From: Andy Polyakov <appro@openssl.org>
Date: Sun, 24 Jun 2012 19:22:06 +0000 (+0000)
Subject: sha512-x86_64.pl: add SIMD code paths.
X-Git-Tag: master-post-reformat~1793
X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=commitdiff_plain;h=a8f3b8b519ddf028f7b1f915813b4ba2ceb858aa;hp=0f229cce65c1e7a04ed114c04327d75169b6dac3

sha512-x86_64.pl: add SIMD code paths.
---

diff --git a/crypto/sha/asm/sha512-x86_64.pl b/crypto/sha/asm/sha512-x86_64.pl
index 6dcdbeafba..0d445def5e 100755
--- a/crypto/sha/asm/sha512-x86_64.pl
+++ b/crypto/sha/asm/sha512-x86_64.pl
@@ -43,7 +43,44 @@
 # May 2012.
 #
 # Optimization including one of Pavel Semjanov's ideas, alternative
-# Maj, resulted in >=5% improvement on most CPUs, 20% on P4.
+# Maj, resulted in >=5% improvement on most CPUs, +20% SHA256 and
+# unfortunately -10% SHA512 on P4 [which nobody should care about
+# that much].
+#
+# June 2012.
+#
+# Add SIMD code paths, see below for improvement coefficients. SSSE3
+# code path was not attempted for SHA512, because improvement is not
+# estimated to be high enough, noticeably less than 9%, to justify
+# the effort, not on pre-AVX processors. [Obviously with exclusion
+# for VIA Nano, but it has SHA512 instruction that is faster and
+# should be used instead.] For reference, corresponding estimated
+# upper limit for improvement for SSSE3 SHA256 is 28%. The fact that
+# higher coefficients are observed on VIA Nano and Bulldozer has more
+# to do with specifics of their architecture [which is topic for
+# separate discussion].
+
+######################################################################
+# 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	13.9(+11%)  -		    10.3    -
+# Westmere	15.1	12.5(+21%)  -		    9.72    -
+# Atom		23.0	21.6(+6%)   -		    14.7    -
+# VIA Nano	23.0	16.3(+41%)  -		    14.7    -
+# Sandy Bridge	17.4	14.0(+24%)  11.6(+50%(**))  11.2    8.10(+38%(**))
+# Ivy Bridge	12.6	10.3(+22%)  10.3(+22%)	    8.17    7.22(+13%)
+# Bulldozer	21.5	13.7(+57%)  13.7(+57%(***)) 13.5    8.58(+57%)
+#
+# (*)	whichever applicable;
+# (**)	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;
@@ -56,6 +93,16 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
 die "can't locate x86_64-xlate.pl";
 
+$avx=1 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
+		=~ /GNU assembler version ([2-9]\.[0-9]+)/ &&
+	   $1>=2.19);
+$avx=1 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
+	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/ &&
+	   $1>=2.09);
+$avx=1 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
+	   `ml64 2>&1` =~ /Version ([0-9]+)\./ &&
+	   $1>=10);
+
 open STDOUT,"| $^X $xlate $flavour $output";
 
 if ($output =~ /512/) {
@@ -179,10 +226,33 @@ ___
 $code=<<___;
 .text
 
+.extern	OPENSSL_ia32cap_P
 .globl	$func
 .type	$func,\@function,4
 .align	16
 $func:
+___
+$code.=<<___ if ($SZ==4 || $avx);
+	lea	OPENSSL_ia32cap_P(%rip),%r11
+	mov	0(%r11),%r10d
+	mov	4(%r11),%r11d
+___
+$code.=<<___ if ($avx && $SZ==8);
+	test	\$`1<<11`,%r11d		# check for XOP
+	jnz	.Lxop_shortcut
+___
+$code.=<<___ if ($avx);
+	and	\$`1<<30`,%r10d		# mask "Intel CPU" bit
+	and	\$`1<<28|1<<9`,%r11d	# mask AVX and SSSE3 bits
+	or	%r10d,%r11d
+	cmp	\$`1<<28|1<<9|1<<30`,%r11d
+	je	.Lavx_shortcut
+___
+$code.=<<___ if ($SZ==4);
+	test	\$`1<<9`,%r11d
+	jnz	.Lssse3_shortcut
+___
+$code.=<<___;
 	push	%rbx
 	push	%rbp
 	push	%r12
@@ -235,7 +305,7 @@ ___
 	}
 
 $code.=<<___;
-	testl	\$-1,($Tbl)
+	cmpb	\$0,`$SZ-1`($Tbl)
 	jnz	.Lrounds_16_xx
 
 	mov	$_ctx,$ctx
@@ -296,7 +366,10 @@ $TABLE:
 	.long	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
 	.long	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
 	.long	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
-	.long	0
+
+	.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f
+	.long	0x03020100,0x0b0a0908,0xffffffff,0xffffffff
+	.long	0xffffffff,0xffffffff,0x03020100,0x0b0a0908
 	.asciz	"SHA256 block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 ___
 } else {
@@ -344,11 +417,1094 @@ $TABLE:
 	.quad	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
 	.quad	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
 	.quad	0x5fcb6fab3ad6faec,0x6c44198c4a475817
-	.long	0
+
+	.quad	0x0001020304050607,0x08090a0b0c0d0e0f
 	.asciz	"SHA512 block transfort for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 ___
 }
 
+######################################################################
+# SIMD code paths
+#
+{{{
+
+my $a4=$T1;
+my ($a,$b,$c,$d,$e,$f,$g,$h);
+
+sub AUTOLOAD()		# thunk [simplified] 32-bit style perlasm
+{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://;
+  my $arg = pop;
+    $arg = "\$$arg" if ($arg*1 eq $arg);
+    $code .= "\t$opcode\t".join(',',$arg,reverse @_)."\n";
+}
+
+sub body_00_15 () {
+	(
+	'($a,$b,$c,$d,$e,$f,$g,$h)=@ROT;'.
+
+	'&ror	($a0,$Sigma1[2]-$Sigma1[1])',
+	'&mov	($a,$a1)',
+	'&mov	($a4,$f)',
+
+	'&xor	($a0,$e)',
+	'&ror	($a1,$Sigma0[2]-$Sigma0[1])',
+	'&xor	($a4,$g)',			# f^g
+
+	'&ror	($a0,$Sigma1[1]-$Sigma1[0])',
+	'&xor	($a1,$a)',
+	'&and	($a4,$e)',			# (f^g)&e
+
+	'&xor	($a0,$e)',
+	'&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
+	'&xor	($a2,$b)',			# a^b, b^c in next round
+
+	'&ror	($a0,$Sigma1[0])',		# Sigma1(e)
+	'&add	($h,$a4)',			# h+=Ch(e,f,g)
+	'&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	($h,$a3)',			# h+=Maj(a,b,c)
+
+	'&mov	($a0,$d)',
+	'&add	($a1,$h);'.			# h+=Sigma0(a)
+	'($a2,$a3) = ($a3,$a2); unshift(@ROT,pop(@ROT)); $i++;'
+	);
+}
+
+######################################################################
+# SSSE3 code path
+#
+if ($SZ==4) {	# SHA256 only
+my @X = map("%xmm$_",(0..3));
+my ($t0,$t1,$t2,$t3, $t4,$t5) = map("%xmm$_",(4..9));
+
+$code.=<<___;
+.type	${func}_ssse3,\@function,4
+.align	64
+${func}_ssse3:
+.Lssse3_shortcut:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	mov	%rsp,%r11		# copy %rsp
+	shl	\$4,%rdx		# num*16
+	sub	\$`$framesz+$win64*16*4`,%rsp
+	lea	($inp,%rdx,$SZ),%rdx	# inp+num*16*$SZ
+	and	\$-64,%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
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,16*$SZ+32(%rsp)
+	movaps	%xmm7,16*$SZ+48(%rsp)
+	movaps	%xmm8,16*$SZ+64(%rsp)
+	movaps	%xmm9,16*$SZ+80(%rsp)
+___
+$code.=<<___;
+.Lprologue_ssse3:
+
+	mov	$SZ*0($ctx),$A
+	mov	$SZ*1($ctx),$B
+	mov	$SZ*2($ctx),$C
+	mov	$SZ*3($ctx),$D
+	mov	$SZ*4($ctx),$E
+	mov	$SZ*5($ctx),$F
+	mov	$SZ*6($ctx),$G
+	mov	$SZ*7($ctx),$H
+___
+
+$code.=<<___;
+	movdqa	$TABLE+`$SZ*$rounds`+16(%rip),$t4
+	movdqa	$TABLE+`$SZ*$rounds`+32(%rip),$t5
+	jmp	.Lloop_ssse3
+.align	16
+.Lloop_ssse3:
+	movdqa	$TABLE+`$SZ*$rounds`(%rip),$t3
+	movdqu	0x00($inp),@X[0]
+	movdqu	0x10($inp),@X[1]
+	movdqu	0x20($inp),@X[2]
+	movdqu	0x30($inp),@X[3]
+	pshufb	$t3,@X[0]
+	lea	$TABLE(%rip),$Tbl
+	pshufb	$t3,@X[1]
+	movdqa	0x00($Tbl),$t0
+	pshufb	$t3,@X[2]
+	movdqa	0x10($Tbl),$t1
+	paddd	@X[0],$t0
+	movdqa	0x20($Tbl),$t2
+	pshufb	$t3,@X[3]
+	movdqa	0x30($Tbl),$t3
+	paddd	@X[1],$t1
+	paddd	@X[2],$t2
+	paddd	@X[3],$t3
+	movdqa	$t0,0x00(%rsp)
+	mov	$A,$a1
+	movdqa	$t1,0x10(%rsp)
+	mov	$B,$a3
+	movdqa	$t2,0x20(%rsp)
+	xor	$C,$a3			# magic
+	movdqa	$t3,0x30(%rsp)
+	mov	$E,$a0
+	jmp	.Lssse3_00_47
+
+.align	16
+.Lssse3_00_47:
+	add	\$16*$SZ,$Tbl
+___
+sub Xupdate_256_SSSE3 () {
+	(
+	'&movdqa	($t0,@X[1]);',
+	'&movdqa	($t3,@X[3])',
+	'&palignr	($t0,@X[0],$SZ)',	# X[1..4]
+	 '&palignr	($t3,@X[2],$SZ);',	# X[9..12]
+	'&movdqa	($t1,$t0)',
+	'&movdqa	($t2,$t0);',
+	'&psrld		($t0,$sigma0[2])',
+	 '&paddd	(@X[0],$t3);',		# X[0..3] += X[9..12]
+	'&psrld		($t2,$sigma0[0])',
+	 '&pshufd	($t3,@X[3],0b11111010)',# X[14..15]
+	'&pslld		($t1,8*$SZ-$sigma0[1]);'.
+	'&pxor		($t0,$t2)',
+	'&psrld		($t2,$sigma0[1]-$sigma0[0]);'.
+	'&pxor		($t0,$t1)',
+	'&pslld		($t1,$sigma0[1]-$sigma0[0]);'.
+	'&pxor		($t0,$t2);',
+	 '&movdqa	($t2,$t3)',
+	'&pxor		($t0,$t1);',		# sigma0(X[1..4])
+	 '&psrld	($t3,$sigma1[2])',
+	'&paddd		(@X[0],$t0);',		# X[0..3] += sigma0(X[1..4])
+	 '&psrlq	($t2,$sigma1[0])',
+	 '&pxor		($t3,$t2);',
+	 '&psrlq	($t2,$sigma1[1]-$sigma1[0])',
+	 '&pxor		($t3,$t2)',
+	 '&pshufb	($t3,$t4)',		# sigma1(X[14..15])
+	'&paddd		(@X[0],$t3)',		# X[0..1] += sigma1(X[14..15])
+	 '&pshufd	($t3,@X[0],0b01010000)',# X[16..17]
+	 '&movdqa	($t2,$t3);',
+	 '&psrld	($t3,$sigma1[2])',
+	 '&psrlq	($t2,$sigma1[0])',
+	 '&pxor		($t3,$t2);',
+	 '&psrlq	($t2,$sigma1[1]-$sigma1[0])',
+	 '&pxor		($t3,$t2);',
+	'&movdqa	($t2,16*$j."($Tbl)")',
+	 '&pshufb	($t3,$t5)',
+	'&paddd		(@X[0],$t3)'		# X[2..3] += sigma1(X[16..17])
+	);
+}
+
+sub SSSE3_256_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body);	# 104 instructions
+
+    if (0) {
+	foreach (Xupdate_256_SSSE3()) {		# 36 instructions
+	    eval;
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	}
+    } else {			# squeeze extra 3% on Westmere and Atom
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	&movdqa		($t0,@X[1]);
+	  eval(shift(@insns));
+	&movdqa		($t3,@X[3]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	&palignr	($t0,@X[0],$SZ);	# X[1..4]
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	 &palignr	($t3,@X[2],$SZ);	# X[9..12]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	&movdqa		($t1,$t0);
+	  eval(shift(@insns));
+	&movdqa		($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	($t3,@X[3],0b11111010);	# X[4..15]
+	  eval(shift(@insns));
+	&pslld		($t1,8*$SZ-$sigma0[1]);
+	  eval(shift(@insns));
+	&pxor		($t0,$t2);
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	&psrld		($t2,$sigma0[1]-$sigma0[0]);
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	&pxor		($t0,$t1);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&pslld		($t1,$sigma0[1]-$sigma0[0]);
+	  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));
+	 &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,$t2);
+	  eval(shift(@insns));
+	  eval(shift(@insns));	#@
+	 &psrlq		($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));
+	 &pshufb	($t3,$t4);		# sigma1(X[14..15])
+	  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));
+	  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));	#@
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &pxor		($t3,$t2);
+	  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*$j."($Tbl)");
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	 &pshufb	($t3,$t5);
+	  eval(shift(@insns));
+	  eval(shift(@insns));	#@
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&paddd		(@X[0],$t3);		# X[2..3] += sigma1(X[16..17])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+    }
+	&paddd		($t2,@X[0]);
+	  foreach (@insns) { eval; }		# remaining instructions
+	&movdqa		(16*$j."(%rsp)",$t2);
+}
+
+    for ($i=0,$j=0; $j<4; $j++) {
+	&SSSE3_256_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+	&cmpb	($SZ-1+16*$SZ."($Tbl)",0);
+	&jne	(".Lssse3_00_47");
+
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+$code.=<<___;
+	mov	$_ctx,$ctx
+	mov	$a1,$A
+
+	add	$SZ*0($ctx),$A
+	lea	16*$SZ($inp),$inp
+	add	$SZ*1($ctx),$B
+	add	$SZ*2($ctx),$C
+	add	$SZ*3($ctx),$D
+	add	$SZ*4($ctx),$E
+	add	$SZ*5($ctx),$F
+	add	$SZ*6($ctx),$G
+	add	$SZ*7($ctx),$H
+
+	cmp	$_end,$inp
+
+	mov	$A,$SZ*0($ctx)
+	mov	$B,$SZ*1($ctx)
+	mov	$C,$SZ*2($ctx)
+	mov	$D,$SZ*3($ctx)
+	mov	$E,$SZ*4($ctx)
+	mov	$F,$SZ*5($ctx)
+	mov	$G,$SZ*6($ctx)
+	mov	$H,$SZ*7($ctx)
+	jb	.Lloop_ssse3
+
+	mov	$_rsp,%rsi
+___
+$code.=<<___ if ($win64);
+	movaps	16*$SZ+32(%rsp),%xmm6
+	movaps	16*$SZ+48(%rsp),%xmm7
+	movaps	16*$SZ+64(%rsp),%xmm8
+	movaps	16*$SZ+80(%rsp),%xmm9
+___
+$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
+.Lepilogue_ssse3:
+	ret
+.size	${func}_ssse3,.-${func}_ssse3
+___
+}
+
+if ($avx) {{
+######################################################################
+# XOP code path
+#
+if ($SZ==8) {	# SHA512 only
+$code.=<<___;
+.type	${func}_xop,\@function,4
+.align	64
+${func}_xop:
+.Lxop_shortcut:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	mov	%rsp,%r11		# copy %rsp
+	shl	\$4,%rdx		# num*16
+	sub	\$`$framesz+$win64*16*($SZ==4?4:6)`,%rsp
+	lea	($inp,%rdx,$SZ),%rdx	# inp+num*16*$SZ
+	and	\$-64,%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
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,16*$SZ+32(%rsp)
+	movaps	%xmm7,16*$SZ+48(%rsp)
+	movaps	%xmm8,16*$SZ+64(%rsp)
+	movaps	%xmm9,16*$SZ+80(%rsp)
+___
+$code.=<<___ if ($win64 && $SZ>4);
+	movaps	%xmm10,16*$SZ+96(%rsp)
+	movaps	%xmm11,16*$SZ+112(%rsp)
+___
+$code.=<<___;
+.Lprologue_xop:
+
+	vzeroall
+	mov	$SZ*0($ctx),$A
+	mov	$SZ*1($ctx),$B
+	mov	$SZ*2($ctx),$C
+	mov	$SZ*3($ctx),$D
+	mov	$SZ*4($ctx),$E
+	mov	$SZ*5($ctx),$F
+	mov	$SZ*6($ctx),$G
+	mov	$SZ*7($ctx),$H
+	jmp	.Lloop_xop
+___
+					if ($SZ==4) {	# SHA256
+    my @X = map("%xmm$_",(0..3));
+    my ($t0,$t1,$t2,$t3) = map("%xmm$_",(4..7));
+
+$code.=<<___;
+.align	16
+.Lloop_xop:
+	vmovdqa	$TABLE+`$SZ*$rounds`(%rip),$t3
+	vmovdqu	0x00($inp),@X[0]
+	vmovdqu	0x10($inp),@X[1]
+	vmovdqu	0x20($inp),@X[2]
+	vmovdqu	0x30($inp),@X[3]
+	vpshufb	$t3,@X[0],@X[0]
+	lea	$TABLE(%rip),$Tbl
+	vpshufb	$t3,@X[1],@X[1]
+	vpshufb	$t3,@X[2],@X[2]
+	vpaddd	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[3],@X[3]
+	vpaddd	0x10($Tbl),@X[1],$t1
+	vpaddd	0x20($Tbl),@X[2],$t2
+	vpaddd	0x30($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	mov	$A,$a1
+	vmovdqa	$t1,0x10(%rsp)
+	mov	$B,$a3
+	vmovdqa	$t2,0x20(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x30(%rsp)
+	mov	$E,$a0
+	jmp	.Lxop_00_47
+
+.align	16
+.Lxop_00_47:
+	add	\$16*$SZ,$Tbl
+___
+sub XOP_256_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body);	# 104 instructions
+
+	&vpalignr	($t0,@X[1],@X[0],$SZ);	# X[1..4]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpalignr	($t3,@X[3],@X[2],$SZ);	# X[9..12]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vprotd		($t1,$t0,8*$SZ-$sigma0[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpsrld		($t0,$t0,$sigma0[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpaddd	(@X[0],@X[0],$t3);	# X[0..3] += X[9..12]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vprotd		($t2,$t1,$sigma0[1]-$sigma0[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpxor		($t0,$t0,$t1);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t3,@X[3],8*$SZ-$sigma1[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpxor		($t0,$t0,$t2);		# sigma0(X[1..4])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpsrld	($t2,@X[3],$sigma1[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		(@X[0],@X[0],$t0);	# X[0..3] += sigma0(X[1..4])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t1,$t3,$sigma1[1]-$sigma1[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t2);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t1);		# sigma1(X[14..15])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpsrldq	($t3,$t3,8);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		(@X[0],@X[0],$t3);	# X[0..1] += sigma1(X[14..15])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t3,@X[0],8*$SZ-$sigma1[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpsrld	($t2,@X[0],$sigma1[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotd	($t1,$t3,$sigma1[1]-$sigma1[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t2);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t1);		# sigma1(X[16..17])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpslldq	($t3,$t3,8);		# 22 instructions
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		(@X[0],@X[0],$t3);	# X[2..3] += sigma1(X[16..17])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddd		($t2,@X[0],16*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	(16*$j."(%rsp)",$t2);
+}
+
+    for ($i=0,$j=0; $j<4; $j++) {
+	&XOP_256_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+	&cmpb	($SZ-1+16*$SZ."($Tbl)",0);
+	&jne	(".Lxop_00_47");
+
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+
+					} else {	# SHA512
+    my @X = map("%xmm$_",(0..7));
+    my ($t0,$t1,$t2,$t3) = map("%xmm$_",(8..11));
+
+$code.=<<___;
+.align	16
+.Lloop_xop:
+	vmovdqa	$TABLE+`$SZ*$rounds`(%rip),$t3
+	vmovdqu	0x00($inp),@X[0]
+	lea	$TABLE(%rip),$Tbl
+	vmovdqu	0x10($inp),@X[1]
+	vmovdqu	0x20($inp),@X[2]
+	vpshufb	$t3,@X[0],@X[0]
+	vmovdqu	0x30($inp),@X[3]
+	vpshufb	$t3,@X[1],@X[1]
+	vmovdqu	0x40($inp),@X[4]
+	vpshufb	$t3,@X[2],@X[2]
+	vmovdqu	0x50($inp),@X[5]
+	vpshufb	$t3,@X[3],@X[3]
+	vmovdqu	0x60($inp),@X[6]
+	vpshufb	$t3,@X[4],@X[4]
+	vmovdqu	0x70($inp),@X[7]
+	vpshufb	$t3,@X[5],@X[5]
+	vpaddq	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[6],@X[6]
+	vpaddq	0x10($Tbl),@X[1],$t1
+	vpshufb	$t3,@X[7],@X[7]
+	vpaddq	0x20($Tbl),@X[2],$t2
+	vpaddq	0x30($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	vpaddq	0x40($Tbl),@X[4],$t0
+	vmovdqa	$t1,0x10(%rsp)
+	vpaddq	0x50($Tbl),@X[5],$t1
+	vmovdqa	$t2,0x20(%rsp)
+	vpaddq	0x60($Tbl),@X[6],$t2
+	vmovdqa	$t3,0x30(%rsp)
+	vpaddq	0x70($Tbl),@X[7],$t3
+	vmovdqa	$t0,0x40(%rsp)
+	mov	$A,$a1
+	vmovdqa	$t1,0x50(%rsp)
+	mov	$B,$a3
+	vmovdqa	$t2,0x60(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x70(%rsp)
+	mov	$E,$a0
+	jmp	.Lxop_00_47
+
+.align	16
+.Lxop_00_47:
+	add	\$16*$SZ,$Tbl
+___
+sub XOP_512_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body);			# 52 instructions
+
+	&vpalignr	($t0,@X[1],@X[0],$SZ);	# X[1..2]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpalignr	($t3,@X[5],@X[4],$SZ);	# X[9..10]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vprotq		($t1,$t0,8*$SZ-$sigma0[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpsrlq		($t0,$t0,$sigma0[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpaddq	(@X[0],@X[0],$t3);	# X[0..1] += X[9..10]
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vprotq		($t2,$t1,$sigma0[1]-$sigma0[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpxor		($t0,$t0,$t1);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotq	($t3,@X[7],8*$SZ-$sigma1[1]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpxor		($t0,$t0,$t2);		# sigma0(X[1..2])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpsrlq	($t2,@X[7],$sigma1[2]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddq		(@X[0],@X[0],$t0);	# X[0..1] += sigma0(X[1..2])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vprotq	($t1,$t3,$sigma1[1]-$sigma1[0]);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t2);
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	 &vpxor		($t3,$t3,$t1);		# sigma1(X[14..15])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddq		(@X[0],@X[0],$t3);	# X[0..1] += sigma1(X[14..15])
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	  eval(shift(@insns));
+	&vpaddq		($t2,@X[0],16*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	(16*$j."(%rsp)",$t2);
+}
+
+    for ($i=0,$j=0; $j<8; $j++) {
+	&XOP_512_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+	&cmpb	($SZ-1+16*$SZ."($Tbl)",0);
+	&jne	(".Lxop_00_47");
+
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+}
+$code.=<<___;
+	mov	$_ctx,$ctx
+	mov	$a1,$A
+
+	add	$SZ*0($ctx),$A
+	lea	16*$SZ($inp),$inp
+	add	$SZ*1($ctx),$B
+	add	$SZ*2($ctx),$C
+	add	$SZ*3($ctx),$D
+	add	$SZ*4($ctx),$E
+	add	$SZ*5($ctx),$F
+	add	$SZ*6($ctx),$G
+	add	$SZ*7($ctx),$H
+
+	cmp	$_end,$inp
+
+	mov	$A,$SZ*0($ctx)
+	mov	$B,$SZ*1($ctx)
+	mov	$C,$SZ*2($ctx)
+	mov	$D,$SZ*3($ctx)
+	mov	$E,$SZ*4($ctx)
+	mov	$F,$SZ*5($ctx)
+	mov	$G,$SZ*6($ctx)
+	mov	$H,$SZ*7($ctx)
+	jb	.Lloop_xop
+
+	mov	$_rsp,%rsi
+	vzeroall
+___
+$code.=<<___ if ($win64);
+	movaps	16*$SZ+32(%rsp),%xmm6
+	movaps	16*$SZ+48(%rsp),%xmm7
+	movaps	16*$SZ+64(%rsp),%xmm8
+	movaps	16*$SZ+80(%rsp),%xmm9
+___
+$code.=<<___ if ($win64 && $SZ>4);
+	movaps	16*$SZ+96(%rsp),%xmm10
+	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
+.Lepilogue_xop:
+	ret
+.size	${func}_xop,.-${func}_xop
+___
+}
+######################################################################
+# AVX+shrd code path
+#
+local *ror = sub { &shrd(@_[0],@_) };
+
+$code.=<<___;
+.type	${func}_avx,\@function,4
+.align	64
+${func}_avx:
+.Lavx_shortcut:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	mov	%rsp,%r11		# copy %rsp
+	shl	\$4,%rdx		# num*16
+	sub	\$`$framesz+$win64*16*($SZ==4?4:6)`,%rsp
+	lea	($inp,%rdx,$SZ),%rdx	# inp+num*16*$SZ
+	and	\$-64,%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
+___
+$code.=<<___ if ($win64);
+	movaps	%xmm6,16*$SZ+32(%rsp)
+	movaps	%xmm7,16*$SZ+48(%rsp)
+	movaps	%xmm8,16*$SZ+64(%rsp)
+	movaps	%xmm9,16*$SZ+80(%rsp)
+___
+$code.=<<___ if ($win64 && $SZ>4);
+	movaps	%xmm10,16*$SZ+96(%rsp)
+	movaps	%xmm11,16*$SZ+112(%rsp)
+___
+$code.=<<___;
+.Lprologue_avx:
+
+	vzeroall
+	mov	$SZ*0($ctx),$A
+	mov	$SZ*1($ctx),$B
+	mov	$SZ*2($ctx),$C
+	mov	$SZ*3($ctx),$D
+	mov	$SZ*4($ctx),$E
+	mov	$SZ*5($ctx),$F
+	mov	$SZ*6($ctx),$G
+	mov	$SZ*7($ctx),$H
+___
+					if ($SZ==4) {	# SHA256
+    my @X = map("%xmm$_",(0..3));
+    my ($t0,$t1,$t2,$t3, $t4,$t5) = map("%xmm$_",(4..9));
+
+$code.=<<___;
+	vmovdqa	$TABLE+`$SZ*$rounds`+16(%rip),$t4
+	vmovdqa	$TABLE+`$SZ*$rounds`+32(%rip),$t5
+	jmp	.Lloop_avx
+.align	16
+.Lloop_avx:
+	vmovdqa	$TABLE+`$SZ*$rounds`(%rip),$t3
+	vmovdqu	0x00($inp),@X[0]
+	vmovdqu	0x10($inp),@X[1]
+	vmovdqu	0x20($inp),@X[2]
+	vmovdqu	0x30($inp),@X[3]
+	vpshufb	$t3,@X[0],@X[0]
+	lea	$TABLE(%rip),$Tbl
+	vpshufb	$t3,@X[1],@X[1]
+	vpshufb	$t3,@X[2],@X[2]
+	vpaddd	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[3],@X[3]
+	vpaddd	0x10($Tbl),@X[1],$t1
+	vpaddd	0x20($Tbl),@X[2],$t2
+	vpaddd	0x30($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	mov	$A,$a1
+	vmovdqa	$t1,0x10(%rsp)
+	mov	$B,$a3
+	vmovdqa	$t2,0x20(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x30(%rsp)
+	mov	$E,$a0
+	jmp	.Lavx_00_47
+
+.align	16
+.Lavx_00_47:
+	add	\$16*$SZ,$Tbl
+___
+sub Xupdate_256_AVX () {
+	(
+	'&vpalignr	($t0,@X[1],@X[0],$SZ)',	# X[1..4]
+	 '&vpalignr	($t3,@X[3],@X[2],$SZ)',	# X[9..12]
+	'&vpsrld	($t2,$t0,$sigma0[0]);',
+	 '&vpaddd	(@X[0],@X[0],$t3)',	# X[0..3] += X[9..12]
+	'&vpsrld	($t3,$t0,$sigma0[2])',
+	'&vpslld	($t1,$t0,8*$SZ-$sigma0[1]);',
+	'&vpxor		($t0,$t3,$t2)',
+	 '&vpshufd	($t3,@X[3],0b11111010)',# X[14..15]
+	'&vpsrld	($t2,$t2,$sigma0[1]-$sigma0[0]);',
+	'&vpxor		($t0,$t0,$t1)',
+	'&vpslld	($t1,$t1,$sigma0[1]-$sigma0[0]);',
+	'&vpxor		($t0,$t0,$t2)',
+	 '&vpsrld	($t2,$t3,$sigma1[2]);',
+	'&vpxor		($t0,$t0,$t1)',		# sigma0(X[1..4])
+	 '&vpsrlq	($t3,$t3,$sigma1[0]);',
+	'&vpaddd	(@X[0],@X[0],$t0)',	# X[0..3] += sigma0(X[1..4])
+	 '&vpxor	($t2,$t2,$t3);',
+	 '&vpsrlq	($t3,$t3,$sigma1[1]-$sigma1[0])',
+	 '&vpxor	($t2,$t2,$t3)',
+	 '&vpshufb	($t2,$t2,$t4)',		# sigma1(X[14..15])
+	'&vpaddd	(@X[0],@X[0],$t2)',	# X[0..1] += sigma1(X[14..15])
+	 '&vpshufd	($t3,@X[0],0b01010000)',# X[16..17]
+	 '&vpsrld	($t2,$t3,$sigma1[2])',
+	 '&vpsrlq	($t3,$t3,$sigma1[0])',
+	 '&vpxor	($t2,$t2,$t3);',
+	 '&vpsrlq	($t3,$t3,$sigma1[1]-$sigma1[0])',
+	 '&vpxor	($t2,$t2,$t3)',
+	 '&vpshufb	($t2,$t2,$t5)',
+	'&vpaddd	(@X[0],@X[0],$t2)'	# X[2..3] += sigma1(X[16..17])
+	);
+}
+
+sub AVX_256_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body,&$body,&$body);	# 104 instructions
+
+	foreach (Xupdate_256_AVX()) {		# 29 instructions
+	    eval;
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	}
+	&vpaddd		($t2,@X[0],16*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	(16*$j."(%rsp)",$t2);
+}
+
+    for ($i=0,$j=0; $j<4; $j++) {
+	&AVX_256_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+	&cmpb	($SZ-1+16*$SZ."($Tbl)",0);
+	&jne	(".Lavx_00_47");
+
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+
+					} else {	# SHA512
+    my @X = map("%xmm$_",(0..7));
+    my ($t0,$t1,$t2,$t3) = map("%xmm$_",(8..11));
+
+$code.=<<___;
+	jmp	.Lloop_avx
+.align	16
+.Lloop_avx:
+	vmovdqa	$TABLE+`$SZ*$rounds`(%rip),$t3
+	vmovdqu	0x00($inp),@X[0]
+	lea	$TABLE(%rip),$Tbl
+	vmovdqu	0x10($inp),@X[1]
+	vmovdqu	0x20($inp),@X[2]
+	vpshufb	$t3,@X[0],@X[0]
+	vmovdqu	0x30($inp),@X[3]
+	vpshufb	$t3,@X[1],@X[1]
+	vmovdqu	0x40($inp),@X[4]
+	vpshufb	$t3,@X[2],@X[2]
+	vmovdqu	0x50($inp),@X[5]
+	vpshufb	$t3,@X[3],@X[3]
+	vmovdqu	0x60($inp),@X[6]
+	vpshufb	$t3,@X[4],@X[4]
+	vmovdqu	0x70($inp),@X[7]
+	vpshufb	$t3,@X[5],@X[5]
+	vpaddq	0x00($Tbl),@X[0],$t0
+	vpshufb	$t3,@X[6],@X[6]
+	vpaddq	0x10($Tbl),@X[1],$t1
+	vpshufb	$t3,@X[7],@X[7]
+	vpaddq	0x20($Tbl),@X[2],$t2
+	vpaddq	0x30($Tbl),@X[3],$t3
+	vmovdqa	$t0,0x00(%rsp)
+	vpaddq	0x40($Tbl),@X[4],$t0
+	vmovdqa	$t1,0x10(%rsp)
+	vpaddq	0x50($Tbl),@X[5],$t1
+	vmovdqa	$t2,0x20(%rsp)
+	vpaddq	0x60($Tbl),@X[6],$t2
+	vmovdqa	$t3,0x30(%rsp)
+	vpaddq	0x70($Tbl),@X[7],$t3
+	vmovdqa	$t0,0x40(%rsp)
+	mov	$A,$a1
+	vmovdqa	$t1,0x50(%rsp)
+	mov	$B,$a3
+	vmovdqa	$t2,0x60(%rsp)
+	xor	$C,$a3			# magic
+	vmovdqa	$t3,0x70(%rsp)
+	mov	$E,$a0
+	jmp	.Lavx_00_47
+
+.align	16
+.Lavx_00_47:
+	add	\$16*$SZ,$Tbl
+___
+sub Xupdate_512_AVX () {
+	(
+	'&vpalignr	($t0,@X[1],@X[0],$SZ)',	# X[1..2]
+	 '&vpalignr	($t3,@X[5],@X[4],$SZ)',	# X[9..10]
+	'&vpsrlq	($t2,$t0,$sigma0[0]);',
+	 '&vpaddq	(@X[0],@X[0],$t3)',	# X[0..1] += X[9..10]
+	'&vpsrlq	($t3,$t0,$sigma0[2])',
+	'&vpsllq	($t1,$t0,8*$SZ-$sigma0[1]);',
+	 '&vpxor	($t0,$t3,$t2)',
+	'&vpsrlq	($t2,$t2,$sigma0[1]-$sigma0[0]);',
+	 '&vpxor	($t0,$t0,$t1)',
+	'&vpsllq	($t1,$t1,$sigma0[1]-$sigma0[0]);',
+	 '&vpxor	($t0,$t0,$t2)',
+	 '&vpsrlq	($t3,@X[7],$sigma1[2]);',
+	'&vpxor		($t0,$t0,$t1)',		# sigma0(X[1..2])
+	 '&vpsllq	($t2,@X[7],8*$SZ-$sigma1[1])',
+	'&vpaddq	(@X[0],@X[0],$t0)',	# X[0..1] += sigma0(X[1..2])
+	 '&vpsrlq	($t1,@X[7],$sigma1[0]);',
+	 '&vpxor	($t3,$t3,$t2)',
+	 '&vpsllq	($t2,$t2,$sigma1[1]-$sigma1[0]);',
+	 '&vpxor	($t3,$t3,$t1)',
+	 '&vpsrlq	($t1,$t1,$sigma1[1]-$sigma1[0]);',
+	 '&vpxor	($t3,$t3,$t2)',
+	 '&vpxor	($t3,$t3,$t1)',		# sigma1(X[14..15])
+	'&vpaddq	(@X[0],@X[0],$t3)',	# X[0..1] += sigma1(X[14..15])
+	);
+}
+
+sub AVX_512_00_47 () {
+my $j = shift;
+my $body = shift;
+my @X = @_;
+my @insns = (&$body,&$body);			# 52 instructions
+
+	foreach (Xupdate_512_AVX()) {		# 23 instructions
+	    eval;
+	    eval(shift(@insns));
+	    eval(shift(@insns));
+	}
+	&vpaddq		($t2,@X[0],16*$j."($Tbl)");
+	  foreach (@insns) { eval; }		# remaining instructions
+	&vmovdqa	(16*$j."(%rsp)",$t2);
+}
+
+    for ($i=0,$j=0; $j<8; $j++) {
+	&AVX_512_00_47($j,\&body_00_15,@X);
+	push(@X,shift(@X));			# rotate(@X)
+    }
+	&cmpb	($SZ-1+16*$SZ."($Tbl)",0);
+	&jne	(".Lavx_00_47");
+
+    for ($i=0; $i<16; ) {
+	foreach(body_00_15()) { eval; }
+    }
+}
+$code.=<<___;
+	mov	$_ctx,$ctx
+	mov	$a1,$A
+
+	add	$SZ*0($ctx),$A
+	lea	16*$SZ($inp),$inp
+	add	$SZ*1($ctx),$B
+	add	$SZ*2($ctx),$C
+	add	$SZ*3($ctx),$D
+	add	$SZ*4($ctx),$E
+	add	$SZ*5($ctx),$F
+	add	$SZ*6($ctx),$G
+	add	$SZ*7($ctx),$H
+
+	cmp	$_end,$inp
+
+	mov	$A,$SZ*0($ctx)
+	mov	$B,$SZ*1($ctx)
+	mov	$C,$SZ*2($ctx)
+	mov	$D,$SZ*3($ctx)
+	mov	$E,$SZ*4($ctx)
+	mov	$F,$SZ*5($ctx)
+	mov	$G,$SZ*6($ctx)
+	mov	$H,$SZ*7($ctx)
+	jb	.Lloop_avx
+
+	mov	$_rsp,%rsi
+	vzeroall
+___
+$code.=<<___ if ($win64);
+	movaps	16*$SZ+32(%rsp),%xmm6
+	movaps	16*$SZ+48(%rsp),%xmm7
+	movaps	16*$SZ+64(%rsp),%xmm8
+	movaps	16*$SZ+80(%rsp),%xmm9
+___
+$code.=<<___ if ($win64 && $SZ>4);
+	movaps	16*$SZ+96(%rsp),%xmm10
+	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
+.Lepilogue_avx:
+	ret
+.size	${func}_avx,.-${func}_avx
+___
+}}}}}
+
 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
 #		CONTEXT *context,DISPATCHER_CONTEXT *disp)
 if ($win64) {
@@ -376,16 +1532,22 @@ se_handler:
 	mov	120($context),%rax	# pull context->Rax
 	mov	248($context),%rbx	# pull context->Rip
 
-	lea	.Lprologue(%rip),%r10
-	cmp	%r10,%rbx		# context->Rip<.Lprologue
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HanderlData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
 	jb	.Lin_prologue
 
 	mov	152($context),%rax	# pull context->Rsp
 
-	lea	.Lepilogue(%rip),%r10
-	cmp	%r10,%rbx		# context->Rip>=.Lepilogue
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
 	jae	.Lin_prologue
 
+	mov	%rax,%rsi		# put aside Rsp
 	mov	16*$SZ+3*8(%rax),%rax	# pull $_rsp
 	lea	48(%rax),%rax
 
@@ -402,6 +1564,15 @@ se_handler:
 	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*$SZ+4*8(%rsi),%rsi	# Xmm6- save area
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$`$SZ==4?8:12`,%ecx
+	.long	0xa548f3fc		# cld; rep movsq
+
 .Lin_prologue:
 	mov	8(%rax),%rdi
 	mov	16(%rax),%rsi
@@ -447,12 +1618,47 @@ se_handler:
 	.rva	.LSEH_begin_$func
 	.rva	.LSEH_end_$func
 	.rva	.LSEH_info_$func
-
+___
+$code.=<<___ if ($SZ==4);
+	.rva	.LSEH_begin_${func}_ssse3
+	.rva	.LSEH_end_${func}_ssse3
+	.rva	.LSEH_info_${func}_ssse3
+___
+$code.=<<___ if ($avx && $SZ==8);
+	.rva	.LSEH_begin_${func}_xop
+	.rva	.LSEH_end_${func}_xop
+	.rva	.LSEH_info_${func}_xop
+___
+$code.=<<___ if ($avx);
+	.rva	.LSEH_begin_${func}_avx
+	.rva	.LSEH_end_${func}_avx
+	.rva	.LSEH_info_${func}_xop
+___
+$code.=<<___;
 .section	.xdata
 .align	8
 .LSEH_info_$func:
 	.byte	9,0,0,0
 	.rva	se_handler
+	.rva	.Lprologue,.Lepilogue			# HandlerData[]
+___
+$code.=<<___ if ($SZ==4);
+.LSEH_info_${func}_ssse3:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_ssse3,.Lepilogue_ssse3	# HandlerData[]
+___
+$code.=<<___ if ($avx && $SZ==8);
+.LSEH_info_${func}_xop:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_xop,.Lepilogue_xop		# HandlerData[]
+___
+$code.=<<___ if ($avx);
+.LSEH_info_${func}_avx:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lprologue_avx,.Lepilogue_avx		# HandlerData[]
 ___
 }