ARM: NEON: fix overflow in h264 16x16 planar pred

Signed-off-by: Mans Rullgard <mans@mansr.com>

diff --git a/libavcodec/arm/h264pred_neon.S b/libavcodec/arm/h264pred_neon.S
index e2c69e1..63c96ee 100644 (file)
--- a/libavcodec/arm/h264pred_neon.S
+++ b/libavcodec/arm/h264pred_neon.S
@@ -131,8 +131,8 @@ function ff_pred16x16_plane_neon, export=1
         vadd.i16        d5,  d6,  d7
         vpadd.i16       d4,  d4,  d5
         vpadd.i16       d4,  d4,  d4
-        vshl.i16        d5,  d4,  #2
-        vaddl.s16       q2,  d4,  d5
+        vshll.s16       q3,  d4,  #2
+        vaddw.s16       q2,  q3,  d4
         vrshrn.s32      d4,  q2,  #6
         mov             r3,  #0
         vtrn.16         d4,  d5