So far atomic objects are aligned according to their default alignment.
For 128 bit scalar types like int128 or long double this results in an
8 byte alignment which is wrong and must be 16 byte.
libstdc++ already computes a correct alignment, though, still adding a
test case in order to make sure that both implementations are
compatible.
Bootstrapped and regtested. Ok for mainline? Since this is an ABI
break, is a backport to GCC 13 reasonable?
gcc/ChangeLog:
* config/s390/s390.cc (TARGET_ATOMIC_ALIGN_FOR_MODE):
New.
(s390_atomic_align_for_mode): New.
gcc/testsuite/ChangeLog:
* g++.target/s390/atomic-align-1.C: New test.
* gcc.target/s390/atomic-align-1.c: New test.
* gcc.target/s390/atomic-align-2.c: New test.
---
gcc/config/s390/s390.cc | 8 ++++++
.../g++.target/s390/atomic-align-1.C | 25 +++++++++++++++++++
.../gcc.target/s390/atomic-align-1.c | 23 +++++++++++++++++
.../gcc.target/s390/atomic-align-2.c | 18 +++++++++++++
4 files changed, 74 insertions(+)
create mode 100644 gcc/testsuite/g++.target/s390/atomic-align-1.C
create mode 100644 gcc/testsuite/gcc.target/s390/atomic-align-1.c
create mode 100644 gcc/testsuite/gcc.target/s390/atomic-align-2.c
diff --git a/gcc/config/s390/s390.cc b/gcc/config/s390/s390.cc
index 505de995da8..4813bf91dc4 100644
--- a/gcc/config/s390/s390.cc
+++ b/gcc/config/s390/s390.cc
@@ -450,6 +450,14 @@ s390_preserve_fpr_arg_p (int regno)
&& regno >= FPR0_REGNUM);
}
+#undef TARGET_ATOMIC_ALIGN_FOR_MODE
+#define TARGET_ATOMIC_ALIGN_FOR_MODE s390_atomic_align_for_mode
+static unsigned int
+s390_atomic_align_for_mode (machine_mode mode)
+{
+ return GET_MODE_BITSIZE (mode);
+}
+
/* A couple of shortcuts. */
#define CONST_OK_FOR_J(x) \
CONST_OK_FOR_CONSTRAINT_P((x), 'J', "J")
diff --git a/gcc/testsuite/g++.target/s390/atomic-align-1.C
b/gcc/testsuite/g++.target/s390/atomic-align-1.C
new file mode 100644
index 00000000000..43aa0bc39ed
--- /dev/null
+++ b/gcc/testsuite/g++.target/s390/atomic-align-1.C
@@ -0,0 +1,25 @@
+/* { dg-do compile { target int128 } } */
+/* { dg-options "-std=c++11" } */
+/* { dg-final { scan-assembler-times {\.align\t2} 2 } } */
+/* { dg-final { scan-assembler-times {\.align\t4} 2 } } */
+/* { dg-final { scan-assembler-times {\.align\t8} 3 } } */
+/* { dg-final { scan-assembler-times {\.align\t16} 2 } } */
+
+#include <atomic>
+
+// 2
+std::atomic<char> var_char;
+std::atomic<short> var_short;
+// 4
+std::atomic<int> var_int;
+// 8
+std::atomic<long> var_long;
+std::atomic<long long> var_long_long;
+// 16
+std::atomic<__int128> var_int128;
+// 4
+std::atomic<float> var_float;
+// 8
+std::atomic<double> var_double;
+// 16
+std::atomic<long double> var_long_double;
diff --git a/gcc/testsuite/gcc.target/s390/atomic-align-1.c
b/gcc/testsuite/gcc.target/s390/atomic-align-1.c
new file mode 100644
index 00000000000..b2e1233e3ee
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/atomic-align-1.c
@@ -0,0 +1,23 @@
+/* { dg-do compile { target int128 } } */
+/* { dg-options "-std=c11" } */
+/* { dg-final { scan-assembler-times {\.align\t2} 2 } } */
+/* { dg-final { scan-assembler-times {\.align\t4} 2 } } */
+/* { dg-final { scan-assembler-times {\.align\t8} 3 } } */
+/* { dg-final { scan-assembler-times {\.align\t16} 2 } } */
+
+// 2
+_Atomic char var_char;
+_Atomic short var_short;
+// 4
+_Atomic int var_int;
+// 8
+_Atomic long var_long;
+_Atomic long long var_long_long;
+// 16
+_Atomic __int128 var_int128;
+// 4
+_Atomic float var_float;
+// 8
+_Atomic double var_double;
+// 16
+_Atomic long double var_long_double;
diff --git a/gcc/testsuite/gcc.target/s390/atomic-align-2.c
b/gcc/testsuite/gcc.target/s390/atomic-align-2.c
new file mode 100644
index 00000000000..0bf17341bf8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/atomic-align-2.c
@@ -0,0 +1,18 @@
+/* { dg-do compile { target int128 } } */
+/* { dg-options "-O -std=c11" } */
+/* { dg-final { scan-assembler-not {abort} } } */
+
+/* The stack is 8 byte aligned which means GCC has to manually align a 16 byte
+ aligned object. This is done by allocating not 16 but rather 24 bytes for
+ variable X and then manually aligning a pointer inside the memory block.
+ Validate this by ensuring that the if-statement is optimized out. */
+
+void bar (_Atomic unsigned __int128 *ptr);
+
+void foo (void) {
+ _Atomic unsigned __int128 x;
+ unsigned long n = (unsigned long)&x;
+ if (n % 16 != 0)
+ __builtin_abort ();
+ bar (&x);
+}
--
2.39.2
