OpenGL Shader Cache Support For RadeonSI Is Making Progress

[phoronix]

Phoronix: OpenGL Shader Cache Support For RadeonSI Is Making Progress

Last week we reported that Collabora developer Timothy Arceri would be trying to support Mesa's OpenGL shader cache with RadeonSI while since then work has begun to materialize...

http://www.phoronix.com/scan.php?page=news_item&px=RadeonSI-Shader-Cache-Progress

[boxie]

I am looking forward to seeing how much of an impact this actually has!

[Pontostroy]

I am looking forward to seeing how much of an impact this actually has!

not so much, dota benchmark +3-4 fps, loading time for Deus Ex, the biggest effect is that with shader cache min fps is much higher and no big fps drops when hundreds of shaders are created at the same time.

[computerquip]

I was going to make a comment on how this shouldn't have much performance gain, if any at all. Then I came to learn that some games compile shaders in the middle of their game, which is a major problem in OpenGL. Why do people do this?

[kparal]

I am looking forward to seeing how much of an impact this actually has!

http://www.gearsongallium.com/?p=3718

[tarceri]

not so much, dota benchmark +3-4 fps, loading time for Deus Ex, the biggest effect is that with shader cache min fps is much higher and no big fps drops when hundreds of shaders are created at the same time.

Well that and load times is the only reason for an on-disk shader cache to exist. It's not intended to bring greater fps, if you are seeing a jump in fps it just means the benchmark is not long enough for the drops to become averaged out, the shaders are really nasty to compile or there is just a bunch of them being compiled throughout the benchmark, none of these should really cause better fps in normal gaming. The cache is just to smooth out those drops and speed up load times.

[dungeon]

I am looking forward to seeing how much of an impact this actually has!

Impact depends how much fast is your CPU, on fastest CPU impact is smaller on slowest CPU it should be bigger... Basically this just avoid CPU to compile shaders non stop, instead compile it once, other times just read from cache.

[Pontostroy]

Dota2 and dirt benchmarks

http://openbenchmarking.org/result/1702043-RI-1702041RI22

[dungeon]

No benchmarks needed, just imagine comfort of driving tank and formula at the same speed

That should be experience difference if some game compile shaders during gameplay.

[ResponseWriter]

I was going to make a comment on how this shouldn't have much performance gain, if any at all. Then I came to learn that some games compile shaders in the middle of their game, which is a major problem in OpenGL. Why do people do this?

Ironically, I believe it's to reduce loading times by loading assets on demand as required rather than pre-loading everything, required or not. Apart from loading times, hopefully it makes games that do this a little smoother, although it should be noted that this probably won't help the pauses in Rocket League and possibly some others.

I think Marek's preferred approached as been to use an in-memory cache to try to help reduce these stalls but that only applies on subsequent compiles so doesn't reduce initial loading times or the stalls when the assets are first compiled during play.

Edit: Forgot to mention Marek was also working on other features that would help speed up compilation and reduce stalls.