// PerChannel: %[[out_2:.*]] = "tf.PartitionedCall"(%arg0, %[[q_w]], %[[scale]], %[[zp]]) <{config = "", config_proto = "", executor_type = "",
// PerChannel-SAME: f = @quantized_conv2d_fn_0}> : (tensor<1x2x2x3xf32>, tensor<2x3x3x2xi8>, tensor<2xf32>, tensor<2xi32>) -> tensor<*xf32>
// PerChannel: return %[[out_1]], %[[out_2]]

}

// -----

    %26 = "tf.Cast"(%25) {Truncate = false} : (tensor<2xi64>) -> tensor<2xf32>
    %27 = "tf.Equal"(%14, %26) {incompatible_shape_error = true} : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xi1>
    %28 = "tf.Cast"(%27) {Truncate = false} : (tensor<2xi1>) -> tensor<2xf32>
    %29 = "tf.Sum"(%28, %6) {keep_dims = false} : (tensor<2xf32>, tensor<1xi32>) -> tensor<f32>

  %1:2 = "tfl.split"(%cst, %0) {num_splits = 2 : i32} : (tensor<i32>, tensor<4xf32>) -> (tensor<2xf32>, tensor<2xf32>)
  %2 = "tfl.quantize"(%1#0) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>
  %3 = "tfl.quantize"(%1#1) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>

// CHECK: return %[[cast_i32]] : !tfr.tensor
}

// CHECK-LABEL: decompose_output_type
func.func @decompose_output_type(%arg0: tensor<2xf32>) -> tensor<2xi32> {
  %0 = "tf.CastFloat"(%arg0) : (tensor<2xf32>) -> tensor<2xi32>
  func.return %0: tensor<2xi32>
// CHECK: %[[i32:.*]] = tfr.constant i32 -> !tfr.attr
// CHECK: tfr.call @tf__cast(%[[casted_arg:.*]], %[[i32]], %false) : (!tfr.tensor, !tfr.attr, i1) -> !tfr.tensor

  %0 = "tf.RealDiv"(%arg0, %arg0) : (tensor<2xi32>, tensor<2xi32>) -> tensor<2xi32>
  func.return %0: tensor<2xi32>
}

// CHECK-LABEL: func @sub
func.func @sub(%arg0: tensor<2xi32>) -> tensor<2xi32> {
  // CHECK-NEXT:  %0 = mhlo.subtract %arg0, %arg0 : tensor<2xi32>
  // CHECK-NEXT:  return %0 : tensor<2xi32>
  %0 = "tf.Sub"(%arg0, %arg0) : (tensor<2xi32>, tensor<2xi32>) -> tensor<2xi32>
  func.return %0: tensor<2xi32>
}

"TANH", input_to_cell_intermediate = tensor<0xf32>, input_to_forget_intermediate = tensor<0xf32>, input_to_input_intermediate = tensor<0xf32>, input_to_output_intermediate = tensor<0xf32>, proj_clip = 0.000000e+00 : f32, time_major = false} : (tensor<?x?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>,...

// -----

// CHECK-LABEL: testDotToDotGeneralMatrixVector
func.func @testDotToDotGeneralMatrixVector(%arg0: tensor<2x3072xf32>, %arg1: tensor<3072xf32>) -> tensor<2xf32> {
  %0 = "mhlo.dot"(%arg0, %arg1) : (tensor<2x3072xf32>, tensor<3072xf32>) -> tensor<2xf32>
  func.return %0 : tensor<2xf32>

// CHECK:      %[[RES:.*]] = "mhlo.dot_general"(%arg0, %arg1) <{
// CHECK-SAME:   dot_dimension_numbers = #mhlo.dot<

func.func @main(tensor<4x4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4x4xf32>, tensor<4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>) -> tensor<4x4x4xf32> {
// CHECK: {
// CHECK-NEXT:   version: 3,
// CHECK-NEXT:   operator_codes: [ {

  %cst_1 = arith.constant dense<0> : tensor<4xi32>
  %cst_2 = arith.constant dense<0.000000e+00> : tensor<1xf32>
  %0 = "tfl.quantize"(%cst_2) {qtype = tensor<1x!quant.uniform<i32:f32, 1.0>>, volatile} : (tensor<1xf32>) -> tensor<1x!quant.uniform<i32:f32, 1.0>>
  %1 = "tfl.dequantize"(%0) : (tensor<1x!quant.uniform<i32:f32, 1.0>>) -> tensor<1xf32>
  %cst_3 = arith.constant dense<[[[[1.0]], [[1.0]], [[1.0]]]]> : tensor<1x3x1x1xf32>

  %1:4 = "tfl.split"(%cst, %0) {num_splits = 4 : i32} : (tensor<i32>, tensor<4xf32>) -> (tensor<2xf32>, tensor<2xf32>,tensor<2xf32>, tensor<2xf32>)
  %2 = "tfl.quantize"(%1#0) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>
  %3 = "tfl.quantize"(%1#1) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>