> On 21 Jul 2026, at 16:05, Richard Biener <[email protected]> wrote:
> 
> On Tue, 21 Jul 2026, [email protected] wrote:
> 
>> From: Kyrylo Tkachov <[email protected]>
>> 
>> create_tmp_var_raw marks a replacement declaration as ignored.  At -O0,
>> this can change use_register_for_decl from memory for the original user
>> declaration to a pseudo for the replacement.  A partition can already own
>> memory from a parameter default definition, which then makes set_rtl reject
>> that memory for another name in the partition.
>> 
>> If the replacement would use a register, copy DECL_IGNORED_P so it keeps the
>> storage choice made during coalescing.  Keep declarations that already use
>> memory unchanged.  Add a checking assertion for the memory invariant.
>> 
>> Bootstrapped and tested on aarch64-none-linux-gnu and x86_64-linux.
>> 
>> Ok for trunk?
>> Thanks,
>> Kyrill
>> 
>> gcc/ChangeLog:
>> 
>> PR middle-end/126341
>> * tree-outof-ssa.cc (split_overlapping_partition_decls): Preserve
>> DECL_IGNORED_P.  Check that the replacement uses memory.
>> 
>> gcc/testsuite/ChangeLog:
>> 
>> PR middle-end/126341
>> * gcc.dg/pr126341.c: New test.
>> 
>> Signed-off-by: Kyrylo Tkachov <[email protected]>
>> ---
>> gcc/testsuite/gcc.dg/pr126341.c | 53 +++++++++++++++++++++++++++++++++
>> gcc/tree-outof-ssa.cc           |  3 ++
>> 2 files changed, 56 insertions(+)
>> create mode 100644 gcc/testsuite/gcc.dg/pr126341.c
>> 
>> diff --git a/gcc/testsuite/gcc.dg/pr126341.c 
>> b/gcc/testsuite/gcc.dg/pr126341.c
>> new file mode 100644
>> index 00000000000..76ed1d889e4
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.dg/pr126341.c
>> @@ -0,0 +1,53 @@
>> +/* { dg-do compile { target bitint575 } } */
>> +/* { dg-options "-fnon-call-exceptions -ftree-coalesce-vars" } */
>> +/* { dg-prune-output "division by zero" } */
>> +
>> +/* PR middle-end/126341 */
>> +
>> +#define N 0x200000000uwb
>> +#define DIV0 (1 / 0)
>> +
>> +struct {
>> +  char j[8];
>> +} k;
>> +char l;
>> +int m, p, q, r;
>> +long n;
>> +_Complex o;
>> +int s;
>> +_BitInt (256) aa;
>> +
>> +void
>> +foo (_BitInt (8) t)
>> +{
>> +  int ah, ai, aj = 1, ak, al = 1, am = 1, an = 1, ao = 1, ap = 1, aq = DIV0;
>> +  int ar = 1, as = 1, at, au = DIV0;
>> +  int av = 1;
>> +  _BitInt (7) c = ak--;
>> +  char aw = 1, ax = 1, ay = 1, az = 1, a = 1, d = 1, e = 1;
>> +  int bfbg = 1, bh = 1, bi = p += ai -= __sync_add_and_fetch (&s, 1);
>> +  int bj;
>> +  int bk, blbm = 1, bt = 1, bu = 1, bv = __atomic_or_fetch (&aa, 1, 1);
>> +  int bn = 1, bobp = 1, br = 1, bs = 1;
>> +  unsigned b, cc;
>> +  char cf, ce;
>> +
>> +  if (bj)
>> +    goto ca;
>> +  b = 1, cc = 1;
>> +  for (;;);
>> +  ce = q;
>> +  cf = o ? 1
>> +    :
>> +    __builtin_stdc_rotate_left (__builtin_stdc_rotate_left
>> + (c - 0x3bfuwb, br), n & 1) ? 3 : 3;
>> +  1 +
>> +    az & 1 | N << -az & m | ((az & 1 | N) << -az & 1 ? 0 : 7u) & 1 |
>> +    ((az & 1 | N) - az & m | az & 1 | N) < 7 > -c & 1;
>> +ca:
>> +  s %= __builtin_stdc_rotate_left (cc, ah);
>> +  at /= r ^= ({ __auto_type cg = 0; __auto_type ch = 0; 0; });
>> +
>> +  long ci = bk % cf;
>> +  _BitInt (8) cj = l ? t : k.j[3];
>> +}
>> diff --git a/gcc/tree-outof-ssa.cc b/gcc/tree-outof-ssa.cc
>> index 918dbbedde3..8a32dab42a5 100644
>> --- a/gcc/tree-outof-ssa.cc
>> +++ b/gcc/tree-outof-ssa.cc
>> @@ -1101,6 +1101,9 @@ split_overlapping_partition_decls (var_map map)
>> continue;
>> 
>>       tree nvar = create_tmp_var_raw (TREE_TYPE (var));
>> +      if (use_register_for_decl (nvar))
> 
> I think this warrants a comment in the code.
> 
> OK with that change.

Thanks, I’ve pushed the fix as g:7459f9136500bfb881902834d56fb110b77909f2 .
I propose to back port it together with the previous out-of-ssa wrong-code fix 
for 16.2 by the end of the week if no fallout occurs?

Kyrill


> Richard.
> 
>> + DECL_IGNORED_P (nvar) = DECL_IGNORED_P (var);
>> +      gcc_checking_assert (!use_register_for_decl (nvar));
>>       DECL_CONTEXT (nvar) = DECL_CONTEXT (var);
>>       DECL_SOURCE_LOCATION (nvar) = DECL_SOURCE_LOCATION (var);
>>       SET_DECL_ALIGN (nvar, DECL_ALIGN (var));
>> 
> 
> -- 
> Richard Biener <[email protected]>
> SUSE Software Solutions Germany GmbH,
> Frankenstrasse 146, 90461 Nuernberg, Germany;
> GF: Jochen Jaser, Andrew McDonald, Abhinav Puri; (HRB 36809, AG Nuernberg)

Reply via email to