[Mlir-commits] [mlir] [mlir][NVVM] Add support for tcgen05.ld.red Op (PR #177330)

[Durgadoss R]

================
@@ -493,33 +516,11 @@ def NVVM_RcpApproxFtzF32Op : NVVM_IntrOp<"rcp.approx.ftz.f", [Pure], 1> {
 // NVVM redux op definitions
 //===----------------------------------------------------------------------===//
 
-def ReduxKindNone : I32EnumAttrCase<"NONE", 0, "none">;
-def ReduxKindAdd  : I32EnumAttrCase<"ADD", 1, "add">;
-def ReduxKindAnd  : I32EnumAttrCase<"AND", 2, "and">;
-def ReduxKindMax  : I32EnumAttrCase<"MAX", 3, "max">;
-def ReduxKindMin  : I32EnumAttrCase<"MIN", 4, "min">;
-def ReduxKindOr   : I32EnumAttrCase<"OR", 5, "or">;
-def ReduxKindUmax : I32EnumAttrCase<"UMAX", 6, "umax">;
-def ReduxKindUmin : I32EnumAttrCase<"UMIN", 7, "umin">;
-def ReduxKindXor  : I32EnumAttrCase<"XOR", 8, "xor">; 
-def ReduxKindFmin : I32EnumAttrCase<"FMIN", 9, "fmin">;
-def ReduxKindFmax : I32EnumAttrCase<"FMAX", 10, "fmax">;
-
-/// Enum attribute of the different kinds.
-def ReduxKind : I32EnumAttr<"ReduxKind", "NVVM redux kind",
-  [ReduxKindAdd, ReduxKindAnd, ReduxKindMax, ReduxKindMin, ReduxKindOr, 
-   ReduxKindUmax, ReduxKindUmin, ReduxKindXor, ReduxKindFmin, ReduxKindFmax]> {
-  let genSpecializedAttr = 0;
-  let cppNamespace = "::mlir::NVVM";
-}
-
-def ReduxKindAttr : EnumAttr<NVVM_Dialect, ReduxKind, "redux_kind">;
----------------
durga4github wrote:

Nice to see this removal.
We should clean up the TMA reduction kind as well, in a subsequent PR

https://github.com/llvm/llvm-project/pull/177330