search

bespoke-silicon-group / bsg_manycore

Tile based architecture designed for computing efficiency, scalability and generality
Other
221 stars 58 forks source link

barrier perf tests #660

Closed tommydcjung closed 1 year ago

tommydcjung commented 2 years ago

Tests to measure AMOADD, HW, tile-group barrier latencies.

tommydcjung commented 2 years ago

bsg_print_int prints the time, and the diff was taken manually for the interest of time, but I can add some scripts to do a sweep and parsing.

mrutt92 commented 2 years ago

Ah got it. Works differently in cuda. Your call on the script. I would do it if you expect someone else to reproduce or update what you've done.