This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: peephole2: dead regs not marked as dead
On 10/25/2010 11:35 AM, Georg Lay wrote:
(insn 22 8 23 2 peep2.c:5 (set (reg:SI 15 d15)
(and:SI (reg:SI 4 d4 [ x ])
(const_int -98305 [0xfffe7fff]))) 143
{*and3_zeroes.insert.{SI}.ic}
(nil))
(insn 23 22 21 2 peep2.c:5 (set (reg:SI 15 d15)
(xor:SI (reg:SI 15 d15)
(reg:SI 4 d4 [ x ]))) 39 {*xorsi3} (nil))
(insn 21 23 10 2 peep2.c:5 (set (reg:SI 4 d4)
(reg:SI 15 d15)) 2 {*movsi_insn} (nil))
(call_insn/j 10 21 11 2 peep2.c:5 (parallel [
(set (reg:SI 2 d2)
(call (mem:HI (symbol_ref:SI ("f") [flags
0x41]<function_decl
0xb76b3280 f>) [0 S2 A16])
(const_int 0 [0x0])))
(use (const_int 1 [0x1]))
]) 92 {call_value_insn} (nil)
(expr_list:REG_DEP_TRUE (use (reg:SI 4 d4))
(nil)))
;; End of basic block 2 -> ( 1)
;; lr out 2 [d2] 26 [SP] 27 [a11]
;; live out 2 [d2] 26 [SP] 27 [a11]
;; Succ edge EXIT [100.0%] (ab,sibcall)
(barrier 11 10 20)
D15, is not marked as dead
True. It should have had a REG_DEAD note. Are you using 4.6 (which
scans forwards in peephole2 and requires REG_DEAD notes) or 4.5 (which
scans backwards)? If the latter, the absence of the note could be a red
herring, because 4.5 didn't need REG_DEAD notes so it didn't compute them.
It is a target port of 4.5.1. I just skimmed peep2_reg_dead_p and it looks
backward.
In this case, as I mentioned the function _doesn't need_ a note to see
that D15 dies. There's no such thing as a register that is "naturally
dead because the function ends" and needs special treatment. Liveness
is computed like everything else from data flow and, when scanning
backwards, can be computed simply from the defs and uses of each
instruction. The notes are not necessary, so their absence is (as I
also have mentioned above already) a red herring.
The definition of liveness is lr_before = lr_after - def + use, which gives:
at the end of basic block lr from dump file = {2,26,27}
insn 10 def = {2}, use = {4} lr before insn 10 = {4,26,27}
insn 21 def = {4}, use = {15} lr before insn 21 = {15,26,27}
insn 23 def = {15}, use = {4,15} lr before insn 23 = {4,15,26,27}
insn 22 def = {15}, use = {4} lr before insn 22 = {4,26,27}
When GCC matches the peephole against 21+23, peep2_reg_dead_p(2, reg)
should test reg against "lr before insn 10". This is {4,26,27}. You
have to set a breakpoint in df_simulate_one_insn_backwards and see which
part of my theory above is incorrect.
Paolo