projects / platform / upstream / libvpx.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 4924934) | patch
Removing rounding from UV MV calculation
authorJohn Koleszar <jkoleszar@google.com>
Mon, 15 Apr 2013 20:18:24 +0000 (13:18 -0700)
committerJohn Koleszar <jkoleszar@google.com>
Fri, 19 Apr 2013 00:47:17 +0000 (17:47 -0700)
commit2987fa1dc130256562e661bafb0136ac9b704abf
tree310eae4de869a0ec216b66d1e70ab82e387b217ctree | snapshot
parent4924934d2bf860a13ebd8b3f28fe78ea7d9ccb25commit | diff
Removing rounding from UV MV calculation

Consider the previous behavior for the MV 1 3/8 (11/8 pel). In the
existing code, the fractional part of the MV is considered separately,
and rounded is applied, giving a result of 6/8. Rounding is not required
in this case, as we're increasing the precision from a q3 to a q4, and
the correct value 11/16 can be represented exactly.

Slight gain observed (+.033 average on derf)

Change-Id: I320e160e8b12f1dd66aa0ce7966b5088870fe9f8
vp9/common/vp9_reconinter.c diff | blob | history
vp9/common/vp9_reconinter.h diff | blob | history
vp9/encoder/vp9_temporal_filter.c diff | blob | history
Domain: Multimedia / Codec;