Such an anonymous variable is referred to via a (possibly implicit)
<a href="#Address_operators">pointer indirection</a>.
</p>

<p>
<i>Structured</i> variables of <a href="#Array_types">array</a>, <a href="#Slice_types">slice</a>,
and <a href="#Struct_types">struct</a> types have elements and fields that may
be <a href="#Address_operators">addressed</a> individually. Each such element
acts like a variable.
</p>

<p>

Such an anonymous variable is referred to via a (possibly implicit)
<a href="#Address_operators">pointer indirection</a>.
</p>

<p>
<i>Structured</i> variables of <a href="#Array_types">array</a>, <a href="#Slice_types">slice</a>,
and <a href="#Struct_types">struct</a> types have elements and fields that may
be <a href="#Address_operators">addressed</a> individually. Each such element
acts like a variable.
</p>

<p>

  auto slice_shape = llvm::to_vector<8>(buffer_type.getShape());
  slice_shape[0] = 1;
  auto slice_type = tensorflow::GetTypeFromTFTensorShape(
      slice_shape, buffer_type.getElementType());
  auto update_slice = element;
  if (element.getType() != slice_type) {
    update_slice = builder.create<TF::ReshapeOp>(
        loc, ArrayRef<Type>{slice_type},
        ArrayRef<Value>{element, GetR1Const(slice_shape, builder, loc)});
  }

// compiler will generate only a single runtime.slicetype data
// structure for all 3 underlying types. It turns out the compiler
// generates just the 251-entry one. There aren't any
// runtime.slicetypes generated for the final two types.
//
// The compiler passes type:[]...[]<...> (251 total "[]") to
// fmt.Sprintf (instead of the correct 253 one). But the data

        slice_size.push_back(-1);
      }
    }

    auto slice_type = mlir::RankedTensorType::get(slice_begin.size(),
                                                  rewriter->getIntegerType(32));
    auto slice_begin_attr =
        mlir::DenseIntElementsAttr::get(slice_type, slice_begin);
    auto slice_size_attr =
        mlir::DenseIntElementsAttr::get(slice_type, slice_size);

    auto slice_begin_const =

                          /*keep_slice_shape=*/true);
    // Add a size-1 leading dimension to elem.
    auto slice_type = original_elem.getType().cast<RankedTensorType>();
    elem = builder.create<TF::ReshapeOp>(
        write.getLoc(), ArrayRef<Type>{slice_type},
        ArrayRef<Value>{elem, cutil::GetR1Const(slice_type.getShape(), builder,
                                                write.getLoc())});
    elem =

    int64_t slice_size = input_dim_size / num_splits;

    // Get each slice's type.
    auto slice_shape = llvm::to_vector<4>(input_type.getShape());
    slice_shape[dim_index] = slice_size;
    Type slice_type = tensorflow::GetTypeFromTFTensorShape(
        slice_shape, input_type.getElementType());

    // Parameters for constructing each slice.
    SmallVector<int64_t, 4> begin_indices(input_rank, 0);

switch(slice[0]){case HistogramBinBoundaries.SLICE_TYPE.LINEAR:binBoundaries.addLinearBins(slice[1],slice[2]);break;case HistogramBinBoundaries.SLICE_TYPE.EXPONENTIAL:binBoundaries.addExponentialBins(slice[1],slice[2]);break;default:throw new Error('Unrecognized HistogramBinBoundaries slice type');}}