[Mlir-commits] [mlir] [MLIR][Linalg] pack, unpack to take memref inputs (PR #129036)

[Han-Chung Wang]

================
@@ -4426,15 +4428,30 @@ static LogicalResult commonVerifierPackAndUnPackOp(OpTy packOrUnPack) {
         tiles, [](OpFoldResult tile) { return isConstantIntValue(tile, 0); });
   };
 
+  // Verify that the source and destination are ranked types.
+  if (!packOrUnPack.getSourceType().hasRank() ||
+      !packOrUnPack.getDestType().hasRank()) {
+    return op->emitError(
+        "expected both source and destination to be shaped types");
+  }
+
   // Verify tiles. Do not allow zero tiles.
   SmallVector<OpFoldResult> mixedTiles = packOrUnPack.getMixedTiles();
   if (hasZeros(mixedTiles))
     return op->emitError("invalid zero tile factor");
 
+  // Verify that the Operation does not have mixed tensor/buffer semantics.
+  if (!packOrUnPack.hasPureBufferSemantics() &&
+      !packOrUnPack.hasPureTensorSemantics()) {
+    return op->emitError("mixing tensor and buffer semantics is not allowed");
----------------
hanhanW wrote:

Can you add a test to invalid.mlir which reflects the check?

https://github.com/llvm/llvm-project/blob/e55164ae1098bbf8ceb87b83a4b282b08bb7bef9/mlir/test/Dialect/Linalg/invalid.mlir#L1489-L1491

https://github.com/llvm/llvm-project/pull/129036