}
// CHECK: func.func @main(%[[ARG:.+]]: tensor<1x3x4x3xf32>) -> tensor<1x3x2x2xf32> {
// CHECK-DAG: %[[CONST_0:.*]] = stablehlo.constant dense<[{{.*}}]> : tensor<2xf32>
// CHECK-DAG: %[[CONST_1:.*]] = stablehlo.constant dense<[{{.*}}]> : tensor<2x3x3x2xf32>
// CHECK-DAG: %[[CONV:.*]] = stablehlo.convolution(%[[ARG]], %[[CONST_1]]) {{.*}} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32>

  %0 = "tf.Conv2D"(%output, %cst) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true}> {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", device = ""} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x2x2x2xf32>

// CHECK:           %[[VAL_20:.*]] = "tfl.quantize"(%[[VAL_19]]) <{qtype = tensor<1x384x128x!quant.uniform<i8:f32, 3.000000e-01:-3>>}> {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x384x128xf32>) -> tensor<1x384x128x!quant.uniform<i8:f32, 3.000000e-01:-3>>

func.func private @conv(%input: tensor<1x3x4x3xf32> {tf._user_specified_name = "input_tensor"}) -> tensor<*xf32> attributes {tf._construction_context = "kEagerRuntime", tf._input_shapes = [#tf_type.shape<1x3x4x3>]} {
  %weight = arith.constant dense_resource<__elided__> : tensor<2x3x3x2xf32>
  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

  %3 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<1x1x384xf32>} : () -> tensor<1x384x384xf32>
  %4 = "tfl.add"(%2, %3) {tac.device = "GPU", tac.inference_type = "FLOAT", fused_activation_function = "NONE"} : (tensor<1x384x384xf32>, tensor<1x384x384xf32>) -> tensor<1x384x384xf32>

  %45 = mhlo.reshape %44 : (tensor<1x64x32x1xf32>) -> tensor<1x64x32xf32>
  %46 = "mhlo.broadcast_in_dim"(%45) <{broadcast_dimensions = dense<[0, 1, 2]> : tensor<3xi64>}> : (tensor<1x64x32xf32>) -> tensor<1x64x32x32xf32>
  %47 = mhlo.multiply %43, %46 : tensor<1x64x32x32xf32>
  %48 = mhlo.add %30, %47 : tensor<1x64x32x32xf32>

}

// -----

func.func @testBatchMatmulHybridQuant(%arg0 : tensor<1x4x384x32xf32>, %arg1 : tensor<1x4x384x32x!quant.uniform<i8:f32, 0.11:-16>>) -> tensor<1x4x384x384xf32> {
  // CHECK: "tfl.batch_matmul"(%arg0, %arg1)
  %0 = "tfl.batch_matmul"(%arg0, %arg1) {adj_x = false, adj_y = true} : (tensor<1x4x384x32xf32>, tensor<1x4x384x32x!quant.uniform<i8:f32, 0.11:-16>>) -> tensor<1x4x384x384xf32>

  // CHECK: %[[CST_0:.*]] = "tfl.no_value"() <{value}> : () -> none
  // CHECK: %[[ARG1:.*]] = "tfl.transpose_conv"(%arg0, %[[ARG0]], %arg2, %[[CST_0]]) <{fused_activation_function = "NONE", padding = "SAME", stride_h = 2 : i32, stride_w = 2 : i32}> : (tensor<4xi32>, tensor<1x3x3x32xf32>, tensor<15x14x14x32xf32>, none) -> tensor<15x28x28x1xf32>