":pjrt_device_compiler_client",
        ":xla_device_compiler_client",
        ":xla_device_context",
        "//tensorflow/core/tfrt/common:create_pjrt_client_util",
        "//tensorflow/core/tfrt/common:global_state",
        "//tensorflow/core/tfrt/common:pjrt_util",
        "//tensorflow/core/tpu:tpu_defs",
        "@com_google_absl//absl/algorithm:container",
        "@com_google_absl//absl/log",

func.func @test(%ch: !tfrt.chain, %arg0: !corert.tensorhandle, %arg1_th: !corert.tensorhandle) {
  %cpu = corert.get_op_handler %ch "cpu"
  %0 = corert.executeop(%cpu) "tf.Relu"(%arg0) { T = f32 } : 1
  %arg1 = tfrt_fallback_async.corert_tensorhandle_to_fallback_tensor %arg1_th {_tfrt_cost = 1 : i64, device = "/CPU:0"} : (!corert.tensorhandle) -> (!tfrt_fallback.tf_tensor)
  %1 = tfrt_fallback_async.executeop key(0) cost(100) device("/CPU:0") "tf.Relu"(%arg1) { T = f32 } : 1

    test_file_exts = ["mlir"],
)

# Bundle together all of the test utilities that are used by tests.
filegroup(
    name = "test_utilities",
    testonly = True,
    data = [
        "//tensorflow/compiler/mlir:tf-mlir-translate",
        "//tensorflow/compiler/mlir:tf-opt",
        "//tensorflow/compiler/mlir/tfrt:tf-tfrt-opt",
        "@llvm-project//llvm:FileCheck",

#ifdef TFRT_GPU_OPS
#else
#define TFRT_GPU_OPS

include "tfrt/tfrt_op_base.td"
include "tfrt/compiler/opdefs/tfrt_op_interfaces.td"
include "tfrt/compiler/opdefs/tfrt_traits.td"
include "tfrt_fallback.td"

def TFRT_GPU_Dialect : Dialect {
  let name = "gpurt";

  let description = [{
    The TFRT GPU Dialect.
  }];

  let cppNamespace = "::tfrt::gpu";
}

class Gpu_Op<string mnemonic, list<Trait> traits = []> :

#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/framework/types.h"
#include "tensorflow/core/tfrt/common/async_value_tensor.h"
#include "tensorflow/core/tfrt/common/global_state.h"
#include "tensorflow/core/tfrt/common/pjrt_state.h"
#include "tensorflow/core/tfrt/common/pjrt_util.h"
#include "tsl/platform/errors.h"
#include "tsl/platform/statusor.h"

namespace tensorflow {

#include "absl/strings/string_view.h"
#include "mlir/IR/BuiltinTypes.h"  // from @llvm-project
#include "tensorflow/compiler/mlir/tfrt/translate/tfrt_compile_options.h"
#include "tensorflow/core/platform/status.h"
#include "tfrt/bef/bef_buffer.h"  // from @tf_runtime
#include "tfrt/core_runtime/tensor_handle.h"  // from @tf_runtime

namespace tfrt {
class CoreRuntime;
}

namespace mlir {
class ModuleOp;
}

namespace tensorflow {

#include "mlir/IR/PatternMatch.h"
#include "tensorflow/compiler/mlir/tfrt/runtime_fallback/runtime_fallback_ops.h"

// This optimizes the following scenario:
// %tft0, %c2 = "tfd.move_dht_to_tft"(%dht0, %c1)
//     : (!dht.host_tensor, !tfrt.chain) -> (!tfd.tf_tensor, !tfrt.chain)
// %dht1, %c3 = "tfd.convert_tft_to_dht"(%tft0, %c2)
//     : (!tfd.tf_tensor, !tfrt.chain) -> (!dht.host_tensor, !tfrt.chain)
// some_op %dht1, %c3
//
// becomes

limitations under the License.
==============================================================================*/

#include "tensorflow/compiler/mlir/tfrt/runtime_fallback/runtime_fallback_ops.h"

#include "tfrt/basic_kernels/opdefs/types.h"  // from @tf_runtime

namespace mlir {
namespace tfd {

//===----------------------------------------------------------------------===//
// TfrtDelegate Dialect

e/runtime_fallback/test/testdata,tensorflow/core/tfrt/stubs,tensorflow/core/tfrt/tfrt_session,tensorflow/core/tfrt/mlrt,tensorflow/core/tfrt/mlrt/attribute,tensorflow/core/tfrt/mlrt/kernel,tensorflow/core/tfrt/mlrt/bytecode,tensorflow/core/tfrt/mlrt/interpreter,tensorflow/compiler/mlir/tfrt/translate/mlrt,tensorflow/compiler/mlir/tfrt/translate/mlrt/testdata,tensorflow/core/tfrt/gpu,tensorflow/core/tfrt/run_handler_thread_pool,tensorflow/core/tfrt/runtime,tensorflow/core/tfrt/saved_model,tensorf...

// RUN: tfrt_fallback_translate -mlir-to-bef %s | tf_bef_executor --work_queue_type=mstd 2>&1 | FileCheck %s

func.func @test_async_op_kernel_thread() -> !tfrt.chain {
  %ch0 = tfrt.new.chain

  %ch1 = tfrt_fallback_async.createop(%ch0) key(0) device("/CPU:0") "tf.Const"()
         { dtype = i32, value = dense<[2]> : tensor<1xi32> } num_args(0)
  %ch2 = tfrt_fallback_async.createop(%ch1) key(1) device("/CPU:0")