https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125173
Stefan Schulze Frielinghaus <stefansf at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Last reconfirmed| |2026-05-04
CC| |stefansf at gcc dot gnu.org
Status|UNCONFIRMED |NEW
--- Comment #1 from Stefan Schulze Frielinghaus <stefansf at gcc dot gnu.org>
---
Confirmed. This is target independent and also reproducible on e.g. s390.
Seems to be the case since the assert was added (at least reproducible for me
with gcc 8/9/10/11/12).
10: r70:DI=`*.LANCHOR0'
REG_EQUIV `*.LANCHOR0'
11:
{[`*.LANCHOR0']=asm_operands;[const(`*.LANCHOR0'+0x8)]=asm_operands;[const(`*.LANCHOR0'+0xc)]=asm_operands;}
REG_DEAD r70:DI
We fail while trying to replace r70 with its equiv in insn 11. In each
asm_operands r70 is referred to. Therefore, while walking over all uses via
rtx_insn *use_insn = 0;
for (df_ref use = DF_REG_USE_CHAIN (regno);
use;
use = DF_REF_NEXT_REG (use))
if (DF_REF_INSN_INFO (use))
{
if (DEBUG_INSN_P (DF_REF_INSN (use)))
continue;
gcc_assert (!use_insn);
use_insn = DF_REF_INSN (use);
}
we see insn 11 multiple times which is why the assert triggers. I think the
assert could be relaxed since what we want to ensure here is that a register is
not referred to by multiple/different insns since otherwise we cannot trivially
move the definition. So maybe something like the attached patch which is
completely untested.
