Does the matconvnet need vl_complienn every time?

[mingtop]

I don't know why ,
Every time I restarted the computer, or I transfer the hole project from sever . I have to re-complie the matconvnet. If I don't complie the matlab crash.
Is it a bug?

[vedaldi]

Hi, restarting should not affect it unless something serious changes in your configuration.

Transferring from another machine may have unpredictable effects. Binary-level compatibility is probably pretty difficult to achieve with CUDA and MATLAB together.

On 25 Jul 2015, at 02:58, Yan Ming notifications@github.com wrote:

I don't know why ,
Every time I restarted the computer, or I transfer the hole project from sever . I have to re-complie the matconvnet. If I don't complie the matlab crash.
Is it a bug?

—
Reply to this email directly or view it on GitHub #217.

[mingtop]

@vedaldi Thank you very much ! Maybe the Tesla is double and my PC is single.
I don't change any thing on my local PC. But after I complied the vl_comple it works well.
It's not a big problem.
Thank you again.

[zhimingluo]

when using the version 1.3, I also meet this matlab carsh problem when matlab is restart.
I find that before using the toolbox , just run the command gpuDevice first and then the problem fixed.

[vedaldi]

Hi, I experienced similar issues as well. I think this happens if the MEX files are linked against a different version of CUDA than the one MATLAB is using (which is common). Indeed initialising with gpuDevice seems to do the trick.

On 27 Aug 2015, at 23:03, Zhiming Luo notifications@github.com wrote:

when using the version 1.3, I also meet this matlab carsh problem when matlab is restart.
I find that before using the toolbox , just run the command gpuDevice first and then the problem fixed.

—
Reply to this email directly or view it on GitHub #217 (comment).

[mingtop]

@zhimingluo Thank you very much!!