Added time measurement output #1322
Conversation
|
Jenkins: OGS-6/Linux-PRs failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs/2377/ |
|
Jenkins: OGS-6/Linux-PRs-dynamic failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs-dynamic/601/ |
|
Jenkins: OGS-6/Linux-PRs-dynamic failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs-dynamic/602/ |
|
Jenkins: OGS-6/Linux-PRs failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs/2378/ |
| auto const start_time_sys = std::chrono::system_clock::to_time_t( | ||
| std::chrono::system_clock::now()); | ||
| std::ostringstream sstr; | ||
| sstr << std::put_time(std::localtime(&start_time_sys), "%F %T %z"); |
There was a problem hiding this comment.
% before T and Z display in red colour, but the expression looks nothing wrong.
|
👍 |
|
However, I think it would be nice to print the accumulated times for assembly, linear solver. To measure the scalability of a parallel computing, we often need these times. |
| } | ||
| } | ||
|
|
||
| std::chrono::steady_clock::now(); |
There was a problem hiding this comment.
Since the return value is not stored here I wonder if this line is necessary? ✅
|
I do not know if it would be a good idea to implement a new loging macro TIME into logog for timing output. What do you think? |
|
After extracting code to BaseLib: 👍 |
|
@wenqing: I'd prefer not to accumulate in ogs. I think such accumulation can be easily done by grepping the measured times from the log and summing them up in a script. |
|
@TomFischer: Since the logog macros and logog log line prefixes are defined in macro.hpp and formatter.cpp in that third party package I don't want to introduce a custom TIME log macro now. |
|
In the last commit I added functionality that prints the OGS version in every run. |
| * Formats the given time point according to RFC 3339 (cf. man-page of the unix | ||
| * date utility). | ||
| * | ||
| * Example: 2006-08-14 02:34:56-06:00 |
There was a problem hiding this comment.
I'd prefer not to print the year and the timezone, which are very unlikely to change in a single run...
There was a problem hiding this comment.
I'd rather keep both of them. 08-14 02:34:56 doesn't look like a real timestamp, does it? 😄
|
Jenkins: OGS-6/Linux-PRs failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs/2435/ |
|
OpenGeoSys development has been moved to GitLab. |
Sorry @wenqing that we did not respond to your time measurement PR #863. But I think time measurement is important, so here is my attempt on it.
This PR provides wall-time measurement of assembly, linear solvers, nonlinear iterations, dirchlet BCs timesteps, output, and overall runtime.
The time is measured for individual "events" and not accumulated over timesteps.