[test] Extend GPU burn test to report GPU node with smallest flops

[jgphpc]

GpuBurnTest

• dom:gpu
  • PrgEnv-gnu
    • num_tasks: 22
    • perf: 4091.0 Gflop/s
    • smallest_flops: 4091.0 GF/s (nid00008)

Goal is to report nidname when the test fails to meet perf. reference
'perf': (4115, -0.10, None, 'Gflop/s')
but I am not sure how to do that... Feel free to push
For now, replacing perf: with the last line gives some info.

[codecov-io]

Codecov Report

Merging #1653 (e3f9de0) into master (4bf3156) will decrease coverage by 0.06%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1653      +/-   ##
==========================================
- Coverage   87.72%   87.65%   -0.07%     
==========================================
  Files          45       45              
  Lines        7477     7485       +8     
==========================================
+ Hits         6559     6561       +2     
- Misses        918      924       +6     

Impacted Files	Coverage Δ
reframe/core/modules.py	58.89% <0.00%> (-0.57%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4bf3156...e3f9de0. Read the comment docs.

[jjotero]

How about adding an optional step after performance with something like self.perf_error_patterns? This would process the output and collect all the information on how and where the test failed. In that way, this additional step would only run if the performance fails, and you would avoid having redundant output when the test passes performance.

[vkarak]

@jjotero Is this a general feature request?

[jgphpc]

you would avoid having redundant output when the test passes performance.

@jjotero
the test always report the smallest flops, I only want to always add the nidname, so the report will not be much different from what we have now:
perf: 4091.0 GF/s (nid00008)
However having self.perf_error_patterns sounds nice.

[jjotero]

@jjotero Is this a general feature request?

I guess it could be. Should I open an issue and move the discussion over there?

[vkarak]

I guess it could be. Should I open an issue and move the discussion over there?

Yes, it would make sense.

[vkarak]

@jgphpc I fixed the PR.