v0.2.0

Release v0.2.0

Both CPU and GPU now construct a simple neighbor list for biases, reducing the performance impact as biases accumulate. The GPU implementation requires a memcpy from the device to the host. Future release will refactor the GPU code to avoid this memcpy.

v0.1.0

Release v0.1.0! 🎉

This marks the first working release of the hoomd-mimse project. This code provides GPU acceleration to the MIMSE protocol, integrated into the HOOMD ecosystem to leverage its large codebase and ease of extensibility.

What needs work

While not perfect, it is ready to start be tested and benchmarked more extensively, prior to being used in research. Some things we still need to get done ...

• Implement neighborlist for biases to reduce penalty for accumulating many biases in the simulation
• Convert particle indexing from tag-based to index-based to improve data-locality (particularly on GPUs)
• Implement RNGs on device for randomKick method
• Migrate to graph and/or dynamic parallelism APIs in some kernel to reduce Host<->Device overhead