[mvapich-discuss] Wrong MPI_Win_allocate calls in OSU benchmark

Subramoni, Hari subramoni.1 at osu.edu
Thu Jun 18 21:18:30 EDT 2020


Dear, Bert.

Many thanks for the report and the patch. We appreciate it.

We will take a look at it and get back to you.

Best,
Hari.

-----Original Message-----
From: mvapich-discuss-bounces at cse.ohio-state.edu <mvapich-discuss-bounces at mailman.cse.ohio-state.edu> On Behalf Of Bert Wesarg
Sent: Thursday, June 18, 2020 6:12 PM
To: mvapich-discuss at cse.ohio-state.edu <mvapich-discuss at mailman.cse.ohio-state.edu>
Subject: Re: [mvapich-discuss] Wrong MPI_Win_allocate calls in OSU benchmark

Dear all,

I needed to revisit this, because valgrind showed me invalid reads/writes and also memory leaks. I come up with the attached second version of the patch.

Please review.

Best,
Bert

On 18.06.20 09:41, Bert Wesarg wrote:
> Dear all,
> 
> while using the OSU benchmark, I noticed that the `baseptr` parameter was wrongly assigned in the calls to `MPI_Win_allocate`. This parameter is marked as an OUT parameter and thus should be a pointer to a pointer. Just like `MPI_Mem_alloc`.
> 
> Please find attached a diff against 5.6.3. The additional changes in mpi/one-sided/osu_get_acc_latency.c are due to the fact, that `rbuf` is not passed as reference to `allocate_memory_get_acc_lat` and this parameter shadows the global one.
> 
> Best,
> Bert
> 

-- 
Dipl.-Inf. Bert Wesarg
wiss. Mitarbeiter

Technische Universität Dresden
Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH)
01062 Dresden

📞 +49 (351) 463-42451
📠 +49 (351) 463-37773
📧 bert.wesarg at tu-dresden.de



More information about the mvapich-discuss mailing list