X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/blobdiff_plain/0f23f75ff53acadf80e9d3dfd2dfd14cb526074f..bc9ac7ebf31d42ea0965ceded383714207746c27:/symm/chacha-x86ish-sse2.S

diff --git a/symm/chacha-x86ish-sse2.S b/symm/chacha-x86ish-sse2.S
index f36bf90f..641ad653 100644
--- a/symm/chacha-x86ish-sse2.S
+++ b/symm/chacha-x86ish-sse2.S
@@ -42,7 +42,7 @@
 /// Main code.
 
 	.arch pentium4
-	.section .text
+	.text
 
 FUNC(chacha_core_x86ish_sse2)
 
@@ -131,7 +131,7 @@ FUNC(chacha_core_x86ish_sse2)
 	movdqa	SAVE2, xmm2
 	movdqa	SAVE3, xmm3
 
-loop:
+0:
 	// Apply a column quarterround to each of the columns simultaneously.
 	// Alas, there doesn't seem to be a packed doubleword rotate, so we
 	// have to synthesize it.
@@ -228,7 +228,7 @@ loop:
 
 	// Decrement the loop counter and see if we should go round again.
 	sub	NR, 2
-	ja	loop
+	ja	0b
 
 	// Almost there.  Firstly, the feedforward addition.
 	paddd	xmm0, SAVE0