[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

Richard Biener  changed:

   What|Removed |Added

   Target Milestone|6.4 |6.5

--- Comment #12 from Richard Biener  ---
GCC 6.4 is being released, adjusting target milestone.

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

Jakub Jelinek  changed:

   What|Removed |Added

   Target Milestone|6.3 |6.4

--- Comment #11 from Jakub Jelinek  ---
GCC 6.3 is being released, adjusting target milestone.

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

--- Comment #10 from Jakub Jelinek  ---
Author: jakub
Date: Fri Oct 14 19:36:58 2016
New Revision: 241182

URL: https://gcc.gnu.org/viewcvs?rev=241182=gcc=rev
Log:
PR middle-end/77959
* expr.c (expand_expr_real_1) : For EXPAND_WRITE
return a MEM.

* gfortran.dg/pr77959.f90: New test.

Added:
trunk/gcc/testsuite/gfortran.dg/pr77959.f90
Modified:
trunk/gcc/ChangeLog
trunk/gcc/expr.c
trunk/gcc/testsuite/ChangeLog

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

Jakub Jelinek  changed:

   What|Removed |Added

 Status|NEW |ASSIGNED
   Assignee|unassigned at gcc dot gnu.org  |jakub at gcc dot gnu.org

--- Comment #9 from Jakub Jelinek  ---
Created attachment 39802
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=39802=edit
gcc7-pr77959.patch

Untested fix.

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

Jakub Jelinek  changed:

   What|Removed |Added

 CC||jakub at gcc dot gnu.org

--- Comment #8 from Jakub Jelinek  ---
expand_assignment is called with
REALPART_EXPR 
as to, where C.3418 is CONST_DECL.
Of course one can't store into a CONST_DECL.
If I compare this to how we compile:
program p
   interface
  subroutine s(x)
 real :: x
  end
   end interface
   call s(1.0)
end
subroutine s(x)
   real :: x
   x = x + 1
end
then the difference is that
to_rtx = expand_expr (to, NULL_RTX, VOIDmode, EXPAND_WRITE);
in that case returns
(mem:SF (reg/f:DI 90) [1 C.3418+0 S4 A32])
and not the CONST_DOUBLE.  I'll have a look.

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

--- Comment #7 from Uroš Bizjak  ---
(In reply to Andrew Pinski from comment #6)
> Maybe turning on RTL checking will find the problem is located.

Indeed, the rtl-checking enabled build says:

internal compiler error: RTL check: expected code 'mem', have 'const_double' in
get_mem_attrs, at rtl.h:3300
0xa8b327 rtl_check_failed_code1(rtx_def const*, rtx_code, char const*, int,
char const*)
../../git/gcc/gcc/rtl.c:811
0x8345d8 get_mem_attrs
../../git/gcc/gcc/rtl.h:3300
0x83038b get_mem_attrs
../../git/gcc/gcc/expr.c:6950
0x83038b store_field
../../git/gcc/gcc/expr.c:6778
0x82d2e3 expand_assignment(tree_node*, tree_node*, bool)
../../git/gcc/gcc/expr.c:5167
0x7253bd expand_gimple_stmt_1
../../git/gcc/gcc/cfgexpand.c:3649
0x7253bd expand_gimple_stmt
../../git/gcc/gcc/cfgexpand.c:3745

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

--- Comment #6 from Andrew Pinski  ---
(In reply to Uroš Bizjak from comment #4)
> Looks like middle-end problem to me:

Maybe turning on RTL checking will find the problem is located.

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

Uroš Bizjak  changed:

   What|Removed |Added

 Target|x86_64-*-*  |x86_64-*-*, alphaev68-*-*

--- Comment #5 from Uroš Bizjak  ---
Also fails for alphaev68-linux-gnu:

internal compiler error: Segmentation fault
0x120948993 crash_signal
/space/homedirs/uros/gcc-svn/trunk/gcc/toplev.c:337
0x120d359b0 alpha_legitimate_address_p
/space/homedirs/uros/gcc-svn/trunk/gcc/config/alpha/alpha.c:857
0x1209439eb default_addr_space_legitimate_address_p(machine_mode, rtx_def*,
bool, unsigned char)
/space/homedirs/uros/gcc-svn/trunk/gcc/targhooks.c:1344
0x12087f613 memory_address_addr_space_p(machine_mode, rtx_def*, unsigned char)
/space/homedirs/uros/gcc-svn/trunk/gcc/recog.c:1336
0x120500963 adjust_address_1(rtx_def*, machine_mode, long, int, int, int, long)
/space/homedirs/uros/gcc-svn/trunk/gcc/emit-rtl.c:2221
0x120552b07 store_field
/space/homedirs/uros/gcc-svn/trunk/gcc/expr.c:6938
0x12054de3b expand_assignment(tree_node*, tree_node*, bool)
/space/homedirs/uros/gcc-svn/trunk/gcc/expr.c:5167
0x1203f251b expand_gimple_stmt_1
/space/homedirs/uros/gcc-svn/trunk/gcc/cfgexpand.c:3649
0x1203f251b expand_gimple_stmt
/space/homedirs/uros/gcc-svn/trunk/gcc/cfgexpand.c:3745
0x1203f908b expand_gimple_basic_block
/space/homedirs/uros/gcc-svn/trunk/gcc/cfgexpand.c:5752
0x1203fc2db execute
/space/homedirs/uros/gcc-svn/trunk/gcc/cfgexpand.c:6363

[Bug middle-end/77959] ICE in ix86_decompose_address, at i386/i386.c:14954

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77959

Uroš Bizjak  changed:

   What|Removed |Added

  Component|target  |middle-end

--- Comment #4 from Uroš Bizjak  ---
Looks like middle-end problem to me:

Program received signal SIGSEGV, Segmentation fault.
0x00ef2ddb in ix86_decompose_address (addr=0x41,
out=out@entry=0x7fffd490) at
/home/uros/gcc-svn/trunk/gcc/config/i386/i386.c:14954
14954 if (TARGET_64BIT && GET_MODE (addr) == DImode)
(gdb) p addr
$10 = (rtx) 0x41

(gdb) bt
#0  0x00ef2ddb in ix86_decompose_address (addr=0x41,
out=out@entry=0x7fffd490) at
/home/uros/gcc-svn/trunk/gcc/config/i386/i386.c:14954
#1  0x00ef4ddc in ix86_legitimate_address_p (addr=,
strict=) at
/home/uros/gcc-svn/trunk/gcc/config/i386/i386.c:15668
#2  0x00b1f4af in memory_address_addr_space_p (mode=mode@entry=SFmode,
addr=addr@entry=0x41, as=)
at /home/uros/gcc-svn/trunk/gcc/recog.c:1336
#3  0x00874930 in adjust_address_1 (memref=memref@entry=0x7fffefdec168,
mode=mode@entry=SFmode, offset=offset@entry=0, validate=validate@entry=1, 
adjust_address=adjust_address@entry=1, adjust_object=adjust_object@entry=0,
size=4) at /home/uros/gcc-svn/trunk/gcc/emit-rtl.c:2221
#4  0x008b3d1c in store_field (target=0x7fffefdec168, bitsize=32,
bitpos=0, bitregion_start=, bitregion_end=,
mode=SFmode, 
exp=0x7fffeffd8c48, alias_set=2, nontemporal=false, reverse=false) at
/home/uros/gcc-svn/trunk/gcc/expr.c:6938
#5  0x008afd2f in expand_assignment (to=to@entry=0x7fffeffc4a60,
from=from@entry=0x7fffeffd8c48, nontemporal=)
at /home/uros/gcc-svn/trunk/gcc/expr.c:5167
#6  0x0079c6ee in expand_gimple_stmt_1 (stmt=0x7fffeffc88c0) at
/home/uros/gcc-svn/trunk/gcc/cfgexpand.c:3649

(gdb) f 3
#3  0x00874930 in adjust_address_1 (memref=memref@entry=0x7fffefdec168,
mode=mode@entry=SFmode, offset=offset@entry=0, validate=validate@entry=1, 
adjust_address=adjust_address@entry=1, adjust_object=adjust_object@entry=0,
size=4) at /home/uros/gcc-svn/trunk/gcc/emit-rtl.c:2221
2221  && (!validate || memory_address_addr_space_p (mode, addr,
(gdb) list
2216size = defattrs->size;
2217
2218  /* If there are no changes, just return the original memory
reference.  */
2219  if (mode == GET_MODE (memref) && !offset
2220  && (size == 0 || (attrs.size_known_p && attrs.size == size))
2221  && (!validate || memory_address_addr_space_p (mode, addr,
attrs.addrspace)))
2223return memref;
2224
2225  /* ??? Prefer to create garbage instead of creating shared rtl.
(gdb) p addr
$8 = (rtx) 0x41

Corrupted RTX is passed from adjust_address_1.