Instruction count difference compared to reference

[muellch]

The OEC compiled baseline for many kernels performs more instructions than the dawn reference experiments, even if both variants execute the kernels in sequence without any inlining or fusion.

[gysit]

I close this since I think parts of this is due to the not optimal nvptx support (fma for integer etc)... I think this is a minor issue at the moment and we should focus on features and correctness