// CHECK:           %[[VAL_6:.*]] = "tfl.reshape"(%[[VAL_1]], %[[VAL_2]]) : (tensor<1xf32>, tensor<4xi32>) -> tensor<1x1x1x1xf32>
// CHECK:           %[[VAL_7:.*]] = "tfl.concatenation"(%[[VAL_5]], %[[VAL_6]]) <{axis = 3 : i32, fused_activation_function = "NONE"}> : (tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x2xf32>

    %1 = "tfl.reshape"(%arg1, %cst) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1xf32>, tensor<4xi32>) -> tensor<1x1x1x1xf32>
    %2 = "tfl.concatenation"(%0, %1) {axis = 3 : i32, fused_activation_function = "NONE", tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x1x1x1xf32>, tensor<1x1x1x1xf32>) -> tensor<1x1x1x2xf32>

// CHECK:           %[[VAL_8:.*]] = "tfl.reshape"(%[[VAL_7]], %[[VAL_3]]) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x1x1x2xf32>, tensor<1xi32>) -> tensor<2xf32>
// CHECK:           %[[VAL_9:.*]] = "tfl.reshape"(%[[VAL_8]], %[[VAL_4]]) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<2xf32>, tensor<2xi32>) -> tensor<2x1xf32>

    %cst = "tf.Const"() {value = dense<3.00000000e-1> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
    %cst_0 = "tf.Const"() {value = dense<4.00000000e-1> : tensor<1x1x1x2xf32>} : () -> tensor<1x1x1x2xf32>
    %0 = "quantfork.stats"(%arg0) {layerStats = dense<[6.00000000e-6, 9.00000000e-1]> : tensor<2xf32>} : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3xf32>
    %1 = "tf.XlaCallModule"(%0, %cst, %cst_0) {

  %custom_2 = "tfl.custom"(%arg0, %dq_w) {custom_code = "CustomTestOp", custom_option = #tfl<const_bytes : "0x">} : (tensor<1x1x1x1xf32>, tensor<1024x1x1x1xf32>) -> tensor<*xf32>
  %custom_3 = "tfl.custom"(%arg0, %dq_w) {custom_code = "CustomTestOp", custom_option = #tfl<const_bytes : "0x">} : (tensor<1x1x1x1xf32>, tensor<1024x1x1x1xf32>) -> tensor<*xf32>

// test stablehlo roundtrip

module {
func.func @main(%arg0: tensor<1x1x1x96xf32>) -> tensor<1x1x1x96xf32> {
  %0 = "vhlo.logistic_v1"(%arg0) : (tensor<1x1x1x96xf32>) -> tensor<1x1x1x96xf32>
  %1 = "tfl.exp"(%0) : (tensor<1x1x1x96xf32>) -> tensor<1x1x1x96xf32> loc("exp")
  func.return %1 : tensor<1x1x1x96xf32>
}
}

// CHECK: func.func @main(%arg0: tensor<1x1x1x96xf32>) -> tensor<1x1x1x96xf32> attributes {tf.entry_function = {inputs = "arg0", outputs = "exp"}} {

}
// CHECK: @main(%[[ARG:.+]]: tensor<1x8x4x4xf32>) -> tensor<1x8x4x4xf32>

// Contains the `stablehlo.transpose` op of the arg (e.g. [b, f, 0, 1] to
// [b, 0, 1, f]). The weight constant is folded into [0, 1, i, o] format.
// CHECK-DAG: %[[CST:.+]] = stablehlo.constant dense<3.000000e+00> : tensor<3x3x8x8xf32>
// CHECK: %[[TRANSPOSE_1:.+]] = stablehlo.transpose %arg0, dims = [0, 2, 3, 1] : (tensor<1x8x4x4xf32>) -> tensor<1x4x4x8xf32>

       } : (tensor<1x32x32x3xf32>, tensor<4xi32>, tensor<1x32x32x8xf32>)
         -> tensor<1x1x3x8xf32>

  func.return %0 : tensor<1x1x3x8xf32>
}

// CHECK-LABEL: func @transposeConv2DBackpropInput
func.func @transposeConv2DBackpropInput(
  %input_sizes: tensor<4xi32>,
  %filter: tensor<1x1x3x8xf32>,
  %out_backprop: tensor<1x32x32x8xf32>
) -> tensor<1x32x32x3xf32> {

  %4 = "quantfork.stats"(%3) {
    layerStats = dense<[0.0, 4.0]> : tensor<2xf32>
  } : (tensor<?x1x1x3xf32>) -> tensor<?x1x1x3xf32>
  %5 = "tfl.sqrt"(%4) : (tensor<?x1x1x3xf32>) -> tensor<?x1x1x3xf32>
  %6 = "quantfork.stats"(%5) {
    layerStats = dense<[0.0, 2.0]> : tensor<2xf32>
  } : (tensor<?x1x1x3xf32>) -> tensor<?x1x1x3xf32>

module attributes {tfl.metadata = {"keep_stablehlo_constant" = "true"}} {
 func.func @main () -> tensor<1x1x1x96xf32> {
  %0 = "vhlo.constant_v1"() <{value = #vhlo.tensor_v1<dense<0.000000e+00> : tensor<f32>>}> : () -> tensor<1x1x1x96xf32>
  func.return %0 : tensor<1x1x1x96xf32>
 }
}

//CHECK: func.func @main() -> tensor<1x1x1x96xf32> attributes {tf.entry_function = {outputs = "vhlo.constant_v1"}} {