func.func @attribute_cast(%arg0: tensor<1x4x4x1xf32>) -> tensor<1x2x2x1xf32> {
  %0 = "tfr.cast"(%arg0) : (tensor<1x4x4x1xf32>) -> !tfr.tensor
  %stride_i32 = arith.constant 2 : i32
  %1 = tfr.call @tf__my_max_pool(%0, %stride_i32, %stride_i32) : (!tfr.tensor, i32, i32) -> !tfr.tensor
  %2 = "tfr.cast"(%1) : (!tfr.tensor) -> tensor<1x2x2x1xf32>
  func.return %2 : tensor<1x2x2x1xf32>
// CHECK: tf__max_pool
}

// are also transposed accordingly.

// CHECK-LABEL: nchw_conv_with_bias_add_max_pool
// CHECK-SAME: %[[ARG:.+]]: tensor<1x2x5x5xf32>
func.func @nchw_conv_with_bias_add_max_pool(%arg0: tensor<1x2x5x5xf32>) -> tensor<1x4x2x2xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<4x2x3x3xf32>
  %1 = stablehlo.constant dense<3.000000e+00> : tensor<1x4x5x5xf32>
  %5 = stablehlo.constant dense<0xFF800000> : tensor<f32>  // -inf

  func.return %mm_s : tensor<1x1x12x3xf32>

// CHECK: %[[w:.*]] = arith.constant dense<1.270000e+02> : tensor<1x1x12x512xf32>
// CHECK: %[[mm:.*]] = "tfl.batch_matmul"(%[[w]], %arg0) <{adj_x = false, adj_y = true}>
// CHECK: return %[[mm:.*]]

// PerTensor: %[[w:.*]] = arith.constant dense<1.270000e+02> : tensor<1x1x12x512xf32>