// expected-error @+1 {{requires orig_output to be rank 4}}
  %result = "tf.MaxPoolGrad"(%orig_input, %orig_output, %grad) {
     data_format = "NHWC",
     ksize = [1, 2, 2, 1],
     padding = "VALID",
     strides = [1, 2, 2, 1]

// DISABLE-ALL-DOT-GENERAL: @main
func.func @main(%arg0: tensor<1x1x167xf32>) -> tensor<1x1x64xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<167x64xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [2] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x1x167xf32>, tensor<167x64xf32>) -> tensor<1x1x64xf32>
  return %1 : tensor<1x1x64xf32>
}

// DISABLE-ALL-DOT-GENERAL: %[[CONST:.+]] = stablehlo.constant dense<2.000000e+00>

  %0 = mhlo.constant dense<0.000000e+00> : tensor<f32>
  %1 = "mhlo.pad"(%arg0, %0) <{edge_padding_high = dense<1> : tensor<3xi64>, edge_padding_low = dense<1> : tensor<3xi64>, interior_padding = dense<0> : tensor<3xi64>}> : (tensor<10x10x10xf32>, tensor<f32>) -> tensor<12x12x12xf32>
  %2 = mhlo.exponential %1 : tensor<12x12x12xf32>
  %3 = mhlo.constant dense<0.000000e+00> : tensor<f32>

  %5 = stablehlo.broadcast_in_dim %1, dims = [2] : (tensor<64xf32>) -> tensor<1x1x64xf32>
  %6 = stablehlo.add %4, %5 : tensor<1x1x64xf32>
  %7 = stablehlo.clamp %2, %6, %3 : tensor<1x1x64xf32>
  func.return %7: tensor<1x1x64xf32>
}
// CHECK: %[[CONST_0:.*]] = stablehlo.constant dense<2.000000e+00>
// CHECK: %[[CONST_1:.*]] = stablehlo.constant dense<2.000000e+00>

    precision_config = #vhlo.array_v1<[#vhlo<precision_v1 DEFAULT>, #vhlo<precision_v1 DEFAULT>]>}> : (tensor<1x1x167xf32>, tensor<167x64xf32>) -> tensor<1x1x64xf32>
  return %0 : tensor<1x1x64xf32>
}

//CHECK:func.func private @dot_general(%arg0: tensor<1x1x167xf32>, %arg1: tensor<167x64xf32>) -> tensor<1x1x64xf32> {
//CHECK-NEXT: %0 = "vhlo.dot_general_v1"(%arg0, %arg1) <{
//CHECK-SAME:    lhs_batching_dimensions = #vhlo.tensor_v1<dense<0> : tensor<1xi64>>,

  %404 = "tfl.batch_matmul"(%402, %403) {adj_x = false, adj_y = false, asymmetric_quantize_inputs = true} : (tensor<2x12x128xf32>, tensor<128x64x!quant.uniform<i8:f32, 0.0047710379585623741>>) -> tensor<2x12x64xf32>
  %405 = "tfl.reshape"(%404, %cst_0) : (tensor<2x12x64xf32>, tensor<2xi32>) -> tensor<2x768xf32>
  return %405 : tensor<2x768xf32>
  // CHECK: %cst = arith.constant dense_resource<__elided__> : tensor<3xi32>