1 |
Donnie Berkholz wrote: |
2 |
|
3 |
> Adam Pi±tyszek wrote: |
4 |
> |
5 |
>>I have just tested your autotooled *-reference ebuilds... |
6 |
>>does not work (cholesky_test from IT++ package goes into an endless loop). |
7 |
> |
8 |
> ... |
9 |
> |
10 |
> That's an odd problem. I'd expect at least undefined functions or |
11 |
> something obvious, not a random loop. I'll take a look at it soon, maybe |
12 |
> I missed something obvious. |
13 |
|
14 |
May that be related to the DLAMC3 problem reported at |
15 |
http://www.theochem.kth.se/dalton/faq.html near the end of the page? The |
16 |
links to the Fedora bugzilla work no more, but as far as I remember the |
17 |
infinite loop arises somehow from incorrect machine precision |
18 |
assumptions due to excessive precision of numbers in registers, not |
19 |
complient to the IEEE standard. So I try (althoug rather with the atlas |
20 |
libraries than with the reference ones) using -mieee-fp compiler flag |
21 |
rather than -ffloat-store recomended at the Dalton page (and -O1 rather |
22 |
than -O2 on my AMD machines -- both athlon xp and k8 architectures -- |
23 |
due to other problem, that is too obscure to me so far for me to be even |
24 |
able to file a reasonable bugzilla entry). |
25 |
If that is the case, than the key to the malfunction may be in the |
26 |
compiler flags, and the ebuilds could be improved by some flags checking |
27 |
and mangling, otherwise being OK. |
28 |
|
29 |
Honza Macháèek |
30 |
-- |
31 |
gentoo-science@g.o mailing list |