https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97759
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Looking at the referenced benchmark, options that would result in popcnt*
instruction aren't there though and in the assembly it calls __popcountdi2.
So, I guess we want at least:
1) for POPCOUNT (x) == 1 use x && (x & (x - 1)) == 0 if optab_handler
(popcount_optab, mode) == CODE_FOR_nothing (a fuzzy case are the double-word
cases where we'd end up with popcount (x >> wordbits) + popcount ((word) x) ==
1
vs. double-word x && (x & (x - 1)) = 0)
2) the case Richi wrote about, optimize POPCOUNT (x) <= 1 or POPCOUNT (x) == 1
|| x == 0 to (x & (x - 1)) == 0 always
