[Bug middle-end/52472] New: ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[Bug middle-end/52472] New: ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

             Bug #: 52472
           Summary: ICE: in convert_debug_memory_address, at
                    cfgexpand.c:2491
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Keywords: addr-space, ice-on-invalid-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: gjl@gcc.gnu.org
                CC: eric.weddington@atmel.com
              Host: mingw32
            Target: avr


typedef unsigned long uint32_t;
typedef unsigned int uint16_t;

extern const __memx unsigned int data[320][12];

#define pgm_read_word_far(addr)                 \
    (__extension__({                            \
        uint32_t __addr32 = (uint32_t)(addr);   \
        uint16_t __result;                      \
        __asm__("; much code"                   \
            : "=r" (__result)                   \
            : "r" (__addr32));                  \
        __result;                               \
    }))

int ice (void)
{
    return pgm_read_word_far (0x20000 + (unsigned long) &data[0][0]);
}

== Command Line ==

>> avr-gcc ice-memx.c -S -Os -g3 -v

== ICE ==

ice-memx.c: In function 'ice':
ice-memx.c:18:12: warning: cast from pointer to integer of different size
[-Wpointer-to-int-cast]
ice-memx.c:16:5: internal compiler error: in convert_debug_memory_address, at
cfgexpand.c:2491


== Configure ==

Using built-in specs.
Target: avr
Configured with: ../gcc-trunk/configure --with-gmp=/usr/local
--prefix=/usr/local/avr --target=avr --enable-languages=c,c++ --with-dwarf2
--disable-shared --disable-libada --disable-libssp --disable-nls
Thread model: single
gcc version 4.7.0 20120217 (experimental) (GCC) 
GNU C (GCC) version 4.7.0 20120217 (experimental) (avr)
    compiled by GNU C version 4.6.2, GMP version 5.0.4, MPFR version 3.1.0, MPC
version 0.9

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-03-03
   Target Milestone|---                         |4.7.0
     Ever Confirmed|0                           |1

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

--- Comment #1 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-03-03 19:38:56 UTC ---
Created attachment 26820
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26820
ice-memx.c: C source

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org
   Target Milestone|4.7.0                       |---

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-03-05 10:22:42 UTC ---
The !POINTERS_EXTEND_UNSIGNED path seems to be not implemented if it would
require codegen.  I suppose instead we should have a way of saying
"not possible" here?

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

--- Comment #3 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-03-05 10:32:26 UTC ---
I thought about setting POINTERS_EXTEND_UNSIGNED but am unsure about other side
effects of that define.

Presumably, POINTERS_EXTEND_UNSIGNED is undefined because it was never needed
before?

What do you think?

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

--- Comment #4 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-03-05 14:05:36 UTC ---
I tried 

#define POINTERS_EXTEND_UNSIGNED 1

but this runs into ICE some lines later in cfgexpand.c:2496

mode = PSI
as = 0
x = symbol_ref:PSI ("data")

Of course PSImode is not a valid pointer mode for generic address space.

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-06 12:35:09 UTC ---
convert_debug_memory_address can just return NULL_RTX if it can't handle it. 
Can't reproduce this though.

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

--- Comment #6 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-03-07 16:09:12 UTC ---
(In reply to comment #5)
> Can't reproduce this though.

Strange. I still see the ICE with current trunk. It's a 32 bit build and host;
might this have an impact?

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|ice-on-invalid-code         |ice-on-valid-code
   Last reconfirmed|2012-03-03 00:00:00         |2012-03-28 0:00
               Host|mingw32                     |

--- Comment #7 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-03-28 15:27:42 UTC ---
Reduced test-case: Compile with:

$ avr-gcc ice-memx.c -S -g -Os


extern const __memx unsigned data[][10];

unsigned long ice (void)
{
    unsigned long addr32;

    return addr32 = (unsigned long) data;
}

[Bug middle-end/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

Georg-Johann Lay <gjl at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |Sergey.Belyashov at gmail
                   |                            |dot com

--- Comment #8 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2013-04-13 14:33:19 UTC ---
*** Bug 56792 has been marked as a duplicate of this bug. ***

[Bug debug/52472] ICE: in convert_debug_memory_address, at cfgexpand.c:2491

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52472

--- Comment #9 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Author: law
Date: Thu Jun  5 18:34:43 2014
New Revision: 211288

URL: http://gcc.gnu.org/viewcvs?rev=211288&root=gcc&view=rev
Log:
    PR target/52472
    * cfgexpand.c (expand_debug_expr): Use address space of nested
    TREE_TYPE for ADDR_EXPR and MEM_REF.

    PR target/52472
    * gcc.target/avr/pr52472.c: New test.

Added:
    trunk/gcc/testsuite/gcc.target/avr/pr52472.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/cfgexpand.c
    trunk/gcc/testsuite/ChangeLog