[Bug fortran/17075] New: galgel crashes if compiled with -foptimize-sibling-calls

[Bug fortran/17075] New: galgel crashes if compiled with -foptimize-sibling-calls

[lei at il dot ibm dot com]

When SPEC 2000 benchmark 178.galgel is compiled with -O2
option, galgel crashes.
This problem is possibly specific to ia32.  I was unable
to reproduce it on ppc.

I don't have a small testcase that reproduces the problem. 

The problem seems to be with the function DNRM2 inside lapak.f90.
If that function is compiled with -O2 -fno-optimize-sibling-calls,
galgel works okay.  The problem seems to be connected to the call to SQRT.

      function dnrm2 ( n, dx, incx)
      real*8 dnrm2
      integer i, incx, ix, j, n, next
      real*8   dx(*), cutlo, cuthi, hitest, sum, xmax,zero,one
      data   zero, one /0.0d0, 1.0d0/
      data cutlo, cuthi / 8.232d-11,  1.304d19 /
      if(n .gt. 0) go to 10
         dnrm2  = zero
         go to 300
   10 assign 30 to next
      sum = zero
      i = 1
      if( incx .lt. 0 )i = (-n+1)*incx + 1
      ix = 1
   20    go to next,(30, 50, 70, 110)
   30 if( dabs(dx(i)) .gt. cutlo) go to 85
      assign 50 to next
      xmax = zero
   50 if( dx(i) .eq. zero) go to 200
      if( dabs(dx(i)) .gt. cutlo) go to 85
      assign 70 to next
      go to 105
  100 continue
      ix = j
      assign 110 to next
      sum = (sum / dx(i)) / dx(i)
  105 xmax = dabs(dx(i))
      go to 115
   70 if( dabs(dx(i)) .gt. cutlo ) go to 75
  110 if( dabs(dx(i)) .le. xmax ) go to 115
         sum = one + sum * (xmax / dx(i))**2
         xmax = dabs(dx(i))
         go to 200
  115 sum = sum + (dx(i)/xmax)**2
      go to 200
   75 sum = (sum * xmax) * xmax
   85 hitest = cuthi/float( n )
      do 95 j = ix,n
      if(dabs(dx(i)) .ge. hitest) go to 100
         sum = sum + dx(i)**2
         i = i + incx
   95 continue
      dnrm2 = dsqrt( sum )
      go to 300
  200 continue
      ix = ix + 1
      i = i + incx
      if( ix .le. n ) go to 20
      dnrm2 = xmax * dsqrt(sum)
  300 continue
      return
      end

-- 
           Summary: galgel crashes if compiled with -foptimize-sibling-calls
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: lei at il dot ibm dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] galgel crashes if compiled with -foptimize-sibling-calls

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |15502
              nThis|                            |
          Component|fortran                     |middle-end
           Keywords|                            |wrong-code


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] galgel crashes if compiled with -foptimize-sibling-calls

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-08-18 15:44 -------
Confirmed (yes it had to do with the sqrt):
Here is a simple testcase:
#include <stdio.h>
double f(double x){return __builtin_sqrt(x);}
int main()
{
  printf("%f\n", f(10));
}

We get printted NAN instead of the correct answer.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
      Known to fail|                            |3.5.0
      Known to work|                            |3.4.0
   Last reconfirmed|0000-00-00 00:00:00         |2004-08-18 15:44:39
               date|                            |
            Summary|galgel crashes if compiled  |[3.5 Regression] galgel
                   |with -foptimize-sibling-    |crashes if compiled with -
                   |calls                       |foptimize-sibling-calls
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] galgel crashes if compiled with -foptimize-sibling-calls

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2004-08-18 15:56 -------
I'll have a look. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |steven at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] galgel crashes if compiled with -foptimize-sibling-calls

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2004-08-22 18:47 -------
This bug is caused by the following patch: 
 
2004-07-21  Richard Henderson  <rth@redhat.com> 
 
        * cfgexpand.c (expand_gimple_tailcall): Fix case where we need 
        to create a new basic block. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] galgel crashes if compiled with -foptimize-sibling-calls

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2004-08-22 21:54 -------
The problem appears to be that the function result assignment is 
removed in this part of the patch: 
 
 
+  while (NEXT_INSN (last)) 
+    { 
+      /* For instance an sqrt builtin expander expands if with 
+         sibcall in the then and label for `else`.  */ 
+      if (LABEL_P (NEXT_INSN (last))) 
+        break; 
+      delete_insn (NEXT_INSN (last)); 
+    } 
 
"before" is just before we enter this loop, after is when we've 
finished the loop: 
 
--- before      2004-08-22 23:43:38.188246232 +0200 
+++ after       2004-08-22 23:44:31.837195217 +0200 
@@ -27,21 +27,9 @@ 
         (nil)) 
     (nil)) 
 
-(barrier 17 15 16) 
+(barrier 17 15 19) 
 
-(insn 16 17 18 (parallel [ 
-            (set (reg/f:SI 7 sp) 
-                (plus:SI (reg/f:SI 7 sp) 
-                    (const_int 8 [0x8]))) 
-            (clobber (reg:CC 17 flags)) 
-        ]) -1 (nil) 
-    (nil)) 
- 
-(insn 18 16 19 (set (reg:DF 61) 
-        (reg:DF 8 st)) -1 (nil) 
-    (nil)) 
- 
-(code_label 19 18 20 2 "" [0 uses]) 
+(code_label 19 17 20 2 "" [0 uses]) 
 
 (insn 20 19 0 (set (reg:DF 58 [ T.0 ]) 
         (reg:DF 61)) -1 (nil) 
 
 
The result of that loop is that the set for reg 61 is deleted, so 
that when we do (set (T.0) (reg 61)), nothing has set reg 61 to 
the function result at that point. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] miscompilation with tail calls in cfgexpand

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2004-08-22 22:05 -------
I bet this could manifest itself with builtins on other targets 
as well. 
(Unfortunately I have no idea what intention rth had 
with the patch that breaks this, as it is supposed to fix another 
problem from an earlier patch. Rth??) 
 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P2                          |P1
            Summary|[3.5 Regression] galgel     |[3.5 Regression]
                   |crashes if compiled with -  |miscompilation with tail
                   |foptimize-sibling-calls     |calls in cfgexpand


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] miscompilation with tail calls in cfgexpand

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2004-08-22 22:34 -------
Testing a patch.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|steven at gcc dot gnu dot   |rth at gcc dot gnu dot org
                   |org                         |
           Priority|P1                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] miscompilation with tail calls in cfgexpand

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-08-23 00:03 -------
Subject: Bug 17075

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rth@gcc.gnu.org	2004-08-23 00:02:58

Modified files:
	gcc            : ChangeLog cfgexpand.c 

Log message:
	PR 17075
	* cfgexpand.c (expand_gimple_tailcall): Detect when we can both
	create new blocks and fall through.
	(expand_gimple_basic_block): Update to match.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.5023&r2=2.5024
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cfgexpand.c.diff?cvsroot=gcc&r1=2.17&r2=2.18



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075

[Bug middle-end/17075] [3.5 Regression] miscompilation with tail calls in cfgexpand

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2004-08-23 00:07 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17075