Dear Mark,
We are testing extensively the LAPACK library and we know more or less the
behavior on most popular machines.
Usually the problem comes from the compilers flags or BLAS library. That's why
I always advice to go with the Reference BLAS first.
We have some (tortured) tests that will never pass on some machines. I agree,
we need to take a closer look on those.
We are working on changing those LAPACK testings as they are confusing to users.
Here would be my guidelines
1) Test completed successfully without seg fault
2) No illegal error
3) No info error
4) No severe error
5) Between 200-500 tests failing (usually around 280 is the minimum [this is
exactly what you have])
and that's it.
If you have a serious problem, failing tests will go above 1000.
Hope it helps
Julie
On Mar 23, 2011, at 6:07 AM, Mark Dixon wrote:
On Tue, 22 Mar 2011, julie langou wrote:
Mark,
Actually if the build finish without a segfaut, you're fine.
Is the bar really that low?!
I would build again the Reference BLAS first and then ATLAS.
Will do.
Here is a little script that may help you.
You should have only Numerical Errors (between 200 and 500) Those are
torture tests and we know some are failing.
Thanks very much for that script. So that I'm absolutely clear and I know
I can jump for joy (and I will), which did you mean:
1) That there will be between 200-500 tests failing
2) The values of the numerical errors should be between 200 and 500
I take it that 'value of largest test error' is where I should be looking
for the E+06 major error criterion?
I'm a bit puzzled though. The values for 'value of largest test error'
are much lower than the last numbers of some of the lines that look like
(output of your script from the same output files appended to this email):
Type= 2, IWA= 5, IWB= 1, IWX= 3, IWY= 1, result 4 is 1.737D+07
Is the line above not a test failure, then?
I'm getting a few "info error"s on some compilers from DGEES, DDRVGG,
ZDRGEV - I guess I'll look into that with the reference BLAS.
Best wishes,
Mark
--
-----------------------------------------------------------------
Mark Dixon Email : m.c.dixon@Domain.Removed
HPC/Grid Systems Support Tel (int): 35429
Information Systems Services Tel (ext): +44(0)113 343 5429
University of Leeds, LS2 9JT, UK
-----------------------------------------------------------------
******************************************************
RESULTS
- numerical failure:
CST drivers: 1 out of 11664 tests failed to pass the threshold
DXV drivers: 200 out of 5000 tests failed to pass the threshold
DST drivers: 1 out of 14256 tests failed to pass the threshold
SXV drivers: 37 out of 5000 tests failed to pass the threshold
ZXV drivers: 24 out of 5000 tests failed to pass the threshold
ZST drivers: 1 out of 11664 tests failed to pass the threshold
CST: 1 out of 4662 tests failed to pass the threshold
CST: 1 out of 4662 tests failed to pass the threshold
DST: 1 out of 4662 tests failed to pass the threshold
DST: 1 out of 4662 tests failed to pass the threshold
SST: 1 out of 4662 tests failed to pass the threshold
SST: 1 out of 4662 tests failed to pass the threshold
ZST: 1 out of 4662 tests failed to pass the threshold
ZST: 1 out of 4662 tests failed to pass the threshold
ZST: 1 out of 4662 tests failed to pass the threshold
value of largest test error = 0.189E+01
value of largest test error = 0.796E+04
value of largest test error = 0.160D+01
value of largest test error = 0.197E+01
value of largest test error = 0.189D+01
- illegal error: 0
- info error: 0
- severe error 0
_______________________________________________
Lapack mailing list
Lapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/lapack
**********************************************
Julie Langou; Research Associate in Computer Science
Innovative Computing Laboratory;
University of Tennessee from Denver, Colorado ;-)
julie@Domain.Removed; http://www.cs.utk.edu/~julie/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.eecs.utk.edu/mailman/private/lapack/attachments/20110323/d81eebca/attachment.html
|