[mvapich-discuss] MVAPICH bug report

[Matthew Koop]

Kevin,

Thanks for this suggestion. I assume you mean that these problems are
occuring when the user manually sets parameters outside of the range of
the device and not that the default values themselves are leading to
problems.

We've updated the SVN trunk with checks to do additional verification of
the parameters.

Thanks,
Matt

On Mon, 7 Aug 2006, Kevin Ball wrote:

> Dear MVAPICH development team,
>
>   We believe we have found a bug in the MVAPICH QP initialization
> process.  In mpid/ch_gen2/viaparam.c a number of efault values are set
> using preprocessor macros from mpid/ch_gen2/ibverbs_const.h.  Many of
> these values, however, can vary by device.  For example,
> viadev_default_mtu, viadev_default_qp_ous_rd_atom, and
> viadev_default_max_sg_list are set to VIADEV_DEFAULT_MTU,
> VIADEV_DEFAULT_QP_OUS_RD_ATOM, and VIADEV_DEFAULT_MAX_SG_LIST.  They are
> later used without checking to see if the values assigned are supported
> by the device, leading to failures at runtime in the case where this is
> not true.
>
>   The values supported by the device can be discovered using the verb
> ibv_query_device.
>
> Thanks!
>
> Kevin Ball
> Senior Engineer
> SIG, QLogic
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at mail.cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>