Re: [PATCH 00/22] add support for Clang LTO

From: Paul E. McKenney
Date: Wed Jul 01 2020 - 10:06:57 EST


On Wed, Jul 01, 2020 at 01:40:27PM +0200, Peter Zijlstra wrote:
> On Wed, Jul 01, 2020 at 11:41:17AM +0200, Marco Elver wrote:
> > On Tue, 30 Jun 2020 at 22:30, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > On Tue, Jun 30, 2020 at 10:12:43PM +0200, Peter Zijlstra wrote:
> > > > On Tue, Jun 30, 2020 at 09:19:31PM +0200, Marco Elver wrote:
>
> > > > > Thoughts?
> > > >
> > > > How hard would it be to creates something that analyzes a build and
> > > > looks for all 'dependent load -> control dependency' transformations
> > > > headed by a volatile (and/or from asm) load and issues a warning for
> > > > them?
> >
> > I was thinking about this, but in the context of the "auto-promote to
> > acquire" which you didn't like. Issuing a warning should certainly be
> > simpler.
> >
> > I think there is no one place where we know these transformations
> > happen, but rather, need to analyze the IR before transformations,
> > take note of all the dependent loads headed by volatile+asm, and then
> > run an analysis after optimizations checking the dependencies are
> > still there.
>
> Urgh, that sounds nasty. The thing is, as I've hinted at in my other
> reply, I would really like a compiler switch to disable this
> optimization entirely -- knowing how relevant the trnaformation is, is
> simply a first step towards that.
>
> In order to control the tranformation, you have to actually know where
> in the optimization passes it happens.
>
> Also, if (big if in my book) we find the optimization is actually
> beneficial, we can invert the warning when using the switch and warn
> about lost optimization possibilities and manually re-write the code to
> use control deps.

There are lots of optimization passes and any of them might decide to
destroy dependencies. :-(

> > > > This would give us an indication of how valuable this transformation is
> > > > for the kernel. I'm hoping/expecting it's vanishingly rare, but what do
> > > > I know.
> > >
> > > This could be quite useful!
> >
> > We might then even be able to say, "if you get this warning, turn on
> > CONFIG_ACQUIRE_READ_DEPENDENCIES" (or however the option will be
> > named).
>
> I was going to suggest: if this happens, employ -fno-wreck-dependencies
> :-)

The current state in the C++ committee is that marking variables
carrying dependencies is the way forward. This is of course not what
the Linux kernel community does, but it should not be hard to have a
-fall-variables-dependent or some such that causes all variables to be
treated as if they were marked. Though I was hoping for only pointers.
Are they -sure- that they -absolutely- need to carry dependencies
through integers???

Anyway, the next step is to provide this functionality in one of the
major compilers. Akshat Garg started this in GCC as a GSoC project
by duplicating "volatile" functionality with a _Dependent_ptr keyword.
Next steps would include removing "volatile" functionality not required
for dependencies. Here is a random posting, which if I remember correctly
raised some doubts as to whether "volatile" was really carried through
everywhere that it needs to for things like LTO:

https://gcc.gnu.org/legacy-ml/gcc/2019-07/msg00139.html

What happened to this effort? Akshat graduated and got an unrelated
job, you know, the usual. ;-)

Thanx, Paul