NVIDIA / cuCollections

Apache License 2.0
486 stars 88 forks source link

[FEA] CUDA stream support #65

Open seunghwak opened 4 years ago

seunghwak commented 4 years ago

Is your feature request related to a problem? Please describe. static_map/dynamic_map currently does not take cudaStream_t stream parameter. This often requires additional synchronization and limits speedup we can get when we wish to concurrently run multiple cuCollection operations using multiple CUDA streams.

Describe the solution you'd like Add CUDA stream support

Additional context cuGraph needs this to run multiple graph kernels concurrently using multiple CUDA streams (for batch processing).

chirayuG-nvidia commented 2 years ago

Stream support for static_map added in commit https://github.com/NVIDIA/cuCollections/commit/76a6ba1d08761f7e14f27c6794ca2d469301b54f