[Bug middle-end/28970] [4.1 Regression] Wrong code for simple loop test case

[rguenth at gcc dot gnu dot org]

--- Comment #6 from rguenth at gcc dot gnu dot org  2006-10-28 11:06 ---
Ah, no - I didn't look at the result of this first hunt and though it was the
switch from -floop-optimize to -floop-optimize2.  Sorry for the noise.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28970

[Bug middle-end/28970] [4.1 Regression] Wrong code for simple loop test case

[rguenth at gcc dot gnu dot org]

--- Comment #4 from rguenth at gcc dot gnu dot org  2006-10-27 09:50 ---
Janis, can you hunt which path introduced this regression relative from 4.0.0
which seems to work?


-- 

rguenth at gcc dot gnu dot org changed:

   What|Removed |Added

 CC||janis at gcc dot gnu dot org
  Known to work|4.0.0   |4.0.0 4.0.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28970

[Bug middle-end/28970] [4.1 Regression] Wrong code for simple loop test case

[janis at gcc dot gnu dot org]

--- Comment #5 from janis at gcc dot gnu dot org  2006-10-27 16:40 ---
The regression hunt results in comment #2 are from mainline during development
of 4.1.  Is there some other hunt that would be useful as well?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28970

[Bug middle-end/28970] [4.1 Regression] Wrong code for simple loop test case

[rguenth at gcc dot gnu dot org]

--- Comment #3 from rguenth at gcc dot gnu dot org  2006-10-16 12:07 ---
This is a bug in loop.c

bb 0:
  pretmp.61 = q - 1;
  outgo = 0;

L0:;
  outgo.62 = outgo + 1;
  outgo = MIN_EXPR pretmp.61, outgo.62;
  j = tar (outgo * bcount);
  if (j != -1) goto L0; else goto L4;

L4:;
  return;

Bug in biv selection, so strength reduction messes up:

Loop from 45 to 46: 12 real insns.
Biv 60: insn 19 const (reg 59 [ pretmp.26 ])
Biv 60: insn 22 const (1)
Biv 60: verified
Biv 60: initialized at insn 13: initial value (0)
Giv 58: insn 17 src reg 60 benefit 4 lifetime 6

60 is not a Biv:

(insn 19 17 20 1 (set (reg/v:SI 60 [ outgo ])
(reg:SI 59 [ pretmp.26 ])) 40 {*movsi_1} (nil)
(nil))

(insn 20 19 21 1 (set (reg:CCGC 17 flags)
(compare:CCGC (reg/v:SI 60 [ outgo ])
(reg/v:SI 58 [ outgo.27 ]))) 5 {*cmpsi_1_insn} (nil)
(nil))

(jump_insn 21 20 41 1 (set (pc)
(if_then_else (le (reg:CCGC 17 flags)
(const_int 0 [0x0]))
(label_ref 23)
(pc))) 531 {*jcc_1} (nil)
(expr_list:REG_BR_PROB (const_int 5000 [0x1388])
(nil)))
;; End of basic block 1, registers live:
 (nil)

;; Start of basic block 2, registers live: (nil)
(note 41 21 22 2 [bb 2] NOTE_INSN_BASIC_BLOCK)

(insn 22 41 23 2 (set (reg/v:SI 60 [ outgo ])
(reg/v:SI 58 [ outgo.27 ])) 40 {*movsi_1} (nil)
(nil))
;; End of basic block 2, registers live:
 (nil)

;; Start of basic block 3, registers live: (nil)
(code_label 23 22 42 3 3  [1 uses])


now strength reduction does:

(insn 19 17 51 1 (set (reg/v:SI 60 [ outgo ])
(reg:SI 59 [ pretmp.26 ])) -1 (nil)
(nil))

(insn 51 19 20 1 (parallel [
(set (reg:SI 66)
(mult:SI (reg:SI 59 [ pretmp.26 ])
(reg/v:SI 63 [ bcount ])))
(clobber (reg:CC 17 flags))
]) -1 (nil)
(nil))

(insn 20 51 21 1 (set (reg:CCGC 17 flags)
(compare:CCGC (reg/v:SI 60 [ outgo ])
(reg/v:SI 58 [ outgo.27 ]))) -1 (nil)
(nil))

(jump_insn 21 20 41 1 (set (pc)
(if_then_else (le (reg:CCGC 17 flags)
(const_int 0 [0x0]))
(label_ref 23)
(pc))) -1 (nil)
(expr_list:REG_BR_PROB (const_int 5000 [0x1388])
(nil)))
;; End of basic block 1, registers live:
 (nil)

;; Start of basic block 2, registers live: (nil)
(note 41 21 22 2 [bb 2] NOTE_INSN_BASIC_BLOCK)

(insn 22 41 50 2 (set (reg/v:SI 60 [ outgo ])
(reg/v:SI 58 [ outgo.27 ])) -1 (nil)
(nil))

(insn 50 22 23 2 (parallel [
(set (reg:SI 66)
(plus:SI (reg:SI 66)
(reg/v:SI 63 [ bcount ])))
(clobber (reg:CC 17 flags))
]) -1 (nil)
(nil))
;; End of basic block 2, registers live:
 (nil)

;; Start of basic block 3, registers live: (nil)
(code_label 23 50 42 3 3  [1 uses])


-- 

rguenth at gcc dot gnu dot org changed:

   What|Removed |Added

 CC||rakdver at gcc dot gnu dot
   ||org
   Last reconfirmed|2006-09-07 05:01:38 |2006-10-16 12:07:36
   date||


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28970

[Bug middle-end/28970] [4.1 Regression] Wrong code for simple loop test case

[janis at gcc dot gnu dot org]

--- Comment #2 from janis at gcc dot gnu dot org  2006-09-07 20:26 ---
A regression hunt on powerpc-linux identified the following patch:

http://gcc.gnu.org/viewcvs?view=revrev=99850

r99850 | rakdver | 2005-05-17 19:55:53 + (Tue, 17 May 2005)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28970

[Bug middle-end/28970] [4.1 Regression] Wrong code for simple loop test case

[pinskia at gcc dot gnu dot org]

--- Comment #1 from pinskia at gcc dot gnu dot org  2006-09-07 05:01 ---
Confirmed, this is a loop.c bug which is why it is not in 4.2.0 at all.


-- 

pinskia at gcc dot gnu dot org changed:

   What|Removed |Added

 Status|UNCONFIRMED |NEW
 Ever Confirmed|0   |1
  GCC build triplet|powerpc64-linux |
   GCC host triplet|powerpc64-linux |
 GCC target triplet|powerpc64-linux |powerpc-linux, i686-linux-
   ||gnu
   Keywords||wrong-code
  Known to fail||4.1.0 4.1.1 4.1.2
  Known to work||4.0.0
   Last reconfirmed|-00-00 00:00:00 |2006-09-07 05:01:38
   date||
Summary|Wrong code for simple loop  |[4.1 Regression] Wrong code
   |test case   |for simple loop test case
   Target Milestone|--- |4.1.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28970