Innovative Computing Laboratory, University of Tennessee

 ______ _______ _____   _______ _______ ______
|   __ \   |   |     |_|     __|   _   |   __ \
|    __/   |   |       |__     |       |      <
|___|  |_______|_______|_______|___|___|___|__|

Parallel Ultra Light Systolic Array Runtime

http://icl.utk.edu/pulsar/

Deferred

  • Provide control to the user over PRT parameters, which are currently hardcoded constants:

    • PRT_PROXY_MAX_RECVS_PER_AGENT (prt_proxy.h),

    • PRT_PROXY_MAX_SENDS_PER_AGENT (prt_proxy.h),

    • PRT_PROXY_MAX_TAGS_PER_NODE (prt_proxy.h),

    • PRT_VSA_GPU_ALLOC_UNIT_SIZE (prt_vsa.h),

    • PRT_VSA_MAX_VDPS_PER_NODE (prt_vsa.h).

  • Enable direct device-to-device communication (currently implemented, but not exposed).

  • Possibly, put device packet allocation in the VDP’s stream instead of executing it immediately (potential reduction of device memory usage).