v0.3.0

What's Changed

• Fix and test copyto! by @jjgarzella in #3
• Max branch by @MaximusMellberg in #2
• Max gpu karatsuba by @MaximusMellberg in #5
• Fixed inverse, backsub, and forwardsub. by @rybplayer in #4
• Add wide option for stripe mul by @rybplayer in #9
• Updated Karatsuba type to work with CuModMatrices by @MaximusMellberg in #6
• Refactor Permutation, PLUQ, and Inverse by @rybplayer in #10
• Jjg/karatsuba test rebase by @jjgarzella in #12

New Contributors

• @MaximusMellberg made their first contribution in #2
• @rybplayer made their first contribution in #4

Full Changelog: v0.2.5...v0.3.0

v0.2.5

Fix rref/pluq functionality

Stop allocating on the cpu in the constructor

v0.2.4

Bug fixes and added tests.

v0.2.3

Merge and fix many tests. Now, RREF should be working.

v0.2.2

Add fill! and zero!, fix bugs that weren't caught due to a few tests not being enabled.

v0.2.1

More tests and debugging for the matrix type, which is now called CuModMatrix. Also implemented a new mod N matmul that breaks the matrix into "stripes" if the matmul would require too many operations for the datatype.

v0.2.0

Second alpha release of the package. Currently the package offers:

1. GpuMatrixModN, a CuArray wrapper that pads to 32, has mod N built in, and has various matrix operations implemented as GPU kernels. This is a refactor of GpuFiniteFieldMatrices from v0.1.0.
2. KaratsubaMatrices, a matrix type implementing Karatsuba matrix multiplication.

Various miscellaneous improvements have been made since v0.1.0, namely

1. The introduction of a dedicated test quite
2. Better docstrings
3. New matrix functions

v0.1.0

Initial release of the package. Currently the package provides a CuArray wrapper type for matrices $$\text{mod } N$$, with

• matrix multiplication
• row reduction
• PLUQ decomposition
• Hensel lifting for pseudoinverses

all on the GPU.