[Bug rtl-optimization/21767] New: if-convert leaves invalid REG_EQUAL notes

[Bug rtl-optimization/21767] New: if-convert leaves invalid REG_EQUAL notes

[amylaar at gcc dot gnu dot org]

if-convert sometimes moves instructions from after to before a conditional
jump.  Some REG_EQUAL are no longer true after this transformation, yet
they are not removed.

-- 
           Summary: if-convert leaves invalid REG_EQUAL notes
           Product: gcc
           Version: 3.4.3
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P2
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: amylaar at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
OtherBugsDependingO 20070
             nThis:


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-05-26 14:44 -------
Created an attachment (id=8971)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=8971&action=view)
Testcase for sh-elf -O2

When you compile this testcase with -O2 -da for sh-elf, you'll see in the
ce2 dump:

(insn 12 7 13 0 (set (reg:SI 161)
	(const_int 1234567 [0x12d687])) 171 {movsi_i} (nil)
    (nil))

(insn 13 12 17 0 (set (reg:SI 147 t)
	(eq:SI (reg/v:SI 160 [ i ])
	    (reg:SI 161))) 1 {cmpeqsi_t} (insn_list:REG_DEP_TRUE 6
(insn_list:RE
G_DEP_TRUE 12 (nil)))
    (expr_list:REG_DEAD (reg:SI 161)
	(expr_list:REG_EQUAL (eq:SI (reg/v:SI 160 [ i ])
		(const_int 1234567 [0x12d687]))
	    (nil))))

(insn 17 13 14 0 (set (reg:SI 158 [ D.1149 ])
	(reg/v:SI 160 [ i ])) 171 {movsi_i} (nil)
    (expr_list:REG_EQUAL (const_int 1234567 [0x12d687])
	(nil)))

(jump_insn 14 17 20 0 (set (pc)
	(if_then_else (ne (reg:SI 147 t)
		(const_int 0 [0x0]))
	    (label_ref:SI 34)
	    (pc))) 207 {branch_true} (insn_list:REG_DEP_TRUE 13 (nil))
    (expr_list:REG_DEAD (reg:SI 147 t)
	(expr_list:REG_BR_PROB (const_int 1675 [0x68b])
	    (nil))))

The REG_EUQAL note on insn 17 has become invalid when this insn was moved
in front of insn 14.  The equality only holds when this branch is taken.

With the patch for PR20070, this invalid transformation is already done in
ce1, which is followed by cse2.  cse2 will use the REG_EQUAL notes to conclude
that insn 14 is always taken.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-06-01 11:39 -------
Subject: Bug 21767

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	amylaar@gcc.gnu.org	2005-06-01 11:38:54

Modified files:
	gcc            : ChangeLog rtl.h reload1.c ifcvt.c 

Log message:
	PR rtl-optimization/21767
	* rtl.h (function_invariant_p): Re-add declaration.
	* reload1.c (function_invariant_p): No longer static.
	* ifcvt.c (dead_or_predicable): Remove REG_EQUAL notes that
	might have become invalid.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.8980&r2=2.8981
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/rtl.h.diff?cvsroot=gcc&r1=1.550&r2=1.551
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/reload1.c.diff?cvsroot=gcc&r1=1.471&r2=1.472
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ifcvt.c.diff?cvsroot=gcc&r1=1.189&r2=1.190



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-06-01 11:45 -------
Fixed on mainline with patch just checked in.
Now we have to decide if we want this fix (without the reload1.c / rtl.h
hunks) in 3.4.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[amylaar at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|20070                       |
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[giovannibajo at libero dot it]

------- Additional Comments From giovannibajo at libero dot it  2005-06-01 18:10 -------
Well, 3.4 is open for regression fixes only. Did the testcase ever work before 
3.4 in any previous GCC version?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |amylaar at gcc dot gnu dot
                   |dot org                     |org
             Status|UNCONFIRMED                 |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-06-01 18:38 -------
(In reply to comment #4)
> Well, 3.4 is open for regression fixes only. Did the testcase ever work before 
> 3.4 in any previous GCC version?

Of course.  if-conversion is a relatively new pass:

2000-04-30  Richard Henderson  <rth@cygnus.com>

        * ifcvt.c: New file.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767

[Bug rtl-optimization/21767] if-convert leaves invalid REG_EQUAL notes

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-06-01 18:53 -------
(In reply to comment #4)
> Well, 3.4 is open for regression fixes only. Did the testcase ever work before 
> 3.4 in any previous GCC version?

P.S.: There used to be an euivalent optimization in jump.c, but that one
explicitly acknowledged the problem with REG_EQUAL notes possibly confusing
cse, and didn't do the transformation if a non-trivial REG_EQUAL note was
present.  For details, do:
cvs -z 9 dif -puw -r1.119 -r1.120 jump.c|less
and search for:
x = a; goto l



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21767