func.func @main() -> (tensor<5x11xf32>) {
      %arg0 = "tf.Const"() {value = dense<-3.0> : tensor<5x7xf32>} : () -> tensor<5x7xf32>
      %arg1 = "tf.Const"() {value = dense<-3.0> : tensor<11x7xf32>} : () -> tensor<11x7xf32>

      %1 = "tf.MatMul"(%arg0, %arg1) {transpose_a = false, transpose_b = true} : (tensor<5x7xf32>, tensor<11x7xf32>) -> tensor<5x11xf32>

      func.return %1 : tensor<5x11xf32>
    }

  // CHECK-NEXT:   %[[ID:.*]] = "tf_device.launch"() <{device = "TPU_REPLICATED_HOST_0"}> ({
  // CHECK-NEXT:     %[[IDINSIDE:.*]] = "tf.Identity"(%[[REPVAR]]) {ici_weight_distribution_mlir_bridge_marker = true} : (tensor<f32>) -> tensor<f32>
  // CHECK-NEXT:     tf_device.return %[[IDINSIDE]] : tensor<f32>
  // CHECK-NEXT:   }) : () -> tensor<f32>
  // CHECK-NEXT:   "tf_device.cluster"() ({

#include "mlir/Dialect/Quant/QuantOps.h"  // from @llvm-project
#include "mlir/Dialect/SCF/IR/SCF.h"  // from @llvm-project
#include "mlir/Dialect/Shape/IR/Shape.h"  // from @llvm-project
#include "mlir/Dialect/Tensor/IR/Tensor.h"  // from @llvm-project
#include "mlir/InitAllPasses.h"  // from @llvm-project
#include "mlir/Support/LogicalResult.h"  // from @llvm-project
#include "mlir/Tools/mlir-opt/MlirOptMain.h"  // from @llvm-project

// This method will actually merge the information contained in the
// types, it is capable of refining:
//   tensor<!tf_type.variant<tensor<?x8xf32>>>
// and:
//   tensor<!tf_type.variant<tensor<10x?xf32>>>
// into:
//   tensor<!tf_type.variant<tensor<10x8xf32>>>
//
// In case of inconsistencies (rank disagreement for example), it returns `lhs`.
Type TypeMeet(Type lhs, Type rhs) {

  return absl::OkStatus();
}

int MiniDynamicBuffer::WriteToBuffer(char** buffer) {
  // Allocate sufficient memory to tensor buffer.
  int32_t num_strings = offset_.size() - 1;
  // Total bytes include:
  //   * size of content (data_.size)
  //   * offset of each tensor (sizeof(int32_t) * num_strings)
  //   * length of whole buffer (int32_t)
  //   * num of strings (int32_t).

// the module.
std::unique_ptr<OperationPass<ModuleOp>> CreateStripNoinlineAttributePass();

// Converts tensor list operations into operations on buffers and sizes. Needs
// static shapes and known max element count.
std::unique_ptr<OperationPass<ModuleOp>> CreateTensorListOpsDecompositionPass();

// Converts tensor array ops into operations on local variables, which can later