Hi,
Thanks wenju for sharing your thoughts. But my guess is somewhat different(GUESS)
IIRC, i read somewhere that AMD hardware has the capability of running multiple kernels simultanously, although it was not exposed properly. If such is the case multiple kernels must be able to atleast reside in the the GPU at a time.
Also if kernels are small they should easily fit in there.
Anyways it is just a guess and it would be good if Micah or Lee Howes can shed some light here.
Also it is been a long time multiple kernel execution is not supported. I hope you are working on it