# limitations under the License.

# shellcheck disable=SC2034 # Variables sourced in other scripts.

# A set of helpers for tests

reset=$(tput sgr0)
bold=$(tput bold)
black=$(tput setaf 0)
red=$(tput setaf 1)
green=$(tput setaf 2)
readonly reset bold black red green

kube::test::clear_all() {
  if kube::test::if_supports_resource "rc" ; then
    # shellcheck disable=SC2154

	if err != nil && err != io.ErrUnexpectedEOF {
		return err
	}

	return nil
}

func osCPUInfoName() string {
	modelName := ""
	cpuMHz := ""

	// The 512-byte buffer is enough to hold the contents of CPU0
	buf := make([]byte, 512)
	err := readLinuxProcCPUInfo(buf)
	if err != nil {
		return ""
	}

	scanner := bufio.NewScanner(bytes.NewReader(buf))
	for scanner.Scan() {

          "tf.opA"() : () -> ()
          tf_device.return
        }) {device = "/TPU:2"} : () -> ()
        tf_device.return
      }
      tf_executor.yield
    }
    tf_executor.fetch
  }
  func.return
}

// CHECK: "tf_device.launch"
// CHECK: device = "/TPU:2"
// CHECK: "tf.opA"
// CHECK: "tf_device.launch"
// CHECK: device = "/TPU:2"
// CHECK: "tf.opA"

inline constexpr llvm::StringRef kReplicationInfoAttr = "_replication_info";
// Marks a node for XLA-TPU compilation. The attribute value indicates the
// associated compilation cluster and replication metadata op.
inline constexpr llvm::StringRef kTpuReplicateAttr = "_tpu_replicate";
// Device types.
inline constexpr llvm::StringRef kTpuDevice = "TPU";
// _xla_outside_compilation
inline constexpr llvm::StringRef kXlaOutsideCompilationAttr =

#include "mlir/IR/Value.h"  // from @llvm-project
#include "mlir/Support/LogicalResult.h"  // from @llvm-project
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_device.h"
#include "xla/xla_data.pb.h"
#include "tensorflow/core/protobuf/tpu/compile_metadata.pb.h"

namespace tensorflow {

inline constexpr llvm::StringRef kInputShardingAttr =
    "input_sharding_configuration";
inline constexpr llvm::StringRef kOutputShardingAttr =

// RUN: not tf-mlir-translate -mlir-to-graphdef %s -o - 2>&1 | FileCheck %s

// CHECK: entry function `main` must be present

func.func @const() {
  tf_executor.graph {
    %0:2 = tf_executor.island wraps "tf.Const"() {device = "TPU:0", name = "const", dtype = "tfdtype$DT_INT32", value = dense<[1, 2]> : tensor<2xi32>} : () -> tensor<2xi32>
    tf_executor.fetch
  }
  func.return

    });
    if (replicate)
      reshard_was_inserted |= HandleReplicateOp(while_op, replicate);
  });
  if (reshard_was_inserted)
    VLOG(1) << "tf-tpu-variable-runtime-reformatting inserted at least one "
               "TPUReshardVariables";
  else
    VLOG(1) << "tf-tpu-variable-runtime-reformatting inserted no "
               "TPUReshardVariables";
}

}  // namespace

std::unique_ptr<OperationPass<ModuleOp>>

    needed to apply space to depth optimization for a model that needs to load a
    pre-train checkpoint. This transform can be done on the host or TPU device
    based on the cost. As the size of the kernel is relatively small, this won't
    add additional cost to TPU device time. Below is the logic to transform the
    kernel of shape [7, 7, 3, 64] to [4, 4, 12, 84].

    ```python
    conv0 = tf.compat.v1.layers.Conv2D(

  func.func @main() {

                        std::vector<int64_t>(),
                        "number of tasks from available TPU devices must be "
                        "'num_tasks' in 'topology' (1), got 2"),
        std::make_tuple(2, 1, TopologyWithMeshShapeAndTasks({1, 1, 1, 1}, 2, 2),
                        std::vector<int64_t>(),
                        "number of TPU devices available per task must be "