Intel's Mesa Driver Is Very Close To Having FP64, Thereby OpenGL 4 Support

[Ancurio]

except because there IS a number change games and other programs will work by default and you won't have to tell your system to spoof a different level of GL. That's critical for game devs and very helpful for users

Nah, you're making it sound as if game devs cared about their games running under Mesa. If they did, all they'd have to do is fix their own code to only query extensions they actually use instead of a whole GL version. That's just a couple of lines of code, vs. the hundreds or even thousands that are required for complex extensions such as fp64.

[microcode]

This branch only has "Enable ARB_gpu_shader_fp64 for gen8+". In the bug entry there is an old statement

So I guess it's still coming later?

I suspect that makes a lot of sense for them. They've been putting this one off because it's daunting and famously unused. Supporting gen7 would likely make it even more daunting; and less likely to be completed in a timely fashion or at all.

My intuition, as a software developer who works in teams about the size of intel's, that gen7 will come faster if they let themselves deliver gen8+ ahead of others.

[haagch]

https://github.com/Igalia/mesa/commits/i965-fp64

This branch only has "Enable ARB_gpu_shader_fp64 for gen8+". In the bug entry there is an old statement

BTW, in case this hasn't been mentioned before: Jason and I agreed that it would be best to get fp64 working on gen8 first, and then add support for gen7 later. Gen7 is liable to have more bugs wrt fp64, and there is some weirdness around immediates, so it's better to focus on the easier gen first.

So I guess it's still coming later?

[SpyroRyder]

It'll just be a number change, and something for phoronix to write yet another "article" about. No games use double prec, and only few other applications do. Nothing will change.

except because there IS a number change games and other programs will work by default and you won't have to tell your system to spoof a different level of GL. That's critical for game devs and very helpful for users

[Ancurio]

It'll just be a number change, and something for phoronix to write yet another "article" about. No games use double prec, and only few other applications do. Nothing will change.

[microcode]

It's really time for Intel to get OpenGL 4 support. You can fake it and run some games but that does not work all the time. It usually can not complete against a powerful card but with 720p on lowest settings you can play basically the same games as with an entry level to midrage card. I hope they finish the missing extension for OpenGL 4.1 too soon.

This is a huge changeset, accounting for about 4,000 new lines of code to author and review, and touching 94(!) individual files. I think there's a good reason it's waited this long. It's easy to lean back and armchair this; but we'd be better off if each of us spent the time learning to contribute.

To be honest, there really aren't a WHOLE lot of applications which use doubles in their shaders (which is why you can usually report GL 4 safely); so it's easy to see why this extension waited so much longer than the many 4.x core extensions which have come first.

[Kano]

It's really time for Intel to get OpenGL 4 support. You can fake it and run some games but that does not work all the time. It usually can not complete against a powerful card but with 720p on lowest settings you can play basically the same games as with an entry level to midrage card. I hope they finish the missing extension for OpenGL 4.1 too soon.

[carewolf]

Sentence almost complete. Thereby headline.

[schmidtbag]

Finally - now all we need is vertex_attrib_64bit and we'll be on our way to full 4.2 support. I hope AMD catches up with shader_image_load_store and shader_atomic_counters first, though.

It'll be exciting for these OGL specs to get pushed out, because that should hopefully get devs to focus on performance and Vulkan.

[boffo]

4.5 seams so close, yet its so far.