Mesa R500 Texture Semaphore Improvements
For the final test with an X1800XL graphics card on the Sandy Bridge system, to make things more interesting, Lightsmark was tested at 800 x 600 and 2560 x 1600, compared to 1920 x 1080 as used with the previous two graphics cards.
When running Lightsmark at 800 x 600, the Tom's tex-sem Mesa work was not of any benefit over mainline until setting the RADEON_TEX_GROUP value above six. With the default value of eight, the Radeon X1800XL was 23% faster in Lightsmark at 800 x 600. Any value above eight continued to not be of any advantage.
Pushing the X1800XL graphics card hard by running Lightsmark at 2560 x 1600, setting RADEON_TEX_GROUP to even a value of two was useful. As the RADEON_TEX_GROUP value went higher, so did the frame-rate, but again it plateaued with a maximum texture look-up count of eight. The X1800XL was around 45% faster with this instruction scheduler / texture semaphore work for R500 hardware work.
The performance improvements thanks to Tom Stellard's instruction scheduler for texture semaphore changes are really great. It's wonderful to see the Lightsmark numbers go up by so much for this older hardware, but non-shader using OpenGL workloads the tex-sem work will likely yield no changes. The default RADEON_TEX_GROUP value of eight also appears to be sane. I hope that the tex-sem branch will be merged to master in the near future. Again, this work is only targeted for Radeon X1000 (R500) series hardware.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.