rsmith added a comment. In D76096#1921842 <https://reviews.llvm.org/D76096#1921842>, @nickdesaulniers wrote:
> > The performance implications of deleting those lines is the complicated > > part. > > Where does compile time performance suffer from this? I guess if we have > massive array initializers, or large struct definitions, or deeply nested > struct definitions, it might take time to recursively evaluate if all members > are constant expressions; but isn't that what I want as a developer, to > offload the calculations to compile time rather than runtime? Or is the cost > way too significant? It's exactly what you suspect: very large global arrays initialized from constant data. For those, it's substantially more efficient for CodeGen to walk the AST representation (the `InitListExpr`) and directly generate an IR constant than it is to create an `APValue` representation of the array. (`APValue` is not especially space-efficient, and the extra copying and data shuffling can be quite slow.) We saw a significant compile time regression on a couple of LNT benchmarks without these early bailouts for C. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D76096/new/ https://reviews.llvm.org/D76096 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
