https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71522
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |wrong-code
Component|c |tree-optimization
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
So it goes "wrong" when transforming
long double d;
__int128 unsigned _3;
MEM[(char * {ref-all})&d] = 0x414141414141414141414141414141;
_3 = MEM[(char * {ref-all})&d];
into SSA as
d_9 = 4.35573826932891467758901725805789285479666446831741854231e+96;
_3 = VIEW_CONVERT_EXPR<__int128 unsigned>(d_9);
similar as to how elsewhere we avoid using FP representation for what is
accessed as integers we should probably do so in update-address-taken
(otherwise FP normalization will apply via build_real and friends).
