[Mlir-commits] [mlir] [mlir][linalg] Avoid asserts in IndexingMapOpInterface (PR #179072)

[Stefan Weigl-Bosker]

================
@@ -217,6 +217,25 @@ func.func @generic_indexing_map_with_symbol(%arg0: tensor<8xf32>) -> tensor<8xf3
   return %0 : tensor<8xf32>
 }
 
+// -----
+
+// Unranked tensor inputs must be diagnosed.
+func.func @generic_unranked_input_tensor(%in: tensor<*xf32>) {
+  %out = tensor.empty() : tensor<16x16xf32>
+  // expected-error @+1 {{expected operand #0 rank (0) to match the result rank of indexing_map (2)}}
----------------
sweiglbosker wrote:

I feel like this diagnostic isn't the greatest (my bad, I think that is my work lol), i.e unranked tensors are not rank 0, that amx tile thing in the og issue is not rank 0 either. Not sure what can be done about this though

https://github.com/llvm/llvm-project/pull/179072