visaljs 0.0.2 → 0.0.4

package/runtime/third_party/boringssl/gen/bcm/aesv8-armv8-win.S

@@ -0,0 +1,798 @@
+// This file is generated from a similarly-named Perl script in the BoringSSL
+// source tree. Do not edit by hand.
+
+#include <openssl/asm_base.h>
+
+#if !defined(OPENSSL_NO_ASM) && defined(OPENSSL_AARCH64) && defined(_WIN32)
+#if __ARM_MAX_ARCH__>=7
+.text
+.arch	armv8-a+crypto
+.section	.rodata
+
+.align	5
+aes_hw_constants:
+Lrcon:
+.long	0x01,0x01,0x01,0x01
+.long	0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d,0x0c0f0e0d	// rotate-n-splat
+.long	0x1b,0x1b,0x1b,0x1b
+
+.text
+
+.globl	aes_hw_set_encrypt_key
+
+.def aes_hw_set_encrypt_key
+   .type 32
+.endef
+.align	5
+aes_hw_set_encrypt_key:
+Lenc_key:
+	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
+	AARCH64_VALID_CALL_TARGET
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+	mov	x3,#-2
+	cmp	w1,#128
+	b.lt	Lenc_key_abort
+	cmp	w1,#256
+	b.gt	Lenc_key_abort
+	tst	w1,#0x3f
+	b.ne	Lenc_key_abort
+
+	adrp	x3,Lrcon
+	add	x3,x3,:lo12:Lrcon
+	cmp	w1,#192
+
+	eor	v0.16b,v0.16b,v0.16b
+	ld1	{v3.16b},[x0],#16
+	mov	w1,#8		// reuse w1
+	ld1	{v1.4s,v2.4s},[x3],#32
+
+	b.lt	Loop128
+	b.eq	L192
+	b	L256
+
+.align	4
+Loop128:
+	tbl	v6.16b,{v3.16b},v2.16b
+	ext	v5.16b,v0.16b,v3.16b,#12
+	st1	{v3.4s},[x2],#16
+	aese	v6.16b,v0.16b
+	subs	w1,w1,#1
+
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v6.16b,v6.16b,v1.16b
+	eor	v3.16b,v3.16b,v5.16b
+	shl	v1.16b,v1.16b,#1
+	eor	v3.16b,v3.16b,v6.16b
+	b.ne	Loop128
+
+	ld1	{v1.4s},[x3]
+
+	tbl	v6.16b,{v3.16b},v2.16b
+	ext	v5.16b,v0.16b,v3.16b,#12
+	st1	{v3.4s},[x2],#16
+	aese	v6.16b,v0.16b
+
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v6.16b,v6.16b,v1.16b
+	eor	v3.16b,v3.16b,v5.16b
+	shl	v1.16b,v1.16b,#1
+	eor	v3.16b,v3.16b,v6.16b
+
+	tbl	v6.16b,{v3.16b},v2.16b
+	ext	v5.16b,v0.16b,v3.16b,#12
+	st1	{v3.4s},[x2],#16
+	aese	v6.16b,v0.16b
+
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v6.16b,v6.16b,v1.16b
+	eor	v3.16b,v3.16b,v5.16b
+	eor	v3.16b,v3.16b,v6.16b
+	st1	{v3.4s},[x2]
+	add	x2,x2,#0x50
+
+	mov	w12,#10
+	b	Ldone
+
+.align	4
+L192:
+	ld1	{v4.8b},[x0],#8
+	movi	v6.16b,#8			// borrow v6.16b
+	st1	{v3.4s},[x2],#16
+	sub	v2.16b,v2.16b,v6.16b	// adjust the mask
+
+Loop192:
+	tbl	v6.16b,{v4.16b},v2.16b
+	ext	v5.16b,v0.16b,v3.16b,#12
+	st1	{v4.8b},[x2],#8
+	aese	v6.16b,v0.16b
+	subs	w1,w1,#1
+
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v3.16b,v3.16b,v5.16b
+
+	dup	v5.4s,v3.s[3]
+	eor	v5.16b,v5.16b,v4.16b
+	eor	v6.16b,v6.16b,v1.16b
+	ext	v4.16b,v0.16b,v4.16b,#12
+	shl	v1.16b,v1.16b,#1
+	eor	v4.16b,v4.16b,v5.16b
+	eor	v3.16b,v3.16b,v6.16b
+	eor	v4.16b,v4.16b,v6.16b
+	st1	{v3.4s},[x2],#16
+	b.ne	Loop192
+
+	mov	w12,#12
+	add	x2,x2,#0x20
+	b	Ldone
+
+.align	4
+L256:
+	ld1	{v4.16b},[x0]
+	mov	w1,#7
+	mov	w12,#14
+	st1	{v3.4s},[x2],#16
+
+Loop256:
+	tbl	v6.16b,{v4.16b},v2.16b
+	ext	v5.16b,v0.16b,v3.16b,#12
+	st1	{v4.4s},[x2],#16
+	aese	v6.16b,v0.16b
+	subs	w1,w1,#1
+
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v3.16b,v3.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v6.16b,v6.16b,v1.16b
+	eor	v3.16b,v3.16b,v5.16b
+	shl	v1.16b,v1.16b,#1
+	eor	v3.16b,v3.16b,v6.16b
+	st1	{v3.4s},[x2],#16
+	b.eq	Ldone
+
+	dup	v6.4s,v3.s[3]		// just splat
+	ext	v5.16b,v0.16b,v4.16b,#12
+	aese	v6.16b,v0.16b
+
+	eor	v4.16b,v4.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v4.16b,v4.16b,v5.16b
+	ext	v5.16b,v0.16b,v5.16b,#12
+	eor	v4.16b,v4.16b,v5.16b
+
+	eor	v4.16b,v4.16b,v6.16b
+	b	Loop256
+
+Ldone:
+	str	w12,[x2]
+	mov	x3,#0
+
+Lenc_key_abort:
+	mov	x0,x3			// return value
+	ldr	x29,[sp],#16
+	ret
+
+
+.globl	aes_hw_set_decrypt_key
+
+.def aes_hw_set_decrypt_key
+   .type 32
+.endef
+.align	5
+aes_hw_set_decrypt_key:
+	AARCH64_SIGN_LINK_REGISTER
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+	bl	Lenc_key
+
+	cmp	x0,#0
+	b.ne	Ldec_key_abort
+
+	sub	x2,x2,#240		// restore original x2
+	mov	x4,#-16
+	add	x0,x2,x12,lsl#4	// end of key schedule
+
+	ld1	{v0.4s},[x2]
+	ld1	{v1.4s},[x0]
+	st1	{v0.4s},[x0],x4
+	st1	{v1.4s},[x2],#16
+
+Loop_imc:
+	ld1	{v0.4s},[x2]
+	ld1	{v1.4s},[x0]
+	aesimc	v0.16b,v0.16b
+	aesimc	v1.16b,v1.16b
+	st1	{v0.4s},[x0],x4
+	st1	{v1.4s},[x2],#16
+	cmp	x0,x2
+	b.hi	Loop_imc
+
+	ld1	{v0.4s},[x2]
+	aesimc	v0.16b,v0.16b
+	st1	{v0.4s},[x0]
+
+	eor	x0,x0,x0		// return value
+Ldec_key_abort:
+	ldp	x29,x30,[sp],#16
+	AARCH64_VALIDATE_LINK_REGISTER
+	ret
+
+.globl	aes_hw_encrypt
+
+.def aes_hw_encrypt
+   .type 32
+.endef
+.align	5
+aes_hw_encrypt:
+	AARCH64_VALID_CALL_TARGET
+	ldr	w3,[x2,#240]
+	ld1	{v0.4s},[x2],#16
+	ld1	{v2.16b},[x0]
+	sub	w3,w3,#2
+	ld1	{v1.4s},[x2],#16
+
+Loop_enc:
+	aese	v2.16b,v0.16b
+	aesmc	v2.16b,v2.16b
+	ld1	{v0.4s},[x2],#16
+	subs	w3,w3,#2
+	aese	v2.16b,v1.16b
+	aesmc	v2.16b,v2.16b
+	ld1	{v1.4s},[x2],#16
+	b.gt	Loop_enc
+
+	aese	v2.16b,v0.16b
+	aesmc	v2.16b,v2.16b
+	ld1	{v0.4s},[x2]
+	aese	v2.16b,v1.16b
+	eor	v2.16b,v2.16b,v0.16b
+
+	st1	{v2.16b},[x1]
+	ret
+
+.globl	aes_hw_decrypt
+
+.def aes_hw_decrypt
+   .type 32
+.endef
+.align	5
+aes_hw_decrypt:
+	AARCH64_VALID_CALL_TARGET
+	ldr	w3,[x2,#240]
+	ld1	{v0.4s},[x2],#16
+	ld1	{v2.16b},[x0]
+	sub	w3,w3,#2
+	ld1	{v1.4s},[x2],#16
+
+Loop_dec:
+	aesd	v2.16b,v0.16b
+	aesimc	v2.16b,v2.16b
+	ld1	{v0.4s},[x2],#16
+	subs	w3,w3,#2
+	aesd	v2.16b,v1.16b
+	aesimc	v2.16b,v2.16b
+	ld1	{v1.4s},[x2],#16
+	b.gt	Loop_dec
+
+	aesd	v2.16b,v0.16b
+	aesimc	v2.16b,v2.16b
+	ld1	{v0.4s},[x2]
+	aesd	v2.16b,v1.16b
+	eor	v2.16b,v2.16b,v0.16b
+
+	st1	{v2.16b},[x1]
+	ret
+
+.globl	aes_hw_cbc_encrypt
+
+.def aes_hw_cbc_encrypt
+   .type 32
+.endef
+.align	5
+aes_hw_cbc_encrypt:
+	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
+	AARCH64_VALID_CALL_TARGET
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+	subs	x2,x2,#16
+	mov	x8,#16
+	b.lo	Lcbc_abort
+	csel	x8,xzr,x8,eq
+
+	cmp	w5,#0			// en- or decrypting?
+	ldr	w5,[x3,#240]
+	and	x2,x2,#-16
+	ld1	{v6.16b},[x4]
+	ld1	{v0.16b},[x0],x8
+
+	ld1	{v16.4s,v17.4s},[x3]		// load key schedule...
+	sub	w5,w5,#6
+	add	x7,x3,x5,lsl#4	// pointer to last 7 round keys
+	sub	w5,w5,#2
+	ld1	{v18.4s,v19.4s},[x7],#32
+	ld1	{v20.4s,v21.4s},[x7],#32
+	ld1	{v22.4s,v23.4s},[x7],#32
+	ld1	{v7.4s},[x7]
+
+	add	x7,x3,#32
+	mov	w6,w5
+	b.eq	Lcbc_dec
+
+	cmp	w5,#2
+	eor	v0.16b,v0.16b,v6.16b
+	eor	v5.16b,v16.16b,v7.16b
+	b.eq	Lcbc_enc128
+
+	ld1	{v2.4s,v3.4s},[x7]
+	add	x7,x3,#16
+	add	x6,x3,#16*4
+	add	x12,x3,#16*5
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	add	x14,x3,#16*6
+	add	x3,x3,#16*7
+	b	Lenter_cbc_enc
+
+.align	4
+Loop_cbc_enc:
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	st1	{v6.16b},[x1],#16
+Lenter_cbc_enc:
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v2.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v16.4s},[x6]
+	cmp	w5,#4
+	aese	v0.16b,v3.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v17.4s},[x12]
+	b.eq	Lcbc_enc192
+
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v16.4s},[x14]
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v17.4s},[x3]
+	nop
+
+Lcbc_enc192:
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	subs	x2,x2,#16
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	csel	x8,xzr,x8,eq
+	aese	v0.16b,v18.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v19.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v16.16b},[x0],x8
+	aese	v0.16b,v20.16b
+	aesmc	v0.16b,v0.16b
+	eor	v16.16b,v16.16b,v5.16b
+	aese	v0.16b,v21.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v17.4s},[x7]		// re-pre-load rndkey[1]
+	aese	v0.16b,v22.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v23.16b
+	eor	v6.16b,v0.16b,v7.16b
+	b.hs	Loop_cbc_enc
+
+	st1	{v6.16b},[x1],#16
+	b	Lcbc_done
+
+.align	5
+Lcbc_enc128:
+	ld1	{v2.4s,v3.4s},[x7]
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	b	Lenter_cbc_enc128
+Loop_cbc_enc128:
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	st1	{v6.16b},[x1],#16
+Lenter_cbc_enc128:
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	subs	x2,x2,#16
+	aese	v0.16b,v2.16b
+	aesmc	v0.16b,v0.16b
+	csel	x8,xzr,x8,eq
+	aese	v0.16b,v3.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v18.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v19.16b
+	aesmc	v0.16b,v0.16b
+	ld1	{v16.16b},[x0],x8
+	aese	v0.16b,v20.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v21.16b
+	aesmc	v0.16b,v0.16b
+	aese	v0.16b,v22.16b
+	aesmc	v0.16b,v0.16b
+	eor	v16.16b,v16.16b,v5.16b
+	aese	v0.16b,v23.16b
+	eor	v6.16b,v0.16b,v7.16b
+	b.hs	Loop_cbc_enc128
+
+	st1	{v6.16b},[x1],#16
+	b	Lcbc_done
+.align	5
+Lcbc_dec:
+	ld1	{v18.16b},[x0],#16
+	subs	x2,x2,#32		// bias
+	add	w6,w5,#2
+	orr	v3.16b,v0.16b,v0.16b
+	orr	v1.16b,v0.16b,v0.16b
+	orr	v19.16b,v18.16b,v18.16b
+	b.lo	Lcbc_dec_tail
+
+	orr	v1.16b,v18.16b,v18.16b
+	ld1	{v18.16b},[x0],#16
+	orr	v2.16b,v0.16b,v0.16b
+	orr	v3.16b,v1.16b,v1.16b
+	orr	v19.16b,v18.16b,v18.16b
+
+Loop3x_cbc_dec:
+	aesd	v0.16b,v16.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v16.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v16.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v16.4s},[x7],#16
+	subs	w6,w6,#2
+	aesd	v0.16b,v17.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v17.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v17.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v17.4s},[x7],#16
+	b.gt	Loop3x_cbc_dec
+
+	aesd	v0.16b,v16.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v16.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v16.16b
+	aesimc	v18.16b,v18.16b
+	eor	v4.16b,v6.16b,v7.16b
+	subs	x2,x2,#0x30
+	eor	v5.16b,v2.16b,v7.16b
+	csel	x6,x2,x6,lo			// x6, w6, is zero at this point
+	aesd	v0.16b,v17.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v17.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v17.16b
+	aesimc	v18.16b,v18.16b
+	eor	v17.16b,v3.16b,v7.16b
+	add	x0,x0,x6		// x0 is adjusted in such way that
+					// at exit from the loop v1.16b-v18.16b
+					// are loaded with last "words"
+	orr	v6.16b,v19.16b,v19.16b
+	mov	x7,x3
+	aesd	v0.16b,v20.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v20.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v20.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v2.16b},[x0],#16
+	aesd	v0.16b,v21.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v21.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v21.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v3.16b},[x0],#16
+	aesd	v0.16b,v22.16b
+	aesimc	v0.16b,v0.16b
+	aesd	v1.16b,v22.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v22.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v19.16b},[x0],#16
+	aesd	v0.16b,v23.16b
+	aesd	v1.16b,v23.16b
+	aesd	v18.16b,v23.16b
+	ld1	{v16.4s},[x7],#16	// re-pre-load rndkey[0]
+	add	w6,w5,#2
+	eor	v4.16b,v4.16b,v0.16b
+	eor	v5.16b,v5.16b,v1.16b
+	eor	v18.16b,v18.16b,v17.16b
+	ld1	{v17.4s},[x7],#16	// re-pre-load rndkey[1]
+	st1	{v4.16b},[x1],#16
+	orr	v0.16b,v2.16b,v2.16b
+	st1	{v5.16b},[x1],#16
+	orr	v1.16b,v3.16b,v3.16b
+	st1	{v18.16b},[x1],#16
+	orr	v18.16b,v19.16b,v19.16b
+	b.hs	Loop3x_cbc_dec
+
+	cmn	x2,#0x30
+	b.eq	Lcbc_done
+	nop
+
+Lcbc_dec_tail:
+	aesd	v1.16b,v16.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v16.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v16.4s},[x7],#16
+	subs	w6,w6,#2
+	aesd	v1.16b,v17.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v17.16b
+	aesimc	v18.16b,v18.16b
+	ld1	{v17.4s},[x7],#16
+	b.gt	Lcbc_dec_tail
+
+	aesd	v1.16b,v16.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v16.16b
+	aesimc	v18.16b,v18.16b
+	aesd	v1.16b,v17.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v17.16b
+	aesimc	v18.16b,v18.16b
+	aesd	v1.16b,v20.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v20.16b
+	aesimc	v18.16b,v18.16b
+	cmn	x2,#0x20
+	aesd	v1.16b,v21.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v21.16b
+	aesimc	v18.16b,v18.16b
+	eor	v5.16b,v6.16b,v7.16b
+	aesd	v1.16b,v22.16b
+	aesimc	v1.16b,v1.16b
+	aesd	v18.16b,v22.16b
+	aesimc	v18.16b,v18.16b
+	eor	v17.16b,v3.16b,v7.16b
+	aesd	v1.16b,v23.16b
+	aesd	v18.16b,v23.16b
+	b.eq	Lcbc_dec_one
+	eor	v5.16b,v5.16b,v1.16b
+	eor	v17.16b,v17.16b,v18.16b
+	orr	v6.16b,v19.16b,v19.16b
+	st1	{v5.16b},[x1],#16
+	st1	{v17.16b},[x1],#16
+	b	Lcbc_done
+
+Lcbc_dec_one:
+	eor	v5.16b,v5.16b,v18.16b
+	orr	v6.16b,v19.16b,v19.16b
+	st1	{v5.16b},[x1],#16
+
+Lcbc_done:
+	st1	{v6.16b},[x4]
+Lcbc_abort:
+	ldr	x29,[sp],#16
+	ret
+
+.globl	aes_hw_ctr32_encrypt_blocks
+
+.def aes_hw_ctr32_encrypt_blocks
+   .type 32
+.endef
+.align	5
+aes_hw_ctr32_encrypt_blocks:
+	// Armv8.3-A PAuth: even though x30 is pushed to stack it is not popped later.
+	AARCH64_VALID_CALL_TARGET
+	stp	x29,x30,[sp,#-16]!
+	add	x29,sp,#0
+	ldr	w5,[x3,#240]
+
+	ldr	w8, [x4, #12]
+	ld1	{v0.4s},[x4]
+
+	ld1	{v16.4s,v17.4s},[x3]		// load key schedule...
+	sub	w5,w5,#4
+	mov	x12,#16
+	cmp	x2,#2
+	add	x7,x3,x5,lsl#4	// pointer to last 5 round keys
+	sub	w5,w5,#2
+	ld1	{v20.4s,v21.4s},[x7],#32
+	ld1	{v22.4s,v23.4s},[x7],#32
+	ld1	{v7.4s},[x7]
+	add	x7,x3,#32
+	mov	w6,w5
+	csel	x12,xzr,x12,lo
+
+	// ARM Cortex-A57 and Cortex-A72 cores running in 32-bit mode are
+	// affected by silicon errata #1742098 [0] and #1655431 [1],
+	// respectively, where the second instruction of an aese/aesmc
+	// instruction pair may execute twice if an interrupt is taken right
+	// after the first instruction consumes an input register of which a
+	// single 32-bit lane has been updated the last time it was modified.
+	//
+	// This function uses a counter in one 32-bit lane. The vmov lines
+	// could write to v1.16b and v18.16b directly, but that trips this bugs.
+	// We write to v6.16b and copy to the final register as a workaround.
+	//
+	// [0] ARM-EPM-049219 v23 Cortex-A57 MPCore Software Developers Errata Notice
+	// [1] ARM-EPM-012079 v11.0 Cortex-A72 MPCore Software Developers Errata Notice
+#ifndef __AARCH64EB__
+	rev	w8, w8
+#endif
+	add	w10, w8, #1
+	orr	v6.16b,v0.16b,v0.16b
+	rev	w10, w10
+	mov	v6.s[3],w10
+	add	w8, w8, #2
+	orr	v1.16b,v6.16b,v6.16b
+	b.ls	Lctr32_tail
+	rev	w12, w8
+	mov	v6.s[3],w12
+	sub	x2,x2,#3		// bias
+	orr	v18.16b,v6.16b,v6.16b
+	b	Loop3x_ctr32
+
+.align	4
+Loop3x_ctr32:
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v16.16b
+	aesmc	v1.16b,v1.16b
+	aese	v18.16b,v16.16b
+	aesmc	v18.16b,v18.16b
+	ld1	{v16.4s},[x7],#16
+	subs	w6,w6,#2
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v17.16b
+	aesmc	v1.16b,v1.16b
+	aese	v18.16b,v17.16b
+	aesmc	v18.16b,v18.16b
+	ld1	{v17.4s},[x7],#16
+	b.gt	Loop3x_ctr32
+
+	aese	v0.16b,v16.16b
+	aesmc	v4.16b,v0.16b
+	aese	v1.16b,v16.16b
+	aesmc	v5.16b,v1.16b
+	ld1	{v2.16b},[x0],#16
+	add	w9,w8,#1
+	aese	v18.16b,v16.16b
+	aesmc	v18.16b,v18.16b
+	ld1	{v3.16b},[x0],#16
+	rev	w9,w9
+	aese	v4.16b,v17.16b
+	aesmc	v4.16b,v4.16b
+	aese	v5.16b,v17.16b
+	aesmc	v5.16b,v5.16b
+	ld1	{v19.16b},[x0],#16
+	mov	x7,x3
+	aese	v18.16b,v17.16b
+	aesmc	v17.16b,v18.16b
+	aese	v4.16b,v20.16b
+	aesmc	v4.16b,v4.16b
+	aese	v5.16b,v20.16b
+	aesmc	v5.16b,v5.16b
+	eor	v2.16b,v2.16b,v7.16b
+	add	w10,w8,#2
+	aese	v17.16b,v20.16b
+	aesmc	v17.16b,v17.16b
+	eor	v3.16b,v3.16b,v7.16b
+	add	w8,w8,#3
+	aese	v4.16b,v21.16b
+	aesmc	v4.16b,v4.16b
+	aese	v5.16b,v21.16b
+	aesmc	v5.16b,v5.16b
+	 // Note the logic to update v0.16b, v1.16b, and v1.16b is written to work
+	 // around a bug in ARM Cortex-A57 and Cortex-A72 cores running in
+	 // 32-bit mode. See the comment above.
+	eor	v19.16b,v19.16b,v7.16b
+	mov	v6.s[3], w9
+	aese	v17.16b,v21.16b
+	aesmc	v17.16b,v17.16b
+	orr	v0.16b,v6.16b,v6.16b
+	rev	w10,w10
+	aese	v4.16b,v22.16b
+	aesmc	v4.16b,v4.16b
+	mov	v6.s[3], w10
+	rev	w12,w8
+	aese	v5.16b,v22.16b
+	aesmc	v5.16b,v5.16b
+	orr	v1.16b,v6.16b,v6.16b
+	mov	v6.s[3], w12
+	aese	v17.16b,v22.16b
+	aesmc	v17.16b,v17.16b
+	orr	v18.16b,v6.16b,v6.16b
+	subs	x2,x2,#3
+	aese	v4.16b,v23.16b
+	aese	v5.16b,v23.16b
+	aese	v17.16b,v23.16b
+
+	eor	v2.16b,v2.16b,v4.16b
+	ld1	{v16.4s},[x7],#16	// re-pre-load rndkey[0]
+	st1	{v2.16b},[x1],#16
+	eor	v3.16b,v3.16b,v5.16b
+	mov	w6,w5
+	st1	{v3.16b},[x1],#16
+	eor	v19.16b,v19.16b,v17.16b
+	ld1	{v17.4s},[x7],#16	// re-pre-load rndkey[1]
+	st1	{v19.16b},[x1],#16
+	b.hs	Loop3x_ctr32
+
+	adds	x2,x2,#3
+	b.eq	Lctr32_done
+	cmp	x2,#1
+	mov	x12,#16
+	csel	x12,xzr,x12,eq
+
+Lctr32_tail:
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v16.16b
+	aesmc	v1.16b,v1.16b
+	ld1	{v16.4s},[x7],#16
+	subs	w6,w6,#2
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v17.16b
+	aesmc	v1.16b,v1.16b
+	ld1	{v17.4s},[x7],#16
+	b.gt	Lctr32_tail
+
+	aese	v0.16b,v16.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v16.16b
+	aesmc	v1.16b,v1.16b
+	aese	v0.16b,v17.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v17.16b
+	aesmc	v1.16b,v1.16b
+	ld1	{v2.16b},[x0],x12
+	aese	v0.16b,v20.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v20.16b
+	aesmc	v1.16b,v1.16b
+	ld1	{v3.16b},[x0]
+	aese	v0.16b,v21.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v21.16b
+	aesmc	v1.16b,v1.16b
+	eor	v2.16b,v2.16b,v7.16b
+	aese	v0.16b,v22.16b
+	aesmc	v0.16b,v0.16b
+	aese	v1.16b,v22.16b
+	aesmc	v1.16b,v1.16b
+	eor	v3.16b,v3.16b,v7.16b
+	aese	v0.16b,v23.16b
+	aese	v1.16b,v23.16b
+
+	cmp	x2,#1
+	eor	v2.16b,v2.16b,v0.16b
+	eor	v3.16b,v3.16b,v1.16b
+	st1	{v2.16b},[x1],#16
+	b.eq	Lctr32_done
+	st1	{v3.16b},[x1]
+
+Lctr32_done:
+	ldr	x29,[sp],#16
+	ret
+
+#endif
+#endif  // !OPENSSL_NO_ASM && defined(OPENSSL_AARCH64) && defined(_WIN32)