Two build warnings in CLANG DataFormats/Math/test/CholeskyInvert_t.cpp

[mrodozov]

In the Clang IB we have two warnings due to

/data/cmsbld/jenkins/workspace/build-any-ib/w/tmp/BUILDROOT/cff53344cac83c1abd88bb561d95fc5c/opt/cmssw/slc7_amd64_gcc900/cms/cmssw/CMSSW_11_3_CLANG_X_2021-01-25-2300/src/DataFormats/Math/test/CholeskyInvert_t.cpp:48:6: warning: loop not vectorized: the optimizer was unable to perform the requested transformation; the transformation might be disabled or specified as part of an unsupported transformation ordering [-Wpass-failed=transform-warning]

which says clang unable to vectorize a loop
probably around the pragmas in the same file
https://github.com/cms-sw/cmssw/blob/master/DataFormats/Math/test/CholeskyInvert_t.cpp#L97

Full log is in:
https://cmssdt.cern.ch/SDT/cgi-bin/buildlogs/slc7_amd64_gcc900/CMSSW_11_3_CLANG_X_2021-01-25-2300/DataFormats/Math

One hint is there is a option to be enabled in our llvm build to enable clang to vectorize loops (have not tried yet if the problem is in the llvm external).

[cmsbuild]

A new Issue was created by @mrodozov Mircho Rodozov.

@Dr15Jones, @dpiparo, @silviodonato, @smuzaffar, @makortel, @qliphy can you please review it and eventually sign/assign? Thanks.

cms-bot commands are listed here

[makortel]

assign reconstruction

[cmsbuild]

New categories assigned: reconstruction

@slava77,@perrotta,@jpata you have been requested to review this Pull request/Issue and eventually sign? Thanks

[makortel]

I see @smuzaffar silenced the warning in #29790 for LLVM 9. Was it the (already distant) upgrade to LLVM 10 that made the warning to reappear?

[mrodozov]

I was reading this which
https://lists.llvm.org/pipermail/llvm-dev/2016-March/096291.html
is old, but it says
"If a loop contains a function the loop cannot be vectorized unless the function is inlined"
which in our case is indeed inlined.
https://github.com/cms-sw/cmssw/blob/master/DataFormats/Math/interface/choleskyInversion.h#L340
If I remove line
https://github.com/cms-sw/cmssw/blob/master/DataFormats/Math/test/CholeskyInvert_t.cpp#L103
the warning is gone. If I change the function with any of this:
https://llvm.org/docs/Vectorizers.html#vectorization-of-function-calls
say creating an array;
int * fff = new int[1024];
and change the loop to this:

for (unsigned int i = 0; i < 1024; ++i) {
       fff[i] = pow(i, 2);
}

compiles without warning. The problem is with the cholesky function not being able to work with the vectorization

[slava77]

@VinInn
it looks like you are the author of the test.
please take a look.
Thank you.

[mrodozov]

@VinInn would you suggest how to proceed with this, please :) A kind reminder

[VinInn]

the doc is this one https://llvm.org/docs/Vectorizers.html
Most probably we can just remove the pragma.
More in general: this type of info in tests is something useful to verify "regression".
They will appear as warning (in gcc as well). If we cannot manage we have just to silence them.