https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68910
Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|target |rtl-optimization
--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
This regression was introduced by:
2014-05-28 Richard Sandiford <rdsandif...@googlemail.com>
Olivier Hainque <hain...@adacore.com>
* rtl.h (set_for_reg_notes): Declare.
* emit-rtl.c (set_for_reg_notes): New function.
(set_unique_reg_note): Use it.
* optabs.c (add_equal_note): Likewise
This patch breaks the reload inheritance trick in find_reloads:
/* If we made a MEM to load (a part of) the stackslot of a pseudo
that didn't get a hard register, emit a USE with a REG_EQUAL
note in front so that we might inherit a previous, possibly
wider reload. */
if (replace
&& MEM_P (op)
&& REG_P (reg)
&& (GET_MODE_SIZE (GET_MODE (reg))
>= GET_MODE_SIZE (GET_MODE (op)))
&& reg_equiv_constant (REGNO (reg)) == 0)
set_unique_reg_note (emit_insn_before (gen_rtx_USE (VOIDmode, reg),
insn),
REG_EQUAL, reg_equiv_memory_loc (REGNO
(reg)));
by emitting a naked USE instead of a USE with a REG_EQUAL note. Then:
/* Make a pass over all the insns and delete all USEs which we inserted
only to tag a REG_EQUAL note on them. Remove all REG_DEAD and REG_UNUSED
notes. Delete all CLOBBER insns, except those that refer to the return
value and the special mem:BLK CLOBBERs added to prevent the scheduler
from misarranging variable-array code, and simplify (subreg (reg))
operands. Strip and regenerate REG_INC notes that may have been moved
around. */
doesn't delete the artificial USEs, thus seriously hindering DSE2.
