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

[Han-Chung Wang]

================
@@ -4471,12 +4488,17 @@ static LogicalResult commonVerifierPackAndUnPackOp(OpTy packOrUnPack) {
   // Verify result shape is greater than the minimum expected
   // by the pack operation, and that the output shape
   // represents full tiles.
-  RankedTensorType expectedPackedType = PackOp::inferPackedType(
-      unpackedType, packOrUnPack.getStaticTiles(), innerDimsPos, outerDimPerm);
-  if (!areAllInBound(expectedPackedType.getShape(), packedType.getShape())) {
-    return op->emitError("the shape of output is not large enough to hold the "
-                         "packed data. Expected at least ")
-           << expectedPackedType << ", got " << packedType;
+  if (hasTensorSemantics) {
+    RankedTensorType expectedPackedType = PackOp::inferPackedTensorType(
+        cast<RankedTensorType>(unpackedType), packOrUnPack.getStaticTiles(),
+        innerDimsPos, outerDimPerm);
+    if (!areAllInBound(expectedPackedType.getShape(), packedType.getShape())) {
+      return op->emitError(
+                 "the shape of output is not large enough to hold the "
+                 "packed data. Expected at least ")
+             << expectedPackedType << ", got " << packedType;
+    }
+  } else {
   }
----------------
hanhanW wrote:

We should check if the shape from `inferPackedMemRefType` is compatible with the shape from `dest`. This makes me think that we should probably need `inferPackedShape` method instead. How about we keep the `inferPackedType` as what it is and we introduce the `inferPackedShape`. Then the new method can be used for both tensor types and memref types verification.

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