You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I think SPIRV tools can get performance improvement by using Profile-Guided Optimization (PGO). I think so because, from my experience, compiler-like things get up to a 20% boost with it (according to Clang and Rustc tests).
Did anyone try to do such optimization on the project? Not sure, if is it possible or not to use tests as a profile since a good profile should work on a "hot" path in general cases.
The text was updated successfully, but these errors were encountered:
We have not tried it. I agree that it would probably give a good improvement, but the tests would not be good training data. To build usefully with PGO we would need a good set of shaders from a variety of source that have the correct licence.
This is not something that the maintainers are interested in perusing at this time. We could accept a PR from someone else, but we would have to look it over carefully.
Hi!
I think SPIRV tools can get performance improvement by using Profile-Guided Optimization (PGO). I think so because, from my experience, compiler-like things get up to a 20% boost with it (according to Clang and Rustc tests).
Did anyone try to do such optimization on the project? Not sure, if is it possible or not to use tests as a profile since a good profile should work on a "hot" path in general cases.
The text was updated successfully, but these errors were encountered: