[SoC] Week 3 OpenCL Integration

[Markus Neteler]

Hi Seth,

On Tue, Jun 8, 2010 at 6:05 AM, Seth Price <seth at pricepages.org> wrote:
> 1) What did I do last week
>
> First, I worked on the API for integrating OpenCL into the GDAL code with my
> mentors. While that was going on, I began work on integrating OpenCL into
> GRASS's r.sun.
>
> r.sun is pretty complex, and there are global variables all over the place,
> so it's hard to sort out the memory. I also have all the r.sun structures
> being dumped into 3 arrays: one that is per-thread, one of int constants,
> and one of float constants. This is a bit tricky because I need to figure
> out which ones are constant, which I can put in constant cache and have a
> <10 cycle access time, vs the ones that are different per-thread in global
> memory, which have a >100 cycle access time. This is tricky because all the
> structures that are passed through the program are often initialized
> differently depending on which cell you're calculating (but not always)

Consider to discuss this on "grass-dev" for feedback.

If any changes are needed for the variable management, please submit them
as trac tickets and, if accepted, we can merge them into SVN asap.

> Another problem that I'm going to have is calculating the min/max
> sunrise/sunset, OpenCL doesn't lend itself well to comparing values between
> threads.
>
> Something else that's tricky about r.sun:
> Some of the global variables are no longer in use, they're only used in
> commented out code.

Sounds like clean-up desired!
Do you work in trunk (7) or GRASS 6.x?

> 2) What do I plan to do this week
>
> I will be continuing my work with the GDAL API, and perhaps implementing it.
> I'll also be continuing my work on r.sun.
>
> 3) Do I have any problems/obstacle which interfere with my work
>
> Does r.sun count? ;)

Sure :)

Markus