[FFmpeg-devel] [PATCH] avcodec/apedec: fix prediction

[Zhao Zhili]

From: Zhao Zhili <zhilizhao at tencent.com>

Fixes ticket #9816.
Regression since ed0001482a74b60f3d5bc5.

Prediction value larger than INT32_MAX should be treated as
negative. The code already depends on undefined right shift
behavior before the patch, which doesn't get fixed by the patch.

Signed-off-by: Zhao Zhili <zhilizhao at tencent.com>
---
 libavcodec/apedec.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/libavcodec/apedec.c b/libavcodec/apedec.c
index a7c38bce1b..5f2af2e147 100644
--- a/libavcodec/apedec.c
+++ b/libavcodec/apedec.c
@@ -1198,7 +1198,7 @@ static av_always_inline int predictor_update_filter(APEPredictor64 *p,
                   p->buf[delayB - 3] * p->coeffsB[filter][3] +
                   p->buf[delayB - 4] * p->coeffsB[filter][4];
 
-    p->lastA[filter] = decoded + ((int64_t)((uint64_t)predictionA + (predictionB >> 1)) >> 10);
+    p->lastA[filter] = decoded + ((int32_t)((uint64_t)predictionA + (predictionB >> 1)) >> 10);
     p->filterA[filter] = p->lastA[filter] + ((int64_t)(p->filterA[filter] * 31ULL) >> 5);
 
     sign = APESIGN(decoded);
-- 
2.35.3