[Bug tree-optimization/95836] New: Failure to remove strchr converted to bool with string literal when bit magic would be much more efficient (and sometimes smaller)

[Bug tree-optimization/95836] New: Failure to remove strchr converted to bool with string literal when bit magic would be much more efficient (and sometimes smaller)

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95836

            Bug ID: 95836
           Summary: Failure to remove strchr converted to bool with string
                    literal when bit magic would be much more efficient
                    (and sometimes smaller)
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gabravier at gmail dot com
  Target Milestone: ---

bool f(char c)
{
    return strchr("1234567890", c);
}

This (and a lot of other similar strchr calls with string literals) can be
optimized to using a few bitwise operations (here, to something like `return
((unsigned)c < 64) & ((((uint64_t)1 << c) & 287948901175001089) != 0);`). This
transformation is done by LLVM, but not by GCC.

[Bug tree-optimization/95836] Failure to remove strchr converted to bool with string literal when bit magic would be much more efficient (and sometimes smaller)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95836

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
With & it would actually invoke UB for c >= 64U.
So it has to be either &&, or << (c & 63), at least until we know exactly what
the behavior on out of bounds shift will be.

[Bug tree-optimization/95836] Failure to remove strchr converted to bool with string literal when bit magic would be much more efficient (and sometimes smaller)

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95836

--- Comment #2 from Gabriel Ravier <gabravier at gmail dot com> ---
Yeah I wasn't sure whether this was legitimate, I was just trying to give a
small example here, I dont think doesn't changes anything to how the
optimization is able to work unless GCC has targets where `a << b` with `b >
BITS_IN_A` traps.

[Bug tree-optimization/95836] Failure to remove strchr converted to bool with string literal when bit magic would be much more efficient (and sometimes smaller)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95836

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Well, it does, we can't keep it in the IL in that form because the compiler
itself might optimize it away on the assumption that UB doesn't happen.