------- Comment #10 from danglin at gcc dot gnu dot org 2008-01-21 17:47
-------
The patch appears to fix the problem, but I think it isn't a good fix.
The problem is the backend doesn't know when the LABEL_NUSES counter
needs to be incorrected and when it doesn't.
I think this is reload bug. We are losing LABEL_NUSES counts when a
reload occurs for insns like the following:
(insn 1838 1854 1615 282 (set (reg/f:SI 1154)
(label_ref:SI 1897)) 46 {*pa.md:2898} (nil)
(expr_list:REG_EQUIV (label_ref:SI 1897)
(insn_list:REG_LABEL 1897 (nil))))
Reloads for insn # 2938
Reload 0: R1_REGS, RELOAD_FOR_INPUT_ADDRESS (opnum = 1), can't combine,
secondary_reload_p
reload_reg_rtx: (reg/f:SI 1 %r1)
Reload 1: reload_in (SI) = (label_ref:SI 1897)
GENERAL_REGS, RELOAD_FOR_INPUT (opnum = 1), can't combine
reload_in_reg: (reg/f:SI 1154)
reload_reg_rtx: (reg/f:SI 3 %r3 [orig:313 #ref#21#111 ] [313])
secondary_in_reload = 0
secondary_in_icode = reload_insi_r
Shouldn't reload update the label uses count when it completes a reload
sequence?
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34150
