v0.8.0

• Improvements to mapreduce kernel
• addition of a generic matmul kernel
• support in broadcasting for one-layer deep lazy wrappers like Transpose, Adjoint and SubArray.
• addition of a @time macro that reports GPU allocations.

v0.7.3: Merge pull request #119 from JuliaGPU/omm/broadcasted

Use `broadcasted` instead of `preprocess`

v0.7.2

update readme

v0.7.1: Merge pull request #112 from JuliaGPU/tb/move_tests

Improve tests

v0.7.0

rm project.toml for now

v0.6.2

dilation kwarg fix

v0.6.1

Add pool allocator management, releasing unused memory after a while (see #83 for more info on the design). This should not impact performance, while reducing idle-time memory usage and lowering GC pressure.

In the case of performance problems, run your code with CUARRAYS_MANAGED_POOL set to respectively true and false, and report the statistics as printed at the end of the session.

v0.6.0

Import of CURAND.jl

v0.5.0

fix activation functions

v0.4.0

pooling