RADV+ACO Look To Your Help For Improving The Vulkan Driver & Linux Gaming Performance

[darkbasic]

Q: Will you actually look at the bug reports, for realz?

A: Yes! We look at all RADV and ACO bug reports as they come in, even if we can't fix them immediately.

This is the most important thing, it wasn't always that way a few years ago.

[syrjala]

IMO filing a new bug is better than piling on to an existing one, unless you're quite sure it's the exact same issue on similar hardware. Marking bugs as duplicates is much easier than trying to detangle a coherent story from a bug report that has turned into a dumping ground for comments along the lines of "/metoo has this sort of similar bug, but on this totally different hardware".

[BNieuwenhuizen]

My biggest hesitation in reporting bugs sometimes is that I dont want to report a duplicate bug and I assume someone smarter with better logs already reported it. Path of Exile for instance. It ran smooth for a while, then they changed something and it's a stuttery mess again (tbh, its pretty awful in windows right now too). I feel like theres no way the Mesa team doesn't know about this, it's a pretty big game. But maybe I'm wrong. I should be better about these things.

In general posting is better than not posting. Would be nice if you do a basic search on game title and check driver of the report (i.e. intel vs amd) but I'd rather have a duplicate report than not knowing about something. (In particular we typically are closer to the issue of people deduping non-equal issues too much than too little)

[AdelKS]

Thank you very much for this post RADV devs, you rock! I must admit that I love how close we can be to you guys, as linux users. I believe a windows user can never interact with the AMD GPU devs about issues he may encounter in his games. Yet here we can just open an issue and we can talk with a RADV dev right away. They read and handle opened issues swiftly as a cherry on top (I have opened a few ones). I gotta love also the ability for vertical integration when trying to win performance, because of the open source nature of things: take for example valve that is getting involved in the kernel and wine (with fsync) and also in the AMD GPU driver with ACO.

Something to note: there's a "fossilize" feature that is installed with steam play for Linux. That is easy to use (as simple as using a env variable before lunching wine), at least for overwatch, to record some precious shader compilation logs for ACO. I don't think it's explained anywhere in the docs but I got pointed to it when I opened my first bug report.

[f0rmat]

My biggest hesitation in reporting bugs sometimes is that I dont want to report a duplicate bug and I assume someone smarter with better logs already reported it. Path of Exile for instance. It ran smooth for a while, then they changed something and it's a stuttery mess again (tbh, its pretty awful in windows right now too). I feel like theres no way the Mesa team doesn't know about this, it's a pretty big game. But maybe I'm wrong. I should be better about these things.

I understand what you are saying there, but I have been told by people that I have reported bugs to that I should report it anyway. That can help determine if the bug is hardware specific or software specific or a combination of both. As an oversimplified example, it I report the same bug on two different types of hardware, then there is a greater probability that it might be software specific. If I keep reporting it for the same hardware, than it has a higher probability of hardware specific. If I manage a project, many people reporting the same bug - particularly across multiple hardware types (and not just GPUs in this case), then it will probably rise in the queue as it appears to affect more people.

Now in all honesty, I do not know how they manage their bug reports, but this seems like a logical conclusion. If only one person out of 100 report a bug, that is 1% of the users. If 30 out of 100 report the same bug - especially if it is under different hardware, that is 30% of the users. If those 30 are also using the same GPU (or GPU family) that provides focus as to what to fix.

That is my 2 cents anyway - and if I am wrong, someone please let me know. Bug fixing seems to be more effective if a "pattern of misconduct" can be determined and/or traced.

[Brisse]

Will do! With that said, RADV+ACO has been super solid for me nowadays. I guess part of that comes down to using an older GPU.

[ms178]

Any news on the Company of Heroes 2 issue (reported here)? As the 32bit version works and is still around, it is not that urgent. But I am just curious why it doesn't work with the 64bit version of that game yet.

[aufkrawall]

Thanks for your work! I'm looking forward to report more bugs once I get RDNA 2 card.

Don't get your hopes up, it seems it works mostly (entirely?) as well as RDNA1 already.

[Venemo]

My biggest hesitation in reporting bugs sometimes is that I dont want to report a duplicate bug and I assume someone smarter with better logs already reported it. Path of Exile for instance. It ran smooth for a while, then they changed something and it's a stuttery mess again (tbh, its pretty awful in windows right now too). I feel like theres no way the Mesa team doesn't know about this, it's a pretty big game. But maybe I'm wrong. I should be better about these things.

I personally haven't player Path of Exile yet, but you can use the search feature in the Mesa bug tracker to find out if someone reported it.

[shmerl]

Thanks for your work! I'm looking forward to report more bugs once I get RDNA 2 card.