}
// 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>

func.func @broadcast_to_to_reshape(%arg0: tensor<4x4x4xf32>, %arg1 : tensor<4xi32>) -> tensor<1x4x4x4xf32> {
  %0 = "tfl.broadcast_to"(%arg0, %arg1) : (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>
  // CHECK: "tfl.reshape"
  // CHECK-SAME: (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>
  func.return %0 : tensor<1x4x4x4xf32>
}

  return %2 : tensor<1x8x4x4xf32>
}
// CHECK-DAG: %[[CONST:.+]] = stablehlo.constant {{.*}} : tensor<3x3x8x8xf32>
// CHECK-DAG: %[[TRANSPOSE_0:.+]] = stablehlo.transpose %[[ARG]], dims = [0, 2, 3, 1] : (tensor<1x8x4x4xf32>) -> tensor<1x4x4x8xf32>

// CHECK:           %[[VAL_6:.*]] = arith.constant dense<[0, 3, 1, 2]> : tensor<4xi32>
// CHECK:           %[[VAL_7:.*]] = "tfl.transpose"(%[[VAL_5]], %[[VAL_6]]) : (tensor<1x4x4x3xf32>, tensor<4xi32>) -> tensor<1x3x4x4xf32>
// CHECK:           %[[VAL_8:.*]] = "tf.Identity"(%[[VAL_7]]) {device = ""} : (tensor<1x3x4x4xf32>) -> tensor<*xf32>

  %2 = "tf.Transpose"(%arg0, %0) : (tensor<1x4x4x8xf32>, tensor<4xi32>) -> tensor<1x8x4x4xf32>
  %3 = "tf.Transpose"(%2, %1) : (tensor<1x8x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x8xf32>

  func.return %3 : tensor<1x4x4x8xf32>
  // CHECK: return %arg0
}

// CHECK-LABEL: @nonCancellableTranspose
func.func @nonCancellableTranspose(%arg0: tensor<1x4x4x8xf32>) -> tensor<4x1x4x8xf32> {

// CHECK-SAME:      %arg0: tensor<2x4x4x1xf32>,
// CHECK-SAME:      %arg1: tensor<2x4x4x2xf32>) -> tensor<2x4x4x1xf32> attributes {tf._implements = "DenseImageWarp"} {
// CHECK-NEXT:    %0 = "tfl.custom"(%arg0, %arg1) <{custom_code = "DenseImageWarp", custom_option = #tfl<const_bytes : "0x">}> : (tensor<2x4x4x1xf32>, tensor<2x4x4x2xf32>) -> tensor<2x4x4x1xf32>
// CHECK-NEXT:    return %0 : tensor<2x4x4x1xf32>
// CHECK-NEXT:  }
}

// -----

      %1 = "tf.XlaCallModule"(%0, %cst_0, %cst_1) <{Sout = [#tf_type.shape<1x4x4x3>], module = "", version = 9 : i64}> {_entry_function = @composite_fn_2, _original_entry_function = "composite_fn_2", _tfl_quant_trait = "fully_quantizable"} : (tensor<1x4x4x3xf32>, tensor<3x1x1x3xf32>, tensor<3xf32>) -> tensor<1x4x4x3xf32>
      return %1 : tensor<1x4x4x3xf32>

  func.return %2 : tensor<4x4x4xf32>
}

func.func @testSingleLstmFloat16(%arg0: tensor<4x4xf32>, %arg1: tensor<4xf32>, %arg2: tensor<4x4x4xf32>) -> tensor<4x4x4xf32> {
  // CHECK-LABEL: testSingleLstm
  // CHECK-DAG:  %[[CST_0:.*]] = "tfl.pseudo_const"() <{value = dense<0.000000e+00> : tensor<4x4xf16>}> : () -> tensor<4x4xf16>

func.func @conv_fn(%arg0: tensor<1x3x3x4xf32>) -> tensor<1x3x3x4xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<3x3x4x4xf32>
  %1 = stablehlo.convolution(%arg0, %0) dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f], window = {pad = [[1, 1], [1, 1]]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<1x3x3x4xf32>, tensor<3x3x4x4xf32>) -> tensor<1x3x3x4xf32>
  func.return %1: tensor<1x3x3x4xf32>
}

func.func @broadcast_multi_dim_add(%arg0: tensor<4x1x1xi32>, %arg1: tensor<4x4x4x4xi32>) -> tensor<4x4x4x4xi32> {
  // CHECK-NEXT: %[[LHS_BCAST:.+]] = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<[1, 2, 3]> : tensor<3xi64>}>
  // CHECK-NEXT: mhlo.add %[[LHS_BCAST]], %arg1
  %0 = "tf.AddV2"(%arg0, %arg1) : (tensor<4x1x1xi32>, tensor<4x4x4x4xi32>) -> tensor<4x4x4x4xi32>
  func.return %0: tensor<4x4x4x4xi32>
}

// CHECK-LABEL: func @add_dynamic