https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89826
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
That REG_UNUSED note on r1 looks weird.
In postreload_jump we have:
(insn 49 11 50 3 (set (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(const_int 0 [0])) "pr89826.c":9:14 181 {*arm_movsi_insn}
(expr_list:REG_EQUAL (const_int 0 [0])
(nil)))
(insn 50 49 12 3 (set (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(const_int 0 [0])) "pr89826.c":9:14 181 {*arm_movsi_insn}
(expr_list:REG_EQUAL (const_int 0 [0])
(nil)))
(insn 12 50 13 3 (set (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(not:SI (reg:SI 2 r2 [orig:124 a.0_1 ] [124]))) "pr89826.c":9:14 156
{*arm_one_cmplsi2}
(nil))
(insn 13 12 37 3 (set (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(not:SI (reg:SI 12 ip [orig:125 a.0_1+4 ] [125]))) "pr89826.c":9:14 156
{*arm_one_cmplsi2}
(nil))
which looks normal, but contains non-DCEd code (that was there since subreg
pass
that lowered:
(insn 35 11 12 3 (set (reg:DI 114 [ iftmp.1_5 ])
(const_int 0 [0])) "pr89826.c":9:14 -1
(nil))
(insn 12 35 13 3 (set (subreg:SI (reg:DI 114 [ iftmp.1_5 ]) 0)
(not:SI (subreg:SI (reg:DI 110 [ a.0_1 ]) 0))) "pr89826.c":9:14 156
{*arm_one_cmplsi2}
(nil))
(insn 13 12 37 3 (set (subreg:SI (reg:DI 114 [ iftmp.1_5 ]) 4)
(not:SI (subreg:SI (reg:DI 110 [ a.0_1 ]) 4))) "pr89826.c":9:14 156
{*arm_one_cmplsi2}
(nil))
).
In jump2 we have:
(insn 50 49 12 3 (set (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])) "pr89826.c":9:14 181
{*arm_movsi_insn}
(expr_list:REG_DEAD (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(expr_list:REG_UNUSED (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(expr_list:REG_EQUAL (const_int 0 [0])
(nil)))))
(insn 12 50 13 3 (set (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(not:SI (reg:SI 2 r2 [orig:124 a.0_1 ] [124]))) "pr89826.c":9:14 156
{*arm_one_cmplsi2}
(nil))
(insn 13 12 37 3 (set (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(not:SI (reg:SI 12 ip [orig:125 a.0_1+4 ] [125]))) "pr89826.c":9:14 156
{*arm_one_cmplsi2}
(nil))
that still has the dead code in there and makes through REG_UNUSED clear it is
unused.
But then ce3 turns it into:
(insn 49 9 50 2 (set (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(const_int 0 [0])) "pr89826.c":9:14 181 {*arm_movsi_insn}
(nil))
(insn 50 49 12 2 (set (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])) "pr89826.c":9:14 181
{*arm_movsi_insn}
(expr_list:REG_DEAD (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(expr_list:REG_UNUSED (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(nil))))
(insn 12 50 13 2 (cond_exec (eq (reg:CC 100 cc)
(const_int 0 [0]))
(set (reg:SI 3 r3 [orig:126 iftmp.1_5 ] [126])
(not:SI (reg:SI 2 r2 [orig:124 a.0_1 ] [124])))) "pr89826.c":9:14
3805 {*p *arm_one_cmplsi2}
(nil))
(insn 13 12 39 2 (cond_exec (eq (reg:CC 100 cc)
(const_int 0 [0]))
(set (reg:SI 1 r1 [orig:127 iftmp.1_5+4 ] [127])
(not:SI (reg:SI 12 ip [orig:125 a.0_1+4 ] [125]))))
"pr89826.c":9:14 3805 {*p *arm_one_cmplsi2}
(nil))
So, if the problem is using REG_UNUSED note, something forgot to recompute the
notes problem (regcprop?).
