Mesa Drops Support For AMD Zen L3 Thread Pinning, Will Develop New Approach
It was just a few months back that the Mesa/RadeonSI open-source AMD Linux driver stack received Zen tuning for that CPU microarchitecture's characteristics. But now AMD's Marek Olšák is going back to the drawing board to work on a new approach for Zen tuning.
Just a few days ago I wrote about another developer wanting to toggle the support around L3 thread pinning as it was found to hurt the RadeonSI Gallium3D performance in at least some Linux games. At that point the goal was to allow making it a DriConf tunable that could then be adjusted a per-game/app basis, but it turns out the gains aren't there to keep it around.
Marek did further testing and today dropped it from Mesa 19.0 Git and also copied the patch for back-porting to Mesa 18.3 for disabling the pinning of driver threads to a specific L3 cache.
It was hoped to deliver optimal performance for Zen CPUs but in the end, Marek wrote on the revert, "This implementation can have massive drawbacks."
He did comment on the mailing list that he has a "better plan, but I don't know if it will work" without elaborating further. We'll be waiting to hear what comes about of his next step for open-source AMD Zen tuning for the Radeon driver stack.
Considering how Marek has dominated Radeon driver Mesa development over the past decade, I have no doubt about his ability to come up with a new and superior solution.
Just a few days ago I wrote about another developer wanting to toggle the support around L3 thread pinning as it was found to hurt the RadeonSI Gallium3D performance in at least some Linux games. At that point the goal was to allow making it a DriConf tunable that could then be adjusted a per-game/app basis, but it turns out the gains aren't there to keep it around.
Marek did further testing and today dropped it from Mesa 19.0 Git and also copied the patch for back-porting to Mesa 18.3 for disabling the pinning of driver threads to a specific L3 cache.
It was hoped to deliver optimal performance for Zen CPUs but in the end, Marek wrote on the revert, "This implementation can have massive drawbacks."
He did comment on the mailing list that he has a "better plan, but I don't know if it will work" without elaborating further. We'll be waiting to hear what comes about of his next step for open-source AMD Zen tuning for the Radeon driver stack.
Considering how Marek has dominated Radeon driver Mesa development over the past decade, I have no doubt about his ability to come up with a new and superior solution.
15 Comments