DenseIntElementsAttr DenseI64AttrToI32Attr(
    const DenseIntElementsAttr& dense_attr, PatternRewriter& builder);

// Returns a NHWC shaped type from an NCHW shaped type op.
// For example- Given a Composite op that wraps a core.aten.avg_pool2d, this
// returns the return type of the tfl.average_pool_2d emitted. Note that the
// aten.avg_pool2d works with the NCHW layout while tfl.average_pool_2d assumes
// NHWC.

#define TENSORFLOW_COMPILER_JIT_XLA_HOST_RECV_DEVICE_CONTEXT_H_

#include "xla/shape.h"
#include "xla/stream_executor/device_memory.h"
#include "xla/stream_executor/stream.h"
#include "tensorflow/core/framework/device_base.h"
#include "tfrt/concurrency/async_value_ref.h"  // from @tf_runtime

namespace tensorflow {

// XlaHostRecvDeviceContext is a DeviceContext that is intended to be

#include "tensorflow/compiler/mlir/quantization/tensorflow/exported_model.pb.h"
#include "tensorflow/compiler/mlir/quantization/tensorflow/quantization_options.pb.h"
#include "tensorflow/core/protobuf/meta_graph.pb.h"

namespace tensorflow::quantization {

// Declares pure virtual member functions for a python-side derived class to
// override. This allows calling python implementations from the C++ layer.

absl::Status InitStreamExecutorPlugin(SEInitPluginFn init_fn,
                                      std::string* device_type,
                                      std::string* platform_name);

// This file implements core stream executor base classes in terms of
// the C API defined in stream_executor.h. A class "CSomething" represents a
// "Something" that can be manipulated via calls in the C interface.
class CPlatform : public Platform {

namespace mlir {
namespace TFL {

// A config that controls which passes get run as part TFLite converter.
struct PassConfig {
  explicit PassConfig(quant::QuantizationSpecs specs)
      : quant_specs(std::move(specs)) {}

  // If `emit_builtin_tflite_ops` is true, TF Lite legalization passes will be
  // added, which produces TF Lite ops.
  bool emit_builtin_tflite_ops = true;

// executable.
//
// Returns a non-OK Status if no unambiguous choice of device exists.
//
// We choose the device using the following rules:
//
//  - It is an error for `device_names` to contain more than one device of the
//    same type.
//  - GPU is preferred over CPU.
//  - If `allow_mixing_unknown_and_cpu` is true then unknown devices are
//    preferred over CPU.

#include "mlir/IR/Operation.h"  // from @llvm-project
#include "tensorflow/compiler/mlir/tensorflow/translate/mlir_roundtrip_flags.h"
#include "tensorflow/core/framework/function.h"
#include "tensorflow/core/framework/graph.pb.h"
#include "tensorflow/core/framework/node_def.pb.h"
#include "tensorflow/core/graph/graph.h"

namespace tensorflow {
namespace tf2xla {
namespace v2 {

// Given an MLIR module, returns a GraphDef.

#include "tensorflow/compiler/tf2xla/xla_expression.h"
#include "xla/client/xla_builder.h"
#include "xla/client/xla_computation.h"
#include "xla/mlir_hlo/mhlo/IR/hlo_ops.h"
#include "tensorflow/core/common_runtime/device_mgr.h"
#include "tensorflow/core/framework/op_kernel.h"

namespace mlir {
namespace mhlo {

class Tf2XlaRewriterTestPeer;

class Tf2XlaRewriter {
 public:
  static mlir::LogicalResult RewriteOp(mlir::Operation* op,

#define TENSORFLOW_COMPILER_MLIR_TFRT_BACKEND_COMPILER_H_

#include "absl/status/status.h"
#include "mlir/IR/BuiltinOps.h"  // from @llvm-project
#include "mlir/IR/DialectRegistry.h"  // from @llvm-project
#include "tensorflow/core/tfrt/runtime/runtime.h"

namespace tensorflow {

class BackendCompiler {
 public:
  virtual ~BackendCompiler();

  virtual void GetDependentDialects(mlir::DialectRegistry& registry) const {}

#include "mlir/IR/Types.h"  // from @llvm-project
#include "mlir/IR/Value.h"  // from @llvm-project
#include "tensorflow/compiler/mlir/lite/schema/schema_generated.h"
#include "xla/statusor.h"
#include "tensorflow/core/framework/types.pb.h"

namespace mlir {
class Builder;
}  // namespace mlir

namespace tflite {
// Convert the MLIR type to the corresponding TFLite tensor.