On 10 February 2008 at 00:06, Ben Goodrich wrote: | -----BEGIN PGP SIGNED MESSAGE----- | Hash: SHA1 | | Dirk Eddelbuettel wrote: | | [snip] | | > I attach an example program below. It runs as follows | > | > [EMAIL PROTECTED]:~/src/progs/C> ./lapack_dgesdd | > info 0 tmp 1.000000 != recd 603 | > 1.725883,0.321633,0.031039,0.001979,0.000088,0.000003,0.000000,0.000000,0.000000 | > Kappa == max(s)/min(s) == 4.861940e+11 | > [EMAIL PROTECTED]:~/src/progs/C> | > | > and works because I override the faulty output from the first call to | > DGESDD. Once you comment that out, it terminates with a similar error as R: | > | > [EMAIL PROTECTED]:~/src/progs/C> ./lapack_dgesdd | > info 0 tmp 1.000000 != recd 603 | > ** On entry to DGEBRD parameter number 10 had an illegal value | > | > The program is below. The key overrride line is | > lwork = recsize; | > which 'fixes' things, ie suppresses the bug. Else it rears its head. | | It appears that this bug is a duplicate of #328890 which was filed in | 2005 against lapack3-dev but never (formally?) addressed. However, it
Indeed, well spotted. | also appears to somehow be fixed in the version of lapack3-dev in | experimental now. When I comment out the key line in Dirk's test, I get Ah, didn't look there myself. Yes, and this is Lapack 3.1 in experimental, and from I've seen the Lapack that R ships is also 3.1 -- so my vote would be to migrate this from experimental to unstable. | [EMAIL PROTECTED]:/tmp/Dirk$ gcc -O3 -Wall -o lapack_dgesdd | lapack_dgesdd.c -llapack | [EMAIL PROTECTED]:/tmp/Dirk$ ./lapack_dgesdd | info 0 tmp 603.000000 != recd 603 | 1.725883,0.321633,0.031039,0.001979,0.000088,0.000003,0.000000,0.000000,0.000000 | Kappa == max(s)/min(s) == 4.861929e+11 Nice -- the 603 we're expecting and getting with R's lapack. | I am sorry that my knowledge in this area is too limited to isolate what | change in the code would lead to this change in the behavior, but | hopefully someone else can. That's no excuse. I simply spent most of the day running through the code getting mad that I couldn't nail it down :) Perseverance eventually payed off. Thanks again for the bug report and the follow-up. I'll let Camm take it from here. Dirk | | Thanks, | Ben | | -----BEGIN PGP SIGNATURE----- | Version: GnuPG v1.4.6 (GNU/Linux) | | iD8DBQFHroZMzQDSXIcN85kRAlGUAJ48bQi0ixjdORGLkCPtFb7DulsMMgCaAxop | pnob8kAuAfvCXhwfJsLs328= | =CZNO | -----END PGP SIGNATURE----- -- Three out of two people have difficulties with fractions. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
