[PyCUDA] c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory

[Rana, Sanjay]

Hi Everyone,

I have an identical set up of pycuda on two computers as follow :

Windows 10 64bit
CUDA 8.0
Pycuda 2017.1+cuda8061-cp27-cp27m-amd64.whl
Visual Studio 2013 Community

This works perfectly on one of the computers but on the other computer, pycuda fails and reports the issue "c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory"

I have attached the script, script run log, and nvcc.profile.

Is there something I am missing in the problematic installation? Or is the issue somewhere else e.g. windows config, some other interfering application e.g. antivirus etc.

I have ran the NVIDIA's CUDA sample applications and they work just fine.

Many thanks,

Sanjay


************************************************************************************
This email and any files transmitted with it are confidential and
intended solely for the use of the individual or entity to whom they
are addressed. If you have received this email in error please notify
the system manager.

This footnote also confirms that this email message has been swept by
Cisco ESA for the presence of computer viruses.
************************************************************************************

[Andreas Kloeckner]

Post by Rana, Sanjay
Hi Everyone,
Windows 10 64bit
CUDA 8.0
Pycuda 2017.1+cuda8061-cp27-cp27m-amd64.whl
Visual Studio 2013 Community
This works perfectly on one of the computers but on the other computer, pycuda fails and reports the issue "c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory"
I have attached the script, script run log, and nvcc.profile.
Is there something I am missing in the problematic installation? Or is the issue somewhere else e.g. windows config, some other interfering application e.g. antivirus etc.
I have ran the NVIDIA's CUDA sample applications and they work just fine.

Does nvcc work the same way on either computer? You could also try
building this branch:

https://gitlab.tiker.net/inducer/pycuda/merge_requests/3

That gets by without needing to call nvcc, which may well solve your
problem.

Andreas

[Rana, Sanjay]

Thanks Andreas for the response. How could I check whether nvcc is work the same way on both computers?

On the problematic installation, I did try running the nvcc from a command prompt. I only get the text "kernel.cu" as response when I had tried to run the following command:

nvcc --cubin -v -arch sm_52 -m64 -Ic:\python2713\lib\site-packages\pycuda\cuda c:\users\crome\appdata\local\temp\tmp4gztsd\kernel.cu

There was no error message reported but perhaps one should have been produced?

Would a log of the run from the successful installation help?

Thanks,

Sanjay Rana | Design Directorate
Rural Payments Agency | Room 200 | North Gate House | Reading | RG1 1AF
Jabber: 02077642065 | ext. 42065 | Email: ***@rpa.gsi.gov.uk
Follow us on Twitter @Ruralpay

-----Original Message-----
From: Andreas Kloeckner [mailto:***@informa.tiker.net]
Sent: 05 June 2017 14:25
To: Rana, Sanjay; '***@tiker.net'
Subject: Re: [PyCUDA] c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory

Post by Rana, Sanjay
Hi Everyone,
Windows 10 64bit
CUDA 8.0
Pycuda 2017.1+cuda8061-cp27-cp27m-amd64.whl
Visual Studio 2013 Community
This works perfectly on one of the computers but on the other computer, pycuda fails and reports the issue "c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory"
I have attached the script, script run log, and nvcc.profile.
Is there something I am missing in the problematic installation? Or is the issue somewhere else e.g. windows config, some other interfering application e.g. antivirus etc.
I have ran the NVIDIA's CUDA sample applications and they work just fine.

Does nvcc work the same way on either computer? You could also try building this branch:

https://gitlab.tiker.net/inducer/pycuda/merge_requests/3

That gets by without needing to call nvcc, which may well solve your problem.

Andreas

______________________________________________________________________
This email has been scanned by the Symantec Email Security.cloud service.
For more information please visit http://www.symanteccloud.com ______________________________________________________________________
************************************************************************************
This email and any files transmitted with it are confidential and
intended solely for the use of the individual or entity to whom they
are addressed. If you have received this email in error please notify
the system manager.

This footnote also confirms that this email message has been swept by
Cisco ESA for the presence of computer viruses.
************************************************************************************

[Andreas Kloeckner]

Post by Rana, Sanjay
Thanks Andreas for the response. How could I check whether nvcc is work the same way on both computers?

Compile some sample cuda code.

Post by Rana, Sanjay
nvcc --cubin -v -arch sm_52 -m64 -Ic:\python2713\lib\site-packages\pycuda\cuda c:\users\crome\appdata\local\temp\tmp4gztsd\kernel.cu
There was no error message reported but perhaps one should have been produced?

That sounds odd. Can you determine which nvcc is being picked up on the
PATH, or if you have multiple nvcc versions installed?

Andreas

[Rana, Sanjay]

Hi Andreas,



Okay - thanks for your help.



After several days of frustatation, I managed to get it working and the reason, which I suspected before while reading through the verbose log, but didn't think could be as simple as that and thus discarded, turns out to be just to do with the non-default value I had set in the registry for the "AutoRun" value of the "Command Processor" entry.



Apparently, nvcc only works with the default windows value for AutoRun i.e. "c:\users\.." ! Surely, a potential item for NVIDIA to enhance/eliminate?



Sanjay Rana | Design Directorate

Rural Payments Agency | Room 200 | North Gate House | Reading | RG1 1AF

Jabber: 02077642065 | ext. 42065 | Email: ***@rpa.gsi.gov.uk

Follow us on Twitter @Ruralpay



-----Original Message-----
From: Andreas Kloeckner [mailto:***@informa.tiker.net]
Sent: 06 June 2017 03:22
To: Rana, Sanjay; '***@tiker.net'
Subject: RE: [PyCUDA] c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory

Post by Rana, Sanjay
Thanks Andreas for the response. How could I check whether nvcc is work the same way on both computers?

Compile some sample cuda code.

Post by Rana, Sanjay
nvcc --cubin -v -arch sm_52 -m64
-Ic:\python2713\lib\site-packages\pycuda\cuda
c:\users\crome\appdata\local\temp\tmp4gztsd\kernel.cu
There was no error message reported but perhaps one should have been produced?

That sounds odd. Can you determine which nvcc is being picked up on the PATH, or if you have multiple nvcc versions installed?



Andreas





______________________________________________________________________

This email has been scanned by the Symantec Email Security.cloud service.

For more information please visit http://www.symanteccloud.com ______________________________________________________________________
************************************************************************************
This email and any files transmitted with it are confidential and
intended solely for the use of the individual or entity to whom they
are addressed. If you have received this email in error please notify
the system manager.

This footnote also confirms that this email message has been swept by
Cisco ESA for the presence of computer viruses.
************************************************************************************

[Andreas Kloeckner]

Hi Sanjay,

Post by Rana, Sanjay
After several days of frustatation, I managed to get it working and
the reason, which I suspected before while reading through the verbose
log, but didn't think could be as simple as that and thus discarded,
turns out to be just to do with the non-default value I had set in the
registry for the "AutoRun" value of the "Command Processor" entry.
Apparently, nvcc only works with the default windows value for AutoRun
i.e. "c:\users\.." ! Surely, a potential item for NVIDIA to
enhance/eliminate?

Good to hear that you solved your problem, and thanks for following up
and letting the group know what the issue was.

Andreas

[Rana, Sanjay]

Hi Andreas



Yeah. Also note that the AutoRun registry setting (i.e. the default folder that opens when cmd is run) also causes problem even building the C++(?) CUDA samples in the Microsoft Visual Studio IDE.

I think it's quite reasonable to have a non-default setting for cmd startup folder so perhaps you could let NVIDIA devs know of this minor issue.



Sanjay Rana

-----Original Message-----
From: Andreas Kloeckner [mailto:***@informa.tiker.net]
Sent: 07 June 2017 13:29
To: Rana, Sanjay; '***@tiker.net'
Subject: RE: [PyCUDA] c1xx : fatal error C1083: Cannot open source file: 'kernel.cu': No such file or directory



Hi Sanjay,

Post by Rana, Sanjay
After several days of frustatation, I managed to get it working and
the reason, which I suspected before while reading through the verbose
log, but didn't think could be as simple as that and thus discarded,
turns out to be just to do with the non-default value I had set in the
registry for the "AutoRun" value of the "Command Processor" entry.
Apparently, nvcc only works with the default windows value for AutoRun
i.e. "c:\users\.." ! Surely, a potential item for NVIDIA to
enhance/eliminate?

Good to hear that you solved your problem, and thanks for following up and letting the group know what the issue was.



Andreas



______________________________________________________________________

This email has been scanned by the Symantec Email Security.cloud service.

For more information please visit http://www.symanteccloud.com ______________________________________________________________________
************************************************************************************
This email and any files transmitted with it are confidential and
intended solely for the use of the individual or entity to whom they
are addressed. If you have received this email in error please notify
the system manager.

This footnote also confirms that this email message has been swept by
Cisco ESA for the presence of computer viruses.
************************************************************************************

[Andreas Kloeckner]

Hi Sanjay,

Post by Rana, Sanjay
Yeah. Also note that the AutoRun registry setting (i.e. the default
folder that opens when cmd is run) also causes problem even building
the C++(?) CUDA samples in the Microsoft Visual Studio IDE.
I think it's quite reasonable to have a non-default setting for cmd
startup folder so perhaps you could let NVIDIA devs know of this minor
issue.

I unfortunately don't have a more direct line to nvidia than anyone
else--just post to their forums.

https://devtalk.nvidia.com/

Andreas