3 ms·
Lift: A functional parallel IR for high-performance GPU code generation (2017)
- dwrodri 5y agoI am very curious what the authors of this work think of MLIR! OpenCL is an interesting compilation target, I was under the impression that vendor implementations ended resulting in high variance in performance results when porting OpenCL to different hardware.