[Mlir-commits] [mlir] Adding to execute_region_op some missing support (PR #164159)

[Mehdi Amini]

================
@@ -1604,6 +1604,28 @@ func.func @func_execute_region_inline_multi_yield() {
 
 // -----
 
+// CHECK-LABEL:   func.func private @canonicalize_execute_region_yeilding_external_value(
+// CHECK-SAME:                                                                           %[[VAL_0:.*]]: tensor<1x120xui8>) -> tensor<1x60xui8> {
+// CHECK:           %[[VAL_1:.*]] = memref.alloc() {alignment = 64 : i64} : memref<1x60xui8>
+// CHECK:           scf.execute_region no_inline {
+// CHECK:             scf.yield
+// CHECK:           }
+// CHECK:           %[[VAL_2:.*]] = bufferization.to_tensor %[[VAL_1]] : memref<1x60xui8> to tensor<1x60xui8>
+// CHECK:           return %[[VAL_2]] : tensor<1x60xui8>
+// CHECK:         }
+
+func.func private @canonicalize_execute_region_yeilding_external_value(%arg0: tensor<1x120xui8>) -> tensor<1x60xui8> {
----------------
joker-eph wrote:

"canonicalize" does not say much about what we're testing here.
Probably need a comment to describe the property to check on.

Can you also minimize the checks please?

https://github.com/llvm/llvm-project/pull/164159