_;
+ my $rex=0;
+
+ $rex|=0x04 if($dst>=8);
+ $rex|=0x01 if($src>=8);
+ unshift @opcode,$rex|0x40 if($rex);
+}
+
+{
+ my %opcodelet = (
+ "sha256rnds2" => 0xcb,
+ "sha256msg1" => 0xcc,
+ "sha256msg2" => 0xcd );
+
+ sub sha256op38 {
+ my $instr = shift;
+
+ if (defined($opcodelet{$instr}) && @_[0] =~ /%xmm([0-7]),\s*%xmm([0-7])/) {
+ my @opcode=(0x0f,0x38);
+ rex(\@opcode,$2,$1);
+ push @opcode,$opcodelet{$instr};
+ push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
+ return ".byte\t".join(',',@opcode);
+ } else {
+ return $instr."\t".@_[0];
+ }
+ }
+}
+
$code =~ s/\`([^\`]*)\`/eval $1/gem;
+$code =~ s/\b(sha256[^\s]*)\s+(.*)/sha256op38($1,$2)/gem;
print $code;
close STDOUT;
projects / openssl.git / blobdiff