cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

pszilard
Adept I

CodeXL Linux-1.0.1042.0-x86_64 segfaults

Hi.

I'm running v1.0.1042.0-x86_64 on Ubuntu Linux 12.10 x86_64 and trying to test profiling of a C application. However, CodeXL crashes with the familiar error:

terminate called after throwing an instance of 'std::length_error'

  what():  basic_string::_S_create

Here's the core file: http://db.tt/0lnLgvKc

Please AMD folks fix this, it looks like it's a pathetic small issue.

Cheers,

Szilard

0 Likes
11 Replies
pszilard
Adept I

PS: if I set an invalid path as a working directory, I get a message saying that I should set the working directory, instead of suggesting that what I set does not exist!

Message was edited by: Doron Ofek, to keep the language appropriate for an open forum.

0 Likes
dorono
Staff

Hi Szilard,

Thanks for using CodeXL. Can you share the specific session type you used for profiling when the exception occurred?

What were you doing when the exception occurred?

0 Likes

Hi,

No matter which profiling type I use, after the execution of the analyzed binary finishes, CodeXL seems to be doing some post-processing which is interrupted by a segfault followed by either the aforementioned or the following message:

terminate called after throwing an instance of 'std::bad_alloc'

  what():  std::bad_alloc

Profiler output is generated (imd, caperf, ebp, & css files).

Note that I can't do sampling-based profiling either, I get a "The driver failed to start profiling. (error code 0x8000ffff)"; yet another brilliant error message.

Cheers,

Szilard

PS:

"Message was edited by: Doron Ofek, to keep the language appropriate for an open forum."

Are you censoring messages and publishing them modified without my consent!? The text of my message might have been strong, but definitely not a pile of curses. Your excuse to substantiate your a pitiful act of "editing" is simply lame. What is this, the forum of the Vatican!?

You can be sure that next time I'll think twice before wasting my time with bug reports! That's not only because your censorship pisses me off, but also because your software engineering dept. needs to get its !@#$% together. Releasing v1.0 software full of ridiculous bugs is something to think of before you get crushed by the big blue (and green).

0 Likes

Turning off unwinding gets rid of the issue.

0 Likes

Hi,

Thanks for the additional information. I can assure you the Developer Tools team takes every bug very seriously. No software is completely bug-free but we're making a great effort to provide a stable and useful product while pushing forward the functionality and added value to developers.

This forum is a platform for exchanging opinions, knowledge and mutual help for developers both in and out of AMD. As moderator I strive to keep the forum a proper environment for achieving these goals. Criticism is welcome and will never be censored, but style matters. Do note that other users flag material as questionable when offensive language is used as well as for other reasons (spam, commercial links and other types of abuse).

Respectfully,

0 Likes

Doron & others,

I'm very disappointed. Theannoying call stack/unwinding bug is, nearly four months and a minor release after the initial report, still present. IMHO, this is definitely not a sign of taking severe bugs seriously, especially that you did not even update me/the developer community on this matter, nor posted a link to a publicly available bugtracker or something.

Regarding your previous comments, I have one thing to say: censoring blatantly without permission != forums moderation. Please read up on censorship if you disagree.

Regards,

Sz.

0 Likes

pszilard wrote:

Note that I can't do sampling-based profiling either, I get a "The driver failed to start profiling. (error code 0x8000ffff)"; yet another brilliant error message.

This was discovered to be an issue with Instruction Based Sampling not working on Linux Kernel version 3.5.5 or later. This issue is fixed in CodeXL 1.1. The notice of this fix is accidentally missing from the 1.1 release notes.

There seems to be the same problem with 3.8 kernels, will that be fixed also? do you have an estimate on when?

0 Likes

Hi,

Thanks for this report.

While CodeXL target platforms do not yet include 3.8 Linux kernel, we ran a pilot using the CodeXL version of the upcoming release, and it is working well on Ubuntu 13.04 which carries a 3.8 Linux kernel.

I cannot discuss future release dates, however given the 3-4 expected releases of CodeXL per year, it should not be a long wait

0 Likes

Already Ubuntu 12.04 and others are using 3.11+ kernels and the only version which uses 3.5 kernel (Ubuntu 12.10) is being EOLed.

Does CodeXL fully support 3.11+ kernels already?

0 Likes
dorono
Staff

Hi,

There has been a lot of work done on the call stack unwinding mechanism for CodeXL 1.3. Can you let us know if this issue still occurs with CodeXL 1.3?

Thanks,

0 Likes