]> granicus.if.org Git - libvpx/commit
Speed up h_predictor_16x16
authorJian Zhou <zhoujian@google.com>
Thu, 26 Nov 2015 01:12:20 +0000 (17:12 -0800)
committerJian Zhou <zhoujian@google.com>
Fri, 4 Dec 2015 20:12:55 +0000 (12:12 -0800)
commite86c7c863e465a2f532cc281ea73392b72ba5c78
treecc1feb1cd33311e67a874b2b9c2b3a1639180d55
parentda3f08fac3f35a4d0a6f2d170ba5a27e9719eb73
Speed up h_predictor_16x16

Relocate the function from SSSE3 to SSE2, Unroll loop from 8 to 4,
and reduce mem access to left.
Speed up by >20% in ./test_intra_pred_speed.

Change-Id: Ie48229c2e32404706b722442942c84983bda74cc
test/test_intra_pred_speed.cc
vpx_dsp/vpx_dsp_rtcd_defs.pl
vpx_dsp/x86/intrapred_sse2.asm
vpx_dsp/x86/intrapred_ssse3.asm