%0 = "mhlo.reshape"(%arg0) : (tensor<1x1x512xf32>) -> tensor<1x512xf32>
  %1 = "mhlo.dot"(%0, %arg1) : (tensor<1x512xf32>, tensor<512x13x!quant.uniform<i8:f32, 0.00285>>) -> tensor<1x13xf32>
  %2 = "mhlo.reshape"(%1) : (tensor<1x13xf32>) -> tensor<1x1x13xf32>
  func.return %2 : tensor<1x1x13xf32>

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

    %1 = stablehlo.constant dense<-1.280000e+02> : tensor<1x1024xf32>
    %2 = stablehlo.constant dense<0.003921567> : tensor<1x1024xf32>
    %3 = stablehlo.divide %arg0, %2 : tensor<1x1024xf32>
    %4 = stablehlo.add %3, %1 : tensor<1x1024xf32>
    %5 = "tf.Identity"(%4) {device = ""} : (tensor<1x1024xf32>) -> tensor<1x1024xf32>
    return %5 : tensor<1x1024xf32>
  }

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

    %1 = "tf.Identity"(%0#3) {device = ""} : (tensor<1x1024xf32>) -> tensor<1x1024xf32>
    func.return %1 : tensor<1x1024xf32>
  }

  %1 = "tf.BiasAdd"(%0, %arg0) {data_format = "NHWC"} : (tensor<*xf32>, tensor<128xf32>) -> tensor<*xf32>
  %2 = "tf.Identity"(%1) : (tensor<*xf32>) -> tensor<*xf32>
  func.return %2 : tensor<*xf32>
}

// CHECK-LABEL: conv2DBiasAdd_reluActivation
func.func @conv2DBiasAdd_reluActivation(%arg0: tensor<128xf32>, %arg1: tensor<1x1x3x128xf32>, %arg2: tensor<8x32x32x3xf32>) -> (tensor<*xf32>) {

  // CHECK-NEXT: [[RESULT:%.*]] = "tf.BiasAdd"([[SQUEEZE]], [[BIAS]]) : (tensor<1x128x8xf32>, tensor<8xf32>) -> tensor<1x128x8xf32>
  // CHECK-NEXT: return [[RESULT]] : tensor<1x128x8xf32>
}

func.func @testDilatedConv1DExpandW(%arg0: tensor<1x128x3xf32>, %arg1: tensor<5x1x3x8xf32>) -> tensor<1x128x8xf32> {
  %cst = "tf.Const"() {value = dense<0> : tensor<1x2xi32>} : () -> tensor<1x2xi32>

  %prelu = "tfl.prelu"(%arg0, %cst) : (tensor<1x10x10x3xf32>, tensor<1x1x3xf32>) -> tensor<1x10x10x3xf32>
  func.return %prelu : tensor<1x10x10x3xf32>

// CHECK: %[[cst:.*]] = arith.constant dense<[{{\[}}[1.66394591, 3.61694336, 2.0382936]]]> : tensor<1x1x3xf32>

    %3 = stablehlo.concatenate %2, %1, dim = 0 : (tensor<1x3xf32>, tensor<1x3xf32>) -> tensor<2x3xf32>
    return %3 : tensor<2x3xf32>
  }
  func.func private @composite_dot_general_fn_1(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module, tf_quant.composite_function} {

}

// -----

module {
func.func private @func_20_GPU_FLOAT(%arg0: tensor<128x128xf32>, %arg1: tensor<3xi32>) -> tensor<1x128x128xf32> attributes {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_20"} {
  %0 = "tfl.reshape"(%arg0, %arg1) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<128x128xf32>, tensor<3xi32>) -> tensor<1x128x128xf32>
  func.return %0 : tensor<1x128x128xf32>
}