[Bug middle-end/23706] New: [4.1 Regression] ICE in rtl_verify_flow_info_1

[Bug middle-end/23706] New: [4.1 Regression] ICE in rtl_verify_flow_info_1

[kkojima at gcc dot gnu dot org]

sh4-unknown-linux-gnu build fails when compiling libjava/interpret.cc.
with "error: missing REG_EH_REGION note in the end of bb".
I've attached a reduced preprocessed file.

-- 
           Summary: [4.1 Regression] ICE in rtl_verify_flow_info_1
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code, build
          Severity: normal
          Priority: P2
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: kkojima at gcc dot gnu dot org
                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: sh4-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug middle-end/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[kkojima at gcc dot gnu dot org]

------- Additional Comments From kkojima at gcc dot gnu dot org  2005-09-02 23:33 -------
Created an attachment (id=9655)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9655&action=view)
testcase

cc1plus -fnon-call-exceptions interpret.i


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug middle-end/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[rth at gcc dot gnu dot org]

------- Additional Comments From rth at gcc dot gnu dot org  2005-09-03 00:27 -------
The true problem is in optimize_mode_switching.  The code about line 605
that inserts code around (but not on) abnormal edges, is incorrect.  This
is immediately obvious in this case because the emit_insn_after must by
definition insert insns after the last insn in the block, resulting in
the observed assertion failure.

The solution will have to involve re-organizing optimize_mode_switching
with respect to abnormal edges.

As for why this is suddenly showing up, I can only imagine that the removal
of EH edges (by marking local stack frame slots non-trapping) have changed
the solution LCM finds to the mode switching problem.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rth at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-09-03 00:27:51
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug middle-end/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-03 00:36 -------
I am going to reduce this, just for fun.  I am also going to apply the testcase for PR 23676 so it will be 
tested without enabling libjava.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug middle-end/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[kkojima at gcc dot gnu dot org]

------- Additional Comments From kkojima at gcc dot gnu dot org  2005-09-03 00:40 -------
I'll try the patch.  Thanks.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug target/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-03 03:29 -------
Here is the reduced testcase:
void run (void) {
  float stack[1];
  float *sp = stack;
  try
  {
    float value2 = ((float) *(--sp));
    float value1 = ((float) *(--sp));
    *(sp++) = (value1 - value2);
  }
  catch (int *ex)
  {
  }
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug target/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[kkojima at gcc dot gnu dot org]

------- Additional Comments From kkojima at gcc dot gnu dot org  2005-09-03 07:07 -------
There are no new failures on c/c++/objc tests with the patch but
it makes 120 new failures on gfortran execution test.  The patched
compiler still fails for libjava/interpret.cc and Andrew's
reduced testcase.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706

[Bug target/23706] [4.1 Regression] ICE in rtl_verify_flow_info_1

[kkojima at gcc dot gnu dot org]

------- Additional Comments From kkojima at gcc dot gnu dot org  2005-09-03 09:20 -------
BTW, I've found that Andrew's testcase fails even on i686-linux with
"-O2 -fnon-call-exceptions":

  internal compiler error: in expand_assignment, at expr.c:3929

I've filed it as PR 23714.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23706