Hi Everyone,

Avadh has pushed a patch to the core-models branch that I contributed that
allows the new stats (i.e. not the ptlstats) mechanism to keep track of time
based stats. You can see the commit here:
https://github.com/avadhpatel/marss/compare/a00c3df705...cb835199ac .

A lot of times I'd like to see how a particular stat is changing over time
instead of just a final snapshot. Just wanted to give everyone a quick tour
of how this works.
First, to enable time-based stats, you should add the filename where you
want these stats to go in your simconfig file. The parameter is called
-time-stats-logfile
This is a simple CSV file, so I usually add:

 -time-stats-logfile foo.csv

to my simconfig file.

All StatObj<T> objects now have a method called set_time_stats(Stat *). This
method is intended to be used with a new global variable called
'n_time_stats'.

An example of how to enable time based stats can be seen
in ptlsim/core/default_core/defcore-stats.h:82 (
https://github.com/avadhpatel/marss/compare/a00c3df705...cb835199ac#L0R82):

complete.set_time_stats(n_time_stats);

Here, I call set_time_stats() on the 'complete' statistic, which will cause
it to dump a the value of the stat every 10k cycles to foo.csv. (10k can
generate rather large files, to change the granularity, see
 ptlsim/sim/machine.cpp:138 )

The resulting foo.csv file looks like this:

sim_cycle,base_machine.ooo_0_0.thread0.frontend.status.complete
0,0
10000,4303
20000,17045
30000,17044

The header contains the full hierarchical name of the statistics (the
simulation cycle number is always first). Then the CSV data follows.

The CSV file is a convenient format for importing into excel, gnuplot, or
using one of the graphing scripts in marss.utils (
https://github.com/dramninjasUMD/marss.utils). Currently, graph_and_send
makes a lot of incorrect assumptions about what things will be named so it
doesn't work out of the box. In the near future we'll be polishing up the
initial versions of these scripts into a more robust set of tools. So stay
tuned for that.

-Paul
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel

Reply via email to