[Mlir-commits] [mlir] [mlir][gpu] Add `gpu.subgroup_uniform` op (PR #157743)

[Jianhui Li]

================
@@ -3255,4 +3255,37 @@ def GPU_SubgroupBroadcastOp : GPU_Op<"subgroup_broadcast",
   let hasVerifier = 1;
 }
 
+def GPU_SubgroupUniformOp : GPU_Op<"subgroup_uniform",
+    [Pure, AllTypesMatch<["result", "src"]>,
+    DeclareOpInterfaceMethods<InferIntRangeInterface, ["inferResultRanges"]>] #
+    ElementwiseMappable.traits>,
+  Arguments<(ins AnyType:$src)> {
+  let summary = "Assumes value is unform across the lanes in subgroup";
+  let description = [{
+    The "subgroup_uniform" op assumes that the value is uniform across all lanes
+    in a subgroup. This means that all active lanes in the subgroup are expected
+    to have the same value.
----------------
Jianhui-Li wrote:

Can you clarify the uniformity is for input or output? it is for all lanes, or for all active lanes. 
My understanding was it is for input and for all lanes, regardless of inactive or not active. So that we can lowered it to scalar register for all lanes, and we can always speculate it - as the speculation means the computation yields the same result even if it is executed in a broad range of threads including those threads which may not use the output (as it becomes inactive at the use point). 
But the second statement wrote as if it for output, only active lanes getting the same value. It invites questions like Fabian raised in https://github.com/llvm/llvm-project/pull/157743/files#r2337838103

https://github.com/llvm/llvm-project/pull/157743