X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/blobdiff_plain/70bc6059902c30dcbd1cddbdb628e4bcbd9cc6f5..7b0d1a63587f3cb1ae3bb8b248bbb1b82bdca7bd:/symm/chacha-arm-neon.S

diff --git a/symm/chacha-arm-neon.S b/symm/chacha-arm-neon.S
index 4c72791e..a900db76 100644
--- a/symm/chacha-arm-neon.S
+++ b/symm/chacha-arm-neon.S
@@ -25,18 +25,19 @@
 /// MA 02111-1307, USA.
 
 ///--------------------------------------------------------------------------
-/// External definitions.
+/// Preliminaries.
 
 #include "config.h"
 #include "asm-common.h"
 
-///--------------------------------------------------------------------------
-/// Main.code.
-
 	.arch	armv7-a
 	.fpu	neon
+
 	.text
 
+///--------------------------------------------------------------------------
+/// Main.code.
+
 FUNC(chacha_core_arm_neon)
 
 	// Arguments are in registers.
@@ -55,7 +56,7 @@ FUNC(chacha_core_arm_neon)
 	// We need a copy for later.  Rather than waste time copying them by
 	// hand, we'll use the three-address nature of the instruction set.
 	// But this means that the main loop is offset by a bit.
-	vldmia	r1, {d24-d31}
+	vldmia	r1, {QQ(q12, q15)}
 
 	// a += b; d ^= a; d <<<= 16
 	vadd.u32 q8, q12, q13
@@ -173,7 +174,7 @@ FUNC(chacha_core_arm_neon)
 	vadd.u32 q11, q11, q15
 
 	// And now we write out the result.
-	vstmia	r2, {d16-d23}
+	vstmia	r2, {QQ(q8, q11)}
 
 	// And with that, we're done.
 	bx	r14