[Bug tree-optimization/52275] New: The polyhedron test air.f90 is miscompiled with '-O2 -floop-flatten' after revision 184265

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52275

             Bug #: 52275
           Summary: The polyhedron test air.f90 is miscompiled with '-O2
                    -floop-flatten' after revision 184265
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassig...@gcc.gnu.org
        ReportedBy: domi...@lps.ens.fr
                CC: bur...@gcc.gnu.org, gros...@gcc.gnu.org,
                    rgue...@gcc.gnu.org
              Host: x86_64-apple-darwin10
            Target: x86_64-apple-darwin10
             Build: x86_64-apple-darwin10


Revision 184265 fixes the ICE I saw on the 2005 polyhedron test air.f90.
However as reported in comment #9 of pr50561, the test is miscompiled with '-O2
-floop-flatten' and higher optimization:

  ITERATION#  TIME             FINAL          MASS RESIDUAL
       10    0.000048597138   0.01000000    3.390685
       23    0.000104076310   0.01000000    3.149714
       37    0.000153111889   0.01000000    3.155547
       53    0.000203760449   0.01000000    3.121681
       70    0.000253653805   0.01000000    3.304361
       90    0.000307361190   0.01000000         NaN
       91  100.000307361190   0.01000000         NaN
 deltat, final t, iterations
   100.00000000000000        100.00030736119007               91
 10   3.3906853070974479     
 23   3.1497139386725364     
 37   3.1555468854046320     
 53   3.1216810612141770     
 70   3.3043611633047654     
 90                       NaN
 91                       NaN

or with -Ofast

  ITERATION#  TIME             FINAL          MASS RESIDUAL
  a zero delta time was detected  0.29999999999999999        0.0000000000000000 
 deltat, final t, iterations
   9.4702850481575148E-006                       NaN           2

These miscompilations disappear if I add -fcheck=bound or -fwhole-program, but
not if I add -flto. Note also that the test pass on powerpc-apple-darwin9.

AFAICT the air.f90 is the only test in the suite having loops candidate for
loop reversal or loop flattening, e.g.,

@@ -880,13 +880,13 @@
          DO iy = 1 , NDY
             maxy = maxy + NPY(iy) + 1
             dtd = 100.0
-            DO i = minx , maxx
-               DO j = miny , maxy
+            DO j = miny , maxy
+               DO i = minx , maxx
                   IF ( dtt(i,j).LE.dtd ) dtd = dtt(i,j)
                ENDDO
             ENDDO
-            DO i = minx , maxx
-               DO j = miny , maxy
+            DO j = miny , maxy
+               DO i = minx , maxx
                   dtt(i,j) = dtd
                ENDDO
             ENDDO

So far graphite has been unable to perform this optimization and the modified
test is also miscompiled.