[Bug target/14322] New: Unrecognizable lo_sum insn with -m64 and -fPIC

[Bug target/14322] New: Unrecognizable lo_sum insn with -m64 and -fPIC

[ehrhardt at mathematik dot uni-ulm dot de]

The following code gives an unrecognizable insn on sparc-sun-solaris
with -m64 -fPIC:

=================================
int foo;
void * f ()
{
        return &foo;
}
=================================

The error message is:

locale.cc: In function `f':
locale.cc:5: error: unrecognizable insn:
(insn 9 8 10 0 (set (reg:DI 112)
        (lo_sum:DI (reg:DI 112)
            (unspec:DI [
                    (symbol_ref:SI ("foo") <var_decl ff202d80 foo>)
                ] 0))) -1 (nil)
    (nil))
locale.cc:5: internal compiler error: in extract_insn, at recog.c:2083
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

-- 
           Summary: Unrecognizable lo_sum insn with -m64 and -fPIC
           Product: gcc
           Version: tree-ssa
            Status: UNCONFIRMED
          Severity: critical
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: ehrhardt at mathematik dot uni-ulm dot de
                CC: gcc-bugs at gcc dot gnu dot org
  GCC host triplet: sparc-sun-solaris2.9


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-02-27 16:56 -------
I can confirm this on the tree-ssa branch (I have not tried the mainline) but this patch fixes it but I do 
not know if the right thing to do.  But it parallels the other uses of UNSPEC_MOVE_PIC.


Index: sparc.md
===============================================================
====
RCS file: /cvs/gcc/gcc/gcc/config/sparc/sparc.md,v
retrieving revision 1.175.2.12
diff -u -p -r1.175.2.12 sparc.md
--- sparc.md    13 Feb 2004 13:13:21 -0000      1.175.2.12
+++ sparc.md    27 Feb 2004 16:53:52 -0000
@@ -1938,7 +1938,7 @@
 (define_insn "movsi_lo_sum_pic"
   [(set (match_operand:SI 0 "register_operand" "=r")
         (lo_sum:SI (match_operand:SI 1 "register_operand" "r")
-                   (unspec:SI [(match_operand:SI 2 "immediate_operand" "in")] UNSPEC_MOVE_PIC)))]
+                   (unspec:SI [(match_operand:SI 2 "" "")] UNSPEC_MOVE_PIC)))]
   "flag_pic"
   "or\t%1, %%lo(%a2), %0")
 
@@ -2240,7 +2240,7 @@
 (define_insn "movdi_lo_sum_pic"
   [(set (match_operand:DI 0 "register_operand" "=r")
         (lo_sum:DI (match_operand:DI 1 "register_operand" "r")
-                   (unspec:DI [(match_operand:DI 2 "immediate_operand" "in")] UNSPEC_MOVE_PIC)))]
+                   (unspec:DI [(match_operand:DI 2 "" "")] UNSPEC_MOVE_PIC)))]
   "TARGET_ARCH64 && flag_pic"
   "or\t%1, %%lo(%a2), %0")
 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |ice-on-valid-code, RTL
      Known to fail|                            |tree-ssa
   Last reconfirmed|0000-00-00 00:00:00         |2004-02-27 16:56:39
               date|                            |
   Target Milestone|---                         |tree-ssa


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-02-27 18:54 -------
Actually can you test this patch with the testsuite as I do not have access to a sparc-sun-
solaris2.9 machine?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING
   GCC host triplet|sparc-sun-solaris2.9        |
 GCC target triplet|                            |sparc-sun-solaris2.9


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-02-27 22:57 -------
This is a laten bug so moving version and milestone.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|tree-ssa                    |3.5.0
            Version|tree-ssa                    |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-02-29 16:24 -------
So I assume the patch works.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Last reconfirmed|2004-02-27 16:56:39         |2004-02-29 16:24:31
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-05 09:02 -------
I don't think Andrew's patch is correct.  I think the problem comes from

(symbol_ref:SI ("foo") <var_decl ff202d80 foo>)

which seems wrong to me because of SImode.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-10 16:01 -------
Christian, I can't reproduce it with either

  gcc version 3.5.0 20040307 (experimental)

or

  gcc version 3.4.0 20040310 (prerelease)

on sparc-sun-solaris2.8.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-10 16:56 -------
That is because it can only be reproduced on the tree-ssa as the RTL is different when it comes to the 
middle-end/back-end.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Last reconfirmed|2004-02-29 16:24:31         |2004-03-10 16:56:59
               date|                            |
            Version|3.5.0                       |tree-ssa


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug middle-end/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-10 18:44 -------
Ah! Ok, thanks for the clarification.

Given the form of the input, I'm leaning towards a middle-end bug.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-10 19:18 -------
Actually I just reproduced it on the mainline (20040310):
./cc1 pr14322.c -m64 -fPIC
 f

pr14322.c: In function `f':
pr14322.c:5: error: unrecognizable insn:
(insn 11 10 12 0 (set (reg:DI 112)
        (lo_sum:DI (reg:DI 112)
            (unspec:DI [
                    (symbol_ref:SI ("foo") <var_decl 0x40eea5e4 foo>)
                ] 0))) -1 (nil)
    (nil))
pr14322.c:5: internal compiler error: in extract_insn, at recog.c:2074
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

Now if I do:
./cc1 pr14322.c -m64 -fPIC -mptr64
It works, so this is a target bug.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target
      Known to fail|tree-ssa                    |tree-ssa 3.5.0
            Version|tree-ssa                    |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-10 19:48 -------
I see.  -m64 is not supported without -mptr64 on SPARC.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-10 19:51 -------
Well then it still should not ICE.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|INVALID                     |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-10 19:52 -------
So how the other targets handle the case where an option depends on another option.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |NEW
           Keywords|ice-on-valid-code, RTL      |accepts-invalid
   Target Milestone|3.5.0                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug middle-end/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-10 19:58 -------
I don't know.  But -m64 simply doesn't make sense without -mptr64 for the SPARC
back-end and this is not going to change.  So either you have a testcase that
fails when passed to the compiler driver and I think it is a middle-end bug or
you don't and the bug can be closed.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-10 20:10 -------
No the bug cannot be closed as GCC should never ICE on any input this is policy, it does not matter if 
-mptr64 is not supplied at all when -m64 is used at all.  See how other targets deal with the problem.
For an example:
-m64 on PPC64 Linux requires that you supply a -mcpu that has 64bit support:
      if (TARGET_64BIT)                                         \
        {                                                       \
          if ((target_flags & MASK_POWERPC64) == 0)             \
            {                                                   \
              target_flags |= MASK_POWERPC64;                   \
              error ("-m64 requires a PowerPC64 cpu");          \
            }                                                   \

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target
   Target Milestone|3.5.0                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-10 20:57 -------
This is specious, because the ICE cannot be triggered on mainline through the
compiler driver.  So either it is really a bug on tree-ssa and I don't think how
it could not come from the middle-end or the summary of the PR is misleading and
it can be closed.  Now if you want to open a PR for every pair of options passed
directly to cc1 that misbehaves, fine, but IMHO this is not worth bothering.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug target/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ehrhardt at mathematik dot uni-ulm dot de]

------- Additional Comments From ehrhardt at mathematik dot uni-ulm dot de  2004-03-13 16:08 -------
Subject: Re:  Unrecognizable lo_sum insn with -m64 and -fPIC


This bug (unrecognizable lo_sum insn) triggered for me during bootstrap
of tree-ssa on sparc-sun-solaris. So this can't be closed IMHO. Maybe I did
someting wrong in reducing the testcase that made the testcase invalid ...
I'm willing to investigate further but can't do so in the next two weeks
(away from mail).         regards   Christian



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug bootstrap/14322] Unrecognizable lo_sum insn with -m64 and -fPIC

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2004-03-15 13:14 -------
Ok, so let's put aside the cc1 problem on mainline and requalify the bug as a
bootstrap failure on tree-ssa.

However, the only way I can think of to produce

  (symbol_ref:SI ("foo"))

with -m64, which is invalid according to the manual

     The `symbol_ref' contains a mode, which is usually `Pmode'.
     Usually that is the only mode for which a symbol is directly valid.

because Pmode == DImode with -m64 in all cases, is to pass -mptr32 to cc1.


On mainline, the problem comes from these lines in convert_memory_address:

    case SYMBOL_REF:
      temp = shallow_copy_rtx (x);
      PUT_MODE (temp, to_mode);
      return temp;
      break;

which appears to go against the manual.  But this can't happen on mainline
through the compiler driver since all 64-bit SPARC ports in the FSF tree are
LP64 and, therefore, pass -mptr64 to cc1.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
          Component|target                      |bootstrap
           Keywords|accepts-invalid             |ice-on-valid-code
      Known to fail|tree-ssa 3.5.0              |tree-ssa
           Priority|P2                          |P1
            Version|3.5.0                       |tree-ssa


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug middle-end/14322] [tree-ssa] bootstrap failure: ICE in extract_insn with -m64 -fPIC

[ehrhardt at mathematik dot uni-ulm dot de]

------- Additional Comments From ehrhardt at mathematik dot uni-ulm dot de  2004-03-26 13:51 -------
I just did a cvs up and can't reproduce the tree-ssa bootstrap problem
anymore. So as far as I'm concerned this is fixed. Thanks for the help.

    regards  Christian



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322

[Bug middle-end/14322] [tree-ssa] bootstrap failure: ICE in extract_insn with -m64 -fPIC

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-03-26 13:56 -------
Closing as fixed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |tree-ssa


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14322