[Mlir-commits] [mlir] [mlir][linalg] Fix crash in linalg-specialize-generic-ops with scalar inputs (PR #189212)

[Adam Siemieniuk]

================
@@ -69,3 +69,36 @@ func.func @broadcast_only(%x : tensor<2x16x32xf32>, %y:  tensor<2x32xf32>, %z :
 // CHECK: %[[X_bc:.+]] = linalg.broadcast ins(%[[Y]] : tensor<2x32xf32>) outs(%[[E0]] : tensor<2x16x32xf32>) dimensions = [1]
 // CHECK: {{.*}} = linalg.div ins(%[[X]], %[[X_bc]] : tensor<2x16x32xf32>, tensor<2x16x32xf32>) outs(%arg2 : tensor<2x16x32xf32>) -> tensor<2x16x32xf32>
 // CHECK-NOT: linalg.generic
+
+// -----
+
+// Verify that linalg.generic with scalar (non-tensor) inputs is not decomposed
+// and does not crash. Scalar inputs have 0-D affine maps and are not
+// RankedTensorType; the pass must handle them gracefully by bailing out.
+// (GitHub issue #122094)
----------------
adam-smnk wrote:

nit: I think reference to the issue here is not necessary

https://github.com/llvm/llvm-project/pull/189212