[Bug middle-end/26996] New: interpret_rhs_modify_expr calls fold_convert (vector_type, -1)
rguenth at gcc dot gnu dot org
gcc-bugzilla@gcc.gnu.org
Mon Apr 3 07:49:00 GMT 2006
We ICE on the air Polyhedron test with -march=pentium4 -ffast-math
-funroll-loops -O3 -ftree-vectorize:
/space/rguenther/tramp3d/pb05/lin/source/air.f90: In function ÂboundÂ:
/space/rguenther/tramp3d/pb05/lin/source/air.f90:1119: internal compiler error:
in fold_convert, at fold-const.c:2089
#1 0x082ab8ca in fold_convert (type=0xb797d730, arg=0xb7cae7e0)
at /space/rguenther/src/svn/gcc/gcc/fold-const.c:2089
2089 gcc_assert (tree_int_cst_equal (TYPE_SIZE (type), TYPE_SIZE
(orig)));
(gdb) call debug_tree(arg)
<integer_cst 0xb7cae7e0 type <integer_type 0xb7cbd284 int4> constant invariant
-1>
(gdb) call debug_tree(type)
<vector_type 0xb797d730
type <real_type 0xb7cbd9b4 real8 DF
size <integer_cst 0xb7cae510 constant invariant 64>
unit size <integer_cst 0xb7cae528 constant invariant 8>
align 64 symtab 0 alias set 4 precision 64
pointer_to_this <pointer_type 0xb7cbdac8>>
V2DF
size <integer_cst 0xb7cae648 type <integer_type 0xb7cbd05c bit_size_type>
constant invariant 128>
unit size <integer_cst 0xb7cae660 type <integer_type 0xb7cbd000> constant
invariant 16>
align 128 symtab 0 alias set -1 nunits 2>
(gdb) up
#2 0x085ec360 in interpret_rhs_modify_expr (loop=0x8a4e110,
at_stmt=0xb7983798, opnd1=0xb797eb20, type=0xb797d730)
at /space/rguenther/src/svn/gcc/gcc/tree-scalar-evolution.c:1659
1658 /* TYPE may be integer, real or complex, so use fold_convert. */
1659 res = chrec_fold_multiply (type, chrec10,
1660 fold_convert (type,
integer_minus_one_node));
But we can only build a zero vector in fold_convert.
--
Summary: interpret_rhs_modify_expr calls fold_convert
(vector_type, -1)
Product: gcc
Version: 4.2.0
Status: UNCONFIRMED
Keywords: ice-on-valid-code
Severity: normal
Priority: P3
Component: middle-end
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: rguenth at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26996
More information about the Gcc-bugs
mailing list