https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102586
qinzhao at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |qinzhao at gcc dot gnu.org
--- Comment #25 from qinzhao at gcc dot gnu.org ---
(In reply to Jakub Jelinek from comment #24)
> I bet with the default -flifetime-dse=2 we would then have a
> var = {};
> var ={v} {CLOBBER}; // Start of ctor clobber
> var.whatever = ...; // Ctor content
> and the zero initialization would be optimized away then.
> I guess ideal would be to emit such zero initialization in the complete
> object ctors right after the -flifetime-dse=2 beginning clobber and ideally
> one that clearly says to the gimplifier and rest of middle-end that it must
> initialize even padding bits.
> Because right now gimplification of var = {} will often happily lower that to
> var.a = 0; var.b = 0; var.c = 0;
> even when there are padding bits,
oh right, I recall that this was the reason why adding "var = {}" to initialize
the whole structure didn't work well for initialize all paddings. therefore we
have to explicitly call __builtin_clear_padding for clear all paddings.
