[Mlir-commits] [mlir] [mlir][vector] Add pattern to break down small reductions into arith ops (PR #75727)

[Jakub Kuderski]

================
@@ -6212,6 +6214,7 @@ bool WarpExecuteOnLane0Op::areTypesCompatible(Type lhs, Type rhs) {
 
 Value mlir::vector::makeArithReduction(OpBuilder &b, Location loc,
                                        CombiningKind kind, Value v1, Value acc,
+                                       arith::FastMathFlagsAttr fastmath,
----------------
kuhar wrote:

I need this to propagate fastmath attributes. I figured this is a relatively small change, but I can split it into a separate change if this makes code review difficult -- let me know.

https://github.com/llvm/llvm-project/pull/75727