/* { dg-do compile } */
/* { dg-options "-O2 -g -m32" } */
union U { unsigned long u; unsigned long long v; };
struct S { long s; long t; };
extern unsigned int foo (unsigned long long *, unsigned int);
void bar (int x, union U y, struct S *z)
{
if ((x & 3) == 2)
{
unsigned long long r = y.v;
unsigned int b = 1000000000L;
unsigned int m;
if ((r >> 32) == 0)
{
m = (unsigned int)r % b;
r = (unsigned int)r / b;
}
else
m = foo (&r, b);
z->t = m;
z->s = r;
}
}
creates following bogus location description for y (which is passed by
invisible reference on this arch).
.byte 0x54 # DW_OP_reg4
.byte 0x93 # DW_OP_piece
.byte 0x4 # uleb128 0x4
.byte 0x74 # DW_OP_breg4
.byte 0x4 # sleb128 4
.byte 0x93 # DW_OP_piece
.byte 0x4 # uleb128 0x4
The 64-bit union is certainly not located partly in reg4 and partly in what
reg4+4 points to, at least not in this case.
The problem seems to be that GCC uses REG_EXPR y for the register that
contains the address of the union:
(reg:SI 4 4 [ y ])
and also (MEM_EXPR) for the MEM that actually contains it, e.g.:
(mem/s:SI (reg:SI 4 4 [ y ]) [6 y+0 S4 A64])
(mem/s:SI (plus:SI (reg:SI 4 4 [ y ]) (const_int 4 [0x4])) [6 y+4 S4 A32])
The REG_EXPR on the register sounds wrong, if we allow REG_EXPR to not contain
only decls, then it could be ADDR_EXPR <y>, otherwise perhaps nothing would
be better than lying about where the argument is.
--
Summary: REG_EXPR wrong for parameters passed by invisible
reference
Product: gcc
Version: 4.0.1
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: debug
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jakub at gcc dot gnu dot org
CC: gcc-bugs at gcc dot gnu dot org,roland at redhat dot com
GCC target triplet: ppc-linux
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21913
