[Mlir-commits] [mlir] [MLIR][XeGPU] Matrix load/store subgroup distribution (PR #165008)

[Artem Kroviakov]

================
@@ -371,34 +383,43 @@ SliceAttr SliceAttr::flatten() const {
 }
 
 FailureOr<SmallVector<Value>>
-SliceAttr::delinearizeSubgroupId(OpBuilder &builder, Location loc,
-                                 Value linearId) {
+SliceAttr::delinearizeId(OpBuilder &builder, Location loc, Value linearId) {
   SliceAttr attr = flatten();
   auto parent = dyn_cast<LayoutAttr>(attr.getParent());
-  return parent.delinearizeSubgroupId(builder, loc, linearId);
+  return parent.delinearizeId(builder, loc, linearId);
 }
 
-/// Implements DistributeLayoutAttr::getOffsets to generate
-/// instructions for computing multi-dimensional offsets when distributed by
-/// SliceAttr.
+// Implements DistributeLayoutAttr::computeDistributedCoords to generate
+// instructions for computing multi-dimensional offsets when distributed by
+// LayoutAttr.
 FailureOr<SmallVector<SmallVector<Value>>>
-SliceAttr::getOffsets(OpBuilder &builder, Location loc, Value linearId,
-                      ArrayRef<int64_t> shape) {
+SliceAttr::computeDistributedCoords(OpBuilder &builder, Location loc,
----------------
akroviakov wrote:

I think the same can be said for other methods as well, so it may be better left for a separate layout attribute PR.

https://github.com/llvm/llvm-project/pull/165008