This is the mail archive of the
cygwin
mailing list for the Cygwin project.
Re: Some Problems about gcc 4.8.2 on cygwin
- From: Tim Prince <n8tm at aol dot com>
- To: cygwin at cygwin dot com
- Date: Fri, 14 Mar 2014 09:52:26 -0400
- Subject: Re: Some Problems about gcc 4.8.2 on cygwin
- Authentication-results: sourceware.org; auth=none
- References: <CAFrJEvN+Rckr6_XDYr5WUWLtYA6T6694bwPZAzUmR8OZdJ-_Zw at mail dot gmail dot com>
- Reply-to: tprince at computer dot org
On 3/14/2014 8:42 AM, rexdf Rexdf wrote: 2.It's about OpenMP
<elided stuff which doesn't quote legibly>
Is there a cygwin related question here? Questions on OpenMP and
clock() might be tolerated on gcc-help.
Advice on how to make meaningful benchmarks is definitely off topic.
With cygwin g++ 4.9 at -O or -O3 on win8.1 I get
time: 0
as evidently the compiler can shortcut your test loop (is that what you
wished?).
For the -O0 -fopenmp build,
time: 2484 (1 thread)
time: 2547 (2 threads)
time: 2828 (3 t)
time: 3187 (4 t)
Much as would be expected, as you are asking for the total time spent
among all threads (usual interpretation of clock()).
The bash time command shows real time decreasing with number of threads
(close to total clock time divided by number of threads). OpenMP
provides the function omp_get_wtime() for performance measurement
(possibly a wrapper for gettimeofday()).
The cygwin library evidently doesn't treat clock() as equivalent to
omp_get_wtime(). Speculation on how you could find a non-cygwin library
which treats them as equivalent is probably off topic here.
Anyway, I think your problem is not with the cygwin gcc, unless you are
looking for the more aggressive optimization of version 4.9.
--
Tim Prince
--
Problem reports: http://cygwin.com/problems.html
FAQ: http://cygwin.com/faq/
Documentation: http://cygwin.com/docs.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple