Add SubView class

[gberg617]

Summary

• This WIP PR adds the SubView class. This class stores a view-type object and a tuple of slices that indicate how to access the view's underlying data. The following slice types are supported:
  • range slice: Indexes a subset of the original dimension's extents. Does not reduce dimensionality.
  • fixed slice: Indexes along a fixed value of a given dimension. Reduces dimensionality.
  • no slice: Indexes according to the original view's layout along the given dimension. Does not reduce dimensionality.

A few simple examples are provided below:

    View<Index_type, Layout<2>> view(&my_data[0][0], Layout<2>(3,3));

    // "sv1 = View[1:3,:]"
    auto sv1 = SubView(view, RangeSlice{1,3}, NoSlice{});

    // "sv2 = View[1:3,1]"
    auto sv2 = SubView(view, RangeSlice{1,3}, FixedSlice{1});

Here, sv1 and sv2 are subviews of dimensions (2,3) and (2) respectively.

Future tasks:

• Add Slices whose extents are determined at compile time
• Add Slice types that use non-affine mappings (e.g. index lists)
• Add ability to collapse nested sub-layouts into a single sub-layout for slices that have affine mappings
• Add ability to distinguish between projected dimension and an empty slice (e.g. RangeSlice{2,2})

[gberg617]

This PR is still a WIP!

[gberg617]

I'm considering separating the SubView holding the view from the slices that are used to access the view, somewhat analogous to the approach of having layouts and Views be separate classes. Any thoughts on this?

[rhornung67]

@gberg617 we changed the copyright header in all RAJA files so when we update each year we only need to touch two files rather than the whole repo. Here is what you should have on all files you add or modify:
https://github.com/llnl/RAJA/blob/develop/test/unit/view-layout/test-indexlayout.cpp#L1

[rhornung67]

See comment about copyright header.

[gberg617]

@MrBurmark @adayton1 I assume that SubViews can be 0D in certain cases where the slice(s) reduce the dimension to a scalar. This leads to compiler errors on Windows (see failing CI) because camp array stores a C-style array of size 0. My thoughts are to either disallow 0D SubViews or add a specialization of camp array for a 0-sized array. My preference is the latter because std::array can be zero-sized, so this addition would make camp::array be more consistent with std::array. Any thoughts?

[MrBurmark]

@MrBurmark @adayton1 I assume that SubViews can be 0D in certain cases where the slice(s) reduce the dimension to a scalar. This leads to compiler errors on Windows (see failing CI) because camp array stores a C-style array of size 0. My thoughts are to either disallow 0D SubViews or add a specialization of camp array for a 0-sized array. My preference is the latter because std::array can be zero-sized, so this addition would make camp::array be more consistent with std::array. Any thoughts?

Adding a camp array specialization for 0-size makes a lot of sense.

[adayton1]

@MrBurmark @adayton1 I assume that SubViews can be 0D in certain cases where the slice(s) reduce the dimension to a scalar. This leads to compiler errors on Windows (see failing CI) because camp array stores a C-style array of size 0. My thoughts are to either disallow 0D SubViews or add a specialization of camp array for a 0-sized array. My preference is the latter because std::array can be zero-sized, so this addition would make camp::array be more consistent with std::array. Any thoughts?

Adding a camp array specialization for 0-size makes a lot of sense.

I can't remember for sure, but I think that I disallowed the 0 size array because nvcc couldn't handle it (or maybe it was some other compiler). So just do plenty of testing.