You are here: home | sciGPGPU | 2.0

sciGPGPU

Gpu Computing for Scilab

(7833 downloads for this version - 21058 downloads for all versions)

Details

Version

2.0

Authors

Cedric Delamarre

Vincent LEJEUNE, Allan CORNET, Cedric DELAMARRE

Owner Organization

Scilab Enterprises

Maintainers

C?dric DELAMARRE

Vincent COUVERT

Categories

Linear algebra

Numerical Maths

Website

http://forge.scilab.org/index.php/p/sciCuda/

License

CeCILL

Creation Date

September 11, 2013

Source created on

Scilab 5.4.x

Binaries available on

Scilab 5.4.x:

Linux 64-bit

Windows 64-bit

Linux 32-bit

Windows 32-bit

Install command

--> atomsInstall("sciGPGPU")

Report a bug

GitLab's issues for sciGPGPU

Description

            This module provides GPU computing capabilities to Scilab. 
It uses an implementation of BLAS (cuBLAS) and FFT (cuFFT) through gpuAdd,
gpuMult, gpuFFT and other functions.
This module uses essentially Cuda but some functions, as gpuBuild, have been
created for build and use kernels developed with OpenCL or Cuda.

Binary versions have been built with Cuda 4.0.

If you need the OpenCL version of this module, please send an email to
contact@scilab-enterprises.com.

Files (5)

[428.33 kB]

sciGPGPU-2.0-1-src.tar.gz

Source code archive

Upload date : 2016-02-11 10:28:58
MD5         : 4866755d130612e4ff8394ee0ec33de1
SHA1        : 46398d31fbf855ea63bedc8add123673c49f57eb
Downloads   : 1794

[648.48 kB]

sciGPGPU_2.0.bin.x64.linux.tar.gz

Linux 64-bit binary for Scilab 5.4.x

Upload date : 2016-02-11 10:27:24
MD5         : 2639708ddc2b35d28a03ab893550c94d
SHA1        : e1a2b67366f1826c8f790c68bc9f0ee434a2644f
Downloads   : 1882

[626.45 kB]

sciGPGPU_2.0.bin.x64.windows.zip

Windows 64-bit binary for Scilab 5.4.x

Upload date : 2016-02-11 10:27:42
MD5         : 0c95ea6e023055bf80605eb0c004ae6c
SHA1        : b640745edbf5c1baee4d0d982768736c45d57247
Downloads   : 1833

[622.88 kB]

sciGPGPU_2.0.bin.x86.linux.tar.gz

Linux 32-bit binary for Scilab 5.4.x

Upload date : 2016-02-11 10:28:06
MD5         : e9f42c49b292db08f46e3c727fbcb578
SHA1        : 87e71a69bc8ea430e08f37f23fae7d25d02f6546
Downloads   : 1256

[594.86 kB]

sciGPGPU_2.0.bin.x86.windows.zip

Windows 32-bit binary for Scilab 5.4.x

Upload date : 2016-02-11 10:28:25
MD5         : 08b70775e5896243ef2df99cb0e45e9d
SHA1        : c26684049bcbdd60379a06986cec4c1c8f4700da
Downloads   : 1068

News (0)

Comments (10) ↑ Leave a comment

Comment from Anker Josefsen -- November 25, 2013, 02:07:27 PM

This comment has been deleted.

Comment from Anker Josefsen -- November 28, 2013, 11:17:11 PM

This comment has been deleted.

Comment from Moldovan Eugen -- December 3, 2013, 05:18:42 PM

Hi Anker,

I've followed your steps but I still have problems with the compiler. 

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 Visual Studio compiler.
 
-->findmsvccompiler()
 ans  =
 
 msvc100express  

Do you have any idea how to fix this?
Thanks,
Eugen

Comment from Moldovan Eugen -- December 5, 2013, 08:50:46 AM

Forgot to mention: win7, 64bit, Nvidia Tesla C1060, scilab 64bit 5.4.1

Eugen

Comment from Anker Josefsen -- December 5, 2013, 04:06:28 PM

This comment has been deleted.

> Hi Eugen,
> 
> Don't rely on me as en expert in this, this is just the experience I had setting up the
> system (-and I really just want to get to playing with my algo's). I noticed that the
> links in the post above point to some general windows homepage even through they was
> copied from where I got the stuff. Also I noticed that even through the first example
> on
> the sciGPGPU homepage did run, the second one using gpuBuild did not.
> 
> For a 64 bits os/scilab best choise for a free compiler that is visual studio 2010 c++
> ekspress with windows sdk 7.1. 
> 
> Both of these needs to have service packs installed, and the installation order
> matters.
> Follow the guide lines in
> http://blogs.msdn.com/b/heaths/archive/2011/04/01/visual-c-2010-sp1-compiler-update-for-the-windows-sdk-7-1.aspx.
> 
> 
> In me the web installer for sdk 7.1 did not work, I had to download the iso's. Note
> that
> three versions exists, and you have to pick the right one, in my case the amd64 bits
> version. 
> 
> Also, after using windows program remove in order to do the correct installation, the
> registry entry in windows was messed up resulting in error's when I tried to do a clean
> installation. I was lucky to find a description of a regedig procedure somewhere that
> did
> work (I forgot where it was).
> 
> After setting up the compiler you have to select sdk 7.1 as the current version for
> visual
> studio. Open up windows sdk 7.1 configuration command prompt (-not the configuration
> tool
> that have bugs that is triggered depending on OS language !). Go to c:\Program
> Files\Microsoft SDKs\v7.1\Setup. Run WindowsSdkVer.exe -version:v7.1
> 
> Finally some path is messed up in the cuda. Follow the guidelines for creating a bat
> file
> outlined here:
> http://stackoverflow.com/questions/2970493/cuda-linking-error-visual-express-2008-nvcc-fatal-due-to-null-configuratio
> 
> You should now be able to run examples listed on sciGPGPU homepage.
> 
> Best regards,
> Anker

Comment from Anker Josefsen -- December 5, 2013, 04:21:27 PM

This comment has been deleted.

> Hi Eugen,
> 
> Don't rely on me as en expert in this, this is just the experience I had setting up the
> system (-and I really just want to get to playing with my algo's). I noticed that the
> links in the post above point to some general windows homepage even through they was
> copied from where I got the stuff. Also I noticed that even through the first example
> on
> the sciGPGPU homepage did run, the second one using gpuBuild did not.
> 
> For a 64 bits os/scilab best choise for a free compiler that is visual studio 2010 c++
> ekspress with windows sdk 7.1. 
> 
> Both of these needs to have service packs installed, and the installation order
> matters.
> Follow the guide lines in
> http://blogs.msdn.com/b/heaths/archive/2011/04/01/visual-c-2010-sp1-compiler-update-for-the-windows-sdk-7-1.aspx.
> 
> 
> The web installer for sdk 7.1 did not work, I had to download the iso's. Note that
> three
> versions exists, and you have to pick the right one, in my case the amd64 bits version.
> 
> 
> Also, after using windows program remove in order to do the correct installation, the
> registry entry in windows was messed up resulting in error's when I tried to do a clean
> installation. I was lucky to find a description of a reg-edit procedure somewhere that
> did
> work (I forgot where it was).
> 
> After setting up the compiler you have to select sdk 7.1 as the current version for
> visual
> studio. Open up windows sdk 7.1 configuration command prompt (-not the configuration
> tool
> that have bugs that is triggered depending on OS language !). Go to c:\Program
> Files\Microsoft SDKs\v7.1\Setup. Run WindowsSdkVer.exe -version:v7.1 For some reasons
> the
> feed back was the vs 2008 was updated, even through only vs 2010 is installed on my
> system.
> 
> Finally some path is messed up in the cuda setup. Follow the guidelines for creating a
> .bat file outlined here:
> http://stackoverflow.com/questions/2970493/cuda-linking-error-visual-express-2008-nvcc-fatal-due-to-null-configuratio
> 
> You should now be able to run examples listed on sciGPGPU homepage.
> 
> Best regards,
> Anker

Comment from Moldovan Eugen -- January 9, 2014, 11:54:01 AM

Thanks Anker,

No luck after trying the additional instructions above. I think the toolkit is not mature
enough and I cannot afford to install a lower version of the CUDA drivers. I'm using the
GPU also for EM simulations and the CST license requires the latest CUDA drivers.

I'll watch the developments and maybe try again to fiddle with scilab in a year or so :(
Eugen

Comment from Anker Josefsen -- January 21, 2014, 09:51:37 PM

Eugen,

Eventually came to the same conclusion here. Currently I am programming directly in cuda
and compiling to a dll that scilab then link.

/Anker

Comment from Samarendra Hedaoo -- May 6, 2016, 11:18:01 AM

Has anyone got this compiled and working on linux 64-bit with cuda 7.5 (or something >=
6.5) ?

After fixing initial errors in build.sce trying to rename non-existent .cu.cpp.ii files, I
am not stuck on compiler errors:

  Running the Makefile
 !--error 10000 
ilib_compile: An error occurred during the compilation:
makecucomplex.cu.cpp: In function 'void writeToCucomplex(double*, double*, int, int,
cuDoubleComplex*)':
makecucomplex.cu.cpp:23:13: error: 'blockIdx' was not declared in this scope
     int x = blockIdx.x*blockDim.x + threadIdx.x;
             ^
makecucomplex.cu.cpp:23:24: error: 'blockDim' was not declared in this scope
     int x = blockIdx.x*blockDim.x + threadIdx.x;
                        ^
makecucomplex.cu.cpp:23:37: error: 'threadIdx' was not declared in this scope
     int x = blockIdx.x*blockDim.x + threadIdx.x;
                                     ^
makecucomplex.cu.cpp: In function 'void readInCucomplex(double*, double*, int, int,
cuDoubleComplex*)':
makecucomplex.cu.cpp:46:13: error: 'blockIdx' was not declared in this scope
     int x = blockIdx.x*blockDim.x + threadIdx.x;
             ^
makecucomplex.cu.cpp:46:24: error: 'blockDim' was not declared in this scope
     int x = blockIdx.x*blockDim.x + threadIdx.x;
                        ^
makecucomplex.cu.cpp:46:37: error: 'threadIdx' was not declared in this scope
     int x = blockIdx.x*blockDim.x + threadIdx.x;
                                     ^
makecucomplex.cu.cpp: In function 'cudaError_t rewritecucomplex(double*, int, int,
cuDoubleComplex*)':
makecucomplex.cu.cpp:73:45: error: 'cudaGetLastError' was not declared in this scope
     cudaError_t cudaStat = cudaGetLastError();
                                             ^
makecucomplex.cu.cpp:76:43: error: 'cudaGetDeviceProperties' was not declared in this
scope
     cudaGetDeviceProperties(&deviceProp, 0);
                                           ^
makecucomplex.cu.cpp:116:27: error: expected primary-expression before '<' token
         writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
                           ^
makecucomplex.cu.cpp:116:41: error: expected primary-expression before '>' token
         writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
                                         ^
makecucomplex.cu.cpp:121:42: error: 'cudaThreadSynchronize' was not declared in this scope
         cudaStat = cudaThreadSynchronize();
                                          ^
makecucomplex.cu.cpp: In function 'cudaError_t createcucomplex(double*, double*, int, int,
cuDoubleComplex*)':
makecucomplex.cu.cpp:140:45: error: 'cudaGetLastError' was not declared in this scope
     cudaError_t cudaStat = cudaGetLastError();
                                             ^
makecucomplex.cu.cpp:143:43: error: 'cudaGetDeviceProperties' was not declared in this
scope
     cudaGetDeviceProperties(&deviceProp, 0);
                                           ^
makecucomplex.cu.cpp:183:27: error: expected primary-expression before '<' token
         writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
                           ^
makecucomplex.cu.cpp:183:41: error: expected primary-expression before '>' token
         writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
                                         ^
makecucomplex.cu.cpp:188:42: error: 'cudaThreadSynchronize' was not declared in this scope
         cudaStat = cudaThreadSynchronize();
                                          ^
makecucomplex.cu.cpp: In function 'cudaError_t writecucomplex(double*, double*, int, int,
cuDoubleComplex*)':
makecucomplex.cu.cpp:211:45: error: 'cudaGetLastError' was not declared in this scope
     cudaError_t cudaStat = cudaGetLastError();
                                             ^
makecucomplex.cu.cpp:214:43: error: 'cudaGetDeviceProperties' was not declared in this
scope
     cudaGetDeviceProperties(&deviceProp, 0);
                                           ^
makecucomplex.cu.cpp:223:56: error: 'cudaMalloc' was not declared in this scope
         cudaMalloc((void**)&d, sizeof(double)*cols*rows);
                                                        ^
makecucomplex.cu.cpp:228:74: error: 'cudaMemcpy' was not declared in this scope
         cudaMemcpy(d, h, rows*cols*sizeof(double), cudaMemcpyHostToDevice);
                                      at line     129 of function ilib_compile called by : 

at line      94 of function ilib_for_link called by :  
at line      28 of function tbx_build_src called by :  
at line     137 of function builder_src_cu called by :  
builder_src_cu();
at line     148 of exec file called by :    

at line      13 of function tbx_builder called by :  
at line      49 of function tbx_builder_src_lang called by :  
tbx_builder_src_lang(languages, sci_src_dir);
at line      10 of exec file called by :    

at line      13 of function tbx_builder called by :  
at line      32 of function tbx_builder_src called by :  
at line     172 of function main_builder called by :  
    main_builder();
at line     304 of exec file called by :    
exec builder.sce

Please let me know if you have got it working.

> Has anyone got this compiled and working on linux 64-bit with cuda 7.5 (or something
> >= 6.5) ?
> 
> After fixing initial errors in build.sce trying to rename non-existent .cu.cpp.ii files,
> I
> am not stuck on compiler errors:
> 
>   Running the Makefile
>  !--error 10000 
> ilib_compile: An error occurred during the compilation:
> makecucomplex.cu.cpp: In function 'void writeToCucomplex(double*, double*, int, int,
> cuDoubleComplex*)':
> makecucomplex.cu.cpp:23:13: error: 'blockIdx' was not declared in this scope
>      int x = blockIdx.x*blockDim.x + threadIdx.x;
>              ^
> makecucomplex.cu.cpp:23:24: error: 'blockDim' was not declared in this scope
>      int x = blockIdx.x*blockDim.x + threadIdx.x;
>                         ^
> makecucomplex.cu.cpp:23:37: error: 'threadIdx' was not declared in this scope
>      int x = blockIdx.x*blockDim.x + threadIdx.x;
>                                      ^
> makecucomplex.cu.cpp: In function 'void readInCucomplex(double*, double*, int, int,
> cuDoubleComplex*)':
> makecucomplex.cu.cpp:46:13: error: 'blockIdx' was not declared in this scope
>      int x = blockIdx.x*blockDim.x + threadIdx.x;
>              ^
> makecucomplex.cu.cpp:46:24: error: 'blockDim' was not declared in this scope
>      int x = blockIdx.x*blockDim.x + threadIdx.x;
>                         ^
> makecucomplex.cu.cpp:46:37: error: 'threadIdx' was not declared in this scope
>      int x = blockIdx.x*blockDim.x + threadIdx.x;
>                                      ^
> makecucomplex.cu.cpp: In function 'cudaError_t rewritecucomplex(double*, int, int,
> cuDoubleComplex*)':
> makecucomplex.cu.cpp:73:45: error: 'cudaGetLastError' was not declared in this scope
>      cudaError_t cudaStat = cudaGetLastError();
>                                              ^
> makecucomplex.cu.cpp:76:43: error: 'cudaGetDeviceProperties' was not declared in this
> scope
>      cudaGetDeviceProperties(&deviceProp, 0);
>                                            ^
> makecucomplex.cu.cpp:116:27: error: expected primary-expression before '<' token
>          writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
>                            ^
> makecucomplex.cu.cpp:116:41: error: expected primary-expression before '>' token
>          writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
>                                          ^
> makecucomplex.cu.cpp:121:42: error: 'cudaThreadSynchronize' was not declared in this
> scope
>          cudaStat = cudaThreadSynchronize();
>                                           ^
> makecucomplex.cu.cpp: In function 'cudaError_t createcucomplex(double*, double*, int,
> int,
> cuDoubleComplex*)':
> makecucomplex.cu.cpp:140:45: error: 'cudaGetLastError' was not declared in this scope
>      cudaError_t cudaStat = cudaGetLastError();
>                                              ^
> makecucomplex.cu.cpp:143:43: error: 'cudaGetDeviceProperties' was not declared in this
> scope
>      cudaGetDeviceProperties(&deviceProp, 0);
>                                            ^
> makecucomplex.cu.cpp:183:27: error: expected primary-expression before '<' token
>          writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
>                            ^
> makecucomplex.cu.cpp:183:41: error: expected primary-expression before '>' token
>          writeToCucomplex<<<grid, block>>>(d,di,rows,cols,d_data);
>                                          ^
> makecucomplex.cu.cpp:188:42: error: 'cudaThreadSynchronize' was not declared in this
> scope
>          cudaStat = cudaThreadSynchronize();
>                                           ^
> makecucomplex.cu.cpp: In function 'cudaError_t writecucomplex(double*, double*, int,
> int,
> cuDoubleComplex*)':
> makecucomplex.cu.cpp:211:45: error: 'cudaGetLastError' was not declared in this scope
>      cudaError_t cudaStat = cudaGetLastError();
>                                              ^
> makecucomplex.cu.cpp:214:43: error: 'cudaGetDeviceProperties' was not declared in this
> scope
>      cudaGetDeviceProperties(&deviceProp, 0);
>                                            ^
> makecucomplex.cu.cpp:223:56: error: 'cudaMalloc' was not declared in this scope
>          cudaMalloc((void**)&d, sizeof(double)*cols*rows);
>                                                         ^
> makecucomplex.cu.cpp:228:74: error: 'cudaMemcpy' was not declared in this scope
>          cudaMemcpy(d, h, rows*cols*sizeof(double), cudaMemcpyHostToDevice);
>                                       at line     129 of function ilib_compile called by
> :  
> at line      94 of function ilib_for_link called by :  
> at line      28 of function tbx_build_src called by :  
> at line     137 of function builder_src_cu called by :  
> builder_src_cu();
> at line     148 of exec file called by :    
> 
> at line      13 of function tbx_builder called by :  
> at line      49 of function tbx_builder_src_lang called by :  
> tbx_builder_src_lang(languages, sci_src_dir);
> at line      10 of exec file called by :    
> 
> at line      13 of function tbx_builder called by :  
> at line      32 of function tbx_builder_src called by :  
> at line     172 of function main_builder called by :  
>     main_builder();
> at line     304 of exec file called by :    
> exec builder.sce
> 
> Please let me know if you have got it working.

Comment from Samarendra Hedaoo -- May 6, 2016, 11:18:50 AM

My scilab version is 5.5.2

Answer from Nik Babichev -- May 20, 2016, 08:06:23 PM

> My scilab version is 5.5.2

I try to install it under windows, CUDA 7.5, but unfortunately.
In requirements written Nvidia : CUDA Toolkit 3.2 to 4.0