Thread: Tetra decoding
View Single Post
  #1172 (permalink)  
Old 02-29-2016, 5:05 PM
skinkie skinkie is offline
Member
   
Join Date: Jan 2016
Location: Leidschendam
Posts: 23
Default

Yeah the volk stuff is how more real hardcore performance programs should have been written. I am also looking forward to VOLK progress into the GPU domain. If the framework is just selecting which algorithm can do the job best (with zero overhead), it is also much easier to benchmark and interchange higher level algorithms with the same output.

You might also try your luck on CFLAGS="-O2 -march=native" (or even -O3 if you feel lucky). This might give a performance benefit (~10%) as well.
Reply With Quote