absl::AbortedError("The input `min` must have float type."));
      OP_REQUIRES(context, context->input_type(i * 3 + 1) == DT_FLOAT,
                  absl::AbortedError("The input `max` must have float type."));
      OP_REQUIRES(
          context, context->input_type(i * 3 + 2) == DT_INT64,
          absl::AbortedError("The input `histogram` must have int64 type."));
    }
  }

    $metadata is a serialized TPUCompileMetadataProto describing the shapes and types of the inputs to the computation, as well as a mapping onto the TPU pod topology.
    $constant_operand_indices are the indices of the inputs that are constant to the TPU program (e.g. weights in inference), the rest of the inputs are input tensors.
    constant_operand_indices is sorted in ascending order.

            // TODO when implemented, this assertion should be replaced with the commented one:
            inputs.expect(allOf(startsWith("Plugin class 'SneakyPluginA': file '${FileUtils.testFileName}'"), containsString(FileUtils.testFileName)))
            /*
            pluginClasses.forEach {
                inputs = inputs.expect(allOf(startsWith("Plugin class '$it':"), containsString(FileUtils.testFileName))))
            }*/
        }

      cost += GetRankedTensorSize(type);
    } else {
      // For unranked tensors, use the max size among the input tensors. This is
      // because the only dynamic information of the function should be the
      // input, so the size of dynamic tensors should be usually capped by
      // inputs' sizes.
      cost += max_arg_size_;
    }
  }

  cost_map_[op] = cost;
}

}  // namespace tfrt_compiler

// RUN: tf-opt %s -split-input-file -verify-diagnostics -tf-xla-validate-inputs

// expected-error @+1 {{expects no nested calls of entry functions as they prevent graph traversal in some passes from working correctly}}
func.func @nested_entry_functions() attributes {tf.entry_function = {}} {
  tf_executor.graph {

}
}

//CHECK: module attributes
//CHECK-SAME: keep_stablehlo_constant = "true"
//CHECK-NEXT:  func.func @main(%arg0: tensor<2x1x2xf32>) -> tensor<2x1x2xf32> attributes {tf.entry_function = {inputs = "arg0", outputs = "vhlo.dynamic_update_slice_v1"}} {
//CHECK-DAG:    %[[c0:.+]] = stablehlo.constant dense<2.000000e+00> : tensor<1x1x2xf32>
//CHECK-DAG:    %[[c1:.+]] = stablehlo.constant dense<1> : tensor<i32>

                outputs.cacheIf { true }

                inputs.files(inputFile)
                    .withPropertyName("dummy-input")
                    .withPathSensitivity(PathSensitivity.RELATIVE)

                outputs.dir(outputDir)
                    .withPropertyName('dummy-output')

                doLast {
                    outputFile.text = inputFile.text
                }
            }

    Attribute inputSpec = std::get<1>(input);
    if (!isValidQuantizationSpec(inputSpec, inputType)) {
      return emitOpError() << "has incompatible specification " << inputSpec
                           << " and input type " << inputType;
    }
  }

  for (auto result : llvm::zip(getResultTypes(), getOutputSpecs())) {
    Type outputType = std::get<0>(result);
    Attribute outputSpec = std::get<1>(result);

    Attribute inputSpec = std::get<1>(input);
    if (!isValidQuantizationSpec(inputSpec, inputType)) {
      return emitOpError() << "has incompatible specification " << inputSpec
                           << " and input type " << inputType;
    }
  }

  for (auto result : llvm::zip(getResultTypes(), getOutputSpecs())) {
    Type outputType = std::get<0>(result);
    Attribute outputSpec = std::get<1>(result);

        // Once we migrate to all-scheduled transforms we should consider if we can avoid having this optimization and use only normalized inputs.
        //
        // Note that we are not capturing this value in the actual inputs of the work; doing so would cause unnecessary cache misses.
        // This is why the hash is captured here and not in visitIdentityInputs().