New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
PPU LLVM: Add FMA accuracy setting #7874
Conversation
6cce131
to
0d4627d
Compare
4f4abb8
to
fb50f45
Compare
fb50f45
to
f7bf99c
Compare
f7bf99c
to
657f727
Compare
Is the performance penalty that big to warrant an accuracy setting? |
It depends on the cpu really, on cpus from the last ~5 years it may result in a slight performance increase actually. |
I think we should avoid the extra setting then. All CPUs from Recommended Specs have FMA as far as I'm aware |
…s, fnmadd, fnmadds, fnmsub, fnmsubs Co-Authored-By: Eladash <elad3356p@gmail.com>
At the very least, it should default to on. But I also agreed with Ani, I don't think a setting is warranted. Maybe make it an option in the other direction? That is, make it "PPU fast fma" and false by default. If someone really needs to they can enable inaccuracy. |
657f727
to
5ddd716
Compare
I think it shouldn't check for m_use_fma. This would make setting work consistently. |
The point is for the setting to be forced on fma enabled CPUs |
|
Followup for #7135, implements as "PPU Accurate FMA" setting in config file.
When this option enabled it changes the PPU cache version changes and if you do not have cache for x option state it will generate new PPU LLVM cache.