Scilab Home Page | Wiki | Bug Tracker | Forge | Mailing List Archives | Scilab Online Help | File Exchange
ATOMS : sciGPGPU details
Login with GitLab

sciGPGPU

Gpu Computing for Scilab
(3581 downloads for this version - 19597 downloads for all versions)
Details
Version
1.0
A more recent valid version with binaries for Scilab exists: 1.2
Authors
Cedric Delamarre
Vincent LEJEUNE, Allan CORNET, Cedric DELAMARRE
Owner Organization
Consortium Scilab
Maintainer
C?dric DELAMARRE
Categories
License
Creation Date
July 21, 2011
Source created on
No source archive has been provided for this toolbox version.
Binaries available on
Scilab 5.3.x:
Linux 64-bit Windows 64-bit Linux 32-bit Windows 32-bit macOS
Install command
--> atomsInstall("sciGPGPU")
Description
            This toolbox provides a gpu computing capabilities at Scilab. 
It uses an implementation of BLAS (cuBLAS) and FFT (cuFFT) through gpuAdd,
gpuMult, gpuFFT and other functions.
This toolbox uses essentially Cuda but some functions, as gpuBuild, have been
created for build and use kernels developed with OpenCL or Cuda.

link to sources : http://forge.scilab.org/index.php/p/sciCuda/
            
Files (5)
[482.81 kB]
Linux 64-bit binary for Scilab 5.3.x

[30.21 MB]
Windows 64-bit binary for Scilab 5.3.x

[466.76 kB]
Linux 32-bit binary for Scilab 5.3.x

[30.07 MB]
Windows 32-bit binary for Scilab 5.3.x

[6.68 MB]
macOS binary for Scilab 5.3.x

News (0)
Comments (4)     Leave a comment 
Comment -- September 27, 2011, 04:15:56 PM    
Hi,

I tried to install the module, but I got an error:


-->atomsInstall('sciGPGPU')
 ans  =
 
!sciGPGPU  1.0-0  allusers  SCI\contrib\sciGPGPU\1.0-0  I  !
 
-->atomsLoad('sciGPGPU')

Start sciGPGPU
	Load macros
	Load help
Shared archive loaded.
Link done.
Shared archive loaded.
Link done.
atomsLoad: An error occurred while loading 'sciGPGPU-1.0-0':
	Please install Nvidia Developer Drivers
 !--error 10000 

at line     337 of function atomsLoad called by :  
atomsLoad('sciGPGPU')

What should I do to make this work ?

Best regards,

Michaël Baudin
Answer from Cédric DELAMARRE -- September 28, 2011, 09:34:16 AM    
Hi Michael,

Before use this module, you need to have a Nvidia CUDA Card double precision capabilities.
Then, downlaod the driver developement and the CUDA toolkit available in Nvidia web site.
(http://developer.nvidia.com/cuda-toolkit-archive)
Finally, install them and try again to load the module.


Cédric Delamarre.
Answer -- September 28, 2011, 02:35:57 PM    
Hi Cédric,

Thank you for your help and instructions. I did what you suggest and get the following 
error:

Start sciGPGPU
	Load macros
	Load help
[...]
	Load gateways
atomsLoad: An error occurred while loading 'sciGPGPU-1.0-0':
	\tYour device not support double precision.

Anyway, that was worth trying.

Regards,

Michaël

> Hi Michael,
> 
> Before use this module, you need to have a Nvidia CUDA Card double precision
> capabilities.
> Then, downlaod the driver developement and the CUDA toolkit available in Nvidia web
> site.
> (http://developer.nvidia.com/cuda-toolkit-archive)
> Finally, install them and try again to load the module.
> 
> 
> Cédric Delamarre.
Comment from Sorin Beldie -- October 5, 2011, 05:46:10 PM    
Hello,

In the same note as Michael Baudin, I have installed the sciGPGPU in Scilab 5.3. 
Obviously, I made sure that my NVidia driver is up to date (developper version, for GPGPU 
support). Just as Michael Baudin, I have the same problem, with double precission. In the 
context that I am aware that my GT 330M NVidia card is only Cuda Compute 1.2 capable, and 
that double precission is only available for Cuda Compute 1.3 capable devices, can I 
still use the sciGPGPU module in Scilab? Or is double precission deeply embedded in this 
module and I can't use it with single precission?

Thank you in advance for answering.

Regards,

Sorin
Answer from Cédric DELAMARRE -- October 6, 2011, 09:29:49 AM    
Hi Sorin,

Indeed, this toolbox uses only double precision, so if your Nvidia card is not greater or
equal to Cuda compute 1.3 capable, you can't use the sciGPGPU module.

Best regards,

Cédric
Answer -- October 6, 2011, 10:02:44 AM    
This answer has been deleted.
Answer -- October 6, 2011, 10:05:31 AM    
Hi Sorin,

This is just a comment on the fact that this module uses doubles. Scilab is based on
doubles. This is why doubles are a 
mandatory option of this module. This cannot be otherwise, i.e. single precision numbers
could not be used in the 
sciGPGPU module, for this would cause a loss of precision during the exchanges of data
between Scilab and the sciGPGPU 
module. In the end, half of the precision of the original numbers would be silently be
lost. For most computations this 
may not change anything, but for some, this will change everything. For example, some
computations based on Scilab and 
on the sciGPGPU modules could be completely different.

Regards,

Michaël
Answer from Cédric DELAMARRE -- October 6, 2011, 10:40:31 AM    
This answer has been deleted.
Comment from rohit mehta -- November 9, 2011, 08:04:35 PM    
Hi I cannot load the sciGPGPU atom.  I have the NVIDIA drivers installed as well as CUDA
4.0.  Our hardware is a Tesla M2050 which supports double precision.  I get the following
error message.  We did install NVIDIA 64 bit drivers, but "file /usr/lib/..."
does call it
a 32-bit LSB shared object.

Start sciGPGPU
	Load macros
Shared archive loaded.
Link done.
Shared archive loaded.
Link done.
Link failed for dynamic library '/usr/lib/libcuda.so'.
An error occurred: /usr/lib/libcuda.so: wrong ELF class: ELFCLASS32
atomsLoad: An error occurred while loading 'sciGPGPU-1.0-0':
	Please install Nvidia Developer Drivers
 
-->

Answer from Cédric DELAMARRE -- November 10, 2011, 10:00:22 AM    
Hi,

On my computer the command "file /usr/lib/libcuda.so.270.41.19" return me :
ELF 64-bit LSB shared object.

Make sure that you have correctily install the dev driver 64 bit:

http://developer.download.nvidia.com/compute/cuda/4_0/drivers/devdriver_4.0_linux_64_270.41.19.run

Then, check that the file libcuda.so is a symbolic link to libcuda.so.1 that is a symbolic
link to libcuda.so.270.41.19

Regards,

Cedric
Comment from Anker Josefsen -- November 24, 2013, 09:21:05 PM    
Hi,

I am experiencing some problem setting up this great 'atoms'. After installing it I get
the following messages when starting Scilab:

Startup execution:
  loading initial environment

Start sciGPGPU
	Load macros
	Load help
Shared archive loaded.
Link done.
Shared archive loaded.
Link done.
atomsLoad: An error occurred while loading 'sciGPGPU-2.0-0':
	Please install CUDA Toolkit 3.2 or more.
	Please install Visual Studio compiler.
 
I am running windows 8.1 (64 bits) on a PC with a NVIDIA GeForce GTX 780. The Scilab
version is scilab-5.4.1 (64 bit).

The compiler I use is Visual studio 2008 prof. Scilab seems to see this compiler, the
scilab command findmsvccompiler() returns 'msvc90pro' that is the code for this compiler.
  
I have updated the graphic card driver recently.

I have installed cude 5.5 on my PC, and tested functionality by opening one of the samples
from the cuda sample browser in visual studio, compiled and executed it. I tried a simple
vector add example, that does the addition on gpu and cpu before comparing the result. In
my case, everything seems to run smoothly, and the test passed with a 'done' message.
  
Any suggestions on how to proceed from here will be highly appreciated.

Best regards,
Anker
Leave a comment
You must register and log in before leaving a comment.
Login with GitLab
Email notifications
Send me email when this toolbox has changes, new files or a new release.
You must register and log in before setting up notifications.