[test] Update OSU microbenchmark test to new syntax and build system

[rsarm]

Closes #327
Closes #390

[codecov-io]

Codecov Report

Merging #442 into master will decrease coverage by 0.02%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #442      +/-   ##
==========================================
- Coverage   91.25%   91.23%   -0.03%     
==========================================
  Files          70       70              
  Lines        8611     8611              
==========================================
- Hits         7858     7856       -2     
- Misses        753      755       +2

Impacted Files	Coverage Δ
reframe/core/config.py	82.72% <0%> (-1.82%)	⬇️

---

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 0e26e8c...5f2ee48. Read the comment docs.

[vkarak]

This and the corresponding latency one are actually G2G tests using CUDA. Name them accordingly, e.g., G2GBandwidthTest.

[rsarm]

To pass the tags on the parameterized test AlltoAllTest, should it be better something like self.tags = set(tags if isinstance(tags, list) else [tags]) to allow passing both ['production'] and [['production']]?

[vkarak]

I didn't mean it like that. The tests should take a single string argument of whether it is production or not. Then you set the tags as usual:

@rfm.parameterized_test(['production'])
class AlltoallTest(rfm.RegressionTest):
    def __init__(self, variant):
        self.tags = {variant}

[vkarak]

You should still pass the CXXFLAGS if you want to be completely correct. CPPFLAGS are the preprocessor flags, i.e., the -D and -I options mostly.

[vkarak]

GPG -> G2G.

[vkarak]

@jenkins-cscs retry dom kesch

[vkarak]

@jenkins-cscs retry dom