[Bug target/89400] [7/8/9 Regression] ICE: output_operand: invalid %-code with -march=armv6kz -mthumb -munaligned-access

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89400

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ramana at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
All the "unaligned_access" instructions use %? in the patterns (except the last
one) where %? is TARGET_32BIT specific.
If the unaligned loads/stores don't really work for thumb1, then

  if (! opts_set->x_unaligned_access)
    {
      opts->x_unaligned_access = (TARGET_32BIT_P (opts->x_target_flags)
                          && arm_arch6 && (arm_arch_notm || arm_arch7));
    }
  else if (opts->x_unaligned_access == 1
           && !(arm_arch6 && (arm_arch_notm || arm_arch7)))
    {
      warning (0, "target CPU does not support unaligned accesses");
     opts->x_unaligned_access = 0;
    }
should be changed to:
  else if (opts->x_unaligned_access == 1
           && (!(arm_arch6 && (arm_arch_notm || arm_arch7))
               || !TARGET_32BIT_P (opts->x_target_flags)))
If the unaligned_access instructions are also for thumb1, then perhaps they
need to set predicable to no for !TARGET_32BIT e.g. based on isa attribute and
avoid using %? in the thumb1 alternative patterns.