Bug 134152 – ACML 2.5.0 fails fastica_test.cpp from itpp-3.10.2

• Status: RESOLVED FIXED
• Product: Gentoo Linux

[Description Adam Piątyszek 2006-05-23 14:38:48 UTC]

Hi!

I would like to report a problem with acml-2.5.0, which is currently marked as stable for x86. I am a developer of the IT++ (itpp) package and I noticed that one of the internal tests (fastica_test.cpp) gives wrong results when IT++ is linked to acml-2.5.0. This test uses some lapack routines. No problems observed when linking IT++ to {blas,lapack}-atlas or {blas,cblas,lapack}-reference.

For testing purposes, I downloaded an updated version - 2.5.3 - of the acml library and rerun the IT++ tests once again. That time all tests PASSED.
I also tried higher relases of the acml library (2.7.0, 3.0.0 and 3.1.0) and all of them work perfectly.

Therefore, I suggest an immediate update of the stable 2.5.0 release to version 2.5.3 at least or marking 2.7.0 or 3.0.0 or 3.1.0 as stable.

How to reproduce the bug:
1) emerge "=sci-libs/acml-2.5.0"
2) blas-config ACML; lapack-config ACML
3) USE="blas cblas lapack fftw" ebuild /usr/portage/sci-libs/itpp/itpp-3.10.2.ebuild test
4) observe the failing test:
[...]
Test `eigen_test' PASSED.
------------------------------------------------------------------------------
Test `fastica_test' FAILED!!!
------------------------------------------------------------------------------
Test `fastmath_test' PASSED.
[...]
5) emerge "sci-libs/acml-2.7.0"
6) cd /var/tmp/portage/itpp-3.10.2/work/itpp-3.10.2
7) make check
8) observe proper results of `fastica_test':
[...]
Test `fastica_test' PASSED.
------------------------------------------------------------------------------
[...]

BR,
/ediap

[Comment 1 Adam Piątyszek 2006-07-19 23:32:45 UTC]

This report is almost two months old now... So I would like to gently remind it ;)

BR,
/ediap

[Comment 2 Danny van Dyk (RETIRED) 2006-08-01 16:29:31 UTC]

ACML-3-0-0 is stable on amd64 and x86 now. Sorry this took so long. But
i don't had an x86-(stable-chroot) at hand :-/

[Comment 3 Adam Piątyszek 2006-08-02 01:20:45 UTC]

Are you sure? After todays sync I have in my portage:

ediap@lespaul acml % grep KEYWORDS acml-3.0.0.ebuild 
KEYWORDS="amd64 ~x86"

BR,
/ediap

[Comment 4 Adam Piątyszek 2006-08-15 03:37:31 UTC]

I decided to reopen this bug report, since acml-2.5.0 is still the only version stable on x86.

[Comment 5 Danny van Dyk (RETIRED) 2006-08-15 14:24:32 UTC]

God. a) i stabled 3.1.0-r1, though i tested 3.0.0.
b) 3.1.0* is still in p.mask.

I reversed the 3.1.0-r1 changes and stable 3.0.0 now. Sorry, must have been
_really_ confused when i did this :-/

I hope this bug can be fixed now.