On 03/06/15 19:34, Jens Timmerman wrote:
Hi Jack,
Sorry I didn't get back to you on this faster, I knew we were doing
some magic,
but didn't know by hart where to look it up, and Kenneth is busy with
his merge sprint.
On 03/06/15 19:20, Jack Perdue wrote:
- I discovered that EB controls mpicc default compiler
via I_MPI_CC. That was the part of equation I was missing.
I was concerned that all this time when building with
ictce/intel toolchains, since I only saw EB using mpicc,
that EB was using the default gcc wrapper. Now I see
that the variable above is set to icc. So, mpicc DOES use
icc by default when building with EB. Good!
I'm glad you found your answer!
+1 and confirmed, we indeed handle this via $I_MPI_CC and co.
The main reason for doing so is because with OpenMPI (and others) this
is the only way (e.g. via $OMPI_CC), i.e. they don't provide dedicated
wrapper commands equivalent to mpiicc.
To clarify: ' set_mpi_wrapper_aliases_intel' is there to deal with the
situation of setting $I_MPI_CC to gcc, yet retaining the expected
behavior of mpiicc to use icc (yes, mpiicc also listens to $I_MPI_CC...).
Obviously I should've known (and trusted) the EB team
to do the right thing. I just needed to confirm to alleviate
my concerns.
Not at all, I'm glad someone who knows what should happen dears to dive
in and take a look at what's actually happening, this is why we open
source things,
so more eyes can look at the code and find bugs. Feel free to keep
asking questions when
you're in doubt! Hopefully someone can answer quicker in the future.
Very much +1, please don't stop questioning what EB currently does, so
we can keep improving it's functionality.
regards,
Kenneth
