[Mlir-commits] [mlir] [MLIR][XeGPU] Matrix load/store subgroup distribution (PR #165008)
Adam Siemieniuk
llvmlistbot at llvm.org
Mon Oct 27 10:40:32 PDT 2025
================
@@ -283,41 +279,53 @@ LayoutAttr::delinearizeSubgroupId(OpBuilder &builder, Location loc,
};
if (!hasDefaultOrder())
return mlir::emitError(loc, "order attribute is currently not supported.");
-
- auto dims =
- llvm::map_to_vector(getEffectiveSgLayoutAsInt(), [&](int64_t d) -> Value {
- return builder.createOrFold<arith::ConstantIndexOp>(loc, d);
- });
+ SmallVector<int64_t> layout;
+ if (idLevel == xegpu::DistributionLevel::SG) {
----------------
adam-smnk wrote:
nit: braces can be skipped
https://github.com/llvm/llvm-project/pull/165008
More information about the Mlir-commits
mailing list