[Bug tree-optimization/69908] recognizing idioms that check for a buffer of all-zeros could make *much* better code

[tetra2005 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69908

--- Comment #5 from Yuri Gribov <tetra2005 at gmail dot com> ---
(In reply to Marc Glisse from comment #4)
> (In reply to Yuri Gribov from comment #3)
> > As noted in comments, this is more about adding new API to Glibc than GCC
> > (they have corresponding
> > https://sourceware.org/bugzilla/show_bug.cgi?id=19920 about this). So can
> > this be closed?
> 
> I don't think so (at least not for that reason). Whatever glibc does, this
> is still a missed optimization in gcc.

Well, as we all know there are a lot of missing optimizations in GCC :) I think
the real question is whether it's ever going to be fixed if there's no standard
API for this code pattern which we can recognize as builtin.

I believe the answer is "No". ATM GCC does not vectorize even the simplest
memcpy equivalent code:
  // gcc tmp.c -O3 -mtune=native -ftree-vectorize -o- -S
  void memcpy_(char * __restrict a, char * __restrict b, unsigned n) {
    unsigned i;
    for (i = 0; i < n; ++i)
      a[i] = b[i];
  }
so I don't think it'll be able to recognize memcchr in reasonable timeframe.

Adding a std. function in Glibc and then treating it as std. builtin in GCC
would be straightforward and cost-effective solution but the ball is on Glibc
side and they didn't sound very positive.

Thoughts?