CRIS defines DATA_ALIGNMENT such that alignment can be
applied differently to different data of the same type, when
"references to it must bind to the current definition"
(varasm.cc:align_variable). Here, it means that more
alignment is then applied to g, but not f, so the test-case
fails because another message is emitted than the expected:
a same-alignment test dominates the not-discardable test,
and we get "Not unifying; original and alias have
incompatible alignments" rather than "Not unifying; alias
cannot be created; target is discardable". Because this
DATA_ALIGNMENT behavior for CRIS depends on target options,
and this test is already artificial by the use of -fcommon,
better skip it.
* gcc.dg/ipa/pr77653.c: Skip for cris-*-*.
---
gcc/testsuite/gcc.dg/ipa/pr77653.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/gcc/testsuite/gcc.dg/ipa/pr77653.c
b/gcc/testsuite/gcc.dg/ipa/pr77653.c
index 2fddb7eab548..16df3fff6a41 100644
--- a/gcc/testsuite/gcc.dg/ipa/pr77653.c
+++ b/gcc/testsuite/gcc.dg/ipa/pr77653.c
@@ -1,5 +1,6 @@
/* { dg-require-alias "" } */
/* { dg-options "-O2 -fcommon -fdump-ipa-icf-details" } */
+/* { dg-skip-if "Can align g more than f" { cris-*-* } } */
int a, b, c, d, e, h, i, j, k, l;
const int f;
--
2.30.2
