When you develop for one GPU, you actually need to be able to be sure it works on all available architectures.
We have a a Buildbot which handles cross-compiling to several architectures, checks code for OpenCL-specific problems and does runtime-tests on each of the devices.
The service includes code-review on how to optimise the code for several architectures, while keeping the code maintainable. Of course we can also do this for you.