%0 = "tf.Softplus"(%arg0) : (tensor<8x16xf16>) -> tensor<8x16xf16>

  // CHECK:     return [[ENTRY_SELECT]] : tensor<8x16xf16>
  func.return %0 : tensor<8x16xf16>
}

// -----

// CHECK-LABEL: func @softplus_bf16
// CHECK-SAME: ([[FEATURES:%.*]]: tensor<8x16xbf16>)
func.func @softplus_bf16(%arg0: tensor<8x16xbf16>) -> tensor<8x16xbf16> {

  func.return %0 : tensor<8x16xf32>

// CHECK-LABEL: floor_div
// CHECK:  tfl.floor_div %arg0, %arg1 : tensor<8x16xf32>
// CHECK:  return
}

func.func @floor_div_i16(tensor<8x16xi16>, tensor<8x16xi16>) -> tensor<8x16xi16> {
^bb0(%arg0: tensor<8x16xi16>, %arg1: tensor<8x16xi16>):
  %0 = "tf.FloorDiv"(%arg0, %arg1) : (tensor<8x16xi16>, tensor<8x16xi16>) -> tensor<8x16xi16>

  // CHECK: return %0
}

// CHECK-LABEL: testAddOfNegRight
func.func @testAddOfNegRight(%arg0: tensor<8x16xf32>, %arg1: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Neg"(%arg1) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  %1 = "tf.Add"(%arg0, %0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>

}

// CHECK-LABEL: softmax
func.func @softmax(%arg0: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Softmax"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
// CHECK: %[[SOFTMAX_0:.*]] = "tf.Softmax"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
// CHECK: return %[[SOFTMAX_0]] : tensor<8x16xf32>
}

// CHECK-LABEL: conv2d_backprop_input_with_add

  // CHECK: %[[DQ_2:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>
  // CHECK: %[[DQ_3:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>
  // CHECK: %[[DQ_4:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>
  // CHECK: %[[DQ_5:.*]] = "tfl.dequantize"({{.*}}) : (tensor<1x1xf16>) -> tensor<1x1xf32>

func.func @softmax(%arg0: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Softmax"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
// CHECK: %[[CUSTOM_0:.*]] = "tfl.custom"(%arg0) <{custom_code = "FlexSoftmax", custom_option = #tfl<const_bytes : "0x07536F66746D617800161207536F66746D61781A002A070A0154120230013200000221191414042801">}> : (tensor<8x16xf32>) -> tensor<8x16xf32>

  %0 = "tfl.conv_2d"(%arg0, %arg1, %cst_0) {dilation_h_factor = 2 : i32, dilation_w_factor = 3 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 4 : i32, stride_w = 5 : i32} : (tensor<256x32x32x3xf16>, tensor<16x3x3x3xf16>, tensor<16xf16>) -> tensor<256x8x7x16xf16>

  func.return %0 : tensor<8x16xf32>
}

// -----

func.func @testCumprod(%arg: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %axis = arith.constant dense<-3> : tensor<i32>
  // expected-error @+1 {{axis operand should be within range [-2, 2)}}
  %0 = "tf.Cumprod"(%arg, %axis) : (tensor<8x16xf32>, tensor<i32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
}

  %1 = stablehlo.transpose %arg0, dims = [1, 0] : (tensor<16x8xf32>) -> tensor<8x16xf32>
  %2 = "stablehlo.reduce_window"(%1, %0) ({
  ^bb0(%arg1: tensor<f32>, %arg2: tensor<f32>):
      %3 = stablehlo.maximum %arg1, %arg2 : tensor<f32>
      stablehlo.return %3 : tensor<f32>
  }) {window_dimensions = array<i64: 2, 2>, window_strides = array<i64: 2, 2>} : (tensor<8x16xf32>, tensor<f32>) -> tensor<4x8xf32>
  return %2 : tensor<4x8xf32>
}

// CHECK: %[[UQ_1:.*]] = stablehlo.uniform_quantize %[[ARG_1]] : (tensor<8x16x4xf32>) -> tensor<8x16x4x!quant.uniform<i8:f32, 2.500000e+00>>