[ofiwg] completion flags as actually defined by OFI

Jason Gunthorpe jgunthorpe at obsidianresearch.com
Wed Apr 15 09:56:34 PDT 2015


On Wed, Apr 15, 2015 at 04:46:59PM +0000, Hefty, Sean wrote:
> > Perhaps all we can do at this point is recognize that new completions may
> > be necessary and we allocate spare bits for them?
> 
> I think this is the real requirement that needs to be met.  OFI has this ability.
> 
> I am removing FI_COMMIT_COMPLETE from the API until things can be resolved.  I think this goes beyond a completion flag and may require changes to memory registration and/or a new operation. 

What's your plan for FI_TRANSMIT_COMPLETE?

Jason



More information about the ofiwg mailing list