From db42bb440e76399b89fc8ae04644441a2a5f6821 Mon Sep 17 00:00:00 2001
From: Andy Polyakov <appro@openssl.org>
Date: Fri, 15 Feb 2019 22:16:41 +0100
Subject: ARM64 assembly pack: make it Windows-friendly.

"Windows friendliness" means a) unified PIC-ification, unified across
all platforms; b) unified commantary delimiter; c) explicit ldur/stur,
as Visual Studio assembler can't automatically encode ldr/str as
ldur/stur when needed.

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/8256)
---
 crypto/bn/asm/armv8-mont.pl | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

(limited to 'crypto/bn/asm')

diff --git a/crypto/bn/asm/armv8-mont.pl b/crypto/bn/asm/armv8-mont.pl
index c09c783e51..c755555d88 100755
--- a/crypto/bn/asm/armv8-mont.pl
+++ b/crypto/bn/asm/armv8-mont.pl
@@ -197,7 +197,7 @@ bn_mul_mont:
 	mul	$nlo,$nj,$m1		// np[j]*m1
 	adds	$lo1,$lo1,$lo0
 	umulh	$nhi,$nj,$m1
-	str	$lo1,[$tp,#-16]		// tp[j-1]
+	stur	$lo1,[$tp,#-16]		// tp[j-1]
 	cbnz	$j,.Linner
 
 .Linner_skip:
@@ -253,13 +253,13 @@ bn_mul_mont:
 	csel	$nj,$tj,$aj,lo		// did it borrow?
 	ldr	$tj,[$tp],#8
 	ldr	$aj,[$rp],#8
-	str	xzr,[$tp,#-16]		// wipe tp
-	str	$nj,[$rp,#-16]
+	stur	xzr,[$tp,#-16]		// wipe tp
+	stur	$nj,[$rp,#-16]
 	cbnz	$num,.Lcond_copy
 
 	csel	$nj,$tj,$aj,lo
-	str	xzr,[$tp,#-8]		// wipe tp
-	str	$nj,[$rp,#-8]
+	stur	xzr,[$tp,#-8]		// wipe tp
+	stur	$nj,[$rp,#-8]
 
 	ldp	x19,x20,[x29,#16]
 	mov	sp,x29
@@ -596,7 +596,7 @@ __bn_sqr8x_mont:
 	ldp	$a4,$a5,[$tp,#8*4]
 	ldp	$a6,$a7,[$tp,#8*6]
 	adds	$acc0,$acc0,$a0
-	ldr	$n0,[$rp,#-8*8]
+	ldur	$n0,[$rp,#-8*8]
 	adcs	$acc1,$acc1,$a1
 	ldp	$a0,$a1,[$ap,#8*0]
 	adcs	$acc2,$acc2,$a2
@@ -794,7 +794,7 @@ $code.=<<___;
 	//adc	$carry,xzr,xzr		// moved below
 	cbz	$cnt,.Lsqr8x8_post_condition
 
-	ldr	$n0,[$tp,#-8*8]
+	ldur	$n0,[$tp,#-8*8]
 	ldp	$a0,$a1,[$np,#8*0]
 	ldp	$a2,$a3,[$np,#8*2]
 	ldp	$a4,$a5,[$np,#8*4]
@@ -852,7 +852,7 @@ $code.=<<___;
 	ldp	$a6,$a7,[$tp,#8*6]
 	cbz	$cnt,.Lsqr8x_tail_break
 
-	ldr	$n0,[$rp,#-8*8]
+	ldur	$n0,[$rp,#-8*8]
 	adds	$acc0,$acc0,$a0
 	adcs	$acc1,$acc1,$a1
 	ldp	$a0,$a1,[$np,#8*0]
-- 
cgit v1.2.3