Re: [WIP 0/3] Memory model and atomic API in Rust

[Paul E. McKenney]

On Mon, Apr 08, 2024 at 06:03:11PM +0100, Matthew Wilcox wrote:
> On Mon, Apr 08, 2024 at 09:55:23AM -0700, Paul E. McKenney wrote:
> > On Mon, Apr 08, 2024 at 05:02:37PM +0100, Matthew Wilcox wrote:
> > > In my ideal world, the compiler would turn this into:
> > > 
> > > 	newfolio->flags |= folio->flags & MIGRATE_MASK;
> > 
> > Why not accumulate the changes in a mask, and then apply the mask the
> > one time?  (In situations where __folio_set_foo() need not apply.)
> 
> Yes, absolutely, we can, should and probably eventually will do this
> when it gets to the top of somebody's todo list.  But it irks me that
> we can't tell the compiler this is a safe transformation for it to make.
> There are a number of places where similar things happen.
> 
> $ git grep folio_test.*folio_test
> 
> will find you 82 of them (where they happen to be on the same line)
> 
>                 if (folio_test_dirty(folio) || folio_test_locked(folio) ||
>                                 folio_test_writeback(folio))
>                         break;
> 
> turns into:
> 
>     1f41:       48 8b 29                mov    (%rcx),%rbp
>     1f44:       48 c1 ed 04             shr    $0x4,%rbp
>     1f48:       83 e5 01                and    $0x1,%ebp
>     1f4b:       0f 85 d5 00 00 00       jne    2026 <filemap_range_has_writeback+0x1a6>
>     1f51:       48 8b 29                mov    (%rcx),%rbp
>     1f54:       83 e5 01                and    $0x1,%ebp
>     1f57:       0f 85 c9 00 00 00       jne    2026 <filemap_range_has_writeback+0x1a6>
>     1f5d:       48 8b 29                mov    (%rcx),%rbp
>     1f60:       48 d1 ed                shr    $1,%rbp
>     1f63:       83 e5 01                and    $0x1,%ebp
>     1f66:       0f 85 ba 00 00 00       jne    2026 <filemap_range_has_writeback+0x1a6>
> 
> rather than _one_ load from rcx and a test against a mask.

Agreed, it would be nice if we could convince the compiler to do this
for us, preferably without breaking anything.

> > If it turns out that we really do need a not-quite-volatile, what exactly
> > does it do?  You clearly want it to be able to be optimized so as to merge
> > similar accesses.  Is there a limit to the number of accesses that can
> > be merged or to the region of code over which such merging is permitted?
> > Either way, how is the compiler informed of these limits?
> 
> Right, like I said, it's not going to be easy to define exactly what we
> want.

Or to convince the usual suspects that any definition we might come up
with is useful/implementable/teacheable/...  :-/

							Thanx, Paul