[RFC] De-type-ification of LLVM IR: why?

As I wrote, that title is just a teaser. The problem that it is highlighting is that all loop optimisations are off by default. So yes, we have them, but not really. We have been trying to enable one, but haven’t succeeded yet. This discussion here illustrates why. People are running loop optimisations in their downstream compilers, but apparently it’s not good enough for upstream. We also don’t have consensus that Fortran is a better behaved language and that these DependenceAnalysis corner cases are not important.

1 Like