Differential D71968

[X86] Re-enable lowerUINT_TO_FP_vXi32 under fast-math by using an FSUB instead of an FADD.
ClosedPublic
Actions

Authored by craig.topper on Dec 28 2019, 4:55 PM.

Details

Reviewers

RKSimon
spatel
scanon

Commits

rGbd46e2974275: [X86] Re-enable lowerUINT_TO_FP_vXi32 under fast-math by using an FSUB instead…

Summary

We previously disabled this under fast math due to aggressive
reassociation by the machine combiner. But I think we can work
around this by using a FSUB instead of FADD for the first
operation.

This matches the similar algorithm we do for uint_to_fp i64->f64
in TargetLowering::expandUINT_TO_FP. If reassociation hasn't
been a problem for that, hopefully its not a problem here.

Diff Detail

Repository: rG LLVM Github Monorepo

Event Timeline

craig.topper created this revision.Dec 28 2019, 4:55 PM

Herald added a project: Restricted Project. · View Herald TranscriptDec 28 2019, 4:55 PM

Herald added a subscriber: hiraditya. · View Herald Transcript

Harbormaster completed remote builds in B43014: Diff 235494.Dec 28 2019, 4:57 PM

LGTM - we could do what the existing code comment suggests, but it's more work, and I don't think it's worth the effort.

llvm/lib/Target/X86/X86ISelLowering.cpp
19089–19091	We actually do have FMF on MIR now - see D52006 and related patches.
19137–19139	Better leave a code comment here to explain that this is fragile and requires subtracting the positive constant rather than adding the negative constant.