[PATCH] D149812: [X86] Avoid usage constant -1 for fminimum/fmaximum lowering

[Serguei Katkov via Phabricator via llvm-commits]

skatkov created this revision.
skatkov added reviewers: RKSimon, pengfei, goldstein.w.n, e-kud.
Herald added a subscriber: hiraditya.
Herald added a project: All.
skatkov requested review of this revision.
Herald added a project: LLVM.

Instead of equality comparison of value to preferred zero we can check just
the sign of value and if sign is set we should put this value as second operand for minimum
and first operand for maximum.
In this case FMIN/FMAX will choose the right result for 0.f and -0.f comparison.

This allows us:

1. avoid loading of big 64-bit constant for fminimum.
2. for double on non-64-nit platform we need to check only high part of value.
3. test against zero to check sign takes less size of instruction

Additionally, if we know that any of value is guaranteed to be non-zero
we should not care about 0.f and -0.f comparison.


https://reviews.llvm.org/D149812

Files:
  llvm/lib/Target/X86/X86ISelLowering.cpp
  llvm/test/CodeGen/X86/avx512fp16-fminimum-fmaximum.ll
  llvm/test/CodeGen/X86/extractelement-fp.ll
  llvm/test/CodeGen/X86/fminimum-fmaximum.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D149812.519346.patch
Type: text/x-patch
Size: 27193 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230504/45e64cd2/attachment.bin>