vp9/x86: idct_32x32_add_ssse3 sub-16x16-idct.
authorRonald S. Bultje <rsbultje@gmail.com>
Sat, 4 Jan 2014 15:29:45 +0000 (10:29 -0500)
committerRonald S. Bultje <rsbultje@gmail.com>
Wed, 8 Jan 2014 01:43:34 +0000 (20:43 -0500)
commit37b001d14d3426c02f032ff3a3f90e3f02a639cd
tree8e2a351facbec918cb090334452cbdfe6f3cb5f5
parente84d14df10d0408b9e06b33b2f71173188279dda
vp9/x86: idct_32x32_add_ssse3 sub-16x16-idct.

Runtime of all IDCTs together goes from 3327 to 2473 cycles (intra, i.e.
~35% faster) or from 2312 to 1448 cycles (inter, i.e. ~60% faster). Total
decode time of ped1080p.webm goes from 8.086sec to 7.974sec (1.4% faster).
libavcodec/x86/vp9itxfm.asm