apr_atomic: Use __atomic builtins when available.

Unlike Intel's atomic builtins (__sync_*), the more recent __atomic builtins
provide atomic load and store for weakly ordered architectures like ARM32 or
powerpc[64], so use them when available (gcc 4.6.3+).



git-svn-id: https://svn.apache.org/repos/asf/apr/apr/trunk@1894621 13f79535-47bb-0310-9956-ffa450edef68

4 files changed, 120 insertions, 21 deletions

diff --git a/atomic/unix/builtins.c b/atomic/unix/builtins.c
index ebf8833d1..d0f1b454c 100644
--- a/atomic/unix/builtins.c
+++ b/atomic/unix/builtins.c
@@ -25,57 +25,97 @@ APR_DECLARE(apr_status_t) apr_atomic_init(apr_pool_t *p)
 
 APR_DECLARE(apr_uint32_t) apr_atomic_read32(volatile apr_uint32_t *mem)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_load_n(mem, __ATOMIC_SEQ_CST);
+#else
     return *mem;
+#endif
 }
 
 APR_DECLARE(void) apr_atomic_set32(volatile apr_uint32_t *mem, apr_uint32_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_store_n(mem, val, __ATOMIC_SEQ_CST);
+#else
     *mem = val;
+#endif
 }
 
 APR_DECLARE(apr_uint32_t) apr_atomic_add32(volatile apr_uint32_t *mem, apr_uint32_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_fetch_add(mem, val, __ATOMIC_SEQ_CST);
+#else
     return __sync_fetch_and_add(mem, val);
+#endif
 }
 
 APR_DECLARE(void) apr_atomic_sub32(volatile apr_uint32_t *mem, apr_uint32_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_fetch_sub(mem, val, __ATOMIC_SEQ_CST);
+#else
     __sync_fetch_and_sub(mem, val);
+#endif
 }
 
 APR_DECLARE(apr_uint32_t) apr_atomic_inc32(volatile apr_uint32_t *mem)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_fetch_add(mem, 1, __ATOMIC_SEQ_CST);
+#else
     return __sync_fetch_and_add(mem, 1);
+#endif
 }
 
 APR_DECLARE(int) apr_atomic_dec32(volatile apr_uint32_t *mem)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_sub_fetch(mem, 1, __ATOMIC_SEQ_CST);
+#else
     return __sync_sub_and_fetch(mem, 1);
+#endif
 }
 
-APR_DECLARE(apr_uint32_t) apr_atomic_cas32(volatile apr_uint32_t *mem, apr_uint32_t with,
+APR_DECLARE(apr_uint32_t) apr_atomic_cas32(volatile apr_uint32_t *mem, apr_uint32_t val,
                                            apr_uint32_t cmp)
 {
-    return __sync_val_compare_and_swap(mem, cmp, with);
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_compare_exchange_n(mem, &cmp, val, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+    return cmp;
+#else
+    return __sync_val_compare_and_swap(mem, cmp, val);
+#endif
 }
 
 APR_DECLARE(apr_uint32_t) apr_atomic_xchg32(volatile apr_uint32_t *mem, apr_uint32_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_exchange_n(mem, val, __ATOMIC_SEQ_CST);
+#else
     __sync_synchronize();
-
     return __sync_lock_test_and_set(mem, val);
+#endif
 }
 
-APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *with, const void *cmp)
+APR_DECLARE(void*) apr_atomic_casptr(void *volatile *mem, void *ptr, const void *cmp)
 {
-    return (void*) __sync_val_compare_and_swap(mem, cmp, with);
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_compare_exchange_n(mem, (void **)&cmp, ptr, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+    return (void *)cmp;
+#else
+    return (void *)__sync_val_compare_and_swap(mem, (void *)cmp, ptr);
+#endif
 }
 
-APR_DECLARE(void*) apr_atomic_xchgptr(void *volatile *mem, void *with)
+APR_DECLARE(void*) apr_atomic_xchgptr(void *volatile *mem, void *ptr)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_exchange_n(mem, ptr, __ATOMIC_SEQ_CST);
+#else
     __sync_synchronize();
-
-    return (void*) __sync_lock_test_and_set(mem, with);
+    return __sync_lock_test_and_set(mem, ptr);
+#endif
 }
 
 #endif /* USE_ATOMICS_BUILTINS */
diff --git a/atomic/unix/builtins64.c b/atomic/unix/builtins64.c
index 4a4b685c7..0ac950c15 100644
--- a/atomic/unix/builtins64.c
+++ b/atomic/unix/builtins64.c
@@ -20,45 +20,77 @@
 
 APR_DECLARE(apr_uint64_t) apr_atomic_read64(volatile apr_uint64_t *mem)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_load_n(mem, __ATOMIC_SEQ_CST);
+#else
     return *mem;
+#endif
 }
 
 APR_DECLARE(void) apr_atomic_set64(volatile apr_uint64_t *mem, apr_uint64_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_store_n(mem, val, __ATOMIC_SEQ_CST);
+#else
     *mem = val;
+#endif
 }
 
 APR_DECLARE(apr_uint64_t) apr_atomic_add64(volatile apr_uint64_t *mem, apr_uint64_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_fetch_add(mem, val, __ATOMIC_SEQ_CST);
+#else
     return __sync_fetch_and_add(mem, val);
+#endif
 }
 
 APR_DECLARE(void) apr_atomic_sub64(volatile apr_uint64_t *mem, apr_uint64_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_fetch_sub(mem, val, __ATOMIC_SEQ_CST);
+#else
     __sync_fetch_and_sub(mem, val);
+#endif
 }
 
 APR_DECLARE(apr_uint64_t) apr_atomic_inc64(volatile apr_uint64_t *mem)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_fetch_add(mem, 1, __ATOMIC_SEQ_CST);
+#else
     return __sync_fetch_and_add(mem, 1);
+#endif
 }
 
 APR_DECLARE(int) apr_atomic_dec64(volatile apr_uint64_t *mem)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_sub_fetch(mem, 1, __ATOMIC_SEQ_CST);
+#else
     return __sync_sub_and_fetch(mem, 1);
+#endif
 }
 
-APR_DECLARE(apr_uint64_t) apr_atomic_cas64(volatile apr_uint64_t *mem, apr_uint64_t with,
+APR_DECLARE(apr_uint64_t) apr_atomic_cas64(volatile apr_uint64_t *mem, apr_uint64_t val,
                                            apr_uint64_t cmp)
 {
-    return __sync_val_compare_and_swap(mem, cmp, with);
+#if HAVE__ATOMIC_BUILTINS
+    __atomic_compare_exchange_n(mem, &cmp, val, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+    return cmp;
+#else
+    return __sync_val_compare_and_swap(mem, cmp, val);
+#endif
 }
 
 APR_DECLARE(apr_uint64_t) apr_atomic_xchg64(volatile apr_uint64_t *mem, apr_uint64_t val)
 {
+#if HAVE__ATOMIC_BUILTINS
+    return __atomic_exchange_n(mem, val, __ATOMIC_SEQ_CST);
+#else
     __sync_synchronize();
-
     return __sync_lock_test_and_set(mem, val);
+#endif
 }
 
 #endif /* USE_ATOMICS_BUILTINS */
diff --git a/atomic/unix/mutex.c b/atomic/unix/mutex.c
index 78ad75336..b9d1cedc9 100644
--- a/atomic/unix/mutex.c
+++ b/atomic/unix/mutex.c
@@ -89,7 +89,7 @@ static APR_INLINE apr_thread_mutex_t *mutex_hash(volatile apr_uint32_t *mem)
 
 APR_DECLARE(apr_status_t) apr_atomic_init(apr_pool_t *p)
 {
-    return apr__atomic_generic64_init(p);
+    return APR_SUCCESS;
 }
 
 #endif /* APR_HAS_THREADS */
diff --git a/configure.in b/configure.in
index 5fa318306..3e4be99cc 100644
--- a/configure.in
+++ b/configure.in
@@ -487,7 +487,7 @@ esac
 
 AC_CACHE_CHECK([whether the compiler provides atomic builtins], [ap_cv_atomic_builtins],
 [AC_TRY_RUN([
-int main()
+int main(int argc, const char *const *argv)
 {
     unsigned long val = 1010, tmp, *mem = &val;
 
@@ -495,7 +495,6 @@ int main()
         return 1;
 
     tmp = val;
-
     if (__sync_fetch_and_sub(mem, 1010) != tmp || val != 1010)
         return 1;
 
@@ -503,28 +502,56 @@ int main()
         return 1;
 
     tmp = 3030;
-
     if (__sync_val_compare_and_swap(mem, 0, tmp) != 0 || val != tmp)
         return 1;
 
+    __sync_synchronize();
     if (__sync_lock_test_and_set(&val, 4040) != 3030)
         return 1;
 
-    mem = &tmp;
+    if (__sync_val_compare_and_swap(&mem, &val, &tmp) != &val || mem != &tmp)
+        return 1;
+
+    return 0;
+}], [ap_cv_atomic_builtins=yes], [ap_cv_atomic_builtins=no], [ap_cv_atomic_builtins=no])])
+
+AC_CACHE_CHECK([whether the compiler provides __atomic builtins], [ap_cv__atomic_builtins],
+[AC_TRY_RUN([
+int main(int argc, const char *const *argv)
+{
+    unsigned long val = 1010, tmp, *mem = &val, *ptmp;
 
-    if (__sync_val_compare_and_swap(&mem, &tmp, &val) != &tmp)
+    if (__atomic_fetch_add(&val, 1010, __ATOMIC_SEQ_CST) != 1010 || val != 2020)
         return 1;
 
-    __sync_synchronize();
+    tmp = val;
+    if (__atomic_fetch_sub(mem, 1010, __ATOMIC_SEQ_CST) != tmp || val != 1010)
+        return 1;
+
+    if (__atomic_sub_fetch(&val, 1010, __ATOMIC_SEQ_CST) != 0 || val != 0)
+        return 1;
 
-    if (mem != &val)
+    tmp = val;
+    if (!__atomic_compare_exchange_n(mem, &tmp, 3030, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)
+            || tmp != 0)
+        return 1;
+
+    if (__atomic_exchange_n(&val, 4040, __ATOMIC_SEQ_CST) != 3030)
+        return 1;
+
+    ptmp = &val;
+    if (!__atomic_compare_exchange_n(&mem, &ptmp, &tmp, 0, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)
+            || ptmp != &val || mem != &tmp)
         return 1;
 
     return 0;
-}], [ap_cv_atomic_builtins=yes], [ap_cv_atomic_builtins=no], [ap_cv_atomic_builtins=no])])
+}], [ap_cv__atomic_builtins=yes], [ap_cv__atomic_builtins=no], [ap_cv__atomic_builtins=no])])
 
-if test "$ap_cv_atomic_builtins" = "yes"; then
+if test "$ap_cv_atomic_builtins" = "yes" -o "$ap_cv__atomic_builtins" = "yes"; then
     AC_DEFINE(HAVE_ATOMIC_BUILTINS, 1, [Define if compiler provides atomic builtins])
+    if test "$ap_cv__atomic_builtins" = "yes"; then
+        AC_DEFINE(HAVE__ATOMIC_BUILTINS, 1, [Define if compiler provides __atomic builtins])
+    fi
 fi
 
 AC_CACHE_CHECK([whether the compiler handles weak symbols], [ap_cv_weak_symbols],