X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/blobdiff_plain/898f32b33bea7f1e26c8ba906facdd5038edcd57..43ea7558955ca453d4687c62d12efcb585124e8a:/symm/salsa20-arm-neon.S?ds=sidebyside

diff --git a/symm/salsa20-arm-neon.S b/symm/salsa20-arm-neon.S
index 9cb40472..9d553d3a 100644
--- a/symm/salsa20-arm-neon.S
+++ b/symm/salsa20-arm-neon.S
@@ -79,7 +79,7 @@ FUNC(salsa20_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)}
 
 	// Apply a column quarterround to each of the columns simultaneously,
 	// moving the results to their working registers.  Alas, there
@@ -234,7 +234,7 @@ FUNC(salsa20_core_arm_neon)
 	vext.32	q3, q9, q9, #1			// 12, 13, 14, 15
 
 	// And with that, we're done.
-	vstmia	r2, {d0-d7}
+	vstmia	r2, {QQ(q0, q3)}
 	bx	r14
 
 ENDFUNC