[vlc-commits] deinterlace: handle multiple of 16 and 32 widths correctly in arm64 NEON merge asm

[Janne Grunau]

vlc | branch: master | Janne Grunau <janne-vlc at jannau.net> | Thu Oct 13 01:03:19 2016 +0200| [e6241c404f0e72c4954d769361e069172a440111] | committer: Jean-Baptiste Kempf

deinterlace: handle multiple of 16 and 32 widths correctly in arm64 NEON merge asm

The tests for 32 and 16 pixels remainder were inverted resulting in
uninitialized data (YUV green) of 16,32 or 48 pixels at the right
picture edge or equally large overreads/overwrites in the last line.

Closes #17486

Signed-off-by: Jean-Baptiste Kempf <jb at videolan.org>

> http://git.videolan.org/gitweb.cgi/vlc.git/?a=commit;h=e6241c404f0e72c4954d769361e069172a440111
---

 modules/video_filter/deinterlace/merge_arm64.S | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/modules/video_filter/deinterlace/merge_arm64.S b/modules/video_filter/deinterlace/merge_arm64.S
index 827d6f5..a893cc2 100644
--- a/modules/video_filter/deinterlace/merge_arm64.S
+++ b/modules/video_filter/deinterlace/merge_arm64.S
@@ -50,14 +50,14 @@ merge8_arm64_neon:
 	st1		{v2.16b,v3.16b}, [DEST], #32
 	b.gt		1b
 2:
-	tbz		SIZE, #32, 3f
+	tbnz		SIZE, #32, 3f
 	ld1		{v0.16b,v1.16b}, [SRC1], #32
 	ld1		{v4.16b,v5.16b}, [SRC2], #32
 	uhadd		v0.16b, v0.16b, v4.16b
 	uhadd		v1.16b, v1.16b, v5.16b
 	st1		{v0.16b,v1.16b}, [DEST], #32
 3:
-	tbz		SIZE, #16, 4f
+	tbnz		SIZE, #16, 4f
 	ld1		{v0.16b},  [SRC1]
 	ld1		{v4.16b},  [SRC2]
 	uhadd		v0.16b, v0.16b, v4.16b
@@ -85,14 +85,14 @@ merge16_arm64_neon:
 	subs		x5,  x5,  #64
 	b.gt		1b
 2:
-	tbz		SIZE, #32, 3f
+	tbnz		SIZE, #32, 3f
 	ld1		{v0.8h,v1.8h}, [SRC1], #32
 	ld1		{v4.8h,v5.8h}, [SRC2], #32
 	uhadd		v0.8h,  v0.8h,  v4.8h
 	uhadd		v1.8h,  v1.8h,  v5.8h
 	st1		{v0.8h,v1.8h}, [DEST], #32
 3:
-	tbz		SIZE, #16, 4f
+	tbnz		SIZE, #16, 4f
 	ld1		{v0.8h},  [SRC1]
 	ld1		{v4.8h},  [SRC2]
 	uhadd		v0.8h,  v0.8h,  v4.8h