OpenCL Basics: Running multiple kernels in OpenCL

This series “Basic concepts” is based on GPGPU-questions we get via email more than once, or when the question is not clearly explained in the books. For one it is obvious, for the other just what they’re missing.

They say that learning a new technique is best done by playing around with working code and then try to combine it. The idea is that when you have Stackoverflowed and Githubed code together, you’ve created so many bugs by design that you’ll learn a lot if you make it work. When applying this to OpenCL, you quickly get to a situation that you want to run one.cl file and then another.cl file. Almost all beginner’s material discuss a single OpenCL-file, so how to do this elegantly?

The answer is in clCreateProgramWithSource, which takes one or more kernels. If multiple cl-files are compiled (or one big concatenated file), one Program-object is created. But how to use this?

We first need to talk about public and private as we know from C++, as this is in OpenCL cl files too. From the perspective of the host the ‘__kernel void‘ functions are public and the normal functions are private. After a Kernel-object for each kernel-function is created, all kernel-functions are callable. Now you can get an output C by running kernel1(A, B) and then kernel2(B, C).

Related Posts

FIA_F1_Austria_2018_Nr._33_Verstappen

The Art of Benchmarking

...  with your team. It is not the full book. There will be multiple blog posts coming in this series, which will be linked at the end ...

5yearsSC

Birthday present! Free 1-day Online GPGPU crash course: CUDA / HIP / OpenCL

...  goal is hands on. We'll provide templates for CUDA/HIP and OpenCL, where you focus on making the core compute part. The third goal ...

network-of-boxes

Problem solving tactic: making black boxes smaller

We are a problem solving company first, specialised in HPC - building software close to the processor. The more projects we finish, the more it's clea ...

stocks

Improving FinanceBench

...  QuantLib is a C++ library. Unfortunately, languages like OpenCL, CUDA, and OpenACC cannot directly operate on C++ data structures, ...