[Mlir-commits] [mlir] [MLIR][NVVM] Add support for f32 in redux.sync Op (PR #128137)

[Durgadoss R]

================
@@ -25,9 +25,17 @@ using namespace mlir;
 using namespace mlir::LLVM;
 using mlir::LLVM::detail::createIntrinsicCall;
 
+#define REDUX_F32_ID_IMPL(op, abs, nan)                                        \
+  hasNaN ? llvm::Intrinsic::nvvm_redux_sync_f##op##abs##nan                    \
+         : llvm::Intrinsic::nvvm_redux_sync_f##op##abs
+
+#define GET_REDUX_F32_ID(op, abs, nan)                                         \
+  hasAbs ? REDUX_F32_ID_IMPL(op, abs, nan) : REDUX_F32_ID_IMPL(op, , nan)
+
----------------
durga4github wrote:

optional:
Since the concat is straight-forward, passing the booleans to the macros directly, makes them self-contained.

https://github.com/llvm/llvm-project/pull/128137