%w = arith.constant dense<-1.0> : tensor<32x3x3x3xf32>
  %3 = "tfl.quantize"(%w) {qtype = tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>} : (tensor<32x3x3x3xf32>) -> tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>
  %4 = "tfl.dequantize"(%3) : (tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>) -> tensor<32x3x3x3xf32>

  func.return %rst : tensor<256x8x7x4xf32>

// LOBIT-DAG: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<4xf32>
// LOBIT-DAG: %[[CONSTANT0:.*]] = arith.constant dense<0.000000e+00> : tensor<4x3x3x3xf32>
// LOBIT: %[[QUANTIZE:.*]] = "tfl.quantize"(%[[CONSTANT0]]) <{qtype = tensor<4x3x3x3x!quant.uniform<u4<1:15>:f32:0, {1.000000e+00:1,1.000000e+00:2,1.000000e+00:7,1.000000e+00:15}>>}>

  %cst = "tf.Const"() {value = dense<1.000000e+00> : tensor<2x3x3x3xf32>} : () -> tensor<2x3x3x3xf32>
  %cst_0 = "tf.Const"() {value = dense<0.500000e+00> : tensor<1x3x2x3xf32>} : () -> tensor<1x3x2x3xf32>
  %0 = "tf.Conv2D"(%arg0, %cst) {data_format = "NHWC", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x3xf32>) -> tensor<1x3x2x3xf32>

// CHECK:         }

// -----

module {
func.func @constWeight(%arg0: tensor<256x32x32x3xf32>) -> tensor<256x30x30x16xf32> {
  %0 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<16x3x3x3xf32>} : () -> tensor<16x3x3x3xf32>
  %1 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<16xf32>} : () -> tensor<16xf32>

// CHECK:  %0 = "tf.Transpose"(%arg1, %[[CONSTANT0]]) : (tensor<3x3x3x16xf32>, tensor<4xi32>) -> tensor<16x3x3x3xf32>
// CHECK:  %1 = "tfl.conv_2d"(%arg0, %0, %[[CONSTANT]]) <{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<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<256x8x7x16xf32>
// CHECK:  %2 = "tf.Conv2D"

  func.return %0 : tensor<256x32x32x16xf32>
}

// -----

func.func @testConv2D4DBias(tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<1x1x1x16xf32>) -> tensor<256x32x32x16xf32> {
^bb0(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<16x3x3x3xf32>, %arg2: tensor<1x1x1x16xf32>):

  func.return %rst : tensor<256x8x7x16xf32>

// CHECK-DAG: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<16xf32>
// CHECK-DAG: %[[CONSTANT0:.*]] = arith.constant dense<0.000000e+00> : tensor<16x3x3x3xf32>
// CHECK: %[[QUANTIZE:.*]] = "tfl.quantize"(%[[CONSTANT0]]) <{qtype = tensor<16x3x3x3x!quant.uniform<u8:f32, 1.000000e+00>>}>
// CHECK: %[[DEQUANTIZE:.*]] = "tfl.dequantize"(%[[QUANTIZE]])

// CHECK:           %[[VAL_2:.*]] = "tf.Conv2D"(%[[VAL_0]], %[[VAL_1]]) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [0, 0, 1, 1, 1, 1, 0, 0], padding = "EXPLICIT", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true}> : (tensor<64x8x8x8xf32>, tensor<8x8x8x64xf32>) -> tensor<64x3x3x64xf32>
// CHECK:           return %[[VAL_2]] : tensor<64x3x3x64xf32>

// -----

// CHECK-LABEL: func @diag_part
// CHECK-SAME: %[[ARG:.*]]: tensor<4x3x4x3xf32>
func.func @diag_part(%arg0: tensor<4x3x4x3xf32>) -> tensor<4x3xf32> {
  // CHECK: %[[RS:.*]] = mhlo.reshape %[[ARG]] : (tensor<4x3x4x3xf32>) -> tensor<12x12xf32>
  // CHECK-DAG: %[[IOTA0:.*]] = "mhlo.iota"() <{iota_dimension = 0 : i64}> : () -> tensor<12x12xi32>