[Outdated] Cleanroom Compute API

[Kuba663]

Leaving this here because the MVP is almost ready. This is untested, I will undraft when the unit tests are done and passing.

Cleanroom Compute API

Run OpenCL kernels from Cleanroom for maximum parallelism.

The goal

Have an easy to use API for running OCL code from Cleanroom, mod developers should be unable to cause memory and state leaks (if state leaks are even possible).

Code for running a kernel should look like this (as of now):

CommandQueue queue = Compute.instance().queueDispatch.dispatch("Queue1"); // Queues shoudl be re-used multiple times
queue.dispatchKernel(kernel, args, offsets, sizes).next(kernel2, args2, offsets2, sizes2).run();
queue.close();

The other goal is to prevent the inferno that would break out if multiple different modders start including their own OCL jar in their mods for optimization.
Also, OpenCL supports SoA so it can work with Kirino a tad better than compute shaders which require AoS conversion.

The whole thing will use JSON files in the resources to red metadata and automatically detect headers.

#include <mod/example.h> // this is ok
#include <example.h> // this is not

TODO

• Initializing OpenCL
• Compiling Programs
• Command Queues
• Kernel invokation
• Buffer and image creation (OCL seems to have problems with non-MemoryStack buffers)
• Handling cases where OCL is unavaliable
• Writing unit tests (priority)
• Cleanup code

Look into discussion #567 for more.

[Kuba663]

I just realized the target branch is wrong.

[Kuba663]

Superseded by #590