[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Public WebGL] Proposing OpenGL ES extensions for inclusion into WebGL

On Wed, Jan 4, 2012 at 12:53 AM, Kenneth Russell <kbr@google.com> wrote:
OES_texture_3D is an example of one which seems like a poor idea to
expose because of the market fragmentation it will imply -- any
application which requires it will work only on desktop hardware to
the best of my knowledge. Fortunately, the lack of native 3D texture
support in WebGL has inspired developers to seek solutions which work
on all platforms. See for example Vicomtech's excellent volume
rendering demonstration at http://demos.vicomtech.org/volren/ .

On Wed, Jan 4, 2012 at 1:09 AM, Gregg Tavares (wrk) <gman@google.com> wrote:
Sure they are not perfect and they are certainly slower than actual 3d textures but for color mapping they might not be prohibitively slow

Many of the extensions can be done in other ways (some cannot like blend minmax). It's true that 3D textures can be done in other ways, but this has two primary drawbacks.
#1 In order to compute the trilinear interpolation/wrapping it is necessary to perform 8 texel fetches in GLSL. The assumption is that if the hardware supports actual interpolated/wrapped 3D textures, that this would be significantly faster (somewhere on the order of 8x faster at least)
#2 I suspect cache coherency of actual 3D textures is better 

In the case of 3D textures like many extensions, it would trigger authors to write multiple rendering paths if they need it. Of course it means that relatively desktops get a performance boon and mobiles do not. But such is the fate of hardware that lacks extensions whose aim is to speed things up.

On Wed, Jan 4, 2012 at 12:53 AM, Kenneth Russell <kbr@google.com> wrote:
It would be most helpful for someone to research the support levels of
the extensions on this list on iOS 5 and Android on a few high end
devices. If support appears to be widespread, or if the extension
enables critical functionality unattainable any other way, then it
should be considered for exposure.
I would like to build some sort "Can I use this" for 3D capabilities. The idea is to quantify particular capabilities (like floating point textures) in small support matrices (maybe by major OS/GPU brand and total)
In order to do that I need some data that could be periodically pulled and parsed. I've asked http://www.glbenchmark.com/ if it would be possible to obtain an anonymized data dump yesterday. No reply so far. Other interesting sources for data would be the steam hardware surveys or mozillas crash database. Let me know if there's any data sources that you know can be queried.