GL_KHR_robustness Lands For RadeonSI / NVC0, OpenGL 4.5 Changes Complete

[mannerov]

I really mean 'without'. The extension forces to use a table, that's what I understood.

[nhaehnle]

Which discussion are you referring to?

The 64-bit value in the extension can just be a GPU address that points to the descriptor. So you do have to maintain a table, but at least for GCN that's basically the same thing as the descriptor tables we have today.

AFAICS, the only potential issue with ARB_bindless_texture is that it's fairly easy to shoot yourself in the foot with it. All resident textures have to be reserved while a CS is in progress, which means that if you use the extension badly, you might end up forcing more work on the kernel (because buffer lists are longer).

[Linuxhippy]

I guess bindless textures could also be a win for glamor, 2d-apps love to blit tons of tiny pixmaps

[Linuxhippy]

Actually bindless textues are what is reffered to as "modern, low-overhead opengl" by many slides published by at least NVidia (and I guess also AMD).

I guess bindless textures would also help glamor - 2d apps love to blit tons of small pixmaps...

[mannerov]

It was an old discussion on irc or on the mailing list, but I've probably misunderstood things then.

[pal666]

There was discussion long ago between mesa devs on the topic, and the concensus was that it is designed to reduce cpu overhead, but increases gpu work, which is thus a bad idea unless cpu limited.

it's not hard to find game which is cpu-limited on mesa

[pal666]

see:

that post was from before mesa 12, so "next release" in it is 12, not 13

[treba]

that post was from before mesa 12, so "next release" in it is 12, not 13

Ahm right confused that somehow, sorry.

[pal666]

Ahm right

10:51 #dri-devel: < nha> arb_enhanced_layouts/execution/component-layout/vs-to-fs-array-interleave.shader_test passes \o/