The plugins provided by the APIs of this code will boost the performance of matrix operations by using OpenCL.
The framework designed is to use the underlying GPU hardware available on your computer, to boost the performance of any piece of code, where parallelization is possible. API support for C/C++ would be provided first, and support for python will follow.The framework is undergoing development, is in the pre-alpha phase, and is available as per the University of Illinois/NCSA Open Source License.
Author: ASquared.
The framework can now be build on a mac/linux/windows machine, using the cmake tool. However, a GPU would be required with OpenCL support.Please make sure that the following libraries/tools are already installed:
1. OpenCL (With API support for NVIDIA GPU/AMD GPU/Intel GPU)
2. cmake
3. <Please add here, if I am missing something>
The following steps are required to run the kernel code on the GPU, on your machine:
1. Clone the boostbyparallel directory on the destination folder on local machine.
2. cd boostbyparallel
3. mkdir build
4. cd build
5. cmake ../src/
6. make
7. ./boostbyparallel
If all the above steps go through, you'll run the sample kernel code on the GPU.
Takes as input n = 10,
vector-a: 0, 1, 2, ..., 9.
vector-b: 0, 2, 4, ..., 18.
Output vector c is expected to be:
vector-c[i] = vector-a[i] + vector-b[i]
At the end, average of all elements of vector-c will also be printed out.
Kernels for the parallelization software are provided in the folder named, ./kernels.
This section will contain the support, I would be adding in the future. Like matrix multiplication sample for matrix operations.That piece of code would be extended soon.