Closed GiovanniBussi closed 4 months ago
Attention: Patch coverage is 1.38889%
with 71 lines
in your changes are missing coverage. Please review.
Project coverage is 84.12%. Comparing base (
923d41b
) to head (5a571e7
).:exclamation: Current head 5a571e7 differs from pull request most recent head 6358441. Consider uploading reports for the commit 6358441 to get more accurate results
Files | Patch % | Lines |
---|---|---|
src/cltools/Benchmark.cpp | 1.38% | 71 Missing :warning: |
:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
This is to add the possibility to simultaneously test multiple kernels with the benchmark application.
In short, one can do
to test a specific kernel. Alternatively
to test two (or more) kernels. Test is alternating from one kernel to the other in random order at every step so as to make the timing homogeneous. The idea is that possible system overloading will affect equally both kernels, and the comparison will be more fair.
I attach a notebook where I show that in this manner timings are reproducible. To do so I use
to compare the kernel of the current PLUMED version with itself. Since it's the same kernel, the speed should be the same. From the figure it can be seen that the timings are significantly more homogeneous when we use the multiple kernel option rather than if we compare two separate runs.
timings.zip