782500 - nvidia Tegra 3 drivers may give bogus errors on unused fragment shader varyings

Status: RESOLVED WORKSFORME

(Core :: Graphics: CanvasWebGL, defect)

Description

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

I don't have a testcase for this yet, but it should be easy enough to write.  I'm seeing something on a much larger codebase that makes me suspect that the Tegra drivers at least on the Nexus 7 (Version: OpenGL ES 2.0 14.01002) aren't following the GLSL ES spec (v1.0.17) section 4.3.5, which explicitly states:

"Only those varying variables statically used (i.e. read) in the fragment shader must be declared in the vertex shader; declaring superfluous varying variables in the vertex shader is permissible."

I suspect there's an unused varying declared in a fragment shader, and no corresponding declaration in the vertex shader... and linkProgram is saying:

L0007: Cannot find vertex shader varying for declared fragment shader varyingworld

Comment 1

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

If this ends up being the case, we should notify nvidia (and also tell them to add a space before their "%s" of the varying name :)... but we may want to tighten up the restriction via ANGLE and require no superfluous declarations.

Comment 2

[Kelsey Gilbert [:jgilbert]]

Indeed. It sounds like we should add a conformance test for this, as well.

I don't think we can tighten anything in ANGLE, because this flexibility is a feature. It might be possible to inject an missing varyings at link time, though.

Comment 3

[Kelsey Gilbert [:jgilbert]]

(In reply to Jeff Gilbert [:jgilbert] from comment #2)
> Indeed. It sounds like we should add a conformance test for this, as well.
> 
> I don't think we can tighten anything in ANGLE, because this flexibility is
> a feature. It might be possible to inject an missing varyings at link time,
> though.

s/an/any

Comment 4

[Kelsey Gilbert [:jgilbert]]

Attachment: testcase, green if unused varyings are ignored. Red and errors otherwise.

Comment 5

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Hmm. This goes green on the Nexus 7 I mentioned earlier; the driver reports the same version (14.01002), though it's now running Android 4.2.1 where when I tested it earlier it was running 4.1 (I think).  Of course, I can't remember where I ran into this in the first place :/

Comment 6

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Green on my 4.1 HTC One X (T3 version) too.  Hrrrm.

Comment 7

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

I'm going to resolve this as WORKSFORME; I remember the thing that drove me to look at this generating shaders via a pretty complex structure.  I was pretty sure I looked at the final generated shaders, but now I'm doubting myself.