[Mlir-commits] [mlir] [mlir][gpu] Add gpu.rotate operation (PR #142796)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Wed Jun 18 08:41:31 PDT 2025
================
@@ -1364,6 +1364,53 @@ def GPU_ShuffleOp : GPU_Op<
];
}
+def GPU_RotateOp : GPU_Op<
+ "rotate", [Pure, AllTypesMatch<["value", "rotateResult"]>]>,
+ Arguments<(ins AnyIntegerOrFloatOr1DVector:$value, I32:$offset, I32:$width)>,
+ Results<(outs AnyIntegerOrFloatOr1DVector:$rotateResult, I1:$valid)> {
+ let summary = "Rotate values within a subgroup.";
+ let description = [{
+ The "rotate" op moves values across lanes (a.k.a., invocations, work items)
+ within the same subgroup. The `width` argument specifies the number of lanes
+ that participate in the rotation, and must be uniform across all lanes.
----------------
Muzammiluddin-Syed-ECE wrote:
I'm not sure I understand what this means. This implies that if for example we had a width of 16, then lanes 16+ won't be participating in the rotation. But if so, what does it mean for this to be uniform across all lanes?
Is this supposed to be `uniform across all participating lanes`?
https://github.com/llvm/llvm-project/pull/142796
More information about the Mlir-commits
mailing list