cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

bananafish
Adept I

OpenGL driver change log available?

I am Chuck Ritola, and 7 out of 10 people would agree that I am a human being.

My main interest here is to improve Radeon HDxxxx support for the GitHub project Terminal Recall (TRCL). TRCL is sensitive to changes in driver behavior: With the release of the Omega driver series Radeon support broke with driver segfaults on startup. A later revision of the driver results in a blank game screen and segfault on exit. As a result, AMD support has been down for 6 months and with the resources of a single-developer project there hasn't been much luck isolating the problem. Automated tests will not run on their own because the fglrx driver is crashing on exit of each test case. Because these tests are run from a JVM, the segfault causes the VM to abort and the test never completes.

The next strategy was to use CodeXL to help isolate the rendering problem, however it is not working properly out-of-the-box, throwing SIGPIPE on Teapot. Originally there was an issue finding the proper 64-bit GL library but with some link hacking I was able to get rid of this error but it still throws SIGPIPE with trouble finding various functions. This is on Ubuntu 64-bit 14.04 w/ Redwood card and fglrx-updates 15.20.1013 using the 64-bit 1.7 .deb. I'm a little confused why it had ELF issues in the first place as other GL apps do not have problems finding the proper GL.so.

Some questions to start are:

- Is there a detailed publicly-available changelog outlining specific changes in the OpenGL driver behavior?

- What might I be doing wrong in CodeXL to cause it to fail?

1 Solution
bananafish
Adept I

Update:

I've isolated the change which was causing the problems. gl_Position is initialized as vec4(0,0,0,0) while it used to be initialized as vec4(0,0,0,1). As a result, unset w coords caused all vertices to be divided by zero. The nVIDIA driver still inits to vec4(0,0,0,1) but the Intel and AMD Omega drivers so far init to vec4(0,0,0,0).

A changelog would still be great in the future.

View solution in original post

0 Likes
3 Replies
jtrudeau
Staff

Hi bananafish​!. Who am I to disagree with majority opinion

You will be whitelisted momentarily, and added to the roster of questionable denizens of this scurrilous place. This could go in a couple of places, but I'm going to drop it in the OpenGL forum, and point the CodeXL people at it there.

Good luck, and thanks for the sense of humor about our little walled garden.

0 Likes
nou
Exemplar

I got that SIGPIPE error too. I think it have something that it pick up libGL.so from mesa. IIRC removing link to mesa version of libGL.so solved this for me.

bananafish
Adept I

Update:

I've isolated the change which was causing the problems. gl_Position is initialized as vec4(0,0,0,0) while it used to be initialized as vec4(0,0,0,1). As a result, unset w coords caused all vertices to be divided by zero. The nVIDIA driver still inits to vec4(0,0,0,1) but the Intel and AMD Omega drivers so far init to vec4(0,0,0,0).

A changelog would still be great in the future.

0 Likes