Hey,
On Tue, 31 Oct 2023, Peter Zijlstra wrote:
> > equivalent to that, then it can't be used in this situation. If you
> > _have_ to use a RmW for other reasons like interrupt safety, then a
> > volatile variable is not the way to force this, as C simply doesn't have
> > that concept and hence can't talk about it. (Of course it can't, as not
> > all architectures could implement such, if it were required).
>
> Yeah, RISC archs typically lack the RmW ops. I can understand C not
> mandating their use. However, on architectures that do have them, using
> them makes a ton of sense.
>
> For us living in the real world, this C abstract machine is mostly a
> pain in the arse :-)
Believe me, without it you would live in a world where only languages like
ECMA script or Rust existed, without any reliable spec at all ("it's
obvious, the language should behave like this-and-that compiler from 2010
implements it! Or was it 2012?"). Even if it sometimes would make life
easier without (also for compilers!), at least you _have_ an arse to feel
pain in! :-) Ahem.
> > So, hmm, I don't quite know what to say, you're between a rock and a hard
> > place, I guess. You have to use volatile for its effects but then are
> > unhappy about its effects :)
>
> Notably, Linux uses a *ton* of volatile and there has historically been
> a lot of grumbling about the GCC stance of 'stupid' codegen the moment
> it sees volatile.
>
> It really would help us (the Linux community) if GCC were to be less
> offended by the whole volatile thing and would try to generate better
> code.
>
> Paul has been on the C/C++ committee meetings and keeps telling me them
> folks hate volatile with a passion up to the point of proposing to
> remove it from the language or somesuch. But the reality is that Linux
> very heavily relies on it and _Atomic simply cannot replace it.
Oh yeah, I agree. People trying to get rid of volatile are misguided.
Of course one can try to capture all the individual aspects of it, and
make individual language constructs for them (_Atomic is one). It makes
arguing about and precisely specifying the aspects much easier. But it
also makes the feature-interoperability matrix explode and the language
more complicated in areas that were already arcane to start with.
But it's precisely _because_ of the large-scale feature set of volatile
and the compilers wish to cater for the real world, that it's mostly left
alone, as is, as written by the author. Sure, one can wish for better
codegen related to volatile. But it's a slippery slope: "here I have
volatile, because I don't want optimizations to happen." - "but here I
want a little optimization to happen" - "but not these" - ... It's ... not
so easy :)
In this specific case I think we have now sufficiently argued that
"load-modify-store --> rmw" on x86 even for volatile accesses is a correct
transformation (and one that has sufficiently local effects that our heads
don't explode while thinking about all consequences). You'd have to do
that for each and every transformation where volatile stuff is involved,
just so to not throw out the baby with the water.
> > If you can confirm the above about validity of the optimization, then at
> > least there'd by a point for adding a peephole in GCC for this, even if
> > current codegen isn't a bug, but I still wouldn't hold my breath.
> > volatile is so ... ewww, it's best left alone.
>
> Confirmed, and please, your SMP computer only works becuase of volatile,
> it *is* important.
Agreed.
Ciao,
Michael.
