[Mlir-commits] [mlir] [mlir][tensor] Apply `InsertSliceOfTransferWriteOpFolder` only when `transfer_write` overwrites all elements of `insert_slice` (PR #108803)

[Benjamin Maxwell]

================
@@ -154,6 +166,25 @@ LogicalResult InsertSliceOfTransferWriteOpFolder::matchAndRewrite(
   return success();
 }
 
+bool InsertSliceOfTransferWriteOpFolder::doesTransferWriteCoverInsertSlice(
+    vector::TransferWriteOp writeOp, tensor::InsertSliceOp insertSliceOp,
+    MLIRContext *context) {
+  auto destType = cast<ShapedType>(writeOp.getOperand(0).getType());
+  auto insertSliceType = insertSliceOp.getSourceType();
+
+  if (destType.hasStaticShape() && insertSliceType.hasStaticShape()) {
+    for (int64_t d = 0, e = insertSliceType.getRank(); d < e; ++d) {
+      if (destType.getDimSize(d) != insertSliceType.getDimSize(d))
+        return false;
+    }
+    return true;
+  }
----------------
MacDue wrote:

I think this can be simplified a little. I think you only need to check the `writeOp` fully overwrites its destination tensor (do you don't need to pass  `insertSliceOp` here). 

https://github.com/llvm/llvm-project/pull/108803