[Mlir-commits] [mlir] [mlir][spirv] Fix crash in ArithToSPIRV trunc-i lowering on tensor types (PR #179009)

[Jakub Kuderski]

================
@@ -199,3 +199,19 @@ func.func @type_conversion_failure(%arg0: i32) {
 }
 
 } // end module
+
+// -----
+
+module attributes {
+  spirv.target_env = #spirv.target_env<
+    #spirv.vce<v1.0, [], []>, #spirv.resource_limits<>>
+} {
+
+// Tensor truncation is not supported; ensure conversion fails gracefully (no crash).
+func.func @unsupported_tensor_trunci(%arg0: tensor<1xi32>) -> tensor<1xi16> {
+  // expected-error at +1 {{failed to legalize operation 'arith.trunci'}}
+  %t = arith.trunci %arg0 : tensor<1xi32> to tensor<1xi16>
+  return %t : tensor<1xi16>
+}
----------------
kuhar wrote:

We need testcases for other ops too

https://github.com/llvm/llvm-project/pull/179009