[Mlir-commits] [mlir] [MLIR][XeGPU][VectorToXeGPU] Lower vector.load/store/transfer_read/transfer_write to new offsets syntax (PR #162095)

[Dmitry Chigarev]

================
@@ -116,38 +115,19 @@ createNdDescriptor(PatternRewriter &rewriter, Location loc,
     for (unsigned i = 0; i < srcRank; ++i)
       sourceDims.push_back(memref::DimOp::create(rewriter, loc, src, i));
 
-    SmallVector<int64_t> constOffsets;
-    SmallVector<Value> dynOffsets;
-    for (Value offset : offsets) {
-      std::optional<int64_t> staticVal = getConstantIntValue(offset);
-      if (!staticVal)
-        dynOffsets.push_back(offset);
-      constOffsets.push_back(staticVal.value_or(ShapedType::kDynamic));
-    }
-
-    SmallVector<Value> dynShapes;
+    SmallVector<OpFoldResult> mixedShapes;
----------------
dchigarev wrote:

you're right!

for some reason I thought that the op traces back to the original memref and returns its metadata, but it doesn't work that way. changed the logic to use shapes from the extracted metadata

https://github.com/llvm/llvm-project/pull/162095