// CHECK-NEXT:  %[[cst:.*]] = arith.constant dense<1> : tensor<i32>
  %cst = arith.constant dense<1> : tensor<i32>
// CHECK-NEXT:  %[[softmax:.*]] = "tfl.softmax"(%arg0) <{beta = 1.000000e+00 : f32}> : (tensor<128x16xf32>) -> tensor<128x16xf32>
  %0 = "tfl.softmax"(%arg0) {beta = 1.000000e+00 : f32} : (tensor<128x16xf32>) -> tensor<128x16xf32>

  func.return %0 : tensor<8x16xf32>

// CHECK-LABEL:minimum
// CHECK:  "tfl.minimum"(%arg0, %arg1) : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>
}

func.func @realDiv(%arg0: tensor<8x16xf32>, %arg1: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.RealDiv"(%arg0, %arg1) : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>

}

// -----

module {
func.func private @func_20_GPU_FLOAT(%arg0: tensor<128x128xf32>, %arg1: tensor<3xi32>) -> tensor<1x128x128xf32> attributes {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_20"} {
  %0 = "tfl.reshape"(%arg0, %arg1) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<128x128xf32>, tensor<3xi32>) -> tensor<1x128x128xf32>
  func.return %0 : tensor<1x128x128xf32>
}

func.func @partitioned_3d(%arg0: tensor<!tf_type.resource<tensor<16x8x16xf32>>>, %arg1: tensor<!tf_type.resource<tensor<16x8x16xf32>>>, %arg2: tensor<!tf_type.resource<tensor<16x8x16xf32>>>, %arg3: tensor<!tf_type.resource<tensor<16x8x16xf32>>>)...

  func.func @main(%arg0: tensor<128x10xf32>, %arg1: tensor<10x1024xf32>, %arg2: tensor<128x1024xf32>) -> (tensor<128x10xf32> {mhlo.sharding = "\08\03\1A\02\01\02\22\02\00\01"}, tensor<10x1024xf32> {mhlo.sharding = "\08\01\1A\01\01\22\01\00"}, tensor<128x1024xf32> {mhlo.sharding = ""}) {
    func.return %arg0, %arg1, %arg2 : tensor<128x10xf32>, tensor<10x1024xf32>, tensor<128x1024xf32>
  }
}

  func.return %1 : tensor<128x32xf32>
  // CHECK: %[[FULLY_CONNECTED:.*]] = "tfl.fully_connected"(%arg0, %arg1, %arg2) <{fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"}> : (tensor<128x64xf32>, tensor<32x64xf32>, tensor<32xf32>) -> tensor<128x32xf32>
  // CHECK: return %[[FULLY_CONNECTED]] : tensor<128x32xf32>
}

  // CHECK-LABEL: func @parallel_execute_with_tiled_input
  // CHECK-SAME: (%[[ARG_0:[a-z0-9]*]]: tensor<128x10xf32>, %[[ARG_1:[a-z0-9]*]]: tensor<128x10xf32>, %[[ARG_2:[a-z0-9]*]]: tensor<*xi32>, %[[ARG_3:[a-z0-9]*]]: tensor<*xi32>)
  func.func @parallel_execute_with_tiled_input(%arg0: tensor<128x10xf32>, %arg1: tensor<128x10xf32>, %arg2: tensor<*xi32>, %arg3: tensor<*xi32>) -> (tensor<*xi32>, tensor<*xi1>) {
    // CHECK: tf_device.replicate

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

  // CHECK: return %0
}

// CHECK-LABEL: testAddOfNegRight
func.func @testAddOfNegRight(%arg0: tensor<8x16xf32>, %arg1: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Neg"(%arg1) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  %1 = "tf.Add"(%arg0, %0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>

}

// CHECK-LABEL: softmax
func.func @softmax(%arg0: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Softmax"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
// CHECK: %[[SOFTMAX_0:.*]] = "tf.Softmax"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
// CHECK: return %[[SOFTMAX_0]] : tensor<8x16xf32>
}

// CHECK-LABEL: conv2d_backprop_input_with_add