Check in pristine copies of upstream's poly1305 assembly.

Taken from 6b2ebe4332e22b4eb7dd6fadf418e3da7b926ca4. These don't do anything
right now but are checked in unmodified to make diffs easier to see.

Change-Id: I4f5bdb7b16f4ac27e7ef175f475540c481b8d593
Reviewed-on: https://boringssl-review.googlesource.com/7224
Reviewed-by: Adam Langley <agl@google.com>

4 files changed, 6181 insertions, 0 deletions

diff --git a/crypto/poly1305/asm/poly1305-armv4.pl b/crypto/poly1305/asm/poly1305-armv4.pl
new file mode 100755
index 00000000..65b79cf1
--- /dev/null
+++ b/crypto/poly1305/asm/poly1305-armv4.pl
@@ -0,0 +1,1216 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+#			IALU(*)/gcc-4.4		NEON
+#
+# ARM11xx(ARMv6)	7.78/+100%		-
+# Cortex-A5		6.30/+130%		2.96
+# Cortex-A8		6.25/+115%		2.36
+# Cortex-A9		5.10/+95%		2.55
+# Cortex-A15		3.79/+85%		1.25(**)
+# Snapdragon S4		5.70/+100%		1.48(**)
+#
+# (*)	this is for -march=armv6, i.e. with bunch of ldrb loading data;
+# (**)	these are trade-off results, they can be improved by ~8% but at
+#	the cost of 15/12% regression on Cortex-A5/A7, it's even possible
+#	to improve Cortex-A9 result, but then A5/A7 loose more than 20%;
+
+$flavour = shift;
+if ($flavour=~/^\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
+else { while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {} }
+
+if ($flavour && $flavour ne "void") {
+    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
+    die "can't locate arm-xlate.pl";
+
+    open STDOUT,"| \"$^X\" $xlate $flavour $output";
+} else {
+    open STDOUT,">$output";
+}
+
+($ctx,$inp,$len,$padbit)=map("r$_",(0..3));
+
+$code.=<<___;
+#include "arm_arch.h"
+
+.text
+#if defined(__thumb2__)
+.syntax	unified
+.thumb
+#else
+.code	32
+#endif
+
+.globl	poly1305_emit
+.globl	poly1305_blocks
+.globl	poly1305_init
+.type	poly1305_init,%function
+.align	5
+poly1305_init:
+.Lpoly1305_init:
+	stmdb	sp!,{r4-r11}
+
+	eor	r3,r3,r3
+	cmp	$inp,#0
+	str	r3,[$ctx,#0]		@ zero hash value
+	str	r3,[$ctx,#4]
+	str	r3,[$ctx,#8]
+	str	r3,[$ctx,#12]
+	str	r3,[$ctx,#16]
+	str	r3,[$ctx,#36]		@ is_base2_26
+	add	$ctx,$ctx,#20
+
+#ifdef	__thumb2__
+	it	eq
+#endif
+	moveq	r0,#0
+	beq	.Lno_key
+
+#if	__ARM_MAX_ARCH__>=7
+	adr	r11,.Lpoly1305_init
+	ldr	r12,.LOPENSSL_armcap
+#endif
+	ldrb	r4,[$inp,#0]
+	mov	r10,#0x0fffffff
+	ldrb	r5,[$inp,#1]
+	and	r3,r10,#-4		@ 0x0ffffffc
+	ldrb	r6,[$inp,#2]
+	ldrb	r7,[$inp,#3]
+	orr	r4,r4,r5,lsl#8
+	ldrb	r5,[$inp,#4]
+	orr	r4,r4,r6,lsl#16
+	ldrb	r6,[$inp,#5]
+	orr	r4,r4,r7,lsl#24
+	ldrb	r7,[$inp,#6]
+	and	r4,r4,r10
+
+#if	__ARM_MAX_ARCH__>=7
+	ldr	r12,[r11,r12]		@ OPENSSL_armcap_P
+# ifdef	__APPLE__
+	ldr	r12,[r12]
+# endif
+#endif
+	ldrb	r8,[$inp,#7]
+	orr	r5,r5,r6,lsl#8
+	ldrb	r6,[$inp,#8]
+	orr	r5,r5,r7,lsl#16
+	ldrb	r7,[$inp,#9]
+	orr	r5,r5,r8,lsl#24
+	ldrb	r8,[$inp,#10]
+	and	r5,r5,r3
+
+#if	__ARM_MAX_ARCH__>=7
+	tst	r12,#1			@ check for NEON
+# ifdef	__APPLE__
+	adr	r9,poly1305_blocks_neon
+	adr	r11,poly1305_blocks
+#  ifdef __thumb2__
+	it	ne
+#  endif
+	movne	r11,r9
+	adr	r12,poly1305_emit
+	adr	r10,poly1305_emit_neon
+#  ifdef __thumb2__
+	it	ne
+#  endif
+	movne	r12,r10
+# else
+#  ifdef __thumb2__
+	itete	eq
+#  endif
+	addeq	r12,r11,#(poly1305_emit-.Lpoly1305_init)
+	addne	r12,r11,#(poly1305_emit_neon-.Lpoly1305_init)
+	addeq	r11,r11,#(poly1305_blocks-.Lpoly1305_init)
+	addne	r11,r11,#(poly1305_blocks_neon-.Lpoly1305_init)
+# endif
+# ifdef	__thumb2__
+	orr	r12,r12,#1	@ thumb-ify address
+	orr	r11,r11,#1
+# endif
+#endif
+	ldrb	r9,[$inp,#11]
+	orr	r6,r6,r7,lsl#8
+	ldrb	r7,[$inp,#12]
+	orr	r6,r6,r8,lsl#16
+	ldrb	r8,[$inp,#13]
+	orr	r6,r6,r9,lsl#24
+	ldrb	r9,[$inp,#14]
+	and	r6,r6,r3
+
+	ldrb	r10,[$inp,#15]
+	orr	r7,r7,r8,lsl#8
+	str	r4,[$ctx,#0]
+	orr	r7,r7,r9,lsl#16
+	str	r5,[$ctx,#4]
+	orr	r7,r7,r10,lsl#24
+	str	r6,[$ctx,#8]
+	and	r7,r7,r3
+	str	r7,[$ctx,#12]
+#if	__ARM_MAX_ARCH__>=7
+	stmia	r2,{r11,r12}		@ fill functions table
+	mov	r0,#1
+#else
+	mov	r0,#0
+#endif
+.Lno_key:
+	ldmia	sp!,{r4-r11}
+#if	__ARM_ARCH__>=5
+	ret				@ bx	lr
+#else
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	poly1305_init,.-poly1305_init
+___
+{
+my ($h0,$h1,$h2,$h3,$h4,$r0,$r1,$r2,$r3)=map("r$_",(4..12));
+my ($s1,$s2,$s3)=($r1,$r2,$r3);
+
+$code.=<<___;
+.type	poly1305_blocks,%function
+.align	5
+poly1305_blocks:
+	stmdb	sp!,{r3-r11,lr}
+
+	ands	$len,$len,#-16
+	beq	.Lno_data
+
+	cmp	$padbit,#0
+	add	$len,$len,$inp		@ end pointer
+	sub	sp,sp,#32
+
+	ldmia	$ctx,{$h0-$r3}		@ load context
+
+	str	$ctx,[sp,#12]		@ offload stuff
+	mov	lr,$inp
+	str	$len,[sp,#16]
+	str	$r1,[sp,#20]
+	str	$r2,[sp,#24]
+	str	$r3,[sp,#28]
+	b	.Loop
+
+.Loop:
+#if __ARM_ARCH__<7
+	ldrb	r0,[lr],#16		@ load input
+# ifdef	__thumb2__
+	it	hi
+# endif
+	addhi	$h4,$h4,#1		@ 1<<128
+	ldrb	r1,[lr,#-15]
+	ldrb	r2,[lr,#-14]
+	ldrb	r3,[lr,#-13]
+	orr	r1,r0,r1,lsl#8
+	ldrb	r0,[lr,#-12]
+	orr	r2,r1,r2,lsl#16
+	ldrb	r1,[lr,#-11]
+	orr	r3,r2,r3,lsl#24
+	ldrb	r2,[lr,#-10]
+	adds	$h0,$h0,r3		@ accumulate input
+
+	ldrb	r3,[lr,#-9]
+	orr	r1,r0,r1,lsl#8
+	ldrb	r0,[lr,#-8]
+	orr	r2,r1,r2,lsl#16
+	ldrb	r1,[lr,#-7]
+	orr	r3,r2,r3,lsl#24
+	ldrb	r2,[lr,#-6]
+	adcs	$h1,$h1,r3
+
+	ldrb	r3,[lr,#-5]
+	orr	r1,r0,r1,lsl#8
+	ldrb	r0,[lr,#-4]
+	orr	r2,r1,r2,lsl#16
+	ldrb	r1,[lr,#-3]
+	orr	r3,r2,r3,lsl#24
+	ldrb	r2,[lr,#-2]
+	adcs	$h2,$h2,r3
+
+	ldrb	r3,[lr,#-1]
+	orr	r1,r0,r1,lsl#8
+	str	lr,[sp,#8]		@ offload input pointer
+	orr	r2,r1,r2,lsl#16
+	add	$s1,$r1,$r1,lsr#2
+	orr	r3,r2,r3,lsl#24
+#else
+	ldr	r0,[lr],#16		@ load input
+# ifdef	__thumb2__
+	it	hi
+# endif
+	addhi	$h4,$h4,#1		@ padbit
+	ldr	r1,[lr,#-12]
+	ldr	r2,[lr,#-8]
+	ldr	r3,[lr,#-4]
+# ifdef	__ARMEB__
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+# endif
+	adds	$h0,$h0,r0		@ accumulate input
+	str	lr,[sp,#8]		@ offload input pointer
+	adcs	$h1,$h1,r1
+	add	$s1,$r1,$r1,lsr#2
+	adcs	$h2,$h2,r2
+#endif
+	add	$s2,$r2,$r2,lsr#2
+	adcs	$h3,$h3,r3
+	add	$s3,$r3,$r3,lsr#2
+
+	umull	r2,r3,$h1,$r0
+	 adc	$h4,$h4,#0
+	umull	r0,r1,$h0,$r0
+	umlal	r2,r3,$h4,$s1
+	umlal	r0,r1,$h3,$s1
+	ldr	$r1,[sp,#20]		@ reload $r1
+	umlal	r2,r3,$h2,$s3
+	umlal	r0,r1,$h1,$s3
+	umlal	r2,r3,$h3,$s2
+	umlal	r0,r1,$h2,$s2
+	umlal	r2,r3,$h0,$r1
+	str	r0,[sp,#0]		@ future $h0
+	 mul	r0,$s2,$h4
+	ldr	$r2,[sp,#24]		@ reload $r2
+	adds	r2,r2,r1		@ d1+=d0>>32
+	 eor	r1,r1,r1
+	adc	lr,r3,#0		@ future $h2
+	str	r2,[sp,#4]		@ future $h1
+
+	mul	r2,$s3,$h4
+	eor	r3,r3,r3
+	umlal	r0,r1,$h3,$s3
+	ldr	$r3,[sp,#28]		@ reload $r3
+	umlal	r2,r3,$h3,$r0
+	umlal	r0,r1,$h2,$r0
+	umlal	r2,r3,$h2,$r1
+	umlal	r0,r1,$h1,$r1
+	umlal	r2,r3,$h1,$r2
+	umlal	r0,r1,$h0,$r2
+	umlal	r2,r3,$h0,$r3
+	ldr	$h0,[sp,#0]
+	mul	$h4,$r0,$h4
+	ldr	$h1,[sp,#4]
+
+	adds	$h2,lr,r0		@ d2+=d1>>32
+	ldr	lr,[sp,#8]		@ reload input pointer
+	adc	r1,r1,#0
+	adds	$h3,r2,r1		@ d3+=d2>>32
+	ldr	r0,[sp,#16]		@ reload end pointer
+	adc	r3,r3,#0
+	add	$h4,$h4,r3		@ h4+=d3>>32
+
+	and	r1,$h4,#-4
+	and	$h4,$h4,#3
+	add	r1,r1,r1,lsr#2		@ *=5
+	adds	$h0,$h0,r1
+	adcs	$h1,$h1,#0
+	adcs	$h2,$h2,#0
+	adc	$h3,$h3,#0
+
+	cmp	r0,lr			@ done yet?
+	bhi	.Loop
+
+	ldr	$ctx,[sp,#12]
+	add	sp,sp,#32
+	stmia	$ctx,{$h0-$h4}		@ store the result
+
+.Lno_data:
+#if	__ARM_ARCH__>=5
+	ldmia	sp!,{r3-r11,pc}
+#else
+	ldmia	sp!,{r3-r11,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	poly1305_blocks,.-poly1305_blocks
+___
+}
+{
+my ($ctx,$mac,$nonce)=map("r$_",(0..2));
+my ($h0,$h1,$h2,$h3,$h4,$g0,$g1,$g2,$g3)=map("r$_",(3..11));
+my $g4=$h4;
+
+$code.=<<___;
+.type	poly1305_emit,%function
+.align	5
+poly1305_emit:
+	stmdb	sp!,{r4-r11}
+.Lpoly1305_emit_enter:
+
+	ldmia	$ctx,{$h0-$h4}
+	adds	$g0,$h0,#5		@ compare to modulus
+	adcs	$g1,$h1,#0
+	adcs	$g2,$h2,#0
+	adcs	$g3,$h3,#0
+	adc	$g4,$h4,#0
+	tst	$g4,#4			@ did it carry/borrow?
+
+#ifdef	__thumb2__
+	it	ne
+#endif
+	movne	$h0,$g0
+	ldr	$g0,[$nonce,#0]
+#ifdef	__thumb2__
+	it	ne
+#endif
+	movne	$h1,$g1
+	ldr	$g1,[$nonce,#4]
+#ifdef	__thumb2__
+	it	ne
+#endif
+	movne	$h2,$g2
+	ldr	$g2,[$nonce,#8]
+#ifdef	__thumb2__
+	it	ne
+#endif
+	movne	$h3,$g3
+	ldr	$g3,[$nonce,#12]
+
+	adds	$h0,$h0,$g0
+	adcs	$h1,$h1,$g1
+	adcs	$h2,$h2,$g2
+	adc	$h3,$h3,$g3
+
+#if __ARM_ARCH__>=7
+# ifdef __ARMEB__
+	rev	$h0,$h0
+	rev	$h1,$h1
+	rev	$h2,$h2
+	rev	$h3,$h3
+# endif
+	str	$h0,[$mac,#0]
+	str	$h1,[$mac,#4]
+	str	$h2,[$mac,#8]
+	str	$h3,[$mac,#12]
+#else
+	strb	$h0,[$mac,#0]
+	mov	$h0,$h0,lsr#8
+	strb	$h1,[$mac,#4]
+	mov	$h1,$h1,lsr#8
+	strb	$h2,[$mac,#8]
+	mov	$h2,$h2,lsr#8
+	strb	$h3,[$mac,#12]
+	mov	$h3,$h3,lsr#8
+
+	strb	$h0,[$mac,#1]
+	mov	$h0,$h0,lsr#8
+	strb	$h1,[$mac,#5]
+	mov	$h1,$h1,lsr#8
+	strb	$h2,[$mac,#9]
+	mov	$h2,$h2,lsr#8
+	strb	$h3,[$mac,#13]
+	mov	$h3,$h3,lsr#8
+
+	strb	$h0,[$mac,#2]
+	mov	$h0,$h0,lsr#8
+	strb	$h1,[$mac,#6]
+	mov	$h1,$h1,lsr#8
+	strb	$h2,[$mac,#10]
+	mov	$h2,$h2,lsr#8
+	strb	$h3,[$mac,#14]
+	mov	$h3,$h3,lsr#8
+
+	strb	$h0,[$mac,#3]
+	strb	$h1,[$mac,#7]
+	strb	$h2,[$mac,#11]
+	strb	$h3,[$mac,#15]
+#endif
+	ldmia	sp!,{r4-r11}
+#if	__ARM_ARCH__>=5
+	ret				@ bx	lr
+#else
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	poly1305_emit,.-poly1305_emit
+___
+{
+my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("d$_",(0..9));
+my ($D0,$D1,$D2,$D3,$D4, $H0,$H1,$H2,$H3,$H4) = map("q$_",(5..14));
+my ($T0,$T1,$MASK) = map("q$_",(15,4,0));
+
+my ($in2,$zeros,$tbl0,$tbl1) = map("r$_",(4..7));
+
+$code.=<<___;
+#if	__ARM_MAX_ARCH__>=7
+.fpu	neon
+
+.type	poly1305_init_neon,%function
+.align	5
+poly1305_init_neon:
+	ldr	r4,[$ctx,#20]		@ load key base 2^32
+	ldr	r5,[$ctx,#24]
+	ldr	r6,[$ctx,#28]
+	ldr	r7,[$ctx,#32]
+
+	and	r2,r4,#0x03ffffff	@ base 2^32 -> base 2^26
+	mov	r3,r4,lsr#26
+	mov	r4,r5,lsr#20
+	orr	r3,r3,r5,lsl#6
+	mov	r5,r6,lsr#14
+	orr	r4,r4,r6,lsl#12
+	mov	r6,r7,lsr#8
+	orr	r5,r5,r7,lsl#18
+	and	r3,r3,#0x03ffffff
+	and	r4,r4,#0x03ffffff
+	and	r5,r5,#0x03ffffff
+
+	vdup.32	$R0,r2			@ r^1 in both lanes
+	add	r2,r3,r3,lsl#2		@ *5
+	vdup.32	$R1,r3
+	add	r3,r4,r4,lsl#2
+	vdup.32	$S1,r2
+	vdup.32	$R2,r4
+	add	r4,r5,r5,lsl#2
+	vdup.32	$S2,r3
+	vdup.32	$R3,r5
+	add	r5,r6,r6,lsl#2
+	vdup.32	$S3,r4
+	vdup.32	$R4,r6
+	vdup.32	$S4,r5
+
+	mov	$zeros,#2		@ counter
+
+.Lsquare_neon:
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+	@ d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	@ d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	@ d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	@ d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+
+	vmull.u32	$D0,$R0,${R0}[1]
+	vmull.u32	$D1,$R1,${R0}[1]
+	vmull.u32	$D2,$R2,${R0}[1]
+	vmull.u32	$D3,$R3,${R0}[1]
+	vmull.u32	$D4,$R4,${R0}[1]
+
+	vmlal.u32	$D0,$R4,${S1}[1]
+	vmlal.u32	$D1,$R0,${R1}[1]
+	vmlal.u32	$D2,$R1,${R1}[1]
+	vmlal.u32	$D3,$R2,${R1}[1]
+	vmlal.u32	$D4,$R3,${R1}[1]
+
+	vmlal.u32	$D0,$R3,${S2}[1]
+	vmlal.u32	$D1,$R4,${S2}[1]
+	vmlal.u32	$D3,$R1,${R2}[1]
+	vmlal.u32	$D2,$R0,${R2}[1]
+	vmlal.u32	$D4,$R2,${R2}[1]
+
+	vmlal.u32	$D0,$R2,${S3}[1]
+	vmlal.u32	$D3,$R0,${R3}[1]
+	vmlal.u32	$D1,$R3,${S3}[1]
+	vmlal.u32	$D2,$R4,${S3}[1]
+	vmlal.u32	$D4,$R1,${R3}[1]
+
+	vmlal.u32	$D3,$R4,${S4}[1]
+	vmlal.u32	$D0,$R1,${S4}[1]
+	vmlal.u32	$D1,$R2,${S4}[1]
+	vmlal.u32	$D2,$R3,${S4}[1]
+	vmlal.u32	$D4,$R0,${R4}[1]
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
+	@ and P. Schwabe
+
+	vshr.u64	$T0,$D3,#26
+	vmovn.i64	$D3#lo,$D3
+	 vshr.u64	$T1,$D0,#26
+	 vmovn.i64	$D0#lo,$D0
+	vadd.i64	$D4,$D4,$T0		@ h3 -> h4
+	vbic.i32	$D3#lo,#0xfc000000	@ &=0x03ffffff
+	 vadd.i64	$D1,$D1,$T1		@ h0 -> h1
+	 vbic.i32	$D0#lo,#0xfc000000
+
+	vshrn.u64	$T0#lo,$D4,#26
+	vmovn.i64	$D4#lo,$D4
+	 vshr.u64	$T1,$D1,#26
+	 vmovn.i64	$D1#lo,$D1
+	 vadd.i64	$D2,$D2,$T1		@ h1 -> h2
+	vbic.i32	$D4#lo,#0xfc000000
+	 vbic.i32	$D1#lo,#0xfc000000
+
+	vadd.i32	$D0#lo,$D0#lo,$T0#lo
+	vshl.u32	$T0#lo,$T0#lo,#2
+	 vshrn.u64	$T1#lo,$D2,#26
+	 vmovn.i64	$D2#lo,$D2
+	vadd.i32	$D0#lo,$D0#lo,$T0#lo	@ h4 -> h0
+	 vadd.i32	$D3#lo,$D3#lo,$T1#lo	@ h2 -> h3
+	 vbic.i32	$D2#lo,#0xfc000000
+
+	vshr.u32	$T0#lo,$D0#lo,#26
+	vbic.i32	$D0#lo,#0xfc000000
+	 vshr.u32	$T1#lo,$D3#lo,#26
+	 vbic.i32	$D3#lo,#0xfc000000
+	vadd.i32	$D1#lo,$D1#lo,$T0#lo	@ h0 -> h1
+	 vadd.i32	$D4#lo,$D4#lo,$T1#lo	@ h3 -> h4
+
+	subs		$zeros,$zeros,#1
+	beq		.Lsquare_break_neon
+
+	add		$tbl0,$ctx,#(48+0*9*4)
+	add		$tbl1,$ctx,#(48+1*9*4)
+
+	vtrn.32		$R0,$D0#lo		@ r^2:r^1
+	vtrn.32		$R2,$D2#lo
+	vtrn.32		$R3,$D3#lo
+	vtrn.32		$R1,$D1#lo
+	vtrn.32		$R4,$D4#lo
+
+	vshl.u32	$S2,$R2,#2		@ *5
+	vshl.u32	$S3,$R3,#2
+	vshl.u32	$S1,$R1,#2
+	vshl.u32	$S4,$R4,#2
+	vadd.i32	$S2,$S2,$R2
+	vadd.i32	$S1,$S1,$R1
+	vadd.i32	$S3,$S3,$R3
+	vadd.i32	$S4,$S4,$R4
+
+	vst4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!
+	vst4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!
+	vst4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
+	vst4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
+	vst1.32		{${S4}[0]},[$tbl0,:32]
+	vst1.32		{${S4}[1]},[$tbl1,:32]
+
+	b		.Lsquare_neon
+
+.align	4
+.Lsquare_break_neon:
+	add		$tbl0,$ctx,#(48+2*4*9)
+	add		$tbl1,$ctx,#(48+3*4*9)
+
+	vmov		$R0,$D0#lo		@ r^4:r^3
+	vshl.u32	$S1,$D1#lo,#2		@ *5
+	vmov		$R1,$D1#lo
+	vshl.u32	$S2,$D2#lo,#2
+	vmov		$R2,$D2#lo
+	vshl.u32	$S3,$D3#lo,#2
+	vmov		$R3,$D3#lo
+	vshl.u32	$S4,$D4#lo,#2
+	vmov		$R4,$D4#lo
+	vadd.i32	$S1,$S1,$D1#lo
+	vadd.i32	$S2,$S2,$D2#lo
+	vadd.i32	$S3,$S3,$D3#lo
+	vadd.i32	$S4,$S4,$D4#lo
+
+	vst4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!
+	vst4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!
+	vst4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
+	vst4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
+	vst1.32		{${S4}[0]},[$tbl0]
+	vst1.32		{${S4}[1]},[$tbl1]
+
+	ret				@ bx	lr
+.size	poly1305_init_neon,.-poly1305_init_neon
+
+.type	poly1305_blocks_neon,%function
+.align	5
+poly1305_blocks_neon:
+	ldr	ip,[$ctx,#36]		@ is_base2_26
+	ands	$len,$len,#-16
+	beq	.Lno_data_neon
+
+	cmp	$len,#64
+	bhs	.Lenter_neon
+	tst	ip,ip			@ is_base2_26?
+	beq	poly1305_blocks
+
+.Lenter_neon:
+	stmdb	sp!,{r4-r7}
+	vstmdb	sp!,{d8-d15}		@ ABI specification says so
+
+	tst	ip,ip			@ is_base2_26?
+	bne	.Lbase2_26_neon
+
+	stmdb	sp!,{r1-r3,lr}
+	bl	poly1305_init_neon
+
+	ldr	r4,[$ctx,#0]		@ load hash value base 2^32
+	ldr	r5,[$ctx,#4]
+	ldr	r6,[$ctx,#8]
+	ldr	r7,[$ctx,#12]
+	ldr	ip,[$ctx,#16]
+
+	and	r2,r4,#0x03ffffff	@ base 2^32 -> base 2^26
+	mov	r3,r4,lsr#26
+	 veor	$D0#lo,$D0#lo,$D0#lo
+	mov	r4,r5,lsr#20
+	orr	r3,r3,r5,lsl#6
+	 veor	$D1#lo,$D1#lo,$D1#lo
+	mov	r5,r6,lsr#14
+	orr	r4,r4,r6,lsl#12
+	 veor	$D2#lo,$D2#lo,$D2#lo
+	mov	r6,r7,lsr#8
+	orr	r5,r5,r7,lsl#18
+	 veor	$D3#lo,$D3#lo,$D3#lo
+	and	r3,r3,#0x03ffffff
+	orr	r6,r6,ip,lsl#24
+	 veor	$D4#lo,$D4#lo,$D4#lo
+	and	r4,r4,#0x03ffffff
+	mov	r1,#1
+	and	r5,r5,#0x03ffffff
+	str	r1,[$ctx,#36]		@ is_base2_26
+
+	vmov.32	$D0#lo[0],r2
+	vmov.32	$D1#lo[0],r3
+	vmov.32	$D2#lo[0],r4
+	vmov.32	$D3#lo[0],r5
+	vmov.32	$D4#lo[0],r6
+	adr	$zeros,.Lzeros
+
+	ldmia	sp!,{r1-r3,lr}
+	b	.Lbase2_32_neon
+
+.align	4
+.Lbase2_26_neon:
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ load hash value
+
+	veor		$D0#lo,$D0#lo,$D0#lo
+	veor		$D1#lo,$D1#lo,$D1#lo
+	veor		$D2#lo,$D2#lo,$D2#lo
+	veor		$D3#lo,$D3#lo,$D3#lo
+	veor		$D4#lo,$D4#lo,$D4#lo
+	vld4.32		{$D0#lo[0],$D1#lo[0],$D2#lo[0],$D3#lo[0]},[$ctx]!
+	adr		$zeros,.Lzeros
+	vld1.32		{$D4#lo[0]},[$ctx]
+	sub		$ctx,$ctx,#16		@ rewind
+
+.Lbase2_32_neon:
+	add		$in2,$inp,#32
+	mov		$padbit,$padbit,lsl#24
+	tst		$len,#31
+	beq		.Leven
+
+	vld4.32		{$H0#lo[0],$H1#lo[0],$H2#lo[0],$H3#lo[0]},[$inp]!
+	vmov.32		$H4#lo[0],$padbit
+	sub		$len,$len,#16
+	add		$in2,$inp,#32
+
+# ifdef	__ARMEB__
+	vrev32.8	$H0,$H0
+	vrev32.8	$H3,$H3
+	vrev32.8	$H1,$H1
+	vrev32.8	$H2,$H2
+# endif
+	vsri.u32	$H4#lo,$H3#lo,#8	@ base 2^32 -> base 2^26
+	vshl.u32	$H3#lo,$H3#lo,#18
+
+	vsri.u32	$H3#lo,$H2#lo,#14
+	vshl.u32	$H2#lo,$H2#lo,#12
+	vadd.i32	$H4#hi,$H4#lo,$D4#lo	@ add hash value and move to #hi
+
+	vbic.i32	$H3#lo,#0xfc000000
+	vsri.u32	$H2#lo,$H1#lo,#20
+	vshl.u32	$H1#lo,$H1#lo,#6
+
+	vbic.i32	$H2#lo,#0xfc000000
+	vsri.u32	$H1#lo,$H0#lo,#26
+	vadd.i32	$H3#hi,$H3#lo,$D3#lo
+
+	vbic.i32	$H0#lo,#0xfc000000
+	vbic.i32	$H1#lo,#0xfc000000
+	vadd.i32	$H2#hi,$H2#lo,$D2#lo
+
+	vadd.i32	$H0#hi,$H0#lo,$D0#lo
+	vadd.i32	$H1#hi,$H1#lo,$D1#lo
+
+	mov		$tbl1,$zeros
+	add		$tbl0,$ctx,#48
+
+	cmp		$len,$len
+	b		.Long_tail
+
+.align	4
+.Leven:
+	subs		$len,$len,#64
+# ifdef	__thumb2__
+	it		lo
+# endif
+	movlo		$in2,$zeros
+
+	vmov.i32	$H4,#1<<24		@ padbit, yes, always
+	vld4.32		{$H0#lo,$H1#lo,$H2#lo,$H3#lo},[$inp]	@ inp[0:1]
+	add		$inp,$inp,#64
+	vld4.32		{$H0#hi,$H1#hi,$H2#hi,$H3#hi},[$in2]	@ inp[2:3] (or 0)
+	add		$in2,$in2,#64
+# ifdef	__thumb2__
+	itt		hi
+# endif
+	addhi		$tbl1,$ctx,#(48+1*9*4)
+	addhi		$tbl0,$ctx,#(48+3*9*4)
+
+# ifdef	__ARMEB__
+	vrev32.8	$H0,$H0
+	vrev32.8	$H3,$H3
+	vrev32.8	$H1,$H1
+	vrev32.8	$H2,$H2
+# endif
+	vsri.u32	$H4,$H3,#8		@ base 2^32 -> base 2^26
+	vshl.u32	$H3,$H3,#18
+
+	vsri.u32	$H3,$H2,#14
+	vshl.u32	$H2,$H2,#12
+
+	vbic.i32	$H3,#0xfc000000
+	vsri.u32	$H2,$H1,#20
+	vshl.u32	$H1,$H1,#6
+
+	vbic.i32	$H2,#0xfc000000
+	vsri.u32	$H1,$H0,#26
+
+	vbic.i32	$H0,#0xfc000000
+	vbic.i32	$H1,#0xfc000000
+
+	bls		.Lskip_loop
+
+	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^2
+	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^4
+	vld4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
+	vld4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
+	b		.Loop_neon
+
+.align	5
+.Loop_neon:
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
+	@ ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
+	@   \___________________/
+	@ ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
+	@ ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
+	@   \___________________/ \____________________/
+	@
+	@ Note that we start with inp[2:3]*r^2. This is because it
+	@ doesn't depend on reduction in previous iteration.
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	@ d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	@ d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	@ d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	@ d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ inp[2:3]*r^2
+
+	vadd.i32	$H2#lo,$H2#lo,$D2#lo	@ accumulate inp[0:1]
+	vmull.u32	$D2,$H2#hi,${R0}[1]
+	vadd.i32	$H0#lo,$H0#lo,$D0#lo
+	vmull.u32	$D0,$H0#hi,${R0}[1]
+	vadd.i32	$H3#lo,$H3#lo,$D3#lo
+	vmull.u32	$D3,$H3#hi,${R0}[1]
+	vmlal.u32	$D2,$H1#hi,${R1}[1]
+	vadd.i32	$H1#lo,$H1#lo,$D1#lo
+	vmull.u32	$D1,$H1#hi,${R0}[1]
+
+	vadd.i32	$H4#lo,$H4#lo,$D4#lo
+	vmull.u32	$D4,$H4#hi,${R0}[1]
+	subs		$len,$len,#64
+	vmlal.u32	$D0,$H4#hi,${S1}[1]
+# ifdef	__thumb2__
+	it		lo
+# endif
+	movlo		$in2,$zeros
+	vmlal.u32	$D3,$H2#hi,${R1}[1]
+	vld1.32		${S4}[1],[$tbl1,:32]
+	vmlal.u32	$D1,$H0#hi,${R1}[1]
+	vmlal.u32	$D4,$H3#hi,${R1}[1]
+
+	vmlal.u32	$D0,$H3#hi,${S2}[1]
+	vmlal.u32	$D3,$H1#hi,${R2}[1]
+	vmlal.u32	$D4,$H2#hi,${R2}[1]
+	vmlal.u32	$D1,$H4#hi,${S2}[1]
+	vmlal.u32	$D2,$H0#hi,${R2}[1]
+
+	vmlal.u32	$D3,$H0#hi,${R3}[1]
+	vmlal.u32	$D0,$H2#hi,${S3}[1]
+	vmlal.u32	$D4,$H1#hi,${R3}[1]
+	vmlal.u32	$D1,$H3#hi,${S3}[1]
+	vmlal.u32	$D2,$H4#hi,${S3}[1]
+
+	vmlal.u32	$D3,$H4#hi,${S4}[1]
+	vmlal.u32	$D0,$H1#hi,${S4}[1]
+	vmlal.u32	$D4,$H0#hi,${R4}[1]
+	vmlal.u32	$D1,$H2#hi,${S4}[1]
+	vmlal.u32	$D2,$H3#hi,${S4}[1]
+
+	vld4.32		{$H0#hi,$H1#hi,$H2#hi,$H3#hi},[$in2]	@ inp[2:3] (or 0)
+	add		$in2,$in2,#64
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ (hash+inp[0:1])*r^4 and accumulate
+
+	vmlal.u32	$D3,$H3#lo,${R0}[0]
+	vmlal.u32	$D0,$H0#lo,${R0}[0]
+	vmlal.u32	$D4,$H4#lo,${R0}[0]
+	vmlal.u32	$D1,$H1#lo,${R0}[0]
+	vmlal.u32	$D2,$H2#lo,${R0}[0]
+	vld1.32		${S4}[0],[$tbl0,:32]
+
+	vmlal.u32	$D3,$H2#lo,${R1}[0]
+	vmlal.u32	$D0,$H4#lo,${S1}[0]
+	vmlal.u32	$D4,$H3#lo,${R1}[0]
+	vmlal.u32	$D1,$H0#lo,${R1}[0]
+	vmlal.u32	$D2,$H1#lo,${R1}[0]
+
+	vmlal.u32	$D3,$H1#lo,${R2}[0]
+	vmlal.u32	$D0,$H3#lo,${S2}[0]
+	vmlal.u32	$D4,$H2#lo,${R2}[0]
+	vmlal.u32	$D1,$H4#lo,${S2}[0]
+	vmlal.u32	$D2,$H0#lo,${R2}[0]
+
+	vmlal.u32	$D3,$H0#lo,${R3}[0]
+	vmlal.u32	$D0,$H2#lo,${S3}[0]
+	vmlal.u32	$D4,$H1#lo,${R3}[0]
+	vmlal.u32	$D1,$H3#lo,${S3}[0]
+	vmlal.u32	$D3,$H4#lo,${S4}[0]
+
+	vmlal.u32	$D2,$H4#lo,${S3}[0]
+	vmlal.u32	$D0,$H1#lo,${S4}[0]
+	vmlal.u32	$D4,$H0#lo,${R4}[0]
+	vmov.i32	$H4,#1<<24		@ padbit, yes, always
+	vmlal.u32	$D1,$H2#lo,${S4}[0]
+	vmlal.u32	$D2,$H3#lo,${S4}[0]
+
+	vld4.32		{$H0#lo,$H1#lo,$H2#lo,$H3#lo},[$inp]	@ inp[0:1]
+	add		$inp,$inp,#64
+# ifdef	__ARMEB__
+	vrev32.8	$H0,$H0
+	vrev32.8	$H1,$H1
+	vrev32.8	$H2,$H2
+	vrev32.8	$H3,$H3
+# endif
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ lazy reduction interleaved with base 2^32 -> base 2^26
+
+	vshr.u64	$T0,$D3,#26
+	vmovn.i64	$D3#lo,$D3
+	 vshr.u64	$T1,$D0,#26
+	 vmovn.i64	$D0#lo,$D0
+	vadd.i64	$D4,$D4,$T0		@ h3 -> h4
+	vbic.i32	$D3#lo,#0xfc000000
+	  vsri.u32	$H4,$H3,#8		@ base 2^32 -> base 2^26
+	 vadd.i64	$D1,$D1,$T1		@ h0 -> h1
+	  vshl.u32	$H3,$H3,#18
+	 vbic.i32	$D0#lo,#0xfc000000
+
+	vshrn.u64	$T0#lo,$D4,#26
+	vmovn.i64	$D4#lo,$D4
+	 vshr.u64	$T1,$D1,#26
+	 vmovn.i64	$D1#lo,$D1
+	 vadd.i64	$D2,$D2,$T1		@ h1 -> h2
+	  vsri.u32	$H3,$H2,#14
+	vbic.i32	$D4#lo,#0xfc000000
+	  vshl.u32	$H2,$H2,#12
+	 vbic.i32	$D1#lo,#0xfc000000
+
+	vadd.i32	$D0#lo,$D0#lo,$T0#lo
+	vshl.u32	$T0#lo,$T0#lo,#2
+	  vbic.i32	$H3,#0xfc000000
+	 vshrn.u64	$T1#lo,$D2,#26
+	 vmovn.i64	$D2#lo,$D2
+	vadd.i32	$D0#lo,$D0#lo,$T0#lo	@ h4 -> h0
+	  vsri.u32	$H2,$H1,#20
+	 vadd.i32	$D3#lo,$D3#lo,$T1#lo	@ h2 -> h3
+	  vshl.u32	$H1,$H1,#6
+	 vbic.i32	$D2#lo,#0xfc000000
+	  vbic.i32	$H2,#0xfc000000
+
+	vshr.u32	$T0#lo,$D0#lo,#26
+	vbic.i32	$D0#lo,#0xfc000000
+	  vsri.u32	$H1,$H0,#26
+	  vbic.i32	$H0,#0xfc000000
+	 vshr.u32	$T1#lo,$D3#lo,#26
+	 vbic.i32	$D3#lo,#0xfc000000
+	vadd.i32	$D1#lo,$D1#lo,$T0#lo	@ h0 -> h1
+	 vadd.i32	$D4#lo,$D4#lo,$T1#lo	@ h3 -> h4
+	  vbic.i32	$H1,#0xfc000000
+
+	bhi		.Loop_neon
+
+.Lskip_loop:
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
+
+	add		$tbl1,$ctx,#(48+0*9*4)
+	add		$tbl0,$ctx,#(48+1*9*4)
+	adds		$len,$len,#32
+# ifdef	__thumb2__
+	it		ne
+# endif
+	movne		$len,#0
+	bne		.Long_tail
+
+	vadd.i32	$H2#hi,$H2#lo,$D2#lo	@ add hash value and move to #hi
+	vadd.i32	$H0#hi,$H0#lo,$D0#lo
+	vadd.i32	$H3#hi,$H3#lo,$D3#lo
+	vadd.i32	$H1#hi,$H1#lo,$D1#lo
+	vadd.i32	$H4#hi,$H4#lo,$D4#lo
+
+.Long_tail:
+	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^1
+	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^2
+
+	vadd.i32	$H2#lo,$H2#lo,$D2#lo	@ can be redundant
+	vmull.u32	$D2,$H2#hi,$R0
+	vadd.i32	$H0#lo,$H0#lo,$D0#lo
+	vmull.u32	$D0,$H0#hi,$R0
+	vadd.i32	$H3#lo,$H3#lo,$D3#lo
+	vmull.u32	$D3,$H3#hi,$R0
+	vadd.i32	$H1#lo,$H1#lo,$D1#lo
+	vmull.u32	$D1,$H1#hi,$R0
+	vadd.i32	$H4#lo,$H4#lo,$D4#lo
+	vmull.u32	$D4,$H4#hi,$R0
+
+	vmlal.u32	$D0,$H4#hi,$S1
+	vld4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
+	vmlal.u32	$D3,$H2#hi,$R1
+	vld4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
+	vmlal.u32	$D1,$H0#hi,$R1
+	vmlal.u32	$D4,$H3#hi,$R1
+	vmlal.u32	$D2,$H1#hi,$R1
+
+	vmlal.u32	$D3,$H1#hi,$R2
+	vld1.32		${S4}[1],[$tbl1,:32]
+	vmlal.u32	$D0,$H3#hi,$S2
+	vld1.32		${S4}[0],[$tbl0,:32]
+	vmlal.u32	$D4,$H2#hi,$R2
+	vmlal.u32	$D1,$H4#hi,$S2
+	vmlal.u32	$D2,$H0#hi,$R2
+
+	vmlal.u32	$D3,$H0#hi,$R3
+# ifdef	__thumb2__
+	it		ne
+# endif
+	 addne		$tbl1,$ctx,#(48+2*9*4)
+	vmlal.u32	$D0,$H2#hi,$S3
+# ifdef	__thumb2__
+	it		ne
+# endif
+	 addne		$tbl0,$ctx,#(48+3*9*4)
+	vmlal.u32	$D4,$H1#hi,$R3
+	vmlal.u32	$D1,$H3#hi,$S3
+	vmlal.u32	$D2,$H4#hi,$S3
+
+	vmlal.u32	$D3,$H4#hi,$S4
+	 vorn		$MASK,$MASK,$MASK	@ all-ones, can be redundant
+	vmlal.u32	$D0,$H1#hi,$S4
+	 vshr.u64	$MASK,$MASK,#38
+	vmlal.u32	$D4,$H0#hi,$R4
+	vmlal.u32	$D1,$H2#hi,$S4
+	vmlal.u32	$D2,$H3#hi,$S4
+
+	beq		.Lshort_tail
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ (hash+inp[0:1])*r^4:r^3 and accumulate
+
+	vld4.32		{${R0}[1],${R1}[1],${S1}[1],${R2}[1]},[$tbl1]!	@ load r^3
+	vld4.32		{${R0}[0],${R1}[0],${S1}[0],${R2}[0]},[$tbl0]!	@ load r^4
+
+	vmlal.u32	$D2,$H2#lo,$R0
+	vmlal.u32	$D0,$H0#lo,$R0
+	vmlal.u32	$D3,$H3#lo,$R0
+	vmlal.u32	$D1,$H1#lo,$R0
+	vmlal.u32	$D4,$H4#lo,$R0
+
+	vmlal.u32	$D0,$H4#lo,$S1
+	vld4.32		{${S2}[1],${R3}[1],${S3}[1],${R4}[1]},[$tbl1]!
+	vmlal.u32	$D3,$H2#lo,$R1
+	vld4.32		{${S2}[0],${R3}[0],${S3}[0],${R4}[0]},[$tbl0]!
+	vmlal.u32	$D1,$H0#lo,$R1
+	vmlal.u32	$D4,$H3#lo,$R1
+	vmlal.u32	$D2,$H1#lo,$R1
+
+	vmlal.u32	$D3,$H1#lo,$R2
+	vld1.32		${S4}[1],[$tbl1,:32]
+	vmlal.u32	$D0,$H3#lo,$S2
+	vld1.32		${S4}[0],[$tbl0,:32]
+	vmlal.u32	$D4,$H2#lo,$R2
+	vmlal.u32	$D1,$H4#lo,$S2
+	vmlal.u32	$D2,$H0#lo,$R2
+
+	vmlal.u32	$D3,$H0#lo,$R3
+	vmlal.u32	$D0,$H2#lo,$S3
+	vmlal.u32	$D4,$H1#lo,$R3
+	vmlal.u32	$D1,$H3#lo,$S3
+	vmlal.u32	$D2,$H4#lo,$S3
+
+	vmlal.u32	$D3,$H4#lo,$S4
+	 vorn		$MASK,$MASK,$MASK	@ all-ones
+	vmlal.u32	$D0,$H1#lo,$S4
+	 vshr.u64	$MASK,$MASK,#38
+	vmlal.u32	$D4,$H0#lo,$R4
+	vmlal.u32	$D1,$H2#lo,$S4
+	vmlal.u32	$D2,$H3#lo,$S4
+
+.Lshort_tail:
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ lazy reduction, but without narrowing
+
+	vshr.u64	$T0,$D3,#26
+	vand.i64	$D3,$D3,$MASK
+	 vshr.u64	$T1,$D0,#26
+	 vand.i64	$D0,$D0,$MASK
+	vadd.i64	$D4,$D4,$T0		@ h3 -> h4
+	 vadd.i64	$D1,$D1,$T1		@ h0 -> h1
+
+	vshr.u64	$T0,$D4,#26
+	vand.i64	$D4,$D4,$MASK
+	 vshr.u64	$T1,$D1,#26
+	 vand.i64	$D1,$D1,$MASK
+	 vadd.i64	$D2,$D2,$T1		@ h1 -> h2
+
+	vadd.i64	$D0,$D0,$T0
+	vshl.u64	$T0,$T0,#2
+	 vshr.u64	$T1,$D2,#26
+	 vand.i64	$D2,$D2,$MASK
+	vadd.i64	$D0,$D0,$T0		@ h4 -> h0
+	 vadd.i64	$D3,$D3,$T1		@ h2 -> h3
+
+	vshr.u64	$T0,$D0,#26
+	vand.i64	$D0,$D0,$MASK
+	 vshr.u64	$T1,$D3,#26
+	 vand.i64	$D3,$D3,$MASK
+	vadd.i64	$D1,$D1,$T0		@ h0 -> h1
+	 vadd.i64	$D4,$D4,$T1		@ h3 -> h4
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ horizontal addition
+
+	vadd.i64	$D2#lo,$D2#lo,$D2#hi
+	vadd.i64	$D0#lo,$D0#lo,$D0#hi
+	vadd.i64	$D3#lo,$D3#lo,$D3#hi
+	vadd.i64	$D1#lo,$D1#lo,$D1#hi
+	vadd.i64	$D4#lo,$D4#lo,$D4#hi
+
+	cmp		$len,#0
+	bne		.Leven
+
+	@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
+	@ store hash value
+
+	vst4.32		{$D0#lo[0],$D1#lo[0],$D2#lo[0],$D3#lo[0]},[$ctx]!
+	vst1.32		{$D4#lo[0]},[$ctx]
+
+	vldmia	sp!,{d8-d15}			@ epilogue
+	ldmia	sp!,{r4-r7}
+.Lno_data_neon:
+	ret					@ bx	lr
+.size	poly1305_blocks_neon,.-poly1305_blocks_neon
+
+.type	poly1305_emit_neon,%function
+.align	5
+poly1305_emit_neon:
+	ldr	ip,[$ctx,#36]		@ is_base2_26
+
+	stmdb	sp!,{r4-r11}
+
+	tst	ip,ip
+	beq	.Lpoly1305_emit_enter
+
+	ldmia	$ctx,{$h0-$h4}
+	eor	$g0,$g0,$g0
+
+	adds	$h0,$h0,$h1,lsl#26	@ base 2^26 -> base 2^32
+	mov	$h1,$h1,lsr#6
+	adcs	$h1,$h1,$h2,lsl#20
+	mov	$h2,$h2,lsr#12
+	adcs	$h2,$h2,$h3,lsl#14
+	mov	$h3,$h3,lsr#18
+	adcs	$h3,$h3,$h4,lsl#8
+	adc	$h4,$g0,$h4,lsr#24	@ can be partially reduced ...
+
+	and	$g0,$h4,#-4		@ ... so reduce
+	and	$h4,$h3,#3
+	add	$g0,$g0,$g0,lsr#2	@ *= 5
+	adds	$h0,$h0,$g0
+	adcs	$h1,$h1,#0
+	adcs	$h2,$h2,#0
+	adc	$h3,$h3,#0
+
+	adds	$g0,$h0,#5		@ compare to modulus
+	adcs	$g1,$h1,#0
+	adcs	$g2,$h2,#0
+	adcs	$g3,$h3,#0
+	adc	$g4,$h4,#0
+	tst	$g4,#4			@ did it carry/borrow?
+
+# ifdef	__thumb2__
+	it	ne
+# endif
+	movne	$h0,$g0
+	ldr	$g0,[$nonce,#0]
+# ifdef	__thumb2__
+	it	ne
+# endif
+	movne	$h1,$g1
+	ldr	$g1,[$nonce,#4]
+# ifdef	__thumb2__
+	it	ne
+# endif
+	movne	$h2,$g2
+	ldr	$g2,[$nonce,#8]
+# ifdef	__thumb2__
+	it	ne
+# endif
+	movne	$h3,$g3
+	ldr	$g3,[$nonce,#12]
+
+	adds	$h0,$h0,$g0		@ accumulate nonce
+	adcs	$h1,$h1,$g1
+	adcs	$h2,$h2,$g2
+	adc	$h3,$h3,$g3
+
+# ifdef __ARMEB__
+	rev	$h0,$h0
+	rev	$h1,$h1
+	rev	$h2,$h2
+	rev	$h3,$h3
+# endif
+	str	$h0,[$mac,#0]		@ store the result
+	str	$h1,[$mac,#4]
+	str	$h2,[$mac,#8]
+	str	$h3,[$mac,#12]
+
+	ldmia	sp!,{r4-r11}
+	ret				@ bx	lr
+.size	poly1305_emit_neon,.-poly1305_emit_neon
+
+.align	5
+.Lzeros:
+.long	0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
+.LOPENSSL_armcap:
+.word	OPENSSL_armcap_P-.Lpoly1305_init
+#endif
+___
+}	}
+$code.=<<___;
+.asciz	"Poly1305 for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
+.align	2
+#if	__ARM_MAX_ARCH__>=7
+.comm   OPENSSL_armcap_P,4,4
+#endif
+___
+
+foreach (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo	or
+	s/\bret\b/bx	lr/go						or
+	s/\bbx\s+lr\b/.word\t0xe12fff1e/go;	# make it possible to compile with -march=armv4
+
+	print $_,"\n";
+}
+close STDOUT; # enforce flush
diff --git a/crypto/poly1305/asm/poly1305-armv8.pl b/crypto/poly1305/asm/poly1305-armv8.pl
new file mode 100755
index 00000000..79185d2b
--- /dev/null
+++ b/crypto/poly1305/asm/poly1305-armv8.pl
@@ -0,0 +1,925 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements Poly1305 hash for ARMv8.
+#
+# June 2015
+#
+# Numbers are cycles per processed byte with poly1305_blocks alone.
+#
+#		IALU/gcc-4.9	NEON
+#
+# Apple A7	1.86/+5%	0.72
+# Cortex-A53	2.63/+58%	1.47
+# Cortex-A57	2.70/+7%	1.14
+# Denver	1.39/+50%	1.18(*)
+# X-Gene	2.00/+68%	2.19
+#
+# (*)	estimate based on resources availability is less than 1.0,
+#	i.e. measured result is worse than expected, presumably binary
+#	translator is not almighty;
+
+$flavour=shift;
+$output=shift;
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
+die "can't locate arm-xlate.pl";
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+my ($ctx,$inp,$len,$padbit) = map("x$_",(0..3));
+my ($mac,$nonce)=($inp,$len);
+
+my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
+
+$code.=<<___;
+#include "arm_arch.h"
+
+.text
+
+// forward "declarations" are required for Apple
+.extern	OPENSSL_armcap_P
+.globl	poly1305_blocks
+.globl	poly1305_emit
+
+.globl	poly1305_init
+.type	poly1305_init,%function
+.align	5
+poly1305_init:
+	cmp	$inp,xzr
+	stp	xzr,xzr,[$ctx]		// zero hash value
+	stp	xzr,xzr,[$ctx,#16]	// [along with is_base2_26]
+
+	csel	x0,xzr,x0,eq
+	b.eq	.Lno_key
+
+#ifdef	__ILP32__
+	ldrsw	$t1,.LOPENSSL_armcap_P
+#else
+	ldr	$t1,.LOPENSSL_armcap_P
+#endif
+	adr	$t0,.LOPENSSL_armcap_P
+
+	ldp	$r0,$r1,[$inp]		// load key
+	mov	$s1,#0xfffffffc0fffffff
+	movk	$s1,#0x0fff,lsl#48
+	ldr	w17,[$t0,$t1]
+#ifdef	__ARMEB__
+	rev	$r0,$r0			// flip bytes
+	rev	$r1,$r1
+#endif
+	and	$r0,$r0,$s1		// &=0ffffffc0fffffff
+	and	$s1,$s1,#-4
+	and	$r1,$r1,$s1		// &=0ffffffc0ffffffc
+	stp	$r0,$r1,[$ctx,#32]	// save key value
+
+	tst	w17,#ARMV7_NEON
+
+	adr	$d0,poly1305_blocks
+	adr	$r0,poly1305_blocks_neon
+	adr	$d1,poly1305_emit
+	adr	$r1,poly1305_emit_neon
+
+	csel	$d0,$d0,$r0,eq
+	csel	$d1,$d1,$r1,eq
+
+	stp	$d0,$d1,[$len]
+
+	mov	x0,#1
+.Lno_key:
+	ret
+.size	poly1305_init,.-poly1305_init
+
+.type	poly1305_blocks,%function
+.align	5
+poly1305_blocks:
+	ands	$len,$len,#-16
+	b.eq	.Lno_data
+
+	ldp	$h0,$h1,[$ctx]		// load hash value
+	ldp	$r0,$r1,[$ctx,#32]	// load key value
+	ldr	$h2,[$ctx,#16]
+	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
+	b	.Loop
+
+.align	5
+.Loop:
+	ldp	$t0,$t1,[$inp],#16	// load input
+	sub	$len,$len,#16
+#ifdef	__ARMEB__
+	rev	$t0,$t0
+	rev	$t1,$t1
+#endif
+	adds	$h0,$h0,$t0		// accumulate input
+	adcs	$h1,$h1,$t1
+
+	mul	$d0,$h0,$r0		// h0*r0
+	adc	$h2,$h2,$padbit
+	umulh	$d1,$h0,$r0
+
+	mul	$t0,$h1,$s1		// h1*5*r1
+	umulh	$t1,$h1,$s1
+
+	adds	$d0,$d0,$t0
+	mul	$t0,$h0,$r1		// h0*r1
+	adc	$d1,$d1,$t1
+	umulh	$d2,$h0,$r1
+
+	adds	$d1,$d1,$t0
+	mul	$t0,$h1,$r0		// h1*r0
+	adc	$d2,$d2,xzr
+	umulh	$t1,$h1,$r0
+
+	adds	$d1,$d1,$t0
+	mul	$t0,$h2,$s1		// h2*5*r1
+	adc	$d2,$d2,$t1
+	mul	$t1,$h2,$r0		// h2*r0
+
+	adds	$d1,$d1,$t0
+	adc	$d2,$d2,$t1
+
+	and	$t0,$d2,#-4		// final reduction
+	and	$h2,$d2,#3
+	add	$t0,$t0,$d2,lsr#2
+	adds	$h0,$d0,$t0
+	adc	$h1,$d1,xzr
+
+	cbnz	$len,.Loop
+
+	stp	$h0,$h1,[$ctx]		// store hash value
+	str	$h2,[$ctx,#16]
+
+.Lno_data:
+	ret
+.size	poly1305_blocks,.-poly1305_blocks
+
+.type	poly1305_emit,%function
+.align	5
+poly1305_emit:
+	ldp	$h0,$h1,[$ctx]		// load hash base 2^64
+	ldr	$h2,[$ctx,#16]
+	ldp	$t0,$t1,[$nonce]	// load nonce
+
+	adds	$d0,$h0,#5		// compare to modulus
+	adcs	$d1,$h1,xzr
+	adc	$d2,$h2,xzr
+
+	tst	$d2,#-4			// see if it's carried/borrowed
+
+	csel	$h0,$h0,$d0,eq
+	csel	$h1,$h1,$d1,eq
+
+#ifdef	__ARMEB__
+	ror	$t0,$t0,#32		// flip nonce words
+	ror	$t1,$t1,#32
+#endif
+	adds	$h0,$h0,$t0		// accumulate nonce
+	adc	$h1,$h1,$t1
+#ifdef	__ARMEB__
+	rev	$h0,$h0			// flip output bytes
+	rev	$h1,$h1
+#endif
+	stp	$h0,$h1,[$mac]		// write result
+
+	ret
+.size	poly1305_emit,.-poly1305_emit
+___
+my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
+my ($IN01_0,$IN01_1,$IN01_2,$IN01_3,$IN01_4) = map("v$_.2s",(9..13));
+my ($IN23_0,$IN23_1,$IN23_2,$IN23_3,$IN23_4) = map("v$_.2s",(14..18));
+my ($ACC0,$ACC1,$ACC2,$ACC3,$ACC4) = map("v$_.2d",(19..23));
+my ($H0,$H1,$H2,$H3,$H4) = map("v$_.2s",(24..28));
+my ($T0,$T1,$MASK) = map("v$_",(29..31));
+
+my ($in2,$zeros)=("x16","x17");
+my $is_base2_26 = $zeros;		# borrow
+
+$code.=<<___;
+.type	poly1305_mult,%function
+.align	5
+poly1305_mult:
+	mul	$d0,$h0,$r0		// h0*r0
+	umulh	$d1,$h0,$r0
+
+	mul	$t0,$h1,$s1		// h1*5*r1
+	umulh	$t1,$h1,$s1
+
+	adds	$d0,$d0,$t0
+	mul	$t0,$h0,$r1		// h0*r1
+	adc	$d1,$d1,$t1
+	umulh	$d2,$h0,$r1
+
+	adds	$d1,$d1,$t0
+	mul	$t0,$h1,$r0		// h1*r0
+	adc	$d2,$d2,xzr
+	umulh	$t1,$h1,$r0
+
+	adds	$d1,$d1,$t0
+	mul	$t0,$h2,$s1		// h2*5*r1
+	adc	$d2,$d2,$t1
+	mul	$t1,$h2,$r0		// h2*r0
+
+	adds	$d1,$d1,$t0
+	adc	$d2,$d2,$t1
+
+	and	$t0,$d2,#-4		// final reduction
+	and	$h2,$d2,#3
+	add	$t0,$t0,$d2,lsr#2
+	adds	$h0,$d0,$t0
+	adc	$h1,$d1,xzr
+
+	ret
+.size	poly1305_mult,.-poly1305_mult
+
+.type	poly1305_splat,%function
+.align	5
+poly1305_splat:
+	and	x12,$h0,#0x03ffffff	// base 2^64 -> base 2^26
+	ubfx	x13,$h0,#26,#26
+	extr	x14,$h1,$h0,#52
+	and	x14,x14,#0x03ffffff
+	ubfx	x15,$h1,#14,#26
+	extr	x16,$h2,$h1,#40
+
+	str	w12,[$ctx,#16*0]	// r0
+	add	w12,w13,w13,lsl#2	// r1*5
+	str	w13,[$ctx,#16*1]	// r1
+	add	w13,w14,w14,lsl#2	// r2*5
+	str	w12,[$ctx,#16*2]	// s1
+	str	w14,[$ctx,#16*3]	// r2
+	add	w14,w15,w15,lsl#2	// r3*5
+	str	w13,[$ctx,#16*4]	// s2
+	str	w15,[$ctx,#16*5]	// r3
+	add	w15,w16,w16,lsl#2	// r4*5
+	str	w14,[$ctx,#16*6]	// s3
+	str	w16,[$ctx,#16*7]	// r4
+	str	w15,[$ctx,#16*8]	// s4
+
+	ret
+.size	poly1305_splat,.-poly1305_splat
+
+.type	poly1305_blocks_neon,%function
+.align	5
+poly1305_blocks_neon:
+	ldr	$is_base2_26,[$ctx,#24]
+	cmp	$len,#128
+	b.hs	.Lblocks_neon
+	cbz	$is_base2_26,poly1305_blocks
+
+.Lblocks_neon:
+	stp	x29,x30,[sp,#-80]!
+	add	x29,sp,#0
+
+	ands	$len,$len,#-16
+	b.eq	.Lno_data_neon
+
+	cbz	$is_base2_26,.Lbase2_64_neon
+
+	ldp	w10,w11,[$ctx]		// load hash value base 2^26
+	ldp	w12,w13,[$ctx,#8]
+	ldr	w14,[$ctx,#16]
+
+	tst	$len,#31
+	b.eq	.Leven_neon
+
+	ldp	$r0,$r1,[$ctx,#32]	// load key value
+
+	add	$h0,x10,x11,lsl#26	// base 2^26 -> base 2^64
+	lsr	$h1,x12,#12
+	adds	$h0,$h0,x12,lsl#52
+	add	$h1,$h1,x13,lsl#14
+	adc	$h1,$h1,xzr
+	lsr	$h2,x14,#24
+	adds	$h1,$h1,x14,lsl#40
+	adc	$d2,$h2,xzr		// can be partially reduced...
+
+	ldp	$d0,$d1,[$inp],#16	// load input
+	sub	$len,$len,#16
+	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
+
+	and	$t0,$d2,#-4		// ... so reduce
+	and	$h2,$d2,#3
+	add	$t0,$t0,$d2,lsr#2
+	adds	$h0,$h0,$t0
+	adc	$h1,$h1,xzr
+
+#ifdef	__ARMEB__
+	rev	$d0,$d0
+	rev	$d1,$d1
+#endif
+	adds	$h0,$h0,$d0		// accumulate input
+	adcs	$h1,$h1,$d1
+	adc	$h2,$h2,$padbit
+
+	bl	poly1305_mult
+	ldr	x30,[sp,#8]
+
+	cbz	$padbit,.Lstore_base2_64_neon
+
+	and	x10,$h0,#0x03ffffff	// base 2^64 -> base 2^26
+	ubfx	x11,$h0,#26,#26
+	extr	x12,$h1,$h0,#52
+	and	x12,x12,#0x03ffffff
+	ubfx	x13,$h1,#14,#26
+	extr	x14,$h2,$h1,#40
+
+	cbnz	$len,.Leven_neon
+
+	stp	w10,w11,[$ctx]		// store hash value base 2^26
+	stp	w12,w13,[$ctx,#8]
+	str	w14,[$ctx,#16]
+	b	.Lno_data_neon
+
+.align	4
+.Lstore_base2_64_neon:
+	stp	$h0,$h1,[$ctx]		// store hash value base 2^64
+	stp	$h2,xzr,[$ctx,#16]	// note that is_base2_26 is zeroed
+	b	.Lno_data_neon
+
+.align	4
+.Lbase2_64_neon:
+	ldp	$r0,$r1,[$ctx,#32]	// load key value
+
+	ldp	$h0,$h1,[$ctx]		// load hash value base 2^64
+	ldr	$h2,[$ctx,#16]
+
+	tst	$len,#31
+	b.eq	.Linit_neon
+
+	ldp	$d0,$d1,[$inp],#16	// load input
+	sub	$len,$len,#16
+	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
+#ifdef	__ARMEB__
+	rev	$d0,$d0
+	rev	$d1,$d1
+#endif
+	adds	$h0,$h0,$d0		// accumulate input
+	adcs	$h1,$h1,$d1
+	adc	$h2,$h2,$padbit
+
+	bl	poly1305_mult
+
+.Linit_neon:
+	and	x10,$h0,#0x03ffffff	// base 2^64 -> base 2^26
+	ubfx	x11,$h0,#26,#26
+	extr	x12,$h1,$h0,#52
+	and	x12,x12,#0x03ffffff
+	ubfx	x13,$h1,#14,#26
+	extr	x14,$h2,$h1,#40
+
+	stp	d8,d9,[sp,#16]		// meet ABI requirements
+	stp	d10,d11,[sp,#32]
+	stp	d12,d13,[sp,#48]
+	stp	d14,d15,[sp,#64]
+
+	fmov	${H0},x10
+	fmov	${H1},x11
+	fmov	${H2},x12
+	fmov	${H3},x13
+	fmov	${H4},x14
+
+	////////////////////////////////// initialize r^n table
+	mov	$h0,$r0			// r^1
+	add	$s1,$r1,$r1,lsr#2	// s1 = r1 + (r1 >> 2)
+	mov	$h1,$r1
+	mov	$h2,xzr
+	add	$ctx,$ctx,#48+12
+	bl	poly1305_splat
+
+	bl	poly1305_mult		// r^2
+	sub	$ctx,$ctx,#4
+	bl	poly1305_splat
+
+	bl	poly1305_mult		// r^3
+	sub	$ctx,$ctx,#4
+	bl	poly1305_splat
+
+	bl	poly1305_mult		// r^4
+	sub	$ctx,$ctx,#4
+	bl	poly1305_splat
+	ldr	x30,[sp,#8]
+
+	add	$in2,$inp,#32
+	adr	$zeros,.Lzeros
+	subs	$len,$len,#64
+	csel	$in2,$zeros,$in2,lo
+
+	mov	x4,#1
+	str	x4,[$ctx,#-24]		// set is_base2_26
+	sub	$ctx,$ctx,#48		// restore original $ctx
+	b	.Ldo_neon
+
+.align	4
+.Leven_neon:
+	add	$in2,$inp,#32
+	adr	$zeros,.Lzeros
+	subs	$len,$len,#64
+	csel	$in2,$zeros,$in2,lo
+
+	stp	d8,d9,[sp,#16]		// meet ABI requirements
+	stp	d10,d11,[sp,#32]
+	stp	d12,d13,[sp,#48]
+	stp	d14,d15,[sp,#64]
+
+	fmov	${H0},x10
+	fmov	${H1},x11
+	fmov	${H2},x12
+	fmov	${H3},x13
+	fmov	${H4},x14
+
+.Ldo_neon:
+	ldp	x8,x12,[$in2],#16	// inp[2:3] (or zero)
+	ldp	x9,x13,[$in2],#48
+
+	lsl	$padbit,$padbit,#24
+	add	x15,$ctx,#48
+
+#ifdef	__ARMEB__
+	rev	x8,x8
+	rev	x12,x12
+	rev	x9,x9
+	rev	x13,x13
+#endif
+	and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
+	and	x5,x9,#0x03ffffff
+	ubfx	x6,x8,#26,#26
+	ubfx	x7,x9,#26,#26
+	add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
+	extr	x8,x12,x8,#52
+	extr	x9,x13,x9,#52
+	add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
+	fmov	$IN23_0,x4
+	and	x8,x8,#0x03ffffff
+	and	x9,x9,#0x03ffffff
+	ubfx	x10,x12,#14,#26
+	ubfx	x11,x13,#14,#26
+	add	x12,$padbit,x12,lsr#40
+	add	x13,$padbit,x13,lsr#40
+	add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
+	fmov	$IN23_1,x6
+	add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
+	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
+	fmov	$IN23_2,x8
+	fmov	$IN23_3,x10
+	fmov	$IN23_4,x12
+
+	ldp	x8,x12,[$inp],#16	// inp[0:1]
+	ldp	x9,x13,[$inp],#48
+
+	ld1	{$R0,$R1,$S1,$R2},[x15],#64
+	ld1	{$S2,$R3,$S3,$R4},[x15],#64
+	ld1	{$S4},[x15]
+
+#ifdef	__ARMEB__
+	rev	x8,x8
+	rev	x12,x12
+	rev	x9,x9
+	rev	x13,x13
+#endif
+	and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
+	and	x5,x9,#0x03ffffff
+	ubfx	x6,x8,#26,#26
+	ubfx	x7,x9,#26,#26
+	add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
+	extr	x8,x12,x8,#52
+	extr	x9,x13,x9,#52
+	add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
+	fmov	$IN01_0,x4
+	and	x8,x8,#0x03ffffff
+	and	x9,x9,#0x03ffffff
+	ubfx	x10,x12,#14,#26
+	ubfx	x11,x13,#14,#26
+	add	x12,$padbit,x12,lsr#40
+	add	x13,$padbit,x13,lsr#40
+	add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
+	fmov	$IN01_1,x6
+	add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
+	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
+	fmov	$IN01_2,x8
+	fmov	$IN01_3,x10
+	fmov	$IN01_4,x12
+
+	b.ls	.Lskip_loop
+
+.align	4
+.Loop_neon:
+	////////////////////////////////////////////////////////////////
+	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
+	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
+	//   \___________________/
+	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
+	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
+	//   \___________________/ \____________________/
+	//
+	// Note that we start with inp[2:3]*r^2. This is because it
+	// doesn't depend on reduction in previous iteration.
+	////////////////////////////////////////////////////////////////
+	// d4 = h0*r4 + h1*r3   + h2*r2   + h3*r1   + h4*r0
+	// d3 = h0*r3 + h1*r2   + h2*r1   + h3*r0   + h4*5*r4
+	// d2 = h0*r2 + h1*r1   + h2*r0   + h3*5*r4 + h4*5*r3
+	// d1 = h0*r1 + h1*r0   + h2*5*r4 + h3*5*r3 + h4*5*r2
+	// d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
+
+	subs	$len,$len,#64
+	umull	$ACC4,$IN23_0,${R4}[2]
+	csel	$in2,$zeros,$in2,lo
+	umull	$ACC3,$IN23_0,${R3}[2]
+	umull	$ACC2,$IN23_0,${R2}[2]
+	 ldp	x8,x12,[$in2],#16	// inp[2:3] (or zero)
+	umull	$ACC1,$IN23_0,${R1}[2]
+	 ldp	x9,x13,[$in2],#48
+	umull	$ACC0,$IN23_0,${R0}[2]
+#ifdef	__ARMEB__
+	 rev	x8,x8
+	 rev	x12,x12
+	 rev	x9,x9
+	 rev	x13,x13
+#endif
+
+	umlal	$ACC4,$IN23_1,${R3}[2]
+	 and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
+	umlal	$ACC3,$IN23_1,${R2}[2]
+	 and	x5,x9,#0x03ffffff
+	umlal	$ACC2,$IN23_1,${R1}[2]
+	 ubfx	x6,x8,#26,#26
+	umlal	$ACC1,$IN23_1,${R0}[2]
+	 ubfx	x7,x9,#26,#26
+	umlal	$ACC0,$IN23_1,${S4}[2]
+	 add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
+
+	umlal	$ACC4,$IN23_2,${R2}[2]
+	 extr	x8,x12,x8,#52
+	umlal	$ACC3,$IN23_2,${R1}[2]
+	 extr	x9,x13,x9,#52
+	umlal	$ACC2,$IN23_2,${R0}[2]
+	 add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
+	umlal	$ACC1,$IN23_2,${S4}[2]
+	 fmov	$IN23_0,x4
+	umlal	$ACC0,$IN23_2,${S3}[2]
+	 and	x8,x8,#0x03ffffff
+
+	umlal	$ACC4,$IN23_3,${R1}[2]
+	 and	x9,x9,#0x03ffffff
+	umlal	$ACC3,$IN23_3,${R0}[2]
+	 ubfx	x10,x12,#14,#26
+	umlal	$ACC2,$IN23_3,${S4}[2]
+	 ubfx	x11,x13,#14,#26
+	umlal	$ACC1,$IN23_3,${S3}[2]
+	 add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
+	umlal	$ACC0,$IN23_3,${S2}[2]
+	 fmov	$IN23_1,x6
+
+	add	$IN01_2,$IN01_2,$H2
+	 add	x12,$padbit,x12,lsr#40
+	umlal	$ACC4,$IN23_4,${R0}[2]
+	 add	x13,$padbit,x13,lsr#40
+	umlal	$ACC3,$IN23_4,${S4}[2]
+	 add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
+	umlal	$ACC2,$IN23_4,${S3}[2]
+	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
+	umlal	$ACC1,$IN23_4,${S2}[2]
+	 fmov	$IN23_2,x8
+	umlal	$ACC0,$IN23_4,${S1}[2]
+	 fmov	$IN23_3,x10
+
+	////////////////////////////////////////////////////////////////
+	// (hash+inp[0:1])*r^4 and accumulate
+
+	add	$IN01_0,$IN01_0,$H0
+	 fmov	$IN23_4,x12
+	umlal	$ACC3,$IN01_2,${R1}[0]
+	 ldp	x8,x12,[$inp],#16	// inp[0:1]
+	umlal	$ACC0,$IN01_2,${S3}[0]
+	 ldp	x9,x13,[$inp],#48
+	umlal	$ACC4,$IN01_2,${R2}[0]
+	umlal	$ACC1,$IN01_2,${S4}[0]
+	umlal	$ACC2,$IN01_2,${R0}[0]
+#ifdef	__ARMEB__
+	 rev	x8,x8
+	 rev	x12,x12
+	 rev	x9,x9
+	 rev	x13,x13
+#endif
+
+	add	$IN01_1,$IN01_1,$H1
+	umlal	$ACC3,$IN01_0,${R3}[0]
+	umlal	$ACC4,$IN01_0,${R4}[0]
+	 and	x4,x8,#0x03ffffff	// base 2^64 -> base 2^26
+	umlal	$ACC2,$IN01_0,${R2}[0]
+	 and	x5,x9,#0x03ffffff
+	umlal	$ACC0,$IN01_0,${R0}[0]
+	 ubfx	x6,x8,#26,#26
+	umlal	$ACC1,$IN01_0,${R1}[0]
+	 ubfx	x7,x9,#26,#26
+
+	add	$IN01_3,$IN01_3,$H3
+	 add	x4,x4,x5,lsl#32		// bfi	x4,x5,#32,#32
+	umlal	$ACC3,$IN01_1,${R2}[0]
+	 extr	x8,x12,x8,#52
+	umlal	$ACC4,$IN01_1,${R3}[0]
+	 extr	x9,x13,x9,#52
+	umlal	$ACC0,$IN01_1,${S4}[0]
+	 add	x6,x6,x7,lsl#32		// bfi	x6,x7,#32,#32
+	umlal	$ACC2,$IN01_1,${R1}[0]
+	 fmov	$IN01_0,x4
+	umlal	$ACC1,$IN01_1,${R0}[0]
+	 and	x8,x8,#0x03ffffff
+
+	add	$IN01_4,$IN01_4,$H4
+	 and	x9,x9,#0x03ffffff
+	umlal	$ACC3,$IN01_3,${R0}[0]
+	 ubfx	x10,x12,#14,#26
+	umlal	$ACC0,$IN01_3,${S2}[0]
+	 ubfx	x11,x13,#14,#26
+	umlal	$ACC4,$IN01_3,${R1}[0]
+	 add	x8,x8,x9,lsl#32		// bfi	x8,x9,#32,#32
+	umlal	$ACC1,$IN01_3,${S3}[0]
+	 fmov	$IN01_1,x6
+	umlal	$ACC2,$IN01_3,${S4}[0]
+	 add	x12,$padbit,x12,lsr#40
+
+	umlal	$ACC3,$IN01_4,${S4}[0]
+	 add	x13,$padbit,x13,lsr#40
+	umlal	$ACC0,$IN01_4,${S1}[0]
+	 add	x10,x10,x11,lsl#32	// bfi	x10,x11,#32,#32
+	umlal	$ACC4,$IN01_4,${R0}[0]
+	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
+	umlal	$ACC1,$IN01_4,${S2}[0]
+	 fmov	$IN01_2,x8
+	umlal	$ACC2,$IN01_4,${S3}[0]
+	 fmov	$IN01_3,x10
+
+	/////////////////////////////////////////////////////////////////
+	// lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
+        // and P. Schwabe
+
+	ushr	$T0.2d,$ACC3,#26
+	 fmov	$IN01_4,x12
+	xtn	$H3,$ACC3
+	 ushr	$T1.2d,$ACC0,#26
+	 xtn	$H0,$ACC0
+	add	$ACC4,$ACC4,$T0.2d	// h3 -> h4
+	bic	$H3,#0xfc,lsl#24	// &=0x03ffffff
+	 add	$ACC1,$ACC1,$T1.2d	// h0 -> h1
+	 bic	$H0,#0xfc,lsl#24
+
+	shrn	$T0.2s,$ACC4,#26
+	xtn	$H4,$ACC4
+	 ushr	$T1.2d,$ACC1,#26
+	 xtn	$H1,$ACC1
+	 add	$ACC2,$ACC2,$T1.2d	// h1 -> h2
+	bic	$H4,#0xfc,lsl#24
+	 bic	$H1,#0xfc,lsl#24
+
+	add	$H0,$H0,$T0.2s
+	shl	$T0.2s,$T0.2s,#2
+	 shrn	$T1.2s,$ACC2,#26
+	 xtn	$H2,$ACC2
+	add	$H0,$H0,$T0.2s		// h4 -> h0
+	 add	$H3,$H3,$T1.2s		// h2 -> h3
+	 bic	$H2,#0xfc,lsl#24
+
+	ushr	$T0.2s,$H0,#26
+	bic	$H0,#0xfc,lsl#24
+	 ushr	$T1.2s,$H3,#26
+	 bic	$H3,#0xfc,lsl#24
+	add	$H1,$H1,$T0.2s		// h0 -> h1
+	 add	$H4,$H4,$T1.2s		// h3 -> h4
+
+	b.hi	.Loop_neon
+
+.Lskip_loop:
+	dup	$IN23_2,${IN23_2}[0]
+	movi	$MASK.2d,#-1
+	add	$IN01_2,$IN01_2,$H2
+	ushr	$MASK.2d,$MASK.2d,#38
+
+	////////////////////////////////////////////////////////////////
+	// multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
+
+	adds	$len,$len,#32
+	b.ne	.Long_tail
+
+	dup	$IN23_2,${IN01_2}[0]
+	add	$IN23_0,$IN01_0,$H0
+	add	$IN23_3,$IN01_3,$H3
+	add	$IN23_1,$IN01_1,$H1
+	add	$IN23_4,$IN01_4,$H4
+
+.Long_tail:
+	dup	$IN23_0,${IN23_0}[0]
+	umull2	$ACC0,$IN23_2,${S3}
+	umull2	$ACC3,$IN23_2,${R1}
+	umull2	$ACC4,$IN23_2,${R2}
+	umull2	$ACC2,$IN23_2,${R0}
+	umull2	$ACC1,$IN23_2,${S4}
+
+	dup	$IN23_1,${IN23_1}[0]
+	umlal2	$ACC0,$IN23_0,${R0}
+	umlal2	$ACC2,$IN23_0,${R2}
+	umlal2	$ACC3,$IN23_0,${R3}
+	umlal2	$ACC4,$IN23_0,${R4}
+	umlal2	$ACC1,$IN23_0,${R1}
+
+	dup	$IN23_3,${IN23_3}[0]
+	umlal2	$ACC0,$IN23_1,${S4}
+	umlal2	$ACC3,$IN23_1,${R2}
+	umlal2	$ACC2,$IN23_1,${R1}
+	umlal2	$ACC4,$IN23_1,${R3}
+	umlal2	$ACC1,$IN23_1,${R0}
+
+	dup	$IN23_4,${IN23_4}[0]
+	umlal2	$ACC3,$IN23_3,${R0}
+	umlal2	$ACC4,$IN23_3,${R1}
+	umlal2	$ACC0,$IN23_3,${S2}
+	umlal2	$ACC1,$IN23_3,${S3}
+	umlal2	$ACC2,$IN23_3,${S4}
+
+	umlal2	$ACC3,$IN23_4,${S4}
+	umlal2	$ACC0,$IN23_4,${S1}
+	umlal2	$ACC4,$IN23_4,${R0}
+	umlal2	$ACC1,$IN23_4,${S2}
+	umlal2	$ACC2,$IN23_4,${S3}
+
+	b.eq	.Lshort_tail
+
+	////////////////////////////////////////////////////////////////
+	// (hash+inp[0:1])*r^4:r^3 and accumulate
+
+	add	$IN01_0,$IN01_0,$H0
+	umlal	$ACC3,$IN01_2,${R1}
+	umlal	$ACC0,$IN01_2,${S3}
+	umlal	$ACC4,$IN01_2,${R2}
+	umlal	$ACC1,$IN01_2,${S4}
+	umlal	$ACC2,$IN01_2,${R0}
+
+	add	$IN01_1,$IN01_1,$H1
+	umlal	$ACC3,$IN01_0,${R3}
+	umlal	$ACC0,$IN01_0,${R0}
+	umlal	$ACC4,$IN01_0,${R4}
+	umlal	$ACC1,$IN01_0,${R1}
+	umlal	$ACC2,$IN01_0,${R2}
+
+	add	$IN01_3,$IN01_3,$H3
+	umlal	$ACC3,$IN01_1,${R2}
+	umlal	$ACC0,$IN01_1,${S4}
+	umlal	$ACC4,$IN01_1,${R3}
+	umlal	$ACC1,$IN01_1,${R0}
+	umlal	$ACC2,$IN01_1,${R1}
+
+	add	$IN01_4,$IN01_4,$H4
+	umlal	$ACC3,$IN01_3,${R0}
+	umlal	$ACC0,$IN01_3,${S2}
+	umlal	$ACC4,$IN01_3,${R1}
+	umlal	$ACC1,$IN01_3,${S3}
+	umlal	$ACC2,$IN01_3,${S4}
+
+	umlal	$ACC3,$IN01_4,${S4}
+	umlal	$ACC0,$IN01_4,${S1}
+	umlal	$ACC4,$IN01_4,${R0}
+	umlal	$ACC1,$IN01_4,${S2}
+	umlal	$ACC2,$IN01_4,${S3}
+
+.Lshort_tail:
+	////////////////////////////////////////////////////////////////
+	// lazy reduction, but without narrowing
+
+	ushr	$T0.2d,$ACC3,#26
+	and	$ACC3,$ACC3,$MASK.2d
+	 ushr	$T1.2d,$ACC0,#26
+	 and	$ACC0,$ACC0,$MASK.2d
+
+	add	$ACC4,$ACC4,$T0.2d	// h3 -> h4
+	 add	$ACC1,$ACC1,$T1.2d	// h0 -> h1
+
+	ushr	$T0.2d,$ACC4,#26
+	and	$ACC4,$ACC4,$MASK.2d
+	 ushr	$T1.2d,$ACC1,#26
+	 and	$ACC1,$ACC1,$MASK.2d
+	 add	$ACC2,$ACC2,$T1.2d	// h1 -> h2
+
+	add	$ACC0,$ACC0,$T0.2d
+	shl	$T0.2d,$T0.2d,#2
+	 ushr	$T1.2d,$ACC2,#26
+	 and	$ACC2,$ACC2,$MASK.2d
+	add	$ACC0,$ACC0,$T0.2d	// h4 -> h0
+	 add	$ACC3,$ACC3,$T1.2d	// h2 -> h3
+
+	ushr	$T0.2d,$ACC0,#26
+	and	$ACC0,$ACC0,$MASK.2d
+	 ushr	$T1.2d,$ACC3,#26
+	 and	$ACC3,$ACC3,$MASK.2d
+	add	$ACC1,$ACC1,$T0.2d	// h0 -> h1
+	 add	$ACC4,$ACC4,$T1.2d	// h3 -> h4
+
+	////////////////////////////////////////////////////////////////
+	// horizontal add
+
+	addp	$ACC2,$ACC2,$ACC2
+	 ldp	d8,d9,[sp,#16]		// meet ABI requirements
+	addp	$ACC0,$ACC0,$ACC0
+	 ldp	d10,d11,[sp,#32]
+	addp	$ACC1,$ACC1,$ACC1
+	 ldp	d12,d13,[sp,#48]
+	addp	$ACC3,$ACC3,$ACC3
+	 ldp	d14,d15,[sp,#64]
+	addp	$ACC4,$ACC4,$ACC4
+
+	////////////////////////////////////////////////////////////////
+	// write the result, can be partially reduced
+
+	st4	{$ACC0,$ACC1,$ACC2,$ACC3}[0],[$ctx],#16
+	st1	{$ACC4}[0],[$ctx]
+
+.Lno_data_neon:
+	ldr	x29,[sp],#80
+	ret
+.size	poly1305_blocks_neon,.-poly1305_blocks_neon
+
+.type	poly1305_emit_neon,%function
+.align	5
+poly1305_emit_neon:
+	ldr	$is_base2_26,[$ctx,#24]
+	cbz	$is_base2_26,poly1305_emit
+
+	ldp	w10,w11,[$ctx]		// load hash value base 2^26
+	ldp	w12,w13,[$ctx,#8]
+	ldr	w14,[$ctx,#16]
+
+	add	$h0,x10,x11,lsl#26	// base 2^26 -> base 2^64
+	lsr	$h1,x12,#12
+	adds	$h0,$h0,x12,lsl#52
+	add	$h1,$h1,x13,lsl#14
+	adc	$h1,$h1,xzr
+	lsr	$h2,x14,#24
+	adds	$h1,$h1,x14,lsl#40
+	adc	$h2,$h2,xzr		// can be partially reduced...
+
+	ldp	$t0,$t1,[$nonce]	// load nonce
+
+	and	$d0,$h2,#-4		// ... so reduce
+	add	$d0,$d0,$h2,lsr#2
+	and	$h2,$h2,#3
+	adds	$h0,$h0,$d0
+	adc	$h1,$h1,xzr
+
+	adds	$d0,$h0,#5		// compare to modulus
+	adcs	$d1,$h1,xzr
+	adc	$d2,$h2,xzr
+
+	tst	$d2,#-4			// see if it's carried/borrowed
+
+	csel	$h0,$h0,$d0,eq
+	csel	$h1,$h1,$d1,eq
+
+#ifdef	__ARMEB__
+	ror	$t0,$t0,#32		// flip nonce words
+	ror	$t1,$t1,#32
+#endif
+	adds	$h0,$h0,$t0		// accumulate nonce
+	adc	$h1,$h1,$t1
+#ifdef	__ARMEB__
+	rev	$h0,$h0			// flip output bytes
+	rev	$h1,$h1
+#endif
+	stp	$h0,$h1,[$mac]		// write result
+
+	ret
+.size	poly1305_emit_neon,.-poly1305_emit_neon
+
+.align	5
+.Lzeros:
+.long	0,0,0,0,0,0,0,0
+.LOPENSSL_armcap_P:
+#ifdef	__ILP32__
+.long	OPENSSL_armcap_P-.
+#else
+.quad	OPENSSL_armcap_P-.
+#endif
+.asciz	"Poly1305 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
+.align	2
+___
+
+foreach (split("\n",$code)) {
+	s/\b(shrn\s+v[0-9]+)\.[24]d/$1.2s/			or
+	s/\b(fmov\s+)v([0-9]+)[^,]*,\s*x([0-9]+)/$1d$2,x$3/	or
+	(m/\bdup\b/ and (s/\.[24]s/.2d/g or 1))			or
+	(m/\b(eor|and)/ and (s/\.[248][sdh]/.16b/g or 1))	or
+	(m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1))		or
+	(m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1))		or
+	(m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));
+
+	s/\.[124]([sd])\[/.$1\[/;
+
+	print $_,"\n";
+}
+close STDOUT;
diff --git a/crypto/poly1305/asm/poly1305-x86.pl b/crypto/poly1305/asm/poly1305-x86.pl
new file mode 100755
index 00000000..7c1aee5f
--- /dev/null
+++ b/crypto/poly1305/asm/poly1305-x86.pl
@@ -0,0 +1,1794 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements Poly1305 hash for x86.
+#
+# April 2015
+#
+# Numbers are cycles per processed byte with poly1305_blocks alone,
+# measured with rdtsc at fixed clock frequency.
+#
+#		IALU/gcc-3.4(*)	SSE2(**)	AVX2
+# Pentium	15.7/+80%	-
+# PIII		6.21/+90%	-
+# P4		19.8/+40%	3.24
+# Core 2	4.85/+90%	1.80
+# Westmere	4.58/+100%	1.43
+# Sandy Bridge	3.90/+100%	1.36
+# Haswell	3.88/+70%	1.18		0.72
+# Silvermont	11.0/+40%	4.80
+# VIA Nano	6.71/+90%	2.47
+# Sledgehammer	3.51/+180%	4.27
+# Bulldozer	4.53/+140%	1.31
+#
+# (*)	gcc 4.8 for some reason generated worse code;
+# (**)	besides SSE2 there are floating-point and AVX options; FP
+#	is deemed unnecessary, because pre-SSE2 processor are too
+#	old to care about, while it's not the fastest option on
+#	SSE2-capable ones; AVX is omitted, because it doesn't give
+#	a lot of improvement, 5-10% depending on processor;
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
+require "x86asm.pl";
+
+&asm_init($ARGV[0],"poly1305-x86.pl",$ARGV[$#ARGV] eq "386");
+
+$sse2=$avx=0;
+for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
+
+if ($sse2) {
+	&static_label("const_sse2");
+	&static_label("enter_blocks");
+	&static_label("enter_emit");
+	&external_label("OPENSSL_ia32cap_P");
+
+	if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
+			=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
+		$avx = ($1>=2.19) + ($1>=2.22);
+	}
+
+	if (!$avx && $ARGV[0] eq "win32n" &&
+	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.09) + ($1>=2.10);
+	}
+
+	if (!$avx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9]\.[0-9]+)/) {
+		$avx = ($2>=3.0) + ($2>3.0);
+	}
+}
+
+########################################################################
+# Layout of opaque area is following.
+#
+#	unsigned __int32 h[5];		# current hash value base 2^32
+#	unsigned __int32 pad;		# is_base2_26 in vector context
+#	unsigned __int32 r[4];		# key value base 2^32
+
+&align(64);
+&function_begin("poly1305_init");
+	&mov	("edi",&wparam(0));		# context
+	&mov	("esi",&wparam(1));		# key
+	&mov	("ebp",&wparam(2));		# function table
+
+	&xor	("eax","eax");
+	&mov	(&DWP(4*0,"edi"),"eax");	# zero hash value
+	&mov	(&DWP(4*1,"edi"),"eax");
+	&mov	(&DWP(4*2,"edi"),"eax");
+	&mov	(&DWP(4*3,"edi"),"eax");
+	&mov	(&DWP(4*4,"edi"),"eax");
+	&mov	(&DWP(4*5,"edi"),"eax");	# is_base2_26
+
+	&cmp	("esi",0);
+	&je	(&label("nokey"));
+
+    if ($sse2) {
+	&call	(&label("pic_point"));
+    &set_label("pic_point");
+	&blindpop("ebx");
+
+	&lea	("eax",&DWP("poly1305_blocks-".&label("pic_point"),"ebx"));
+	&lea	("edx",&DWP("poly1305_emit-".&label("pic_point"),"ebx"));
+
+	&picmeup("edi","OPENSSL_ia32cap_P","ebx",&label("pic_point"));
+	&mov	("ecx",&DWP(0,"edi"));
+	&and	("ecx",1<<26|1<<24);
+	&cmp	("ecx",1<<26|1<<24);		# SSE2 and XMM?
+	&jne	(&label("no_sse2"));
+
+	&lea	("eax",&DWP("_poly1305_blocks_sse2-".&label("pic_point"),"ebx"));
+	&lea	("edx",&DWP("_poly1305_emit_sse2-".&label("pic_point"),"ebx"));
+
+      if ($avx>1) {
+	&mov	("ecx",&DWP(8,"edi"));
+	&test	("ecx",1<<5);			# AVX2?
+	&jz	(&label("no_sse2"));
+
+	&lea	("eax",&DWP("_poly1305_blocks_avx2-".&label("pic_point"),"ebx"));
+      }
+    &set_label("no_sse2");
+	&mov	("edi",&wparam(0));		# reload context
+	&mov	(&DWP(0,"ebp"),"eax");		# fill function table
+	&mov	(&DWP(4,"ebp"),"edx");
+    }
+
+	&mov	("eax",&DWP(4*0,"esi"));	# load input key
+	&mov	("ebx",&DWP(4*1,"esi"));
+	&mov	("ecx",&DWP(4*2,"esi"));
+	&mov	("edx",&DWP(4*3,"esi"));
+	&and	("eax",0x0fffffff);
+	&and	("ebx",0x0ffffffc);
+	&and	("ecx",0x0ffffffc);
+	&and	("edx",0x0ffffffc);
+	&mov	(&DWP(4*6,"edi"),"eax");
+	&mov	(&DWP(4*7,"edi"),"ebx");
+	&mov	(&DWP(4*8,"edi"),"ecx");
+	&mov	(&DWP(4*9,"edi"),"edx");
+
+	&mov	("eax",$sse2);
+&set_label("nokey");
+&function_end("poly1305_init");
+
+($h0,$h1,$h2,$h3,$h4,
+ $d0,$d1,$d2,$d3,
+ $r0,$r1,$r2,$r3,
+     $s1,$s2,$s3)=map(4*$_,(0..15));
+
+&function_begin("poly1305_blocks");
+	&mov	("edi",&wparam(0));		# ctx
+	&mov	("esi",&wparam(1));		# inp
+	&mov	("ecx",&wparam(2));		# len
+&set_label("enter_blocks");
+	&and	("ecx",-15);
+	&jz	(&label("nodata"));
+
+	&stack_push(16);
+	&mov	("eax",&DWP(4*6,"edi"));	# r0
+	&mov	("ebx",&DWP(4*7,"edi"));	# r1
+	 &lea	("ebp",&DWP(0,"esi","ecx"));	# end of input
+	&mov	("ecx",&DWP(4*8,"edi"));	# r2
+	&mov	("edx",&DWP(4*9,"edi"));	# r3
+
+	&mov	(&wparam(2),"ebp");
+	&mov	("ebp","esi");
+
+	&mov	(&DWP($r0,"esp"),"eax");	# r0
+	&mov	("eax","ebx");
+	&shr	("eax",2);
+	&mov	(&DWP($r1,"esp"),"ebx");	# r1
+	&add	("eax","ebx");			# s1
+	&mov	("ebx","ecx");
+	&shr	("ebx",2);
+	&mov	(&DWP($r2,"esp"),"ecx");	# r2
+	&add	("ebx","ecx");			# s2
+	&mov	("ecx","edx");
+	&shr	("ecx",2);
+	&mov	(&DWP($r3,"esp"),"edx");	# r3
+	&add	("ecx","edx");			# s3
+	&mov	(&DWP($s1,"esp"),"eax");	# s1
+	&mov	(&DWP($s2,"esp"),"ebx");	# s2
+	&mov	(&DWP($s3,"esp"),"ecx");	# s3
+
+	&mov	("eax",&DWP(4*0,"edi"));	# load hash value
+	&mov	("ebx",&DWP(4*1,"edi"));
+	&mov	("ecx",&DWP(4*2,"edi"));
+	&mov	("esi",&DWP(4*3,"edi"));
+	&mov	("edi",&DWP(4*4,"edi"));
+	&jmp	(&label("loop"));
+
+&set_label("loop",32);
+	&add	("eax",&DWP(4*0,"ebp"));	# accumulate input
+	&adc	("ebx",&DWP(4*1,"ebp"));
+	&adc	("ecx",&DWP(4*2,"ebp"));
+	&adc	("esi",&DWP(4*3,"ebp"));
+	&lea	("ebp",&DWP(4*4,"ebp"));
+	&adc	("edi",&wparam(3));		# padbit
+
+	&mov	(&DWP($h0,"esp"),"eax");	# put aside hash[+inp]
+	&mov	(&DWP($h3,"esp"),"esi");
+
+	&mul	(&DWP($r0,"esp"));		# h0*r0
+	 &mov	(&DWP($h4,"esp"),"edi");
+	&mov	("edi","eax");
+	&mov	("eax","ebx");			# h1
+	&mov	("esi","edx");
+	&mul	(&DWP($s3,"esp"));		# h1*s3
+	&add	("edi","eax");
+	&mov	("eax","ecx");			# h2
+	&adc	("esi","edx");
+	&mul	(&DWP($s2,"esp"));		# h2*s2
+	&add	("edi","eax");
+	&mov	("eax",&DWP($h3,"esp"));
+	&adc	("esi","edx");
+	&mul	(&DWP($s1,"esp"));		# h3*s1
+	&add	("edi","eax");
+	 &mov	("eax",&DWP($h0,"esp"));
+	&adc	("esi","edx");
+
+	&mul	(&DWP($r1,"esp"));		# h0*r1
+	 &mov	(&DWP($d0,"esp"),"edi");
+	&xor	("edi","edi");
+	&add	("esi","eax");
+	&mov	("eax","ebx");			# h1
+	&adc	("edi","edx");
+	&mul	(&DWP($r0,"esp"));		# h1*r0
+	&add	("esi","eax");
+	&mov	("eax","ecx");			# h2
+	&adc	("edi","edx");
+	&mul	(&DWP($s3,"esp"));		# h2*s3
+	&add	("esi","eax");
+	&mov	("eax",&DWP($h3,"esp"));
+	&adc	("edi","edx");
+	&mul	(&DWP($s2,"esp"));		# h3*s2
+	&add	("esi","eax");
+	&mov	("eax",&DWP($h4,"esp"));
+	&adc	("edi","edx");
+	&imul	("eax",&DWP($s1,"esp"));	# h4*s1
+	&add	("esi","eax");
+	 &mov	("eax",&DWP($h0,"esp"));
+	&adc	("edi",0);
+
+	&mul	(&DWP($r2,"esp"));		# h0*r2
+	 &mov	(&DWP($d1,"esp"),"esi");
+	&xor	("esi","esi");
+	&add	("edi","eax");
+	&mov	("eax","ebx");			# h1
+	&adc	("esi","edx");
+	&mul	(&DWP($r1,"esp"));		# h1*r1
+	&add	("edi","eax");
+	&mov	("eax","ecx");			# h2
+	&adc	("esi","edx");
+	&mul	(&DWP($r0,"esp"));		# h2*r0
+	&add	("edi","eax");
+	&mov	("eax",&DWP($h3,"esp"));
+	&adc	("esi","edx");
+	&mul	(&DWP($s3,"esp"));		# h3*s3
+	&add	("edi","eax");
+	&mov	("eax",&DWP($h4,"esp"));
+	&adc	("esi","edx");
+	&imul	("eax",&DWP($s2,"esp"));	# h4*s2
+	&add	("edi","eax");
+	 &mov	("eax",&DWP($h0,"esp"));
+	&adc	("esi",0);
+
+	&mul	(&DWP($r3,"esp"));		# h0*r3
+	 &mov	(&DWP($d2,"esp"),"edi");
+	&xor	("edi","edi");
+	&add	("esi","eax");
+	&mov	("eax","ebx");			# h1
+	&adc	("edi","edx");
+	&mul	(&DWP($r2,"esp"));		# h1*r2
+	&add	("esi","eax");
+	&mov	("eax","ecx");			# h2
+	&adc	("edi","edx");
+	&mul	(&DWP($r1,"esp"));		# h2*r1
+	&add	("esi","eax");
+	&mov	("eax",&DWP($h3,"esp"));
+	&adc	("edi","edx");
+	&mul	(&DWP($r0,"esp"));		# h3*r0
+	&add	("esi","eax");
+	 &mov	("ecx",&DWP($h4,"esp"));
+	&adc	("edi","edx");
+
+	&mov	("edx","ecx");
+	&imul	("ecx",&DWP($s3,"esp"));	# h4*s3
+	&add	("esi","ecx");
+	 &mov	("eax",&DWP($d0,"esp"));
+	&adc	("edi",0);
+
+	&imul	("edx",&DWP($r0,"esp"));	# h4*r0
+	&add	("edx","edi");
+
+	&mov	("ebx",&DWP($d1,"esp"));
+	&mov	("ecx",&DWP($d2,"esp"));
+
+	&mov	("edi","edx");			# last reduction step
+	&shr	("edx",2);
+	&and	("edi",3);
+	&lea	("edx",&DWP(0,"edx","edx",4));	# *5
+	&add	("eax","edx");
+	&adc	("ebx",0);
+	&adc	("ecx",0);
+	&adc	("esi",0);
+
+	&cmp	("ebp",&wparam(2));		# done yet?
+	&jne	(&label("loop"));
+
+	&mov	("edx",&wparam(0));		# ctx
+	&stack_pop(16);
+	&mov	(&DWP(4*0,"edx"),"eax");	# store hash value
+	&mov	(&DWP(4*1,"edx"),"ebx");
+	&mov	(&DWP(4*2,"edx"),"ecx");
+	&mov	(&DWP(4*3,"edx"),"esi");
+	&mov	(&DWP(4*4,"edx"),"edi");
+&set_label("nodata");
+&function_end("poly1305_blocks");
+
+&function_begin("poly1305_emit");
+	&mov	("ebp",&wparam(0));		# context
+&set_label("enter_emit");
+	&mov	("edi",&wparam(1));		# output
+	&mov	("eax",&DWP(4*0,"ebp"));	# load hash value
+	&mov	("ebx",&DWP(4*1,"ebp"));
+	&mov	("ecx",&DWP(4*2,"ebp"));
+	&mov	("edx",&DWP(4*3,"ebp"));
+	&mov	("esi",&DWP(4*4,"ebp"));
+
+	&add	("eax",5);			# compare to modulus
+	&adc	("ebx",0);
+	&adc	("ecx",0);
+	&adc	("edx",0);
+	&adc	("esi",0);
+	&shr	("esi",2);			# did it carry/borrow?
+	&neg	("esi");			# do we choose hash-modulus?
+
+	&and	("eax","esi");
+	&and	("ebx","esi");
+	&and	("ecx","esi");
+	&and	("edx","esi");
+	&mov	(&DWP(4*0,"edi"),"eax");
+	&mov	(&DWP(4*1,"edi"),"ebx");
+	&mov	(&DWP(4*2,"edi"),"ecx");
+	&mov	(&DWP(4*3,"edi"),"edx");
+
+	&not	("esi");			# or original hash value?
+	&mov	("eax",&DWP(4*0,"ebp"));
+	&mov	("ebx",&DWP(4*1,"ebp"));
+	&mov	("ecx",&DWP(4*2,"ebp"));
+	&mov	("edx",&DWP(4*3,"ebp"));
+	&mov	("ebp",&wparam(2));
+	&and	("eax","esi");
+	&and	("ebx","esi");
+	&and	("ecx","esi");
+	&and	("edx","esi");
+	&or	("eax",&DWP(4*0,"edi"));
+	&or	("ebx",&DWP(4*1,"edi"));
+	&or	("ecx",&DWP(4*2,"edi"));
+	&or	("edx",&DWP(4*3,"edi"));
+
+	&add	("eax",&DWP(4*0,"ebp"));	# accumulate key
+	&adc	("ebx",&DWP(4*1,"ebp"));
+	&adc	("ecx",&DWP(4*2,"ebp"));
+	&adc	("edx",&DWP(4*3,"ebp"));
+
+	&mov	(&DWP(4*0,"edi"),"eax");
+	&mov	(&DWP(4*1,"edi"),"ebx");
+	&mov	(&DWP(4*2,"edi"),"ecx");
+	&mov	(&DWP(4*3,"edi"),"edx");
+&function_end("poly1305_emit");
+
+if ($sse2) {
+########################################################################
+# Layout of opaque area is following.
+#
+#	unsigned __int32 h[5];		# current hash value base 2^26
+#	unsigned __int32 is_base2_26;
+#	unsigned __int32 r[4];		# key value base 2^32
+#	unsigned __int32 pad[2];
+#	struct { unsigned __int32 r^4, r^3, r^2, r^1; } r[9];
+#
+# where r^n are base 2^26 digits of degrees of multiplier key. There are
+# 5 digits, but last four are interleaved with multiples of 5, totalling
+# in 9 elements: r0, r1, 5*r1, r2, 5*r2, r3, 5*r3, r4, 5*r4.
+
+my ($D0,$D1,$D2,$D3,$D4,$T0,$T1,$T2)=map("xmm$_",(0..7));
+my $MASK=$T2;	# borrow and keep in mind
+
+&align	(32);
+&function_begin_B("_poly1305_init_sse2");
+	&movdqu		($D4,&QWP(4*6,"edi"));		# key base 2^32
+	&lea		("edi",&DWP(16*3,"edi"));	# size optimization
+	&mov		("ebp","esp");
+	&sub		("esp",16*(9+5));
+	&and		("esp",-16);
+
+	#&pand		($D4,&QWP(96,"ebx"));		# magic mask
+	&movq		($MASK,&QWP(64,"ebx"));
+
+	&movdqa		($D0,$D4);
+	&movdqa		($D1,$D4);
+	&movdqa		($D2,$D4);
+
+	&pand		($D0,$MASK);			# -> base 2^26
+	&psrlq		($D1,26);
+	&psrldq		($D2,6);
+	&pand		($D1,$MASK);
+	&movdqa		($D3,$D2);
+	&psrlq		($D2,4)
+	&psrlq		($D3,30);
+	&pand		($D2,$MASK);
+	&pand		($D3,$MASK);
+	&psrldq		($D4,13);
+
+	&lea		("edx",&DWP(16*9,"esp"));	# size optimization
+	&mov		("ecx",2);
+&set_label("square");
+	&movdqa		(&QWP(16*0,"esp"),$D0);
+	&movdqa		(&QWP(16*1,"esp"),$D1);
+	&movdqa		(&QWP(16*2,"esp"),$D2);
+	&movdqa		(&QWP(16*3,"esp"),$D3);
+	&movdqa		(&QWP(16*4,"esp"),$D4);
+
+	&movdqa		($T1,$D1);
+	&movdqa		($T0,$D2);
+	&pslld		($T1,2);
+	&pslld		($T0,2);
+	&paddd		($T1,$D1);			# *5
+	&paddd		($T0,$D2);			# *5
+	&movdqa		(&QWP(16*5,"esp"),$T1);
+	&movdqa		(&QWP(16*6,"esp"),$T0);
+	&movdqa		($T1,$D3);
+	&movdqa		($T0,$D4);
+	&pslld		($T1,2);
+	&pslld		($T0,2);
+	&paddd		($T1,$D3);			# *5
+	&paddd		($T0,$D4);			# *5
+	&movdqa		(&QWP(16*7,"esp"),$T1);
+	&movdqa		(&QWP(16*8,"esp"),$T0);
+
+	&pshufd		($T1,$D0,0b01000100);
+	&movdqa		($T0,$D1);
+	&pshufd		($D1,$D1,0b01000100);
+	&pshufd		($D2,$D2,0b01000100);
+	&pshufd		($D3,$D3,0b01000100);
+	&pshufd		($D4,$D4,0b01000100);
+	&movdqa		(&QWP(16*0,"edx"),$T1);
+	&movdqa		(&QWP(16*1,"edx"),$D1);
+	&movdqa		(&QWP(16*2,"edx"),$D2);
+	&movdqa		(&QWP(16*3,"edx"),$D3);
+	&movdqa		(&QWP(16*4,"edx"),$D4);
+
+	################################################################
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	&pmuludq	($D4,$D0);			# h4*r0
+	&pmuludq	($D3,$D0);			# h3*r0
+	&pmuludq	($D2,$D0);			# h2*r0
+	&pmuludq	($D1,$D0);			# h1*r0
+	&pmuludq	($D0,$T1);			# h0*r0
+
+sub pmuladd {
+my $load = shift;
+my $base = shift; $base = "esp" if (!defined($base));
+
+	################################################################
+	# As for choice to "rotate" $T0-$T2 in order to move paddq
+	# past next multiplication. While it makes code harder to read
+	# and doesn't have significant effect on most processors, it
+	# makes a lot of difference on Atom, up to 30% improvement.
+
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&QWP(16*3,$base));		# r1*h3
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&QWP(16*2,$base));		# r1*h2
+	&paddq		($D4,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&QWP(16*1,$base));		# r1*h1
+	&paddq		($D3,$T1);
+	&$load		($T1,5);			# s1
+	&pmuludq	($T0,&QWP(16*0,$base));		# r1*h0
+	&paddq		($D2,$T2);
+	&pmuludq	($T1,&QWP(16*4,$base));		# s1*h4
+	 &$load		($T2,2);			# r2^n
+	&paddq		($D1,$T0);
+
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&QWP(16*2,$base));		# r2*h2
+	 &paddq		($D0,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&QWP(16*1,$base));		# r2*h1
+	&paddq		($D4,$T2);
+	&$load		($T2,6);			# s2^n
+	&pmuludq	($T1,&QWP(16*0,$base));		# r2*h0
+	&paddq		($D3,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&QWP(16*4,$base));		# s2*h4
+	&paddq		($D2,$T1);
+	&pmuludq	($T0,&QWP(16*3,$base));		# s2*h3
+	 &$load		($T1,3);			# r3^n
+	&paddq		($D1,$T2);
+
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&QWP(16*1,$base));		# r3*h1
+	 &paddq		($D0,$T0);
+	&$load		($T0,7);			# s3^n
+	&pmuludq	($T2,&QWP(16*0,$base));		# r3*h0
+	&paddq		($D4,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&QWP(16*4,$base));		# s3*h4
+	&paddq		($D3,$T2);
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&QWP(16*3,$base));		# s3*h3
+	&paddq		($D2,$T0);
+	&pmuludq	($T2,&QWP(16*2,$base));		# s3*h2
+	 &$load		($T0,4);			# r4^n
+	&paddq		($D1,$T1);
+
+	&$load		($T1,8);			# s4^n
+	&pmuludq	($T0,&QWP(16*0,$base));		# r4*h0
+	 &paddq		($D0,$T2);
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&QWP(16*4,$base));		# s4*h4
+	&paddq		($D4,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&QWP(16*1,$base));		# s4*h1
+	&paddq		($D3,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&QWP(16*2,$base));		# s4*h2
+	&paddq		($D0,$T2);
+	&pmuludq	($T1,&QWP(16*3,$base));		# s4*h3
+	 &movdqa	($MASK,&QWP(64,"ebx"));
+	&paddq		($D1,$T0);
+	&paddq		($D2,$T1);
+}
+	&pmuladd	(sub {	my ($reg,$i)=@_;
+				&movdqa ($reg,&QWP(16*$i,"esp"));
+			     },"edx");
+
+sub lazy_reduction {
+	################################################################
+	# lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
+	# and P. Schwabe
+
+	 &movdqa	($T0,$D3);
+	 &pand		($D3,$MASK);
+	 &psrlq		($T0,26);
+	 &paddq		($T0,$D4);			# h3 -> h4
+	&movdqa		($T1,$D0);
+	&pand		($D0,$MASK);
+	&psrlq		($T1,26);
+	 &movdqa	($D4,$T0);
+	&paddq		($T1,$D1);			# h0 -> h1
+	 &psrlq		($T0,26);
+	 &pand		($D4,$MASK);
+	&movdqa		($D1,$T1);
+	&psrlq		($T1,26);
+	 &paddd		($D0,$T0);			# favour paddd when
+							# possible, because
+							# paddq is "broken"
+							# on Atom
+	&pand		($D1,$MASK);
+	&paddq		($T1,$D2);			# h1 -> h2
+	 &psllq		($T0,2);
+	&movdqa		($D2,$T1);
+	&psrlq		($T1,26);
+	 &paddd		($T0,$D0);			# h4 -> h0
+	&pand		($D2,$MASK);
+	&paddd		($T1,$D3);			# h2 -> h3
+	 &movdqa	($D0,$T0);
+	 &psrlq		($T0,26);
+	&movdqa		($D3,$T1);
+	&psrlq		($T1,26);
+	 &pand		($D0,$MASK);
+	 &paddd		($D1,$T0);			# h0 -> h1
+	&pand		($D3,$MASK);
+	&paddd		($D4,$T1);			# h3 -> h4
+}
+	&lazy_reduction	();
+
+	&dec		("ecx");
+	&jz		(&label("square_break"));
+
+	&punpcklqdq	($D0,&QWP(16*0,"esp"));		# 0:r^1:0:r^2
+	&punpcklqdq	($D1,&QWP(16*1,"esp"));
+	&punpcklqdq	($D2,&QWP(16*2,"esp"));
+	&punpcklqdq	($D3,&QWP(16*3,"esp"));
+	&punpcklqdq	($D4,&QWP(16*4,"esp"));
+	&jmp		(&label("square"));
+
+&set_label("square_break");
+	&psllq		($D0,32);			# -> r^3:0:r^4:0
+	&psllq		($D1,32);
+	&psllq		($D2,32);
+	&psllq		($D3,32);
+	&psllq		($D4,32);
+	&por		($D0,&QWP(16*0,"esp"));		# r^3:r^1:r^4:r^2
+	&por		($D1,&QWP(16*1,"esp"));
+	&por		($D2,&QWP(16*2,"esp"));
+	&por		($D3,&QWP(16*3,"esp"));
+	&por		($D4,&QWP(16*4,"esp"));
+
+	&pshufd		($D0,$D0,0b10001101);		# -> r^1:r^2:r^3:r^4
+	&pshufd		($D1,$D1,0b10001101);
+	&pshufd		($D2,$D2,0b10001101);
+	&pshufd		($D3,$D3,0b10001101);
+	&pshufd		($D4,$D4,0b10001101);
+
+	&movdqu		(&QWP(16*0,"edi"),$D0);		# save the table
+	&movdqu		(&QWP(16*1,"edi"),$D1);
+	&movdqu		(&QWP(16*2,"edi"),$D2);
+	&movdqu		(&QWP(16*3,"edi"),$D3);
+	&movdqu		(&QWP(16*4,"edi"),$D4);
+
+	&movdqa		($T1,$D1);
+	&movdqa		($T0,$D2);
+	&pslld		($T1,2);
+	&pslld		($T0,2);
+	&paddd		($T1,$D1);			# *5
+	&paddd		($T0,$D2);			# *5
+	&movdqu		(&QWP(16*5,"edi"),$T1);
+	&movdqu		(&QWP(16*6,"edi"),$T0);
+	&movdqa		($T1,$D3);
+	&movdqa		($T0,$D4);
+	&pslld		($T1,2);
+	&pslld		($T0,2);
+	&paddd		($T1,$D3);			# *5
+	&paddd		($T0,$D4);			# *5
+	&movdqu		(&QWP(16*7,"edi"),$T1);
+	&movdqu		(&QWP(16*8,"edi"),$T0);
+
+	&mov		("esp","ebp");
+	&lea		("edi",&DWP(-16*3,"edi"));	# size de-optimization
+	&ret		();
+&function_end_B("_poly1305_init_sse2");
+
+&align	(32);
+&function_begin("_poly1305_blocks_sse2");
+	&mov	("edi",&wparam(0));			# ctx
+	&mov	("esi",&wparam(1));			# inp
+	&mov	("ecx",&wparam(2));			# len
+
+	&mov	("eax",&DWP(4*5,"edi"));		# is_base2_26
+	&and	("ecx",-16);
+	&jz	(&label("nodata"));
+	&cmp	("ecx",64);
+	&jae	(&label("enter_sse2"));
+	&test	("eax","eax");				# is_base2_26?
+	&jz	(&label("enter_blocks"));
+
+&set_label("enter_sse2",16);
+	&call	(&label("pic_point"));
+&set_label("pic_point");
+	&blindpop("ebx");
+	&lea	("ebx",&DWP(&label("const_sse2")."-".&label("pic_point"),"ebx"));
+
+	&test	("eax","eax");				# is_base2_26?
+	&jnz	(&label("base2_26"));
+
+	&call	("_poly1305_init_sse2");
+
+	################################################# base 2^32 -> base 2^26
+	&mov	("eax",&DWP(0,"edi"));
+	&mov	("ecx",&DWP(3,"edi"));
+	&mov	("edx",&DWP(6,"edi"));
+	&mov	("esi",&DWP(9,"edi"));
+	&mov	("ebp",&DWP(13,"edi"));
+	&mov	(&DWP(4*5,"edi"),1);			# is_base2_26
+
+	&shr	("ecx",2);
+	&and	("eax",0x3ffffff);
+	&shr	("edx",4);
+	&and	("ecx",0x3ffffff);
+	&shr	("esi",6);
+	&and	("edx",0x3ffffff);
+
+	&movd	($D0,"eax");
+	&movd	($D1,"ecx");
+	&movd	($D2,"edx");
+	&movd	($D3,"esi");
+	&movd	($D4,"ebp");
+
+	&mov	("esi",&wparam(1));			# [reload] inp
+	&mov	("ecx",&wparam(2));			# [reload] len
+	&jmp	(&label("base2_32"));
+
+&set_label("base2_26",16);
+	&movd	($D0,&DWP(4*0,"edi"));			# load hash value
+	&movd	($D1,&DWP(4*1,"edi"));
+	&movd	($D2,&DWP(4*2,"edi"));
+	&movd	($D3,&DWP(4*3,"edi"));
+	&movd	($D4,&DWP(4*4,"edi"));
+	&movdqa	($MASK,&QWP(64,"ebx"));
+
+&set_label("base2_32");
+	&mov	("eax",&wparam(3));			# padbit
+	&mov	("ebp","esp");
+
+	&sub	("esp",16*(5+5+5+9+9));
+	&and	("esp",-16);
+
+	&lea	("edi",&DWP(16*3,"edi"));		# size optimization
+	&shl	("eax",24);				# padbit
+
+	&test	("ecx",31);
+	&jz	(&label("even"));
+
+	################################################################
+	# process single block, with SSE2, because it's still faster
+	# even though half of result is discarded
+
+	&movdqu		($T1,&QWP(0,"esi"));		# input
+	&lea		("esi",&DWP(16,"esi"));
+
+	&movdqa		($T0,$T1);			# -> base 2^26 ...
+	&pand		($T1,$MASK);
+	&paddd		($D0,$T1);			# ... and accumuate
+
+	&movdqa		($T1,$T0);
+	&psrlq		($T0,26);
+	&psrldq		($T1,6);
+	&pand		($T0,$MASK);
+	&paddd		($D1,$T0);
+
+	&movdqa		($T0,$T1);
+	&psrlq		($T1,4);
+	&pand		($T1,$MASK);
+	&paddd		($D2,$T1);
+
+	&movdqa		($T1,$T0);
+	&psrlq		($T0,30);
+	&pand		($T0,$MASK);
+	&psrldq		($T1,7);
+	&paddd		($D3,$T0);
+
+	&movd		($T0,"eax");			# padbit
+	&paddd		($D4,$T1);
+	 &movd		($T1,&DWP(16*0+12,"edi"));	# r0
+	&paddd		($D4,$T0);
+
+	&movdqa		(&QWP(16*0,"esp"),$D0);
+	&movdqa		(&QWP(16*1,"esp"),$D1);
+	&movdqa		(&QWP(16*2,"esp"),$D2);
+	&movdqa		(&QWP(16*3,"esp"),$D3);
+	&movdqa		(&QWP(16*4,"esp"),$D4);
+
+	################################################################
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	&pmuludq	($D0,$T1);			# h4*r0
+	&pmuludq	($D1,$T1);			# h3*r0
+	&pmuludq	($D2,$T1);			# h2*r0
+	 &movd		($T0,&DWP(16*1+12,"edi"));	# r1
+	&pmuludq	($D3,$T1);			# h1*r0
+	&pmuludq	($D4,$T1);			# h0*r0
+
+	&pmuladd	(sub {	my ($reg,$i)=@_;
+				&movd ($reg,&DWP(16*$i+12,"edi"));
+			     });
+
+	&lazy_reduction	();
+
+	&sub		("ecx",16);
+	&jz		(&label("done"));
+
+&set_label("even");
+	&lea		("edx",&DWP(16*(5+5+5+9),"esp"));# size optimization
+	&lea		("eax",&DWP(-16*2,"esi"));
+	&sub		("ecx",64);
+
+	################################################################
+	# expand and copy pre-calculated table to stack
+
+	&movdqu		($T0,&QWP(16*0,"edi"));		# r^1:r^2:r^3:r^4
+	&pshufd		($T1,$T0,0b01000100);		# duplicate r^3:r^4
+	&cmovb		("esi","eax");
+	&pshufd		($T0,$T0,0b11101110);		# duplicate r^1:r^2
+	&movdqa		(&QWP(16*0,"edx"),$T1);
+	&lea		("eax",&DWP(16*10,"esp"));
+	&movdqu		($T1,&QWP(16*1,"edi"));
+	&movdqa		(&QWP(16*(0-9),"edx"),$T0);
+	&pshufd		($T0,$T1,0b01000100);
+	&pshufd		($T1,$T1,0b11101110);
+	&movdqa		(&QWP(16*1,"edx"),$T0);
+	&movdqu		($T0,&QWP(16*2,"edi"));
+	&movdqa		(&QWP(16*(1-9),"edx"),$T1);
+	&pshufd		($T1,$T0,0b01000100);
+	&pshufd		($T0,$T0,0b11101110);
+	&movdqa		(&QWP(16*2,"edx"),$T1);
+	&movdqu		($T1,&QWP(16*3,"edi"));
+	&movdqa		(&QWP(16*(2-9),"edx"),$T0);
+	&pshufd		($T0,$T1,0b01000100);
+	&pshufd		($T1,$T1,0b11101110);
+	&movdqa		(&QWP(16*3,"edx"),$T0);
+	&movdqu		($T0,&QWP(16*4,"edi"));
+	&movdqa		(&QWP(16*(3-9),"edx"),$T1);
+	&pshufd		($T1,$T0,0b01000100);
+	&pshufd		($T0,$T0,0b11101110);
+	&movdqa		(&QWP(16*4,"edx"),$T1);
+	&movdqu		($T1,&QWP(16*5,"edi"));
+	&movdqa		(&QWP(16*(4-9),"edx"),$T0);
+	&pshufd		($T0,$T1,0b01000100);
+	&pshufd		($T1,$T1,0b11101110);
+	&movdqa		(&QWP(16*5,"edx"),$T0);
+	&movdqu		($T0,&QWP(16*6,"edi"));
+	&movdqa		(&QWP(16*(5-9),"edx"),$T1);
+	&pshufd		($T1,$T0,0b01000100);
+	&pshufd		($T0,$T0,0b11101110);
+	&movdqa		(&QWP(16*6,"edx"),$T1);
+	&movdqu		($T1,&QWP(16*7,"edi"));
+	&movdqa		(&QWP(16*(6-9),"edx"),$T0);
+	&pshufd		($T0,$T1,0b01000100);
+	&pshufd		($T1,$T1,0b11101110);
+	&movdqa		(&QWP(16*7,"edx"),$T0);
+	&movdqu		($T0,&QWP(16*8,"edi"));
+	&movdqa		(&QWP(16*(7-9),"edx"),$T1);
+	&pshufd		($T1,$T0,0b01000100);
+	&pshufd		($T0,$T0,0b11101110);
+	&movdqa		(&QWP(16*8,"edx"),$T1);
+	&movdqa		(&QWP(16*(8-9),"edx"),$T0);
+
+sub load_input {
+my ($inpbase,$offbase)=@_;
+
+	&movdqu		($T0,&QWP($inpbase+0,"esi"));	# load input
+	&movdqu		($T1,&QWP($inpbase+16,"esi"));
+	&lea		("esi",&DWP(16*2,"esi"));
+
+	&movdqa		(&QWP($offbase+16*2,"esp"),$D2);
+	&movdqa		(&QWP($offbase+16*3,"esp"),$D3);
+	&movdqa		(&QWP($offbase+16*4,"esp"),$D4);
+
+	&movdqa		($D2,$T0);			# splat input
+	&movdqa		($D3,$T1);
+	&psrldq		($D2,6);
+	&psrldq		($D3,6);
+	&movdqa		($D4,$T0);
+	&punpcklqdq	($D2,$D3);			# 2:3
+	&punpckhqdq	($D4,$T1);			# 4
+	&punpcklqdq	($T0,$T1);			# 0:1
+
+	&movdqa		($D3,$D2);
+	&psrlq		($D2,4);
+	&psrlq		($D3,30);
+	&movdqa		($T1,$T0);
+	&psrlq		($D4,40);			# 4
+	&psrlq		($T1,26);
+	&pand		($T0,$MASK);			# 0
+	&pand		($T1,$MASK);			# 1
+	&pand		($D2,$MASK);			# 2
+	&pand		($D3,$MASK);			# 3
+	&por		($D4,&QWP(0,"ebx"));		# padbit, yes, always
+
+	&movdqa		(&QWP($offbase+16*0,"esp"),$D0)	if ($offbase);
+	&movdqa		(&QWP($offbase+16*1,"esp"),$D1)	if ($offbase);
+}
+	&load_input	(16*2,16*5);
+
+	&jbe		(&label("skip_loop"));
+	&jmp		(&label("loop"));
+
+&set_label("loop",32);
+	################################################################
+	# ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
+	# ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
+	#   \___________________/
+	# ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
+	# ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
+	#   \___________________/ \____________________/
+	################################################################
+
+	&movdqa		($T2,&QWP(16*(0-9),"edx"));	# r0^2
+	&movdqa		(&QWP(16*1,"eax"),$T1);
+	&movdqa		(&QWP(16*2,"eax"),$D2);
+	&movdqa		(&QWP(16*3,"eax"),$D3);
+	&movdqa		(&QWP(16*4,"eax"),$D4);
+
+	################################################################
+	# d4 = h4*r0 + h0*r4   + h1*r3   + h2*r2   + h3*r1
+	# d3 = h3*r0 + h0*r3   + h1*r2   + h2*r1   + h4*5*r4
+	# d2 = h2*r0 + h0*r2   + h1*r1   + h3*5*r4 + h4*5*r3
+	# d1 = h1*r0 + h0*r1   + h2*5*r4 + h3*5*r3 + h4*5*r2
+	# d0 = h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
+
+	&movdqa		($D1,$T0);
+	&pmuludq	($T0,$T2);			# h0*r0
+	&movdqa		($D0,$T1);
+	&pmuludq	($T1,$T2);			# h1*r0
+	&pmuludq	($D2,$T2);			# h2*r0
+	&pmuludq	($D3,$T2);			# h3*r0
+	&pmuludq	($D4,$T2);			# h4*r0
+
+sub pmuladd_alt {
+my $addr = shift;
+
+	&pmuludq	($D0,&$addr(8));		# h1*s4
+	&movdqa		($T2,$D1);
+	&pmuludq	($D1,&$addr(1));		# h0*r1
+	&paddq		($D0,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&$addr(2));		# h0*r2
+	&paddq		($D1,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&$addr(3));		# h0*r3
+	&paddq		($D2,$T2);
+	 &movdqa	($T2,&QWP(16*1,"eax"));		# pull h1
+	&pmuludq	($T1,&$addr(4));		# h0*r4
+	&paddq		($D3,$T0);
+
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&$addr(1));		# h1*r1
+	 &paddq		($D4,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&$addr(2));		# h1*r2
+	&paddq		($D2,$T2);
+	&movdqa		($T2,&QWP(16*2,"eax"));		# pull h2
+	&pmuludq	($T1,&$addr(3));		# h1*r3
+	&paddq		($D3,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&$addr(7));		# h2*s3
+	&paddq		($D4,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&$addr(8));		# h2*s4
+	&paddq		($D0,$T2);
+
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&$addr(1));		# h2*r1
+	 &paddq		($D1,$T0);
+	&movdqa		($T0,&QWP(16*3,"eax"));		# pull h3
+	&pmuludq	($T2,&$addr(2));		# h2*r2
+	&paddq		($D3,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&$addr(6));		# h3*s2
+	&paddq		($D4,$T2);
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&$addr(7));		# h3*s3
+	&paddq		($D0,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&$addr(8));		# h3*s4
+	&paddq		($D1,$T1);
+
+	&movdqa		($T1,&QWP(16*4,"eax"));		# pull h4
+	&pmuludq	($T0,&$addr(1));		# h3*r1
+	 &paddq		($D2,$T2);
+	&movdqa		($T2,$T1);
+	&pmuludq	($T1,&$addr(8));		# h4*s4
+	&paddq		($D4,$T0);
+	&movdqa		($T0,$T2);
+	&pmuludq	($T2,&$addr(5));		# h4*s1
+	&paddq		($D3,$T1);
+	&movdqa		($T1,$T0);
+	&pmuludq	($T0,&$addr(6));		# h4*s2
+	&paddq		($D0,$T2);
+	 &movdqa	($MASK,&QWP(64,"ebx"));
+	&pmuludq	($T1,&$addr(7));		# h4*s3
+	&paddq		($D1,$T0);
+	&paddq		($D2,$T1);
+}
+	&pmuladd_alt	(sub {	my $i=shift; &QWP(16*($i-9),"edx");	});
+
+	&load_input	(-16*2,0);
+	&lea		("eax",&DWP(-16*2,"esi"));
+	&sub		("ecx",64);
+
+	&paddd		($T0,&QWP(16*(5+0),"esp"));	# add hash value
+	&paddd		($T1,&QWP(16*(5+1),"esp"));
+	&paddd		($D2,&QWP(16*(5+2),"esp"));
+	&paddd		($D3,&QWP(16*(5+3),"esp"));
+	&paddd		($D4,&QWP(16*(5+4),"esp"));
+
+	&cmovb		("esi","eax");
+	&lea		("eax",&DWP(16*10,"esp"));
+
+	&movdqa		($T2,&QWP(16*0,"edx"));		# r0^4
+	&movdqa		(&QWP(16*1,"esp"),$D1);
+	&movdqa		(&QWP(16*1,"eax"),$T1);
+	&movdqa		(&QWP(16*2,"eax"),$D2);
+	&movdqa		(&QWP(16*3,"eax"),$D3);
+	&movdqa		(&QWP(16*4,"eax"),$D4);
+
+	################################################################
+	# d4 += h4*r0 + h0*r4   + h1*r3   + h2*r2   + h3*r1
+	# d3 += h3*r0 + h0*r3   + h1*r2   + h2*r1   + h4*5*r4
+	# d2 += h2*r0 + h0*r2   + h1*r1   + h3*5*r4 + h4*5*r3
+	# d1 += h1*r0 + h0*r1   + h2*5*r4 + h3*5*r3 + h4*5*r2
+	# d0 += h0*r0 + h1*5*r4 + h2*5*r3 + h3*5*r2 + h4*5*r1
+
+	&movdqa		($D1,$T0);
+	&pmuludq	($T0,$T2);			# h0*r0
+	&paddq		($T0,$D0);
+	&movdqa		($D0,$T1);
+	&pmuludq	($T1,$T2);			# h1*r0
+	&pmuludq	($D2,$T2);			# h2*r0
+	&pmuludq	($D3,$T2);			# h3*r0
+	&pmuludq	($D4,$T2);			# h4*r0
+
+	&paddq		($T1,&QWP(16*1,"esp"));
+	&paddq		($D2,&QWP(16*2,"esp"));
+	&paddq		($D3,&QWP(16*3,"esp"));
+	&paddq		($D4,&QWP(16*4,"esp"));
+
+	&pmuladd_alt	(sub {	my $i=shift; &QWP(16*$i,"edx");	});
+
+	&lazy_reduction	();
+
+	&load_input	(16*2,16*5);
+
+	&ja		(&label("loop"));
+
+&set_label("skip_loop");
+	################################################################
+	# multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
+
+	 &pshufd	($T2,&QWP(16*(0-9),"edx"),0x10);# r0^n
+	&add		("ecx",32);
+	&jnz		(&label("long_tail"));
+
+	&paddd		($T0,$D0);			# add hash value
+	&paddd		($T1,$D1);
+	&paddd		($D2,&QWP(16*7,"esp"));
+	&paddd		($D3,&QWP(16*8,"esp"));
+	&paddd		($D4,&QWP(16*9,"esp"));
+
+&set_label("long_tail");
+
+	&movdqa		(&QWP(16*0,"eax"),$T0);
+	&movdqa		(&QWP(16*1,"eax"),$T1);
+	&movdqa		(&QWP(16*2,"eax"),$D2);
+	&movdqa		(&QWP(16*3,"eax"),$D3);
+	&movdqa		(&QWP(16*4,"eax"),$D4);
+
+	################################################################
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	&pmuludq	($T0,$T2);			# h0*r0
+	&pmuludq	($T1,$T2);			# h1*r0
+	&pmuludq	($D2,$T2);			# h2*r0
+	&movdqa		($D0,$T0);
+	 &pshufd	($T0,&QWP(16*(1-9),"edx"),0x10);# r1^n
+	&pmuludq	($D3,$T2);			# h3*r0
+	&movdqa		($D1,$T1);
+	&pmuludq	($D4,$T2);			# h4*r0
+
+	&pmuladd	(sub {	my ($reg,$i)=@_;
+				&pshufd ($reg,&QWP(16*($i-9),"edx"),0x10);
+			     },"eax");
+
+	&jz		(&label("short_tail"));
+
+	&load_input	(-16*2,0);
+
+	 &pshufd	($T2,&QWP(16*0,"edx"),0x10);	# r0^n
+	&paddd		($T0,&QWP(16*5,"esp"));		# add hash value
+	&paddd		($T1,&QWP(16*6,"esp"));
+	&paddd		($D2,&QWP(16*7,"esp"));
+	&paddd		($D3,&QWP(16*8,"esp"));
+	&paddd		($D4,&QWP(16*9,"esp"));
+
+	################################################################
+	# multiply inp[0:1] by r^4:r^3 and accumulate
+
+	&movdqa		(&QWP(16*0,"esp"),$T0);
+	&pmuludq	($T0,$T2);			# h0*r0
+	&movdqa		(&QWP(16*1,"esp"),$T1);
+	&pmuludq	($T1,$T2);			# h1*r0
+	&paddq		($D0,$T0);
+	&movdqa		($T0,$D2);
+	&pmuludq	($D2,$T2);			# h2*r0
+	&paddq		($D1,$T1);
+	&movdqa		($T1,$D3);
+	&pmuludq	($D3,$T2);			# h3*r0
+	&paddq		($D2,&QWP(16*2,"esp"));
+	&movdqa		(&QWP(16*2,"esp"),$T0);
+	 &pshufd	($T0,&QWP(16*1,"edx"),0x10);	# r1^n
+	&paddq		($D3,&QWP(16*3,"esp"));
+	&movdqa		(&QWP(16*3,"esp"),$T1);
+	&movdqa		($T1,$D4);
+	&pmuludq	($D4,$T2);			# h4*r0
+	&paddq		($D4,&QWP(16*4,"esp"));
+	&movdqa		(&QWP(16*4,"esp"),$T1);
+
+	&pmuladd	(sub {	my ($reg,$i)=@_;
+				&pshufd ($reg,&QWP(16*$i,"edx"),0x10);
+			     });
+
+&set_label("short_tail");
+
+	&lazy_reduction	();
+
+	################################################################
+	# horizontal addition
+
+	&pshufd		($T1,$D0,0b01001110);
+	&pshufd		($T0,$D1,0b01001110);
+	&paddd		($D0,$T1);
+	&pshufd		($T1,$D2,0b01001110);
+	&paddd		($D1,$T0);
+	&pshufd		($T0,$D3,0b01001110);
+	&paddd		($D2,$T1);
+	&pshufd		($T1,$D4,0b01001110);
+	&paddd		($D3,$T0);
+	&paddd		($D4,$T1);
+
+&set_label("done");
+	&movd		(&DWP(-16*3+4*0,"edi"),$D0);	# store hash value
+	&movd		(&DWP(-16*3+4*1,"edi"),$D1);
+	&movd		(&DWP(-16*3+4*2,"edi"),$D2);
+	&movd		(&DWP(-16*3+4*3,"edi"),$D3);
+	&movd		(&DWP(-16*3+4*4,"edi"),$D4);
+&set_label("nodata");
+	&mov	("esp","ebp");
+&function_end("_poly1305_blocks_sse2");
+
+&align	(32);
+&function_begin("_poly1305_emit_sse2");
+	&mov	("ebp",&wparam(0));		# context
+
+	&cmp	(&DWP(4*5,"ebp"),0);		# is_base2_26?
+	&je	(&label("enter_emit"));
+
+	&mov	("eax",&DWP(4*0,"ebp"));	# load hash value
+	&mov	("edi",&DWP(4*1,"ebp"));
+	&mov	("ecx",&DWP(4*2,"ebp"));
+	&mov	("edx",&DWP(4*3,"ebp"));
+	&mov	("esi",&DWP(4*4,"ebp"));
+
+	&mov	("ebx","edi");			# base 2^26 -> base 2^32
+	&shl	("edi",26);
+	&shr	("ebx",6);
+	&add	("eax","edi");
+	&mov	("edi","ecx");
+	&adc	("ebx",0);
+
+	&shl	("edi",20);
+	&shr	("ecx",12);
+	&add	("ebx","edi");
+	&mov	("edi","edx");
+	&adc	("ecx",0);
+
+	&shl	("edi",14);
+	&shr	("edx",18);
+	&add	("ecx","edi");
+	&mov	("edi","esi");
+	&adc	("edx",0);
+
+	&shl	("edi",8);
+	&shr	("esi",24);
+	&add	("edx","edi");
+	&adc	("esi",0);			# can be partially reduced
+
+	&mov	("edi","esi");			# final reduction
+	&and	("esi",3);
+	&shr	("edi",2);
+	&lea	("ebp",&DWP(0,"edi","edi",4));	# *5
+	 &mov	("edi",&wparam(1));		# output
+	add	("eax","ebp");
+	 &mov	("ebp",&wparam(2));		# key
+	adc	("ebx",0);
+	adc	("ecx",0);
+	adc	("edx",0);
+
+	&movd	($D0,"eax");			# offload original hash value
+	&add	("eax",5);			# compare to modulus
+	&movd	($D1,"ebx");
+	&adc	("ebx",0);
+	&movd	($D2,"ecx");
+	&adc	("ecx",0);
+	&movd	($D3,"edx");
+	&adc	("edx",0);
+	&adc	("esi",0);
+	&shr	("esi",2);			# did it carry/borrow?
+
+	&neg	("esi");			# do we choose (hash-modulus) ...
+	&and	("eax","esi");
+	&and	("ebx","esi");
+	&and	("ecx","esi");
+	&and	("edx","esi");
+	&mov	(&DWP(4*0,"edi"),"eax");
+	&movd	("eax",$D0);
+	&mov	(&DWP(4*1,"edi"),"ebx");
+	&movd	("ebx",$D1);
+	&mov	(&DWP(4*2,"edi"),"ecx");
+	&movd	("ecx",$D2);
+	&mov	(&DWP(4*3,"edi"),"edx");
+	&movd	("edx",$D3);
+
+	&not	("esi");			# ... or original hash value?
+	&and	("eax","esi");
+	&and	("ebx","esi");
+	&or	("eax",&DWP(4*0,"edi"));
+	&and	("ecx","esi");
+	&or	("ebx",&DWP(4*1,"edi"));
+	&and	("edx","esi");
+	&or	("ecx",&DWP(4*2,"edi"));
+	&or	("edx",&DWP(4*3,"edi"));
+
+	&add	("eax",&DWP(4*0,"ebp"));	# accumulate key
+	&adc	("ebx",&DWP(4*1,"ebp"));
+	&mov	(&DWP(4*0,"edi"),"eax");
+	&adc	("ecx",&DWP(4*2,"ebp"));
+	&mov	(&DWP(4*1,"edi"),"ebx");
+	&adc	("edx",&DWP(4*3,"ebp"));
+	&mov	(&DWP(4*2,"edi"),"ecx");
+	&mov	(&DWP(4*3,"edi"),"edx");
+&function_end("_poly1305_emit_sse2");
+
+if ($avx>1) {
+########################################################################
+# Note that poly1305_init_avx2 operates on %xmm, I could have used
+# poly1305_init_sse2...
+
+&align	(32);
+&function_begin_B("_poly1305_init_avx2");
+	&vmovdqu	($D4,&QWP(4*6,"edi"));		# key base 2^32
+	&lea		("edi",&DWP(16*3,"edi"));	# size optimization
+	&mov		("ebp","esp");
+	&sub		("esp",16*(9+5));
+	&and		("esp",-16);
+
+	#&vpand		($D4,$D4,&QWP(96,"ebx"));	# magic mask
+	&vmovdqa	($MASK,&QWP(64,"ebx"));
+
+	&vpand		($D0,$D4,$MASK);		# -> base 2^26
+	&vpsrlq		($D1,$D4,26);
+	&vpsrldq	($D3,$D4,6);
+	&vpand		($D1,$D1,$MASK);
+	&vpsrlq		($D2,$D3,4)
+	&vpsrlq		($D3,$D3,30);
+	&vpand		($D2,$D2,$MASK);
+	&vpand		($D3,$D3,$MASK);
+	&vpsrldq	($D4,$D4,13);
+
+	&lea		("edx",&DWP(16*9,"esp"));	# size optimization
+	&mov		("ecx",2);
+&set_label("square");
+	&vmovdqa	(&QWP(16*0,"esp"),$D0);
+	&vmovdqa	(&QWP(16*1,"esp"),$D1);
+	&vmovdqa	(&QWP(16*2,"esp"),$D2);
+	&vmovdqa	(&QWP(16*3,"esp"),$D3);
+	&vmovdqa	(&QWP(16*4,"esp"),$D4);
+
+	&vpslld		($T1,$D1,2);
+	&vpslld		($T0,$D2,2);
+	&vpaddd		($T1,$T1,$D1);			# *5
+	&vpaddd		($T0,$T0,$D2);			# *5
+	&vmovdqa	(&QWP(16*5,"esp"),$T1);
+	&vmovdqa	(&QWP(16*6,"esp"),$T0);
+	&vpslld		($T1,$D3,2);
+	&vpslld		($T0,$D4,2);
+	&vpaddd		($T1,$T1,$D3);			# *5
+	&vpaddd		($T0,$T0,$D4);			# *5
+	&vmovdqa	(&QWP(16*7,"esp"),$T1);
+	&vmovdqa	(&QWP(16*8,"esp"),$T0);
+
+	&vpshufd	($T0,$D0,0b01000100);
+	&vmovdqa	($T1,$D1);
+	&vpshufd	($D1,$D1,0b01000100);
+	&vpshufd	($D2,$D2,0b01000100);
+	&vpshufd	($D3,$D3,0b01000100);
+	&vpshufd	($D4,$D4,0b01000100);
+	&vmovdqa	(&QWP(16*0,"edx"),$T0);
+	&vmovdqa	(&QWP(16*1,"edx"),$D1);
+	&vmovdqa	(&QWP(16*2,"edx"),$D2);
+	&vmovdqa	(&QWP(16*3,"edx"),$D3);
+	&vmovdqa	(&QWP(16*4,"edx"),$D4);
+
+	################################################################
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	&vpmuludq	($D4,$D4,$D0);			# h4*r0
+	&vpmuludq	($D3,$D3,$D0);			# h3*r0
+	&vpmuludq	($D2,$D2,$D0);			# h2*r0
+	&vpmuludq	($D1,$D1,$D0);			# h1*r0
+	&vpmuludq	($D0,$T0,$D0);			# h0*r0
+
+	&vpmuludq	($T0,$T1,&QWP(16*3,"edx"));	# r1*h3
+	&vpaddq		($D4,$D4,$T0);
+	&vpmuludq	($T2,$T1,&QWP(16*2,"edx"));	# r1*h2
+	&vpaddq		($D3,$D3,$T2);
+	&vpmuludq	($T0,$T1,&QWP(16*1,"edx"));	# r1*h1
+	&vpaddq		($D2,$D2,$T0);
+	&vmovdqa	($T2,&QWP(16*5,"esp"));		# s1
+	&vpmuludq	($T1,$T1,&QWP(16*0,"edx"));	# r1*h0
+	&vpaddq		($D1,$D1,$T1);
+	 &vmovdqa	($T0,&QWP(16*2,"esp"));		# r2
+	&vpmuludq	($T2,$T2,&QWP(16*4,"edx"));	# s1*h4
+	&vpaddq		($D0,$D0,$T2);
+
+	&vpmuludq	($T1,$T0,&QWP(16*2,"edx"));	# r2*h2
+	&vpaddq		($D4,$D4,$T1);
+	&vpmuludq	($T2,$T0,&QWP(16*1,"edx"));	# r2*h1
+	&vpaddq		($D3,$D3,$T2);
+	&vmovdqa	($T1,&QWP(16*6,"esp"));		# s2
+	&vpmuludq	($T0,$T0,&QWP(16*0,"edx"));	# r2*h0
+	&vpaddq		($D2,$D2,$T0);
+	&vpmuludq	($T2,$T1,&QWP(16*4,"edx"));	# s2*h4
+	&vpaddq		($D1,$D1,$T2);
+	 &vmovdqa	($T0,&QWP(16*3,"esp"));		# r3
+	&vpmuludq	($T1,$T1,&QWP(16*3,"edx"));	# s2*h3
+	&vpaddq		($D0,$D0,$T1);
+
+	&vpmuludq	($T2,$T0,&QWP(16*1,"edx"));	# r3*h1
+	&vpaddq		($D4,$D4,$T2);
+	&vmovdqa	($T1,&QWP(16*7,"esp"));		# s3
+	&vpmuludq	($T0,$T0,&QWP(16*0,"edx"));	# r3*h0
+	&vpaddq		($D3,$D3,$T0);
+	&vpmuludq	($T2,$T1,&QWP(16*4,"edx"));	# s3*h4
+	&vpaddq		($D2,$D2,$T2);
+	&vpmuludq	($T0,$T1,&QWP(16*3,"edx"));	# s3*h3
+	&vpaddq		($D1,$D1,$T0);
+	 &vmovdqa	($T2,&QWP(16*4,"esp"));		# r4
+	&vpmuludq	($T1,$T1,&QWP(16*2,"edx"));	# s3*h2
+	&vpaddq		($D0,$D0,$T1);
+
+	&vmovdqa	($T0,&QWP(16*8,"esp"));		# s4
+	&vpmuludq	($T2,$T2,&QWP(16*0,"edx"));	# r4*h0
+	&vpaddq		($D4,$D4,$T2);
+	&vpmuludq	($T1,$T0,&QWP(16*4,"edx"));	# s4*h4
+	&vpaddq		($D3,$D3,$T1);
+	&vpmuludq	($T2,$T0,&QWP(16*1,"edx"));	# s4*h1
+	&vpaddq		($D0,$D0,$T2);
+	&vpmuludq	($T1,$T0,&QWP(16*2,"edx"));	# s4*h2
+	&vpaddq		($D1,$D1,$T1);
+	 &vmovdqa	($MASK,&QWP(64,"ebx"));
+	&vpmuludq	($T0,$T0,&QWP(16*3,"edx"));	# s4*h3
+	&vpaddq		($D2,$D2,$T0);
+
+	################################################################
+	# lazy reduction
+	 &vpsrlq	($T0,$D3,26);
+	 &vpand		($D3,$D3,$MASK);
+	&vpsrlq		($T1,$D0,26);
+	&vpand		($D0,$D0,$MASK);
+	 &vpaddq	($D4,$D4,$T0);			# h3 -> h4
+	&vpaddq		($D1,$D1,$T1);			# h0 -> h1
+	 &vpsrlq	($T0,$D4,26);
+	 &vpand		($D4,$D4,$MASK);
+	&vpsrlq		($T1,$D1,26);
+	&vpand		($D1,$D1,$MASK);
+	&vpaddq		($D2,$D2,$T1);			# h1 -> h2
+	 &vpaddd	($D0,$D0,$T0);
+	 &vpsllq	($T0,$T0,2);
+	&vpsrlq		($T1,$D2,26);
+	&vpand		($D2,$D2,$MASK);
+	 &vpaddd	($D0,$D0,$T0);			# h4 -> h0
+	&vpaddd		($D3,$D3,$T1);			# h2 -> h3
+	&vpsrlq		($T1,$D3,26);
+	 &vpsrlq	($T0,$D0,26);
+	 &vpand		($D0,$D0,$MASK);
+	&vpand		($D3,$D3,$MASK);
+	 &vpaddd	($D1,$D1,$T0);			# h0 -> h1
+	&vpaddd		($D4,$D4,$T1);			# h3 -> h4
+
+	&dec		("ecx");
+	&jz		(&label("square_break"));
+
+	&vpunpcklqdq	($D0,$D0,&QWP(16*0,"esp"));	# 0:r^1:0:r^2
+	&vpunpcklqdq	($D1,$D1,&QWP(16*1,"esp"));
+	&vpunpcklqdq	($D2,$D2,&QWP(16*2,"esp"));
+	&vpunpcklqdq	($D3,$D3,&QWP(16*3,"esp"));
+	&vpunpcklqdq	($D4,$D4,&QWP(16*4,"esp"));
+	&jmp		(&label("square"));
+
+&set_label("square_break");
+	&vpsllq		($D0,$D0,32);			# -> r^3:0:r^4:0
+	&vpsllq		($D1,$D1,32);
+	&vpsllq		($D2,$D2,32);
+	&vpsllq		($D3,$D3,32);
+	&vpsllq		($D4,$D4,32);
+	&vpor		($D0,$D0,&QWP(16*0,"esp"));	# r^3:r^1:r^4:r^2
+	&vpor		($D1,$D1,&QWP(16*1,"esp"));
+	&vpor		($D2,$D2,&QWP(16*2,"esp"));
+	&vpor		($D3,$D3,&QWP(16*3,"esp"));
+	&vpor		($D4,$D4,&QWP(16*4,"esp"));
+
+	&vpshufd	($D0,$D0,0b10001101);		# -> r^1:r^2:r^3:r^4
+	&vpshufd	($D1,$D1,0b10001101);
+	&vpshufd	($D2,$D2,0b10001101);
+	&vpshufd	($D3,$D3,0b10001101);
+	&vpshufd	($D4,$D4,0b10001101);
+
+	&vmovdqu	(&QWP(16*0,"edi"),$D0);		# save the table
+	&vmovdqu	(&QWP(16*1,"edi"),$D1);
+	&vmovdqu	(&QWP(16*2,"edi"),$D2);
+	&vmovdqu	(&QWP(16*3,"edi"),$D3);
+	&vmovdqu	(&QWP(16*4,"edi"),$D4);
+
+	&vpslld		($T1,$D1,2);
+	&vpslld		($T0,$D2,2);
+	&vpaddd		($T1,$T1,$D1);			# *5
+	&vpaddd		($T0,$T0,$D2);			# *5
+	&vmovdqu	(&QWP(16*5,"edi"),$T1);
+	&vmovdqu	(&QWP(16*6,"edi"),$T0);
+	&vpslld		($T1,$D3,2);
+	&vpslld		($T0,$D4,2);
+	&vpaddd		($T1,$T1,$D3);			# *5
+	&vpaddd		($T0,$T0,$D4);			# *5
+	&vmovdqu	(&QWP(16*7,"edi"),$T1);
+	&vmovdqu	(&QWP(16*8,"edi"),$T0);
+
+	&mov		("esp","ebp");
+	&lea		("edi",&DWP(-16*3,"edi"));	# size de-optimization
+	&ret		();
+&function_end_B("_poly1305_init_avx2");
+
+########################################################################
+# now it's time to switch to %ymm
+
+my ($D0,$D1,$D2,$D3,$D4,$T0,$T1,$T2)=map("ymm$_",(0..7));
+my $MASK=$T2;
+
+sub X { my $reg=shift; $reg=~s/^ymm/xmm/; $reg; }
+
+&align	(32);
+&function_begin("_poly1305_blocks_avx2");
+	&mov	("edi",&wparam(0));			# ctx
+	&mov	("esi",&wparam(1));			# inp
+	&mov	("ecx",&wparam(2));			# len
+
+	&mov	("eax",&DWP(4*5,"edi"));		# is_base2_26
+	&and	("ecx",-16);
+	&jz	(&label("nodata"));
+	&cmp	("ecx",64);
+	&jae	(&label("enter_avx2"));
+	&test	("eax","eax");				# is_base2_26?
+	&jz	(&label("enter_blocks"));
+
+&set_label("enter_avx2",16);
+	&vzeroupper	();
+
+	&call	(&label("pic_point"));
+&set_label("pic_point");
+	&blindpop("ebx");
+	&lea	("ebx",&DWP(&label("const_sse2")."-".&label("pic_point"),"ebx"));
+
+	&test	("eax","eax");				# is_base2_26?
+	&jnz	(&label("base2_26"));
+
+	&call	("_poly1305_init_avx2");
+
+	################################################# base 2^32 -> base 2^26
+	&mov	("eax",&DWP(0,"edi"));
+	&mov	("ecx",&DWP(3,"edi"));
+	&mov	("edx",&DWP(6,"edi"));
+	&mov	("esi",&DWP(9,"edi"));
+	&mov	("ebp",&DWP(13,"edi"));
+
+	&shr	("ecx",2);
+	&and	("eax",0x3ffffff);
+	&shr	("edx",4);
+	&and	("ecx",0x3ffffff);
+	&shr	("esi",6);
+	&and	("edx",0x3ffffff);
+
+	&mov	(&DWP(4*0,"edi"),"eax");
+	&mov	(&DWP(4*1,"edi"),"ecx");
+	&mov	(&DWP(4*2,"edi"),"edx");
+	&mov	(&DWP(4*3,"edi"),"esi");
+	&mov	(&DWP(4*4,"edi"),"ebp");
+	&mov	(&DWP(4*5,"edi"),1);			# is_base2_26
+
+	&mov	("esi",&wparam(1));			# [reload] inp
+	&mov	("ecx",&wparam(2));			# [reload] len
+
+&set_label("base2_26");
+	&mov	("eax",&wparam(3));			# padbit
+	&mov	("ebp","esp");
+
+	&sub	("esp",32*(5+9));
+	&and	("esp",-512);				# ensure that frame
+							# doesn't cross page
+							# boundary, which is
+							# essential for
+							# misaligned 32-byte
+							# loads
+
+	################################################################
+        # expand and copy pre-calculated table to stack
+
+	&vmovdqu	(&X($D0),&QWP(16*(3+0),"edi"));
+	&lea		("edx",&DWP(32*5+128,"esp"));	# +128 size optimization
+	&vmovdqu	(&X($D1),&QWP(16*(3+1),"edi"));
+	&vmovdqu	(&X($D2),&QWP(16*(3+2),"edi"));
+	&vmovdqu	(&X($D3),&QWP(16*(3+3),"edi"));
+	&vmovdqu	(&X($D4),&QWP(16*(3+4),"edi"));
+	&lea		("edi",&DWP(16*3,"edi"));	# size optimization
+	&vpermq		($D0,$D0,0b01000000);		# 00001234 -> 12343434
+	&vpermq		($D1,$D1,0b01000000);
+	&vpermq		($D2,$D2,0b01000000);
+	&vpermq		($D3,$D3,0b01000000);
+	&vpermq		($D4,$D4,0b01000000);
+	&vpshufd	($D0,$D0,0b11001000);		# 12343434 -> 14243444
+	&vpshufd	($D1,$D1,0b11001000);
+	&vpshufd	($D2,$D2,0b11001000);
+	&vpshufd	($D3,$D3,0b11001000);
+	&vpshufd	($D4,$D4,0b11001000);
+	&vmovdqa	(&QWP(32*0-128,"edx"),$D0);
+	&vmovdqu	(&X($D0),&QWP(16*5,"edi"));
+	&vmovdqa	(&QWP(32*1-128,"edx"),$D1);
+	&vmovdqu	(&X($D1),&QWP(16*6,"edi"));
+	&vmovdqa	(&QWP(32*2-128,"edx"),$D2);
+	&vmovdqu	(&X($D2),&QWP(16*7,"edi"));
+	&vmovdqa	(&QWP(32*3-128,"edx"),$D3);
+	&vmovdqu	(&X($D3),&QWP(16*8,"edi"));
+	&vmovdqa	(&QWP(32*4-128,"edx"),$D4);
+	&vpermq		($D0,$D0,0b01000000);
+	&vpermq		($D1,$D1,0b01000000);
+	&vpermq		($D2,$D2,0b01000000);
+	&vpermq		($D3,$D3,0b01000000);
+	&vpshufd	($D0,$D0,0b11001000);
+	&vpshufd	($D1,$D1,0b11001000);
+	&vpshufd	($D2,$D2,0b11001000);
+	&vpshufd	($D3,$D3,0b11001000);
+	&vmovdqa	(&QWP(32*5-128,"edx"),$D0);
+	&vmovd		(&X($D0),&DWP(-16*3+4*0,"edi"));# load hash value
+	&vmovdqa	(&QWP(32*6-128,"edx"),$D1);
+	&vmovd		(&X($D1),&DWP(-16*3+4*1,"edi"));
+	&vmovdqa	(&QWP(32*7-128,"edx"),$D2);
+	&vmovd		(&X($D2),&DWP(-16*3+4*2,"edi"));
+	&vmovdqa	(&QWP(32*8-128,"edx"),$D3);
+	&vmovd		(&X($D3),&DWP(-16*3+4*3,"edi"));
+	&vmovd		(&X($D4),&DWP(-16*3+4*4,"edi"));
+	&vmovdqa	($MASK,&QWP(64,"ebx"));
+	&neg		("eax");			# padbit
+
+	&test		("ecx",63);
+	&jz		(&label("even"));
+
+	&mov		("edx","ecx");
+	&and		("ecx",-64);
+	&and		("edx",63);
+
+	&vmovdqu	(&X($T0),&QWP(16*0,"esi"));
+	&cmp		("edx",32);
+	&jb		(&label("one"));
+
+	&vmovdqu	(&X($T1),&QWP(16*1,"esi"));
+	&je		(&label("two"));
+
+	&vinserti128	($T0,$T0,&QWP(16*2,"esi"),1);
+	&lea		("esi",&DWP(16*3,"esi"));
+	&lea		("ebx",&DWP(8,"ebx"));		# three padbits
+	&lea		("edx",&DWP(32*5+128+8,"esp"));	# --:r^1:r^2:r^3 (*)
+	&jmp		(&label("tail"));
+
+&set_label("two");
+	&lea		("esi",&DWP(16*2,"esi"));
+	&lea		("ebx",&DWP(16,"ebx"));		# two padbits
+	&lea		("edx",&DWP(32*5+128+16,"esp"));# --:--:r^1:r^2 (*)
+	&jmp		(&label("tail"));
+
+&set_label("one");
+	&lea		("esi",&DWP(16*1,"esi"));
+	&vpxor		($T1,$T1,$T1);
+	&lea		("ebx",&DWP(32,"ebx","eax",8));	# one or no padbits
+	&lea		("edx",&DWP(32*5+128+24,"esp"));# --:--:--:r^1 (*)
+	&jmp		(&label("tail"));
+
+# (*)	spots marked with '--' are data from next table entry, but they
+#	are multiplied by 0 and therefore rendered insignificant
+
+&set_label("even",32);
+	&vmovdqu	(&X($T0),&QWP(16*0,"esi"));	# load input
+	&vmovdqu	(&X($T1),&QWP(16*1,"esi"));
+	&vinserti128	($T0,$T0,&QWP(16*2,"esi"),1);
+	&vinserti128	($T1,$T1,&QWP(16*3,"esi"),1);
+	&lea		("esi",&DWP(16*4,"esi"));
+	&sub		("ecx",64);
+	&jz		(&label("tail"));
+
+&set_label("loop");
+	################################################################
+	# ((inp[0]*r^4+r[4])*r^4+r[8])*r^4
+	# ((inp[1]*r^4+r[5])*r^4+r[9])*r^3
+	# ((inp[2]*r^4+r[6])*r^4+r[10])*r^2
+	# ((inp[3]*r^4+r[7])*r^4+r[11])*r^1
+	#   \________/ \_______/
+	################################################################
+
+sub vsplat_input {
+	&vmovdqa	(&QWP(32*2,"esp"),$D2);
+	&vpsrldq	($D2,$T0,6);			# splat input
+	&vmovdqa	(&QWP(32*0,"esp"),$D0);
+	&vpsrldq	($D0,$T1,6);
+	&vmovdqa	(&QWP(32*1,"esp"),$D1);
+	&vpunpckhqdq	($D1,$T0,$T1);			# 4
+	&vpunpcklqdq	($T0,$T0,$T1);			# 0:1
+	&vpunpcklqdq	($D2,$D2,$D0);			# 2:3
+
+	&vpsrlq		($D0,$D2,30);
+	&vpsrlq		($D2,$D2,4);
+	&vpsrlq		($T1,$T0,26);
+	&vpsrlq		($D1,$D1,40);			# 4
+	&vpand		($D2,$D2,$MASK);		# 2
+	&vpand		($T0,$T0,$MASK);		# 0
+	&vpand		($T1,$T1,$MASK);		# 1
+	&vpand		($D0,$D0,$MASK);		# 3 (*)
+	&vpor		($D1,$D1,&QWP(0,"ebx"));	# padbit, yes, always
+
+	# (*)	note that output is counterintuitive, inp[3:4] is
+	#	returned in $D1-2, while $D3-4 are preserved;
+}
+	&vsplat_input	();
+
+sub vpmuladd {
+my $addr = shift;
+
+	&vpaddq		($D2,$D2,&QWP(32*2,"esp"));	# add hash value
+	&vpaddq		($T0,$T0,&QWP(32*0,"esp"));
+	&vpaddq		($T1,$T1,&QWP(32*1,"esp"));
+	&vpaddq		($D0,$D0,$D3);
+	&vpaddq		($D1,$D1,$D4);
+
+	################################################################
+	# d3 = h2*r1   + h0*r3 + h1*r2   + h3*r0   + h4*5*r4
+	# d4 = h2*r2   + h0*r4 + h1*r3   + h3*r1   + h4*r0
+	# d0 = h2*5*r3 + h0*r0 + h1*5*r4 + h3*5*r2 + h4*5*r1
+	# d1 = h2*5*r4 + h0*r1 + h1*r0   + h3*5*r3 + h4*5*r2
+	# d2 = h2*r0   + h0*r2 + h1*r1   + h3*5*r4 + h4*5*r3
+
+	&vpmuludq	($D3,$D2,&$addr(1));		# d3 = h2*r1
+	 &vmovdqa	(QWP(32*1,"esp"),$T1);
+	&vpmuludq	($D4,$D2,&$addr(2));		# d4 = h2*r2
+	 &vmovdqa	(QWP(32*3,"esp"),$D0);
+	&vpmuludq	($D0,$D2,&$addr(7));		# d0 = h2*s3
+	 &vmovdqa	(QWP(32*4,"esp"),$D1);
+	&vpmuludq	($D1,$D2,&$addr(8));		# d1 = h2*s4
+	&vpmuludq	($D2,$D2,&$addr(0));		# d2 = h2*r0
+
+	&vpmuludq	($T2,$T0,&$addr(3));		# h0*r3
+	&vpaddq		($D3,$D3,$T2);			# d3 += h0*r3
+	&vpmuludq	($T1,$T0,&$addr(4));		# h0*r4
+	&vpaddq		($D4,$D4,$T1);			# d4 + h0*r4
+	&vpmuludq	($T2,$T0,&$addr(0));		# h0*r0
+	&vpaddq		($D0,$D0,$T2);			# d0 + h0*r0
+	 &vmovdqa	($T2,&QWP(32*1,"esp"));		# h1
+	&vpmuludq	($T1,$T0,&$addr(1));		# h0*r1
+	&vpaddq		($D1,$D1,$T1);			# d1 += h0*r1
+	&vpmuludq	($T0,$T0,&$addr(2));		# h0*r2
+	&vpaddq		($D2,$D2,$T0);			# d2 += h0*r2
+
+	&vpmuludq	($T1,$T2,&$addr(2));		# h1*r2
+	&vpaddq		($D3,$D3,$T1);			# d3 += h1*r2
+	&vpmuludq	($T0,$T2,&$addr(3));		# h1*r3
+	&vpaddq		($D4,$D4,$T0);			# d4 += h1*r3
+	&vpmuludq	($T1,$T2,&$addr(8));		# h1*s4
+	&vpaddq		($D0,$D0,$T1);			# d0 += h1*s4
+	 &vmovdqa	($T1,&QWP(32*3,"esp"));		# h3
+	&vpmuludq	($T0,$T2,&$addr(0));		# h1*r0
+	&vpaddq		($D1,$D1,$T0);			# d1 += h1*r0
+	&vpmuludq	($T2,$T2,&$addr(1));		# h1*r1
+	&vpaddq		($D2,$D2,$T2);			# d2 += h1*r1
+
+	&vpmuludq	($T0,$T1,&$addr(0));		# h3*r0
+	&vpaddq		($D3,$D3,$T0);			# d3 += h3*r0
+	&vpmuludq	($T2,$T1,&$addr(1));		# h3*r1
+	&vpaddq		($D4,$D4,$T2);			# d4 += h3*r1
+	&vpmuludq	($T0,$T1,&$addr(6));		# h3*s2
+	&vpaddq		($D0,$D0,$T0);			# d0 += h3*s2
+	 &vmovdqa	($T0,&QWP(32*4,"esp"));		# h4
+	&vpmuludq	($T2,$T1,&$addr(7));		# h3*s3
+	&vpaddq		($D1,$D1,$T2);			# d1+= h3*s3
+	&vpmuludq	($T1,$T1,&$addr(8));		# h3*s4
+	&vpaddq		($D2,$D2,$T1);			# d2 += h3*s4
+
+	&vpmuludq	($T2,$T0,&$addr(8));		# h4*s4
+	&vpaddq		($D3,$D3,$T2);			# d3 += h4*s4
+	&vpmuludq	($T1,$T0,&$addr(5));		# h4*s1
+	&vpaddq		($D0,$D0,$T1);			# d0 += h4*s1
+	&vpmuludq	($T2,$T0,&$addr(0));		# h4*r0
+	&vpaddq		($D4,$D4,$T2);			# d4 += h4*r0
+	 &vmovdqa	($MASK,&QWP(64,"ebx"));
+	&vpmuludq	($T1,$T0,&$addr(6));		# h4*s2
+	&vpaddq		($D1,$D1,$T1);			# d1 += h4*s2
+	&vpmuludq	($T0,$T0,&$addr(7));		# h4*s3
+	&vpaddq		($D2,$D2,$T0);			# d2 += h4*s3
+}
+	&vpmuladd	(sub {	my $i=shift; &QWP(32*$i-128,"edx");	});
+
+sub vlazy_reduction {
+	################################################################
+	# lazy reduction
+
+	 &vpsrlq	($T0,$D3,26);
+	 &vpand		($D3,$D3,$MASK);
+	&vpsrlq		($T1,$D0,26);
+	&vpand		($D0,$D0,$MASK);
+	 &vpaddq	($D4,$D4,$T0);			# h3 -> h4
+	&vpaddq		($D1,$D1,$T1);			# h0 -> h1
+	 &vpsrlq	($T0,$D4,26);
+	 &vpand		($D4,$D4,$MASK);
+	&vpsrlq		($T1,$D1,26);
+	&vpand		($D1,$D1,$MASK);
+	&vpaddq		($D2,$D2,$T1);			# h1 -> h2
+	 &vpaddd	($D0,$D0,$T0);
+	 &vpsllq	($T0,$T0,2);
+	&vpsrlq		($T1,$D2,26);
+	&vpand		($D2,$D2,$MASK);
+	 &vpaddd	($D0,$D0,$T0);			# h4 -> h0
+	&vpaddd		($D3,$D3,$T1);			# h2 -> h3
+	&vpsrlq		($T1,$D3,26);
+	 &vpsrlq	($T0,$D0,26);
+	 &vpand		($D0,$D0,$MASK);
+	&vpand		($D3,$D3,$MASK);
+	 &vpaddd	($D1,$D1,$T0);			# h0 -> h1
+	&vpaddd		($D4,$D4,$T1);			# h3 -> h4
+}
+	&vlazy_reduction();
+
+	&vmovdqu	(&X($T0),&QWP(16*0,"esi"));	# load input
+	&vmovdqu	(&X($T1),&QWP(16*1,"esi"));
+	&vinserti128	($T0,$T0,&QWP(16*2,"esi"),1);
+	&vinserti128	($T1,$T1,&QWP(16*3,"esi"),1);
+	&lea		("esi",&DWP(16*4,"esi"));
+	&sub		("ecx",64);
+	&jnz		(&label("loop"));
+
+&set_label("tail");
+	&vsplat_input	();
+	&and		("ebx",-64);			# restore pointer
+
+	&vpmuladd	(sub {	my $i=shift; &QWP(4+32*$i-128,"edx");	});
+
+	&vlazy_reduction();
+
+	################################################################
+	# horizontal addition
+
+	&vpsrldq	($T0,$D0,8);
+	&vpsrldq	($T1,$D1,8);
+	&vpaddq		($D0,$D0,$T0);
+	&vpsrldq	($T0,$D2,8);
+	&vpaddq		($D1,$D1,$T1);
+	&vpsrldq	($T1,$D3,8);
+	&vpaddq		($D2,$D2,$T0);
+	&vpsrldq	($T0,$D4,8);
+	&vpaddq		($D3,$D3,$T1);
+	&vpermq		($T1,$D0,2);			# keep folding
+	&vpaddq		($D4,$D4,$T0);
+	&vpermq		($T0,$D1,2);
+	&vpaddq		($D0,$D0,$T1);
+	&vpermq		($T1,$D2,2);
+	&vpaddq		($D1,$D1,$T0);
+	&vpermq		($T0,$D3,2);
+	&vpaddq		($D2,$D2,$T1);
+	&vpermq		($T1,$D4,2);
+	&vpaddq		($D3,$D3,$T0);
+	&vpaddq		($D4,$D4,$T1);
+
+	&cmp		("ecx",0);
+	&je		(&label("done"));
+
+	################################################################
+	# clear all but single word
+
+	&vpshufd	(&X($D0),&X($D0),0b11111100);
+	&lea		("edx",&DWP(32*5+128,"esp"));	# restore pointer
+	&vpshufd	(&X($D1),&X($D1),0b11111100);
+	&vpshufd	(&X($D2),&X($D2),0b11111100);
+	&vpshufd	(&X($D3),&X($D3),0b11111100);
+	&vpshufd	(&X($D4),&X($D4),0b11111100);
+	&jmp		(&label("even"));
+
+&set_label("done",16);
+	&vmovd		(&DWP(-16*3+4*0,"edi"),"xmm0");	# store hash value
+	&vmovd		(&DWP(-16*3+4*1,"edi"),"xmm1");
+	&vmovd		(&DWP(-16*3+4*2,"edi"),"xmm2");
+	&vmovd		(&DWP(-16*3+4*3,"edi"),"xmm3");
+	&vmovd		(&DWP(-16*3+4*4,"edi"),"xmm4");
+	&vzeroupper	();
+&set_label("nodata");
+	&mov	("esp","ebp");
+&function_end("_poly1305_blocks_avx2");
+}
+&set_label("const_sse2",64);
+	&data_word(1<<24,0,	1<<24,0,	1<<24,0,	1<<24,0);
+	&data_word(0,0,		0,0,		0,0,		0,0);
+	&data_word(0x03ffffff,0,0x03ffffff,0,	0x03ffffff,0,	0x03ffffff,0);
+	&data_word(0x0fffffff,0x0ffffffc,0x0ffffffc,0x0ffffffc);
+}
+&asciz	("Poly1305 for x86, CRYPTOGAMS by <appro\@openssl.org>");
+&align	(4);
+
+&asm_finish();
diff --git a/crypto/poly1305/asm/poly1305-x86_64.pl b/crypto/poly1305/asm/poly1305-x86_64.pl
new file mode 100755
index 00000000..b827d24b
--- /dev/null
+++ b/crypto/poly1305/asm/poly1305-x86_64.pl
@@ -0,0 +1,2246 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements Poly1305 hash for x86_64.
+#
+# March 2015
+#
+# Numbers are cycles per processed byte with poly1305_blocks alone,
+# measured with rdtsc at fixed clock frequency.
+#
+#		IALU/gcc-4.8(*)	AVX(**)		AVX2
+# P4		4.90/+120%      -
+# Core 2	2.39/+90%	-
+# Westmere	1.86/+120%	-
+# Sandy Bridge	1.39/+140%	1.10
+# Haswell	1.10/+175%	1.11		0.65
+# Skylake	1.12/+120%	0.96		0.51
+# Silvermont	2.83/+95%	-
+# VIA Nano	1.82/+150%	-
+# Sledgehammer	1.38/+160%	-
+# Bulldozer	2.21/+130%	0.97
+#
+# (*)	improvement coefficients relative to clang are more modest and
+#	are ~50% on most processors, in both cases we are comparing to
+#	__int128 code;
+# (**)	SSE2 implementation was attempted, but among non-AVX processors
+#	it was faster than integer-only code only on older Intel P4 and
+#	Core processors, 50-30%, less newer processor is, but slower on
+#	contemporary ones, for example almost 2x slower on Atom, and as
+#	former are naturally disappearing, SSE2 is deemed unnecessary;
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
+		=~ /GNU assembler version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.19) + ($1>=2.22);
+}
+
+if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
+	   `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
+	$avx = ($1>=2.09) + ($1>=2.10);
+}
+
+if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
+	   `ml64 2>&1` =~ /Version ([0-9]+)\./) {
+	$avx = ($1>=10) + ($1>=12);
+}
+
+if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
+	$avx = ($2>=3.0) + ($2>3.0);
+}
+
+open OUT,"| \"$^X\" $xlate $flavour $output";
+*STDOUT=*OUT;
+
+my ($ctx,$inp,$len,$padbit)=("%rdi","%rsi","%rdx","%rcx");
+my ($mac,$nonce)=($inp,$len);	# *_emit arguments
+my ($d1,$d2,$d3, $r0,$r1,$s1)=map("%r$_",(8..13));
+my ($h0,$h1,$h2)=("%r14","%rbx","%rbp");
+
+sub poly1305_iteration {
+# input:	copy of $r1 in %rax, $h0-$h2, $r0-$r1
+# output:	$h0-$h2 *= $r0-$r1
+$code.=<<___;
+	mulq	$h0			# h0*r1
+	mov	%rax,$d2
+	 mov	$r0,%rax
+	mov	%rdx,$d3
+
+	mulq	$h0			# h0*r0
+	mov	%rax,$h0		# future $h0
+	 mov	$r0,%rax
+	mov	%rdx,$d1
+
+	mulq	$h1			# h1*r0
+	add	%rax,$d2
+	 mov	$s1,%rax
+	adc	%rdx,$d3
+
+	mulq	$h1			# h1*s1
+	 mov	$h2,$h1			# borrow $h1
+	add	%rax,$h0
+	adc	%rdx,$d1
+
+	imulq	$s1,$h1			# h2*s1
+	add	$h1,$d2
+	 mov	$d1,$h1
+	adc	\$0,$d3
+
+	imulq	$r0,$h2			# h2*r0
+	add	$d2,$h1
+	mov	\$-4,%rax		# mask value
+	adc	$h2,$d3
+
+	and	$d3,%rax		# last reduction step
+	mov	$d3,$h2
+	shr	\$2,$d3
+	and	\$3,$h2
+	add	$d3,%rax
+	add	%rax,$h0
+	adc	\$0,$h1
+___
+}
+
+########################################################################
+# Layout of opaque area is following.
+#
+#	unsigned __int64 h[3];		# current hash value base 2^64
+#	unsigned __int64 r[2];		# key value base 2^64
+
+$code.=<<___;
+.text
+
+.extern	OPENSSL_ia32cap_P
+
+.globl	poly1305_init
+.globl	poly1305_blocks
+.globl	poly1305_emit
+.type	poly1305_init,\@function,3
+.align	32
+poly1305_init:
+	xor	%rax,%rax
+	mov	%rax,0($ctx)		# initialize hash value
+	mov	%rax,8($ctx)
+	mov	%rax,16($ctx)
+
+	cmp	\$0,$inp
+	je	.Lno_key
+
+	lea	poly1305_blocks(%rip),%r10
+	lea	poly1305_emit(%rip),%r11
+___
+$code.=<<___	if ($avx);
+	mov	OPENSSL_ia32cap_P+4(%rip),%r9
+	lea	poly1305_blocks_avx(%rip),%rax
+	lea	poly1305_emit_avx(%rip),%rcx
+	bt	\$`60-32`,%r9		# AVX?
+	cmovc	%rax,%r10
+	cmovc	%rcx,%r11
+___
+$code.=<<___	if ($avx>1);
+	lea	poly1305_blocks_avx2(%rip),%rax
+	bt	\$`5+32`,%r9		# AVX2?
+	cmovc	%rax,%r10
+___
+$code.=<<___;
+	mov	\$0x0ffffffc0fffffff,%rax
+	mov	\$0x0ffffffc0ffffffc,%rcx
+	and	0($inp),%rax
+	and	8($inp),%rcx
+	mov	%rax,24($ctx)
+	mov	%rcx,32($ctx)
+
+	mov	%r10,0(%rdx)
+	mov	%r11,8(%rdx)
+
+	mov	\$1,%eax
+.Lno_key:
+	ret
+.size	poly1305_init,.-poly1305_init
+
+.type	poly1305_blocks,\@function,4
+.align	32
+poly1305_blocks:
+.Lblocks:
+	sub	\$16,$len		# too short?
+	jc	.Lno_data
+
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+.Lblocks_body:
+
+	mov	$len,%r15		# reassign $len
+
+	mov	24($ctx),$r0		# load r
+	mov	32($ctx),$s1
+
+	mov	0($ctx),$h0		# load hash value
+	mov	8($ctx),$h1
+	mov	16($ctx),$h2
+
+	mov	$s1,$r1
+	shr	\$2,$s1
+	mov	$r1,%rax
+	add	$r1,$s1			# s1 = r1 + (r1 >> 2)
+	jmp	.Loop
+
+.align	32
+.Loop:
+	add	0($inp),$h0		# accumulate input
+	adc	8($inp),$h1
+	lea	16($inp),$inp
+	adc	$padbit,$h2
+___
+	&poly1305_iteration();
+$code.=<<___;
+	mov	$r1,%rax
+	sub	\$16,%r15		# len-=16
+	jnc	.Loop
+
+	mov	$h0,0($ctx)		# store hash value
+	mov	$h1,8($ctx)
+	mov	$h2,16($ctx)
+
+	mov	0(%rsp),%r15
+	mov	8(%rsp),%r14
+	mov	16(%rsp),%r13
+	mov	24(%rsp),%r12
+	mov	32(%rsp),%rbp
+	mov	40(%rsp),%rbx
+	lea	48(%rsp),%rsp
+.Lno_data:
+.Lblocks_epilogue:
+	ret
+.size	poly1305_blocks,.-poly1305_blocks
+
+.type	poly1305_emit,\@function,3
+.align	32
+poly1305_emit:
+.Lemit:
+	mov	0($ctx),%r8	# load hash value
+	mov	8($ctx),%r9
+	mov	16($ctx),%r10
+
+	mov	%r8,%rax
+	add	\$5,%r8		# compare to modulus
+	mov	%r9,%rcx
+	adc	\$0,%r9
+	adc	\$0,%r10
+	shr	\$2,%r10	# did 130-bit value overfow?
+	cmovnz	%r8,%rax
+	cmovnz	%r9,%rcx
+
+	add	0($nonce),%rax	# accumulate nonce
+	adc	8($nonce),%rcx
+	mov	%rax,0($mac)	# write result
+	mov	%rcx,8($mac)
+
+	ret
+.size	poly1305_emit,.-poly1305_emit
+___
+if ($avx) {
+
+########################################################################
+# Layout of opaque area is following.
+#
+#	unsigned __int32 h[5];		# current hash value base 2^26
+#	unsigned __int32 is_base2_26;
+#	unsigned __int64 r[2];		# key value base 2^64
+#	unsigned __int64 pad;
+#	struct { unsigned __int32 r^2, r^1, r^4, r^3; } r[9];
+#
+# where r^n are base 2^26 digits of degrees of multiplier key. There are
+# 5 digits, but last four are interleaved with multiples of 5, totalling
+# in 9 elements: r0, r1, 5*r1, r2, 5*r2, r3, 5*r3, r4, 5*r4.
+
+my ($H0,$H1,$H2,$H3,$H4, $T0,$T1,$T2,$T3,$T4, $D0,$D1,$D2,$D3,$D4, $MASK) =
+    map("%xmm$_",(0..15));
+
+$code.=<<___;
+.type	__poly1305_block,\@abi-omnipotent
+.align	32
+__poly1305_block:
+___
+	&poly1305_iteration();
+$code.=<<___;
+	ret
+.size	__poly1305_block,.-__poly1305_block
+
+.type	__poly1305_init_avx,\@abi-omnipotent
+.align	32
+__poly1305_init_avx:
+	mov	$r0,$h0
+	mov	$r1,$h1
+	xor	$h2,$h2
+
+	lea	48+64($ctx),$ctx	# size optimization
+
+	mov	$r1,%rax
+	call	__poly1305_block	# r^2
+
+	mov	\$0x3ffffff,%eax	# save interleaved r^2 and r base 2^26
+	mov	\$0x3ffffff,%edx
+	mov	$h0,$d1
+	and	$h0#d,%eax
+	mov	$r0,$d2
+	and	$r0#d,%edx
+	mov	%eax,`16*0+0-64`($ctx)
+	shr	\$26,$d1
+	mov	%edx,`16*0+4-64`($ctx)
+	shr	\$26,$d2
+
+	mov	\$0x3ffffff,%eax
+	mov	\$0x3ffffff,%edx
+	and	$d1#d,%eax
+	and	$d2#d,%edx
+	mov	%eax,`16*1+0-64`($ctx)
+	lea	(%rax,%rax,4),%eax	# *5
+	mov	%edx,`16*1+4-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	mov	%eax,`16*2+0-64`($ctx)
+	shr	\$26,$d1
+	mov	%edx,`16*2+4-64`($ctx)
+	shr	\$26,$d2
+
+	mov	$h1,%rax
+	mov	$r1,%rdx
+	shl	\$12,%rax
+	shl	\$12,%rdx
+	or	$d1,%rax
+	or	$d2,%rdx
+	and	\$0x3ffffff,%eax
+	and	\$0x3ffffff,%edx
+	mov	%eax,`16*3+0-64`($ctx)
+	lea	(%rax,%rax,4),%eax	# *5
+	mov	%edx,`16*3+4-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	mov	%eax,`16*4+0-64`($ctx)
+	mov	$h1,$d1
+	mov	%edx,`16*4+4-64`($ctx)
+	mov	$r1,$d2
+
+	mov	\$0x3ffffff,%eax
+	mov	\$0x3ffffff,%edx
+	shr	\$14,$d1
+	shr	\$14,$d2
+	and	$d1#d,%eax
+	and	$d2#d,%edx
+	mov	%eax,`16*5+0-64`($ctx)
+	lea	(%rax,%rax,4),%eax	# *5
+	mov	%edx,`16*5+4-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	mov	%eax,`16*6+0-64`($ctx)
+	shr	\$26,$d1
+	mov	%edx,`16*6+4-64`($ctx)
+	shr	\$26,$d2
+
+	mov	$h2,%rax
+	shl	\$24,%rax
+	or	%rax,$d1
+	mov	$d1#d,`16*7+0-64`($ctx)
+	lea	($d1,$d1,4),$d1		# *5
+	mov	$d2#d,`16*7+4-64`($ctx)
+	lea	($d2,$d2,4),$d2		# *5
+	mov	$d1#d,`16*8+0-64`($ctx)
+	mov	$d2#d,`16*8+4-64`($ctx)
+
+	mov	$r1,%rax
+	call	__poly1305_block	# r^3
+
+	mov	\$0x3ffffff,%eax	# save r^3 base 2^26
+	mov	$h0,$d1
+	and	$h0#d,%eax
+	shr	\$26,$d1
+	mov	%eax,`16*0+12-64`($ctx)
+
+	mov	\$0x3ffffff,%edx
+	and	$d1#d,%edx
+	mov	%edx,`16*1+12-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	shr	\$26,$d1
+	mov	%edx,`16*2+12-64`($ctx)
+
+	mov	$h1,%rax
+	shl	\$12,%rax
+	or	$d1,%rax
+	and	\$0x3ffffff,%eax
+	mov	%eax,`16*3+12-64`($ctx)
+	lea	(%rax,%rax,4),%eax	# *5
+	mov	$h1,$d1
+	mov	%eax,`16*4+12-64`($ctx)
+
+	mov	\$0x3ffffff,%edx
+	shr	\$14,$d1
+	and	$d1#d,%edx
+	mov	%edx,`16*5+12-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	shr	\$26,$d1
+	mov	%edx,`16*6+12-64`($ctx)
+
+	mov	$h2,%rax
+	shl	\$24,%rax
+	or	%rax,$d1
+	mov	$d1#d,`16*7+12-64`($ctx)
+	lea	($d1,$d1,4),$d1		# *5
+	mov	$d1#d,`16*8+12-64`($ctx)
+
+	mov	$r1,%rax
+	call	__poly1305_block	# r^4
+
+	mov	\$0x3ffffff,%eax	# save r^4 base 2^26
+	mov	$h0,$d1
+	and	$h0#d,%eax
+	shr	\$26,$d1
+	mov	%eax,`16*0+8-64`($ctx)
+
+	mov	\$0x3ffffff,%edx
+	and	$d1#d,%edx
+	mov	%edx,`16*1+8-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	shr	\$26,$d1
+	mov	%edx,`16*2+8-64`($ctx)
+
+	mov	$h1,%rax
+	shl	\$12,%rax
+	or	$d1,%rax
+	and	\$0x3ffffff,%eax
+	mov	%eax,`16*3+8-64`($ctx)
+	lea	(%rax,%rax,4),%eax	# *5
+	mov	$h1,$d1
+	mov	%eax,`16*4+8-64`($ctx)
+
+	mov	\$0x3ffffff,%edx
+	shr	\$14,$d1
+	and	$d1#d,%edx
+	mov	%edx,`16*5+8-64`($ctx)
+	lea	(%rdx,%rdx,4),%edx	# *5
+	shr	\$26,$d1
+	mov	%edx,`16*6+8-64`($ctx)
+
+	mov	$h2,%rax
+	shl	\$24,%rax
+	or	%rax,$d1
+	mov	$d1#d,`16*7+8-64`($ctx)
+	lea	($d1,$d1,4),$d1		# *5
+	mov	$d1#d,`16*8+8-64`($ctx)
+
+	lea	-48-64($ctx),$ctx	# size [de-]optimization
+	ret
+.size	__poly1305_init_avx,.-__poly1305_init_avx
+
+.type	poly1305_blocks_avx,\@function,4
+.align	32
+poly1305_blocks_avx:
+	mov	20($ctx),%r8d		# is_base2_26
+	cmp	\$128,$len
+	jae	.Lblocks_avx
+	test	%r8d,%r8d
+	jz	.Lblocks
+
+.Lblocks_avx:
+	and	\$-16,$len
+	jz	.Lno_data_avx
+
+	vzeroupper
+
+	test	%r8d,%r8d
+	jz	.Lbase2_64_avx
+
+	test	\$31,$len
+	jz	.Leven_avx
+
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+.Lblocks_avx_body:
+
+	mov	$len,%r15		# reassign $len
+
+	mov	0($ctx),$d1		# load hash value
+	mov	8($ctx),$d2
+	mov	16($ctx),$h2#d
+
+	mov	24($ctx),$r0		# load r
+	mov	32($ctx),$s1
+
+	################################# base 2^26 -> base 2^64
+	mov	$d1#d,$h0#d
+	and	\$-1<<31,$d1
+	mov	$d2,$r1			# borrow $r1
+	mov	$d2#d,$h1#d
+	and	\$-1<<31,$d2
+
+	shr	\$6,$d1
+	shl	\$52,$r1
+	add	$d1,$h0
+	shr	\$12,$h1
+	shr	\$18,$d2
+	add	$r1,$h0
+	adc	$d2,$h1
+
+	mov	$h2,$d1
+	shl	\$40,$d1
+	shr	\$24,$h2
+	add	$d1,$h1
+	adc	\$0,$h2			# can be partially reduced...
+
+	mov	\$-4,$d2		# ... so reduce
+	mov	$h2,$d1
+	and	$h2,$d2
+	shr	\$2,$d1
+	and	\$3,$h2
+	add	$d2,$d1			# =*5
+	add	$d1,$h0
+	adc	\$0,$h1
+
+	mov	$s1,$r1
+	mov	$s1,%rax
+	shr	\$2,$s1
+	add	$r1,$s1			# s1 = r1 + (r1 >> 2)
+
+	add	0($inp),$h0		# accumulate input
+	adc	8($inp),$h1
+	lea	16($inp),$inp
+	adc	$padbit,$h2
+
+	call	__poly1305_block
+
+	test	$padbit,$padbit		# if $padbit is zero,
+	jz	.Lstore_base2_64_avx	# store hash in base 2^64 format
+
+	################################# base 2^64 -> base 2^26
+	mov	$h0,%rax
+	mov	$h0,%rdx
+	shr	\$52,$h0
+	mov	$h1,$r0
+	mov	$h1,$r1
+	shr	\$26,%rdx
+	and	\$0x3ffffff,%rax	# h[0]
+	shl	\$12,$r0
+	and	\$0x3ffffff,%rdx	# h[1]
+	shr	\$14,$h1
+	or	$r0,$h0
+	shl	\$24,$h2
+	and	\$0x3ffffff,$h0		# h[2]
+	shr	\$40,$r1
+	and	\$0x3ffffff,$h1		# h[3]
+	or	$r1,$h2			# h[4]
+
+	sub	\$16,%r15
+	jz	.Lstore_base2_26_avx
+
+	vmovd	%rax#d,$H0
+	vmovd	%rdx#d,$H1
+	vmovd	$h0#d,$H2
+	vmovd	$h1#d,$H3
+	vmovd	$h2#d,$H4
+	jmp	.Lproceed_avx
+
+.align	32
+.Lstore_base2_64_avx:
+	mov	$h0,0($ctx)
+	mov	$h1,8($ctx)
+	mov	$h2,16($ctx)		# note that is_base2_26 is zeroed
+	jmp	.Ldone_avx
+
+.align	16
+.Lstore_base2_26_avx:
+	mov	%rax#d,0($ctx)		# store hash value base 2^26
+	mov	%rdx#d,4($ctx)
+	mov	$h0#d,8($ctx)
+	mov	$h1#d,12($ctx)
+	mov	$h2#d,16($ctx)
+.align	16
+.Ldone_avx:
+	mov	0(%rsp),%r15
+	mov	8(%rsp),%r14
+	mov	16(%rsp),%r13
+	mov	24(%rsp),%r12
+	mov	32(%rsp),%rbp
+	mov	40(%rsp),%rbx
+	lea	48(%rsp),%rsp
+.Lno_data_avx:
+.Lblocks_avx_epilogue:
+	ret
+
+.align	32
+.Lbase2_64_avx:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+.Lbase2_64_avx_body:
+
+	mov	$len,%r15		# reassign $len
+
+	mov	24($ctx),$r0		# load r
+	mov	32($ctx),$s1
+
+	mov	0($ctx),$h0		# load hash value
+	mov	8($ctx),$h1
+	mov	16($ctx),$h2#d
+
+	mov	$s1,$r1
+	mov	$s1,%rax
+	shr	\$2,$s1
+	add	$r1,$s1			# s1 = r1 + (r1 >> 2)
+
+	test	\$31,$len
+	jz	.Linit_avx
+
+	add	0($inp),$h0		# accumulate input
+	adc	8($inp),$h1
+	lea	16($inp),$inp
+	adc	$padbit,$h2
+	sub	\$16,%r15
+
+	call	__poly1305_block
+
+.Linit_avx:
+	################################# base 2^64 -> base 2^26
+	mov	$h0,%rax
+	mov	$h0,%rdx
+	shr	\$52,$h0
+	mov	$h1,$d1
+	mov	$h1,$d2
+	shr	\$26,%rdx
+	and	\$0x3ffffff,%rax	# h[0]
+	shl	\$12,$d1
+	and	\$0x3ffffff,%rdx	# h[1]
+	shr	\$14,$h1
+	or	$d1,$h0
+	shl	\$24,$h2
+	and	\$0x3ffffff,$h0		# h[2]
+	shr	\$40,$d2
+	and	\$0x3ffffff,$h1		# h[3]
+	or	$d2,$h2			# h[4]
+
+	vmovd	%rax#d,$H0
+	vmovd	%rdx#d,$H1
+	vmovd	$h0#d,$H2
+	vmovd	$h1#d,$H3
+	vmovd	$h2#d,$H4
+	movl	\$1,20($ctx)		# set is_base2_26
+
+	call	__poly1305_init_avx
+
+.Lproceed_avx:
+	mov	%r15,$len
+
+	mov	0(%rsp),%r15
+	mov	8(%rsp),%r14
+	mov	16(%rsp),%r13
+	mov	24(%rsp),%r12
+	mov	32(%rsp),%rbp
+	mov	40(%rsp),%rbx
+	lea	48(%rsp),%rax
+	lea	48(%rsp),%rsp
+.Lbase2_64_avx_epilogue:
+	jmp	.Ldo_avx
+
+.align	32
+.Leven_avx:
+	vmovd		4*0($ctx),$H0		# load hash value
+	vmovd		4*1($ctx),$H1
+	vmovd		4*2($ctx),$H2
+	vmovd		4*3($ctx),$H3
+	vmovd		4*4($ctx),$H4
+
+.Ldo_avx:
+___
+$code.=<<___	if (!$win64);
+	lea		-0x58(%rsp),%r11
+	sub		\$0x178,%rsp
+___
+$code.=<<___	if ($win64);
+	lea		-0xf8(%rsp),%r11
+	sub		\$0x218,%rsp
+	vmovdqa		%xmm6,0x50(%r11)
+	vmovdqa		%xmm7,0x60(%r11)
+	vmovdqa		%xmm8,0x70(%r11)
+	vmovdqa		%xmm9,0x80(%r11)
+	vmovdqa		%xmm10,0x90(%r11)
+	vmovdqa		%xmm11,0xa0(%r11)
+	vmovdqa		%xmm12,0xb0(%r11)
+	vmovdqa		%xmm13,0xc0(%r11)
+	vmovdqa		%xmm14,0xd0(%r11)
+	vmovdqa		%xmm15,0xe0(%r11)
+.Ldo_avx_body:
+___
+$code.=<<___;
+	sub		\$64,$len
+	lea		-32($inp),%rax
+	cmovc		%rax,$inp
+
+	vmovdqu		`16*3`($ctx),$D4	# preload r0^2
+	lea		`16*3+64`($ctx),$ctx	# size optimization
+	lea		.Lconst(%rip),%rcx
+
+	################################################################
+	# load input
+	vmovdqu		16*2($inp),$T0
+	vmovdqu		16*3($inp),$T1
+	vmovdqa		64(%rcx),$MASK		# .Lmask26
+
+	vpsrldq		\$6,$T0,$T2		# splat input
+	vpsrldq		\$6,$T1,$T3
+	vpunpckhqdq	$T1,$T0,$T4		# 4
+	vpunpcklqdq	$T1,$T0,$T0		# 0:1
+	vpunpcklqdq	$T3,$T2,$T3		# 2:3
+
+	vpsrlq		\$40,$T4,$T4		# 4
+	vpsrlq		\$26,$T0,$T1
+	vpand		$MASK,$T0,$T0		# 0
+	vpsrlq		\$4,$T3,$T2
+	vpand		$MASK,$T1,$T1		# 1
+	vpsrlq		\$30,$T3,$T3
+	vpand		$MASK,$T2,$T2		# 2
+	vpand		$MASK,$T3,$T3		# 3
+	vpor		32(%rcx),$T4,$T4	# padbit, yes, always
+
+	jbe		.Lskip_loop_avx
+
+	# expand and copy pre-calculated table to stack
+	vmovdqu		`16*1-64`($ctx),$D1
+	vmovdqu		`16*2-64`($ctx),$D2
+	vpshufd		\$0xEE,$D4,$D3		# 34xx -> 3434
+	vpshufd		\$0x44,$D4,$D0		# xx12 -> 1212
+	vmovdqa		$D3,-0x90(%r11)
+	vmovdqa		$D0,0x00(%rsp)
+	vpshufd		\$0xEE,$D1,$D4
+	vmovdqu		`16*3-64`($ctx),$D0
+	vpshufd		\$0x44,$D1,$D1
+	vmovdqa		$D4,-0x80(%r11)
+	vmovdqa		$D1,0x10(%rsp)
+	vpshufd		\$0xEE,$D2,$D3
+	vmovdqu		`16*4-64`($ctx),$D1
+	vpshufd		\$0x44,$D2,$D2
+	vmovdqa		$D3,-0x70(%r11)
+	vmovdqa		$D2,0x20(%rsp)
+	vpshufd		\$0xEE,$D0,$D4
+	vmovdqu		`16*5-64`($ctx),$D2
+	vpshufd		\$0x44,$D0,$D0
+	vmovdqa		$D4,-0x60(%r11)
+	vmovdqa		$D0,0x30(%rsp)
+	vpshufd		\$0xEE,$D1,$D3
+	vmovdqu		`16*6-64`($ctx),$D0
+	vpshufd		\$0x44,$D1,$D1
+	vmovdqa		$D3,-0x50(%r11)
+	vmovdqa		$D1,0x40(%rsp)
+	vpshufd		\$0xEE,$D2,$D4
+	vmovdqu		`16*7-64`($ctx),$D1
+	vpshufd		\$0x44,$D2,$D2
+	vmovdqa		$D4,-0x40(%r11)
+	vmovdqa		$D2,0x50(%rsp)
+	vpshufd		\$0xEE,$D0,$D3
+	vmovdqu		`16*8-64`($ctx),$D2
+	vpshufd		\$0x44,$D0,$D0
+	vmovdqa		$D3,-0x30(%r11)
+	vmovdqa		$D0,0x60(%rsp)
+	vpshufd		\$0xEE,$D1,$D4
+	vpshufd		\$0x44,$D1,$D1
+	vmovdqa		$D4,-0x20(%r11)
+	vmovdqa		$D1,0x70(%rsp)
+	vpshufd		\$0xEE,$D2,$D3
+	 vmovdqa	0x00(%rsp),$D4		# preload r0^2
+	vpshufd		\$0x44,$D2,$D2
+	vmovdqa		$D3,-0x10(%r11)
+	vmovdqa		$D2,0x80(%rsp)
+
+	jmp		.Loop_avx
+
+.align	32
+.Loop_avx:
+	################################################################
+	# ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
+	# ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
+	#   \___________________/
+	# ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
+	# ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
+	#   \___________________/ \____________________/
+	#
+	# Note that we start with inp[2:3]*r^2. This is because it
+	# doesn't depend on reduction in previous iteration.
+	################################################################
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+	#
+	# though note that $Tx and $Hx are "reversed" in this section,
+	# and $D4 is preloaded with r0^2...
+
+	vpmuludq	$T0,$D4,$D0		# d0 = h0*r0
+	vpmuludq	$T1,$D4,$D1		# d1 = h1*r0
+	  vmovdqa	$H2,0x20(%r11)				# offload hash
+	vpmuludq	$T2,$D4,$D2		# d3 = h2*r0
+	 vmovdqa	0x10(%rsp),$H2		# r1^2
+	vpmuludq	$T3,$D4,$D3		# d3 = h3*r0
+	vpmuludq	$T4,$D4,$D4		# d4 = h4*r0
+
+	  vmovdqa	$H0,0x00(%r11)				#
+	vpmuludq	0x20(%rsp),$T4,$H0	# h4*s1
+	  vmovdqa	$H1,0x10(%r11)				#
+	vpmuludq	$T3,$H2,$H1		# h3*r1
+	vpaddq		$H0,$D0,$D0		# d0 += h4*s1
+	vpaddq		$H1,$D4,$D4		# d4 += h3*r1
+	  vmovdqa	$H3,0x30(%r11)				#
+	vpmuludq	$T2,$H2,$H0		# h2*r1
+	vpmuludq	$T1,$H2,$H1		# h1*r1
+	vpaddq		$H0,$D3,$D3		# d3 += h2*r1
+	 vmovdqa	0x30(%rsp),$H3		# r2^2
+	vpaddq		$H1,$D2,$D2		# d2 += h1*r1
+	  vmovdqa	$H4,0x40(%r11)				#
+	vpmuludq	$T0,$H2,$H2		# h0*r1
+	 vpmuludq	$T2,$H3,$H0		# h2*r2
+	vpaddq		$H2,$D1,$D1		# d1 += h0*r1
+
+	 vmovdqa	0x40(%rsp),$H4		# s2^2
+	vpaddq		$H0,$D4,$D4		# d4 += h2*r2
+	vpmuludq	$T1,$H3,$H1		# h1*r2
+	vpmuludq	$T0,$H3,$H3		# h0*r2
+	vpaddq		$H1,$D3,$D3		# d3 += h1*r2
+	 vmovdqa	0x50(%rsp),$H2		# r3^2
+	vpaddq		$H3,$D2,$D2		# d2 += h0*r2
+	vpmuludq	$T4,$H4,$H0		# h4*s2
+	vpmuludq	$T3,$H4,$H4		# h3*s2
+	vpaddq		$H0,$D1,$D1		# d1 += h4*s2
+	 vmovdqa	0x60(%rsp),$H3		# s3^2
+	vpaddq		$H4,$D0,$D0		# d0 += h3*s2
+
+	 vmovdqa	0x80(%rsp),$H4		# s4^2
+	vpmuludq	$T1,$H2,$H1		# h1*r3
+	vpmuludq	$T0,$H2,$H2		# h0*r3
+	vpaddq		$H1,$D4,$D4		# d4 += h1*r3
+	vpaddq		$H2,$D3,$D3		# d3 += h0*r3
+	vpmuludq	$T4,$H3,$H0		# h4*s3
+	vpmuludq	$T3,$H3,$H1		# h3*s3
+	vpaddq		$H0,$D2,$D2		# d2 += h4*s3
+	 vmovdqu	16*0($inp),$H0				# load input
+	vpaddq		$H1,$D1,$D1		# d1 += h3*s3
+	vpmuludq	$T2,$H3,$H3		# h2*s3
+	 vpmuludq	$T2,$H4,$T2		# h2*s4
+	vpaddq		$H3,$D0,$D0		# d0 += h2*s3
+
+	 vmovdqu	16*1($inp),$H1				#
+	vpaddq		$T2,$D1,$D1		# d1 += h2*s4
+	vpmuludq	$T3,$H4,$T3		# h3*s4
+	vpmuludq	$T4,$H4,$T4		# h4*s4
+	 vpsrldq	\$6,$H0,$H2				# splat input
+	vpaddq		$T3,$D2,$D2		# d2 += h3*s4
+	vpaddq		$T4,$D3,$D3		# d3 += h4*s4
+	 vpsrldq	\$6,$H1,$H3				#
+	vpmuludq	0x70(%rsp),$T0,$T4	# h0*r4
+	vpmuludq	$T1,$H4,$T0		# h1*s4
+	 vpunpckhqdq	$H1,$H0,$H4		# 4
+	vpaddq		$T4,$D4,$D4		# d4 += h0*r4
+	 vmovdqa	-0x90(%r11),$T4		# r0^4
+	vpaddq		$T0,$D0,$D0		# d0 += h1*s4
+
+	vpunpcklqdq	$H1,$H0,$H0		# 0:1
+	vpunpcklqdq	$H3,$H2,$H3		# 2:3
+
+	#vpsrlq		\$40,$H4,$H4		# 4
+	vpsrldq		\$`40/8`,$H4,$H4	# 4
+	vpsrlq		\$26,$H0,$H1
+	vpand		$MASK,$H0,$H0		# 0
+	vpsrlq		\$4,$H3,$H2
+	vpand		$MASK,$H1,$H1		# 1
+	vpand		0(%rcx),$H4,$H4		# .Lmask24
+	vpsrlq		\$30,$H3,$H3
+	vpand		$MASK,$H2,$H2		# 2
+	vpand		$MASK,$H3,$H3		# 3
+	vpor		32(%rcx),$H4,$H4	# padbit, yes, always
+
+	vpaddq		0x00(%r11),$H0,$H0	# add hash value
+	vpaddq		0x10(%r11),$H1,$H1
+	vpaddq		0x20(%r11),$H2,$H2
+	vpaddq		0x30(%r11),$H3,$H3
+	vpaddq		0x40(%r11),$H4,$H4
+
+	lea		16*2($inp),%rax
+	lea		16*4($inp),$inp
+	sub		\$64,$len
+	cmovc		%rax,$inp
+
+	################################################################
+	# Now we accumulate (inp[0:1]+hash)*r^4
+	################################################################
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	vpmuludq	$H0,$T4,$T0		# h0*r0
+	vpmuludq	$H1,$T4,$T1		# h1*r0
+	vpaddq		$T0,$D0,$D0
+	vpaddq		$T1,$D1,$D1
+	 vmovdqa	-0x80(%r11),$T2		# r1^4
+	vpmuludq	$H2,$T4,$T0		# h2*r0
+	vpmuludq	$H3,$T4,$T1		# h3*r0
+	vpaddq		$T0,$D2,$D2
+	vpaddq		$T1,$D3,$D3
+	vpmuludq	$H4,$T4,$T4		# h4*r0
+	 vpmuludq	-0x70(%r11),$H4,$T0	# h4*s1
+	vpaddq		$T4,$D4,$D4
+
+	vpaddq		$T0,$D0,$D0		# d0 += h4*s1
+	vpmuludq	$H2,$T2,$T1		# h2*r1
+	vpmuludq	$H3,$T2,$T0		# h3*r1
+	vpaddq		$T1,$D3,$D3		# d3 += h2*r1
+	 vmovdqa	-0x60(%r11),$T3		# r2^4
+	vpaddq		$T0,$D4,$D4		# d4 += h3*r1
+	vpmuludq	$H1,$T2,$T1		# h1*r1
+	vpmuludq	$H0,$T2,$T2		# h0*r1
+	vpaddq		$T1,$D2,$D2		# d2 += h1*r1
+	vpaddq		$T2,$D1,$D1		# d1 += h0*r1
+
+	 vmovdqa	-0x50(%r11),$T4		# s2^4
+	vpmuludq	$H2,$T3,$T0		# h2*r2
+	vpmuludq	$H1,$T3,$T1		# h1*r2
+	vpaddq		$T0,$D4,$D4		# d4 += h2*r2
+	vpaddq		$T1,$D3,$D3		# d3 += h1*r2
+	 vmovdqa	-0x40(%r11),$T2		# r3^4
+	vpmuludq	$H0,$T3,$T3		# h0*r2
+	vpmuludq	$H4,$T4,$T0		# h4*s2
+	vpaddq		$T3,$D2,$D2		# d2 += h0*r2
+	vpaddq		$T0,$D1,$D1		# d1 += h4*s2
+	 vmovdqa	-0x30(%r11),$T3		# s3^4
+	vpmuludq	$H3,$T4,$T4		# h3*s2
+	 vpmuludq	$H1,$T2,$T1		# h1*r3
+	vpaddq		$T4,$D0,$D0		# d0 += h3*s2
+
+	 vmovdqa	-0x10(%r11),$T4		# s4^4
+	vpaddq		$T1,$D4,$D4		# d4 += h1*r3
+	vpmuludq	$H0,$T2,$T2		# h0*r3
+	vpmuludq	$H4,$T3,$T0		# h4*s3
+	vpaddq		$T2,$D3,$D3		# d3 += h0*r3
+	vpaddq		$T0,$D2,$D2		# d2 += h4*s3
+	 vmovdqu	16*2($inp),$T0				# load input
+	vpmuludq	$H3,$T3,$T2		# h3*s3
+	vpmuludq	$H2,$T3,$T3		# h2*s3
+	vpaddq		$T2,$D1,$D1		# d1 += h3*s3
+	 vmovdqu	16*3($inp),$T1				#
+	vpaddq		$T3,$D0,$D0		# d0 += h2*s3
+
+	vpmuludq	$H2,$T4,$H2		# h2*s4
+	vpmuludq	$H3,$T4,$H3		# h3*s4
+	 vpsrldq	\$6,$T0,$T2				# splat input
+	vpaddq		$H2,$D1,$D1		# d1 += h2*s4
+	vpmuludq	$H4,$T4,$H4		# h4*s4
+	 vpsrldq	\$6,$T1,$T3				#
+	vpaddq		$H3,$D2,$H2		# h2 = d2 + h3*s4
+	vpaddq		$H4,$D3,$H3		# h3 = d3 + h4*s4
+	vpmuludq	-0x20(%r11),$H0,$H4	# h0*r4
+	vpmuludq	$H1,$T4,$H0
+	 vpunpckhqdq	$T1,$T0,$T4		# 4
+	vpaddq		$H4,$D4,$H4		# h4 = d4 + h0*r4
+	vpaddq		$H0,$D0,$H0		# h0 = d0 + h1*s4
+
+	vpunpcklqdq	$T1,$T0,$T0		# 0:1
+	vpunpcklqdq	$T3,$T2,$T3		# 2:3
+
+	#vpsrlq		\$40,$T4,$T4		# 4
+	vpsrldq		\$`40/8`,$T4,$T4	# 4
+	vpsrlq		\$26,$T0,$T1
+	 vmovdqa	0x00(%rsp),$D4		# preload r0^2
+	vpand		$MASK,$T0,$T0		# 0
+	vpsrlq		\$4,$T3,$T2
+	vpand		$MASK,$T1,$T1		# 1
+	vpand		0(%rcx),$T4,$T4		# .Lmask24
+	vpsrlq		\$30,$T3,$T3
+	vpand		$MASK,$T2,$T2		# 2
+	vpand		$MASK,$T3,$T3		# 3
+	vpor		32(%rcx),$T4,$T4	# padbit, yes, always
+
+	################################################################
+	# lazy reduction as discussed in "NEON crypto" by D.J. Bernstein
+	# and P. Schwabe
+
+	vpsrlq		\$26,$H3,$D3
+	vpand		$MASK,$H3,$H3
+	vpaddq		$D3,$H4,$H4		# h3 -> h4
+
+	vpsrlq		\$26,$H0,$D0
+	vpand		$MASK,$H0,$H0
+	vpaddq		$D0,$D1,$H1		# h0 -> h1
+
+	vpsrlq		\$26,$H4,$D0
+	vpand		$MASK,$H4,$H4
+
+	vpsrlq		\$26,$H1,$D1
+	vpand		$MASK,$H1,$H1
+	vpaddq		$D1,$H2,$H2		# h1 -> h2
+
+	vpaddq		$D0,$H0,$H0
+	vpsllq		\$2,$D0,$D0
+	vpaddq		$D0,$H0,$H0		# h4 -> h0
+
+	vpsrlq		\$26,$H2,$D2
+	vpand		$MASK,$H2,$H2
+	vpaddq		$D2,$H3,$H3		# h2 -> h3
+
+	vpsrlq		\$26,$H0,$D0
+	vpand		$MASK,$H0,$H0
+	vpaddq		$D0,$H1,$H1		# h0 -> h1
+
+	vpsrlq		\$26,$H3,$D3
+	vpand		$MASK,$H3,$H3
+	vpaddq		$D3,$H4,$H4		# h3 -> h4
+
+	ja		.Loop_avx
+
+.Lskip_loop_avx:
+	################################################################
+	# multiply (inp[0:1]+hash) or inp[2:3] by r^2:r^1
+
+	vpshufd		\$0x10,$D4,$D4		# r0^n, xx12 -> x1x2
+	add		\$32,$len
+	jnz		.Long_tail_avx
+
+	vpaddq		$H2,$T2,$T2
+	vpaddq		$H0,$T0,$T0
+	vpaddq		$H1,$T1,$T1
+	vpaddq		$H3,$T3,$T3
+	vpaddq		$H4,$T4,$T4
+
+.Long_tail_avx:
+	vmovdqa		$H2,0x20(%r11)
+	vmovdqa		$H0,0x00(%r11)
+	vmovdqa		$H1,0x10(%r11)
+	vmovdqa		$H3,0x30(%r11)
+	vmovdqa		$H4,0x40(%r11)
+
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+
+	vpmuludq	$T2,$D4,$D2		# d2 = h2*r0
+	vpmuludq	$T0,$D4,$D0		# d0 = h0*r0
+	 vpshufd	\$0x10,`16*1-64`($ctx),$H2		# r1^n
+	vpmuludq	$T1,$D4,$D1		# d1 = h1*r0
+	vpmuludq	$T3,$D4,$D3		# d3 = h3*r0
+	vpmuludq	$T4,$D4,$D4		# d4 = h4*r0
+
+	vpmuludq	$T3,$H2,$H0		# h3*r1
+	vpaddq		$H0,$D4,$D4		# d4 += h3*r1
+	 vpshufd	\$0x10,`16*2-64`($ctx),$H3		# s1^n
+	vpmuludq	$T2,$H2,$H1		# h2*r1
+	vpaddq		$H1,$D3,$D3		# d3 += h2*r1
+	 vpshufd	\$0x10,`16*3-64`($ctx),$H4		# r2^n
+	vpmuludq	$T1,$H2,$H0		# h1*r1
+	vpaddq		$H0,$D2,$D2		# d2 += h1*r1
+	vpmuludq	$T0,$H2,$H2		# h0*r1
+	vpaddq		$H2,$D1,$D1		# d1 += h0*r1
+	vpmuludq	$T4,$H3,$H3		# h4*s1
+	vpaddq		$H3,$D0,$D0		# d0 += h4*s1
+
+	 vpshufd	\$0x10,`16*4-64`($ctx),$H2		# s2^n
+	vpmuludq	$T2,$H4,$H1		# h2*r2
+	vpaddq		$H1,$D4,$D4		# d4 += h2*r2
+	vpmuludq	$T1,$H4,$H0		# h1*r2
+	vpaddq		$H0,$D3,$D3		# d3 += h1*r2
+	 vpshufd	\$0x10,`16*5-64`($ctx),$H3		# r3^n
+	vpmuludq	$T0,$H4,$H4		# h0*r2
+	vpaddq		$H4,$D2,$D2		# d2 += h0*r2
+	vpmuludq	$T4,$H2,$H1		# h4*s2
+	vpaddq		$H1,$D1,$D1		# d1 += h4*s2
+	 vpshufd	\$0x10,`16*6-64`($ctx),$H4		# s3^n
+	vpmuludq	$T3,$H2,$H2		# h3*s2
+	vpaddq		$H2,$D0,$D0		# d0 += h3*s2
+
+	vpmuludq	$T1,$H3,$H0		# h1*r3
+	vpaddq		$H0,$D4,$D4		# d4 += h1*r3
+	vpmuludq	$T0,$H3,$H3		# h0*r3
+	vpaddq		$H3,$D3,$D3		# d3 += h0*r3
+	 vpshufd	\$0x10,`16*7-64`($ctx),$H2		# r4^n
+	vpmuludq	$T4,$H4,$H1		# h4*s3
+	vpaddq		$H1,$D2,$D2		# d2 += h4*s3
+	 vpshufd	\$0x10,`16*8-64`($ctx),$H3		# s4^n
+	vpmuludq	$T3,$H4,$H0		# h3*s3
+	vpaddq		$H0,$D1,$D1		# d1 += h3*s3
+	vpmuludq	$T2,$H4,$H4		# h2*s3
+	vpaddq		$H4,$D0,$D0		# d0 += h2*s3
+
+	vpmuludq	$T0,$H2,$H2		# h0*r4
+	vpaddq		$H2,$D4,$D4		# h4 = d4 + h0*r4
+	vpmuludq	$T4,$H3,$H1		# h4*s4
+	vpaddq		$H1,$D3,$D3		# h3 = d3 + h4*s4
+	vpmuludq	$T3,$H3,$H0		# h3*s4
+	vpaddq		$H0,$D2,$D2		# h2 = d2 + h3*s4
+	vpmuludq	$T2,$H3,$H1		# h2*s4
+	vpaddq		$H1,$D1,$D1		# h1 = d1 + h2*s4
+	vpmuludq	$T1,$H3,$H3		# h1*s4
+	vpaddq		$H3,$D0,$D0		# h0 = d0 + h1*s4
+
+	jz		.Lshort_tail_avx
+
+	vmovdqu		16*0($inp),$H0		# load input
+	vmovdqu		16*1($inp),$H1
+
+	vpsrldq		\$6,$H0,$H2		# splat input
+	vpsrldq		\$6,$H1,$H3
+	vpunpckhqdq	$H1,$H0,$H4		# 4
+	vpunpcklqdq	$H1,$H0,$H0		# 0:1
+	vpunpcklqdq	$H3,$H2,$H3		# 2:3
+
+	vpsrlq		\$40,$H4,$H4		# 4
+	vpsrlq		\$26,$H0,$H1
+	vpand		$MASK,$H0,$H0		# 0
+	vpsrlq		\$4,$H3,$H2
+	vpand		$MASK,$H1,$H1		# 1
+	vpsrlq		\$30,$H3,$H3
+	vpand		$MASK,$H2,$H2		# 2
+	vpand		$MASK,$H3,$H3		# 3
+	vpor		32(%rcx),$H4,$H4	# padbit, yes, always
+
+	vpshufd		\$0x32,`16*0-64`($ctx),$T4	# r0^n, 34xx -> x3x4
+	vpaddq		0x00(%r11),$H0,$H0
+	vpaddq		0x10(%r11),$H1,$H1
+	vpaddq		0x20(%r11),$H2,$H2
+	vpaddq		0x30(%r11),$H3,$H3
+	vpaddq		0x40(%r11),$H4,$H4
+
+	################################################################
+	# multiply (inp[0:1]+hash) by r^4:r^3 and accumulate
+
+	vpmuludq	$H0,$T4,$T0		# h0*r0
+	vpaddq		$T0,$D0,$D0		# d0 += h0*r0
+	vpmuludq	$H1,$T4,$T1		# h1*r0
+	vpaddq		$T1,$D1,$D1		# d1 += h1*r0
+	vpmuludq	$H2,$T4,$T0		# h2*r0
+	vpaddq		$T0,$D2,$D2		# d2 += h2*r0
+	 vpshufd	\$0x32,`16*1-64`($ctx),$T2		# r1^n
+	vpmuludq	$H3,$T4,$T1		# h3*r0
+	vpaddq		$T1,$D3,$D3		# d3 += h3*r0
+	vpmuludq	$H4,$T4,$T4		# h4*r0
+	vpaddq		$T4,$D4,$D4		# d4 += h4*r0
+
+	vpmuludq	$H3,$T2,$T0		# h3*r1
+	vpaddq		$T0,$D4,$D4		# d4 += h3*r1
+	 vpshufd	\$0x32,`16*2-64`($ctx),$T3		# s1
+	vpmuludq	$H2,$T2,$T1		# h2*r1
+	vpaddq		$T1,$D3,$D3		# d3 += h2*r1
+	 vpshufd	\$0x32,`16*3-64`($ctx),$T4		# r2
+	vpmuludq	$H1,$T2,$T0		# h1*r1
+	vpaddq		$T0,$D2,$D2		# d2 += h1*r1
+	vpmuludq	$H0,$T2,$T2		# h0*r1
+	vpaddq		$T2,$D1,$D1		# d1 += h0*r1
+	vpmuludq	$H4,$T3,$T3		# h4*s1
+	vpaddq		$T3,$D0,$D0		# d0 += h4*s1
+
+	 vpshufd	\$0x32,`16*4-64`($ctx),$T2		# s2
+	vpmuludq	$H2,$T4,$T1		# h2*r2
+	vpaddq		$T1,$D4,$D4		# d4 += h2*r2
+	vpmuludq	$H1,$T4,$T0		# h1*r2
+	vpaddq		$T0,$D3,$D3		# d3 += h1*r2
+	 vpshufd	\$0x32,`16*5-64`($ctx),$T3		# r3
+	vpmuludq	$H0,$T4,$T4		# h0*r2
+	vpaddq		$T4,$D2,$D2		# d2 += h0*r2
+	vpmuludq	$H4,$T2,$T1		# h4*s2
+	vpaddq		$T1,$D1,$D1		# d1 += h4*s2
+	 vpshufd	\$0x32,`16*6-64`($ctx),$T4		# s3
+	vpmuludq	$H3,$T2,$T2		# h3*s2
+	vpaddq		$T2,$D0,$D0		# d0 += h3*s2
+
+	vpmuludq	$H1,$T3,$T0		# h1*r3
+	vpaddq		$T0,$D4,$D4		# d4 += h1*r3
+	vpmuludq	$H0,$T3,$T3		# h0*r3
+	vpaddq		$T3,$D3,$D3		# d3 += h0*r3
+	 vpshufd	\$0x32,`16*7-64`($ctx),$T2		# r4
+	vpmuludq	$H4,$T4,$T1		# h4*s3
+	vpaddq		$T1,$D2,$D2		# d2 += h4*s3
+	 vpshufd	\$0x32,`16*8-64`($ctx),$T3		# s4
+	vpmuludq	$H3,$T4,$T0		# h3*s3
+	vpaddq		$T0,$D1,$D1		# d1 += h3*s3
+	vpmuludq	$H2,$T4,$T4		# h2*s3
+	vpaddq		$T4,$D0,$D0		# d0 += h2*s3
+
+	vpmuludq	$H0,$T2,$T2		# h0*r4
+	vpaddq		$T2,$D4,$D4		# d4 += h0*r4
+	vpmuludq	$H4,$T3,$T1		# h4*s4
+	vpaddq		$T1,$D3,$D3		# d3 += h4*s4
+	vpmuludq	$H3,$T3,$T0		# h3*s4
+	vpaddq		$T0,$D2,$D2		# d2 += h3*s4
+	vpmuludq	$H2,$T3,$T1		# h2*s4
+	vpaddq		$T1,$D1,$D1		# d1 += h2*s4
+	vpmuludq	$H1,$T3,$T3		# h1*s4
+	vpaddq		$T3,$D0,$D0		# d0 += h1*s4
+
+.Lshort_tail_avx:
+	################################################################
+	# lazy reduction
+
+	vpsrlq		\$26,$D3,$H3
+	vpand		$MASK,$D3,$D3
+	vpaddq		$H3,$D4,$D4		# h3 -> h4
+
+	vpsrlq		\$26,$D0,$H0
+	vpand		$MASK,$D0,$D0
+	vpaddq		$H0,$D1,$D1		# h0 -> h1
+
+	vpsrlq		\$26,$D4,$H4
+	vpand		$MASK,$D4,$D4
+
+	vpsrlq		\$26,$D1,$H1
+	vpand		$MASK,$D1,$D1
+	vpaddq		$H1,$D2,$D2		# h1 -> h2
+
+	vpaddq		$H4,$D0,$D0
+	vpsllq		\$2,$H4,$H4
+	vpaddq		$H4,$D0,$D0		# h4 -> h0
+
+	vpsrlq		\$26,$D2,$H2
+	vpand		$MASK,$D2,$D2
+	vpaddq		$H2,$D3,$D3		# h2 -> h3
+
+	vpsrlq		\$26,$D0,$H0
+	vpand		$MASK,$D0,$D0
+	vpaddq		$H0,$D1,$D1		# h0 -> h1
+
+	vpsrlq		\$26,$D3,$H3
+	vpand		$MASK,$D3,$D3
+	vpaddq		$H3,$D4,$D4		# h3 -> h4
+
+	################################################################
+	# horizontal addition
+
+	vpsrldq		\$8,$D2,$T2
+	vpsrldq		\$8,$D0,$T0
+	vpsrldq		\$8,$D1,$T1
+	vpsrldq		\$8,$D3,$T3
+	vpsrldq		\$8,$D4,$T4
+	vpaddq		$T2,$D2,$H2
+	vpaddq		$T0,$D0,$H0
+	vpaddq		$T1,$D1,$H1
+	vpaddq		$T3,$D3,$H3
+	vpaddq		$T4,$D4,$H4
+
+	vmovd		$H0,`4*0-48-64`($ctx)	# save partially reduced
+	vmovd		$H1,`4*1-48-64`($ctx)
+	vmovd		$H2,`4*2-48-64`($ctx)
+	vmovd		$H3,`4*3-48-64`($ctx)
+	vmovd		$H4,`4*4-48-64`($ctx)
+___
+$code.=<<___	if ($win64);
+	vmovdqa		0x50(%r11),%xmm6
+	vmovdqa		0x60(%r11),%xmm7
+	vmovdqa		0x70(%r11),%xmm8
+	vmovdqa		0x80(%r11),%xmm9
+	vmovdqa		0x90(%r11),%xmm10
+	vmovdqa		0xa0(%r11),%xmm11
+	vmovdqa		0xb0(%r11),%xmm12
+	vmovdqa		0xc0(%r11),%xmm13
+	vmovdqa		0xd0(%r11),%xmm14
+	vmovdqa		0xe0(%r11),%xmm15
+	lea		0xf8(%r11),%rsp
+.Ldo_avx_epilogue:
+___
+$code.=<<___	if (!$win64);
+	lea		0x58(%r11),%rsp
+___
+$code.=<<___;
+	vzeroupper
+	ret
+.size	poly1305_blocks_avx,.-poly1305_blocks_avx
+
+.type	poly1305_emit_avx,\@function,3
+.align	32
+poly1305_emit_avx:
+	cmpl	\$0,20($ctx)	# is_base2_26?
+	je	.Lemit
+
+	mov	0($ctx),%eax	# load hash value base 2^26
+	mov	4($ctx),%ecx
+	mov	8($ctx),%r8d
+	mov	12($ctx),%r11d
+	mov	16($ctx),%r10d
+
+	shl	\$26,%rcx	# base 2^26 -> base 2^64
+	mov	%r8,%r9
+	shl	\$52,%r8
+	add	%rcx,%rax
+	shr	\$12,%r9
+	add	%rax,%r8	# h0
+	adc	\$0,%r9
+
+	shl	\$14,%r11
+	mov	%r10,%rax
+	shr	\$24,%r10
+	add	%r11,%r9
+	shl	\$40,%rax
+	add	%rax,%r9	# h1
+	adc	\$0,%r10	# h2
+
+	mov	%r10,%rax	# could be partially reduced, so reduce
+	mov	%r10,%rcx
+	and	\$3,%r10
+	shr	\$2,%rax
+	and	\$-4,%rcx
+	add	%rcx,%rax
+	add	%rax,%r8
+	adc	\$0,%r9
+
+	mov	%r8,%rax
+	add	\$5,%r8		# compare to modulus
+	mov	%r9,%rcx
+	adc	\$0,%r9
+	adc	\$0,%r10
+	shr	\$2,%r10	# did 130-bit value overfow?
+	cmovnz	%r8,%rax
+	cmovnz	%r9,%rcx
+
+	add	0($nonce),%rax	# accumulate nonce
+	adc	8($nonce),%rcx
+	mov	%rax,0($mac)	# write result
+	mov	%rcx,8($mac)
+
+	ret
+.size	poly1305_emit_avx,.-poly1305_emit_avx
+___
+
+if ($avx>1) {
+my ($H0,$H1,$H2,$H3,$H4, $MASK, $T4,$T0,$T1,$T2,$T3, $D0,$D1,$D2,$D3,$D4) =
+    map("%ymm$_",(0..15));
+my $S4=$MASK;
+
+$code.=<<___;
+.type	poly1305_blocks_avx2,\@function,4
+.align	32
+poly1305_blocks_avx2:
+	mov	20($ctx),%r8d		# is_base2_26
+	cmp	\$128,$len
+	jae	.Lblocks_avx2
+	test	%r8d,%r8d
+	jz	.Lblocks
+
+.Lblocks_avx2:
+	and	\$-16,$len
+	jz	.Lno_data_avx2
+
+	vzeroupper
+
+	test	%r8d,%r8d
+	jz	.Lbase2_64_avx2
+
+	test	\$63,$len
+	jz	.Leven_avx2
+
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+.Lblocks_avx2_body:
+
+	mov	$len,%r15		# reassign $len
+
+	mov	0($ctx),$d1		# load hash value
+	mov	8($ctx),$d2
+	mov	16($ctx),$h2#d
+
+	mov	24($ctx),$r0		# load r
+	mov	32($ctx),$s1
+
+	################################# base 2^26 -> base 2^64
+	mov	$d1#d,$h0#d
+	and	\$-1<<31,$d1
+	mov	$d2,$r1			# borrow $r1
+	mov	$d2#d,$h1#d
+	and	\$-1<<31,$d2
+
+	shr	\$6,$d1
+	shl	\$52,$r1
+	add	$d1,$h0
+	shr	\$12,$h1
+	shr	\$18,$d2
+	add	$r1,$h0
+	adc	$d2,$h1
+
+	mov	$h2,$d1
+	shl	\$40,$d1
+	shr	\$24,$h2
+	add	$d1,$h1
+	adc	\$0,$h2			# can be partially reduced...
+
+	mov	\$-4,$d2		# ... so reduce
+	mov	$h2,$d1
+	and	$h2,$d2
+	shr	\$2,$d1
+	and	\$3,$h2
+	add	$d2,$d1			# =*5
+	add	$d1,$h0
+	adc	\$0,$h1
+
+	mov	$s1,$r1
+	mov	$s1,%rax
+	shr	\$2,$s1
+	add	$r1,$s1			# s1 = r1 + (r1 >> 2)
+
+.Lbase2_26_pre_avx2:
+	add	0($inp),$h0		# accumulate input
+	adc	8($inp),$h1
+	lea	16($inp),$inp
+	adc	$padbit,$h2
+	sub	\$16,%r15
+
+	call	__poly1305_block
+	mov	$r1,%rax
+
+	test	\$63,%r15
+	jnz	.Lbase2_26_pre_avx2
+
+	test	$padbit,$padbit		# if $padbit is zero,
+	jz	.Lstore_base2_64_avx2	# store hash in base 2^64 format
+
+	################################# base 2^64 -> base 2^26
+	mov	$h0,%rax
+	mov	$h0,%rdx
+	shr	\$52,$h0
+	mov	$h1,$r0
+	mov	$h1,$r1
+	shr	\$26,%rdx
+	and	\$0x3ffffff,%rax	# h[0]
+	shl	\$12,$r0
+	and	\$0x3ffffff,%rdx	# h[1]
+	shr	\$14,$h1
+	or	$r0,$h0
+	shl	\$24,$h2
+	and	\$0x3ffffff,$h0		# h[2]
+	shr	\$40,$r1
+	and	\$0x3ffffff,$h1		# h[3]
+	or	$r1,$h2			# h[4]
+
+	test	%r15,%r15
+	jz	.Lstore_base2_26_avx2
+
+	vmovd	%rax#d,%x#$H0
+	vmovd	%rdx#d,%x#$H1
+	vmovd	$h0#d,%x#$H2
+	vmovd	$h1#d,%x#$H3
+	vmovd	$h2#d,%x#$H4
+	jmp	.Lproceed_avx2
+
+.align	32
+.Lstore_base2_64_avx2:
+	mov	$h0,0($ctx)
+	mov	$h1,8($ctx)
+	mov	$h2,16($ctx)		# note that is_base2_26 is zeroed
+	jmp	.Ldone_avx2
+
+.align	16
+.Lstore_base2_26_avx2:
+	mov	%rax#d,0($ctx)		# store hash value base 2^26
+	mov	%rdx#d,4($ctx)
+	mov	$h0#d,8($ctx)
+	mov	$h1#d,12($ctx)
+	mov	$h2#d,16($ctx)
+.align	16
+.Ldone_avx2:
+	mov	0(%rsp),%r15
+	mov	8(%rsp),%r14
+	mov	16(%rsp),%r13
+	mov	24(%rsp),%r12
+	mov	32(%rsp),%rbp
+	mov	40(%rsp),%rbx
+	lea	48(%rsp),%rsp
+.Lno_data_avx2:
+.Lblocks_avx2_epilogue:
+	ret
+
+.align	32
+.Lbase2_64_avx2:
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+.Lbase2_64_avx2_body:
+
+	mov	$len,%r15		# reassign $len
+
+	mov	24($ctx),$r0		# load r
+	mov	32($ctx),$s1
+
+	mov	0($ctx),$h0		# load hash value
+	mov	8($ctx),$h1
+	mov	16($ctx),$h2#d
+
+	mov	$s1,$r1
+	mov	$s1,%rax
+	shr	\$2,$s1
+	add	$r1,$s1			# s1 = r1 + (r1 >> 2)
+
+	test	\$63,$len
+	jz	.Linit_avx2
+
+.Lbase2_64_pre_avx2:
+	add	0($inp),$h0		# accumulate input
+	adc	8($inp),$h1
+	lea	16($inp),$inp
+	adc	$padbit,$h2
+	sub	\$16,%r15
+
+	call	__poly1305_block
+	mov	$r1,%rax
+
+	test	\$63,%r15
+	jnz	.Lbase2_64_pre_avx2
+
+.Linit_avx2:
+	################################# base 2^64 -> base 2^26
+	mov	$h0,%rax
+	mov	$h0,%rdx
+	shr	\$52,$h0
+	mov	$h1,$d1
+	mov	$h1,$d2
+	shr	\$26,%rdx
+	and	\$0x3ffffff,%rax	# h[0]
+	shl	\$12,$d1
+	and	\$0x3ffffff,%rdx	# h[1]
+	shr	\$14,$h1
+	or	$d1,$h0
+	shl	\$24,$h2
+	and	\$0x3ffffff,$h0		# h[2]
+	shr	\$40,$d2
+	and	\$0x3ffffff,$h1		# h[3]
+	or	$d2,$h2			# h[4]
+
+	vmovd	%rax#d,%x#$H0
+	vmovd	%rdx#d,%x#$H1
+	vmovd	$h0#d,%x#$H2
+	vmovd	$h1#d,%x#$H3
+	vmovd	$h2#d,%x#$H4
+	movl	\$1,20($ctx)		# set is_base2_26
+
+	call	__poly1305_init_avx
+
+.Lproceed_avx2:
+	mov	%r15,$len
+
+	mov	0(%rsp),%r15
+	mov	8(%rsp),%r14
+	mov	16(%rsp),%r13
+	mov	24(%rsp),%r12
+	mov	32(%rsp),%rbp
+	mov	40(%rsp),%rbx
+	lea	48(%rsp),%rax
+	lea	48(%rsp),%rsp
+.Lbase2_64_avx2_epilogue:
+	jmp	.Ldo_avx2
+
+.align	32
+.Leven_avx2:
+	vmovd		4*0($ctx),%x#$H0	# load hash value base 2^26
+	vmovd		4*1($ctx),%x#$H1
+	vmovd		4*2($ctx),%x#$H2
+	vmovd		4*3($ctx),%x#$H3
+	vmovd		4*4($ctx),%x#$H4
+
+.Ldo_avx2:
+___
+$code.=<<___	if (!$win64);
+	lea		-8(%rsp),%r11
+	sub		\$0x128,%rsp
+___
+$code.=<<___	if ($win64);
+	lea		-0xf8(%rsp),%r11
+	sub		\$0x1c8,%rsp
+	vmovdqa		%xmm6,0x50(%r11)
+	vmovdqa		%xmm7,0x60(%r11)
+	vmovdqa		%xmm8,0x70(%r11)
+	vmovdqa		%xmm9,0x80(%r11)
+	vmovdqa		%xmm10,0x90(%r11)
+	vmovdqa		%xmm11,0xa0(%r11)
+	vmovdqa		%xmm12,0xb0(%r11)
+	vmovdqa		%xmm13,0xc0(%r11)
+	vmovdqa		%xmm14,0xd0(%r11)
+	vmovdqa		%xmm15,0xe0(%r11)
+.Ldo_avx2_body:
+___
+$code.=<<___;
+	lea		48+64($ctx),$ctx	# size optimization
+	lea		.Lconst(%rip),%rcx
+
+	# expand and copy pre-calculated table to stack
+	vmovdqu		`16*0-64`($ctx),%x#$T2
+	and		\$-512,%rsp
+	vmovdqu		`16*1-64`($ctx),%x#$T3
+	vmovdqu		`16*2-64`($ctx),%x#$T4
+	vmovdqu		`16*3-64`($ctx),%x#$D0
+	vmovdqu		`16*4-64`($ctx),%x#$D1
+	vmovdqu		`16*5-64`($ctx),%x#$D2
+	vmovdqu		`16*6-64`($ctx),%x#$D3
+	vpermq		\$0x15,$T2,$T2		# 00003412 -> 12343434
+	vmovdqu		`16*7-64`($ctx),%x#$D4
+	vpermq		\$0x15,$T3,$T3
+	vpshufd		\$0xc8,$T2,$T2		# 12343434 -> 14243444
+	vmovdqu		`16*8-64`($ctx),%x#$MASK
+	vpermq		\$0x15,$T4,$T4
+	vpshufd		\$0xc8,$T3,$T3
+	vmovdqa		$T2,0x00(%rsp)
+	vpermq		\$0x15,$D0,$D0
+	vpshufd		\$0xc8,$T4,$T4
+	vmovdqa		$T3,0x20(%rsp)
+	vpermq		\$0x15,$D1,$D1
+	vpshufd		\$0xc8,$D0,$D0
+	vmovdqa		$T4,0x40(%rsp)
+	vpermq		\$0x15,$D2,$D2
+	vpshufd		\$0xc8,$D1,$D1
+	vmovdqa		$D0,0x60(%rsp)
+	vpermq		\$0x15,$D3,$D3
+	vpshufd		\$0xc8,$D2,$D2
+	vmovdqa		$D1,0x80(%rsp)
+	vpermq		\$0x15,$D4,$D4
+	vpshufd		\$0xc8,$D3,$D3
+	vmovdqa		$D2,0xa0(%rsp)
+	vpermq		\$0x15,$MASK,$MASK
+	vpshufd		\$0xc8,$D4,$D4
+	vmovdqa		$D3,0xc0(%rsp)
+	vpshufd		\$0xc8,$MASK,$MASK
+	vmovdqa		$D4,0xe0(%rsp)
+	vmovdqa		$MASK,0x100(%rsp)
+	vmovdqa		64(%rcx),$MASK		# .Lmask26
+
+	################################################################
+	# load input
+	vmovdqu		16*0($inp),%x#$T0
+	vmovdqu		16*1($inp),%x#$T1
+	vinserti128	\$1,16*2($inp),$T0,$T0
+	vinserti128	\$1,16*3($inp),$T1,$T1
+	lea		16*4($inp),$inp
+
+	vpsrldq		\$6,$T0,$T2		# splat input
+	vpsrldq		\$6,$T1,$T3
+	vpunpckhqdq	$T1,$T0,$T4		# 4
+	vpunpcklqdq	$T3,$T2,$T2		# 2:3
+	vpunpcklqdq	$T1,$T0,$T0		# 0:1
+
+	vpsrlq		\$30,$T2,$T3
+	vpsrlq		\$4,$T2,$T2
+	vpsrlq		\$26,$T0,$T1
+	vpsrlq		\$40,$T4,$T4		# 4
+	vpand		$MASK,$T2,$T2		# 2
+	vpand		$MASK,$T0,$T0		# 0
+	vpand		$MASK,$T1,$T1		# 1
+	vpand		$MASK,$T3,$T3		# 3
+	vpor		32(%rcx),$T4,$T4	# padbit, yes, always
+
+	lea		0x90(%rsp),%rax		# size optimization
+	vpaddq		$H2,$T2,$H2		# accumulate input
+	sub		\$64,$len
+	jz		.Ltail_avx2
+	jmp		.Loop_avx2
+
+.align	32
+.Loop_avx2:
+	################################################################
+	# ((inp[0]*r^4+r[4])*r^4+r[8])*r^4
+	# ((inp[1]*r^4+r[5])*r^4+r[9])*r^3
+	# ((inp[2]*r^4+r[6])*r^4+r[10])*r^2
+	# ((inp[3]*r^4+r[7])*r^4+r[11])*r^1
+	#   \________/\________/
+	################################################################
+	#vpaddq		$H2,$T2,$H2		# accumulate input
+	vpaddq		$H0,$T0,$H0
+	vmovdqa		`32*0`(%rsp),$T0	# r0^4
+	vpaddq		$H1,$T1,$H1
+	vmovdqa		`32*1`(%rsp),$T1	# r1^4
+	vpaddq		$H3,$T3,$H3
+	vmovdqa		`32*3`(%rsp),$T2	# r2^4
+	vpaddq		$H4,$T4,$H4
+	vmovdqa		`32*6-0x90`(%rax),$T3	# s3^4
+	vmovdqa		`32*8-0x90`(%rax),$S4	# s4^4
+
+	# d4 = h4*r0 + h3*r1   + h2*r2   + h1*r3   + h0*r4
+	# d3 = h3*r0 + h2*r1   + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0 + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3 + h2*5*r4
+	# d0 = h0*r0 + h4*5*r1 + h3*5*r2 + h2*5*r3 + h1*5*r4
+	#
+	# however, as h2 is "chronologically" first one available pull
+	# corresponding operations up, so it's
+	#
+	# d4 = h2*r2   + h4*r0 + h3*r1             + h1*r3   + h0*r4
+	# d3 = h2*r1   + h3*r0           + h1*r2   + h0*r3   + h4*5*r4
+	# d2 = h2*r0           + h1*r1   + h0*r2   + h4*5*r3 + h3*5*r4
+	# d1 = h2*5*r4 + h1*r0 + h0*r1   + h4*5*r2 + h3*5*r3
+	# d0 = h2*5*r3 + h0*r0 + h4*5*r1 + h3*5*r2           + h1*5*r4
+
+	vpmuludq	$H2,$T0,$D2		# d2 = h2*r0
+	vpmuludq	$H2,$T1,$D3		# d3 = h2*r1
+	vpmuludq	$H2,$T2,$D4		# d4 = h2*r2
+	vpmuludq	$H2,$T3,$D0		# d0 = h2*s3
+	vpmuludq	$H2,$S4,$D1		# d1 = h2*s4
+
+	vpmuludq	$H0,$T1,$T4		# h0*r1
+	vpmuludq	$H1,$T1,$H2		# h1*r1, borrow $H2 as temp
+	vpaddq		$T4,$D1,$D1		# d1 += h0*r1
+	vpaddq		$H2,$D2,$D2		# d2 += h1*r1
+	vpmuludq	$H3,$T1,$T4		# h3*r1
+	vpmuludq	`32*2`(%rsp),$H4,$H2	# h4*s1
+	vpaddq		$T4,$D4,$D4		# d4 += h3*r1
+	vpaddq		$H2,$D0,$D0		# d0 += h4*s1
+	 vmovdqa	`32*4-0x90`(%rax),$T1	# s2
+
+	vpmuludq	$H0,$T0,$T4		# h0*r0
+	vpmuludq	$H1,$T0,$H2		# h1*r0
+	vpaddq		$T4,$D0,$D0		# d0 += h0*r0
+	vpaddq		$H2,$D1,$D1		# d1 += h1*r0
+	vpmuludq	$H3,$T0,$T4		# h3*r0
+	vpmuludq	$H4,$T0,$H2		# h4*r0
+	 vmovdqu	16*0($inp),%x#$T0	# load input
+	vpaddq		$T4,$D3,$D3		# d3 += h3*r0
+	vpaddq		$H2,$D4,$D4		# d4 += h4*r0
+	 vinserti128	\$1,16*2($inp),$T0,$T0
+
+	vpmuludq	$H3,$T1,$T4		# h3*s2
+	vpmuludq	$H4,$T1,$H2		# h4*s2
+	 vmovdqu	16*1($inp),%x#$T1
+	vpaddq		$T4,$D0,$D0		# d0 += h3*s2
+	vpaddq		$H2,$D1,$D1		# d1 += h4*s2
+	 vmovdqa	`32*5-0x90`(%rax),$H2	# r3
+	vpmuludq	$H1,$T2,$T4		# h1*r2
+	vpmuludq	$H0,$T2,$T2		# h0*r2
+	vpaddq		$T4,$D3,$D3		# d3 += h1*r2
+	vpaddq		$T2,$D2,$D2		# d2 += h0*r2
+	 vinserti128	\$1,16*3($inp),$T1,$T1
+	 lea		16*4($inp),$inp
+
+	vpmuludq	$H1,$H2,$T4		# h1*r3
+	vpmuludq	$H0,$H2,$H2		# h0*r3
+	 vpsrldq	\$6,$T0,$T2		# splat input
+	vpaddq		$T4,$D4,$D4		# d4 += h1*r3
+	vpaddq		$H2,$D3,$D3		# d3 += h0*r3
+	vpmuludq	$H3,$T3,$T4		# h3*s3
+	vpmuludq	$H4,$T3,$H2		# h4*s3
+	 vpsrldq	\$6,$T1,$T3
+	vpaddq		$T4,$D1,$D1		# d1 += h3*s3
+	vpaddq		$H2,$D2,$D2		# d2 += h4*s3
+	 vpunpckhqdq	$T1,$T0,$T4		# 4
+
+	vpmuludq	$H3,$S4,$H3		# h3*s4
+	vpmuludq	$H4,$S4,$H4		# h4*s4
+	 vpunpcklqdq	$T1,$T0,$T0		# 0:1
+	vpaddq		$H3,$D2,$H2		# h2 = d2 + h3*r4
+	vpaddq		$H4,$D3,$H3		# h3 = d3 + h4*r4
+	 vpunpcklqdq	$T3,$T2,$T3		# 2:3
+	vpmuludq	`32*7-0x90`(%rax),$H0,$H4	# h0*r4
+	vpmuludq	$H1,$S4,$H0		# h1*s4
+	vmovdqa		64(%rcx),$MASK		# .Lmask26
+	vpaddq		$H4,$D4,$H4		# h4 = d4 + h0*r4
+	vpaddq		$H0,$D0,$H0		# h0 = d0 + h1*s4
+
+	################################################################
+	# lazy reduction (interleaved with tail of input splat)
+
+	vpsrlq		\$26,$H3,$D3
+	vpand		$MASK,$H3,$H3
+	vpaddq		$D3,$H4,$H4		# h3 -> h4
+
+	vpsrlq		\$26,$H0,$D0
+	vpand		$MASK,$H0,$H0
+	vpaddq		$D0,$D1,$H1		# h0 -> h1
+
+	vpsrlq		\$26,$H4,$D4
+	vpand		$MASK,$H4,$H4
+
+	 vpsrlq		\$4,$T3,$T2
+
+	vpsrlq		\$26,$H1,$D1
+	vpand		$MASK,$H1,$H1
+	vpaddq		$D1,$H2,$H2		# h1 -> h2
+
+	vpaddq		$D4,$H0,$H0
+	vpsllq		\$2,$D4,$D4
+	vpaddq		$D4,$H0,$H0		# h4 -> h0
+
+	 vpand		$MASK,$T2,$T2		# 2
+	 vpsrlq		\$26,$T0,$T1
+
+	vpsrlq		\$26,$H2,$D2
+	vpand		$MASK,$H2,$H2
+	vpaddq		$D2,$H3,$H3		# h2 -> h3
+
+	 vpaddq		$T2,$H2,$H2		# modulo-scheduled
+	 vpsrlq		\$30,$T3,$T3
+
+	vpsrlq		\$26,$H0,$D0
+	vpand		$MASK,$H0,$H0
+	vpaddq		$D0,$H1,$H1		# h0 -> h1
+
+	 vpsrlq		\$40,$T4,$T4		# 4
+
+	vpsrlq		\$26,$H3,$D3
+	vpand		$MASK,$H3,$H3
+	vpaddq		$D3,$H4,$H4		# h3 -> h4
+
+	 vpand		$MASK,$T0,$T0		# 0
+	 vpand		$MASK,$T1,$T1		# 1
+	 vpand		$MASK,$T3,$T3		# 3
+	 vpor		32(%rcx),$T4,$T4	# padbit, yes, always
+
+	sub		\$64,$len
+	jnz		.Loop_avx2
+
+	.byte		0x66,0x90
+.Ltail_avx2:
+	################################################################
+	# while above multiplications were by r^4 in all lanes, in last
+	# iteration we multiply least significant lane by r^4 and most
+	# significant one by r, so copy of above except that references
+	# to the precomputed table are displaced by 4...
+
+	#vpaddq		$H2,$T2,$H2		# accumulate input
+	vpaddq		$H0,$T0,$H0
+	vmovdqu		`32*0+4`(%rsp),$T0	# r0^4
+	vpaddq		$H1,$T1,$H1
+	vmovdqu		`32*1+4`(%rsp),$T1	# r1^4
+	vpaddq		$H3,$T3,$H3
+	vmovdqu		`32*3+4`(%rsp),$T2	# r2^4
+	vpaddq		$H4,$T4,$H4
+	vmovdqu		`32*6+4-0x90`(%rax),$T3	# s3^4
+	vmovdqu		`32*8+4-0x90`(%rax),$S4	# s4^4
+
+	vpmuludq	$H2,$T0,$D2		# d2 = h2*r0
+	vpmuludq	$H2,$T1,$D3		# d3 = h2*r1
+	vpmuludq	$H2,$T2,$D4		# d4 = h2*r2
+	vpmuludq	$H2,$T3,$D0		# d0 = h2*s3
+	vpmuludq	$H2,$S4,$D1		# d1 = h2*s4
+
+	vpmuludq	$H0,$T1,$T4		# h0*r1
+	vpmuludq	$H1,$T1,$H2		# h1*r1
+	vpaddq		$T4,$D1,$D1		# d1 += h0*r1
+	vpaddq		$H2,$D2,$D2		# d2 += h1*r1
+	vpmuludq	$H3,$T1,$T4		# h3*r1
+	vpmuludq	`32*2+4`(%rsp),$H4,$H2	# h4*s1
+	vpaddq		$T4,$D4,$D4		# d4 += h3*r1
+	vpaddq		$H2,$D0,$D0		# d0 += h4*s1
+
+	vpmuludq	$H0,$T0,$T4		# h0*r0
+	vpmuludq	$H1,$T0,$H2		# h1*r0
+	vpaddq		$T4,$D0,$D0		# d0 += h0*r0
+	 vmovdqu	`32*4+4-0x90`(%rax),$T1	# s2
+	vpaddq		$H2,$D1,$D1		# d1 += h1*r0
+	vpmuludq	$H3,$T0,$T4		# h3*r0
+	vpmuludq	$H4,$T0,$H2		# h4*r0
+	vpaddq		$T4,$D3,$D3		# d3 += h3*r0
+	vpaddq		$H2,$D4,$D4		# d4 += h4*r0
+
+	vpmuludq	$H3,$T1,$T4		# h3*s2
+	vpmuludq	$H4,$T1,$H2		# h4*s2
+	vpaddq		$T4,$D0,$D0		# d0 += h3*s2
+	vpaddq		$H2,$D1,$D1		# d1 += h4*s2
+	 vmovdqu	`32*5+4-0x90`(%rax),$H2	# r3
+	vpmuludq	$H1,$T2,$T4		# h1*r2
+	vpmuludq	$H0,$T2,$T2		# h0*r2
+	vpaddq		$T4,$D3,$D3		# d3 += h1*r2
+	vpaddq		$T2,$D2,$D2		# d2 += h0*r2
+
+	vpmuludq	$H1,$H2,$T4		# h1*r3
+	vpmuludq	$H0,$H2,$H2		# h0*r3
+	vpaddq		$T4,$D4,$D4		# d4 += h1*r3
+	vpaddq		$H2,$D3,$D3		# d3 += h0*r3
+	vpmuludq	$H3,$T3,$T4		# h3*s3
+	vpmuludq	$H4,$T3,$H2		# h4*s3
+	vpaddq		$T4,$D1,$D1		# d1 += h3*s3
+	vpaddq		$H2,$D2,$D2		# d2 += h4*s3
+
+	vpmuludq	$H3,$S4,$H3		# h3*s4
+	vpmuludq	$H4,$S4,$H4		# h4*s4
+	vpaddq		$H3,$D2,$H2		# h2 = d2 + h3*r4
+	vpaddq		$H4,$D3,$H3		# h3 = d3 + h4*r4
+	vpmuludq	`32*7+4-0x90`(%rax),$H0,$H4		# h0*r4
+	vpmuludq	$H1,$S4,$H0		# h1*s4
+	vmovdqa		64(%rcx),$MASK		# .Lmask26
+	vpaddq		$H4,$D4,$H4		# h4 = d4 + h0*r4
+	vpaddq		$H0,$D0,$H0		# h0 = d0 + h1*s4
+
+	################################################################
+	# lazy reduction
+
+	vpsrlq		\$26,$H3,$D3
+	vpand		$MASK,$H3,$H3
+	vpaddq		$D3,$H4,$H4		# h3 -> h4
+
+	vpsrlq		\$26,$H0,$D0
+	vpand		$MASK,$H0,$H0
+	vpaddq		$D0,$D1,$H1		# h0 -> h1
+
+	vpsrlq		\$26,$H4,$D4
+	vpand		$MASK,$H4,$H4
+
+	vpsrlq		\$26,$H1,$D1
+	vpand		$MASK,$H1,$H1
+	vpaddq		$D1,$H2,$H2		# h1 -> h2
+
+	vpaddq		$D4,$H0,$H0
+	vpsllq		\$2,$D4,$D4
+	vpaddq		$D4,$H0,$H0		# h4 -> h0
+
+	vpsrlq		\$26,$H2,$D2
+	vpand		$MASK,$H2,$H2
+	vpaddq		$D2,$H3,$H3		# h2 -> h3
+
+	vpsrlq		\$26,$H0,$D0
+	vpand		$MASK,$H0,$H0
+	vpaddq		$D0,$H1,$H1		# h0 -> h1
+
+	vpsrlq		\$26,$H3,$D3
+	vpand		$MASK,$H3,$H3
+	vpaddq		$D3,$H4,$H4		# h3 -> h4
+
+	################################################################
+	# horizontal addition
+
+	vpsrldq		\$8,$H2,$T2
+	vpsrldq		\$8,$H0,$T0
+	vpsrldq		\$8,$H1,$T1
+	vpsrldq		\$8,$H3,$T3
+	vpsrldq		\$8,$H4,$T4
+	vpaddq		$T2,$H2,$H2
+	vpaddq		$T0,$H0,$H0
+	vpaddq		$T1,$H1,$H1
+	vpaddq		$T3,$H3,$H3
+	vpaddq		$T4,$H4,$H4
+
+	vpermq		\$0x2,$H2,$T2
+	vpermq		\$0x2,$H0,$T0
+	vpermq		\$0x2,$H1,$T1
+	vpermq		\$0x2,$H3,$T3
+	vpermq		\$0x2,$H4,$T4
+	vpaddq		$T2,$H2,$H2
+	vpaddq		$T0,$H0,$H0
+	vpaddq		$T1,$H1,$H1
+	vpaddq		$T3,$H3,$H3
+	vpaddq		$T4,$H4,$H4
+
+	vmovd		%x#$H0,`4*0-48-64`($ctx)# save partially reduced
+	vmovd		%x#$H1,`4*1-48-64`($ctx)
+	vmovd		%x#$H2,`4*2-48-64`($ctx)
+	vmovd		%x#$H3,`4*3-48-64`($ctx)
+	vmovd		%x#$H4,`4*4-48-64`($ctx)
+___
+$code.=<<___	if ($win64);
+	vmovdqa		0x50(%r11),%xmm6
+	vmovdqa		0x60(%r11),%xmm7
+	vmovdqa		0x70(%r11),%xmm8
+	vmovdqa		0x80(%r11),%xmm9
+	vmovdqa		0x90(%r11),%xmm10
+	vmovdqa		0xa0(%r11),%xmm11
+	vmovdqa		0xb0(%r11),%xmm12
+	vmovdqa		0xc0(%r11),%xmm13
+	vmovdqa		0xd0(%r11),%xmm14
+	vmovdqa		0xe0(%r11),%xmm15
+	lea		0xf8(%r11),%rsp
+.Ldo_avx2_epilogue:
+___
+$code.=<<___	if (!$win64);
+	lea		8(%r11),%rsp
+___
+$code.=<<___;
+	vzeroupper
+	ret
+.size	poly1305_blocks_avx2,.-poly1305_blocks_avx2
+___
+}
+$code.=<<___;
+.align	64
+.Lconst:
+.Lmask24:
+.long	0x0ffffff,0,0x0ffffff,0,0x0ffffff,0,0x0ffffff,0
+.L129:
+.long	1<<24,0,1<<24,0,1<<24,0,1<<24,0
+.Lmask26:
+.long	0x3ffffff,0,0x3ffffff,0,0x3ffffff,0,0x3ffffff,0
+.Lfive:
+.long	5,0,5,0,5,0,5,0
+___
+}
+
+$code.=<<___;
+.asciz	"Poly1305 for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align	16
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	se_handler,\@abi-omnipotent
+.align	16
+se_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
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<.Lprologue
+	jb	.Lcommon_seh_tail
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=.Lepilogue
+	jae	.Lcommon_seh_tail
+
+	lea	48(%rax),%rax
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R14
+
+	jmp	.Lcommon_seh_tail
+.size	se_handler,.-se_handler
+
+.type	avx_handler,\@abi-omnipotent
+.align	16
+avx_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lcommon_seh_tail
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lcommon_seh_tail
+
+	mov	208($context),%rax	# pull context->R11
+
+	lea	0x50(%rax),%rsi
+	lea	0xf8(%rax),%rax
+	lea	512($context),%rdi	# &context.Xmm6
+	mov	\$20,%ecx
+	.long	0xa548f3fc		# cld; rep movsq
+
+.Lcommon_seh_tail:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$154,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	avx_handler,.-avx_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_poly1305_init
+	.rva	.LSEH_end_poly1305_init
+	.rva	.LSEH_info_poly1305_init
+
+	.rva	.LSEH_begin_poly1305_blocks
+	.rva	.LSEH_end_poly1305_blocks
+	.rva	.LSEH_info_poly1305_blocks
+
+	.rva	.LSEH_begin_poly1305_emit
+	.rva	.LSEH_end_poly1305_emit
+	.rva	.LSEH_info_poly1305_emit
+___
+$code.=<<___ if ($avx);
+	.rva	.LSEH_begin_poly1305_blocks_avx
+	.rva	.Lbase2_64_avx
+	.rva	.LSEH_info_poly1305_blocks_avx_1
+
+	.rva	.Lbase2_64_avx
+	.rva	.Leven_avx
+	.rva	.LSEH_info_poly1305_blocks_avx_2
+
+	.rva	.Leven_avx
+	.rva	.LSEH_end_poly1305_blocks_avx
+	.rva	.LSEH_info_poly1305_blocks_avx_3
+
+	.rva	.LSEH_begin_poly1305_emit_avx
+	.rva	.LSEH_end_poly1305_emit_avx
+	.rva	.LSEH_info_poly1305_emit_avx
+___
+$code.=<<___ if ($avx>1);
+	.rva	.LSEH_begin_poly1305_blocks_avx2
+	.rva	.Lbase2_64_avx2
+	.rva	.LSEH_info_poly1305_blocks_avx2_1
+
+	.rva	.Lbase2_64_avx2
+	.rva	.Leven_avx2
+	.rva	.LSEH_info_poly1305_blocks_avx2_2
+
+	.rva	.Leven_avx2
+	.rva	.LSEH_end_poly1305_blocks_avx2
+	.rva	.LSEH_info_poly1305_blocks_avx2_3
+___
+$code.=<<___;
+.section	.xdata
+.align	8
+.LSEH_info_poly1305_init:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.LSEH_begin_poly1305_init,.LSEH_begin_poly1305_init
+
+.LSEH_info_poly1305_blocks:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lblocks_body,.Lblocks_epilogue
+
+.LSEH_info_poly1305_emit:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.LSEH_begin_poly1305_emit,.LSEH_begin_poly1305_emit
+___
+$code.=<<___ if ($avx);
+.LSEH_info_poly1305_blocks_avx_1:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lblocks_avx_body,.Lblocks_avx_epilogue		# HandlerData[]
+
+.LSEH_info_poly1305_blocks_avx_2:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lbase2_64_avx_body,.Lbase2_64_avx_epilogue	# HandlerData[]
+
+.LSEH_info_poly1305_blocks_avx_3:
+	.byte	9,0,0,0
+	.rva	avx_handler
+	.rva	.Ldo_avx_body,.Ldo_avx_epilogue			# HandlerData[]
+
+.LSEH_info_poly1305_emit_avx:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.LSEH_begin_poly1305_emit_avx,.LSEH_begin_poly1305_emit_avx
+___
+$code.=<<___ if ($avx>1);
+.LSEH_info_poly1305_blocks_avx2_1:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lblocks_avx2_body,.Lblocks_avx2_epilogue	# HandlerData[]
+
+.LSEH_info_poly1305_blocks_avx2_2:
+	.byte	9,0,0,0
+	.rva	se_handler
+	.rva	.Lbase2_64_avx2_body,.Lbase2_64_avx2_epilogue	# HandlerData[]
+
+.LSEH_info_poly1305_blocks_avx2_3:
+	.byte	9,0,0,0
+	.rva	avx_handler
+	.rva	.Ldo_avx2_body,.Ldo_avx2_epilogue		# HandlerData[]
+___
+}
+
+foreach (split('\n',$code)) {
+	s/\`([^\`]*)\`/eval($1)/ge;
+	s/%r([a-z]+)#d/%e$1/g;
+	s/%r([0-9]+)#d/%r$1d/g;
+	s/%x#%y/%x/g;
+
+	print $_,"\n";
+}
+close STDOUT;