#define TENSORFLOW_COMPILER_JIT_DEVICE_COMPILER_CLIENT_H_

#include <optional>
#include <string>
#include <variant>

#include "tensorflow/compiler/tf2xla/xla_compiler.h"
#include "xla/client/executable_build_options.h"

namespace tensorflow {

template <typename ExecutableType, typename ClientType>
class DeviceCompilerClient {
 public:
  DeviceCompilerClient() = default;

    const std::vector<TensorShape>& arg_shapes,
    std::vector<tpu::ShardingAndIndex>* arg_core_mapping,
    std::vector<std::vector<xla::Shape>>* per_core_arg_shapes);

// Compiles a serialized MLIR module into XLA HLO, generates all accompanying
// metadata and stores them in CompilationResult.
absl::StatusOr<XlaCompilationResult> LegalizeWithMlirBridge(
    const tpu::MlirToHloArgs& computation,

#include "absl/container/flat_hash_map.h"
#include "absl/numeric/bits.h"
#include "absl/strings/string_view.h"
#include "absl/types/span.h"
#include "tensorflow/compiler/tf2xla/xla_op_registry.h"
#include "xla/status_macros.h"
#include "xla/statusor.h"
#include "tensorflow/core/framework/types.h"

namespace tensorflow {
namespace jit {
class DeviceInfoCache;
class DeviceSet;

        "@com_google_absl//absl/strings",
        "@local_xla//xla/service:hlo_proto_cc",
    ],
)

tf_cc_test(
    name = "device_compiler_serialize_test",
    srcs = [
        "device_compiler_serialize_test.cc",
    ],
    tags = [
        "config-cuda-only",
        "no_oss",  # This test only runs with GPU.
        "requires-gpu-nvidia",
        "xla",
    ],
    deps = [
        ":device_compiler_test_helper",

        "@local_tsl//tsl/platform:status",
        "@local_tsl//tsl/platform:statusor",
        "@local_xla//xla:xla_proto_cc",
        "@local_xla//xla/client:compile_only_client",
        "@local_xla//xla/hlo/ir:hlo",
        "@local_xla//xla/mlir_hlo:hlo_dialect_registration",
        "@local_xla//xla/pjrt:compile_options_proto_cc",
        "@stablehlo//:register",
    ],
)

tf_cc_test(
    name = "legalize_tf_test",

 limitations under the License.
 ==============================================================================*/
// Rewrites computations generated by the xla.compile() Python code into
// XlaLaunch nodes.
//
// xla.compile() does two main things:
// a) marks operators that make up an XLA computation with the attribute
//    _xla_compile_id=XYZ, where XYZ is a unique key.
// b) adds XlaClusterOutput nodes to represent outputs of the computation.

        "@com_google_absl//absl/status",
        "@com_google_googletest//:gtest_main",
        "@local_tsl//tsl/platform:statusor",
        "@local_xla//xla/client:xla_computation",
        "@local_xla//xla/service:hlo_proto_cc",
    ],
)

cc_library(
    name = "mlir_pass_instrumentation",
    srcs = ["mlir_pass_instrumentation.cc"],
    hdrs = ["mlir_pass_instrumentation.h"],
    deps = [

#include "tensorflow/compiler/mlir/tf2xla/internal/utils/test_metadata_config.h"
#include "tensorflow/compiler/tf2xla/xla_compiler.h"
#include "tensorflow/compiler/tf2xla/xla_helpers.h"
#include "xla/client/client_library.h"
#include "xla/stream_executor/platform_manager.h"
#include "tensorflow/core/lib/monitoring/cell_reader.h"
#include "tensorflow/core/lib/monitoring/test_utils.h"
#include "tensorflow/core/platform/env.h"

namespace tensorflow {

// Compiles/lowers a given Tensorflow graph/function/cluster into a compiled XLA
// compilation (HLO) using the XlaCompiler and compiles the resulting
// XlaCompilationResult into an `ExecutableType` (eg. xla::LocalExecutable) by
// calling `ClientType` (eg. xla::LocalClient).
//
// Caches the compiled XlaCompilationResult and Executable using a

    return "quant-replace-cast-hacks-with-tf-xla-ops";
  }
  StringRef getDescription() const final {
    // This is a brief description of the pass.
    return "Replace mixed-type Conv and Matmul cast hacks with TF XLA ops.";
  }

  void runOnOperation() override;
};

// Generates params for the XLA Convolution op.
void PrepareXlaConvParams(OpBuilder &builder, Location loc, ArrayAttr strides,