On Fri, 2016-02-26 at 20:10 +0100, Torvald Riegel wrote:
> On Fri, 2016-02-26 at 11:49 +0100, Richard Biener wrote:
> > On Thu, Feb 25, 2016 at 6:33 PM, Torvald Riegel <trie...@redhat.com> wrote:
> > > On Wed, 2016-02-24 at 13:14 +0100, Richard Biener wrote:
> > >> On Tue, Feb 23, 2016 at 8:38 PM, Torvald Riegel <trie...@redhat.com>
> > >> wrote:
> > >> > I'd like to know, based on the GCC experience, how important we
> > >> > consider
> > >> > optimizations that may turn data dependencies of pointers into control
> > >> > dependencies. I'm thinking about all optimizations or transformations
> > >> > that guess that a pointer might have a specific value, and then create
> > >> > (specialized) code that assumes this value that is only executed if the
> > >> > pointer actually has this value. For example:
> > >> >
> > >> > int d[2] = {23, compute_something()};
> > >> >
> > >> > int compute(int v) {
> > >> > if (likely(v == 23)) return 23;
> > >> > else <lots of stuff>;
> > >> > }
> > >> >
> > >> > int bar() {
> > >> > int *p = ptr.load(memory_order_consume);
> > >> > size_t reveal_that_p_is_in_d = p - d[0];
> > >> > return compute(*p);
> > >> > }
> > >> >
> > >> > Could be transformed to (after inlining compute(), and specializing for
> > >> > the likely path):
> > >> >
> > >> > int bar() {
> > >> > int *p = ptr.load(memory_order_consume);
> > >> > if (p == d) return 23;
> > >> > else <lots of stuff(*p)>;
> > >> > }
> > >>
[snip]
> What about the example I gave above? Is it unrealistic for compilers do
> ever do something like this, or is it just unlikely to gain much
> performance, or is it just that GCC does not do this today?
Modification of d happens-before the mo_consume load in this case, so
it's "immutable" data and the this specific example would be okay.
Sorry for the noise.
However, I find it nontrivial to figure out whether a compiler would
_not_ be able to replace a data dependency with a control dependency.
Such impossibility results are hard. Yet this is something that users
of the new mo_consume proposal would have to do, and I'm not sure we'll
end up at a state were this is relatively simple to figure out. If we
don't know when a programmer intends to use a dependency, we also cannot
warn if there's none.
