OpenCL backend is way too slow compared to futhark bench

diku-dk / bfast

GPU Implementation for BFAST

GNU General Public License v3.0

37 stars 17 forks source link

OpenCL backend is way too slow compared to futhark bench #34

Closed mortvest closed 3 years ago

mortvest commented 3 years ago

There are problems with tuning thresholds in base.py which make it perform much slower (up to 5x speedup is possible with proper tuning). Need to look into retuning the code with the newest version of the Futhark compiler

mortvest commented 3 years ago

fixed in #38