Cherry-picked from D71767 with added tests.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/lib/Target/AArch64/AArch64ISelLowering.cpp | ||
---|---|---|
3680 | I doubt you need the usesSVEFor... test or llvm_unreachable part. I went a bit overboard for my original patch. LowerToPredicatedOp has enough asserts to raise an alarm if something weird starts happening. | |
llvm/test/CodeGen/AArch64/sve-fixed-length-fp-minmax.ll | ||
559 | I'm happy with the tests, but just wanted to raise awareness of the "other" style of writing tests as shown by D85724. Without going silly with covering all possible cases there's two options. Stronger type legalisation tests as used by this patch and stronger "end result" testing as used by D85724. Both are equally valid, it just comes down to whether you think type legalisation is already pretty much covered by other instruction tests and possible also which allows for better protection. As I say, I'm not expecting any changes here, I'm just presenting the options. |
llvm/test/CodeGen/AArch64/sve-fixed-length-fp-minmax.ll | ||
---|---|---|
559 | Ah, ok. Makes sense. I had originally based the tests off of FADD, so that's where I got the legalization style from. Updated to the end result style now. Hopefully I didn't make any mistakes. Updating all these fixed VLs is mind numbing... |
I doubt you need the usesSVEFor... test or llvm_unreachable part. I went a bit overboard for my original patch. LowerToPredicatedOp has enough asserts to raise an alarm if something weird starts happening.