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

Re: [Public WebGL] Minefield brings system to halt



Just to clarify, there were two problems: one with shading in general, this now works with disabling the validator in Firefox Beta. The other one are these "need at least 12 bytes, but have only 0 ..." messages in combination with the system halt occurring after a while in Minefield nightly - this still exists :(


Am 21.09.2010 17:47, schrieb Benoit Jacob:
Yes, but that is precisely what is (in principle) implemented in the code that I pasted in this thread. Since disabling the shader translator fixed it for Yvonne, we concluded that the problem was probably with ANGLE. In principle the issues with attrib 0 on desktop GL are solved in Minefield (and have been solved for a longer time in WebKit).

Benoit

----- Original Message -----
I raised a problem a few weeks ago where attrib 0 has to be bound to
the vertex position. I got around this by simply ensuring that I
always
bind vertex position to attrib 0. Ken Russell/Vlad discussed this and
I
believe their intention was to ensure that ES2 behavior applied (which
allows any attribute to be bound to 0). I haven't removed my patch, so
I've no idea if this was resolved. If you search the list you should
find the conversations.

HTH

Alan


On 9/21/2010 8:23 AM, Benoit Jacob wrote:
----- Original Message -----
Hi Benoit,

thank you for your fast response.

Am 21.09.2010 16:30, schrieb Benoit Jacob:
Another thing that's worth trying is setting
webgl.shader_validator
to false, to see if it's an issue with our ANGLE copy or the way
that we're using it.
I first started testing in Firefox Beta: when disabling the shader
validator both mentioned demos (the shadows and the cube map/
multi-texture shader) both work again as expected :-) (though the
console still tells s.th. about "been bound to a different
target").

This really hints to an ANGLE problem, because I then double
checked
my
Chromium settings and remembered that there I have had to activate
the
"--use-gl=desktop" flag to make it work (Chrome Beta is ok
without)...
Ah! Very interesting.

I would like to report to ANGLE but this is quite nontrivial as we
would probably need a small testcase. If we could bisect it to a
precise shader, that would probably help!

Alternatively, there's a good hope that some ANGLE devs are reading
this list anyway ... ;-)

Concerning NeedFakeVertexAttrib0() I'd assume that this check is
simply
too vague,
Well the check that results in the error message is in the part of
the code that I elided. There are still questions that could not be
easily explained by a bug there:
   - why does the VBO have no data when drawElements is called? Is it
   a race condition in loading resources, as was suspected last week
   on this list about the Caves demo?
   - why does drawElements get called to draw only 1 vertex?

Cheers,
Benoit

since there hasn't been any problem with rendering before,
at
least on those machines with a fairly good GPU -- even in the case
attribute 0 wasn't set (what I admittedly don't know because I
haven't
checked it as we dynamically assign the locations based on the name
of
the respective attribute in the shader via getAttribLocation).

Cheers
Yvonne


-- Yvonne Jung Fraunhofer-IGD | Tel.: ++49-6151-155-290 Fraunhoferstr. 5 | Fax.: ++49-6151-155-196 64283 Darmstadt | email: yvonne.jung@igd.fraunhofer.de Germany | http://www.igd.fhg.de/igd-a4
-----------------------------------------------------------
You are currently subscribed to public_webgl@khronos.org.
To unsubscribe, send an email to majordomo@khronos.org with
the following command in the body of your email:



--
Yvonne Jung
Fraunhofer-IGD    | Tel.: ++49-6151-155-290
Fraunhoferstr. 5  | Fax.: ++49-6151-155-196
64283 Darmstadt   | email: yvonne.jung@igd.fraunhofer.de
Germany           | http://www.igd.fhg.de/igd-a4
-----------------------------------------------------------
You are currently subscribed to public_webgl@khronos.org.
To unsubscribe, send an email to majordomo@khronos.org with
the following command in the body of your email: