auto input = block->getArgument(0);
    auto size = block->getArgument(3);

    // Subtract `input_rank` by 1 to get the item's rank, which is used as
    // `partial_position_shape`.
    auto input_rank = rewriter->create<TF::RankOp>(
        loc, tensorflow::GetTypeFromTFTensorShape({}, shape_dtype), input);
    auto partial_position_shape = rewriter->create<TF::SubOp>(

// in handling tf.If op which would end up with excessive host-device
// communication.

// In this test, all ops other than tf.Rank are marked with outside_compilation
// . So the TPU program should contain tf.Rank op and there should be no
// host-device communication.
// CHECK: tf._TPUCompileMlir
// CHECK-SAME: tf.Rank
// CHECK-NOT: tf._XlaHostComputeMlir
// CHECK-NOT: tf._XlaRecvAtHost
// CHECK-NOT: tf._XlaSendFromHost

           << tensor_type << ".";
  }

  int rank = tensor_type.getRank();
  if (rank <= partition_dim) {
    return op->emitError() << "cannot partition " << first_operand_type
                           << " (rank = " << rank << ") along dimension "
                           << partition_dim << ".";
  }

  llvm::SmallVector<int64_t, 4> partition_dims(is_replicated ? 0 : rank, 1);
  if (!is_replicated) {

  if (!input_type) return success();
  int64_t rank = input_type.getRank();

  DenseIntElementsAttr dims_attr;
  if (!matchPattern(dims, m_Constant(&dims_attr))) return success();
  for (const auto &dim_pair : llvm::enumerate(dims_attr)) {
    int64_t cur_dim = dim_pair.value().getSExtValue();
    if (cur_dim < -rank || cur_dim >= rank)
      return emitError(loc)

type byEvictionPriority []evictionapi.Threshold

func (a byEvictionPriority) Len() int      { return len(a) }
func (a byEvictionPriority) Swap(i, j int) { a[i], a[j] = a[j], a[i] }

// Less ranks memory before all other resources, and ranks thresholds with no resource to reclaim last
func (a byEvictionPriority) Less(i, j int) bool {
	_, jSignalHasResource := signalToResource[a[j].Signal]

// Helper function for reduction ops.
//
// input_shape: 1-D Tensor, the shape of the Tensor being reduced.
// axes: 1-D Tensor, the reduction axes.
//   Note that the reduction indices are in the range
//   -rank(input_shape), rank(input_shape)
// returns a 1-D Tensor, the output shape as if keep_dims were set to True.
Output ReducedShapeHelper(const Scope& scope, const Output& input_shape,
                          const Output& reduction_axes);

  clean

  // Ensure ordering
  compileDebugSwift -> compileReleaseSwift -> clean [style=invis]
  {rank=same compileDebugSwift compileReleaseSwift clean}

  assemble -> empty1 -> empty2 -> check [style=invis]
  {rank=same assemble check empty1 empty2}
  empty1[shape=plain, label=""]
  empty2[shape=plain, label=""]

  compileReleaseCpp -> linkRelease -> assembleRelease [dir=back]
  clean

  // Ensure ordering
  compileDebugCpp -> compileReleaseCpp -> clean [style=invis]
  {rank=same compileDebugCpp compileReleaseCpp clean}

  assemble -> empty1 -> empty2 -> check [style=invis]
  {rank=same assemble check empty1 empty2}
  empty1[shape=plain, label=""]
  empty2[shape=plain, label=""]

  clean

  // Ensure ordering
  compileDebugSwift -> compileReleaseSwift -> clean [style=invis]
  {rank=same compileDebugSwift compileReleaseSwift clean}

  assemble -> empty1 -> empty2 -> check [style=invis]
  {rank=same assemble check empty1 empty2}
  empty1[shape=plain, label=""]
  empty2[shape=plain, label=""]

    }

    /** Equivalent to {@code newReentrantReadWriteLock(rank, false)}. */
    public ReentrantReadWriteLock newReentrantReadWriteLock(E rank) {
      return newReentrantReadWriteLock(rank, false);
    }

    /**
     * Creates a {@link ReentrantReadWriteLock} with the given fairness policy and rank. The values