Productive OpenCL with Intel Xeon Phi Coprocessors
2 likes969 views
The document discusses how software libraries like ArrayFire can make programming with OpenCL and Intel Xeon Phi coprocessors more productive and easier. It notes that libraries reduce code, are optimized by experts, well-tested and maintained. ArrayFire in particular provides high performance, is easy to program, portable across devices, allows for easy scaling to multiple devices, and has a large community for support. It demonstrates how ArrayFire can be used for tasks like parallel matrix multiplication and 3D surface plotting with only a few lines of code.
![Scalability
Multi-Device is 1-line of code
array *y = new array[n];
for (int i = 0; i < n; ++i) {
deviceset(i); // change device
array x = randu(5,5); // add work to device’s queue
y[i] = fft(x); // more work in queue
}
// all devices are now computing simultaneously](https://image.slidesharecdn.com/accelereyes-at-intel-isc13-theater-130618110004-phpapp02/85/Productive-OpenCL-with-Intel-Xeon-Phi-Coprocessors-17-320.jpg)
Productive OpenCL with Intel Xeon Phi Coprocessors
- 1. Productive OpenCL with Intel Xeon Phi Coprocessors and ArrayFire by John Melonakos
- 2. Why Software Libraries Are Great Reduce many lines of code to one line Obsessively tuned by experts; faster than DIY Well-tested and maintained Continuously improving
- 3. 12 Minutes of Library Love Dusty servers = bad, very bad Software struggles lead to under utilization Everyone loses
- 4. Five Influencers (besides price) Portability Scalability Community ProgrammabilityPerformance
- 5. Faster Time- consuming SSE or AVX Slower Easy-to-use Performance & Programmability
- 6. Faster Time- consuming Writing Kernels SSE or AVX Slower Easy-to-use Performance & Programmability
- 7. Faster Time- consuming Writing Kernels Compiler Directives SSE or AVX Slower Easy-to-use Performance & Programmability
- 8. Faster Time- consuming Writing Kernels Using Libraries Compiler Directives SSE or AVX Slower Easy-to-use Performance & Programmability
- 9. Portability Flavors of portability HW options CPU fallback High-performance mobile computing Libraries can provide portability
- 10. Scalability Always start with one device Potential headaches of adding devices Performance hit Development complexity Libraries can make scaling easy
- 11. Community What do you do when bugs arise? Continuous refinement Someone to answer questions Libraries can have great community support
- 12. Benefits of Using a Library Development Documentation Test and QA Maintenance Porting TIME COST TIME COST ArrayFire eliminates hidden costs of software development Pain Pleasure
- 13. Beware of Bad Libraries Bad libraries are a time sink Ruin the “library” reputation
- 14. Here’s One Good Library
- 15. Performance & Programmability Super easy to program Highly optimized
- 16. Portability
- 17. Scalability Multi-Device is 1-line of code array *y = new array[n]; for (int i = 0; i < n; ++i) { deviceset(i); // change device array x = randu(5,5); // add work to device’s queue y[i] = fft(x); // more work in queue } // all devices are now computing simultaneously
- 18. Community Over 8,000 posts at http://forums.accelereyes.com Nightly library update releases Stable releases a few times a year v2.0 coming at the end of summer
- 19. Graphics Functions #include <arrayfire.h> using namespace af; int main() { // random 3d surface const int n = 256; while (1) { array x = randu(n,n); // 3d surface plot surface(x); } return 0; }
- 20. GFOR Parallel Loops BA(,,1:3)C(,,1:3) *= *= *= gfor (array i, 3) C(span,span,i) = A(span,span,i) * B; Parallel matrix multiplications (1 kernel launch)
- 21. Software Shop for CUDA & OpenCL Two ways to work with us: Use Hire our CUDA & OpenCL developers Code development; CUDA & OpenCL training
- 22. Take Home Messages Free version of ArrayFire http://accelereyes.com Need acceleration? Try a library!
- 23. Productive OpenCL with Intel Xeon Phi Coprocessors and ArrayFire by John Melonakos