From: Daniel Hu <Daniel.Hu@arm.com>
Date: Mon, 14 Feb 2022 14:36:34 +0000 (+0000)
Subject: SM4 optimization for ARM by ASIMD
X-Git-Tag: openssl-3.2.0-alpha1~2770
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=4908787f21f4f5fa24b721ed3ebbc4d3e93ef70c;p=thirdparty%2Fopenssl.git

SM4 optimization for ARM by ASIMD

This patch optimizes SM4 for ARM processor using ASIMD instruction

It will improve performance if both of following conditions are met:
1) Input data equal to or more than 4 blocks
2) Cipher mode allows parallelism, including ECB,CTR,GCM or CBC decryption

This patch implements SM4 SBOX lookup in vector registers, with the
benefit of constant processing time over existing C implementation.

It is only enabled for micro-architecture N1/V1. In the ideal scenario,
performance can reach up to 2.7X

When either of above two conditions is not met, e.g. single block input
or CFB/OFB mode, CBC encryption, performance could drop about 50%.

The assembly code has been reviewed internally by ARM engineer
Fangming.Fang@arm.com

Signed-off-by: Daniel Hu <Daniel.Hu@arm.com>

Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/17951)
---

diff --git a/crypto/evp/e_sm4.c b/crypto/evp/e_sm4.c
index bff79ff197..c8e8cfe9c9 100644
--- a/crypto/evp/e_sm4.c
+++ b/crypto/evp/e_sm4.c
@@ -76,6 +76,17 @@ static int sm4_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
                 dat->stream.ecb = (ecb128_f) HWSM4_ecb_encrypt;
 # endif
         } else
+#endif
+#ifdef VPSM4_CAPABLE
+        if (VPSM4_CAPABLE) {
+            vpsm4_set_decrypt_key(key, &dat->ks.ks);
+            dat->block = (block128_f) vpsm4_decrypt;
+            dat->stream.cbc = NULL;
+            if (mode == EVP_CIPH_CBC_MODE)
+                dat->stream.cbc = (cbc128_f) vpsm4_cbc_encrypt;
+            else if (mode == EVP_CIPH_ECB_MODE)
+                dat->stream.ecb = (ecb128_f) vpsm4_ecb_encrypt;
+        } else
 #endif
         {
             dat->block = (block128_f) ossl_sm4_decrypt;
@@ -104,6 +115,19 @@ static int sm4_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 # endif
             (void)0;            /* terminate potentially open 'else' */
     } else
+#endif
+#ifdef VPSM4_CAPABLE
+    if (VPSM4_CAPABLE) {
+        vpsm4_set_encrypt_key(key, &dat->ks.ks);
+        dat->block = (block128_f) vpsm4_encrypt;
+        dat->stream.cbc = NULL;
+        if (mode == EVP_CIPH_CBC_MODE)
+            dat->stream.cbc = (cbc128_f) vpsm4_cbc_encrypt;
+        else if (mode == EVP_CIPH_ECB_MODE)
+            dat->stream.ecb = (ecb128_f) vpsm4_ecb_encrypt;
+        else if (mode == EVP_CIPH_CTR_MODE)
+            dat->stream.ctr = (ctr128_f) vpsm4_ctr32_encrypt_blocks;
+    } else
 #endif
     {
         dat->block = (block128_f) ossl_sm4_encrypt;
diff --git a/crypto/sm4/asm/vpsm4-armv8.pl b/crypto/sm4/asm/vpsm4-armv8.pl
new file mode 100755
index 0000000000..095d9dae64
--- /dev/null
+++ b/crypto/sm4/asm/vpsm4-armv8.pl
@@ -0,0 +1,1118 @@
+#! /usr/bin/env perl
+# Copyright 2020-2021 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the Apache License 2.0 (the "License").  You may not use
+# this file except in compliance with the License.  You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
+#
+# This module implements SM4 with ASIMD on aarch64
+#
+# Feb 2022
+#
+
+# $output is the last argument if it looks like a file (it has an extension)
+# $flavour is the first argument if it doesn't look like a file
+$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
+$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $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\""
+    or die "can't call $xlate: $!";
+*STDOUT=*OUT;
+
+$prefix="vpsm4";
+my @vtmp=map("v$_",(0..3));
+my @data=map("v$_",(4..7));
+my @datax=map("v$_",(8..11));
+my ($rk0,$rk1)=("v12","v13");
+my ($rka,$rkb)=("v14","v15");
+my @vtmpx=map("v$_",(12..15));
+my @sbox=map("v$_",(16..31));
+my ($inp,$outp,$blocks,$rks)=("x0","x1","w2","x3");
+my ($tmpw,$tmp,$wtmp0,$wtmp1,$wtmp2)=("w6","x6","w7","w8","w9");
+my ($ptr,$counter)=("x10","w11");
+my ($word0,$word1,$word2,$word3)=("w12","w13","w14","w15");
+
+sub rev32() {
+	my $dst = shift;
+	my $src = shift;
+
+	if ($src and ("$src" ne "$dst")) {
+$code.=<<___;
+#ifndef __ARMEB__
+	rev32	$dst.16b,$src.16b
+#else
+	mov	$dst.16b,$src.16b
+#endif
+___
+	} else {
+$code.=<<___;
+#ifndef __ARMEB__
+	rev32	$dst.16b,$dst.16b
+#endif
+___
+	}
+}
+
+sub transpose() {
+	my ($dat0,$dat1,$dat2,$dat3,$vt0,$vt1,$vt2,$vt3) = @_;
+
+$code.=<<___;
+	zip1	$vt0.4s,$dat0.4s,$dat1.4s
+	zip2	$vt1.4s,$dat0.4s,$dat1.4s
+	zip1	$vt2.4s,$dat2.4s,$dat3.4s
+	zip2	$vt3.4s,$dat2.4s,$dat3.4s
+	zip1	$dat0.2d,$vt0.2d,$vt2.2d
+	zip2	$dat1.2d,$vt0.2d,$vt2.2d
+	zip1	$dat2.2d,$vt1.2d,$vt3.2d
+	zip2	$dat3.2d,$vt1.2d,$vt3.2d
+___
+}
+
+# sbox operations for 4-lane of words
+sub sbox() {
+	my $dat = shift;
+
+$code.=<<___;
+	movi	@vtmp[0].16b,#64
+	movi	@vtmp[1].16b,#128
+	movi	@vtmp[2].16b,#192
+	sub	@vtmp[0].16b,$dat.16b,@vtmp[0].16b
+	sub	@vtmp[1].16b,$dat.16b,@vtmp[1].16b
+	sub	@vtmp[2].16b,$dat.16b,@vtmp[2].16b
+	tbl	$dat.16b,{@sbox[0].16b,@sbox[1].16b,@sbox[2].16b,@sbox[3].16b},$dat.16b
+	tbl	@vtmp[0].16b,{@sbox[4].16b,@sbox[5].16b,@sbox[6].16b,@sbox[7].16b},@vtmp[0].16b
+	tbl	@vtmp[1].16b,{@sbox[8].16b,@sbox[9].16b,@sbox[10].16b,@sbox[11].16b},@vtmp[1].16b
+	tbl	@vtmp[2].16b,{@sbox[12].16b,@sbox[13].16b,@sbox[14].16b,@sbox[15].16b},@vtmp[2].16b
+	add	@vtmp[0].2d,@vtmp[0].2d,@vtmp[1].2d
+	add	@vtmp[2].2d,@vtmp[2].2d,$dat.2d
+	add	$dat.2d,@vtmp[0].2d,@vtmp[2].2d
+
+	ushr	@vtmp[0].4s,$dat.4s,32-2
+	sli	@vtmp[0].4s,$dat.4s,2
+	ushr	@vtmp[2].4s,$dat.4s,32-10
+	eor	@vtmp[1].16b,@vtmp[0].16b,$dat.16b
+	sli	@vtmp[2].4s,$dat.4s,10
+	eor	@vtmp[1].16b,@vtmp[2].16b,$vtmp[1].16b
+	ushr	@vtmp[0].4s,$dat.4s,32-18
+	sli	@vtmp[0].4s,$dat.4s,18
+	ushr	@vtmp[2].4s,$dat.4s,32-24
+	eor	@vtmp[1].16b,@vtmp[0].16b,$vtmp[1].16b
+	sli	@vtmp[2].4s,$dat.4s,24
+	eor	$dat.16b,@vtmp[2].16b,@vtmp[1].16b
+___
+}
+
+# sbox operation for 8-lane of words
+sub sbox_double() {
+	my $dat = shift;
+	my $datx = shift;
+
+$code.=<<___;
+	movi	@vtmp[3].16b,#64
+	sub	@vtmp[0].16b,$dat.16b,@vtmp[3].16b
+	sub	@vtmp[1].16b,@vtmp[0].16b,@vtmp[3].16b
+	sub	@vtmp[2].16b,@vtmp[1].16b,@vtmp[3].16b
+	tbl	$dat.16b,{@sbox[0].16b,@sbox[1].16b,@sbox[2].16b,@sbox[3].16b},$dat.16b
+	tbl	@vtmp[0].16b,{@sbox[4].16b,@sbox[5].16b,@sbox[6].16b,@sbox[7].16b},@vtmp[0].16b
+	tbl	@vtmp[1].16b,{@sbox[8].16b,@sbox[9].16b,@sbox[10].16b,@sbox[11].16b},@vtmp[1].16b
+	tbl	@vtmp[2].16b,{@sbox[12].16b,@sbox[13].16b,@sbox[14].16b,@sbox[15].16b},@vtmp[2].16b
+	add	@vtmp[1].2d,@vtmp[0].2d,@vtmp[1].2d
+	add	$dat.2d,@vtmp[2].2d,$dat.2d
+	add	$dat.2d,@vtmp[1].2d,$dat.2d
+
+	sub	@vtmp[0].16b,$datx.16b,@vtmp[3].16b
+	sub	@vtmp[1].16b,@vtmp[0].16b,@vtmp[3].16b
+	sub	@vtmp[2].16b,@vtmp[1].16b,@vtmp[3].16b
+	tbl	$datx.16b,{@sbox[0].16b,@sbox[1].16b,@sbox[2].16b,@sbox[3].16b},$datx.16b
+	tbl	@vtmp[0].16b,{@sbox[4].16b,@sbox[5].16b,@sbox[6].16b,@sbox[7].16b},@vtmp[0].16b
+	tbl	@vtmp[1].16b,{@sbox[8].16b,@sbox[9].16b,@sbox[10].16b,@sbox[11].16b},@vtmp[1].16b
+	tbl	@vtmp[2].16b,{@sbox[12].16b,@sbox[13].16b,@sbox[14].16b,@sbox[15].16b},@vtmp[2].16b
+	add	@vtmp[1].2d,@vtmp[0].2d,@vtmp[1].2d
+	add	$datx.2d,@vtmp[2].2d,$datx.2d
+	add	$datx.2d,@vtmp[1].2d,$datx.2d
+
+	ushr	@vtmp[0].4s,$dat.4s,32-2
+	sli	@vtmp[0].4s,$dat.4s,2
+	ushr	@vtmp[2].4s,$datx.4s,32-2
+	eor	@vtmp[1].16b,@vtmp[0].16b,$dat.16b
+	sli	@vtmp[2].4s,$datx.4s,2
+
+	ushr	@vtmp[0].4s,$dat.4s,32-10
+	eor	@vtmp[3].16b,@vtmp[2].16b,$datx.16b
+	sli	@vtmp[0].4s,$dat.4s,10
+	ushr	@vtmp[2].4s,$datx.4s,32-10
+	eor	@vtmp[1].16b,@vtmp[0].16b,$vtmp[1].16b
+	sli	@vtmp[2].4s,$datx.4s,10
+
+	ushr	@vtmp[0].4s,$dat.4s,32-18
+	eor	@vtmp[3].16b,@vtmp[2].16b,$vtmp[3].16b
+	sli	@vtmp[0].4s,$dat.4s,18
+	ushr	@vtmp[2].4s,$datx.4s,32-18
+	eor	@vtmp[1].16b,@vtmp[0].16b,$vtmp[1].16b
+	sli	@vtmp[2].4s,$datx.4s,18
+
+	ushr	@vtmp[0].4s,$dat.4s,32-24
+	eor	@vtmp[3].16b,@vtmp[2].16b,$vtmp[3].16b
+	sli	@vtmp[0].4s,$dat.4s,24
+	ushr	@vtmp[2].4s,$datx.4s,32-24
+	eor	$dat.16b,@vtmp[0].16b,@vtmp[1].16b
+	sli	@vtmp[2].4s,$datx.4s,24
+	eor	$datx.16b,@vtmp[2].16b,@vtmp[3].16b
+___
+}
+
+# sbox operation for one single word
+sub sbox_1word () {
+	my $word = shift;
+
+$code.=<<___;
+	movi	@vtmp[1].16b,#64
+	movi	@vtmp[2].16b,#128
+	movi	@vtmp[3].16b,#192
+	mov	@vtmp[0].s[0],$word
+
+	sub	@vtmp[1].16b,@vtmp[0].16b,@vtmp[1].16b
+	sub	@vtmp[2].16b,@vtmp[0].16b,@vtmp[2].16b
+	sub	@vtmp[3].16b,@vtmp[0].16b,@vtmp[3].16b
+
+	tbl	@vtmp[0].16b,{@sbox[0].16b,@sbox[1].16b,@sbox[2].16b,@sbox[3].16b},@vtmp[0].16b
+	tbl	@vtmp[1].16b,{@sbox[4].16b,@sbox[5].16b,@sbox[6].16b,@sbox[7].16b},@vtmp[1].16b
+	tbl	@vtmp[2].16b,{@sbox[8].16b,@sbox[9].16b,@sbox[10].16b,@sbox[11].16b},@vtmp[2].16b
+	tbl	@vtmp[3].16b,{@sbox[12].16b,@sbox[13].16b,@sbox[14].16b,@sbox[15].16b},@vtmp[3].16b
+
+	mov	$word,@vtmp[0].s[0]
+	mov	$wtmp0,@vtmp[1].s[0]
+	mov	$wtmp2,@vtmp[2].s[0]
+	add	$wtmp0,$word,$wtmp0
+	mov	$word,@vtmp[3].s[0]
+	add	$wtmp0,$wtmp0,$wtmp2
+	add	$wtmp0,$wtmp0,$word
+
+	eor	$word,$wtmp0,$wtmp0,ror #32-2
+	eor	$word,$word,$wtmp0,ror #32-10
+	eor	$word,$word,$wtmp0,ror #32-18
+	eor	$word,$word,$wtmp0,ror #32-24
+___
+}
+
+# sm4 for one block of data, in scalar registers word0/word1/word2/word3
+sub sm4_1blk () {
+	my $kptr = shift;
+
+$code.=<<___;
+	ldp	$wtmp0,$wtmp1,[$kptr],8
+	// B0 ^= SBOX(B1 ^ B2 ^ B3 ^ RK0)
+	eor	$tmpw,$word2,$word3
+	eor	$wtmp2,$wtmp0,$word1
+	eor	$tmpw,$tmpw,$wtmp2
+___
+	&sbox_1word($tmpw);
+$code.=<<___;
+	eor	$word0,$word0,$tmpw
+	// B1 ^= SBOX(B0 ^ B2 ^ B3 ^ RK1)
+	eor	$tmpw,$word2,$word3
+	eor	$wtmp2,$word0,$wtmp1
+	eor	$tmpw,$tmpw,$wtmp2
+___
+	&sbox_1word($tmpw);
+$code.=<<___;
+	ldp	$wtmp0,$wtmp1,[$kptr],8
+	eor	$word1,$word1,$tmpw
+	// B2 ^= SBOX(B0 ^ B1 ^ B3 ^ RK2)
+	eor	$tmpw,$word0,$word1
+	eor	$wtmp2,$wtmp0,$word3
+	eor	$tmpw,$tmpw,$wtmp2
+___
+	&sbox_1word($tmpw);
+$code.=<<___;
+	eor	$word2,$word2,$tmpw
+	// B3 ^= SBOX(B0 ^ B1 ^ B2 ^ RK3)
+	eor	$tmpw,$word0,$word1
+	eor	$wtmp2,$word2,$wtmp1
+	eor	$tmpw,$tmpw,$wtmp2
+___
+	&sbox_1word($tmpw);
+$code.=<<___;
+	eor	$word3,$word3,$tmpw
+___
+}
+
+# sm4 for 4-lanes of data, in neon registers data0/data1/data2/data3
+sub sm4_4blks () {
+	my $kptr = shift;
+
+$code.=<<___;
+	ldp	$wtmp0,$wtmp1,[$kptr],8
+	dup	$rk0.4s,$wtmp0
+	dup	$rk1.4s,$wtmp1
+
+	// B0 ^= SBOX(B1 ^ B2 ^ B3 ^ RK0)
+	eor	$rka.16b,@data[2].16b,@data[3].16b
+	eor	$rk0.16b,@data[1].16b,$rk0.16b
+	eor	$rk0.16b,$rka.16b,$rk0.16b
+___
+	&sbox($rk0);
+$code.=<<___;
+	eor	@data[0].16b,@data[0].16b,$rk0.16b
+
+	// B1 ^= SBOX(B0 ^ B2 ^ B3 ^ RK1)
+	eor	$rka.16b,$rka.16b,@data[0].16b
+	eor	$rk1.16b,$rka.16b,$rk1.16b
+___
+	&sbox($rk1);
+$code.=<<___;
+	ldp	$wtmp0,$wtmp1,[$kptr],8
+	eor	@data[1].16b,@data[1].16b,$rk1.16b
+
+	dup	$rk0.4s,$wtmp0
+	dup	$rk1.4s,$wtmp1
+
+	// B2 ^= SBOX(B0 ^ B1 ^ B3 ^ RK2)
+	eor	$rka.16b,@data[0].16b,@data[1].16b
+	eor	$rk0.16b,@data[3].16b,$rk0.16b
+	eor	$rk0.16b,$rka.16b,$rk0.16b
+___
+	&sbox($rk0);
+$code.=<<___;
+	eor	@data[2].16b,@data[2].16b,$rk0.16b
+
+	// B3 ^= SBOX(B0 ^ B1 ^ B2 ^ RK3)
+	eor	$rka.16b,$rka.16b,@data[2].16b
+	eor	$rk1.16b,$rka.16b,$rk1.16b
+___
+	&sbox($rk1);
+$code.=<<___;
+	eor	@data[3].16b,@data[3].16b,$rk1.16b
+___
+}
+
+# sm4 for 8 lanes of data, in neon registers
+# data0/data1/data2/data3 datax0/datax1/datax2/datax3
+sub sm4_8blks () {
+	my $kptr = shift;
+
+$code.=<<___;
+	ldp	$wtmp0,$wtmp1,[$kptr],8
+	// B0 ^= SBOX(B1 ^ B2 ^ B3 ^ RK0)
+	dup	$rk0.4s,$wtmp0
+	eor	$rka.16b,@data[2].16b,@data[3].16b
+	eor	$rkb.16b,@datax[2].16b,@datax[3].16b
+	eor	@vtmp[0].16b,@data[1].16b,$rk0.16b
+	eor	@vtmp[1].16b,@datax[1].16b,$rk0.16b
+	eor	$rk0.16b,$rka.16b,@vtmp[0].16b
+	eor	$rk1.16b,$rkb.16b,@vtmp[1].16b
+___
+	&sbox_double($rk0,$rk1);
+$code.=<<___;
+	eor	@data[0].16b,@data[0].16b,$rk0.16b
+	eor	@datax[0].16b,@datax[0].16b,$rk1.16b
+
+	// B1 ^= SBOX(B0 ^ B2 ^ B3 ^ RK1)
+	dup	$rk1.4s,$wtmp1
+	eor	$rka.16b,$rka.16b,@data[0].16b
+	eor	$rkb.16b,$rkb.16b,@datax[0].16b
+	eor	$rk0.16b,$rka.16b,$rk1.16b
+	eor	$rk1.16b,$rkb.16b,$rk1.16b
+___
+	&sbox_double($rk0,$rk1);
+$code.=<<___;
+	ldp	$wtmp0,$wtmp1,[$kptr],8
+	eor	@data[1].16b,@data[1].16b,$rk0.16b
+	eor	@datax[1].16b,@datax[1].16b,$rk1.16b
+
+	// B2 ^= SBOX(B0 ^ B1 ^ B3 ^ RK2)
+	dup	$rk0.4s,$wtmp0
+	eor	$rka.16b,@data[0].16b,@data[1].16b
+	eor	$rkb.16b,@datax[0].16b,@datax[1].16b
+	eor	@vtmp[0].16b,@data[3].16b,$rk0.16b
+	eor	@vtmp[1].16b,@datax[3].16b,$rk0.16b
+	eor	$rk0.16b,$rka.16b,@vtmp[0].16b
+	eor	$rk1.16b,$rkb.16b,@vtmp[1].16b
+___
+	&sbox_double($rk0,$rk1);
+$code.=<<___;
+	eor	@data[2].16b,@data[2].16b,$rk0.16b
+	eor	@datax[2].16b,@datax[2].16b,$rk1.16b
+
+	// B3 ^= SBOX(B0 ^ B1 ^ B2 ^ RK3)
+	dup	$rk1.4s,$wtmp1
+	eor	$rka.16b,$rka.16b,@data[2].16b
+	eor	$rkb.16b,$rkb.16b,@datax[2].16b
+	eor	$rk0.16b,$rka.16b,$rk1.16b
+	eor	$rk1.16b,$rkb.16b,$rk1.16b
+___
+	&sbox_double($rk0,$rk1);
+$code.=<<___;
+	eor	@data[3].16b,@data[3].16b,$rk0.16b
+	eor	@datax[3].16b,@datax[3].16b,$rk1.16b
+___
+}
+
+sub encrypt_1blk_norev() {
+	my $dat = shift;
+
+$code.=<<___;
+	mov	$ptr,$rks
+	mov	$counter,#8
+	mov	$word0,$dat.s[0]
+	mov	$word1,$dat.s[1]
+	mov	$word2,$dat.s[2]
+	mov	$word3,$dat.s[3]
+10:
+___
+	&sm4_1blk($ptr);
+$code.=<<___;
+	subs	$counter,$counter,#1
+	b.ne	10b
+	mov	$dat.s[0],$word3
+	mov	$dat.s[1],$word2
+	mov	$dat.s[2],$word1
+	mov	$dat.s[3],$word0
+___
+}
+
+sub encrypt_1blk() {
+	my $dat = shift;
+
+	&encrypt_1blk_norev($dat);
+	&rev32($dat,$dat);
+}
+
+sub encrypt_4blks() {
+$code.=<<___;
+	mov	$ptr,$rks
+	mov	$counter,#8
+10:
+___
+	&sm4_4blks($ptr);
+$code.=<<___;
+	subs	$counter,$counter,#1
+	b.ne	10b
+___
+	&rev32(@vtmp[3],@data[0]);
+	&rev32(@vtmp[2],@data[1]);
+	&rev32(@vtmp[1],@data[2]);
+	&rev32(@vtmp[0],@data[3]);
+}
+
+sub encrypt_8blks() {
+$code.=<<___;
+	mov	$ptr,$rks
+	mov	$counter,#8
+10:
+___
+	&sm4_8blks($ptr);
+$code.=<<___;
+	subs	$counter,$counter,#1
+	b.ne	10b
+___
+	&rev32(@vtmp[3],@data[0]);
+	&rev32(@vtmp[2],@data[1]);
+	&rev32(@vtmp[1],@data[2]);
+	&rev32(@vtmp[0],@data[3]);
+	&rev32(@data[3],@datax[0]);
+	&rev32(@data[2],@datax[1]);
+	&rev32(@data[1],@datax[2]);
+	&rev32(@data[0],@datax[3]);
+}
+
+sub load_sbox () {
+	my $data = shift;
+
+$code.=<<___;
+	adr	$ptr,.Lsbox
+	ld1	{@sbox[0].4s,@sbox[1].4s,@sbox[2].4s,@sbox[3].4s},[$ptr],#64
+	ld1	{@sbox[4].4s,@sbox[5].4s,@sbox[6].4s,@sbox[7].4s},[$ptr],#64
+	ld1	{@sbox[8].4s,@sbox[9].4s,@sbox[10].4s,@sbox[11].4s},[$ptr],#64
+	ld1	{@sbox[12].4s,@sbox[13].4s,@sbox[14].4s,@sbox[15].4s},[$ptr]
+___
+}
+
+$code=<<___;
+#include "arm_arch.h"
+.arch	armv8-a
+.text
+
+.type	_vpsm4_consts,%object
+.align	7
+_vpsm4_consts:
+.Lsbox:
+	.byte 0xD6,0x90,0xE9,0xFE,0xCC,0xE1,0x3D,0xB7,0x16,0xB6,0x14,0xC2,0x28,0xFB,0x2C,0x05
+	.byte 0x2B,0x67,0x9A,0x76,0x2A,0xBE,0x04,0xC3,0xAA,0x44,0x13,0x26,0x49,0x86,0x06,0x99
+	.byte 0x9C,0x42,0x50,0xF4,0x91,0xEF,0x98,0x7A,0x33,0x54,0x0B,0x43,0xED,0xCF,0xAC,0x62
+	.byte 0xE4,0xB3,0x1C,0xA9,0xC9,0x08,0xE8,0x95,0x80,0xDF,0x94,0xFA,0x75,0x8F,0x3F,0xA6
+	.byte 0x47,0x07,0xA7,0xFC,0xF3,0x73,0x17,0xBA,0x83,0x59,0x3C,0x19,0xE6,0x85,0x4F,0xA8
+	.byte 0x68,0x6B,0x81,0xB2,0x71,0x64,0xDA,0x8B,0xF8,0xEB,0x0F,0x4B,0x70,0x56,0x9D,0x35
+	.byte 0x1E,0x24,0x0E,0x5E,0x63,0x58,0xD1,0xA2,0x25,0x22,0x7C,0x3B,0x01,0x21,0x78,0x87
+	.byte 0xD4,0x00,0x46,0x57,0x9F,0xD3,0x27,0x52,0x4C,0x36,0x02,0xE7,0xA0,0xC4,0xC8,0x9E
+	.byte 0xEA,0xBF,0x8A,0xD2,0x40,0xC7,0x38,0xB5,0xA3,0xF7,0xF2,0xCE,0xF9,0x61,0x15,0xA1
+	.byte 0xE0,0xAE,0x5D,0xA4,0x9B,0x34,0x1A,0x55,0xAD,0x93,0x32,0x30,0xF5,0x8C,0xB1,0xE3
+	.byte 0x1D,0xF6,0xE2,0x2E,0x82,0x66,0xCA,0x60,0xC0,0x29,0x23,0xAB,0x0D,0x53,0x4E,0x6F
+	.byte 0xD5,0xDB,0x37,0x45,0xDE,0xFD,0x8E,0x2F,0x03,0xFF,0x6A,0x72,0x6D,0x6C,0x5B,0x51
+	.byte 0x8D,0x1B,0xAF,0x92,0xBB,0xDD,0xBC,0x7F,0x11,0xD9,0x5C,0x41,0x1F,0x10,0x5A,0xD8
+	.byte 0x0A,0xC1,0x31,0x88,0xA5,0xCD,0x7B,0xBD,0x2D,0x74,0xD0,0x12,0xB8,0xE5,0xB4,0xB0
+	.byte 0x89,0x69,0x97,0x4A,0x0C,0x96,0x77,0x7E,0x65,0xB9,0xF1,0x09,0xC5,0x6E,0xC6,0x84
+	.byte 0x18,0xF0,0x7D,0xEC,0x3A,0xDC,0x4D,0x20,0x79,0xEE,0x5F,0x3E,0xD7,0xCB,0x39,0x48
+.Lck:
+	.long 0x00070E15, 0x1C232A31, 0x383F464D, 0x545B6269
+	.long 0x70777E85, 0x8C939AA1, 0xA8AFB6BD, 0xC4CBD2D9
+	.long 0xE0E7EEF5, 0xFC030A11, 0x181F262D, 0x343B4249
+	.long 0x50575E65, 0x6C737A81, 0x888F969D, 0xA4ABB2B9
+	.long 0xC0C7CED5, 0xDCE3EAF1, 0xF8FF060D, 0x141B2229
+	.long 0x30373E45, 0x4C535A61, 0x686F767D, 0x848B9299
+	.long 0xA0A7AEB5, 0xBCC3CAD1, 0xD8DFE6ED, 0xF4FB0209
+	.long 0x10171E25, 0x2C333A41, 0x484F565D, 0x646B7279
+.Lfk:
+	.dword 0x56aa3350a3b1bac6,0xb27022dc677d9197
+.Lshuffles:
+	.dword 0x0B0A090807060504,0x030201000F0E0D0C
+
+.size	_vpsm4_consts,.-_vpsm4_consts
+___
+
+{{{
+my ($key,$keys,$enc)=("x0","x1","w2");
+my ($pointer,$schedules,$wtmp,$roundkey)=("x5","x6","w7","w8");
+my ($vkey,$vfk,$vmap)=("v5","v6","v7");
+$code.=<<___;
+.type	_vpsm4_set_key,%function
+.align	4
+_vpsm4_set_key:
+	AARCH64_VALID_CALL_TARGET
+	ld1	{$vkey.4s},[$key]
+___
+	&load_sbox();
+	&rev32($vkey,$vkey);
+$code.=<<___;
+	adr	$pointer,.Lshuffles
+	ld1	{$vmap.4s},[$pointer]
+	adr	$pointer,.Lfk
+	ld1	{$vfk.4s},[$pointer]
+	eor	$vkey.16b,$vkey.16b,$vfk.16b
+	mov	$schedules,#32
+	adr	$pointer,.Lck
+	movi	@vtmp[0].16b,#64
+	cbnz	$enc,1f
+	add	$keys,$keys,124
+1:
+	mov	$wtmp,$vkey.s[1]
+	ldr	$roundkey,[$pointer],#4
+	eor	$roundkey,$roundkey,$wtmp
+	mov	$wtmp,$vkey.s[2]
+	eor	$roundkey,$roundkey,$wtmp
+	mov	$wtmp,$vkey.s[3]
+	eor	$roundkey,$roundkey,$wtmp
+	// sbox lookup
+	mov	@data[0].s[0],$roundkey
+	tbl	@vtmp[1].16b,{@sbox[0].16b,@sbox[1].16b,@sbox[2].16b,@sbox[3].16b},@data[0].16b
+	sub	@data[0].16b,@data[0].16b,@vtmp[0].16b
+	tbx	@vtmp[1].16b,{@sbox[4].16b,@sbox[5].16b,@sbox[6].16b,@sbox[7].16b},@data[0].16b
+	sub	@data[0].16b,@data[0].16b,@vtmp[0].16b
+	tbx	@vtmp[1].16b,{@sbox[8].16b,@sbox[9].16b,@sbox[10].16b,@sbox[11].16b},@data[0].16b
+	sub	@data[0].16b,@data[0].16b,@vtmp[0].16b
+	tbx	@vtmp[1].16b,{@sbox[12].16b,@sbox[13].16b,@sbox[14].16b,@sbox[15].16b},@data[0].16b
+	mov	$wtmp,@vtmp[1].s[0]
+	eor	$roundkey,$wtmp,$wtmp,ror #19
+	eor	$roundkey,$roundkey,$wtmp,ror #9
+	mov	$wtmp,$vkey.s[0]
+	eor	$roundkey,$roundkey,$wtmp
+	mov	$vkey.s[0],$roundkey
+	cbz	$enc,2f
+	str	$roundkey,[$keys],#4
+	b	3f
+2:
+	str	$roundkey,[$keys],#-4
+3:
+	tbl	$vkey.16b,{$vkey.16b},$vmap.16b
+	subs	$schedules,$schedules,#1
+	b.ne	1b
+	ret
+.size	_vpsm4_set_key,.-_vpsm4_set_key
+___
+}}}
+
+
+{{{
+$code.=<<___;
+.type	_vpsm4_enc_4blks,%function
+.align	4
+_vpsm4_enc_4blks:
+	AARCH64_VALID_CALL_TARGET
+___
+	&encrypt_4blks();
+$code.=<<___;
+	ret
+.size	_vpsm4_enc_4blks,.-_vpsm4_enc_4blks
+___
+}}}
+
+{{{
+$code.=<<___;
+.type	_vpsm4_enc_8blks,%function
+.align	4
+_vpsm4_enc_8blks:
+	AARCH64_VALID_CALL_TARGET
+___
+	&encrypt_8blks();
+$code.=<<___;
+	ret
+.size	_vpsm4_enc_8blks,.-_vpsm4_enc_8blks
+___
+}}}
+
+
+{{{
+my ($key,$keys)=("x0","x1");
+$code.=<<___;
+.globl	${prefix}_set_encrypt_key
+.type	${prefix}_set_encrypt_key,%function
+.align	5
+${prefix}_set_encrypt_key:
+	AARCH64_SIGN_LINK_REGISTER
+	stp	x29,x30,[sp,#-16]!
+	mov	w2,1
+	bl	_vpsm4_set_key
+	ldp	x29,x30,[sp],#16
+	AARCH64_VALIDATE_LINK_REGISTER
+	ret
+.size	${prefix}_set_encrypt_key,.-${prefix}_set_encrypt_key
+___
+}}}
+
+{{{
+my ($key,$keys)=("x0","x1");
+$code.=<<___;
+.globl	${prefix}_set_decrypt_key
+.type	${prefix}_set_decrypt_key,%function
+.align	5
+${prefix}_set_decrypt_key:
+	AARCH64_SIGN_LINK_REGISTER
+	stp	x29,x30,[sp,#-16]!
+	mov	w2,0
+	bl	_vpsm4_set_key
+	ldp	x29,x30,[sp],#16
+	AARCH64_VALIDATE_LINK_REGISTER
+	ret
+.size	${prefix}_set_decrypt_key,.-${prefix}_set_decrypt_key
+___
+}}}
+
+{{{
+sub gen_block () {
+	my $dir = shift;
+	my ($inp,$outp,$rk)=map("x$_",(0..2));
+
+$code.=<<___;
+.globl	${prefix}_${dir}crypt
+.type	${prefix}_${dir}crypt,%function
+.align	5
+${prefix}_${dir}crypt:
+	AARCH64_VALID_CALL_TARGET
+	ld1	{@data[0].16b},[$inp]
+___
+	&load_sbox();
+	&rev32(@data[0],@data[0]);
+$code.=<<___;
+	mov	$rks,x2
+___
+	&encrypt_1blk(@data[0]);
+$code.=<<___;
+	st1	{@data[0].16b},[$outp]
+	ret
+.size	${prefix}_${dir}crypt,.-${prefix}_${dir}crypt
+___
+}
+&gen_block("en");
+&gen_block("de");
+}}}
+
+{{{
+my ($enc) = ("w4");
+my @dat=map("v$_",(16..23));
+
+$code.=<<___;
+.globl	${prefix}_ecb_encrypt
+.type	${prefix}_ecb_encrypt,%function
+.align	5
+${prefix}_ecb_encrypt:
+	AARCH64_SIGN_LINK_REGISTER
+	// convert length into blocks
+	lsr	x2,x2,4
+	stp	d8,d9,[sp,#-80]!
+	stp	d10,d11,[sp,#16]
+	stp	d12,d13,[sp,#32]
+	stp	d14,d15,[sp,#48]
+	stp	x29,x30,[sp,#64]
+___
+	&load_sbox();
+$code.=<<___;
+.Lecb_8_blocks_process:
+	cmp	$blocks,#8
+	b.lt	.Lecb_4_blocks_process
+	ld4	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$inp],#64
+	ld4	{@datax[0].4s,$datax[1].4s,@datax[2].4s,@datax[3].4s},[$inp],#64
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+	&rev32(@datax[0],@datax[0]);
+	&rev32(@datax[1],@datax[1]);
+	&rev32(@datax[2],@datax[2]);
+	&rev32(@datax[3],@datax[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_8blks
+	st4	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s,@vtmp[3].4s},[$outp],#64
+	st4	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$outp],#64
+	subs	$blocks,$blocks,#8
+	b.gt	.Lecb_8_blocks_process
+	b	100f
+.Lecb_4_blocks_process:
+	cmp	$blocks,#4
+	b.lt	1f
+	ld4	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$inp],#64
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_4blks
+	st4	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s,@vtmp[3].4s},[$outp],#64
+	sub	$blocks,$blocks,#4
+1:
+	// process last block
+	cmp	$blocks,#1
+	b.lt	100f
+	b.gt	1f
+	ld1	{@data[0].16b},[$inp]
+___
+	&rev32(@data[0],@data[0]);
+	&encrypt_1blk(@data[0]);
+$code.=<<___;
+	st1	{@data[0].16b},[$outp]
+	b	100f
+1:	// process last 2 blocks
+	ld4	{@data[0].s,@data[1].s,@data[2].s,@data[3].s}[0],[$inp],#16
+	ld4	{@data[0].s,@data[1].s,@data[2].s,@data[3].s}[1],[$inp],#16
+	cmp	$blocks,#2
+	b.gt	1f
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_4blks
+	st4	{@vtmp[0].s-@vtmp[3].s}[0],[$outp],#16
+	st4	{@vtmp[0].s-@vtmp[3].s}[1],[$outp]
+	b	100f
+1:	// process last 3 blocks
+	ld4	{@data[0].s,@data[1].s,@data[2].s,@data[3].s}[2],[$inp],#16
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_4blks
+	st4	{@vtmp[0].s-@vtmp[3].s}[0],[$outp],#16
+	st4	{@vtmp[0].s-@vtmp[3].s}[1],[$outp],#16
+	st4	{@vtmp[0].s-@vtmp[3].s}[2],[$outp]
+100:
+	ldp	d10,d11,[sp,#16]
+	ldp	d12,d13,[sp,#32]
+	ldp	d14,d15,[sp,#48]
+	ldp	x29,x30,[sp,#64]
+	ldp	d8,d9,[sp],#80
+	AARCH64_VALIDATE_LINK_REGISTER
+	ret
+.size	${prefix}_ecb_encrypt,.-${prefix}_ecb_encrypt
+___
+}}}
+
+{{{
+my ($len,$ivp,$enc)=("x2","x4","w5");
+my $ivec0=("v3");
+my $ivec1=("v15");
+
+$code.=<<___;
+.globl	${prefix}_cbc_encrypt
+.type	${prefix}_cbc_encrypt,%function
+.align	5
+${prefix}_cbc_encrypt:
+	AARCH64_VALID_CALL_TARGET
+	lsr	$len,$len,4
+___
+	&load_sbox();
+$code.=<<___;
+	cbz	$enc,.Ldec
+	ld1	{$ivec0.4s},[$ivp]
+.Lcbc_4_blocks_enc:
+	cmp	$blocks,#4
+	b.lt	1f
+	ld1	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$inp],#64
+	eor	@data[0].16b,@data[0].16b,$ivec0.16b
+___
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+	&encrypt_1blk_norev(@data[0]);
+$code.=<<___;
+	eor	@data[1].16b,@data[1].16b,@data[0].16b
+___
+	&encrypt_1blk_norev(@data[1]);
+	&rev32(@data[0],@data[0]);
+
+$code.=<<___;
+	eor	@data[2].16b,@data[2].16b,@data[1].16b
+___
+	&encrypt_1blk_norev(@data[2]);
+	&rev32(@data[1],@data[1]);
+$code.=<<___;
+	eor	@data[3].16b,@data[3].16b,@data[2].16b
+___
+	&encrypt_1blk_norev(@data[3]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+$code.=<<___;
+	orr	$ivec0.16b,@data[3].16b,@data[3].16b
+	st1	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$outp],#64
+	subs	$blocks,$blocks,#4
+	b.ne	.Lcbc_4_blocks_enc
+	b	2f
+1:
+	subs	$blocks,$blocks,#1
+	b.lt	2f
+	ld1	{@data[0].4s},[$inp],#16
+	eor	$ivec0.16b,$ivec0.16b,@data[0].16b
+___
+	&rev32($ivec0,$ivec0);
+	&encrypt_1blk($ivec0);
+$code.=<<___;
+	st1	{$ivec0.16b},[$outp],#16
+	b	1b
+2:
+	// save back IV
+	st1	{$ivec0.16b},[$ivp]
+	ret
+
+.Ldec:
+	// decryption mode starts
+	AARCH64_SIGN_LINK_REGISTER
+	stp	d8,d9,[sp,#-80]!
+	stp	d10,d11,[sp,#16]
+	stp	d12,d13,[sp,#32]
+	stp	d14,d15,[sp,#48]
+	stp	x29,x30,[sp,#64]
+.Lcbc_8_blocks_dec:
+	cmp	$blocks,#8
+	b.lt	1f
+	ld4	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$inp]
+	add	$ptr,$inp,#64
+	ld4	{@datax[0].4s,@datax[1].4s,@datax[2].4s,@datax[3].4s},[$ptr]
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],$data[3]);
+	&rev32(@datax[0],@datax[0]);
+	&rev32(@datax[1],@datax[1]);
+	&rev32(@datax[2],@datax[2]);
+	&rev32(@datax[3],$datax[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_8blks
+___
+	&transpose(@vtmp,@datax);
+	&transpose(@data,@datax);
+$code.=<<___;
+	ld1	{$ivec1.16b},[$ivp]
+	ld1	{@datax[0].4s,@datax[1].4s,@datax[2].4s,@datax[3].4s},[$inp],#64
+	// note ivec1 and vtmpx[3] are resuing the same register
+	// care needs to be taken to avoid conflict
+	eor	@vtmp[0].16b,@vtmp[0].16b,$ivec1.16b
+	ld1	{@vtmpx[0].4s,@vtmpx[1].4s,@vtmpx[2].4s,@vtmpx[3].4s},[$inp],#64
+	eor	@vtmp[1].16b,@vtmp[1].16b,@datax[0].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@datax[1].16b
+	eor	@vtmp[3].16b,$vtmp[3].16b,@datax[2].16b
+	// save back IV
+	st1	{$vtmpx[3].16b}, [$ivp]
+	eor	@data[0].16b,@data[0].16b,$datax[3].16b
+	eor	@data[1].16b,@data[1].16b,@vtmpx[0].16b
+	eor	@data[2].16b,@data[2].16b,@vtmpx[1].16b
+	eor	@data[3].16b,$data[3].16b,@vtmpx[2].16b
+	st1	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s,@vtmp[3].4s},[$outp],#64
+	st1	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$outp],#64
+	subs	$blocks,$blocks,#8
+	b.gt	.Lcbc_8_blocks_dec
+	b.eq	100f
+1:
+	ld1	{$ivec1.16b},[$ivp]
+.Lcbc_4_blocks_dec:
+	cmp	$blocks,#4
+	b.lt	1f
+	ld4	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$inp]
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],$data[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_4blks
+	ld1	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$inp],#64
+___
+	&transpose(@vtmp,@datax);
+$code.=<<___;
+	eor	@vtmp[0].16b,@vtmp[0].16b,$ivec1.16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@data[0].16b
+	orr	$ivec1.16b,@data[3].16b,@data[3].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@data[1].16b
+	eor	@vtmp[3].16b,$vtmp[3].16b,@data[2].16b
+	st1	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s,@vtmp[3].4s},[$outp],#64
+	subs	$blocks,$blocks,#4
+	b.gt	.Lcbc_4_blocks_dec
+	// save back IV
+	st1	{@vtmp[3].16b}, [$ivp]
+	b	100f
+1:	// last block
+	subs	$blocks,$blocks,#1
+	b.lt	100f
+	b.gt	1f
+	ld1	{@data[0].4s},[$inp],#16
+	// save back IV
+	st1	{$data[0].16b}, [$ivp]
+___
+	&rev32(@datax[0],@data[0]);
+	&encrypt_1blk(@datax[0]);
+$code.=<<___;
+	eor	@datax[0].16b,@datax[0].16b,$ivec1.16b
+	st1	{@datax[0].16b},[$outp],#16
+	b	100f
+1:	// last two blocks
+	ld4	{@data[0].s,@data[1].s,@data[2].s,@data[3].s}[0],[$inp]
+	add	$ptr,$inp,#16
+	ld4	{@data[0].s,@data[1].s,@data[2].s,@data[3].s}[1],[$ptr],#16
+	subs	$blocks,$blocks,1
+	b.gt	1f
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_4blks
+	ld1	{@data[0].4s,@data[1].4s},[$inp],#32
+___
+	&transpose(@vtmp,@datax);
+$code.=<<___;
+	eor	@vtmp[0].16b,@vtmp[0].16b,$ivec1.16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@data[0].16b
+	st1	{@vtmp[0].4s,@vtmp[1].4s},[$outp],#32
+	// save back IV
+	st1	{@data[1].16b}, [$ivp]
+	b	100f
+1:	// last 3 blocks
+	ld4	{@data[0].s,@data[1].s,@data[2].s,@data[3].s}[2],[$ptr]
+___
+	&rev32(@data[0],@data[0]);
+	&rev32(@data[1],@data[1]);
+	&rev32(@data[2],@data[2]);
+	&rev32(@data[3],@data[3]);
+$code.=<<___;
+	bl	_vpsm4_enc_4blks
+	ld1	{@data[0].4s,@data[1].4s,@data[2].4s},[$inp],#48
+___
+	&transpose(@vtmp,@datax);
+$code.=<<___;
+	eor	@vtmp[0].16b,@vtmp[0].16b,$ivec1.16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@data[0].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@data[1].16b
+	st1	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s},[$outp],#48
+	// save back IV
+	st1	{@data[2].16b}, [$ivp]
+100:
+	ldp	d10,d11,[sp,#16]
+	ldp	d12,d13,[sp,#32]
+	ldp	d14,d15,[sp,#48]
+	ldp	x29,x30,[sp,#64]
+	ldp	d8,d9,[sp],#80
+	AARCH64_VALIDATE_LINK_REGISTER
+	ret
+.size	${prefix}_cbc_encrypt,.-${prefix}_cbc_encrypt
+___
+}}}
+
+{{{
+my ($ivp)=("x4");
+my ($ctr)=("w5");
+my $ivec=("v3");
+
+$code.=<<___;
+.globl	${prefix}_ctr32_encrypt_blocks
+.type	${prefix}_ctr32_encrypt_blocks,%function
+.align	5
+${prefix}_ctr32_encrypt_blocks:
+	AARCH64_VALID_CALL_TARGET
+	ld1	{$ivec.4s},[$ivp]
+___
+	&rev32($ivec,$ivec);
+	&load_sbox();
+$code.=<<___;
+	cmp	$blocks,#1
+	b.ne	1f
+	// fast processing for one single block without
+	// context saving overhead
+___
+	&encrypt_1blk($ivec);
+$code.=<<___;
+	ld1	{@data[0].16b},[$inp]
+	eor	@data[0].16b,@data[0].16b,$ivec.16b
+	st1	{@data[0].16b},[$outp]
+	ret
+1:
+	AARCH64_SIGN_LINK_REGISTER
+	stp	d8,d9,[sp,#-80]!
+	stp	d10,d11,[sp,#16]
+	stp	d12,d13,[sp,#32]
+	stp	d14,d15,[sp,#48]
+	stp	x29,x30,[sp,#64]
+	mov	$word0,$ivec.s[0]
+	mov	$word1,$ivec.s[1]
+	mov	$word2,$ivec.s[2]
+	mov	$ctr,$ivec.s[3]
+.Lctr32_4_blocks_process:
+	cmp	$blocks,#4
+	b.lt	1f
+	dup	@data[0].4s,$word0
+	dup	@data[1].4s,$word1
+	dup	@data[2].4s,$word2
+	mov	@data[3].s[0],$ctr
+	add	$ctr,$ctr,#1
+	mov	$data[3].s[1],$ctr
+	add	$ctr,$ctr,#1
+	mov	@data[3].s[2],$ctr
+	add	$ctr,$ctr,#1
+	mov	@data[3].s[3],$ctr
+	add	$ctr,$ctr,#1
+	cmp	$blocks,#8
+	b.ge	.Lctr32_8_blocks_process
+	bl	_vpsm4_enc_4blks
+	ld4	{@vtmpx[0].4s,@vtmpx[1].4s,@vtmpx[2].4s,@vtmpx[3].4s},[$inp],#64
+	eor	@vtmp[0].16b,@vtmp[0].16b,@vtmpx[0].16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@vtmpx[1].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@vtmpx[2].16b
+	eor	@vtmp[3].16b,@vtmp[3].16b,@vtmpx[3].16b
+	st4	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s,@vtmp[3].4s},[$outp],#64
+	subs	$blocks,$blocks,#4
+	b.ne	.Lctr32_4_blocks_process
+	b	100f
+.Lctr32_8_blocks_process:
+	dup	@datax[0].4s,$word0
+	dup	@datax[1].4s,$word1
+	dup	@datax[2].4s,$word2
+	mov	@datax[3].s[0],$ctr
+	add	$ctr,$ctr,#1
+	mov	$datax[3].s[1],$ctr
+	add	$ctr,$ctr,#1
+	mov	@datax[3].s[2],$ctr
+	add	$ctr,$ctr,#1
+	mov	@datax[3].s[3],$ctr
+	add	$ctr,$ctr,#1
+	bl	_vpsm4_enc_8blks
+	ld4	{@vtmpx[0].4s,@vtmpx[1].4s,@vtmpx[2].4s,@vtmpx[3].4s},[$inp],#64
+	ld4	{@datax[0].4s,@datax[1].4s,@datax[2].4s,@datax[3].4s},[$inp],#64
+	eor	@vtmp[0].16b,@vtmp[0].16b,@vtmpx[0].16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@vtmpx[1].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@vtmpx[2].16b
+	eor	@vtmp[3].16b,@vtmp[3].16b,@vtmpx[3].16b
+	eor	@data[0].16b,@data[0].16b,@datax[0].16b
+	eor	@data[1].16b,@data[1].16b,@datax[1].16b
+	eor	@data[2].16b,@data[2].16b,@datax[2].16b
+	eor	@data[3].16b,@data[3].16b,@datax[3].16b
+	st4	{@vtmp[0].4s,@vtmp[1].4s,@vtmp[2].4s,@vtmp[3].4s},[$outp],#64
+	st4	{@data[0].4s,@data[1].4s,@data[2].4s,@data[3].4s},[$outp],#64
+	subs	$blocks,$blocks,#8
+	b.ne	.Lctr32_4_blocks_process
+	b	100f
+1:	// last block processing
+	subs	$blocks,$blocks,#1
+	b.lt	100f
+	b.gt	1f
+	mov	$ivec.s[0],$word0
+	mov	$ivec.s[1],$word1
+	mov	$ivec.s[2],$word2
+	mov	$ivec.s[3],$ctr
+___
+	&encrypt_1blk($ivec);
+$code.=<<___;
+	ld1	{@data[0].16b},[$inp]
+	eor	@data[0].16b,@data[0].16b,$ivec.16b
+	st1	{@data[0].16b},[$outp]
+	b	100f
+1:	// last 2 blocks processing
+	dup	@data[0].4s,$word0
+	dup	@data[1].4s,$word1
+	dup	@data[2].4s,$word2
+	mov	@data[3].s[0],$ctr
+	add	$ctr,$ctr,#1
+	mov	@data[3].s[1],$ctr
+	subs	$blocks,$blocks,#1
+	b.ne	1f
+	bl	_vpsm4_enc_4blks
+	ld4	{@vtmpx[0].s,@vtmpx[1].s,@vtmpx[2].s,@vtmpx[3].s}[0],[$inp],#16
+	ld4	{@vtmpx[0].s,@vtmpx[1].s,@vtmpx[2].s,@vtmpx[3].s}[1],[$inp],#16
+	eor	@vtmp[0].16b,@vtmp[0].16b,@vtmpx[0].16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@vtmpx[1].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@vtmpx[2].16b
+	eor	@vtmp[3].16b,@vtmp[3].16b,@vtmpx[3].16b
+	st4	{@vtmp[0].s,@vtmp[1].s,@vtmp[2].s,@vtmp[3].s}[0],[$outp],#16
+	st4	{@vtmp[0].s,@vtmp[1].s,@vtmp[2].s,@vtmp[3].s}[1],[$outp],#16
+	b	100f
+1:	// last 3 blocks processing
+	add	$ctr,$ctr,#1
+	mov	@data[3].s[2],$ctr
+	bl	_vpsm4_enc_4blks
+	ld4	{@vtmpx[0].s,@vtmpx[1].s,@vtmpx[2].s,@vtmpx[3].s}[0],[$inp],#16
+	ld4	{@vtmpx[0].s,@vtmpx[1].s,@vtmpx[2].s,@vtmpx[3].s}[1],[$inp],#16
+	ld4	{@vtmpx[0].s,@vtmpx[1].s,@vtmpx[2].s,@vtmpx[3].s}[2],[$inp],#16
+	eor	@vtmp[0].16b,@vtmp[0].16b,@vtmpx[0].16b
+	eor	@vtmp[1].16b,@vtmp[1].16b,@vtmpx[1].16b
+	eor	@vtmp[2].16b,@vtmp[2].16b,@vtmpx[2].16b
+	eor	@vtmp[3].16b,@vtmp[3].16b,@vtmpx[3].16b
+	st4	{@vtmp[0].s,@vtmp[1].s,@vtmp[2].s,@vtmp[3].s}[0],[$outp],#16
+	st4	{@vtmp[0].s,@vtmp[1].s,@vtmp[2].s,@vtmp[3].s}[1],[$outp],#16
+	st4	{@vtmp[0].s,@vtmp[1].s,@vtmp[2].s,@vtmp[3].s}[2],[$outp],#16
+100:
+	ldp	d10,d11,[sp,#16]
+	ldp	d12,d13,[sp,#32]
+	ldp	d14,d15,[sp,#48]
+	ldp	x29,x30,[sp,#64]
+	ldp	d8,d9,[sp],#80
+	AARCH64_VALIDATE_LINK_REGISTER
+	ret
+.size	${prefix}_ctr32_encrypt_blocks,.-${prefix}_ctr32_encrypt_blocks
+___
+}}}
+########################################
+open SELF,$0;
+while(<SELF>) {
+        next if (/^#!/);
+        last if (!s/^#/\/\// and !/^$/);
+        print;
+}
+close SELF;
+
+foreach(split("\n",$code)) {
+	s/\`([^\`]*)\`/eval($1)/ge;
+	print $_,"\n";
+}
+
+close STDOUT or die "error closing STDOUT: $!";
diff --git a/crypto/sm4/build.info b/crypto/sm4/build.info
index e27aa49e67..75a215ab80 100644
--- a/crypto/sm4/build.info
+++ b/crypto/sm4/build.info
@@ -1,8 +1,8 @@
 LIBS=../../libcrypto
 
 IF[{- !$disabled{asm} -}]
-  $SM4DEF_aarch64=SM4_ASM
-  $SM4ASM_aarch64=sm4-armv8.S
+  $SM4DEF_aarch64=SM4_ASM VPSM4_ASM
+  $SM4ASM_aarch64=sm4-armv8.S vpsm4-armv8.S
 
   # Now that we have defined all the arch specific variables, use the
   # appropriate one, and define the appropriate macros
@@ -29,4 +29,6 @@ IF[{- !$disabled{module} && !$disabled{shared} -}]
 ENDIF
 
 GENERATE[sm4-armv8.S]=asm/sm4-armv8.pl
+GENERATE[vpsm4-armv8.S]=asm/vpsm4-armv8.pl
 INCLUDE[sm4-armv8.o]=..
+INCLUDE[vpsm4-armv8.o]=..
diff --git a/include/crypto/sm4_platform.h b/include/crypto/sm4_platform.h
index 42c8b44a43..11f9b9d88b 100644
--- a/include/crypto/sm4_platform.h
+++ b/include/crypto/sm4_platform.h
@@ -15,6 +15,16 @@
 #  if (defined(__arm__) || defined(__arm) || defined(__aarch64__))
 #   include "arm_arch.h"
 #   if __ARM_MAX_ARCH__>=8
+extern unsigned int OPENSSL_arm_midr;
+static inline int vpsm4_capable(void)
+{
+    return (OPENSSL_armcap_P & ARMV8_CPUID) &&
+            (MIDR_IS_CPU_MODEL(OPENSSL_arm_midr, ARM_CPU_IMP_ARM, ARM_CPU_PART_V1) ||
+             MIDR_IS_CPU_MODEL(OPENSSL_arm_midr, ARM_CPU_IMP_ARM, ARM_CPU_PART_N1));
+}
+#    if defined(VPSM4_ASM)
+#     define VPSM4_CAPABLE vpsm4_capable()
+#    endif
 #    define HWSM4_CAPABLE (OPENSSL_armcap_P & ARMV8_SM4)
 #    define HWSM4_set_encrypt_key sm4_v8_set_encrypt_key
 #    define HWSM4_set_decrypt_key sm4_v8_set_decrypt_key
@@ -45,4 +55,23 @@ void HWSM4_ctr32_encrypt_blocks(const unsigned char *in, unsigned char *out,
                                 const unsigned char ivec[16]);
 # endif /* HWSM4_CAPABLE */
 
+#ifdef VPSM4_CAPABLE
+int vpsm4_set_encrypt_key(const unsigned char *userKey, SM4_KEY *key);
+int vpsm4_set_decrypt_key(const unsigned char *userKey, SM4_KEY *key);
+void vpsm4_encrypt(const unsigned char *in, unsigned char *out,
+                   const SM4_KEY *key);
+void vpsm4_decrypt(const unsigned char *in, unsigned char *out,
+                   const SM4_KEY *key);
+void vpsm4_cbc_encrypt(const unsigned char *in, unsigned char *out,
+                       size_t length, const SM4_KEY *key,
+                       unsigned char *ivec, const int enc);
+void vpsm4_ecb_encrypt(const unsigned char *in, unsigned char *out,
+                       size_t length, const SM4_KEY *key,
+                       const int enc);
+void vpsm4_ctr32_encrypt_blocks(const unsigned char *in, unsigned char *out,
+                                size_t len, const void *key,
+                                const unsigned char ivec[16]);
+# endif /* VPSM4_CAPABLE */
+
+
 #endif /* OSSL_SM4_PLATFORM_H */
diff --git a/providers/implementations/ciphers/cipher_sm4_gcm_hw.c b/providers/implementations/ciphers/cipher_sm4_gcm_hw.c
index b9633f83ed..db7fe0fe2f 100644
--- a/providers/implementations/ciphers/cipher_sm4_gcm_hw.c
+++ b/providers/implementations/ciphers/cipher_sm4_gcm_hw.c
@@ -32,6 +32,13 @@ static int sm4_gcm_initkey(PROV_GCM_CTX *ctx, const unsigned char *key,
 #  endif
     } else
 # endif /* HWSM4_CAPABLE */
+# ifdef VPSM4_CAPABLE
+    if (VPSM4_CAPABLE) {
+        vpsm4_set_encrypt_key(key, ks);
+        CRYPTO_gcm128_init(&ctx->gcm, ks, (block128_f) vpsm4_encrypt);
+        ctx->ctr = (ctr128_f) vpsm4_ctr32_encrypt_blocks;
+    } else
+# endif /* VPSM4_CAPABLE */
     {
         ossl_sm4_set_key(key, ks);
         CRYPTO_gcm128_init(&ctx->gcm, ks, (block128_f)ossl_sm4_encrypt);
diff --git a/providers/implementations/ciphers/cipher_sm4_hw.c b/providers/implementations/ciphers/cipher_sm4_hw.c
index 4cd3d3d669..9a2e99f67c 100644
--- a/providers/implementations/ciphers/cipher_sm4_hw.c
+++ b/providers/implementations/ciphers/cipher_sm4_hw.c
@@ -41,6 +41,19 @@ static int cipher_hw_sm4_initkey(PROV_CIPHER_CTX *ctx,
 #endif
             (void)0;            /* terminate potentially open 'else' */
         } else
+#endif
+#ifdef VPSM4_CAPABLE
+        if (VPSM4_CAPABLE) {
+            vpsm4_set_encrypt_key(key, ks);
+            ctx->block = (block128_f)vpsm4_encrypt;
+            ctx->stream.cbc = NULL;
+            if (ctx->mode == EVP_CIPH_CBC_MODE)
+                ctx->stream.cbc = (cbc128_f)vpsm4_cbc_encrypt;
+            else if (ctx->mode == EVP_CIPH_ECB_MODE)
+                ctx->stream.ecb = (ecb128_f)vpsm4_ecb_encrypt;
+            else if (ctx->mode == EVP_CIPH_CTR_MODE)
+                ctx->stream.ctr = (ctr128_f)vpsm4_ctr32_encrypt_blocks;
+        } else
 #endif
         {
             ossl_sm4_set_key(key, ks);
@@ -61,6 +74,17 @@ static int cipher_hw_sm4_initkey(PROV_CIPHER_CTX *ctx,
                 ctx->stream.ecb = (ecb128_f)HWSM4_ecb_encrypt;
 #endif
         } else
+#endif
+#ifdef VPSM4_CAPABLE
+        if (VPSM4_CAPABLE) {
+            vpsm4_set_decrypt_key(key, ks);
+            ctx->block = (block128_f)vpsm4_decrypt;
+            ctx->stream.cbc = NULL;
+            if (ctx->mode == EVP_CIPH_CBC_MODE)
+                ctx->stream.cbc = (cbc128_f)vpsm4_cbc_encrypt;
+        else if (ctx->mode == EVP_CIPH_ECB_MODE)
+                ctx->stream.ecb = (ecb128_f)vpsm4_ecb_encrypt;
+        } else
 #endif
         {
             ossl_sm4_set_key(key, ks);