https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96245
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org,
| |vries at gcc dot gnu.org
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I think it isn't really a too special case, of course one shouldn't take it
literally and look for a particular call etc.
But, generally try to see if all the switch cases are the same except some
constant or two (same e.g. in the ICF sense) and if that constant or two is
linear vs. the corresponding case.
It can be done either in switchconv, or could be perhaps some kind of
GIMPLE_SWITCH aware parametrized tail merging/cross jumping.
As it would (or could) turn constant expressions into non-constant, I guess
we'd need to take some conservative approach with __builtin_constant_p, punt if
there is a __builtin_constant_p in the block, either altogether, or if it is
among (recursively) the immediate uses of the stmt where the constant(s)
appear.
