hoomd.util¶
Overview
hoomd.util.cuda_profile_start |
Start CUDA profiling. |
hoomd.util.cuda_profile_stop |
Stop CUDA profiling. |
hoomd.util.quiet_status |
Quiet the status line output. |
hoomd.util.unquiet_status |
Resume the status line output. |
Details
Utilities.
-
hoomd.util.
cuda_profile_start
()¶ Start CUDA profiling.
When using nvvp to profile CUDA kernels in hoomd jobs, you usually don’t care about all the initialization and startup. cuda_profile_start() allows you to not even record that. To use, uncheck the box “start profiling on application start” in your nvvp session configuration. Then, call cuda_profile_start() in your hoomd script when you want nvvp to start collecting information.
Example:
from hoomd import * init.read_xml("init.xml"); # setup.... run(30000); # warm up and auto-tune kernel block sizes option.set_autotuner_params(enable=False); # prevent block sizes from further autotuning cuda_profile_start(); run(100);
-
hoomd.util.
cuda_profile_stop
()¶ Stop CUDA profiling.
See also
-
hoomd.util.
quiet_status
()¶ Quiet the status line output.
After calling
hoomd.util.quiet_status()
, hoomd will no longer print out the line of code that executes each hoomd script command. Callhoomd.util.unquiet_status()
to enable the status messages again. Messages are only enabled after a number ofhoomd.util.unquiet_status()
calls equal to the number of priorhoomd.util.quiet_status()
calls.
-
hoomd.util.
unquiet_status
()¶ Resume the status line output.
See also