ETC2 Texture Compression Looks Good For OpenGL

[MaxToTheMax]

AFAIK, while ETC2 must be supported by the OpenGL implementation, most major desktop graphics vendors will implement it in software. (I know for a fact that nVidia is doing it this way and plans to continue doing it this way for future hardware as well.) This allows them to be "OpenGL compliant," but without actually making ETC2 a viable usable choice.

This quote from the article is wrong:

Graphics texture compression reduces memory usage and avoids congesting the bus, thereby trying to avoid a performance bottleneck.

It should say:

Hardware-accelerated graphics texture compression reduces memory usage and avoids congesting the bus, thereby trying to avoid a performance bottleneck. When implemented in software, all it does is add another decoding step which the OpenGL driver must implement, wasting memory and not doing anything at all for bus congestion; this is similar to the reasons why you should always use GL_RGBA instead of GL_RGB, even when you're not using the alpha channel.

If we're going to wait for a texture compression scheme to get widespread and implemented commonly in hardware, we might as well wait for ASTC. From what I know it's better than ETC2.

[FourDMusic]

Here's a link to the slides without the Phoronix watermark: http://www.khronos.org/assets/upload...RAPH_Aug12.pdf

Is it even legal to have the watermark on the slides since it's copyrighted by Ericsson International??? Unless Michael received permission...

[elanthis]

If only games were developed for PCs and not ported from aging consoles...

The tide is ebbing, no worries.

[Kayden]

interesting reading, i would like to know if Sandy/Ivy Bridge will support it soon?

None of our currently shipping hardware has native support for ETC2, which means we have to implement it in software. That means that textures are uploaded and stored to the GPU in an uncompressed format, which isn't terribly efficient, but there's not much else we can do. That will at least make applications using ETC2 compressed textures work. (I believe Chad's volunteered to write the software decoder, unless someone beats him to it.)

[MaxToTheMax]

I realize the Intel driver department and hardware department may not communicate at this level, so you may not know this, but it's worth a shot: are you going to be prioritizing hardware ASTC or hardware ECT2 or both for future chipsets? It makes a difference for me personally in my work on Alien Arena, knowing what will and won't be hardware accelerated in the future.

[ptall]

Michael, I don't think you can say royalty free is the same as not patented. Ericsson probably has it completely covered but has been nice enough to let Khronos use it in their APIs.

[renox]

This is good but, when will wayland will full suport OpenGL so we can use it in he future?

Wayland is about sharing buffers between a client and a server, you can use whatever you want in the client to draw the buffer so Wayland is irrelevant here.
I repeat Wayland has *no* drawing API, and the article is about an improvement to a drawing API..

[Setlec]

None of our currently shipping hardware has native support for ETC2, which means we have to implement it in software. That means that textures are uploaded and stored to the GPU in an uncompressed format, which isn't terribly efficient, but there's not much else we can do. That will at least make applications using ETC2 compressed textures work. (I believe Chad's volunteered to write the software decoder, unless someone beats him to it.)

Thank you for the quick answer. you guys are impressing me!

[russofris]

None of our currently shipping hardware has native support for ETC2, which means we have to implement it in software. That means that textures are uploaded and stored to the GPU in an uncompressed format, which isn't terribly efficient, but there's not much else we can do. That will at least make applications using ETC2 compressed textures work. (I believe Chad's volunteered to write the software decoder, unless someone beats him to it.)

Kayden,

Is ETC2 arithmetically similar enough to a supported texture-compression format that you might be able to do partial acceleration via an intermediate format? Basically, take a loss on the CPU which would transcode ETC2 to the intermediate, and make up, or exceed the loss on the GPU?

The example that comes to mind is when we were able to get an old SigmaDesign hardware MPEG1 decoder to partially accelerate MPEG2 video.

F

[oibaf]

Is ETC2 really required by GL 4.3 as stated in the article? I don't see it listed here: http://cgit.freedesktop.org/mesa/mesa/tree/docs/GL3.txt