[Mlir-commits] [clang] [llvm] [mlir] [clang][NVPTX][MLIR][NVVM] Add overloaded fma intrinsics (PR #232672)

[Srinivasa Ravi]

================
@@ -4344,16 +4301,19 @@ static Constant *ConstantFoldScalarCall3(StringRef Name,
       const bool IsFMul = IntrinsicID == Intrinsic::nvvm_fmul ||
                           IntrinsicID == Intrinsic::nvvm_fmul_ftz;
       if (IsFAdd || IsFMul) {
+        APFloat::roundingMode RoundMode =
+            nvvm::GetRoundingModeFromImmArg(Operands[2]);
+        // Do not fold combinations that codegen rejects.
+        if (!nvvm::FPArithIsSupported(IntrinsicID, Ty, RoundMode))
----------------
Wolfram70 wrote:

Makes sense, moved these checks to `VerifierNVVM.cpp` in the latest revision. Thanks!

https://github.com/llvm/llvm-project/pull/232672