(Memory measured using /usr/bin/time -f "%M" [command] and thus subject to some sampling problems but likely reliable. Running with valgrind would be better but would take forever.)
On my laptop (which has a relatively small cache) this results in a performance improvement - this is not evidenced on the machine I tested on (cspc276), which has a larger cache.
[x] Find the bug which affects the p&r time - probably something in "projects_to".
Rewrite the transmission parameters to reduce code complexity and save memory (on host).
Profiling this for a single instance of a 512-D memory recall benchmark indicates:
(Memory measured using
/usr/bin/time -f "%M" [command]
and thus subject to some sampling problems but likely reliable. Running with valgrind would be better but would take forever.)On my laptop (which has a relatively small cache) this results in a performance improvement - this is not evidenced on the machine I tested on (
cspc276
), which has a larger cache.