efriedma added a comment. The problem with a change like that is that it's not clear what the underlying semantic model is. If we add a flag that says "try to generate code that matches Linux kernel developers' mental model of the underlying machine", or "loop unrolling should try to preserve undefined behavior", or "out-of-bounds memory accesses are well-defined", the semantics are unclear. The point of having well-defined IR is that a developer can tell whether a transform is legal or not based on reading LangRef, as opposed to trying to blindly guess whether a transform is going to break some user's code.
Making clang generate different IR for specific load/division/etc. operators is much easier to specify; IR has the same meaning it always has, we're just generating IR that's well-defined in more cases. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D146466/new/ https://reviews.llvm.org/D146466 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
