[Bug fortran/19551] New: LAPACK routine claic1.f bug

[Bug fortran/19551] New: LAPACK routine claic1.f bug

[billingd at gcc dot gnu dot org]

I have a reduced test case for LAPACK CLS Drivers testsuite failures with 
gfortran.  See PR 5900 for background.  The following assumes some knowledge 
of the LAPACK testsuite

Step 1 was to reduce the ctest.in input file for the xlintstc test routine 
(below).

I then identified a point where the g77-3.4 and gfortran results diverged.  
The problem is in routine CLAIC1, which is called by CGELSX and CGELSY, and 
occurs at line 285 of claic1.f, which is

    TMP = SQRT( SINE*CONJG( SINE )+COSINE*CONJG( COSINE ) )

The reduced testcase below:
 o passes with g77-3.4 on cygwin and irix
 o fails with gfortran-20050120 on cygwin

################ test case ##########
      program claic1_bug
      real  x, correct
      parameter(correct=2.30457878)
      call  claic1(x)
      if ( abs(x-correct) .gt. 1.0e-4 ) then
         write(6,*) 'x = ', x,' expected ',correct
         call abort
      end if
      end

      subroutine claic1( tmp )
      implicit none
      real      tmp
      complex   cosine, sine
      sine = cmplx(0.2331063,-0.488660812)
      cosine = cmplx(2.2400794,0.)
      tmp = sqrt( sine*conjg( sine )+cosine*conjg( cosine ) )
      end

############## reduced input file for xlintstc ###################
Data file for testing COMPLEX LAPACK linear equation routines
1                      Number of values of M
3                      Values of M (row dimension)
1                      Number of values of N
3                      Values of N (column dimension)
1                      Number of values of NRHS
1                      Values of NRHS (number of right hand sides)
1                      Number of values of NB
1                      Values of NB (the blocksize)
1                      Values of NX (crossover point)
30.0                   Threshold value of test ratio
T                      Put T to test the LAPACK routines
T                      Put T to test the driver routines
F                      Put T to test the error exits
CLS    1               List types on next line if 0 < NTYPES <  6
1
CEQ

-- 
           Summary: LAPACK routine claic1.f bug
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: billingd at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-cygwin
  GCC host triplet: i686-pc-cygwin
GCC target triplet: i686-pc-cygwin
OtherBugsDependingO 5900
             nThis:


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug fortran/19551] LAPACK routine claic1.f bug

[bdavis at gcc dot gnu dot org]

------- Additional Comments From bdavis at gcc dot gnu dot org  2005-01-21 03:15 -------
works OK on AMD-64
fails on i686


--bud


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-01-21 03:15:27
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug target/19551] LAPACK routine claic1.f bug

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-21 03:17 -------
Hmm, sounds like an extensive precision problem.

Is there any way for a self contained example.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|fortran                     |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug target/19551] LAPACK routine claic1.f bug

[david dot billinghurst at comalco dot riotinto dot com dot au]

------- Additional Comments From david dot billinghurst at comalco dot riotinto dot com dot au  2005-01-21 03:46 -------
Subject: RE:  LAPACK routine claic1.f bug

Pardon?  It is a self contained example.  Took me several hours to reduce it.

I doubt it is an excess precision problem as the numbers involved are all O(1)
and there is no possibility of cancellation.

	David


NOTICE
This e-mail and any attachments are private and confidential and may contain privileged information. If you are not an authorised recipient, the copying or distribution of this e-mail and any attachments is prohibited and you must not read, print or act in reliance on this e-mail or attachments.
This notice should not be removed.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-21 04:15 -------
Hmm (sorry about that), never mind about not being a self contained or excess precision problem 
because it can be  reproduced on 32bit powerpc also.  The problem comes from .16.life so this is not a 
fortran related problem.
This can also reproduced with the following C code (but only at -O1 and above) (the fortran example 
can be reproduce with -O0):
_Complex float csqrtf(_Complex float);
float claic1 (float *tmp)
{
  _Complex float D435;
  _Complex float D433;
  __imag D433 = 0.0;
  __real D433 = 5.31108379364013671875e+0;
  D435 = csqrtf (D433);
  *tmp = __real D435;
}
And that is at least a regression from 3.3.2.
On x86, it also fails with 3.4.0.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
          Component|target                      |middle-end
  GCC build triplet|i686-pc-cygwin              |
   GCC host triplet|i686-pc-cygwin              |
 GCC target triplet|i686-pc-cygwin              |i686-pc-cygwin, powerpc-
                   |                            |darwin
           Keywords|                            |wrong-code
      Known to fail|                            |4.0.0
      Known to work|                            |3.3.2
           Priority|P2                          |P1
            Summary|LAPACK routine claic1.f bug |[3.4/4.0 Regression] LAPACK
                   |                            |routine claic1.f bug
   Target Milestone|---                         |3.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-21 04:17 -------
Note the full C self contained example is:
_Complex float csqrtf(_Complex float);
float claic1 (float *tmp);
void abort (void);

int main (void)
{
  float x;
  float x1;

  x = 10000.0;
  claic1 (&x);
  x1 =x- 2.3045787811279296875;
  x1 = x1>0?x1:-x1;
  if (x1 > 9.99999974737875163555145263671875e-5)
    abort ();
  return 0;
}

float claic1 (float *tmp)
{
  _Complex float D435;
  _Complex float D433;

  __imag D433 = 0.0;
  __real D433 = 5.31108379364013671875e+0;
  D435 = csqrtf (D433);
  *tmp = __real D435;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[Thomas dot Koenig at online dot de]

------- Additional Comments From Thomas dot Koenig at online dot de  2005-01-21 08:14 -------
The test cases from the original description
and from commen #5 work correctly at -O0, -O1,
-O2 and -O3 on ia64-unknown-linux-gnu with the
20050116 snapshot.

       Thomas

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-21 14:32 -------
Hmm, adding -fno-builtin "fixes" the failure, maybe this is libcall problem then.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-01-21 15:46 -------
-fno-builtin only matters in not adding pure attribute to csqrtf.
If you start the #5 testcase with
_Complex float csqrtf(_Complex float) __attribute__((pure));
instead, it fails even with -fno-builtin.
Reproduced on i386-redhat-linux, looking into it.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-01-21 16:13 -------
Simplified testcase:
extern void abort ();

__attribute__((pure)) _Complex float
foo (int x)
{
  _Complex float r;
  __real r = x + 1;
  __imag r = x - 1;
  return r;
}

void
bar (float *f)
{
  *f = foo (5);
}

int
main (void)
{
  float f;
  bar (&f);
  if (f != 6.0)
    abort ();
  return 0;
}

Likely the culprit is in libcall_dead_p.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] LAPACK routine claic1.f bug

[jakub at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2005-01-21 03:15:27         |2005-01-21 18:11:43
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] pure (complex types) function call removed as dead (LAPACK routine claic1.f bug)

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-21 21:37 -------
Jakub you are a life saver some times.  Thanks for looking into this.

Patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-01/msg01529.html>.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch
            Summary|[3.4/4.0 Regression] pure   |[3.4/4.0 Regression] pure
                   |function call removed as    |(complex types) function
                   |dead (LAPACK routine        |call removed as dead (LAPACK
                   |claic1.f bug)               |routine claic1.f bug)


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] pure (complex types) function call removed as dead (LAPACK routine claic1.f bug)

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-24 08:55 -------
Subject: Bug 19551

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	jakub@gcc.gnu.org	2005-01-24 08:54:26

Modified files:
	gcc            : ChangeLog flow.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.c-torture/execute: 20050121-1.c 
	gcc/testsuite/gcc.dg: 20050121-2.c 

Log message:
	* flow.c (propagate_one_insn): Formatting.
	
	PR middle-end/19551
	* flow.c (libcall_dead_p): Be more conservative if unsure.
	If there are any instructions between insn and call, see if they are
	all dead before saying the libcall is dead.
	
	* gcc.c-torture/execute/20050121-1.c: New test.
	* gcc.dg/20050121-2.c: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7255&r2=2.7256
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/flow.c.diff?cvsroot=gcc&r1=1.614&r2=1.615
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.4928&r2=1.4929
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.c-torture/execute/20050121-1.c.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/20050121-2.c.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] pure (complex types) function call removed as dead (LAPACK routine claic1.f bug)

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-24 09:11 -------
Subject: Bug 19551

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_4-branch
Changes by:	jakub@gcc.gnu.org	2005-01-24 09:10:53

Modified files:
	gcc            : ChangeLog flow.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.c-torture/execute: 20050121-1.c 

Log message:
	* flow.c (propagate_one_insn): Formatting.
	
	PR middle-end/19551
	* flow.c (libcall_dead_p): Be more conservative if unsure.
	If there are any instructions between insn and call, see if they are
	all dead before saying the libcall is dead.
	
	* gcc.c-torture/execute/20050121-1.c: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=2.2326.2.780&r2=2.2326.2.781
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/flow.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.572.4.3&r2=1.572.4.4
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.3389.2.351&r2=1.3389.2.352
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.c-torture/execute/20050121-1.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=NONE&r2=1.1.2.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] pure (complex types) function call removed as dead (LAPACK routine claic1.f bug)

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-01-24 09:12 -------
Should be fixed in CVS.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] pure (complex types) function call removed as dead (LAPACK routine claic1.f bug)

[billingd at gcc dot gnu dot org]

------- Additional Comments From billingd at gcc dot gnu dot org  2005-01-24 22:18 -------
I can confirm that the original problem is fixed in CVS.  All the LAPACK CLS 
Driver tests now pass on i686-pc-cygwin.

Thanks you Jakub for the quick fix.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551

[Bug middle-end/19551] [3.4/4.0 Regression] pure (complex types) function call removed as dead (LAPACK routine claic1.f bug)

[jkj at sco dot com]

------- Additional Comments From jkj at sco dot com  2005-08-04 00:57 -------
Just FYI, this still fails for me, from 3.4.4 all teh way up to the mainline,
when using -fPIC. Platform is UnixWare, but thats not significantly different
from the test point of view to any other x86 target. I dont have others I can
test it on. Any ideas why PIC may be causing a problem (other than the obvious
that an extra register is being used by the PIC stuff)?

If it will help I can include -S output from PIC and non-PIC compiles, or any
other debug info required. Any help greatly appreciated.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19551