func.func @strided_slice(%arg0 : tensor<*xf32>) -> tensor<*xf32> {
  %18 = "tf.Const"() {value = dense<1> : tensor<4xi32>} : () -> tensor<4xi32>
  %57 = "tf.Const"() {value = dense<0> : tensor<4xi32>} : () -> tensor<4xi32>
  %534 = "tf.StridedSlice"(%arg0, %57, %57, %18) {begin_mask = 11 : i64, device = "", ellipsis_mask = 0 : i64, end_mask = 11 : i64, new_axis_mask = 4 : i64, shrink_axis_mask = 0 : i64} : (tensor<*xf32>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<*xf32>

        "@llvm-project//mlir:Support",
    ],
)

tf_cc_test(
    name = "error_collector_inst_test",
    srcs = ["error_collector_inst_test.cc"],
    data = [
        "testdata/strided_slice.mlir",
    ],
    deps = [
        ":error_collector",
        ":error_collector_inst",
        ":types_util",
        "//tensorflow/compiler/mlir/tensorflow",
        "//tensorflow/core:test",

  %6 = "tfl.dequantize"(%5) : (tensor<1x34x1x3x!quant.uniform<i8:f32, 1.0>>) -> tensor<1x34x1x3xf32>

}

TEST(ErrorCollectorTest, TessSuccessPass) {
  std::string input_file = tensorflow::GetDataDependencyFilepath(
      "tensorflow/compiler/mlir/lite/metrics/testdata/strided_slice.mlir");
  MLIRContext context;
  context.getOrLoadDialect<mlir::func::FuncDialect>();
  context.getOrLoadDialect<TF::TensorFlowDialect>();
  context.enableMultithreading();

    %cst_3 = arith.constant dense<0> : tensor<1xi32>
    %0 = "tfl.shape"(%arg2) {tac.device = "DARWINN", tac.inference_type = "FLOAT"} : (tensor<?x?xi32>) -> tensor<2xi32>

  %3 = "tfl.strided_slice"(%arg0, %1, %0, %2) {begin_mask = 0 : i32, ellipsis_mask = 0 : i32, end_mask = 0 : i32, new_axis_mask = 0 : i32, shrink_axis_mask = 0 : i32, offset = false} : (tensor<10x10x10xf32>, tensor<3xi32>, tensor<3xi32>, tensor<3xi32>) -> tensor<9x9x9xf32>
  func.return %3 : tensor<9x9x9xf32>
  // CHECK: %[[STRIDED_SLICE:.*]] = "tfl.strided_slice"
  // CHECK:  return %[[STRIDED_SLICE]]
}

// masks will complicate the strided_slice computation logic, we can simplify
// the logic by inserting a reshape op to pad the inputs so strided_slice can
// be easier to handle.
//
// So the graph may looks like below:
//   original_input -> strided_slice -> output
//      (transforms)
//   original_input -> reshape -> strided_slice -> output
//
// And the new shape is computed based on the masks.

// tfl.custom / tfl.mean / tfl.minimum / tfl.pad / tfl.pow / tfl.prelu /
// tfl.relu / tfl.relu6 / tfl.rsqrt / tfl.sin / tfl.slice / tfl.softmax /
// tfl.space_to_depth / tfl.sqrt / tfl.square / tfl.squared_difference /
// tfl.strided_slice / tfl.tanh / tfl.transpose / tfl.transpose_conv
class GpuBasicSupportedOpNoCost : public TargetHardwareOperation {
  double GetOpCost(mlir::Operation* op) const override { return 0; }

  func.return %0 : tensor<21x15x2xf32>

  // CHECK-DAG: %[[CST:.*]] = arith.constant dense<0> : tensor<3xi32>
  // CHECK-DAG: %[[CST_0:.*]] = arith.constant dense<1> : tensor<3xi32>

  // CHECK-DAG:  %[[cst:.*]] = arith.constant dense<1> : tensor<1xi32>
  // CHECK-DAG:  %[[c0:.*]] = "tfl.pseudo_const"() <{value = dense<1> : tensor<1xi32>}> : () -> tensor<1xi32>