[Bug target/20322] New: Miscompilation of libcpp/expr.c at -O2+

[Bug target/20322] New: Miscompilation of libcpp/expr.c at -O2+

[jakub at gcc dot gnu dot org]

libcpp/expr.c is miscompiled when compiled with gcc-4_0-branch GCC on x86-64
(and likely i386 as well).
expr.c miscompiled that way causes a testsuite regression on arith-3.c:
FAIL: gcc.dg/cpp/arith-3.c  (test for bogus messages, line 257)

Self-contained testcase that reproduces the problem is:
extern void abort (void);

typedef unsigned long long T;
typedef struct
{
  T a, b;
  unsigned char c, d;
} S;

#define M (sizeof (T) * 4)

S __attribute__((noinline))
foo (T x, T y)
{
  S e;
  T f[2], g;

  e.b = (x & (~(T) 0 >> M)) * (y & (~(T) 0 >> M));
  e.a = (x >> M) * (y >> M);

  f[0] = (x & (~(T) 0 >> M)) * (y >> M);
  f[1] = (x >> M) * (y & (~(T) 0 >> M));

  g = e.b;
  e.b += (f[0] & (~(T) 0 >> M)) << M;
  if (e.b < g)
    e.a++;

  g = e.b;
  e.b += (f[1] & (~(T) 0 >> M)) << M;
  if (e.b < g)
    e.a++;

  e.a += (f[0] >> M);
  e.a += (f[1] >> M);
  e.c = 1;
  e.d = 0;

  return e;
}

int
main (void)
{
  T x = 1ULL << (M * 2 - 1);
  S y = foo (1, x);
  if (y.a || y.b != x || y.c != 1 || y.d)
    abort ();
  return 0;
}

-- 
           Summary: Miscompilation of libcpp/expr.c at -O2+
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jakub at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
GCC target triplet: x86_64-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0 Regression] Miscompilation of libcpp/expr.c at -O2+

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
           Keywords|                            |wrong-code
            Summary|Miscompilation of           |[4.0 Regression]
                   |libcpp/expr.c at -O2+       |Miscompilation of
                   |                            |libcpp/expr.c at -O2+
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.0 Regression]            |[4.0/4.1 Regression]
                   |Miscompilation of           |Miscompilation of
                   |libcpp/expr.c at -O2+       |libcpp/expr.c at -O2+


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-03-04 18:30 -------
>From initial skimming, this sounds like combiner bug.
(insn 78 41 79 0 (set (reg:CC 17 flags)
        (compare:CC (reg:DI 58 [ e$b.9 ])
            (reg:DI 59 [ e$b.8 ]))) 2 {cmpdi_1_insn_rex64}
(insn_list:REG_DEP_TRUE 41 (nil))
    (expr_list:REG_DEAD (reg:DI 59 [ e$b.8 ])
        (nil)))

(insn 79 78 52 0 (parallel [
            (set (reg:DI 64 [ e$a ])
                (plus:DI (plus:DI (ltu:DI (reg:CC 17 flags)
                            (const_int 0 [0x0]))
                        (reg:DI 64 [ e$a ]))
                    (const_int 0 [0x0])))
            (clobber (reg:CC 17 flags))
        ]) 174 {adddi3_carry_rex64} (insn_list:REG_DEP_TRUE 77
(insn_list:REG_DEP_TRUE 78 (nil)))
    (expr_list:REG_DEAD (reg:CC 17 flags)
        (expr_list:REG_UNUSED (reg:CC 17 flags)
            (nil))))

(insn 52 79 53 0 (parallel [
            (set (reg:DI 76)
                (lshiftrt:DI (reg:DI 61 [ f$0 ])
                    (const_int 32 [0x20])))
            (clobber (reg:CC 17 flags))
        ]) 435 {*lshrdi3_1_rex64} (nil)
    (expr_list:REG_DEAD (reg:DI 61 [ f$0 ])
        (expr_list:REG_UNUSED (reg:CC 17 flags)
            (nil))))

(insn 53 52 55 0 (parallel [
            (set (reg:DI 60 [ e$a.6 ])
                (plus:DI (reg:DI 64 [ e$a ])
                    (reg:DI 76)))
            (clobber (reg:CC 17 flags))
        ]) 192 {*adddi_1_rex64} (insn_list:REG_DEP_TRUE 79
(insn_list:REG_DEP_TRUE 52 (nil)))
    (expr_list:REG_DEAD (reg:DI 64 [ e$a ])
        (expr_list:REG_DEAD (reg:DI 76)
            (expr_list:REG_UNUSED (reg:CC 17 flags)
                (nil)))))

before combine is turned into:
(insn 78 41 79 0 (set (reg:CC 17 flags)
        (compare:CC (reg:DI 58 [ e$b.9 ])
            (reg:DI 59 [ e$b.8 ]))) 2 {cmpdi_1_insn_rex64}
(insn_list:REG_DEP_TRUE 41 (nil))
    (expr_list:REG_DEAD (reg:DI 59 [ e$b.8 ])
        (nil)))

(note 79 78 52 0 NOTE_INSN_DELETED)

(insn 52 79 53 0 (parallel [
            (set (reg:DI 76)
                (lshiftrt:DI (reg:DI 61 [ f$0 ])
                    (const_int 32 [0x20])))
            (clobber (reg:CC 17 flags))
        ]) 435 {*lshrdi3_1_rex64} (nil)
    (expr_list:REG_UNUSED (reg:CC 17 flags)
        (expr_list:REG_DEAD (reg:DI 61 [ f$0 ])
            (nil))))

(insn 53 52 55 0 (parallel [
            (set (reg:DI 60 [ e$a.6 ])
                (plus:DI (plus:DI (ltu:DI (reg:CC 17 flags)
                            (const_int 0 [0x0]))
                        (reg:DI 64 [ e$a ]))
                    (reg:DI 76)))
            (clobber (reg:CC 17 flags))
        ]) 174 {adddi3_carry_rex64} (insn_list:REG_DEP_TRUE 77
(insn_list:REG_DEP_TRUE 52 (nil)))
    (expr_list:REG_UNUSED (reg:CC 17 flags)
        (expr_list:REG_DEAD (reg:DI 76)
            (expr_list:REG_DEAD (reg:DI 64 [ e$a ])
                (nil)))))

which shouldn't be done, because the middle instruction is kept and clobbers
(reg:CC 17 flags), so combiner can't move use of %eflags from the first to the
last insn accross the clobber.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-03-07 16:06 -------
Patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-03/msg00647.html>

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|                            |1
           Keywords|                            |patch
   Last reconfirmed|0000-00-00 00:00:00         |2005-03-07 16:06:06
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-03-10 21:26 -------
Subject: Bug 20322

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	jakub@gcc.gnu.org	2005-03-10 21:26:12

Modified files:
	gcc            : ChangeLog combine.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.dg: 20050307-1.c 

Log message:
	PR target/20322
	* combine.c (try_combine): If recog_for_combine added CLOBBERs
	to NEWI2PAT, fail to combine if they are used by NEWPAT.
	
	* gcc.dg/20050307-1.c: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7775&r2=2.7776
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/combine.c.diff?cvsroot=gcc&r1=1.478&r2=1.479
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.5138&r2=1.5139
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/20050307-1.c.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-03-10 21:30 -------
Subject: Bug 20322

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-4_0-branch
Changes by:	jakub@gcc.gnu.org	2005-03-10 21:29:56

Modified files:
	gcc            : ChangeLog combine.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.dg: 20050307-1.c 

Log message:
	PR target/20322
	* combine.c (try_combine): If recog_for_combine added CLOBBERs
	to NEWI2PAT, fail to combine if they are used by NEWPAT.
	
	* gcc.dg/20050307-1.c: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=2.7592.2.39&r2=2.7592.2.40
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/combine.c.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=1.475&r2=1.475.4.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=1.5084.2.33&r2=1.5084.2.34
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/20050307-1.c.diff?cvsroot=gcc&only_with_tag=gcc-4_0-branch&r1=NONE&r2=1.1.2.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-03-10 21:31 -------
Fixed in CVS.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322

[Bug target/20322] [4.0/4.1 Regression] Miscompilation of libcpp/expr.c at -O2+

[uros at kss-loka dot si]

------- Additional Comments From uros at kss-loka dot si  2005-03-11 14:01 -------
Just FYI: arith-3.c still fails for me on pentium4. The sistem is fairly
unpatched Red Hat Linux release 8.0 (Psyche):

Linux moron 2.4.18-14 #1 Wed Sep 4 13:35:50 EDT 2002 i686 i686 i386 GNU/Linux

Compiler version: 4.1.0 20050310 (experimental) 
Platform: i686-pc-linux-gnu
libc-2.2.93.

Running target unix
FAIL: gcc.dg/cpp/arith-3.c  (test for bogus messages, line 257)
XPASS: gcc.dg/vect/vect-22.c scan-tree-dump-times vectorized 3 loops 1

However, testcase from description compiles and runs OK with "gcc -O2 pr20322.c".


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20322