%0 = "tfl.reshape"(%arg0, %cst) : (tensor<128x64xf32>, tensor<3xi32>) -> tensor<1x128x64xf32>
  %1 = "tfl.fully_connected"(%0, %arg1, %arg2) {fused_activation_function = "NONE", keep_num_dims = true, weights_format = "DEFAULT"} : (tensor<1x128x64xf32>, tensor<32x64xf32>, tensor<32xf32>) -> tensor<1x128x32xf32>
  func.return %1 : tensor<1x128x32xf32>
  // CHECK-DAG: %[[CST:.*]] = arith.constant dense<[1, 128, 64]> : tensor<3xi32>

  // CHECK-NEXT: [[RESULT:%.*]] = "tf.BiasAdd"([[SQUEEZE]], [[BIAS]]) : (tensor<1x128x8xf32>, tensor<8xf32>) -> tensor<1x128x8xf32>
  // CHECK-NEXT: return [[RESULT]] : tensor<1x128x8xf32>
}

func.func @testDilatedConv1DExpandW(%arg0: tensor<1x128x3xf32>, %arg1: tensor<5x1x3x8xf32>) -> tensor<1x128x8xf32> {
  %cst = "tf.Const"() {value = dense<0> : tensor<1x2xi32>} : () -> tensor<1x2xi32>

// 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>

// -----

// CHECK-LABEL: testLiftDotConcatLHSAndRHS
func.func @testLiftDotConcatLHSAndRHS(%arg0: tensor<1x72x128xf32>, %arg1: tensor<1x128x72xf32>, %arg2: tensor<1x72x128xf32>, %arg3: tensor<1x128x72xf32>, %arg4: tensor<1x72x128xf32>, %arg5: tensor<1x128x72xf32>, %arg6: tensor<1x72x128xf32>, %arg7: tensor<1x128x72xf32>) -> tensor<4x72x72xf32> {
  %0 = "mhlo.dot_general"(%arg0, %arg1) {
    dot_dimension_numbers = #mhlo.dot<

  %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>

  // CHECK: %[[TRANSPOSE_2:.*]] = "tf.Transpose"(%arg1, %[[CST_0]]) : (tensor<8x64x64xf32>, tensor<3xi32>) -> tensor<64x8x64xf32>
  // CHECK: %[[ADD:.*]] = "tf.AddV2"(%[[TRANSPOSE_1]], %[[TRANSPOSE_2]]) {device = ""} : (tensor<1x8x64xf32>, tensor<64x8x64xf32>) -> tensor<64x8x64xf32>

func.func @Int64SliceBeginSize(%arg0: tensor<4x128x32xf32>) -> tensor<1x128x32xf32> {
  %0 = "tfl.pseudo_const"() {value = dense<0> : tensor<3xi64>} : () -> tensor<3xi64>
  %1 = "tfl.pseudo_const"() {value = dense<[1, 128, 32]> : tensor<3xi64>} : () -> tensor<3xi64>
  %2 = "tfl.slice"(%arg0, %0, %1) : (tensor<4x128x32xf32>, tensor<3xi64>, tensor<3xi64>) -> tensor<1x128x32xf32>
  func.return %2 : tensor<1x128x32xf32>

    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>>,

  // 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]

// dimensions into a single dimension. For example,
//
//   %shape = arith.constant dense<[1, 128, 64]> : tensor<3xi32>
//   %reshape = tfl.reshape(%input, %shape) // %input: tensor<128x64xf32>
//   %fc = tfl.fully_connected(%reshape, %filter, %bias)
//           {keep_num_dims = false, weights_format = "DEFAULT"}
//
// can be canonicalized to
//