https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67513
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |law at gcc dot gnu.org
--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Yury Gribov from comment #7)
> FYI I'd prefer to keep current BIT_IOR_EXPR approach in
> asan_expand_check_ifn as it allows for efficient implementation for ARM
> targets (as compared to two successive branches currently used in LLVM).
Ugh. Then perhaps we should decide based on some costs.
We have LOGICAL_OP_NON_SHORT_CIRCUIT and BRANCH_COST, unfortunately the former
is to decide on very cheap computation with roughly the same probabilities,
rather than this case where the cheaper condition is much more likely, and
BRANCH_COST can't be really compared to insn costs.
So perhaps base this say on targetm.have_conditional_execution () ?
