One scenario is meteorological data analysis, where
non-linear operations
are run for huge data matrices. In this case, we would
be able to define the
operations as Lua codelets, and run them parallel on the
data. I guess only
Lua of the current scripting languages (and Forth) would
be small enough to
effectively fit in on the CUDA/OpenCL. Even there, some
limitations s.a. no
recursive calls may be too much. But we'll see.
Are you suggesting running Lua instructions in OpenCL?
like:
-- begin codelet
for i=1, 10000 do
for j=1, 10000 do
-- some big fat equation in pure lua that runs in
OpenCL
end
end
-- end codelet
What about making some form of OpenCL flavored shader as
in GLSL or Cg
shaders? Basically some C codelet that is dynamically
compiled and
run in OpenCL. I haven't read the docs in a while, but
from what I
remember from a presentation this summer, code can be
dynamically
compiled and executed.
wes
