https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88846
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Priority|P3 |P1
CC| |jakub at gcc dot gnu.org,
| |vmakarov at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Seems entirely like RA bug to me. The MEM_ALIAS_SET on the REG_EQUIV is
correct, it is the mode of the memory the SET_DEST of the insn is going to be
stored into. What is weird is that it is added to an instruction before it is
stored there, and if that is reasonable, it should actually only use it if it
moves the store from where it has actually been done to where it wants with no
other memory accesses that would prevent that (in this case there are ones that
conflict with it, so it must not do that). Probably latent before though.
