Good point, and the binaries do indeed exist.

I’m running on a 2008 MacPro, with pre-Nehalem Xeons that are one major 
generation behind the Xserves. So that makes me an ideal candidate to test.

Now it’s simply a matter of putting together a suite of tests that thoroughly 
exercises OpenMPI. Stay tuned...


> On 2021-04-27-T, at 19:20, Ryan Schmidt <[email protected]> wrote:
> 
>> On Apr 27, 2021, at 18:15, Christopher Nielsen wrote:
>> 
>> [My previous reply looks like it may be stuck in the mailman queue, and/or 
>> rejected. So re-sending with additional info.]
>> 
>> Well, the investigation is all on me, barring a definitive answer from 
>> someone more knowledgable.
>> 
>> But after a lot of scouring, I haven’t found anything in the OpenMPI 
>> documentation - or on the web in-general - that suggests binaries are tied 
>> to a build machine’s specific hardware.
>> 
>> So what I’m asking is, should we just go ahead and update openmpi-* to 
>> publish binaries?
> 
> We already publish the binaries; the port is just configured not to look for 
> them. So if you want to test first, and you have a machine with a processor 
> that's older than the Xeon in the 2009 Xserves we use to build, you could 
> remove the archive_sites clearing line from the port and try to install the 
> binary. If openmpi crashes or doesn't work, that would be an indication that 
> we should not change the port.

Reply via email to