projects / platform / upstream / libvpx.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 0207f17) | patch
quantize fp: neon implementation
authorJohann <johannkoenig@google.com>
Thu, 10 Aug 2017 22:02:22 +0000 (15:02 -0700)
committerJohann <johannkoenig@google.com>
Wed, 23 Aug 2017 15:01:30 +0000 (08:01 -0700)
commite83d99d7b8e009fd847ff91319f57abd5f7f96eb
treea6020dfc9d8673f40182995d3fa71a2dc1133e46tree | snapshot
parent0207f17144f213a52512cfd8287e26946e30d7adcommit | diff
quantize fp: neon implementation

About 4x faster when values are below the dequant threshold and 10x
faster if everything needs to be calculated.

Both numbers would improve if the division for dqcoeff could be
simplified.

BUG=webm:1426

Change-Id: I8da67c1f3fcb4abed8751990c1afe00bc841f4b2
test/vp9_quantize_test.cc diff | blob | history
vp9/common/vp9_rtcd_defs.pl diff | blob | history
vp9/encoder/arm/neon/vp9_quantize_neon.c diff | blob | history
Domain: Multimedia / Codec;