[Beowulf] Re: the comparison between OpenMP and MPI

Joe Landman landman at scalableinformatics.com
Mon May 21 09:47:31 PDT 2007

Florent Calvayrac wrote:

> I once discovered a bug in the Cray Fortran compiler for T3E.
> Without optimization the code was running fine but at O2/O3
> results were wrong. It turned out that some code lines I had autogenerated
> from Maple were implying a large number of variables, exceeding
> the number of registers available and wrapping on the first register.
> Without optimization no registers were used so the code was giving
> correct results and at O1 code was optimized on a line per line basis
> and not procedure-wide.
> So even commercial code can be wrong...

FWIW:  recently (within the last month) we worked on building a 64 bit 
version of a code for a customer.  F90.  Intel happily compiled it 
(latest and greatest 9.1.x).  PathScale compiler complained about 
problems.  PathScale was in fact, correct.

That Intel compiled and "ran" this code w/o nary a peep worries me.  The 
issue was a somewhat esoteric type safety issue (never assume logical is 
an integer and vice versa, especially when moving from 32 bit to 64 bit 
code, and copying data ...).  Regardless, the PathScale compiler was 
correct for raising the issue, and the Intel compiler was not for 
failing to raise it.

Didn't try it with g95/gfortran.

Another problem we have found with the intel compilers is the way they 
set particular code paths, specifically with regards to testing for 
capability.  Rather than testing for sse2/3/4 instruction results, they 
test processor strings/ids.  Which means their generated code often 
fails to run, or runs with slow code paths selected on Opterons.  Which 
means we tend to wave people off these compilers unless their code is 
only ever going to run on an Intel platform.  That and the compiler 
doesn't yet seem to do much with NUMA as the Intel platform doesn't need 
to worry about it.




Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452 or +1 866 888 3112
cell : +1 734 612 4615

More information about the Beowulf mailing list