[cudamapper] Overlap index fetching with overlap computation

mimaric commented 4 years ago

When creating indices generation of one Index is overlapped with transferring of previously generated Index to host memory.

When calculating overlaps while one batch of device indices in being processed the other batch is being loaded into device memory simultaneously.

TODO:

Create pool allocator for pinned host memory: #319
When creating indices in IndexCache synchronize generation and communication streams using events, not cudaStreamSynchronize(): #524
When copying indices from host to device use stream callback functions to update the state indices upon copy completion: #525

Closes #318 Closes #526

mimaric commented 4 years ago

rerun tests

mimaric commented 4 years ago

Addressing #526 in this PR as well

mimaric commented 4 years ago

rerun tests

mimaric commented 4 years ago

rerun tests

mimaric commented 4 years ago

@ahehn-nv This is not the final commit, no need to re-review yet

mimaric commented 4 years ago

rerun tests

mimaric commented 4 years ago

rerun tests

ohadmo commented 4 years ago

rerun tests

mimaric commented 4 years ago

@ahehn-nv While re-reviewing can you also check those few unresolved comments from previous reviews?

NVIDIA-Genomics-Research / GenomeWorks