diff options
| author | Ronald S. Bultje <rsbultje@gmail.com> | 2016-10-14 13:01:27 -0400 |
|---|---|---|
| committer | Ronald S. Bultje <rsbultje@gmail.com> | 2016-10-24 16:02:39 -0400 |
| commit | be885da3427c5d9a6fa68229d16318afffe67193 (patch) | |
| tree | a3203f05788df88cf61f01e5008c43a9ebfffb78 | |
| parent | 490c6bda0e35498a24936fd1524317aeebed026b (diff) | |
| download | ffmpeg-be885da3427c5d9a6fa68229d16318afffe67193.tar.gz | |
vp9: change order of operations in adapt_prob().
This is intended to workaround bug "665 Integer Divide Instruction May
Cause Unpredictable Behavior" on some early AMD CPUs, which causes a
div-by-zero in this codepath, such as reported in Mozilla bug #1293996.
Note that this isn't guaranteed to fix the bug, since a compiler is free
to reorder instructions that don't depend on each other. However, it
appears to fix the bug in Firefox, and a similar patch was applied to
libvpx also (see Chrome bug #599899).
| -rw-r--r-- | libavcodec/vp9.c | 5 |
1 files changed, 2 insertions, 3 deletions
diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 779f2d50f3..1aab6ba03c 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -3705,11 +3705,10 @@ static av_always_inline void adapt_prob(uint8_t *p, unsigned ct0, unsigned ct1, if (!ct) return; + update_factor = FASTDIV(update_factor * FFMIN(ct, max_count), max_count); p1 = *p; - p2 = ((ct0 << 8) + (ct >> 1)) / ct; + p2 = ((((int64_t) ct0) << 8) + (ct >> 1)) / ct; p2 = av_clip(p2, 1, 255); - ct = FFMIN(ct, max_count); - update_factor = FASTDIV(update_factor * ct, max_count); // (p1 * (256 - update_factor) + p2 * update_factor + 128) >> 8 *p = p1 + (((p2 - p1) * update_factor + 128) >> 8); |