Thanks for the info. That is interesting.  And yes, Having a CUDA aware MPI 
API list would be very useful.
--Junchao Zhang

On Mon, Aug 19, 2019 at 10:23 AM Fang, Leo 
<leof...@bnl.gov<mailto:leof...@bnl.gov>> wrote:

Hi Junchao,

First, for your second question, the answer is here: 
https://www.mail-archive.com/users@lists.open-mpi.org/msg33279.html. I know 
this because I also asked it earlier 😊 It'd be nice to have this documented in 
the Q&A though.

As for your first question, I am also interested. It'd be nice for Open MPI 
core devs to keep the supported API list up-to-date. We recently added support 
of CUDA-aware MPI to mpi4py, and such a list is important for us to keep track 
upstream support so that we know whether a test fails due to lack of 
CUDA-awareness or because we messed up (much less likely).



Yao-Lung Leo Fang
Assistant Computational Scientist
Computational Science Initiative
Brookhaven National Laboratory
Bldg. 725, Room 2-169
P.O. Box 5000, Upton, NY 11973-5000
Office: (631) 344-3265
Email: leof...@bnl.gov<mailto:leof...@bnl.gov>
Website: https://leofang.github.io/
寄件者: users 
<users-boun...@lists.open-mpi.org<mailto:users-boun...@lists.open-mpi.org>> 代表 
Zhang, Junchao via users 
寄件日期: 2019年8月15日 上午 11:52:56
收件者: Open MPI Users <users@lists.open-mpi.org<mailto:users@lists.open-mpi.org>>
副本: Zhang, Junchao <jczh...@mcs.anl.gov<mailto:jczh...@mcs.anl.gov>>
主旨: Re: [OMPI users] CUDA supported APIs

Another question: If MPI_Allgatherv(const void *sendbuf, int sendcount, 
MPI_Datatype sendtype, void *recvbuf, const int recvcounts[],const int 
displs[], MPI_Datatype recvtype, MPI_Comm comm) is cuda aware, are recvcounts, 
displs in CPU memory or GPU memory?

--Junchao Zhang

On Thu, Aug 15, 2019 at 9:55 AM Junchao Zhang 
<jczh...@mcs.anl.gov<mailto:jczh...@mcs.anl.gov>> wrote:
   Are the APIs at 
 latest? I could not find MPI_Neighbor_xxx and MPI_Reduce_local.

--Junchao Zhang
users mailing list

Reply via email to