block-grid gives you a fixed size, two-dimensional array, with a blocked memory representation. This has the sweet benefit of being much more cache-friendly if you're often accessing nearby coordinates.

This library works and is stable, and will not be worked on actively. I am more than happy to review PRs if there's a need :)

Features

• Can store any type
• Generic compile-time block sizes
• Indexing with (row, col): (usize, usize)
• Block level access with Block and BlockMut
• Constructors from row-major and column-major order arrays
• Iterators for in-memory and row-major order, and by block
• no_std and serde support
• Also supports no blocks (i.e. classic row-major)

Example

use block_grid::{BlockGrid, CoordsIterator, U2};

fn main() {
    let data: Vec<_> = (0..(4 * 6)).collect();

    // Construct from row-major ordered data
    let grid = BlockGrid::<usize, U2>::from_row_major(4, 6, &data).unwrap();

    // The 2D grid looks like:
    // +-----------------------+
    // |  0  1 |  2  3 |  4  5 |
    // |  6  7 |  8  9 | 10 11 |
    // |-------+-------+-------|
    // | 12 13 | 14 15 | 16 17 |
    // | 18 19 | 20 21 | 22 23 |
    // +-----------------------+

    // Indexing
    assert_eq!(grid[(1, 3)], 9);

    // Access raw array
    let first_five = &grid.raw()[..5];
    assert_eq!(first_five, &[0, 1, 6, 7, 2]);

    // Iterate over blocks, and access the last
    let block = grid.block_iter().last().unwrap();
    assert_eq!(block[(0, 1)], 17);

    // Iterate in row-major order
    for (i, &x) in grid.row_major_iter().enumerate() {
        assert_eq!(x, i);
    }

    // Iterate in memory order, with coordinates
    for ((row, col), &x) in grid.each_iter().coords() {
        assert_eq!(row * 6 + col, x);
    }
}

Why

TODO: Stuff about caches

Trade-offs

• Non-resizable, and grid dimensions have to be a multiple of the block size.
• Currently, only square blocks, and power-of-two block sizes are supported.
• Computing the modified index takes just a bit more time.
• There are still cache misses when you cross tile boundaries.
• No support for strides or general subsets.

Changelog

See CHANGELOG.md.

License

block-grid is licensed under the MIT license.

Alternatives

If your access patterns suit a typical row-major memory representation, you can still use block-grid! If you truly desire alternatives, however, check out array2d, imgref, grid, or toodee. The last two support dynamic resizing. For matrices and linear algebra, there's also nalgebra.