Krzysztof Sopyła – strona domowa Dla studentów, o moich badaniach.

cuSVM for CUDA 5.0 and Matlab x64

This page shows how to build cuSVM, GPU accelerated SVM with dense format. Library has been written by AUSTIN CARPENTER. The procedure use CUDA 5.0, MATLAB x64 and Visual Studio 2010. The code and project files were modified in order to compile and link library, many steps were taken from http://www.parallelcoding.com/2012/02/09/cusvm-in-visual-studio-2010-with-cuda-4-0/

Modifications:

1. Add matlab variables:
   1. cuSVMTrainIter – contains number of iteration the solver does
   2. cuSVMTrainObj –  contains the final objective function value after the trainning
2. In file cuSVMSolver.cu lines 869-874 all calls of cudaMemcpyToSymbol was changed, because of changes made in CUDA 5.0 runtime library – http://stackoverflow.com/questions/12947914/error-in-cudamemcpytosymbol-using-cuda-5
   before the change:
   mxCUDA_SAFE_CALL(cudaMemcpyToSymbol(„taumin”, &h_taumin, sizeof(float) ));
   after the change:
   mxCUDA_SAFE_CALL(cudaMemcpyToSymbol(taumin, &h_taumin, sizeof(float) ));
3. In functions FindBI, FindBJ, FindStoppingJ – change the way reduction in shared memory was done (http://stackoverflow.com/questions/6510427/cuda-finding-max-using-reduction-error)
4. The kernel cache size is constrained to 400MB, if you want bigger cache you can modify cuSVMSolver.cu line 24
   #define KERNEL_CACHE_SIZE (400*1024*1024)

Build Procedure

Download preconfigure cuSVM Visual Studio 2010 solution with LibSVM and matlab scritp for classification. 

All steps describe below are done, you have to check if all paths are set correctly and yours GPU computational capability is set properly.

My setup:

• windows 7 x64
• visual studio 2010
• CUDA 5.0
• Matlab R2011b
• the code was tested on GeForce GT 330M and Geforce GTX 690

Prerequisites:

Determine paths:

1. Matlab include path, mine is „D:\Program Files\MATLAB\R2011b\extern\include” (Matlab was installed on drive d:\)
2. Matlab library path: „D:\Program Files\MATLAB\R2011b\extern\lib\win64\microsoft”
3. CUDA toolkit include path: „C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v5.0\include”
4. GPU compute capability, mine is 1.2 in case of GeForce GT 330M(compute_12,sm_12), and 3.0 in case GeForce GTX 690 (compute_30,sm_30)

Eventually you can check if it is „Release” or „Debug” build.

How to use cuSVM

The zip package contains two folders:

• cuSVM – Visual Studio 2010 solution
• cuSVMmatlab – contains:

1. libsvm,
2. compile cuSVMTrain.mexw64 and cuSVMPredict.mexw64 in Lib folder,
3. sample datasets in data folder
4. matlab script cuSVMTest.m