Search found 6 matches

by mkbane
Thu Jan 05, 2012 2:53 am
Forum: OpenMP 3.0 API Specifications
Topic: how much cost of openMP 3.0 ?
Replies: 0
Views: 49684

Re: how much cost of openMP 3.0 ?

It's a specification of extensions to the likes of FORTRAN and C/C++ so OpenMP itself is free.
To make use of the parallel directives you will need an OpenMP compliant compiler, which may or may not be free.

Michael
@mkbane_mcr
by mkbane
Mon Jul 18, 2011 2:03 pm
Forum: Using OpenMP
Topic: combined PARALLEL DO slower than uncombined
Replies: 4
Views: 5648

Re: combined PARALLEL DO slower than uncombined

Sorry for the delay but I can confirm that with Intel 11.1 that the OMP PARALLEL DO REDUCTION (single combined directive) does take much longer than the OMP PARALLEL REDUCTION followed immediately by a OMP DO viz threads par-do-reduce par reduce;do 1 68 sec 64 sec 2 134 sec 32 sec 4 145 sec 16 sec 6...
by mkbane
Sat Apr 23, 2011 5:05 am
Forum: OpenMP 3.0 API Specifications
Topic: GPU Directives
Replies: 1
Views: 23453

GPU Directives

Just wondering if any directives for GPUs will be included this time around?
Michael
@mkbane_mcr
by mkbane
Fri Mar 25, 2011 9:57 am
Forum: Using OpenMP
Topic: combined PARALLEL DO slower than uncombined
Replies: 4
Views: 5648

Re: combined PARALLEL DO slower than uncombined

I need to re-check that previous claim on the box in question since just tried ifort 11.0 and ifort 11.1 on another Sci Linux box and see the problem with 11.0 but not with 11.1 (and more strangely the 11.0 on 1 thread is 20s whereas 11.1 on 40 sec on 1 and 20 sec on 2)...
by mkbane
Fri Mar 25, 2011 2:05 am
Forum: Using OpenMP
Topic: combined PARALLEL DO slower than uncombined
Replies: 4
Views: 5648

Re: combined PARALLEL DO slower than uncombined

Tried Intel 11.0 and 11.1, on Scientific Linux.
by mkbane
Thu Mar 24, 2011 5:35 pm
Forum: Using OpenMP
Topic: combined PARALLEL DO slower than uncombined
Replies: 4
Views: 5648

combined PARALLEL DO slower than uncombined

Perhaps I'm missing something here, but why does the following slow-down on 2 or more threads whereas if I split the PARALLEL DO it scales as expected? Thanks, M program calcpi USE omp_lib implicit none double precision:: h,x,sum,pi integer:: n,i double precision:: f f(x) = 4.0/(1.0+x**2) n = 210000...