Fix issues on new_testing

Try to fix issues reported by @furmento

• Enable the possibility to give both gemm and [sdcz]gemm as operator
• Add the operation option in the .in files
• Fix issue when calling ./dnew-testing -h
• Move CHAMELEON_My_Mpi_Rank() as deprecated in profit of CHAMELEON_Comm_rank()
• Add the missing -l --niter option to set the number of iteration of each test.
• Fix issue with the conversions rules that were not converting alpha/beta parameters, as well as read/print functions

added Feature Tests/CTest labels

added 3 commits

• 7d441eb9...41060486 - 2 commits from branch solverstack:master
• b61201cf - Enable both [sdcz]xxxx and xxxx as operations, and to declare the operator withinh the input file

Compare with previous version

the output header is different if the option --check is set or not. could the header always have the column RETURN and have a n/a value when --check is not set ?

and what about the parameter --iter ? is it no longer available ?

added 3 commits

• c8ab9a27 - Remove the use of CHAMELEON_My_Mpi_Rank to replace it by CHAMELEON_Comm_rank
• b849fccc - Add the niter parameter
• 35d48058 - Fix mistake in starpu

Compare with previous version

added 1 commit

• 3feb9ade - Fix conversion rules, and issue with the niter

Compare with previous version

changed the description

added 1 commit

• 063b70a1 - Cleanup some format

Compare with previous version

• About the return parameter, it is not printed when no check is performed to avoid confusion. We could print SUCCESS/FAILURE when check is not enabled to say the testing returned a correct value, but I think this could be confusing for users. We would think that some checks have been performed when it is not the case. (@ltaief, @agullo, @thibault, @fpruvost what do you think ?)

This other question is do you really need this column ? or you suggest we should print that no check has been performed ?

• The niter parameter has been restored in this PR.

for the return parameter, yes my suggestion was to print n/a to say that no check has been performed.

this way, the number of columns in the output is always the same no matter if check was enabled or disabled.

thanks for the niter parameter

No that's the mistake, the number of columns will vary. For now, not all testings have it, but the checks are adding information like the norm of the residual, the norm of the input matrix, the value of the scaled residual that is used to check the validity, ... You can try the QR testings, and the LU testings. You'll see that the number of columns will vary with and without check.

The idea is to detect the columns by their header, not by their place in the print :)

okay, i did not know, thanks for the information. So yes no need to always have the result column.

No problem? I was too lazy to update each testing one by one to have them included from the first PR :) It will probably be integrated as long as issues are detected and need more information in their analysis.

so let's merge!

merged

mentioned in commit f2c3f027

another question, what is the content of the gflops column when the computation fails ? it used to be a list of - if i am not mistaken.

Good question. I would say, it never fails :D but I think that will be an incoherent value, nothing else. Maybe we to change that to -1 for example.

mentioned in commit debfb49a

Fixed in commit debfb49a