ARM assembly pack: add ChaCha20 and Poly1305 modules.

Reviewed-by: Richard Levitte <levitte@openssl.org>

3 files changed, 2273 insertions, 0 deletions

diff --git a/crypto/chacha/Makefile.in b/crypto/chacha/Makefile.in
index 6fb63c1..dd0f36c 100644
--- a/crypto/chacha/Makefile.in
+++ b/crypto/chacha/Makefile.in
@@ -43,6 +43,9 @@ chacha-x86_64.s:	asm/chacha-x86_64.pl
 
 chacha-%.S:	asm/chacha-%.pl;	$(PERL) $< $(PERLASM_SCHEME) $@
 
+chacha-armv4.o:	chacha-armv4.S
+chacha-armv8.o:	chacha-armv8.S
+
 files:
 	$(PERL) $(TOP)/util/files.pl Makefile >> $(TOP)/MINFO
 
diff --git a/crypto/chacha/asm/chacha-armv4.pl b/crypto/chacha/asm/chacha-armv4.pl
new file mode 100755
index 0000000..4d234b7
--- /dev/null
+++ b/crypto/chacha/asm/chacha-armv4.pl
@@ -0,0 +1,1144 @@
+#!/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/.
+# ====================================================================
+#
+# December 2014
+# 
+# ChaCha20 for ARMv4.
+#
+# Performance in cycles per byte out of large buffer.
+#
+#			IALU/gcc-4.4    1xNEON      3xNEON+1xIALU
+#
+# Cortex-A5		19.3(*)/+95%    21.8        14.1
+# Cortex-A8		10.5(*)/+160%   13.9        6.35
+# Cortex-A9		12.9(**)/+110%  14.3        6.50
+# Cortex-A15		11.0/+40%       16.0        5.00
+# Snapdragon S4		11.5/+125%      13.6        4.90
+#
+# (*)	most "favourable" result for aligned data on little-endian
+#	processor, result for misaligned data is 10-15% lower;
+# (**)	this result is a trade-off: it can be improved by 20%,
+#	but then Snapdragon S4 and Cortex-A8 results get
+#	20-25% worse;
+
+$flavour = shift;
+if ($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";
+}
+
+sub AUTOLOAD()		# thunk [simplified] x86-style perlasm
+{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./;
+  my $arg = pop;
+    $arg = "#$arg" if ($arg*1 eq $arg);
+    $code .= "\t$opcode\t".join(',',@_,$arg)."\n";
+}
+
+my @x=map("r$_",(0..7,"x","x","x","x",12,"x",14,"x"));
+my @t=map("r$_",(8..11));
+
+sub ROUND {
+my ($a0,$b0,$c0,$d0)=@_;
+my ($a1,$b1,$c1,$d1)=map(($_&~3)+(($_+1)&3),($a0,$b0,$c0,$d0));
+my ($a2,$b2,$c2,$d2)=map(($_&~3)+(($_+1)&3),($a1,$b1,$c1,$d1));
+my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2));
+my $odd = $d0&1;
+my ($xc,$xc_) = (@t[0..1]);
+my ($xd,$xd_) = $odd ? (@t[2],@x[$d1]) : (@x[$d0],@t[2]);
+my @ret;
+
+	# Consider order in which variables are addressed by their
+	# index:
+	#
+	#       a   b   c   d
+	#
+	#       0   4   8  12 < even round
+	#       1   5   9  13
+	#       2   6  10  14
+	#       3   7  11  15
+	#       0   5  10  15 < odd round
+	#       1   6  11  12
+	#       2   7   8  13
+	#       3   4   9  14
+	#
+	# 'a', 'b' are permanently allocated in registers, @x[0..7],
+	# while 'c's and pair of 'd's are maintained in memory. If
+	# you observe 'c' column, you'll notice that pair of 'c's is
+	# invariant between rounds. This means that we have to reload
+	# them once per round, in the middle. This is why you'll see
+	# bunch of 'c' stores and loads in the middle, but none in
+	# the beginning or end. If you observe 'd' column, you'll
+	# notice that 15 and 13 are reused in next pair of rounds.
+	# This is why these two are chosen for offloading to memory,
+	# to make loads count more.
+							push @ret,(
+	"&add	(@x[$a0],@x[$a0],@x[$b0])",
+	"&mov	($xd,$xd,'ror#16')",
+	 "&add	(@x[$a1],@x[$a1],@x[$b1])",
+	 "&mov	($xd_,$xd_,'ror#16')",
+	"&eor	($xd,$xd,@x[$a0],'ror#16')",
+	 "&eor	($xd_,$xd_,@x[$a1],'ror#16')",
+
+	"&add	($xc,$xc,$xd)",
+	"&mov	(@x[$b0],@x[$b0],'ror#20')",
+	 "&add	($xc_,$xc_,$xd_)",
+	 "&mov	(@x[$b1],@x[$b1],'ror#20')",
+	"&eor	(@x[$b0],@x[$b0],$xc,'ror#20')",
+	 "&eor	(@x[$b1],@x[$b1],$xc_,'ror#20')",
+
+	"&add	(@x[$a0],@x[$a0],@x[$b0])",
+	"&mov	($xd,$xd,'ror#24')",
+	 "&add	(@x[$a1],@x[$a1],@x[$b1])",
+	 "&mov	($xd_,$xd_,'ror#24')",
+	"&eor	($xd,$xd,@x[$a0],'ror#24')",
+	 "&eor	($xd_,$xd_,@x[$a1],'ror#24')",
+
+	"&add	($xc,$xc,$xd)",
+	"&mov	(@x[$b0],@x[$b0],'ror#25')"		);
+							push @ret,(
+	"&str	($xd,'[sp,#4*(16+$d0)]')",
+	"&ldr	($xd,'[sp,#4*(16+$d2)]')"		) if ($odd);
+							push @ret,(
+	 "&add	($xc_,$xc_,$xd_)",
+	 "&mov	(@x[$b1],@x[$b1],'ror#25')"		);
+							push @ret,(
+	 "&str	($xd_,'[sp,#4*(16+$d1)]')",
+	 "&ldr	($xd_,'[sp,#4*(16+$d3)]')"		) if (!$odd);
+							push @ret,(
+	"&eor	(@x[$b0],@x[$b0],$xc,'ror#25')",
+	 "&eor	(@x[$b1],@x[$b1],$xc_,'ror#25')"	);
+
+	$xd=@x[$d2]					if (!$odd);
+	$xd_=@x[$d3]					if ($odd);
+							push @ret,(
+	"&str	($xc,'[sp,#4*(16+$c0)]')",
+	"&ldr	($xc,'[sp,#4*(16+$c2)]')",
+	"&add	(@x[$a2],@x[$a2],@x[$b2])",
+	"&mov	($xd,$xd,'ror#16')",
+	 "&str	($xc_,'[sp,#4*(16+$c1)]')",
+	 "&ldr	($xc_,'[sp,#4*(16+$c3)]')",
+	 "&add	(@x[$a3],@x[$a3],@x[$b3])",
+	 "&mov	($xd_,$xd_,'ror#16')",
+	"&eor	($xd,$xd,@x[$a2],'ror#16')",
+	 "&eor	($xd_,$xd_,@x[$a3],'ror#16')",
+
+	"&add	($xc,$xc,$xd)",
+	"&mov	(@x[$b2],@x[$b2],'ror#20')",
+	 "&add	($xc_,$xc_,$xd_)",
+	 "&mov	(@x[$b3],@x[$b3],'ror#20')",
+	"&eor	(@x[$b2],@x[$b2],$xc,'ror#20')",
+	 "&eor	(@x[$b3],@x[$b3],$xc_,'ror#20')",
+
+	"&add	(@x[$a2],@x[$a2],@x[$b2])",
+	"&mov	($xd,$xd,'ror#24')",
+	 "&add	(@x[$a3],@x[$a3],@x[$b3])",
+	 "&mov	($xd_,$xd_,'ror#24')",
+	"&eor	($xd,$xd,@x[$a2],'ror#24')",
+	 "&eor	($xd_,$xd_,@x[$a3],'ror#24')",
+
+	"&add	($xc,$xc,$xd)",
+	"&mov	(@x[$b2],@x[$b2],'ror#25')",
+	 "&add	($xc_,$xc_,$xd_)",
+	 "&mov	(@x[$b3],@x[$b3],'ror#25')",
+	"&eor	(@x[$b2],@x[$b2],$xc,'ror#25')",
+	 "&eor	(@x[$b3],@x[$b3],$xc_,'ror#25')"	);
+
+	@ret;
+}
+
+$code.=<<___;
+#include "arm_arch.h"
+
+.text
+#if defined(__thumb2__)
+.syntax	unified
+.thumb
+#else
+.code	32
+#endif
+
+#if defined(__thumb2__) || defined(__clang__)
+#define ldrhsb	ldrbhs
+#endif
+
+.align	5
+.Lsigma:
+.long	0x61707865,0x3320646e,0x79622d32,0x6b206574	@ endian-neutral
+.Lone:
+.long	1,0,0,0
+#if __ARM_MAX_ARCH__>=7
+.LOPENSSL_armcap:
+.word   OPENSSL_armcap_P-.LChaCha20_ctr32
+#else
+.word	-1
+#endif
+
+.globl	ChaCha20_ctr32
+.type	ChaCha20_ctr32,%function
+.align	5
+ChaCha20_ctr32:
+.LChaCha20_ctr32:
+	ldr	r12,[sp,#0]		@ pull pointer to counter and nonce
+	stmdb	sp!,{r0-r2,r4-r11,lr}
+#if __ARM_ARCH__<7 && !defined(__thumb2__)
+	sub	r14,pc,#16		@ ChaCha20_ctr32
+#else
+	adr	r14,.LChaCha20_ctr32
+#endif
+#if __ARM_MAX_ARCH__>=7
+	cmp	r2,#192			@ test len
+	bls	.Lshort
+	ldr	r4,[r14,#-32]
+	ldr	r4,[r14,r4]
+# ifdef	__APPLE__
+	ldr	r4,[r4]
+# endif
+	tst	r4,#1
+	bne	.LChaCha20_neon
+.Lshort:
+#endif
+	ldmia	r12,{r4-r7}		@ load counter and nonce
+	sub	sp,sp,#4*(16)		@ off-load area
+	sub	r14,r14,#64		@ .Lsigma
+	stmdb	sp!,{r4-r7}		@ copy counter and nonce
+	ldmia	r3,{r4-r11}		@ load key
+	ldmia	r14,{r0-r3}		@ load sigma
+	stmdb	sp!,{r4-r11}		@ copy key
+	stmdb	sp!,{r0-r3}		@ copy sigma
+	str	r10,[sp,#4*(16+10)]	@ off-load "@x[10]"
+	str	r11,[sp,#4*(16+11)]	@ off-load "@x[11]"
+	b	.Loop_outer_enter
+
+.align	4
+.Loop_outer:
+	ldmia	sp,{r0-r9}		@ load key material
+	str	@t[3],[sp,#4*(32+2)]	@ save len
+	str	r12,  [sp,#4*(32+1)]	@ save inp
+	str	r14,  [sp,#4*(32+0)]	@ save out
+.Loop_outer_enter:
+	ldr	@t[3], [sp,#4*(15)]
+	ldr	@x[12],[sp,#4*(12)]	@ modulo-scheduled load
+	ldr	@t[2], [sp,#4*(13)]
+	ldr	@x[14],[sp,#4*(14)]
+	str	@t[3], [sp,#4*(16+15)]
+	mov	@t[3],#10
+	b	.Loop
+
+.align	4
+.Loop:
+	subs	@t[3],@t[3],#1
+___
+	foreach (&ROUND(0, 4, 8,12)) { eval; }
+	foreach (&ROUND(0, 5,10,15)) { eval; }
+$code.=<<___;
+	bne	.Loop
+
+	ldr	@t[3],[sp,#4*(32+2)]	@ load len
+
+	str	@t[0], [sp,#4*(16+8)]	@ modulo-scheduled store
+	str	@t[1], [sp,#4*(16+9)]
+	str	@x[12],[sp,#4*(16+12)]
+	str	@t[2], [sp,#4*(16+13)]
+	str	@x[14],[sp,#4*(16+14)]
+
+	@ at this point we have first half of 512-bit result in
+	@ @x[0-7] and second half at sp+4*(16+8)
+
+	cmp	@t[3],#64		@ done yet?
+#ifdef	__thumb2__
+	itete	lo
+#endif
+	addlo	r12,sp,#4*(0)		@ shortcut or ...
+	ldrhs	r12,[sp,#4*(32+1)]	@ ... load inp
+	addlo	r14,sp,#4*(0)		@ shortcut or ...
+	ldrhs	r14,[sp,#4*(32+0)]	@ ... load out
+
+	ldr	@t[0],[sp,#4*(0)]	@ load key material
+	ldr	@t[1],[sp,#4*(1)]
+
+#if __ARM_ARCH__>=6 || !defined(__ARMEB__)
+# if __ARM_ARCH__<7
+	orr	@t[2],r12,r14
+	tst	@t[2],#3		@ are input and output aligned?
+	ldr	@t[2],[sp,#4*(2)]
+	bne	.Lunaligned
+	cmp	@t[3],#64		@ restore flags
+# else
+	ldr	@t[2],[sp,#4*(2)]
+# endif
+	ldr	@t[3],[sp,#4*(3)]
+
+	add	@x[0],@x[0],@t[0]	@ accumulate key material
+	add	@x[1],@x[1],@t[1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+
+	add	@x[2],@x[2],@t[2]
+	add	@x[3],@x[3],@t[3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[0],@x[0]
+	rev	@x[1],@x[1]
+	rev	@x[2],@x[2]
+	rev	@x[3],@x[3]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[0],@x[0],@t[0]	@ xor with input
+	eorhs	@x[1],@x[1],@t[1]
+	 add	@t[0],sp,#4*(4)
+	str	@x[0],[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[2],@x[2],@t[2]
+	eorhs	@x[3],@x[3],@t[3]
+	 ldmia	@t[0],{@t[0]-@t[3]}	@ load key material
+	str	@x[1],[r14,#-12]
+	str	@x[2],[r14,#-8]
+	str	@x[3],[r14,#-4]
+
+	add	@x[4],@x[4],@t[0]	@ accumulate key material
+	add	@x[5],@x[5],@t[1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+	add	@x[6],@x[6],@t[2]
+	add	@x[7],@x[7],@t[3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[4],@x[4]
+	rev	@x[5],@x[5]
+	rev	@x[6],@x[6]
+	rev	@x[7],@x[7]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[4],@x[4],@t[0]
+	eorhs	@x[5],@x[5],@t[1]
+	 add	@t[0],sp,#4*(8)
+	str	@x[4],[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[6],@x[6],@t[2]
+	eorhs	@x[7],@x[7],@t[3]
+	str	@x[5],[r14,#-12]
+	 ldmia	@t[0],{@t[0]-@t[3]}	@ load key material
+	str	@x[6],[r14,#-8]
+	 add	@x[0],sp,#4*(16+8)
+	str	@x[7],[r14,#-4]
+
+	ldmia	@x[0],{@x[0]-@x[7]}	@ load second half
+
+	add	@x[0],@x[0],@t[0]	@ accumulate key material
+	add	@x[1],@x[1],@t[1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	 strhi	@t[2],[sp,#4*(16+10)]	@ copy "@x[10]" while at it
+	 strhi	@t[3],[sp,#4*(16+11)]	@ copy "@x[11]" while at it
+	add	@x[2],@x[2],@t[2]
+	add	@x[3],@x[3],@t[3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[0],@x[0]
+	rev	@x[1],@x[1]
+	rev	@x[2],@x[2]
+	rev	@x[3],@x[3]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[0],@x[0],@t[0]
+	eorhs	@x[1],@x[1],@t[1]
+	 add	@t[0],sp,#4*(12)
+	str	@x[0],[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[2],@x[2],@t[2]
+	eorhs	@x[3],@x[3],@t[3]
+	str	@x[1],[r14,#-12]
+	 ldmia	@t[0],{@t[0]-@t[3]}	@ load key material
+	str	@x[2],[r14,#-8]
+	str	@x[3],[r14,#-4]
+
+	add	@x[4],@x[4],@t[0]	@ accumulate key material
+	add	@x[5],@x[5],@t[1]
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	 addhi	@t[0],@t[0],#1		@ next counter value
+	 strhi	@t[0],[sp,#4*(12)]	@ save next counter value
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+	add	@x[6],@x[6],@t[2]
+	add	@x[7],@x[7],@t[3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[4],@x[4]
+	rev	@x[5],@x[5]
+	rev	@x[6],@x[6]
+	rev	@x[7],@x[7]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[4],@x[4],@t[0]
+	eorhs	@x[5],@x[5],@t[1]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 ldrhi	@t[0],[sp,#4*(32+2)]	@ re-load len
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[6],@x[6],@t[2]
+	eorhs	@x[7],@x[7],@t[3]
+	str	@x[4],[r14],#16		@ store output
+	str	@x[5],[r14,#-12]
+# ifdef	__thumb2__
+	it	hs
+# endif
+	 subhs	@t[3],@t[0],#64		@ len-=64
+	str	@x[6],[r14,#-8]
+	str	@x[7],[r14,#-4]
+	bhi	.Loop_outer
+
+	beq	.Ldone
+# if __ARM_ARCH__<7
+	b	.Ltail
+
+.align	4
+.Lunaligned:				@ unaligned endian-neutral path
+	cmp	@t[3],#64		@ restore flags
+# endif
+#endif
+#if __ARM_ARCH__<7
+	ldr	@t[3],[sp,#4*(3)]
+___
+for ($i=0;$i<16;$i+=4) {
+my $j=$i&0x7;
+
+$code.=<<___	if ($i==4);
+	add	@x[0],sp,#4*(16+8)
+___
+$code.=<<___	if ($i==8);
+	ldmia	@x[0],{@x[0]-@x[7]}		@ load second half
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	strhi	@t[2],[sp,#4*(16+10)]		@ copy "@x[10]"
+	strhi	@t[3],[sp,#4*(16+11)]		@ copy "@x[11]"
+___
+$code.=<<___;
+	add	@x[$j+0],@x[$j+0],@t[0]		@ accumulate key material
+___
+$code.=<<___	if ($i==12);
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	addhi	@t[0],@t[0],#1			@ next counter value
+	strhi	@t[0],[sp,#4*(12)]		@ save next counter value
+___
+$code.=<<___;
+	add	@x[$j+1],@x[$j+1],@t[1]
+	add	@x[$j+2],@x[$j+2],@t[2]
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	@t[0],@t[0],@t[0]		@ zero or ...
+	ldrhsb	@t[0],[r12],#16			@ ... load input
+	eorlo	@t[1],@t[1],@t[1]
+	ldrhsb	@t[1],[r12,#-12]
+
+	add	@x[$j+3],@x[$j+3],@t[3]
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	@t[2],@t[2],@t[2]
+	ldrhsb	@t[2],[r12,#-8]
+	eorlo	@t[3],@t[3],@t[3]
+	ldrhsb	@t[3],[r12,#-4]
+
+	eor	@x[$j+0],@t[0],@x[$j+0]		@ xor with input (or zero)
+	eor	@x[$j+1],@t[1],@x[$j+1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[0],[r12,#-15]		@ load more input
+	ldrhsb	@t[1],[r12,#-11]
+	eor	@x[$j+2],@t[2],@x[$j+2]
+	 strb	@x[$j+0],[r14],#16		@ store output
+	eor	@x[$j+3],@t[3],@x[$j+3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[2],[r12,#-7]
+	ldrhsb	@t[3],[r12,#-3]
+	 strb	@x[$j+1],[r14,#-12]
+	eor	@x[$j+0],@t[0],@x[$j+0],lsr#8
+	 strb	@x[$j+2],[r14,#-8]
+	eor	@x[$j+1],@t[1],@x[$j+1],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[0],[r12,#-14]		@ load more input
+	ldrhsb	@t[1],[r12,#-10]
+	 strb	@x[$j+3],[r14,#-4]
+	eor	@x[$j+2],@t[2],@x[$j+2],lsr#8
+	 strb	@x[$j+0],[r14,#-15]
+	eor	@x[$j+3],@t[3],@x[$j+3],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[2],[r12,#-6]
+	ldrhsb	@t[3],[r12,#-2]
+	 strb	@x[$j+1],[r14,#-11]
+	eor	@x[$j+0],@t[0],@x[$j+0],lsr#8
+	 strb	@x[$j+2],[r14,#-7]
+	eor	@x[$j+1],@t[1],@x[$j+1],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[0],[r12,#-13]		@ load more input
+	ldrhsb	@t[1],[r12,#-9]
+	 strb	@x[$j+3],[r14,#-3]
+	eor	@x[$j+2],@t[2],@x[$j+2],lsr#8
+	 strb	@x[$j+0],[r14,#-14]
+	eor	@x[$j+3],@t[3],@x[$j+3],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[2],[r12,#-5]
+	ldrhsb	@t[3],[r12,#-1]
+	 strb	@x[$j+1],[r14,#-10]
+	 strb	@x[$j+2],[r14,#-6]
+	eor	@x[$j+0],@t[0],@x[$j+0],lsr#8
+	 strb	@x[$j+3],[r14,#-2]
+	eor	@x[$j+1],@t[1],@x[$j+1],lsr#8
+	 strb	@x[$j+0],[r14,#-13]
+	eor	@x[$j+2],@t[2],@x[$j+2],lsr#8
+	 strb	@x[$j+1],[r14,#-9]
+	eor	@x[$j+3],@t[3],@x[$j+3],lsr#8
+	 strb	@x[$j+2],[r14,#-5]
+	 strb	@x[$j+3],[r14,#-1]
+___
+$code.=<<___	if ($i<12);
+	add	@t[0],sp,#4*(4+$i)
+	ldmia	@t[0],{@t[0]-@t[3]}		@ load key material
+___
+}
+$code.=<<___;
+# ifdef	__thumb2__
+	it	hi
+# endif
+	ldrhi	@t[0],[sp,#4*(32+2)]		@ re-load len
+# ifdef	__thumb2__
+	it	hs
+# endif
+	subhs	@t[3],@t[0],#64			@ len-=64
+	bhi	.Loop_outer
+
+	beq	.Ldone
+#endif
+
+.Ltail:
+	ldr	r12,[sp,#4*(32+1)]	@ load inp
+	add	@t[2],sp,#4*(0)
+	ldr	r14,[sp,#4*(32+0)]	@ load out
+
+.Loop_tail:
+	ldrb	@t[0],[@t[2]],#1	@ read buffer on stack
+	ldrb	@t[1],[r12],#1		@ read input
+	subs	@t[3],@t[3],#1
+	eor	@t[0],@t[0],@t[1]
+	strb	@t[0],[r14],#1		@ store output
+	bne	.Loop_tail
+
+.Ldone:
+	add	sp,sp,#4*(32+3)
+	ldmia	sp!,{r4-r11,pc}
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+___
+
+{{{
+my ($a0,$b0,$c0,$d0,$a1,$b1,$c1,$d1,$a2,$b2,$c2,$d2,$t0,$t1,$t2,$t3) =
+    map("q$_",(0..15));
+
+sub NEONROUND {
+my $odd = pop;
+my ($a,$b,$c,$d,$t)=@_;
+
+	(
+	"&vadd_i32	($a,$a,$b)",
+	"&veor		($d,$d,$a)",
+	"&vrev32_16	($d,$d)",	# vrot ($d,16)
+
+	"&vadd_i32	($c,$c,$d)",
+	"&veor		($t,$b,$c)",
+	"&vshr_u32	($b,$t,20)",
+	"&vsli_32	($b,$t,12)",
+
+	"&vadd_i32	($a,$a,$b)",
+	"&veor		($t,$d,$a)",
+	"&vshr_u32	($d,$t,24)",
+	"&vsli_32	($d,$t,8)",
+
+	"&vadd_i32	($c,$c,$d)",
+	"&veor		($t,$b,$c)",
+	"&vshr_u32	($b,$t,25)",
+	"&vsli_32	($b,$t,7)",
+
+	"&vext_8	($c,$c,$c,8)",
+	"&vext_8	($b,$b,$b,$odd?12:4)",
+	"&vext_8	($d,$d,$d,$odd?4:12)"
+	);
+}
+
+$code.=<<___;
+#if __ARM_MAX_ARCH__>=7
+.arch	armv7-a
+.fpu	neon
+
+.type	ChaCha20_neon,%function
+.align	5
+ChaCha20_neon:
+	ldr		r12,[sp,#0]		@ pull pointer to counter and nonce
+	stmdb		sp!,{r0-r2,r4-r11,lr}
+.LChaCha20_neon:
+	adr		r14,.Lsigma
+	vstmdb		sp!,{d8-d15}		@ ABI spec says so
+	stmdb		sp!,{r0-r3}
+
+	vld1.32		{$b0-$c0},[r3]		@ load key
+	ldmia		r3,{r4-r11}		@ load key
+
+	sub		sp,sp,#4*(16+16)
+	vld1.32		{$d0},[r12]		@ load counter and nonce
+	add		r12,sp,#4*8
+	ldmia		r14,{r0-r3}		@ load sigma
+	vld1.32		{$a0},[r14]!		@ load sigma
+	vld1.32		{$t0},[r14]		@ one
+	vst1.32		{$c0-$d0},[r12]		@ copy 1/2key|counter|nonce
+	vst1.32		{$a0-$b0},[sp]		@ copy sigma|1/2key
+
+	str		r10,[sp,#4*(16+10)]	@ off-load "@x[10]"
+	str		r11,[sp,#4*(16+11)]	@ off-load "@x[11]"
+	vshl.i32	$t1#lo,$t0#lo,#1	@ two
+	vstr		$t0#lo,[sp,#4*(16+0)]
+	vshl.i32	$t2#lo,$t0#lo,#2	@ four
+	vstr		$t1#lo,[sp,#4*(16+2)]
+	vmov		$a1,$a0
+	vstr		$t2#lo,[sp,#4*(16+4)]
+	vmov		$a2,$a0
+	vmov		$b1,$b0
+	vmov		$b2,$b0
+	b		.Loop_neon_enter
+
+.align	4
+.Loop_neon_outer:
+	ldmia		sp,{r0-r9}		@ load key material
+	cmp		@t[3],#64*2		@ if len<=64*2
+	bls		.Lbreak_neon		@ switch to integer-only
+	vmov		$a1,$a0
+	str		@t[3],[sp,#4*(32+2)]	@ save len
+	vmov		$a2,$a0
+	str		r12,  [sp,#4*(32+1)]	@ save inp
+	vmov		$b1,$b0
+	str		r14,  [sp,#4*(32+0)]	@ save out
+	vmov		$b2,$b0
+.Loop_neon_enter:
+	ldr		@t[3], [sp,#4*(15)]
+	vadd.i32	$d1,$d0,$t0		@ counter+1
+	ldr		@x[12],[sp,#4*(12)]	@ modulo-scheduled load
+	vmov		$c1,$c0
+	ldr		@t[2], [sp,#4*(13)]
+	vmov		$c2,$c0
+	ldr		@x[14],[sp,#4*(14)]
+	vadd.i32	$d2,$d1,$t0		@ counter+2
+	str		@t[3], [sp,#4*(16+15)]
+	mov		@t[3],#10
+	add		@x[12],@x[12],#3	@ counter+3 
+	b		.Loop_neon
+
+.align	4
+.Loop_neon:
+	subs		@t[3],@t[3],#1
+___
+	my @thread0=&NEONROUND($a0,$b0,$c0,$d0,$t0,0);
+	my @thread1=&NEONROUND($a1,$b1,$c1,$d1,$t1,0);
+	my @thread2=&NEONROUND($a2,$b2,$c2,$d2,$t2,0);
+	my @thread3=&ROUND(0,4,8,12);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+
+	@thread0=&NEONROUND($a0,$b0,$c0,$d0,$t0,1);
+	@thread1=&NEONROUND($a1,$b1,$c1,$d1,$t1,1);
+	@thread2=&NEONROUND($a2,$b2,$c2,$d2,$t2,1);
+	@thread3=&ROUND(0,5,10,15);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+$code.=<<___;
+	bne		.Loop_neon
+
+	add		@t[3],sp,#32
+	vld1.32		{$t0-$t1},[sp]		@ load key material
+	vld1.32		{$t2-$t3},[@t[3]]
+
+	ldr		@t[3],[sp,#4*(32+2)]	@ load len
+
+	str		@t[0], [sp,#4*(16+8)]	@ modulo-scheduled store
+	str		@t[1], [sp,#4*(16+9)]
+	str		@x[12],[sp,#4*(16+12)]
+	str		@t[2], [sp,#4*(16+13)]
+	str		@x[14],[sp,#4*(16+14)]
+
+	@ at this point we have first half of 512-bit result in
+	@ @x[0-7] and second half at sp+4*(16+8)
+
+	ldr		r12,[sp,#4*(32+1)]	@ load inp
+	ldr		r14,[sp,#4*(32+0)]	@ load out
+
+	vadd.i32	$a0,$a0,$t0		@ accumulate key material
+	vadd.i32	$a1,$a1,$t0
+	vadd.i32	$a2,$a2,$t0
+	vldr		$t0#lo,[sp,#4*(16+0)]	@ one
+
+	vadd.i32	$b0,$b0,$t1
+	vadd.i32	$b1,$b1,$t1
+	vadd.i32	$b2,$b2,$t1
+	vldr		$t1#lo,[sp,#4*(16+2)]	@ two
+
+	vadd.i32	$c0,$c0,$t2
+	vadd.i32	$c1,$c1,$t2
+	vadd.i32	$c2,$c2,$t2
+	vadd.i32	$d1#lo,$d1#lo,$t0#lo	@ counter+1
+	vadd.i32	$d2#lo,$d2#lo,$t1#lo	@ counter+2
+
+	vadd.i32	$d0,$d0,$t3
+	vadd.i32	$d1,$d1,$t3
+	vadd.i32	$d2,$d2,$t3
+
+	cmp		@t[3],#64*4
+	blo		.Ltail_neon
+
+	vld1.8		{$t0-$t1},[r12]!	@ load input
+	 mov		@t[3],sp
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0		@ xor with input
+	veor		$b0,$b0,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a1,$a1,$t0
+	 vst1.8		{$a0-$b0},[r14]!	@ store output
+	veor		$b1,$b1,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c1,$c1,$t2
+	 vst1.8		{$c0-$d0},[r14]!
+	veor		$d1,$d1,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a2,$a2,$t0
+	 vld1.32	{$a0-$b0},[@t[3]]!	@ load for next iteration
+	 veor		$t0#hi,$t0#hi,$t0#hi
+	 vldr		$t0#lo,[sp,#4*(16+4)]	@ four
+	veor		$b2,$b2,$t1
+	 vld1.32	{$c0-$d0},[@t[3]]
+	veor		$c2,$c2,$t2
+	 vst1.8		{$a1-$b1},[r14]!
+	veor		$d2,$d2,$t3
+	 vst1.8		{$c1-$d1},[r14]!
+
+	vadd.i32	$d0#lo,$d0#lo,$t0#lo	@ next counter value
+	vldr		$t0#lo,[sp,#4*(16+0)]	@ one
+
+	ldmia		sp,{@t[0]-@t[3]}	@ load key material
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	ldr		@t[0],[r12],#16		@ load input
+	 vst1.8		{$a2-$b2},[r14]!
+	add		@x[1],@x[1],@t[1]
+	ldr		@t[1],[r12,#-12]
+	 vst1.8		{$c2-$d2},[r14]!
+	add		@x[2],@x[2],@t[2]
+	ldr		@t[2],[r12,#-8]
+	add		@x[3],@x[3],@t[3]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+# endif
+	eor		@x[0],@x[0],@t[0]	@ xor with input
+	 add		@t[0],sp,#4*(4)
+	eor		@x[1],@x[1],@t[1]
+	str		@x[0],[r14],#16		@ store output
+	eor		@x[2],@x[2],@t[2]
+	str		@x[1],[r14,#-12]
+	eor		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+	str		@x[2],[r14,#-8]
+	str		@x[3],[r14,#-4]
+
+	add		@x[4],@x[4],@t[0]	@ accumulate key material
+	ldr		@t[0],[r12],#16		@ load input
+	add		@x[5],@x[5],@t[1]
+	ldr		@t[1],[r12,#-12]
+	add		@x[6],@x[6],@t[2]
+	ldr		@t[2],[r12,#-8]
+	add		@x[7],@x[7],@t[3]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	eor		@x[4],@x[4],@t[0]
+	 add		@t[0],sp,#4*(8)
+	eor		@x[5],@x[5],@t[1]
+	str		@x[4],[r14],#16		@ store output
+	eor		@x[6],@x[6],@t[2]
+	str		@x[5],[r14,#-12]
+	eor		@x[7],@x[7],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+	str		@x[6],[r14,#-8]
+	 add		@x[0],sp,#4*(16+8)
+	str		@x[7],[r14,#-4]
+
+	ldmia		@x[0],{@x[0]-@x[7]}	@ load second half
+
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	ldr		@t[0],[r12],#16		@ load input
+	add		@x[1],@x[1],@t[1]
+	ldr		@t[1],[r12,#-12]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 strhi		@t[2],[sp,#4*(16+10)]	@ copy "@x[10]" while at it
+	add		@x[2],@x[2],@t[2]
+	ldr		@t[2],[r12,#-8]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 strhi		@t[3],[sp,#4*(16+11)]	@ copy "@x[11]" while at it
+	add		@x[3],@x[3],@t[3]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+# endif
+	eor		@x[0],@x[0],@t[0]
+	 add		@t[0],sp,#4*(12)
+	eor		@x[1],@x[1],@t[1]
+	str		@x[0],[r14],#16		@ store output
+	eor		@x[2],@x[2],@t[2]
+	str		@x[1],[r14,#-12]
+	eor		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+	str		@x[2],[r14,#-8]
+	str		@x[3],[r14,#-4]
+
+	add		@x[4],@x[4],@t[0]	@ accumulate key material
+	 add		@t[0],@t[0],#4		@ next counter value
+	add		@x[5],@x[5],@t[1]
+	 str		@t[0],[sp,#4*(12)]	@ save next counter value
+	ldr		@t[0],[r12],#16		@ load input
+	add		@x[6],@x[6],@t[2]
+	 add		@x[4],@x[4],#3		@ counter+3
+	ldr		@t[1],[r12,#-12]
+	add		@x[7],@x[7],@t[3]
+	ldr		@t[2],[r12,#-8]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	eor		@x[4],@x[4],@t[0]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 ldrhi		@t[0],[sp,#4*(32+2)]	@ re-load len
+	eor		@x[5],@x[5],@t[1]
+	eor		@x[6],@x[6],@t[2]
+	str		@x[4],[r14],#16		@ store output
+	eor		@x[7],@x[7],@t[3]
+	str		@x[5],[r14,#-12]
+	 sub		@t[3],@t[0],#64*4	@ len-=64*4
+	str		@x[6],[r14,#-8]
+	str		@x[7],[r14,#-4]
+	bhi		.Loop_neon_outer
+
+	b		.Ldone_neon
+
+.align	4
+.Lbreak_neon:
+	@ harmonize NEON and integer-only stack frames: load data
+	@ from NEON frame, but save to integer-only one; distance
+	@ between the two is 4*(32+4+16-32)=4*(20).
+
+	str		@t[3], [sp,#4*(20+32+2)]	@ save len
+	 add		@t[3],sp,#4*(32+4)
+	str		r12,   [sp,#4*(20+32+1)]	@ save inp
+	str		r14,   [sp,#4*(20+32+0)]	@ save out
+
+	ldr		@x[12],[sp,#4*(16+10)]
+	ldr		@x[14],[sp,#4*(16+11)]
+	 vldmia		@t[3],{d8-d15}			@ fulfill ABI requirement
+	str		@x[12],[sp,#4*(20+16+10)]	@ copy "@x[10]"
+	str		@x[14],[sp,#4*(20+16+11)]	@ copy "@x[11]"
+
+	ldr		@t[3], [sp,#4*(15)]
+	ldr		@x[12],[sp,#4*(12)]		@ modulo-scheduled load
+	ldr		@t[2], [sp,#4*(13)]
+	ldr		@x[14],[sp,#4*(14)]
+	str		@t[3], [sp,#4*(20+16+15)]
+	add		@t[3],sp,#4*(20)
+	vst1.32		{$a0-$b0},[@t[3]]!		@ copy key
+	add		sp,sp,#4*(20)			@ switch frame
+	vst1.32		{$c0-$d0},[@t[3]]
+	mov		@t[3],#10
+	b		.Loop				@ go integer-only
+
+.align	4
+.Ltail_neon:
+	cmp		@t[3],#64*3
+	bhs		.L192_or_more_neon
+	cmp		@t[3],#64*2
+	bhs		.L128_or_more_neon
+	cmp		@t[3],#64*1
+	bhs		.L64_or_more_neon
+
+	add		@t[0],sp,#4*(8)
+	vst1.8		{$a0-$b0},[sp]
+	add		@t[2],sp,#4*(0)
+	vst1.8		{$c0-$d0},[@t[0]]
+	b		.Loop_tail_neon
+
+.align	4
+.L64_or_more_neon:
+	vld1.8		{$t0-$t1},[r12]!
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0
+	veor		$b0,$b0,$t1
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vst1.8		{$a0-$b0},[r14]!
+	vst1.8		{$c0-$d0},[r14]!
+
+	beq		.Ldone_neon
+
+	add		@t[0],sp,#4*(8)
+	vst1.8		{$a1-$b1},[sp]
+	add		@t[2],sp,#4*(0)
+	vst1.8		{$c1-$d1},[@t[0]]
+	sub		@t[3],@t[3],#64*1	@ len-=64*1
+	b		.Loop_tail_neon
+
+.align	4
+.L128_or_more_neon:
+	vld1.8		{$t0-$t1},[r12]!
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0
+	veor		$b0,$b0,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a1,$a1,$t0
+	veor		$b1,$b1,$t1
+	 vst1.8		{$a0-$b0},[r14]!
+	veor		$c1,$c1,$t2
+	 vst1.8		{$c0-$d0},[r14]!
+	veor		$d1,$d1,$t3
+	vst1.8		{$a1-$b1},[r14]!
+	vst1.8		{$c1-$d1},[r14]!
+
+	beq		.Ldone_neon
+
+	add		@t[0],sp,#4*(8)
+	vst1.8		{$a2-$b2},[sp]
+	add		@t[2],sp,#4*(0)
+	vst1.8		{$c2-$d2},[@t[0]]
+	sub		@t[3],@t[3],#64*2	@ len-=64*2
+	b		.Loop_tail_neon
+
+.align	4
+.L192_or_more_neon:
+	vld1.8		{$t0-$t1},[r12]!
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0
+	veor		$b0,$b0,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a1,$a1,$t0
+	veor		$b1,$b1,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c1,$c1,$t2
+	 vst1.8		{$a0-$b0},[r14]!
+	veor		$d1,$d1,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a2,$a2,$t0
+	 vst1.8		{$c0-$d0},[r14]!
+	veor		$b2,$b2,$t1
+	 vst1.8		{$a1-$b1},[r14]!
+	veor		$c2,$c2,$t2
+	 vst1.8		{$c1-$d1},[r14]!
+	veor		$d2,$d2,$t3
+	vst1.8		{$a2-$b2},[r14]!
+	vst1.8		{$c2-$d2},[r14]!
+
+	beq		.Ldone_neon
+
+	ldmia		sp,{@t[0]-@t[3]}	@ load key material
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	 add		@t[0],sp,#4*(4)
+	add		@x[1],@x[1],@t[1]
+	add		@x[2],@x[2],@t[2]
+	add		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+
+	add		@x[4],@x[4],@t[0]	@ accumulate key material
+	 add		@t[0],sp,#4*(8)
+	add		@x[5],@x[5],@t[1]
+	add		@x[6],@x[6],@t[2]
+	add		@x[7],@x[7],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	stmia		sp,{@x[0]-@x[7]}
+	 add		@x[0],sp,#4*(16+8)
+
+	ldmia		@x[0],{@x[0]-@x[7]}	@ load second half
+
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	 add		@t[0],sp,#4*(12)
+	add		@x[1],@x[1],@t[1]
+	add		@x[2],@x[2],@t[2]
+	add		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+
+	add		@x[4],@x[4],@t[0]	@ accumulate key material
+	 add		@t[0],sp,#4*(8)
+	add		@x[5],@x[5],@t[1]
+	 add		@x[4],@x[4],#3		@ counter+3
+	add		@x[6],@x[6],@t[2]
+	add		@x[7],@x[7],@t[3]
+	 ldr		@t[3],[sp,#4*(32+2)]	@ re-load len
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	stmia		@t[0],{@x[0]-@x[7]}
+	 add		@t[2],sp,#4*(0)
+	 sub		@t[3],@t[0],#64*3	@ len-=64*3
+
+.Loop_tail_neon:
+	ldrb		@t[0],[@t[2]],#1	@ read buffer on stack
+	ldrb		@t[1],[r12],#1		@ read input
+	subs		@t[3],@t[3],#1
+	eor		@t[0],@t[0],@t[1]
+	strb		@t[0],[r14],#1		@ store ouput
+	bne		.Loop_tail_neon
+
+.Ldone_neon:
+	add		sp,sp,#4*(32+4)
+	vldmia		sp,{d8-d15}
+	add		sp,sp,#4*(16+3)
+	ldmia		sp!,{r4-r11,pc}
+.size	ChaCha20_neon,.-ChaCha20_neon
+.comm	OPENSSL_armcap_P,4,4
+#endif
+___
+}}}
+
+foreach (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
+
+	print $_,"\n";
+}
+close STDOUT;
diff --git a/crypto/chacha/asm/chacha-armv8.pl b/crypto/chacha/asm/chacha-armv8.pl
new file mode 100755
index 0000000..6ddb31f
--- /dev/null
+++ b/crypto/chacha/asm/chacha-armv8.pl
@@ -0,0 +1,1126 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# June 2015
+# 
+# ChaCha20 for ARMv8.
+#
+# Performance in cycles per byte out of large buffer.
+#
+#			IALU/gcc-4.9    3xNEON+1xIALU	6xNEON+2xIALU
+#
+# Apple A7		5.50/+49%       3.33            1.70
+# Cortex-A53		8.40/+80%       4.72		4.72(*)
+# Cortex-A57		8.06/+43%       4.90            4.43(**)
+# Denver		4.50/+82%       2.63		2.67(*)
+# X-Gene		9.50/+46%       8.82		8.89(*)
+#
+# (*)	it's expected that doubling interleave factor doesn't help
+#	all processors, only those with higher NEON latency and
+#	higher instruction issue rate;
+# (**)	expected improvement was actually higher;
+
+$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;
+
+sub AUTOLOAD()		# thunk [simplified] x86-style perlasm
+{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./;
+  my $arg = pop;
+    $arg = "#$arg" if ($arg*1 eq $arg);
+    $code .= "\t$opcode\t".join(',',@_,$arg)."\n";
+}
+
+my ($out,$inp,$len,$key,$ctr) = map("x$_",(0..4));
+
+my @x=map("x$_",(5..17,19..21));
+my @d=map("x$_",(22..28,30));
+
+sub ROUND {
+my ($a0,$b0,$c0,$d0)=@_;
+my ($a1,$b1,$c1,$d1)=map(($_&~3)+(($_+1)&3),($a0,$b0,$c0,$d0));
+my ($a2,$b2,$c2,$d2)=map(($_&~3)+(($_+1)&3),($a1,$b1,$c1,$d1));
+my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2));
+
+    (
+	"&add_32	(@x[$a0],@x[$a0],@x[$b0])",
+	 "&add_32	(@x[$a1],@x[$a1],@x[$b1])",
+	  "&add_32	(@x[$a2],@x[$a2],@x[$b2])",
+	   "&add_32	(@x[$a3],@x[$a3],@x[$b3])",
+	"&eor_32	(@x[$d0],@x[$d0],@x[$a0])",
+	 "&eor_32	(@x[$d1],@x[$d1],@x[$a1])",
+	  "&eor_32	(@x[$d2],@x[$d2],@x[$a2])",
+	   "&eor_32	(@x[$d3],@x[$d3],@x[$a3])",
+	"&ror_32	(@x[$d0],@x[$d0],16)",
+	 "&ror_32	(@x[$d1],@x[$d1],16)",
+	  "&ror_32	(@x[$d2],@x[$d2],16)",
+	   "&ror_32	(@x[$d3],@x[$d3],16)",
+
+	"&add_32	(@x[$c0],@x[$c0],@x[$d0])",
+	 "&add_32	(@x[$c1],@x[$c1],@x[$d1])",
+	  "&add_32	(@x[$c2],@x[$c2],@x[$d2])",
+	   "&add_32	(@x[$c3],@x[$c3],@x[$d3])",
+	"&eor_32	(@x[$b0],@x[$b0],@x[$c0])",
+	 "&eor_32	(@x[$b1],@x[$b1],@x[$c1])",
+	  "&eor_32	(@x[$b2],@x[$b2],@x[$c2])",
+	   "&eor_32	(@x[$b3],@x[$b3],@x[$c3])",
+	"&ror_32	(@x[$b0],@x[$b0],20)",
+	 "&ror_32	(@x[$b1],@x[$b1],20)",
+	  "&ror_32	(@x[$b2],@x[$b2],20)",
+	   "&ror_32	(@x[$b3],@x[$b3],20)",
+
+	"&add_32	(@x[$a0],@x[$a0],@x[$b0])",
+	 "&add_32	(@x[$a1],@x[$a1],@x[$b1])",
+	  "&add_32	(@x[$a2],@x[$a2],@x[$b2])",
+	   "&add_32	(@x[$a3],@x[$a3],@x[$b3])",
+	"&eor_32	(@x[$d0],@x[$d0],@x[$a0])",
+	 "&eor_32	(@x[$d1],@x[$d1],@x[$a1])",
+	  "&eor_32	(@x[$d2],@x[$d2],@x[$a2])",
+	   "&eor_32	(@x[$d3],@x[$d3],@x[$a3])",
+	"&ror_32	(@x[$d0],@x[$d0],24)",
+	 "&ror_32	(@x[$d1],@x[$d1],24)",
+	  "&ror_32	(@x[$d2],@x[$d2],24)",
+	   "&ror_32	(@x[$d3],@x[$d3],24)",
+
+	"&add_32	(@x[$c0],@x[$c0],@x[$d0])",
+	 "&add_32	(@x[$c1],@x[$c1],@x[$d1])",
+	  "&add_32	(@x[$c2],@x[$c2],@x[$d2])",
+	   "&add_32	(@x[$c3],@x[$c3],@x[$d3])",
+	"&eor_32	(@x[$b0],@x[$b0],@x[$c0])",
+	 "&eor_32	(@x[$b1],@x[$b1],@x[$c1])",
+	  "&eor_32	(@x[$b2],@x[$b2],@x[$c2])",
+	   "&eor_32	(@x[$b3],@x[$b3],@x[$c3])",
+	"&ror_32	(@x[$b0],@x[$b0],25)",
+	 "&ror_32	(@x[$b1],@x[$b1],25)",
+	  "&ror_32	(@x[$b2],@x[$b2],25)",
+	   "&ror_32	(@x[$b3],@x[$b3],25)"
+    );
+}
+
+$code.=<<___;
+#include "arm_arch.h"
+
+.text
+
+.extern	OPENSSL_armcap_P
+
+.align	5
+.Lsigma:
+.quad	0x3320646e61707865,0x6b20657479622d32		// endian-neutral
+.Lone:
+.long	1,0,0,0
+.LOPENSSL_armcap_P:
+#ifdef	__ILP32__
+.long	OPENSSL_armcap_P-.
+#else
+.quad	OPENSSL_armcap_P-.
+#endif
+.asciz	"ChaCha20 for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
+
+.globl	ChaCha20_ctr32
+.type	ChaCha20_ctr32,%function
+.align	5
+ChaCha20_ctr32:
+	cbz	$len,.Labort
+	adr	@x[0],.LOPENSSL_armcap_P
+	cmp	$len,#192
+	b.lo	.Lshort
+#ifdef	__ILP32__
+	ldrsw	@x[1],[@x[0]]
+#else
+	ldr	@x[1],[@x[0]]
+#endif
+	ldr	w17,[@x[1],@x[0]]
+	tst	w17,#ARMV7_NEON
+	b.ne	ChaCha20_neon
+
+.Lshort:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	@x[0],.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+	sub	sp,sp,#64
+
+	ldp	@d[0],@d[1],[@x[0]]		// load sigma
+	ldp	@d[2],@d[3],[$key]		// load key
+	ldp	@d[4],@d[5],[$key,#16]
+	ldp	@d[6],@d[7],[$ctr]		// load counter
+#ifdef	__ARMEB__
+	ror	@d[2],@d[2],#32
+	ror	@d[3],@d[3],#32
+	ror	@d[4],@d[4],#32
+	ror	@d[5],@d[5],#32
+	ror	@d[6],@d[6],#32
+	ror	@d[7],@d[7],#32
+#endif
+
+.Loop_outer:
+	mov.32	@x[0],@d[0]			// unpack key block
+	lsr	@x[1],@d[0],#32
+	mov.32	@x[2],@d[1]
+	lsr	@x[3],@d[1],#32
+	mov.32	@x[4],@d[2]
+	lsr	@x[5],@d[2],#32
+	mov.32	@x[6],@d[3]
+	lsr	@x[7],@d[3],#32
+	mov.32	@x[8],@d[4]
+	lsr	@x[9],@d[4],#32
+	mov.32	@x[10],@d[5]
+	lsr	@x[11],@d[5],#32
+	mov.32	@x[12],@d[6]
+	lsr	@x[13],@d[6],#32
+	mov.32	@x[14],@d[7]
+	lsr	@x[15],@d[7],#32
+
+	mov	$ctr,#10
+	subs	$len,$len,#64
+.Loop:
+	sub	$ctr,$ctr,#1	
+___
+	foreach (&ROUND(0, 4, 8,12)) { eval; }
+	foreach (&ROUND(0, 5,10,15)) { eval; }
+$code.=<<___;
+	cbnz	$ctr,.Loop
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	add	@x[1],@x[1],@d[0],lsr#32
+	add.32	@x[2],@x[2],@d[1]
+	add	@x[3],@x[3],@d[1],lsr#32
+	add.32	@x[4],@x[4],@d[2]
+	add	@x[5],@x[5],@d[2],lsr#32
+	add.32	@x[6],@x[6],@d[3]
+	add	@x[7],@x[7],@d[3],lsr#32
+	add.32	@x[8],@x[8],@d[4]
+	add	@x[9],@x[9],@d[4],lsr#32
+	add.32	@x[10],@x[10],@d[5]
+	add	@x[11],@x[11],@d[5],lsr#32
+	add.32	@x[12],@x[12],@d[6]
+	add	@x[13],@x[13],@d[6],lsr#32
+	add.32	@x[14],@x[14],@d[7]
+	add	@x[15],@x[15],@d[7],lsr#32
+
+	b.lo	.Ltail
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__ARMEB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	eor	@x[10],@x[10],@x[11]
+	eor	@x[12],@x[12],@x[13]
+	eor	@x[14],@x[14],@x[15]
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#1			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	stp	@x[8],@x[10],[$out,#32]
+	stp	@x[12],@x[14],[$out,#48]
+	add	$out,$out,#64
+
+	b.hi	.Loop_outer
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+.Labort:
+	ret
+
+.align	4
+.Ltail:
+	add	$len,$len,#64
+.Less_than_64:
+	sub	$out,$out,#1
+	add	$inp,$inp,$len
+	add	$out,$out,$len
+	add	$ctr,sp,$len
+	neg	$len,$len
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+#ifdef	__ARMEB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	stp	@x[0],@x[2],[sp,#0]
+	stp	@x[4],@x[6],[sp,#16]
+	stp	@x[8],@x[10],[sp,#32]
+	stp	@x[12],@x[14],[sp,#48]
+
+.Loop_tail:
+	ldrb	w10,[$inp,$len]
+	ldrb	w11,[$ctr,$len]
+	add	$len,$len,#1
+	eor	w10,w10,w11
+	strb	w10,[$out,$len]
+	cbnz	$len,.Loop_tail
+
+	stp	xzr,xzr,[sp,#0]
+	stp	xzr,xzr,[sp,#16]
+	stp	xzr,xzr,[sp,#32]
+	stp	xzr,xzr,[sp,#48]
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+___
+
+{{{
+my ($A0,$B0,$C0,$D0,$A1,$B1,$C1,$D1,$A2,$B2,$C2,$D2,$T0,$T1,$T2,$T3) =
+    map("v$_.4s",(0..7,16..23));
+my (@K)=map("v$_.4s",(24..30));
+my $ONE="v31.4s";
+
+sub NEONROUND {
+my $odd = pop;
+my ($a,$b,$c,$d,$t)=@_;
+
+	(
+	"&add		('$a','$a','$b')",
+	"&eor		('$d','$d','$a')",
+	"&rev32_16	('$d','$d')",		# vrot ($d,16)
+
+	"&add		('$c','$c','$d')",
+	"&eor		('$t','$b','$c')",
+	"&ushr		('$b','$t',20)",
+	"&sli		('$b','$t',12)",
+
+	"&add		('$a','$a','$b')",
+	"&eor		('$t','$d','$a')",
+	"&ushr		('$d','$t',24)",
+	"&sli		('$d','$t',8)",
+
+	"&add		('$c','$c','$d')",
+	"&eor		('$t','$b','$c')",
+	"&ushr		('$b','$t',25)",
+	"&sli		('$b','$t',7)",
+
+	"&ext		('$c','$c','$c',8)",
+	"&ext		('$d','$d','$d',$odd?4:12)",
+	"&ext		('$b','$b','$b',$odd?12:4)"
+	);
+}
+
+$code.=<<___;
+
+.type	ChaCha20_neon,%function
+.align	5
+ChaCha20_neon:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	@x[0],.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+	cmp	$len,#512
+	b.hs	.L512_or_more_neon
+
+	sub	sp,sp,#64
+
+	ldp	@d[0],@d[1],[@x[0]]		// load sigma
+	ld1	{@K[0]},[@x[0]],#16
+	ldp	@d[2],@d[3],[$key]		// load key
+	ldp	@d[4],@d[5],[$key,#16]
+	ld1	{@K[1],@K[2]},[$key]
+	ldp	@d[6],@d[7],[$ctr]		// load counter
+	ld1	{@K[3]},[$ctr]
+	ld1	{$ONE},[@x[0]]
+#ifdef	__ARMEB__
+	rev64	@K[0],@K[0]
+	ror	@d[2],@d[2],#32
+	ror	@d[3],@d[3],#32
+	ror	@d[4],@d[4],#32
+	ror	@d[5],@d[5],#32
+	ror	@d[6],@d[6],#32
+	ror	@d[7],@d[7],#32
+#endif
+	add	@K[3],@K[3],$ONE		// += 1
+	add	@K[4],@K[3],$ONE
+	add	@K[5],@K[4],$ONE
+	shl	$ONE,$ONE,#2			// 1 -> 4
+
+.Loop_outer_neon:
+	mov.32	@x[0],@d[0]			// unpack key block
+	lsr	@x[1],@d[0],#32
+	 mov	$A0,@K[0]
+	mov.32	@x[2],@d[1]
+	lsr	@x[3],@d[1],#32
+	 mov	$A1,@K[0]
+	mov.32	@x[4],@d[2]
+	lsr	@x[5],@d[2],#32
+	 mov	$A2,@K[0]
+	mov.32	@x[6],@d[3]
+	 mov	$B0,@K[1]
+	lsr	@x[7],@d[3],#32
+	 mov	$B1,@K[1]
+	mov.32	@x[8],@d[4]
+	 mov	$B2,@K[1]
+	lsr	@x[9],@d[4],#32
+	 mov	$D0,@K[3]
+	mov.32	@x[10],@d[5]
+	 mov	$D1,@K[4]
+	lsr	@x[11],@d[5],#32
+	 mov	$D2,@K[5]
+	mov.32	@x[12],@d[6]
+	 mov	$C0,@K[2]
+	lsr	@x[13],@d[6],#32
+	 mov	$C1,@K[2]
+	mov.32	@x[14],@d[7]
+	 mov	$C2,@K[2]
+	lsr	@x[15],@d[7],#32
+
+	mov	$ctr,#10
+	subs	$len,$len,#256
+.Loop_neon:
+	sub	$ctr,$ctr,#1
+___
+	my @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0);
+	my @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0);
+	my @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0);
+	my @thread3=&ROUND(0,4,8,12);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1);
+	@thread3=&ROUND(0,5,10,15);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+$code.=<<___;
+	cbnz	$ctr,.Loop_neon
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	 add	$A0,$A0,@K[0]
+	add	@x[1],@x[1],@d[0],lsr#32
+	 add	$A1,$A1,@K[0]
+	add.32	@x[2],@x[2],@d[1]
+	 add	$A2,$A2,@K[0]
+	add	@x[3],@x[3],@d[1],lsr#32
+	 add	$C0,$C0,@K[2]
+	add.32	@x[4],@x[4],@d[2]
+	 add	$C1,$C1,@K[2]
+	add	@x[5],@x[5],@d[2],lsr#32
+	 add	$C2,$C2,@K[2]
+	add.32	@x[6],@x[6],@d[3]
+	 add	$D0,$D0,@K[3]
+	add	@x[7],@x[7],@d[3],lsr#32
+	add.32	@x[8],@x[8],@d[4]
+	 add	$D1,$D1,@K[4]
+	add	@x[9],@x[9],@d[4],lsr#32
+	add.32	@x[10],@x[10],@d[5]
+	 add	$D2,$D2,@K[5]
+	add	@x[11],@x[11],@d[5],lsr#32
+	add.32	@x[12],@x[12],@d[6]
+	 add	$B0,$B0,@K[1]
+	add	@x[13],@x[13],@d[6],lsr#32
+	add.32	@x[14],@x[14],@d[7]
+	 add	$B1,$B1,@K[1]
+	add	@x[15],@x[15],@d[7],lsr#32
+	 add	$B2,$B2,@K[1]
+
+	b.lo	.Ltail_neon
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__ARMEB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	 eor	$A0,$A0,$T0
+	eor	@x[10],@x[10],@x[11]
+	 eor	$B0,$B0,$T1
+	eor	@x[12],@x[12],@x[13]
+	 eor	$C0,$C0,$T2
+	eor	@x[14],@x[14],@x[15]
+	 eor	$D0,$D0,$T3
+	 ld1.8	{$T0-$T3},[$inp],#64
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#4			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	 add	@K[3],@K[3],$ONE		// += 4
+	stp	@x[8],@x[10],[$out,#32]
+	 add	@K[4],@K[4],$ONE
+	stp	@x[12],@x[14],[$out,#48]
+	 add	@K[5],@K[5],$ONE
+	add	$out,$out,#64
+
+	st1.8	{$A0-$D0},[$out],#64
+	ld1.8	{$A0-$D0},[$inp],#64
+
+	eor	$A1,$A1,$T0
+	eor	$B1,$B1,$T1
+	eor	$C1,$C1,$T2
+	eor	$D1,$D1,$T3
+	st1.8	{$A1-$D1},[$out],#64
+
+	eor	$A2,$A2,$A0
+	eor	$B2,$B2,$B0
+	eor	$C2,$C2,$C0
+	eor	$D2,$D2,$D0
+	st1.8	{$A2-$D2},[$out],#64
+
+	b.hi	.Loop_outer_neon
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+
+.Ltail_neon:
+	add	$len,$len,#256
+	cmp	$len,#64
+	b.lo	.Less_than_64
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__ARMEB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	eor	@x[10],@x[10],@x[11]
+	eor	@x[12],@x[12],@x[13]
+	eor	@x[14],@x[14],@x[15]
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#4			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	stp	@x[8],@x[10],[$out,#32]
+	stp	@x[12],@x[14],[$out,#48]
+	add	$out,$out,#64
+	b.eq	.Ldone_neon
+	sub	$len,$len,#64
+	cmp	$len,#64
+	b.lo	.Less_than_128
+
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	$A0,$A0,$T0
+	eor	$B0,$B0,$T1
+	eor	$C0,$C0,$T2
+	eor	$D0,$D0,$T3
+	st1.8	{$A0-$D0},[$out],#64
+	b.eq	.Ldone_neon
+	sub	$len,$len,#64
+	cmp	$len,#64
+	b.lo	.Less_than_192
+
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	$A1,$A1,$T0
+	eor	$B1,$B1,$T1
+	eor	$C1,$C1,$T2
+	eor	$D1,$D1,$T3
+	st1.8	{$A1-$D1},[$out],#64
+	b.eq	.Ldone_neon
+	sub	$len,$len,#64
+
+	st1.8	{$A2-$D2},[sp]
+	b	.Last_neon
+
+.Less_than_128:
+	st1.8	{$A0-$D0},[sp]
+	b	.Last_neon
+.Less_than_192:
+	st1.8	{$A1-$D1},[sp]
+	b	.Last_neon
+
+.align	4
+.Last_neon:
+	sub	$out,$out,#1
+	add	$inp,$inp,$len
+	add	$out,$out,$len
+	add	$ctr,sp,$len
+	neg	$len,$len
+
+.Loop_tail_neon:
+	ldrb	w10,[$inp,$len]
+	ldrb	w11,[$ctr,$len]
+	add	$len,$len,#1
+	eor	w10,w10,w11
+	strb	w10,[$out,$len]
+	cbnz	$len,.Loop_tail_neon
+
+	stp	xzr,xzr,[sp,#0]
+	stp	xzr,xzr,[sp,#16]
+	stp	xzr,xzr,[sp,#32]
+	stp	xzr,xzr,[sp,#48]
+
+.Ldone_neon:
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_neon,.-ChaCha20_neon
+___
+{
+my ($T0,$T1,$T2,$T3,$T4,$T5)=@K;
+my ($A0,$B0,$C0,$D0,$A1,$B1,$C1,$D1,$A2,$B2,$C2,$D2,
+    $A3,$B3,$C3,$D3,$A4,$B4,$C4,$D4,$A5,$B5,$C5,$D5) = map("v$_.4s",(0..23));
+
+$code.=<<___;
+.type	ChaCha20_512_neon,%function
+.align	5
+ChaCha20_512_neon:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	@x[0],.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+
+.L512_or_more_neon:
+	sub	sp,sp,#128+64
+
+	ldp	@d[0],@d[1],[@x[0]]		// load sigma
+	ld1	{@K[0]},[@x[0]],#16
+	ldp	@d[2],@d[3],[$key]		// load key
+	ldp	@d[4],@d[5],[$key,#16]
+	ld1	{@K[1],@K[2]},[$key]
+	ldp	@d[6],@d[7],[$ctr]		// load counter
+	ld1	{@K[3]},[$ctr]
+	ld1	{$ONE},[@x[0]]
+#ifdef	__ARMEB__
+	rev64	@K[0],@K[0]
+	ror	@d[2],@d[2],#32
+	ror	@d[3],@d[3],#32
+	ror	@d[4],@d[4],#32
+	ror	@d[5],@d[5],#32
+	ror	@d[6],@d[6],#32
+	ror	@d[7],@d[7],#32
+#endif
+	add	@K[3],@K[3],$ONE		// += 1
+	stp	@K[0],@K[1],[sp,#0]		// off-load key block, invariant part
+	add	@K[3],@K[3],$ONE		// not typo
+	str	@K[2],[sp,#32]
+	add	@K[4],@K[3],$ONE
+	add	@K[5],@K[4],$ONE
+	add	@K[6],@K[5],$ONE
+	shl	$ONE,$ONE,#2			// 1 -> 4
+
+	stp	d8,d9,[sp,#128+0]		// meet ABI requirements
+	stp	d10,d11,[sp,#128+16]
+	stp	d12,d13,[sp,#128+32]
+	stp	d14,d15,[sp,#128+48]
+
+	sub	$len,$len,#512			// not typo
+
+.Loop_outer_512_neon:
+	 mov	$A0,@K[0]
+	 mov	$A1,@K[0]
+	 mov	$A2,@K[0]
+	 mov	$A3,@K[0]
+	 mov	$A4,@K[0]
+	 mov	$A5,@K[0]
+	 mov	$B0,@K[1]
+	mov.32	@x[0],@d[0]			// unpack key block
+	 mov	$B1,@K[1]
+	lsr	@x[1],@d[0],#32
+	 mov	$B2,@K[1]
+	mov.32	@x[2],@d[1]
+	 mov	$B3,@K[1]
+	lsr	@x[3],@d[1],#32
+	 mov	$B4,@K[1]
+	mov.32	@x[4],@d[2]
+	 mov	$B5,@K[1]
+	lsr	@x[5],@d[2],#32
+	 mov	$D0,@K[3]
+	mov.32	@x[6],@d[3]
+	 mov	$D1,@K[4]
+	lsr	@x[7],@d[3],#32
+	 mov	$D2,@K[5]
+	mov.32	@x[8],@d[4]
+	 mov	$D3,@K[6]
+	lsr	@x[9],@d[4],#32
+	 mov	$C0,@K[2]
+	mov.32	@x[10],@d[5]
+	 mov	$C1,@K[2]
+	lsr	@x[11],@d[5],#32
+	 add	$D4,$D0,$ONE			// +4
+	mov.32	@x[12],@d[6]
+	 add	$D5,$D1,$ONE			// +4
+	lsr	@x[13],@d[6],#32
+	 mov	$C2,@K[2]
+	mov.32	@x[14],@d[7]
+	 mov	$C3,@K[2]
+	lsr	@x[15],@d[7],#32
+	 mov	$C4,@K[2]
+	 stp	@K[3],@K[4],[sp,#48]		// off-load key block, variable part
+	 mov	$C5,@K[2]
+	 str	@K[5],[sp,#80]
+
+	mov	$ctr,#5
+	subs	$len,$len,#512
+.Loop_upper_neon:
+	sub	$ctr,$ctr,#1
+___
+	my @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0);
+	my @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0);
+	my @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0);
+	my @thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,0);
+	my @thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,0);
+	my @thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,0);
+	my @thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+	my $diff = ($#thread0+1)*6 - $#thread67 - 1;
+	my $i = 0;
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1);
+	@thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,1);
+	@thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,1);
+	@thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,1);
+	@thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+$code.=<<___;
+	cbnz	$ctr,.Loop_upper_neon
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	add	@x[1],@x[1],@d[0],lsr#32
+	add.32	@x[2],@x[2],@d[1]
+	add	@x[3],@x[3],@d[1],lsr#32
+	add.32	@x[4],@x[4],@d[2]
+	add	@x[5],@x[5],@d[2],lsr#32
+	add.32	@x[6],@x[6],@d[3]
+	add	@x[7],@x[7],@d[3],lsr#32
+	add.32	@x[8],@x[8],@d[4]
+	add	@x[9],@x[9],@d[4],lsr#32
+	add.32	@x[10],@x[10],@d[5]
+	add	@x[11],@x[11],@d[5],lsr#32
+	add.32	@x[12],@x[12],@d[6]
+	add	@x[13],@x[13],@d[6],lsr#32
+	add.32	@x[14],@x[14],@d[7]
+	add	@x[15],@x[15],@d[7],lsr#32
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__ARMEB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	eor	@x[10],@x[10],@x[11]
+	eor	@x[12],@x[12],@x[13]
+	eor	@x[14],@x[14],@x[15]
+
+	 stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#1			// increment counter
+	mov.32	@x[0],@d[0]			// unpack key block
+	lsr	@x[1],@d[0],#32
+	 stp	@x[4],@x[6],[$out,#16]
+	mov.32	@x[2],@d[1]
+	lsr	@x[3],@d[1],#32
+	 stp	@x[8],@x[10],[$out,#32]
+	mov.32	@x[4],@d[2]
+	lsr	@x[5],@d[2],#32
+	 stp	@x[12],@x[14],[$out,#48]
+	 add	$out,$out,#64
+	mov.32	@x[6],@d[3]
+	lsr	@x[7],@d[3],#32
+	mov.32	@x[8],@d[4]
+	lsr	@x[9],@d[4],#32
+	mov.32	@x[10],@d[5]
+	lsr	@x[11],@d[5],#32
+	mov.32	@x[12],@d[6]
+	lsr	@x[13],@d[6],#32
+	mov.32	@x[14],@d[7]
+	lsr	@x[15],@d[7],#32
+
+	mov	$ctr,#5
+.Loop_lower_neon:
+	sub	$ctr,$ctr,#1
+___
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0);
+	@thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,0);
+	@thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,0);
+	@thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,0);
+	@thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1);
+	@thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,1);
+	@thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,1);
+	@thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,1);
+	@thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+$code.=<<___;
+	cbnz	$ctr,.Loop_lower_neon
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	 ldp	@K[0],@K[1],[sp,#0]
+	add	@x[1],@x[1],@d[0],lsr#32
+	 ldp	@K[2],@K[3],[sp,#32]
+	add.32	@x[2],@x[2],@d[1]
+	 ldp	@K[4],@K[5],[sp,#64]
+	add	@x[3],@x[3],@d[1],lsr#32
+	 add	$A0,$A0,@K[0]
+	add.32	@x[4],@x[4],@d[2]
+	 add	$A1,$A1,@K[0]
+	add	@x[5],@x[5],@d[2],lsr#32
+	 add	$A2,$A2,@K[0]
+	add.32	@x[6],@x[6],@d[3]
+	 add	$A3,$A3,@K[0]
+	add	@x[7],@x[7],@d[3],lsr#32
+	 add	$A4,$A4,@K[0]
+	add.32	@x[8],@x[8],@d[4]
+	 add	$A5,$A5,@K[0]
+	add	@x[9],@x[9],@d[4],lsr#32
+	 add	$C0,$C0,@K[2]
+	add.32	@x[10],@x[10],@d[5]
+	 add	$C1,$C1,@K[2]
+	add	@x[11],@x[11],@d[5],lsr#32
+	 add	$C2,$C2,@K[2]
+	add.32	@x[12],@x[12],@d[6]
+	 add	$C3,$C3,@K[2]
+	add	@x[13],@x[13],@d[6],lsr#32
+	 add	$C4,$C4,@K[2]
+	add.32	@x[14],@x[14],@d[7]
+	 add	$C5,$C5,@K[2]
+	add	@x[15],@x[15],@d[7],lsr#32
+	 add	$D4,$D4,$ONE			// +4
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	 add	$D5,$D5,$ONE			// +4
+	add	@x[2],@x[2],@x[3],lsl#32
+	 add	$D0,$D0,@K[3]
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	 add	$D1,$D1,@K[4]
+	add	@x[4],@x[4],@x[5],lsl#32
+	 add	$D2,$D2,@K[5]
+	add	@x[6],@x[6],@x[7],lsl#32
+	 add	$D3,$D3,@K[6]
+	ldp	@x[5],@x[7],[$inp,#16]
+	 add	$D4,$D4,@K[3]
+	add	@x[8],@x[8],@x[9],lsl#32
+	 add	$D5,$D5,@K[4]
+	add	@x[10],@x[10],@x[11],lsl#32
+	 add	$B0,$B0,@K[1]
+	ldp	@x[9],@x[11],[$inp,#32]
+	 add	$B1,$B1,@K[1]
+	add	@x[12],@x[12],@x[13],lsl#32
+	 add	$B2,$B2,@K[1]
+	add	@x[14],@x[14],@x[15],lsl#32
+	 add	$B3,$B3,@K[1]
+	ldp	@x[13],@x[15],[$inp,#48]
+	 add	$B4,$B4,@K[1]
+	add	$inp,$inp,#64
+	 add	$B5,$B5,@K[1]
+
+#ifdef	__ARMEB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	 eor	$A0,$A0,$T0
+	eor	@x[10],@x[10],@x[11]
+	 eor	$B0,$B0,$T1
+	eor	@x[12],@x[12],@x[13]
+	 eor	$C0,$C0,$T2
+	eor	@x[14],@x[14],@x[15]
+	 eor	$D0,$D0,$T3
+	 ld1.8	{$T0-$T3},[$inp],#64
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#7			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	stp	@x[8],@x[10],[$out,#32]
+	stp	@x[12],@x[14],[$out,#48]
+	add	$out,$out,#64
+	st1.8	{$A0-$D0},[$out],#64
+
+	ld1.8	{$A0-$D0},[$inp],#64
+	eor	$A1,$A1,$T0
+	eor	$B1,$B1,$T1
+	eor	$C1,$C1,$T2
+	eor	$D1,$D1,$T3
+	st1.8	{$A1-$D1},[$out],#64
+
+	ld1.8	{$A1-$D1},[$inp],#64
+	eor	$A2,$A2,$A0
+	 ldp	@K[0],@K[1],[sp,#0]
+	eor	$B2,$B2,$B0
+	 ldp	@K[2],@K[3],[sp,#32]
+	eor	$C2,$C2,$C0
+	eor	$D2,$D2,$D0
+	st1.8	{$A2-$D2},[$out],#64
+
+	ld1.8	{$A2-$D2},[$inp],#64
+	eor	$A3,$A3,$A1
+	eor	$B3,$B3,$B1
+	eor	$C3,$C3,$C1
+	eor	$D3,$D3,$D1
+	st1.8	{$A3-$D3},[$out],#64
+
+	ld1.8	{$A3-$D3},[$inp],#64
+	eor	$A4,$A4,$A2
+	eor	$B4,$B4,$B2
+	eor	$C4,$C4,$C2
+	eor	$D4,$D4,$D2
+	st1.8	{$A4-$D4},[$out],#64
+
+	shl	$A0,$ONE,#1			// 4 -> 8
+	eor	$A5,$A5,$A3
+	eor	$B5,$B5,$B3
+	eor	$C5,$C5,$C3
+	eor	$D5,$D5,$D3
+	st1.8	{$A5-$D5},[$out],#64
+
+	add	@K[3],@K[3],$A0			// += 8
+	add	@K[4],@K[4],$A0
+	add	@K[5],@K[5],$A0
+	add	@K[6],@K[6],$A0
+
+	b.hs	.Loop_outer_512_neon
+
+	adds	$len,$len,#512
+	ushr	$A0,$ONE,#2			// 4 -> 1
+
+	ldp	d8,d9,[sp,#128+0]		// meet ABI requirements
+	ldp	d10,d11,[sp,#128+16]
+	ldp	d12,d13,[sp,#128+32]
+	ldp	d14,d15,[sp,#128+48]
+
+	stp	@K[0],$ONE,[sp,#0]		// wipe off-load area
+	stp	@K[0],$ONE,[sp,#32]
+	stp	@K[0],$ONE,[sp,#64]
+
+	b.eq	.Ldone_512_neon
+
+	cmp	$len,#192
+	sub	@K[3],@K[3],$A0			// -= 1
+	sub	@K[4],@K[4],$A0
+	sub	@K[5],@K[5],$A0
+	add	sp,sp,#128
+	b.hs	.Loop_outer_neon
+
+	eor	@K[1],@K[1],@K[1]
+	eor	@K[2],@K[2],@K[2]
+	eor	@K[3],@K[3],@K[3]
+	eor	@K[4],@K[4],@K[4]
+	eor	@K[5],@K[5],@K[5]
+	eor	@K[6],@K[6],@K[6]
+	b	.Loop_outer
+
+.Ldone_512_neon:
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#128+64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_512_neon,.-ChaCha20_512_neon
+___
+}
+}}}
+
+foreach (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	(s/\b([a-z]+)\.32\b/$1/ and (s/x([0-9]+)/w$1/g or 1))	or
+	(m/\b(eor|ext|mov)\b/ and (s/\.4s/\.16b/g or 1))	or
+	(s/\b((?:ld|st)1)\.8\b/$1/ and (s/\.4s/\.16b/g or 1))	or
+	(m/\b(ld|st)[rp]\b/ and (s/v([0-9]+)\.4s/q$1/g or 1))	or
+	(s/\brev32\.16\b/rev32/ and (s/\.4s/\.8h/g or 1));
+
+	#s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
+
+	print $_,"\n";
+}