[Bug middle-end/34093] New: [4.3 Regression] ICE in ssa_operand_alloc, at tree-ssa-operands.c:484

["rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

(I realize this is a dup)

struct X { int i; int j; };
#define FOO struct X
#define FOO10(x) FOO x ## 0; FOO x ## 1; FOO x ## 2; FOO x ## 3; FOO x ## 4;
FOO x ## 5; FOO x ## 6; FOO x ## 7; FOO x ## 8; FOO x ## 9;
#define FOO100(x) FOO10(x ## 0) FOO10(x ## 1) FOO10(x ## 2) FOO10(x ## 3)
FOO10(x ## 4) FOO10(x ## 5) FOO10(x ## 6) FOO10(x ## 7) FOO10(x ## 8) FOO10(x
## 9)
  FOO100(x0)
  FOO100(x1)
  FOO100(x2)
  FOO100(x3)
  FOO100(x4)
  FOO100(x5)
  FOO100(x6)
  FOO100(x7)
  FOO100(x8)
  FOO100(x9)

#define COO(n,f) case n: p = &f; break;
#define COO10(n,f) COO(n ## 0, f ## 0) COO(n ## 1, f ## 1) COO(n ## 2, f ## 2)
COO(n ## 3, f ## 3) COO(n ## 4, f ## 4) COO(n ## 5, f ## 5) COO(n ## 6, f ## 6)
COO(n ## 7, f ## 7) COO(n ## 8, f ## 8) COO(n ## 9, f ## 9)
#define COO100(n,f) COO10(n ## 0, f ## 0) COO10(n ## 1, f ## 1) COO10(n ## 2, f
## 2) COO10(n ## 3, f ## 3) COO10(n ## 4, f ## 4) COO10(n ## 5, f ## 5) COO10(n
## 6, f ## 6) COO10(n ## 7, f ## 7) COO10(n ## 8, f ## 8) COO10(n ## 9, f ## 9)

int foo(int i)
{
  struct X *p = 0;
  x000.i = 0;
  x599.j = 0;
  switch (i)
    {
  COO100(1, x0)
  COO100(2, x1)
  COO100(3, x2)
  COO100(4, x3)
  COO100(5, x4)
  COO100(6, x5)
  COO100(7, x6)
  COO100(8, x7)
  COO100(9, x8)
  COO100(10, x9)
    }
  return p->j;
}

./cc1 -quiet t.c -O  
t.c: In function 'foo':
t.c:21: internal compiler error: in ssa_operand_alloc, at
tree-ssa-operands.c:484
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.


this now happens because with the fix for PR33870 we mark all of the
pointed-to SFTs as not partitionable, so partitioning doesn't touch
the aliases of the single memory reference in this testcase.

There are two possible solutions - allow partitioning here (but make
sure to include all SFTs of all structures in this case to not re-open
PR33870 again) or to implement the fallback suggested in tree-ssa-operands.c:

      /* Fail if there is not enough space.  If there are this many operands
         required, first make sure there isn't a different problem causing this
         many operands.  If the decision is that this is OK, then we can
         specially allocate a buffer just for this request.  */
      gcc_assert (size <= gimple_ssa_operands (cfun)->ssa_operand_mem_size);

even if having that many VOPs is silly in the first place.

Note that we can construct a similar testcase without SFTs just adding the
DECLs tags.


-- 
           Summary: [4.3 Regression] ICE in ssa_operand_alloc, at tree-ssa-
                    operands.c:484
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org
 BugsThisDependsOn: 33870


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34093