Re: [PATCH v9] pgo: add clang's Profile Guided Optimization infrastructure

[Peter Zijlstra]

On Sat, Jun 12, 2021 at 12:10:03PM -0700, Bill Wendling wrote:
> > You're modifying a lot of x86 files, you don't think it's good to let us
> > know?  Worse, afaict this -fprofile-generate changes code generation,
> > and we definitely want to know about that.
> >
> I got the list of people to add from the scripts/get_maintainer.pl.

$ ./scripts/get_maintainer.pl -f arch/x86/Makefile
Thomas Gleixner <tglx@xxxxxxxxxxxxx> (maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT))
Ingo Molnar <mingo@xxxxxxxxxx> (maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT))
Borislav Petkov <bp@xxxxxxxxx> (maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT))
x86@xxxxxxxxxx (maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT))

> there's one intel people CC'ed, but he didn't sign off on it.

Intel does not employ the main x86 maintainers, even it if did, mailing
a random Google person won't get the mail to you either, would it?

> These patches were available for review for months now,

Which doesn't help if you don't Cc the right people, does it. *nobody*
has time to read LKML.

> and posted to all of the lists and CC'ed to the people from
> scripts/get_maintainers.pl. Perhaps that program should be improved?

I suspect operator error, see above.

> > Supposedly -fprofile-generate adds instrumentation to the generated
> > code. noinstr *MUST* disable that. If not, this is a complete
> > non-starter for x86.
> 
> "noinstr" has "notrace", which is defined as
> "__attribute__((__no_instrument_function__))", which is honored by
> both gcc and clang.

Yes it is, but is that sufficient in this case? It very much isn't for
KASAN, UBSAN, and a whole host of other instrumentation crud. They all
needed their own 'bugger-off' attributes.

> > We've got KCOV and GCOV support already. Coverage is also not an
> > argument mentioned anywhere else. Coverage can go pound sand, we really
> > don't need a third means of getting that.
> >
> Those aren't useful for clang-based implementations. And I like to
> look forward to potential improvements.

I look forward to less things doing the same over and over. The obvious
solution if of course to make clang use what we have, not the other way
around.

> > Do you have actual numbers that back up the sampling vs instrumented
> > argument? Having the instrumentation will affect performance which can
> > scew the profile just the same.
> >
> Instrumentation counts the number of times a branch is taken. Sampling
> is at a gross level, where if the sampling time is fine enough, you
> can get an idea of where the hot spots are, but it won't give you the
> fine-grained information that clang finds useful. Essentially, while
> sampling can "capture the hot spots very well", relying solely on
> sampling is basically leaving optimization on the floor.
> 
> Our optimizations experts here have determined, through data of
> course, that instrumentation is the best option for PGO.

It would be very good to post some of that data and explicit examples.
Hear-say don't carry much weight.