WebGL texture-mips conformance test fails with ANGLE across different GPUs

[GoogleCodeExporter]

One of the WebGL texture-mips conformance tests fails on both an NVIDIA and ATI 
GPUs and in both Chrome and Firefox on Windows XP 32. The test passes with 
Chrome "--use-gl=desktop" on the NVIDIA GPU but 3 of the tests fail on the ATI 
GPU. It also passes under Windows 7 on a different system with a different ATI 
GPU. The failure looks like an ANGLE bug on Windows XP.

What steps will reproduce the problem?
1. Visit 
https://cvs.khronos.org/svn/repos/registry/trunk/public/webgl/conformance-suites
/1.0.1/webgl-conformance-tests.html
2. Scroll down to conformance/textures/texture-mips.html
3. Click the run button

What is the expected output? What do you see instead?
The test should say PASS on a green background. Instead it says "failed: 
texture that is only using the smallest 2 mips should draw with green" on a red 
background.

What version of the product are you using? On what operating system?
ANGLE revision 1243 in Chrome 22.0.1217.0 (Official Build 148295) canary
ANGLE revision 1041 in Firefox 15.0

Fails on both of these GPUs
NVIDIA 9800GT with 301 series driver
ATI FirePro V7800 with driver version 8.911.3.4

This is on Windows XP. 

Original issue reported on code.google.com by goo...@erewhon0.net on 26 Jul 2012 at 4:13

[GoogleCodeExporter]

Looks like a GPU specific issue. The test passes with 22.0.1128.0 (Official 
Build 148524) Canary with NVIDIA 301.42 drivers and a Quadro FX 380 on Windows 
7 64-bit.

Original comment by kbr@chromium.org on 26 Jul 2012 at 6:00

[GoogleCodeExporter]

I know it passes on both AMD and NV on Win7.  It's possible there is an ANGLE 
bug in the non-9ex path which would only really be exercised on WinXP.  Either 
that, or it's a driver issue.  One way to test would be to run the tests with 
d3d9ex disabled on Windows 7.

Original comment by dan...@transgaming.com on 26 Jul 2012 at 7:54

[GoogleCodeExporter]

> Looks like a GPU specific issue
Specific to 2 GPUs from different makers? I don't think so. Furthermore one of 
those on which it fails on Windows XP is a GeForce 9800GT with the NVIDIA 
301.42 driver. I think, as I wrote in initial report, that this is an ANGLE bug 
on Windows XP. 

> Either that, or it's a driver issue.
What driver is common to both GPUs?

> I know it passes on ... AMD
You have to be careful making statements like that because AMD has 2 different 
drivers with different version numbering schemes for the pro (workstation) 
cards and the consumer (desktop) cards.

> One way to test would be to run the tests with d3d9ex disabled on Windows 7
How do I do that?

Original comment by goo...@erewhon0.net on 27 Jul 2012 at 1:56

[GoogleCodeExporter]

Yeah, Mark is correct.

I just build chromium with D3D9EX disabled and on my Win7 with AMD GPU, 
texture-mips.html began to fail, whereas the test is passing with D3D9EX 
enabled.

Original comment by z...@google.com on 28 Jul 2012 at 12:41

[GoogleCodeExporter]

I was able to reproduce this as well on Windows 7 with D3D9EX disabled. I'll 
have a closer look at it...

Original comment by nicolas....@gmail.com on 30 Jul 2012 at 8:11

• Changed state: Accepted

[GoogleCodeExporter]

Original comment by nicolas....@gmail.com on 24 Sep 2012 at 8:29

• Changed state: Started

[GoogleCodeExporter]

This is fixed in r1288.  This issue was with managed textures which need to be 
explicitly marked dirty when upper levels are modified.

Original comment by dan...@transgaming.com on 27 Sep 2012 at 5:56

• Changed state: Fixed