diff options
| author | YunQiang Su <syq@debian.org> | 2024-06-21 11:53:08 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-06-21 11:53:08 +0800 |
| commit | 89881480030f48f83af668175b70a9798edca2fb (patch) | |
| tree | dcc9680d4a5d3bb7093aad3d15b4d368dd752c7a /llvm/lib/CodeGen/ExpandVectorPredication.cpp | |
| parent | f3c4dae6f2c44f1a7f130c4cf4b2861b62402b48 (diff) | |
| download | llvm-89881480030f48f83af668175b70a9798edca2fb.zip llvm-89881480030f48f83af668175b70a9798edca2fb.tar.gz llvm-89881480030f48f83af668175b70a9798edca2fb.tar.bz2 | |
Intrinsic: introduce minimumnum and maximumnum (#93841)
Currently, on different platform, the behaivor of llvm.minnum is
different if one operand is sNaN:
When we compare sNaN vs NUM:
ARM/AArch64/PowerPC: follow the IEEE754-2008's minNUM: return qNaN.
RISC-V/Hexagon follow the IEEE754-2019's minimumNumber: return NUM. X86:
Returns NUM but not same with IEEE754-2019's minimumNumber as
+0.0 is not always greater than -0.0.
MIPS/LoongArch/Generic: return NUM.
LIBCALL: returns qNaN.
So, let's introduce llvm.minmumnum/llvm.maximumnum, which always follow
IEEE754-2019's minimumNumber/maximumNumber.
Half-fix: #93033
Diffstat (limited to 'llvm/lib/CodeGen/ExpandVectorPredication.cpp')
| -rw-r--r-- | llvm/lib/CodeGen/ExpandVectorPredication.cpp | 18 |
1 files changed, 17 insertions, 1 deletions
diff --git a/llvm/lib/CodeGen/ExpandVectorPredication.cpp b/llvm/lib/CodeGen/ExpandVectorPredication.cpp index dc35f33..c8a9f90 100644 --- a/llvm/lib/CodeGen/ExpandVectorPredication.cpp +++ b/llvm/lib/CodeGen/ExpandVectorPredication.cpp @@ -331,7 +331,9 @@ Value *CachingVPExpander::expandPredicationToFPCall( return NewOp; } case Intrinsic::maxnum: - case Intrinsic::minnum: { + case Intrinsic::minnum: + case Intrinsic::maximumnum: + case Intrinsic::minimumnum: { Value *Op0 = VPI.getOperand(0); Value *Op1 = VPI.getOperand(1); Function *Fn = Intrinsic::getDeclaration( @@ -498,6 +500,18 @@ CachingVPExpander::expandPredicationInReduction(IRBuilder<> &Builder, Reduction = Builder.CreateBinaryIntrinsic(Intrinsic::minimum, Reduction, Start); break; + case Intrinsic::vp_reduce_fmaximumnum: + Reduction = Builder.CreateFPMaximumReduce(RedOp); + transferDecorations(*Reduction, VPI); + Reduction = + Builder.CreateBinaryIntrinsic(Intrinsic::maximumnum, Reduction, Start); + break; + case Intrinsic::vp_reduce_fminimumnum: + Reduction = Builder.CreateFPMinimumReduce(RedOp); + transferDecorations(*Reduction, VPI); + Reduction = + Builder.CreateBinaryIntrinsic(Intrinsic::minimumnum, Reduction, Start); + break; case Intrinsic::vp_reduce_fadd: Reduction = Builder.CreateFAddReduce(Start, RedOp); break; @@ -755,6 +769,8 @@ Value *CachingVPExpander::expandPredication(VPIntrinsic &VPI) { case Intrinsic::vp_minnum: case Intrinsic::vp_maximum: case Intrinsic::vp_minimum: + case Intrinsic::vp_maximumnum: + case Intrinsic::vp_minimumnum: case Intrinsic::vp_fma: case Intrinsic::vp_fmuladd: return expandPredicationToFPCall(Builder, VPI, |