[PATCH] LoongArch: Fix atomic_exchange make comparison and may jump out

[Jinyang He <hejinyang@loongson.cn>]

gcc/ChangeLog:

* config/loongarch/sync.md:
Add atomic_cas_value_exchange_and_7<mode> and fix atomic_exchange<mode>.

gcc/testsuite/ChangeLog:

* gcc.target/loongarch/sync-1.c: New test.
---
 gcc/config/loongarch/sync.md                |  27 ++++-
 gcc/testsuite/gcc.target/loongarch/sync-1.c | 104 ++++++++++++++++++++
 2 files changed, 129 insertions(+), 2 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/loongarch/sync-1.c

diff --git a/gcc/config/loongarch/sync.md b/gcc/config/loongarch/sync.md
index 0c4f1983e..8a8e6247b 100644
--- a/gcc/config/loongarch/sync.md
+++ b/gcc/config/loongarch/sync.md
@@ -448,6 +448,29 @@
 }
   [(set (attr "length") (const_int 32))])
 
+(define_insn "atomic_cas_value_exchange_and_7_<mode>"
+  [(set (match_operand:GPR 0 "register_operand" "=&r")
+	(match_operand:GPR 1 "memory_operand" "+ZC"))
+   (set (match_dup 1)
+	(unspec_volatile:GPR [(match_operand:GPR 2 "reg_or_0_operand" "rJ")
+			      (match_operand:GPR 3 "reg_or_0_operand" "rJ")
+			      (match_operand:GPR 4 "reg_or_0_operand" "rJ")
+			      (match_operand:GPR 5 "reg_or_0_operand"  "rJ")
+			      (match_operand:SI 6 "const_int_operand")] ;; model
+	 UNSPEC_SYNC_EXCHANGE))
+   (clobber (match_scratch:GPR 7 "=&r"))]
+  ""
+{
+  return "%G6\\n\\t"
+	 "1:\\n\\t"
+	 "ll.<amo>\\t%0,%1\\n\\t"
+	 "and\\t%7,%0,%z3\\n\\t"
+	 "or%i5\\t%7,%7,%5\\n\\t"
+	 "sc.<amo>\\t%7,%1\\n\\t"
+	 "beqz\\t%7,1b\\n\\t";
+}
+  [(set (attr "length") (const_int 20))])
+
 (define_expand "atomic_exchange<mode>"
   [(set (match_operand:SHORT 0 "register_operand")
 	(unspec_volatile:SHORT
@@ -459,9 +482,9 @@
   ""
 {
   union loongarch_gen_fn_ptrs generator;
-  generator.fn_7 = gen_atomic_cas_value_cmp_and_7_si;
+  generator.fn_7 = gen_atomic_cas_value_exchange_and_7_si;
   loongarch_expand_atomic_qihi (generator, operands[0], operands[1],
-				operands[1], operands[2], operands[3]);
+				const0_rtx, operands[2], operands[3]);
   DONE;
 })
 
diff --git a/gcc/testsuite/gcc.target/loongarch/sync-1.c b/gcc/testsuite/gcc.target/loongarch/sync-1.c
new file mode 100644
index 000000000..cebed6a9b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/loongarch/sync-1.c
@@ -0,0 +1,104 @@
+/* Test __sync_test_and_set in atomic_exchange */
+/* { dg-do run } */
+/* { dg-options "-lpthread -std=c11" } */
+
+#include <pthread.h>
+#include <sched.h>
+#include <stdatomic.h>
+#include <stdlib.h>
+
+#define NR_THREAD 16
+#define NR_DATA 1000
+#define ITER_COUNT 10000
+
+static int _data[NR_DATA];
+static char _lock;
+static int _overcnt;
+
+static inline void proc_yield(int cnt)
+{
+  __asm__ __volatile__("":::"memory");
+}
+
+static void unlock()
+{
+  return atomic_store_explicit(&_lock, 0, memory_order_seq_cst);
+}
+
+static int trylock()
+{
+  return atomic_exchange_explicit(&_lock, 1, memory_order_acquire) == 0;
+}
+
+static void lockslow()
+{
+  for (int i = 0;; i++) {
+    if (i < 10)
+      proc_yield(i);
+    else
+      sched_yield();
+    if (atomic_load_explicit(&_lock, memory_order_relaxed) == 0
+      && atomic_exchange_explicit(&_lock, 1, memory_order_acquire) == 0)
+      return;
+  }
+}
+
+static void lock()
+{
+  if (trylock())
+    return;
+  lockslow();
+}
+
+static void checkeq(int a, int b)
+{
+  if (a != b)
+    __builtin_abort();
+}
+
+static void adddata()
+{
+  int i, v;
+  lock();
+  v = _data[0];
+  for (i = 0; i < NR_DATA; i++) {
+    checkeq(_data[i], v);
+    _data[i]++;
+  }
+  unlock();
+}
+
+static void backoff()
+{
+  int i, data[NR_DATA] = {0};
+  for (i = 0; i < NR_DATA; i++) {
+    data[i]++;
+    checkeq(data[i], 1);
+  }
+}
+
+static void *write_mutex_thread(void *unused)
+{
+  int i;
+  for (i = 0; i < ITER_COUNT; i++) {
+    adddata();
+    backoff();
+  }
+  atomic_fetch_add(&_overcnt, 1);
+}
+
+int main()
+{
+  int cnt;
+
+  pthread_t threads[NR_THREAD];
+  for (int i = 0; i < NR_THREAD; i++)
+    pthread_create(&threads[i], 0, write_mutex_thread, NULL);
+  for (int i = 0; i < NR_THREAD; i++)
+    pthread_detach(threads[i]);
+  while(cnt != NR_THREAD) {
+    sched_yield();
+    cnt = atomic_load(&_overcnt);
+  }
+  return 0;
+}
-- 
2.34.3