[Llvm-bgq-discuss] Status of OpenMP support in BG/Q

Schlottke, Michael m.schlottke at fz-juelich.de
Sat Feb 8 03:27:41 CST 2014


>>> mpibgclang++/*: -std=c++11 -stdlib=libc++  -O3 -DNDEBUG -mtune=native
>>> -fvectorize -fslp-vectorize
>>
>> -mtune=native seems wrong in general here: you're cross compiling! (I don't think it does anything with bgclang, but if it did, if would cause you to optimize for the P7 and not the A2, and the instruction scheduling is very different for those two cores).
>
> I would argue that mpibgclang should provide native=a2qpx not
> power7vsx because the mpi/bg compilers are only for cross-compiling
> anyways.
We have the -mtune=… setting from our default setting for clang on other hosts. A quick check about a year ago did not show any performance changes, so we just left it in there… But I will remove it anyways, since it is "just wrong" :)

> Does LLVM support -march= such that the code will not run on the wrong
> processor?  That would prevent certain accidental errors.
Indeed: mpibgclang++ will complain about "unused command line option -march=…"

> Indeed, no one should make decisions based upon measurements with the
> 2012 XLC, for example.
For your reference, our benchmarks were done using Xl 12.1. For clang, however, we were still using the relatively old clang/3.4.r193439 version.

Regards,

Michael


------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------



More information about the llvm-bgq-discuss mailing list