matlab trouble [SOLVED]

stevyn · n00b Joined: 26 Aug 2004 Posts: 70 Location: USA

I'm having some trouble running Matlab 7R14. When I run it from the command line, I only get this information:

seidren · n00b Joined: 29 Sep 2004 Posts: 9

hmmm... I am not sure if this will help or not. But I couldnt get matlab working properly the first time because of nptl and I had to do call matlab like this.

stevyn · n00b Joined: 26 Aug 2004 Posts: 70 Location: USA

Thanks for your suggestion, but that didn't work. Do you have an NVIDIA graphics card?

seidren · n00b Joined: 29 Sep 2004 Posts: 9

No I dont have an Nvidia card. I have an Intel.
I googled around and found that this is because of the nvidia kernel module.... Apparently this occurs only in kernel 2.6. If you can give kernel 2.4 a shot.. it might work.
Also check if glibc is compiled with +nptl and -nptlonly. The -nptlonly option is important because then, gentoo installs botht the nptl enabled and disabled versions of glibc. Then you will be able to use 'LD_ASSUME_KERNEL=2.4.1 matlab' and make the program use the nptl disabled libraries. Its the 'tls' bit that makes me believe this is an nptl problem, but it may be not.

stevyn · n00b Joined: 26 Aug 2004 Posts: 70 Location: USA

I added the "-nptlonly" use flag and I'm recompiling glibc now. Is this the sort of thing that requires me to rebuild other packages, like the kernel? As for kernel 2.4, I'd rather temporarily jump back to the nv driver than switch kernels. I'll update after glibc is finished compiling.

caslca · Tux's lil' helper Joined: 24 Aug 2003 Posts: 85

Same problem here.... I get the splash screen, but desktop crashes (also matlab -nojvm won't run in cmd mode alone either)

I compiled glibc without nptlonly (as i had it before), but the LD_ASSUME.... thing still doesn't help - matlab crashes again.

My system is ~x86, with nptl glibc-2.3.4.20041102, gcc 3.4.3-20050110, and kernel 2.6.10-r6 gentoo-dev-sources.

I can confirm that matlab starts and some protions work if I use "nv" instead of "nvidia", but some portions still won't function (symbolic math for example throws errors related to glibc).

I can confirm that matlab works fine on a x86 system.

I can confirm that symbolic math(and seems everything else) works if i use "nv" and the LD_ASSUME thing.

my matlab build is 7.0.0.19901 (R14) May 06 2004

any help would be greatly appreciated!

stevyn · n00b Joined: 26 Aug 2004 Posts: 70 Location: USA

I'm able to run Matlab now if I run "opengl-update xorg-x11", but then I can't run anything with opengl. Opengl in matlab still doesn't work and X crashed if I run glxgears or opengl screensavers. It's barely a work around, but It's better than using the nv driver because I have a setting in my xorg.conf file that enlarges the fonts to make them readable on my display. So as long as I'm taking classes that require matlab, I'll have to deal with this I guess. I'd prefer just to remove matlab's dependency on opengl since I don't think I'll be needing it anyway.

CarloJekko · Posted: Thu Jun 16, 2005 3:59 pm Post subject:

You can run symbolic?
It doesn't work to me...
I think is a problem with glibc-2.3.5 and kernel 2.6...
_________________
C.

johntramp · Guru Joined: 03 Feb 2004 Posts: 457 Location: New Zealand

Hi I am having the exact same problem with nvidia and the same build of matlab.

When I ran opengl-update xorg-x11 it starts, as stevyn said and then the first thing I am greeted with is this.

raf · Apprentice Joined: 16 Jan 2005 Posts: 158

Hi,

The tls error you are getting is actually due to the fact that one of the patches from glibc-2.3.4 has not been applied to glibc-2.3.5 and in fact there is also a glic-2.0 compatibility issue with glibc-2.3.5. I have submitted bug reports on these and the nice people from gentoo have fixed both of these issues in the latest (still masked) version of glibc called glibc-2.3.5-r1.

So here is what you do:
1) add "glibc-compat20 nptl -nptlonly" to your use flags
2) Emerge glibc-2.3.5-r1 (might have to unmask it)
3) Before running these non nptl binaries such as matlab or Cadence run export "LD_ASSUME_KERNEL=2.4.19"

Now I'm able to run matlab just fine without having to change to xorg-x11 driver. I still get an error in matlab complaining about opengl libraries but matlab runs fine. As far as I can tell all figures appear, all functions work, so even though I'd rather not see the error at all things seem to run ok.

Let us know if this works for you.
_________________
-Raf

depontius · Advocate Joined: 05 May 2004 Posts: 3509

I'm glad I found this post, since Cadence stopped working for me, right after upgrading to glibc-2.3.5. Moving to glibc-2.3.5-r1 has indeed fixed my problems.

Now for the question about LD_ASSUME_KERNEL...

I've used this in the past, but at the moment am not. Cadence appears to start and run just fine on kernel 2.6, for me. I have had one snag in the past, however. When trying to edit a "config" view, *sometimes* after electing to either read or edit the hierarchy view, X would die. (especially after electing to edit both schematic and hierarchy views) I never had this happen under kernel 2.4, and it didn't always happen under kernel 2.6. But using the new glibc, I haven't had it happen yet today. I never tried using LD_ASSUME_KERNEL, since I didn't have a good guess about which kernel to assume.

Do you find LD_ASSUME_KERNEL necessary for Cadence? Why? How/where did you come up with 2.4.19?

By the way, today I found it necessary to install elf-utils, in order to get /usr/lib/libelf.so.1 available. That wasn't necessary prior to the glibc upgrade.
_________________
.sigs waste space and bandwidth

raf · Apprentice Joined: 16 Jan 2005 Posts: 158

Hi,

depontius · Advocate Joined: 05 May 2004 Posts: 3509

I have nptl in my USE flags, and Cadence has always run for me, without LD_ASSUME_KERNEL.

I would expect it to, since Cadence is "certified" on RHEL Workstation 3.0. That is a 2.4 kernel, but RedHat kernels have traditionally been heavily patched, and in particular have included nptl since RH9, IIRC. I used the hierarchy editor a fair amount today, with no X crashes. If/when I do get an X crash, I'll switch to LD_ASSUME_KERNEL and see if that helps.
_________________
.sigs waste space and bandwidth

depontius · Advocate Joined: 05 May 2004 Posts: 3509

Shortly after my last post, I got another first crash of the Cadence hierarchy editor. So I added LD_ASSUME_KERNEL, and still get the crashes. If there's any sort of remoteness to X, it hasn't crashed, yet. So it appears I've got 4 more experiments to do. I need to export the display through at tcp socket instead of a Unix socket. So I need to try the box address and loopback, with and without LD_ASSUME_KERNEL. I still believe this is a timing issue, and I'm guessing the different path of a tcp socket will change things significantly.
_________________
.sigs waste space and bandwidth