func.return %a: tensor<*xf32>
  }

// CHECK-LABEL: same_scale_test
// CHECK: %[[maxpool:.*]] = "tf.MaxPool"
// CHECK: %[[q1:.*]] = "quantfork.qcast"(%[[maxpool]])
// CHECK-SAME: quant.uniform<i8:f32, 5.000000e-02:-10>
// CHECK: %[[dq1:.*]] = "quantfork.dcast"(%[[q1]])
// CHECK-SAME: quant.uniform<i8:f32, 5.000000e-02:-10>

// CHECK: %[[q0:.*]] = "quantfork.qcast"(%arg0)
// CHECK: %[[dq0:.*]] = "quantfork.dcast"(%[[q0]])
// CHECK-SAME: quant.uniform<i8:f32, 0.010039215461880554:-1>
// CHECK: %[[maxpool:.*]] = "tf.MaxPool"(%[[dq0]])
// CHECK: %[[q1:.*]] = "quantfork.qcast"(%[[maxpool]])
// CHECK-SAME: quant.uniform<i8:f32, 0.010039215461880554:-1>
// CHECK: %[[dq1:.*]] = "quantfork.dcast"(%[[q1]])
// CHECK-SAME: quant.uniform<i8:f32, 0.010039215461880554:-1>

func.func @fold_into_max_pool(%arg0: tensor<1x64x112x112xf32>) -> tensor<1x56x56x64xf32> {

  // MaxPool operand transpose must be folded into the op and MaxPool
  // must use NCHW data format with updated kernel size and strides.

  // CHECK: %[[RES_PERM:.*]] = "tf.Const"() <{value = dense<[0, 2, 3, 1]> : tensor<4xi32>}>

  // CHECK: %[[R0:.*]] = "tf.Transpose"(%arg0, %[[CST]])
  // CHECK: %[[R1:.*]] = "tf.MaxPool"(%[[R0]]) <{data_format = "NHWC", explicit_paddings = [], ksize = [1, 3, 3, 1], padding = "SAME", strides = [1, 2, 2, 1]}>
  // CHECK: %[[CST_0:.*]] = "tf.Const"() <{value = dense<[0, 3, 1, 2]> : tensor<4xi64>}>
  // CHECK: "tf.Transpose"(%[[R1]], %[[CST_0]])
  %0 = "tf.MaxPool"(%arg0)
       {
         data_format = "NCHW",
         ksize = [1, 1, 3, 3],

 BiasAdd 1
 BiasAddGrad 1
 Cast 3
 Const 357
 Conv2D 53
 Conv2DBackpropFilter 53
 Conv2DBackpropInput 52
 DivNoNan 1
 Equal 1
 FusedBatchNorm 53
 FusedBatchNormGrad 53
 Identity 2
 MatMul 3
 MaxPool 1
 MaxPoolGrad 1
 Mean 1
 Mul 164
 Pad 1
 ReadVariableOp 646
 Relu 49
 ReluGrad 49
 Reshape 2
 ResourceApplyKerasMomentum 161
 ShapeN 50
 Softmax 1
 SparseSoftmaxCrossEntropyWithLogits 1

  %0 = "quantfork.qcast"(%arg0) : (tensor<*xf32>) -> tensor<*x!quant.uniform<i8:f32, 5.000000e-02:-10>>
  %1 = "quantfork.dcast"(%0) : (tensor<*x!quant.uniform<i8:f32, 5.000000e-02:-10>>) -> tensor<*xf32>
  %2 = "tf.MaxPool"(%1) {data_format = "NHWC", device = "", explicit_paddings = [], ksize = [1, 2, 2, 1], padding = "VALID", strides = [1, 2, 2, 1]} : (tensor<*xf32>) -> tensor<*xf32>

    }
  }
  attr {
    key: "strides"
    value {
      list {
        i: 1
        i: 2
        i: 2
        i: 1
      }
    }
  }
}
node {
  name: "MaxPool"
  op: "MaxPool"
  input: "MobilenetV1/MobilenetV1/Conv2d_0/Conv2D"
  attr {
    key: "strides"
    value {
      list {
        i: 1
        i: 2
        i: 2
        i: 1
      }
    }
  }

 AssignAddVariableOp 1
 BiasAdd 1
 BiasAddGrad 1
 Cast 115
 Const 407
 Conv2D 53
 Conv2DBackpropFilter 53
 Conv2DBackpropInput 52
 Equal 1
 FusedBatchNormGradV2 53
 FusedBatchNormV2 53
 MatMul 3
 MaxPool 1
 MaxPoolGrad 1
 Mean 1
 Mul 218
 Pad 2
 ReadVariableOp 538
 Relu 49
 ReluGrad 49
 Reshape 2
 ResourceApplyKerasMomentum 161
 Slice 1
 Softmax 1
 SparseSoftmaxCrossEntropyWithLogits 1

  %0 = "quantfork.qcast"(%arg0) : (tensor<*xf32>) -> tensor<*x!quant.uniform<i8:f32, 5.000000e-02:-10>>
  %1 = "quantfork.dcast"(%0) : (tensor<*x!quant.uniform<i8:f32, 5.000000e-02:-10>>) -> tensor<*xf32>
  %2 = "tf.MaxPool"(%1) {data_format = "NHWC", device = "", explicit_paddings = [], ksize = [1, 2, 2, 1], padding = "VALID", strides = [1, 2, 2, 1]} : (tensor<*xf32>) -> tensor<*xf32>

  let description = [{
    Operation supports folding operand(s) transposes into the operation itself.

    (1) Operation might have layout dependent operands and results...

      Example:  MaxPool(Transpose($arg, $perm))
                  -> Transpose(MaxPool($arg, $perm))

    (2) ... or it might have only layout dependent operands:

      Example: Mean(Transpose($arg, $reduction_dims))
                 -> Mean($arg, Transpose($reduction_dims))