Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] [OMPI svn-full] svn:open-mpi r19628
From: Tim Mattox (timattox_at_[hidden])
Date: 2008-09-24 15:40:30


Ok, regression might not technically be the right word, but from a user's
perspective, it would seem like it.

That is, if I understand the implications of the bug that was fixed in 1.2.6 via
https://svn.open-mpi.org/trac/ompi/changeset/17883
Since, prior to that fix, the IBM eHCA got lucky and worked because
we were comparing to the wrong parameter which was much smaller, AFAIK.
I'm open to a better wording in the NEWS file. Something like "a
fix in 1.2.6 revealed a latent bug that affected the IBM eHCA?"

On Wed, Sep 24, 2008 at 2:32 PM, Jeff Squyres <jsquyres_at_[hidden]> wrote:
> For the NEWS: I don't think this was a regression. Did we really have those
> casts initially and remove them in v1.2.6? I thought Brad said they only
> first *noticed* it in v1.2.6 -- I don't know where the change actually came
> from (it could actually be in the hardware or the driver reporting much
> larger numbers that they used to).
>
>
>
> On Sep 24, 2008, at 1:59 PM, timattox_at_[hidden] wrote:
>
>> Author: timattox
>> Date: 2008-09-24 13:59:00 EDT (Wed, 24 Sep 2008)
>> New Revision: 19628
>> URL: https://svn.open-mpi.org/trac/ompi/changeset/19628
>>
>> Log:
>> Closes #1526, #1515: Fix handling of max_cqe, which makes IBM eHCA work
>> again.
>>
>> Submitted by bbenton, Reviewed by jsquyres, RM-Approved by tdd.
>>
>> Patch applied which adds some typecasts to btl_openib.c.
>>
>> Also, bumped the VERSION.
>>
>> Text files modified:
>> branches/v1.2/NEWS | 6 ++++++
>> branches/v1.2/VERSION | 4 ++--
>> branches/v1.2/ompi/mca/btl/openib/btl_openib.c | 8 ++++----
>> 3 files changed, 12 insertions(+), 6 deletions(-)
>>
>> Modified: branches/v1.2/NEWS
>>
>> ==============================================================================
>> --- branches/v1.2/NEWS (original)
>> +++ branches/v1.2/NEWS 2008-09-24 13:59:00 EDT (Wed, 24 Sep 2008)
>> @@ -26,6 +26,12 @@
>> bug fixes (and other actions) for each version of Open MPI since
>> version 1.0.
>>
>> +1.2.8
>> +-----
>> +
>> +- Fix a regession introduced in 1.2.6 for the IBM eHCA. See ticket #1526.
>> +
>> +
>> 1.2.7
>> -----
>>
>>
>> Modified: branches/v1.2/VERSION
>>
>> ==============================================================================
>> --- branches/v1.2/VERSION (original)
>> +++ branches/v1.2/VERSION 2008-09-24 13:59:00 EDT (Wed, 24 Sep 2008)
>> @@ -8,7 +8,7 @@
>>
>> major=1
>> minor=2
>> -release=7
>> +release=8
>>
>> # greek is used for alpha or beta release tags. If it is non-empty,
>> # it will be appended to the version number. It does not have to be
>> @@ -17,7 +17,7 @@
>> # only requirement is that it must be entirely printable ASCII
>> # characters and have no white space.
>>
>> -greek=rc6
>> +greek=b1
>>
>> # If want_svn=1, then the SVN r number will be included in the overall
>> # Open MPI version number in some form.
>>
>> Modified: branches/v1.2/ompi/mca/btl/openib/btl_openib.c
>>
>> ==============================================================================
>> --- branches/v1.2/ompi/mca/btl/openib/btl_openib.c (original)
>> +++ branches/v1.2/ompi/mca/btl/openib/btl_openib.c 2008-09-24
>> 13:59:00 EDT (Wed, 24 Sep 2008)
>> @@ -257,8 +257,8 @@
>>
>>
>> if(min_cq_size > (int32_t) mca_btl_openib_component.ib_cq_size) {
>> - mca_btl_openib_component.ib_cq_size = min_cq_size >
>> openib_btl->hca->ib_dev_attr.max_cqe ?
>> - openib_btl->hca->ib_dev_attr.max_cqe : min_cq_size;
>> + mca_btl_openib_component.ib_cq_size = min_cq_size >
>> (uint32_t)openib_btl->hca->ib_dev_attr.max_cqe ?
>> + (uint32_t)openib_btl->hca->ib_dev_attr.max_cqe : min_cq_size;
>> #if OMPI_MCA_BTL_OPENIB_HAVE_RESIZE_CQ
>> if(!first_time) {
>> rc = ibv_resize_cq(openib_btl->ib_cq[BTL_OPENIB_LP_QP],
>> mca_btl_openib_component.ib_cq_size);
>> @@ -280,8 +280,8 @@
>> if(first_time) {
>> /* never been here before, setup cq and srq */
>> mca_btl_openib_component.ib_cq_size = (int)
>> mca_btl_openib_component.ib_cq_size >
>> - openib_btl->hca->ib_dev_attr.max_cqe ?
>> - openib_btl->hca->ib_dev_attr.max_cqe :
>> + (uint32_t)openib_btl->hca->ib_dev_attr.max_cqe ?
>> + (uint32_t)openib_btl->hca->ib_dev_attr.max_cqe :
>> (int) mca_btl_openib_component.ib_cq_size;
>> return mca_btl_openib_create_cq_srq(openib_btl);
>> }
>> _______________________________________________
>> svn-full mailing list
>> svn-full_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/svn-full
>
>
> --
> Jeff Squyres
> Cisco Systems
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>

-- 
Tim Mattox, Ph.D. - http://homepage.mac.com/tmattox/
 tmattox_at_[hidden] || timattox_at_[hidden]
 I'm a bright... http://www.the-brights.net/