[Bug bootstrap/37086] New: Cross-compilers built with GCC 3.4 do not work

[Bug bootstrap/37086] New: Cross-compilers built with GCC 3.4 do not work

[drow at gcc dot gnu dot org]

Building libgcc, I get:

/scratch/dan/mips-nonpic-upstream/src/gcc-mainline/libgcc/../gcc/unwind-pe.h:
In function 'size_of_encoded_value':
/scratch/dan/mips-nonpic-upstream/src/gcc-mainline/libgcc/../gcc/unwind-pe.h:78:
internal compiler error: tree check: accessed elt 7 of tree_vec with 5 elts in
find_switch_asserts, at tree-vrp.c:4346

/scratch/dan/mips-nonpic-upstream/src/gcc-mainline/libgcc/../gcc/unwind-dw2.c:1257:
internal compiler error: tree check: accessed elt 9 of tree_vec with 7 elts in
find_switch_asserts, at tree-vrp.c:4346

If I recompile tree-vrp.o with -O0, then libgcc compiles OK.

Joseph suggests this came in at the time of the tuples merge.  If possible, it
would be nice if we could avoid the problematic construct...

The code in question is:

      /* If there are multiple case labels with the same destination
         we need to combine them to a single value range for the edge.  */
      if (idx + 1 < n
          && CASE_LABEL (cl) == CASE_LABEL (TREE_VEC_ELT (vec2, idx + 1)))
        {
          /* Skip labels until the last of the group.  */
          do {
            ++idx;
          } while (idx < n
                   && CASE_LABEL (cl) == CASE_LABEL (TREE_VEC_ELT (vec2,
idx)));
          --idx;

When we reach the outer if, idx == 3 and n == 5.  We go around twice, but idx
goes up to 7.

It looks like a loop optimization issue:

0x84fa0fd <find_switch_asserts+1309>:   lea    0x2(%edi),%esi
0x84fa100 <find_switch_asserts+1312>:   mov    -0x40(%ebp),%edi
0x84fa103 <find_switch_asserts+1315>:   cmp    -0x3c(%ebp),%edi
0x84fa106 <find_switch_asserts+1318>:   mov    %esi,-0x40(%ebp)
0x84fa109 <find_switch_asserts+1321>:   jb     0x84fa080
<find_switch_asserts+1184>
0x84fa10f <find_switch_asserts+1327>:   mov    -0x38(%ebp),%ecx
0x84fa112 <find_switch_asserts+1330>:   dec    %esi
0x84fa113 <find_switch_asserts+1331>:   mov    %esi,-0x40(%ebp)

-0x40(%ebp) is idx, and idx+1 is in %edi before the above code.


-- 
           Summary: Cross-compilers built with GCC 3.4 do not work
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: drow at gcc dot gnu dot org
  GCC host triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] Cross-compilers built with GCC 3.4 do not work

[joseph at codesourcery dot com]

------- Comment #1 from joseph at codesourcery dot com  2008-08-11 22:19 -------
Subject: Re:   New: Cross-compilers built with GCC 3.4
 do not work

On Mon, 11 Aug 2008, drow at gcc dot gnu dot org wrote:

> Joseph suggests this came in at the time of the tuples merge.  If possible, it
> would be nice if we could avoid the problematic construct...

Specifically, I did a binary search that determined the tuples merge as 
the revision at which this broke.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] Cross-compilers built with GCC 3.4 do not work

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2008-08-12 08:41 -------
As this is a bug in GCC 3.4 let's close this as invalid (can't we do a
C only host bootstrap for building cross-compilers to work around issues
with host compilers?)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] Cross-compilers built with GCC 3.4 do not work

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2008-08-12 08:42 -------
*** Bug 37091 has been marked as a duplicate of this bug. ***


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |nightstrike at gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] Cross-compilers built with GCC 3.4 do not work

[drow at gcc dot gnu dot org]

------- Comment #4 from drow at gcc dot gnu dot org  2008-08-12 11:55 -------
In the past, we've worked around bugs like this in widely used compilers - they
make life difficult.  Obviously it depends how nasty a workaround is found!


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] Cross-compilers built with GCC 3.4 do not work

[joseph at codesourcery dot com]

------- Comment #5 from joseph at codesourcery dot com  2008-08-12 14:00 -------
Subject: Re:  Cross-compilers built with GCC 3.4 do not
 work

On Tue, 12 Aug 2008, rguenth at gcc dot gnu dot org wrote:

> As this is a bug in GCC 3.4 let's close this as invalid (can't we do a
> C only host bootstrap for building cross-compilers to work around issues
> with host compilers?)

Existing documented practice is to support building cross compilers using 
2.95 or later.  Again, this (and keeping the requirement to 3.4 as 
previously discussed if C++ use is introduced, and ensuring bootstrap from 
a compiler with a different C++ ABI and runtime works) is not about what 
builds are possible but about what builds are not made more complicated 
than necessary.

By now we can probably remove the workaround

    This paragraph is here to try to keep Sun CC from dying.
    The number of chars here seems crucial!!!!  */

in gcc.c though.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] Cross-compilers built with GCC 3.4 do not work

[nightstrike at gmail dot com]

------- Comment #6 from nightstrike at gmail dot com  2008-08-13 14:23 -------
Is 2.95 still the minimum gcc required for the build?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[pinskia at gcc dot gnu dot org]

------- Comment #7 from pinskia at gcc dot gnu dot org  2008-08-16 22:20 -------
Note I agree with Richard here.  All my builds for cross compilers (except for
testing the trunk for spu-elf which I am going to change soon anyways), I
bootstrap first the compiler I am going to compile and then use that to compile
the cross.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |build
            Summary|Cross-compilers built with  |[4.4 Regression] GCC 3.4
                   |GCC 3.4 do not work         |miscompiles trunk (for cross
                   |                            |compiling)
   Target Milestone|---                         |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[nightstrike at gmail dot com]

------- Comment #8 from nightstrike at gmail dot com  2008-08-22 06:47 -------
I can confirm this bug (seeing as how the one I wrote got duped to here).  Can
someone update the status to confirmed?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[rguenth at gcc dot gnu dot org]

------- Comment #9 from rguenth at gcc dot gnu dot org  2008-08-22 15:20 -------
This is not a gcc bug.


-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[nightstrike at gmail dot com]

------- Comment #10 from nightstrike at gmail dot com  2008-08-22 17:41 -------
How can you close this if the tuples merge makes it impossible to build gcc
with version 3.4?  The minimum gcc is still 2.95.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[joseph at codesourcery dot com]

------- Comment #11 from joseph at codesourcery dot com  2008-08-27 17:59 -------
Subject: Re:  [4.4 Regression] GCC 3.4 miscompiles trunk
 (for cross compiling)

On Fri, 22 Aug 2008, nightstrike at gmail dot com wrote:

> How can you close this if the tuples merge makes it impossible to build gcc
> with version 3.4?  The minimum gcc is still 2.95.

You can always submit a workaround patch to gcc-patches.  The obvious 
first thing to try would be making the loop index volatile if building 
with GCC <= 3.4.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[rguenth at gcc dot gnu dot org]

------- Comment #12 from rguenth at gcc dot gnu dot org  2008-08-27 19:21 -------
I can close this because this is a bug in GCC 3.4 which is out of maintainance
since years and not a bug in the mainline GCC.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[jsm28 at gcc dot gnu dot org]

------- Comment #13 from jsm28 at gcc dot gnu dot org  2008-08-29 23:31 -------
Subject: Bug 37086

Author: jsm28
Date: Fri Aug 29 23:30:18 2008
New Revision: 139792

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=139792
Log:
        PR bootstrap/37086
        * tree-vrp.c (find_switch_asserts): Make idx volatile for GCC
        versions before 4.0.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-vrp.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[nightstrike at gmail dot com]

------- Comment #14 from nightstrike at gmail dot com  2008-09-12 05:47 -------
I just wanted to confirm that Joesph's patch does indeed work (thank you!), and
ask that the status of this PR change from INVALID to FIXED.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[drow at gcc dot gnu dot org]

------- Comment #15 from drow at gcc dot gnu dot org  2008-09-12 12:43 -------
Reopening...


-- 

drow at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086

[Bug bootstrap/37086] [4.4 Regression] GCC 3.4 miscompiles trunk (for cross compiling)

[drow at gcc dot gnu dot org]

------- Comment #16 from drow at gcc dot gnu dot org  2008-09-12 12:43 -------
Fixed.


-- 

drow at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37086