[Openmp-dev] runtime/test/api/omp_get_wtime.c tolerances

[Peyton, Jonathan L via Openmp-dev]

Jack,

Can you see if this patch fixes the rpath issues (it also changes the tolerance for the timer as well)?
The choice of 1% for the timer tolerance is arbitrary (it had worked for me on all the OS X 10.8 and 10.9 versions I tried as well as Linux).  It is in the University of Houston's testsuite.  I don't mind changing it to 3% to accommodate later OSX versions.

-- Johnny

-----Original Message-----
From: Jack Howarth [mailto:howarth.mailing.lists at gmail.com] 
Sent: Wednesday, September 23, 2015 8:02 AM
To: openmp-dev at dcs-maillist2.engr.illinois.edu; Peyton, Jonathan L
Subject: runtime/test/api/omp_get_wtime.c tolerances

Jonathan,
         As I mentioned in http://reviews.llvm.org/D11821, the current default tolerances of...

return ((measured_time > 0.99 * wait_time) && (measured_time < 1.01 *
wait_time)) ;

cause the  runtime/test/api/omp_get_wtime.c testcase to fail its execution test on OS X 10.11 but not on 10.10 and earlier. Execution of the test case repeatedly on a dual quad-core MacPro 3,1 produces exit codes ranging from 2 to 6. Increasing this tolerance to...

return ((measured_time > 0.98 * wait_time) && (measured_time < 1.02 *
wait_time)) ;

results in the exit code ranging from 0 to 1. Only increasing the tolerances to...

return ((measured_time > 0.97 * wait_time) && (measured_time < 1.03 *
wait_time)) ;

cause the test case to always return the exit code of 0.
        Where exactly did you come up with this 1% timer tolerance and is it part of a particular OpenMP specification? If so, I'll open a radar bug report against OS X 10.11 on this issue. Thanks in advance for any clarification.
                Jack
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mac_test_fixes.patch
Type: application/octet-stream
Size: 1848 bytes
Desc: mac_test_fixes.patch
URL: <http://lists.llvm.org/pipermail/openmp-dev/attachments/20150923/3b00247f/attachment.obj>