Tuesday, September 25, 2012

Finally Holidays... and some GPU code!

During the next week I'll be visiting London.

In the meanwhile enjoy a code preview of A GPU Exercise part 1 and part 2, and stay tuned for part 3 :)

To get, compile, and test the code, clone my funproject repository on github.
git clone git://github.com/cybercase/funproject.git
To compile the code is required the NVIDIA CUDA Toolkit. Then run:
cd funproject
cmake .
If CUDA Toolkit was found by CMake, then the lev_distance target should be built into the experiments directory. To quickly test the performances run:
cd experiments
./lev_distance lev_distance.cu thread.h
The following result is achieved on my late 2010 MBA
lev_distance.cu size is 8716 bytes
thread.h size is 17818 bytes
elapsed time: 9.240 (s)
elapsed time: 0.872 (s)
Results matches!
Distance: 14332
Please note that this is an educational implementation, and several improvements can be done to exploit all the features of the underlying architecture...

But before talking about this... holidays! :)

No comments: