While GPU compiler phase optimization has been shown to improve performance significantly (e.g. https://link.springer.com/chapter/10.1007/978-3-319-75178-8_35), there has been very little follow-up research on this topic. Does anyone have any idea why that is the case?

  • 2
    $\begingroup$ Would shading work here? I feel like we need a more general tag for targeting GPUs $\endgroup$ Jul 12, 2023 at 18:14
  • $\begingroup$ A while ago ARM experimented with using a genetic algorithm to select LLVM passes sequence for OpenCL on Mali GPU. It managed to be no better (and no worse) than a manually selected sequence of passes. $\endgroup$
    – SK-logic
    Jul 13, 2023 at 8:19
  • $\begingroup$ I would go for either "GPU" as a specific tag or heterogenous computing as the more general term and tag. $\endgroup$ Jul 15, 2023 at 0:41
  • $\begingroup$ This question should be clarified to say whether you mean optimising for the performance of the compiler or the performance of the compiled code; I also presume that the GPU is the target architecture for the compiled code, not the architecture on which the compiler should run, but this could be spelled out. I'm sure this information is included in the paper you linked to, but questions should make sense on their own. $\endgroup$
    – kaya3
    Jul 24, 2023 at 13:12


You must log in to answer this question.

Browse other questions tagged .