This patch sits in my trees for quite some years and I always forget
to push it - it usually gets triggered by weird targets (PSImode
pointers/sizetype) which run into GIMPLE IL checking asserts for
pointer -> integer conversions and the "sizetype" special-case
not triggering. That special-case should not exist but instead
it should be something like the patch below - the whole point
of the IL consistecy check is to ensure we know how to extend
from pointer to integer.
I've bootstrapped / tested the patch on x86_64 where it is a no-op.
But from the looks, can you punch any holes in it? (it was able
to resolve reporters issues at least).
Thanks,
Richard.
-
The existing check doesn't reflect the actual reason why it exists,
the patch makes us to use POINTERS_EXTEND_UNSIGNED instead which
is specified for ptr_mode and word_mode/Pmode precision.
* tree-cfg.c (verify_gimple_assign_unary): Adjust integer
to/from pointer conversion checking.
---
gcc/tree-cfg.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/gcc/tree-cfg.c b/gcc/tree-cfg.c
index e99fb9ff5d1..c6140af50e7 100644
--- a/gcc/tree-cfg.c
+++ b/gcc/tree-cfg.c
@@ -3591,13 +3591,21 @@ verify_gimple_assign_unary (gassign *stmt)
/* Allow conversions from pointer type to integral type only if
there is no sign or zero extension involved.
For targets were the precision of ptrofftype doesn't match that
- of pointers we need to allow arbitrary conversions to ptrofftype. */
+ of pointers we allow conversions to types where
+ POINTERS_EXTEND_UNSIGNED specifies how that works. */
if ((POINTER_TYPE_P (lhs_type)
&& INTEGRAL_TYPE_P (rhs1_type))
|| (POINTER_TYPE_P (rhs1_type)
&& INTEGRAL_TYPE_P (lhs_type)
&& (TYPE_PRECISION (rhs1_type) >= TYPE_PRECISION (lhs_type)
- || ptrofftype_p (lhs_type))))
+#if defined(POINTERS_EXTEND_UNSIGNED)
+ || (TYPE_MODE (rhs1_type) == ptr_mode
+ && (TYPE_PRECISION (lhs_type)
+ == BITS_PER_WORD /* word_mode */
+ || (TYPE_PRECISION (lhs_type)
+ == GET_MODE_PRECISION (Pmode))))
+#endif
+ )))
return false;
/* Allow conversion from integral to offset type and vice versa. */
--
2.16.4
