[PATCH] D70550: [LegalizeTypes] Mutate strict float operations to non-strict when softening

[John Brawn via Phabricator via llvm-commits]

john.brawn added a comment.

In D70550#1755470 <https://reviews.llvm.org/D70550#1755470>, @craig.topper wrote:

> I was going to start working on fixing this correctly next week. Can you wait for that?


Sure. There's also a problem in SelectionDAGLegalize::ConvertNodeToLibcall where not all strict opcodes are handled (you can see errors caused by this if you run the CodeGen/ARM/fp-intrinsics.ll test with -mattr=fp-armv8sp, as that means we have to convert double-precision operations to libcalls), are you going to be looking at that as well or should I take a look at that? I've had a brief look so far and doing the obvious thing (add e.g. case ISD::STRICT_FADD to where we currently have ISD::FADD) causes failures in the X86 tests, so it looks like it'll need to take some work to fix.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70550/new/

https://reviews.llvm.org/D70550