Well, not sure if it is the right way to go, but I tried creating
SSA_NAMEs for the function's arguments and use them in the later
assignments rather than the parm_decls themselves.
Now an assert triggers at out-of-ssa phase, at ssaexpand.h:59, in
get_rtx_for_ssa_name().
My code changed like this:
-------------
tree var, copy_par;
...
g = gimple_build_nop();
var = make_ssa_name(p, g);
gimple_seq_add_stmt_without_update (&seq, g);
copy_par = create_tmp_var(TREE_TYPE(p), NULL);
add_referenced_var(copy_par);
copy_par = make_ssa_name(copy_par, NULL);
g = gimple_build_assign_with_ops(
NOP_EXPR, copy_par, var, NULL_TREE);
<the rest is the same as before>
-------------
Because there are no real statements to define these SSA_NAMEs, I added
GIMPLE_NOPs for that purpose.
The assert triggers at ssaexpand.h:59, in get_rtx_for_ssa_name():
(gdb) l
53 static inline rtx
54 get_rtx_for_ssa_name (tree exp)
55 {
56 int p = partition_find (SA.map->var_partition, SSA_NAME_VERSION (exp));
57 if (SA.map->partition_to_view)
58 p = SA.map->partition_to_view[p];
59 gcc_assert (p != NO_PARTITION);
60 return SA.partition_to_pseudo[p];
61 }
partition_find() returns 19 at line 56 but SA.map->partition_to_view[p]
is -1 (NO_PARTITION) in this case, and gcc_assert() fires.
The function being processed is rather simple:
------------------
static void
other_func(const char *p, int x, unsigned long u, unsigned char s,
unsigned long other[])
{
printf("x = %d, p = %s, u = %lu, args: %lx, %lx\n",
x, p, u, other[0], other[1]);
printf("%lu\n", *(unsigned long *)&s);
}
------------------
When debugging this with GDB, I can see that GCC first processes the
SSA_NAMEs for 'p', 'x' and 'u' - no crash, SA.map->partition_to_view[p]
is not -1 for these. After that, 's' is processed and this is when the
problem happens.
Here is what debug_tree() shows for the SSA_NAME for 's' at that point,
if it helps:
------------------
<ssa_name 0xb6fd27bc
type <integer_type 0xb6f621e0 unsigned char public unsigned
string-flag QI
size <integer_cst 0xb6f5071c constant 8>
unit size <integer_cst 0xb6f50738 constant 1>
align 8 symtab -1224807280 alias set -1 canonical type
0xb6f621e0 precision 8 min <integer_cst 0xb6f50754 0> max <integer_cst
0xb6f506e4 255> context <translation_unit_decl 0xb7002d80 D.1837>
pointer_to_this <pointer_type 0xb7020f60>>
var <parm_decl 0xb701cec4 s>def_stmt GIMPLE_NOP
version 19>
------------------
Any ideas what could possibly be wrong here?
Regards,
Eugene
--
Eugene Shatokhin, ROSA
www.rosalab.com
