[Mlir-commits] [mlir] [mlir][arith] Add rounding mode flags to binary arithmetic operations (PR #188458)

[Matthias Springer]

================
@@ -157,6 +157,11 @@ class AttrConverterConstrainedFPToLLVM {
       convertedAttr.set(TargetOp::getRoundingModeAttrName(),
                         convertArithRoundingModeAttrToLLVM(arithAttr));
     }
+    // Constrained intrinsics do not support fastmath flags. Remove the
+    // arith fastmath attribute if present.
+    if constexpr (SourceOp::template hasTrait<
----------------
matthias-springer wrote:

This is needed because fast_math flags cannot be specified together with rounding modes when lowering to LLVM.

https://github.com/llvm/llvm-project/pull/188458