Bug fixing the MESA 3D drivers

Most of our projects are around performance optimisation, but we’re cleaning up bugs too. This is because you can only speed up software when certain types of bugs are cleared out. A few months ago, we got a different type of request. If we could solve bugs in MESA 3D that appear in games.

Yes, we wanted to try that and got a list of bugs to solve. And as you can read, we were successful.

Below you found a detailed description of one of the 5 bugs we solved by digging deep into the different games and the MESA 3D drivers. At the end of the blog post you’ll find the full list with links to issues in MESA’s bugtracker.

Fixing the Render Error in Wasteland 2

Here’s a peek in our kitchen. The text below is taken from our issue-tracker, but does not include the logs of the tools like ApiTrace.

Problem Description

There are black rectangles instead of blood splashes. They are rendered in call 1570549, for example.

Fragments’ color values are NaNs. The fragment shader (1337, call 1080782) receives NaN values for two inputs: xlv_TEXCOORD2 and xlv_TEXCOORD2_1.

The vertex shader (1336, call 1080779) produces these values by calling normalize on TANGENT attribute.

TANGENT is mapped to 1 generic vertex attribute. There is 1570545 @1 glDisableVertexAttribArrayARB(index = 1) before 1570549 @1 glDrawElements. I.e. the attribute uses the current value, but this value is never set directly (using glVertexAttrib4f, for example), so its value is (0, 0, 0, 1) according to the OpenGL specification.

Problem Analysis

In the OpenGL specifications normalize is not defined for zero-vectors:

Returns a vector in the same direction as x but with a length of 1.

The behavior of all drivers is correct:

  • Mesa returns NaNs
  • NVIDIA proprietary driver returns NaNs
  • AMD proprietary driver returns zeros

Why it works on NVIDIA: although normalize returns NaNs for zero vectors, for some reason GL_CURRENT_VERTEX_ATTRIB is (1, 0, 0, 1), this behavior violates the specification!

Why it works on AMD: normalize returns zeros for zero vectors, so other values are not “damaged”.
Catalyst uses v_mul_legacy here which treats NaNs as zeros (Mesa uses v_mul).

Work-around

Here is a workaround that makes Mesa return zeros for zeros vectors (as AMD on Catalyst does) :


How the screen looks like with this workaround:

This workaround also fixes the bug in “Megamaggots’ Trails”.

In my opinion the bug is in the game engine, not Mesa. glVertexAttrib4(1, …) should be added after glDisableVertexAttribArray(1) to set correct values for TANGENT attribute. It’s a coincidence that it works well on proprietary NVIDIA (because of its incorrect behavior) and AMD (because of its normalize implementation).

MESA 3D bugs we fixed

We fixed the following bugs:

At your service! We like it that more of our work is really visible out there.

 

Related Posts

<!--:en-->nvidia_geforce_gtx_580_vs_amd_radeon_hd_6870<!--:-->

Installing both NVidia GTX and AMD Radeon on Linux for OpenCL

...  pretty easy (with some luck). You need to install both drivers and watch out that libglx.so isn't overwritten by NVidia's driver as ...

save

Mobile Processor OpenCL drivers (Q3 2013) + rating

...  does not reflect reality. ZiiLabs SDK-page@StreamHPCDrivers can be delivered by Creative, when you pledge to order ...

<!--:en-->chip-check<!--:-->

Do your (X86) CPU and GPU support OpenCL?

...  hardware, just read this paragraph. If you have recent drivers for your GPU, you can be sure OpenCL is already supported and you ...

<!--:en-->trends-cuda-vs-opencl-adjusted<!--:-->

OpenCL vs CUDA Misconceptions

...  copies. Comparing CUDA to OpenCL 1.0 (since NVIDIA's 1.1-drivers a a year old and not been updated since) is just not fair. What is ...