On Thu, Apr 09, 2020 at 10:57:46PM +0200, Christophe Lyon via Gcc-patches wrote:
> This patch makes GCC fail to build newlib when configured for
> aarch64_be-none-elf:
> 0x10c488c vt_expand_var_loc_chain
>
> /tmp/9192639_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/var-tracking.c:8355
Seems it is the cselib_record_sp_cfa_base_equiv part of the patch (reverting
the add_stores change doesn't make the ICE go away).
The following fixes it, but don't have cycles to properly test it right now:
2020-04-10 Jakub Jelinek <ja...@redhat.com>
PR debug/94495
* cselib.c (cselib_record_sp_cfa_base_equiv): Set PRESERVED_VALUE_P on
val->val_rtx.
--- gcc/cselib.c.jj 2020-04-09 21:20:59.457529812 +0200
+++ gcc/cselib.c 2020-04-10 11:49:45.820124893 +0200
@@ -2695,7 +2695,10 @@ cselib_record_sp_cfa_base_equiv (HOST_WI
= cselib_lookup_from_insn (plus_constant (Pmode, sp_derived_value, offset),
Pmode, 1, VOIDmode, insn);
if (val != NULL)
- cselib_record_set (stack_pointer_rtx, val, NULL);
+ {
+ PRESERVED_VALUE_P (val->val_rtx) = 1;
+ cselib_record_set (stack_pointer_rtx, val, NULL);
+ }
}
/* Return true if V is SP_DERIVED_VALUE_P (or SP_DERIVED_VALUE_P + CONST_INT)
Jakub
