Looks like put and get functions should be added if possible. The MTL
layer looks like it is designed for two-sided only with no intention of
supporting one-sided.
-Nathan
On Thu, Nov 06, 2014 at 03:21:32PM -0700, Nathan Hjelm wrote:
>
> Great! We should probably try to figure out how the mtl l
Great! We should probably try to figure out how the mtl layer can be
modified to expose those atomics. If possible this should be done before
the 1.9 branch to ensure the feature is available in the next release
series.
-Nathan
On Thu, Nov 06, 2014 at 05:15:30PM -0500, Joshua Ladd wrote:
>MX
MXM supports atomics.
On Thursday, November 6, 2014, Nathan Hjelm wrote:
> I haven't look at that yet. Would be great to get the new osc component
> working over both btls and mtls. I know portals supports atomics but I
> don't know whether psm does.
>
> -Nathan
>
> On Thu, Nov 06, 2014 at 08:
Not handling the multi-rail case at this point. Only issue atomics and
rdma operations over a single btl module (which should be a single
HCA).
-Nathan
On Thu, Nov 06, 2014 at 12:15:13PM -0700, Howard Pritchard wrote:
>HI Nathan,
>How would you get things right with atomics and multirail?
HI Nathan,
How would you get things right with atomics and multirail?
Getting the memory consistency right would be really difficult.
You'd have to keep issuing zero length rdma reads and hoping
that that would have the effect of a pci-e flush in the case of
multiple updates to a given target add
I haven't look at that yet. Would be great to get the new osc component
working over both btls and mtls. I know portals supports atomics but I
don't know whether psm does.
-Nathan
On Thu, Nov 06, 2014 at 08:45:15PM +0200, Mike Dubman wrote:
>btw, do you plan to add atomics API to MTL layer as
btw, do you plan to add atomics API to MTL layer as well?
On Thu, Nov 6, 2014 at 5:23 PM, Nathan Hjelm wrote:
> At the moment I select the lowest latency BTL that can reach all of the
> ranks in the communicator used to create the window. I can add code to
> round-robin windows over the availab
At the moment I select the lowest latency BTL that can reach all of the
ranks in the communicator used to create the window. I can add code to
round-robin windows over the available BTLs on multi-rail systems.
-Nathan
On Wed, Nov 05, 2014 at 06:38:25PM -0800, Paul Hargrove wrote:
>All atomics
All atomics must be done through not just "the same btl" but the same btl
MODULE, since atomics from two IB HCAs, for instance, are not necessarily
coherent. So, how is the "best" one to be selected?
-Paul [Sent from my phone]
On Nov 5, 2014 7:15 AM, "Nathan Hjelm" wrote:
>
> In the new osc com
In the osc component, no. Though it would be straightforward to add that
feature.
-Nathan
On Wed, Nov 05, 2014 at 11:05:48AM -0500, Joshua Ladd wrote:
>Does this mean that you maintain a separate channel for 'put' and 'gets'
>that can use multiple transports and another for atomics?
>
>
Does this mean that you maintain a separate channel for 'put' and 'gets'
that can use multiple transports and another for atomics?
Josh
On Wed, Nov 5, 2014 at 10:15 AM, Nathan Hjelm wrote:
>
> In the new osc component I don't try to handle that case. All atomics
> have to be done through the sa
In the new osc component I don't try to handle that case. All atomics
have to be done through the same btl (including atomics on self). I did
this because with the default setup of Gemini they can not be mixed. If
it is possible to mix them with other networks I would be happy to add
an atomic fla
Quick question. Out of curiosity, how do you handle the (common) case of
mixing network atomics with CPU atomics? Say for a single target with two
initiators, one initiator is on host with the target, so goes through the
SM BTL, and the other initiator is off host, so goes through the network
BTL.
What: Completely revamp the BTL RDMA interface (btl_put, btl_get) to
better match what is needed for MPI one-sided.
Why: I am preparing to push an enhanced MPI-3 one-sided component that
makes use of network rdma and atomic operations to provide a fast truely
one-sided implementation. Before I ca
14 matches
Mail list logo
