[Mlir-commits] [mlir] [MLIR][XeGPU] Allow load/store/prefetch uses [memref+offset] instead of tdesc (PR #150576)
Charitha Saumya
llvmlistbot at llvm.org
Tue Jul 29 10:53:00 PDT 2025
================
@@ -638,25 +638,44 @@ def XeGPU_PrefetchOp : XeGPU_Op<"prefetch", []> {
}];
- let arguments = (ins XeGPU_TensorDesc: $TensorDesc,
+ let arguments = (ins XeGPU_TensorDescOrMemRef: $source,
+ Optional<XeGPU_OffsetType>: $offsets,
OptionalAttr<XeGPU_CacheHintAttr>: $l1_hint,
OptionalAttr<XeGPU_CacheHintAttr>: $l2_hint,
OptionalAttr<XeGPU_CacheHintAttr>: $l3_hint);
let extraClassDeclaration = extraBaseClassDeclaration # [{
+ Type getSourceType() {
+ return getSource().getType();
+ }
+
+ Value getTensorDesc() {
----------------
charithaintc wrote:
nit: this is bit risky. because the caller might expect a tensor_desc but this may return a memref or I64. I would add a TODO note.
https://github.com/llvm/llvm-project/pull/150576
More information about the Mlir-commits
mailing list