Add initial support for compute shaders.

[Roldak]

Hi!

This PR is to add the base support for compute shaders, which already allows performing arbitrary computation in SSBO objects on the GPU (see test). Binding some of the "memory barrier" API is necessary to make compute shaders usable, so I also did that.

A possible future evolution is to bind the Image units API, to allow compute shaders to perform arbitrary read/writes to textures (well, images), but I didn't need them personally yet.

Note: I added the GL.Compute and GL.Memory_Barriers packages because I couldn't think of an existing package that would be a good fit the new APIs. If you have better ideas on how to expose those I'll happily make the changes!

[flyx]

Thanks for this! I will hopefully find time to give this a proper review in the coming week. A quick thought would be that GL.Memory_Barriers.Memory_Barrier could be shortened to GL.Memory.Barrier without any loss on intent. Besides, could you add a wrapper for glMemoryBarrierByRegion since that would basically be the same as the one you did already? My rule of thumb is „always add the full extend of one API reference page, never just a part of it“ because otherwise, exotic alternatives that frequently exist will never get added.

[Roldak]

A quick thought would be that GL.Memory_Barriers.Memory_Barrier could be shortened to GL.Memory.Barrier without any loss on intent. Besides, could you add a wrapper for glMemoryBarrierByRegion since that would basically be the same as the one you did already?

Done!

I will hopefully find time to give this a proper review in the coming week

Sounds good!

[flyx]

Sorry for the long silence. I am still figuring out how to be able to continue maintaining this library since

• GNAT isn't natively available on aarch64-darwin
• OpenGL 4.2+ features aren't available on macOS

The situation is that I currently cannot test anything. There are solutions like Apple's Rosetta 2 to run GNAT (creating amd64 binaries which it can then run) but that's certainly not ideal (and doesn't get me newer OpenGL features). qemu seems to be able to run an amd64 Linux but that also is a layer that can introduce additional bugs. I'll have a solution eventually but this doesn't have much priority for me since it is for one library that doesn't have much activity. It may take some more time, sorry for that.

[flyx]

I'm afraid I have decided to stop maintaining OpenGLAda. This is my only active Ada project and I am not using it for anything, hence keeping up with the language and ecosystem is taking up too much of my time and I lost motivation.

I am very sorry for this. You can keep using your own fork, there won't be any upstream development happening anymore.