[Bug ada/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[Bug ada/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-09-25 17:22:21 UTC ---
The workaround is to remove -g from the command line.

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|ada                         |rtl-optimization

--- Comment #6 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-09-25 17:46:49 UTC ---
The combiner swaps 2 insns.  Before:

(debug_insn 227 226 228 42 (var_location:SI D.1502 (reg:SI 85 [ D.1501 ]))
p.adb:21 -1
     (nil))

(insn 228 227 229 42 (set (mem/s/j:SI (reg/f:SI 84 [ D.1519 ]) [7
D.1519_197->BOUNDS.LB0+0 S4 A32])
        (reg:SI 85 [ D.1501 ])) p.adb:21 67 {*movsi_esa}
     (expr_list:REG_DEAD (reg:SI 85 [ D.1501 ])
        (expr_list:REG_EH_REGION (const_int 1 [0x1])
            (nil))))

After:

(insn 228 226 227 42 (parallel [
            (set (mem/s/j:SI (reg/f:SI 84 [ D.1519 ]) [7
D.1519_197->BOUNDS.LB0+0 S4 A32])
                (plus:SI (reg/v:SI 63 [ p__current_parameter__B_1__TTpSP1___L
])
                    (const_int 1 [0x1])))
            (clobber (reg:CC 33 %cc))
        ]) p.adb:21 250 {*addsi3}
     (expr_list:REG_UNUSED (reg:CC 33 %cc)
        (expr_list:REG_DEAD (reg/v:SI 63 [
p__current_parameter__B_1__TTpSP1___L ])
            (expr_list:REG_EH_REGION (const_int 1 [0x1])
                (nil)))))

(debug_insn 227 228 229 42 (var_location:SI D.1502 (mem/s/j:SI (reg/f:SI 84 [
D.1519 ]) [7 D.1519_197->BOUNDS.LB0+0 S4 A32])) p.adb:21 -1
     (nil))

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
                 CC|                            |aoliva at gcc dot gnu.org

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-09-30 13:00:38 UTC ---
For the records, I have a reduced testcase and a patch.  ETA is next week.

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-06 14:25:11 UTC ---
Author: ebotcazou
Date: Wed Oct  6 14:25:04 2010
New Revision: 165030

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=165030
Log:
    PR rtl-optimization/45394
    * combine.c (cleanup_auto_inc_dec): Remove AFTER parameter and adjust.
    (struct rtx_subst_pair): Remove AFTER field.
    (propagate_for_debug_subst): Adjust call to cleanup_auto_inc_dec.
    (propagate_for_debug): Remove MOVE parameter and adjust.
    (try_combine): In the special case where I2 is a PARALLEL, combine only
    destinations of SETs and set I2SRC to the I2 source.
    Adjust calls to propagate_for_debug throughout.
    * reload1.c (eliminate_regs_1) <ASM_OPERANDS>: Do not abort for debug
    insns.

Added:
    trunk/gcc/testsuite/gnat.dg/opt6.adb
    trunk/gcc/testsuite/gnat.dg/opt6.ads
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/combine.c
    trunk/gcc/reload1.c
    trunk/gcc/testsuite/ChangeLog

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2010-10/msg00457.htm
                   |                            |l
         Resolution|                            |FIXED

--- Comment #9 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-06 14:29:36 UTC ---
Presumably.

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

--- Comment #10 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2010-10-07 06:15:15 UTC ---
(In reply to comment #6)
> The combiner swaps 2 insns.

This swap looks correct to me.  Presumably reg 85 is being dropped, so moving
the debug_insn down to a point at which it can refer to the used value should
be fine, no?  Is this move even related with the bug?  Why?

I'm trying to figure out why this change that drops “after”, presumably needed
for this very kind of situation, is correct, but I can't easily get an
environment in which I can duplicate the problem.  A cross compiler (without
target libs) doesn't seem to be enough (or at least I don't know how to run it
to trigger the problem).  Help?

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

--- Comment #11 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2010-10-07 06:18:21 UTC ---
Aah, I think I got it: the MEM reference may throw, right?  Well, if i2src is
still available, I guess keeping the debug insn where it is, suitably adjusted,
is even better.  Did you check that it still makes sense after combine?

[Bug rtl-optimization/45394] [4.6 regression] gnat fails to build on s390, trunk 20100918

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45394

--- Comment #12 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-07 11:49:07 UTC ---
> Aah, I think I got it: the MEM reference may throw, right?  Well, if i2src is
> still available, I guess keeping the debug insn where it is, suitably adjusted,
> is even better.  Did you check that it still makes sense after combine?

Yes, the debug insn is adjusted like in all the other cases.