From: Niels Möller <nisse@lysator.liu.se>
Date: Thu, 16 May 2013 08:32:45 +0000 (+0200)
Subject: arm/v6: Alignment tweaks to aes code.
X-Git-Tag: nettle_3.0_release_20140607~237
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=8b1d9263f5f26bf30c1678e9baa533365453f01d;p=thirdparty%2Fnettle.git

arm/v6: Alignment tweaks to aes code.
---

diff --git a/ChangeLog b/ChangeLog
index 6bcf524a..38c26f6d 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,10 @@
+2013-05-16  Niels Möller  <nisse@lysator.liu.se>
+
+	* arm/v6/aes-encrypt-internal.asm: Use ALIGN macro. Use 16-byte
+	alignment for loops.
+	* arm/v6/aes-decrypt-internal.asm: Likewise. Also added a nop
+	which mysteriously improves benchmark performance on Cortex-A9.
+
 2013-05-15  Niels Möller  <nisse@lysator.liu.se>
 
 	* configure.ac (asm_path): Handle armv6 and armv7 differently from
diff --git a/arm/v6/aes-decrypt-internal.asm b/arm/v6/aes-decrypt-internal.asm
index 22df3aa8..651636cc 100644
--- a/arm/v6/aes-decrypt-internal.asm
+++ b/arm/v6/aes-decrypt-internal.asm
@@ -50,13 +50,16 @@ define(<X3>, <r14>)	C lr
 	C	       size_t length, uint8_t *dst,
 	C	       uint8_t *src)
 	.text
-	.align 2
+	ALIGN(4)
 PROLOGUE(_nettle_aes_decrypt)
 	teq	LENGTH, #0
 	beq	.Lend
 	ldr	SRC, [sp]
 
 	push	{r4,r5,r6,r7,r8,r10,r11,lr}
+	nop	C For some mysterious reason, taking out this nop
+		C slows this function down on Cortex-A9.
+	ALIGN(16)
 .Lblock_loop:
 	mov	KEY, CTX
 	AES_LOAD(SRC,KEY,W0)
@@ -69,7 +72,7 @@ PROLOGUE(_nettle_aes_decrypt)
 	add	TABLE, TABLE, #AES_TABLE0
 
 	b	.Lentry
-	.align 2
+	ALIGN(16)
 .Lround_loop:
 	C	Transform X -> W
 	AES_DECRYPT_ROUND(X0, X1, X2, X3, W0, W1, W2, W3, KEY)
diff --git a/arm/v6/aes-encrypt-internal.asm b/arm/v6/aes-encrypt-internal.asm
index 81eb6d04..15cf1bb0 100644
--- a/arm/v6/aes-encrypt-internal.asm
+++ b/arm/v6/aes-encrypt-internal.asm
@@ -52,13 +52,14 @@ define(<X3>, <r14>)	C lr
 	C	       size_t length, uint8_t *dst,
 	C	       uint8_t *src)
 	.text
-	.align 2
+	ALIGN(4)
 PROLOGUE(_nettle_aes_encrypt)
 	teq	LENGTH, #0
 	beq	.Lend
 	ldr	SRC, [sp]
 
 	push	{r4,r5,r6,r7,r8,r10,r11,lr}
+	ALIGN(16)
 .Lblock_loop:
 	mov	KEY, CTX
 	AES_LOAD(SRC,KEY,W0)
@@ -71,7 +72,7 @@ PROLOGUE(_nettle_aes_encrypt)
 	add	TABLE, TABLE, #AES_TABLE0
 
 	b	.Lentry
-	.align 2
+	ALIGN(16)
 .Lround_loop:
 	C	Transform X -> W
 	AES_ENCRYPT_ROUND(X0, X1, X2, X3, W0, W1, W2, W3, KEY)