Please add ANGLE_instanced_arrays support to GLES2 wrapper

[floooh]

The GLES2 wrapper seems to be missing support for ANGLE_instanced_arrays:

• the GLES2 headers are missing the following prototypes:
  glVertexAttribDivisorANGLE()
  glDrawElementsInstancedANGLE()
  glDrawArraysInstancedANGLE()
• library_gl.js only seems to have support for the instancing functions in the desktop OpenGL wrapper

I tried copying the latest GL headers from here (http://www.khronos.org/registry/gles/api/GLES2/) to emscripten/system/include/GLES2, this worked for compilation, but the linker failed to find the ANGLE-postfixed GLES2 functions (since they are missing in library_gl.js).

Since WebGL is closer to GLES2 then Desktop GL I'd like to continue using the GLES2 headers and emulation wrapper for the emscripten port, which would require support for ANGLE_instanced_arrays though.

Thanks!
-Floh.

[kripken]

cc @juj

Sounds good, can you perhaps make a testcase for this?

[floooh]

Yes, I'll try to derive a testcase from one of the existing tests, might take until the weekend until I get around this though.

Also, ANGLE_instanced_arrays isn't implemented in FF on Windows right now, which is strange because it works on OSX, and also on Chrome/Windows. There's a fairly old ticket here: https://bugzilla.mozilla.org/show_bug.cgi?id=843673

Would be good to poke the right people about this ;)

[kripken]

I see @jdashg assigned it to himself today, so the right people have certainly been poked ;)

[kdashg]

The patch is ready to commit, so it should be in nightly by the weekend.
On Jul 10, 2014 5:44 PM, "Alon Zakai" [email protected] wrote:

I see @jdashg https://github.com/jdashg assigned it to himself today,
so the right people have certainly been poked ;)

—
Reply to this email directly or view it on GitHub
#2510 (comment).

[juj]

I can help here, or @floooh , did you already have a test in the works?

[floooh]

Hi @juj I didn't get around to write a test yet, I was planning to cobble something together on the weekend, so if you're more familiar writing emscripten GL tests you're welcome to take over :)

I was thinking that a general instanced-arrays test would make sense, which can be switched between Core-GL, ARB_instanced_arrays and ANGLE_instanced_arrays with a define (I think the extension is in Core-GL since 3.1 or 3.2). The implementation would be the same, only the method names for glVertexAttribDivisor(), glDrawElementsInstanced() and glDrawArraysInstanced() would be different. I'm not sure whether other extensions or Core-GL has more DrawXXXInstanced() variants (I think ANGLE/ARB_instanced_arrays only adds those 3 methods).

[floooh]

On further thought, I think handling ARB_instanced_arrays is not necessary, since the emscripten desktop GL wrapper already implements the Core-GL calls.

There's also another problem in the current Desktop-GL implementation (at least in my demo here: http://floooh.github.io/oryol/Instancing.html) with a runtime error because closure seems to remove some code (more details here: https://groups.google.com/forum/#!topic/emscripten-discuss/Lcm7RzIs7UY), you'll probably also stumble across this when running your test and use closure in the linker stage.

[AbigailBuccaneer]

Surely the GLES2 wrapper should expose EXT_instanced_arrays - the GLES2 extension - rather than ANGLE_instanced_arrays or ARB_instanced_arrays?

[floooh]

The GLES extension registry lists both EXT_instanced_arrays and ANGLE_instanced_arrays (and a third NV_instanced_arrays). I think the reason that there is a separate ANGLE extension is that this has slight under-the-hood limitations because it needs to be compatible with D3D9. So the right way would be to expose both EXT_instanced_arrays and ANGLE_instanced_arrays I think.

I think we can skip the NV extension because this was very likely the blueprint for the EXT extension.

The official GLES2 extension header (http://www.khronos.org/registry/gles/api/GLES2/gl2ext.h) has all 3 instanced_arrays extensions.

[AbigailBuccaneer]

Ah, I didn't realise that ANGLE_instanced_arrays existed as a GLES extension as well as a WebGL extension. My bad. I guess if there are subtle differences or limitations then only the ANGLE extension should be exposed.

From a quick scan through, there are very minor differences: the ANGLE extension contains an extra limitation, as you mentioned:

INVALID_OPERATION is generated by DrawArraysInstancedANGLE or DrawElementsInstancedANGLE if there is not at least one enabled vertex attribute array that has a divisor of zero and is bound to an active generic attribute value in the program used for the draw command."

[stale]

This issue has been automatically marked as stale because there has been no activity in the past 2 years. It will be closed automatically if no further activity occurs in the next 7 days. Feel free to re-open at any time if this issue is still relevant.

[floooh]

FWIW I think this is properly fixed for quite a while now. All the required ANGLE extension prototypes are in the GLES2/gl2ext.h header.

IMHO the ticket can be closed as fixed.