[mvapich-discuss] Compilation error for mvapich2-2.0.1 with disabled C++ bindings.

[Jonathan Perkins]

On Tue, Jan 13, 2015 at 05:33:04PM -0600, Chris Green wrote:
> Hi,
> 
> On 1/13/15 4:36 PM, Jonathan Perkins wrote:
> >There's a great chance. This patch is queued to be included in testing
> >tonight.
> Great, thank you. One other question and an observation, since I obviously
> have an expert on the line:
> 
>  * Will my previously mentioned configure find and use my InifiniBand
>    support (Mellanox card, mlx4 driver) and use it optimally, or am I
>    missing a --with option or two? Repeated here for ease of reference:
> 
>    ./configure --prefix=/usr/local/mvapich2-2.0.1 --enable-fast=O3,ndebug --enable-f77 --enable-fc \
>                 --disable-cxx --enable-romio --enable-versioning --enable-threads=runtime --enable-registration-cache \
>                 --enable-rsh --enable-shared --enable-static --enable-yield=sched_yield --enable-rdma-cm --with-pm=hydra

It will detect the support available with your installation of OFED (or
distro provided libraries) and enable useful features such as multicast
and xrc if they are available.  After that MVAPICH2 detects the HCA
support at runtime and use settings which will provide you very good
results in most cases.

If you'd like to know more about which features or environment variables
may need to be tweaked for your setup we can start a separate thread
about that.  Unfortunately I'm not an expert in that area but I do work
with a few who may be able to give you good advice.

> 
>  * If I use --enable-strict, the configure fails during a test compile
>    for the rdma-cm feature -- something breaks down with the rdma_cma.h
>    header's use of its own unioned structs. Since it wasn't necessary
>    to get me the feature set I wanted, I just dropped the strict option
>    and continued. If you like, I can send you the header and config.log
>    out of band.

I'll have to look at this.  In the past we changed --enable-strict to
use the C99 dialect instead of C89 because the infiniband headers use
C99 features.  I'm not sure if we had a regression here or if this is a
separate new issue.

I can try to reproduce this myself but I'd appreciate it if you can
forward along any output that you may have from the --enable-strict
failure.

> 
> Thanks again for your prompt responses.

No problem.

> 
> Best,
> Chris.
> 
> -- 
> Chris Green <greenc at fnal.gov>, FNAL CS/SCD/ADSS/SSI/SSD;
> 'phone (630) 840-2167; Skype: chris.h.green;
> IM: greenc at jabber.fnal.gov, chissgreen (AIM, Yahoo),
> chissg at hotmail.com (MSNM), chris.h.green (Google Talk).
> 

-- 
Jonathan Perkins