Handle multiple GPUs, avoid NVCC/GCC compiler warnings

gunrock / graphblast

High-Performance Linear Algebra-based Graph Primitives on GPUs

Apache License 2.0

211 stars 27 forks source link

Handle multiple GPUs, avoid NVCC/GCC compiler warnings #10

Closed BenBrock closed 4 years ago

BenBrock commented 4 years ago

Descriptors now create a moderngpu context on the current device, instead of always using device 0.

Added GrB_NOT_IMPLEMENTED and added a number of missing return statements.

ctcyang commented 4 years ago

Thanks for the pull request!

Be aware that this will still results in a double cudaFree when a + b are aliased to A + B and their underlying GPU memory is freed, so I've added an issue to track the fix for that (#11).