From: Niels Möller <nisse@lysator.liu.se>
Date: Wed, 6 Feb 2013 12:13:29 +0000 (+0100)
Subject: ARM: Optmized aligned case of memxor.
X-Git-Tag: nettle_2.7_release_20130424~117
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=537f64ba9926a7743c8a2353e0fb533718f90910;p=thirdparty%2Fnettle.git

ARM: Optmized aligned case of memxor.
---

diff --git a/ChangeLog b/ChangeLog
index e05dedbf..0d881de6 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,8 @@
 2013-02-06  Niels Möller  <nisse@lysator.liu.se>
 
+	* armv7/memxor.asm (memxor): Optimized aligned case, now runs at
+	0.75 cycles/byte.
+
 	* armv7/README: New file.
 	* armv7/machine.m4: New (empty) file.
 	* armv7/memxor.asm: Initial assembly implementation.
diff --git a/armv7/memxor.asm b/armv7/memxor.asm
index 78762d03..80e725e7 100644
--- a/armv7/memxor.asm
+++ b/armv7/memxor.asm
@@ -131,14 +131,36 @@ PROLOGUE(memxor)
 	b	.Lmemxor_bytes
 
 .Lmemxor_same:
+	tst	N, #4
+	it	ne
+	subne	N, #4
+	bne	.Lmemxor_same_loop
+
 	ldr	r3, [SRC], #+4
 	ldr	r4, [DST]
 	eor	r3, r4
 	str	r3, [DST], #+4
-
-	subs	N, #4
-	bcs	.Lmemxor_same
-	adds	N, #4
+	
+	subs	N, #8
+	bcc	.Lmemxor_same_end
+
+.Lmemxor_same_loop:
+	C 6 cycles per iteration, 0.75 cycles/byte
+	ldr	r4, [SRC, #+4]
+	ldr	r3, [SRC], #+8
+	ldr	r6, [DST, #+4]
+	ldr	r5, [DST]
+	
+	eor	r4, r6
+	eor	r3, r5
+	subs	N, #8
+	
+	str	r4, [DST, #+4]
+	str	r3, [DST], #+8
+	bcs	.Lmemxor_same_loop
+	
+.Lmemxor_same_end:
+	adds	N, #8
 	beq	.Lmemxor_done
 	b	.Lmemxor_bytes
 EPILOGUE(memxor)