Merge pull request #1330 from rodgert/master

Use GCC/Clang compiler intrinsics for atomic operations

acinclude.m4

@@ -603,6 +603,26 @@ int main (int argc, char *argv [])
     )
 }])
 
+dnl ################################################################################
+dnl # LIBZMQ_CHECK_ATOMIC_INSTRINSICS([action-if-found], [action-if-not-found])    #
+dnl # Check if compiler supoorts __atomic_Xxx intrinsics                           #
+dnl ################################################################################
+AC_DEFUN([LIBZMQ_CHECK_ATOMIC_INTRINSICS], [{
+    AC_MSG_CHECKING(whether compiler supports __atomic_Xxx intrinsics)
+    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
+/* atomic intrinsics test */
+int v = 0;
+int main (int, char **)
+{
+    int t = __atomic_add_fetch (&v, 1, __ATOMIC_ACQ_REL);
+    return t;
+}
+    ])],
+    [AC_MSG_RESULT(yes) ; libzmq_cv_has_atomic_instrisics="yes" ; $1],
+    [AC_MSG_RESULT(no)  ; libzmq_cv_has_atomic_instrisics="no"  ; $2]
+    )
+}])
+
 dnl ################################################################################
 dnl # LIBZMQ_CHECK_SO_KEEPALIVE([action-if-found], [action-if-not-found])          #
 dnl # Check if SO_KEEPALIVE is supported                                           #
@@ -764,7 +784,7 @@ kqueue();
 dnl ################################################################################
 dnl # LIBZMQ_CHECK_POLLER_EPOLL_RUN([action-if-found], [action-if-not-found])      #
 dnl # Checks epoll polling system can actually run #
-dnl # For cross-compile, only requires that epoll can link # 
+dnl # For cross-compile, only requires that epoll can link #
 dnl ################################################################################
 AC_DEFUN([LIBZMQ_CHECK_POLLER_EPOLL], [{
     AC_RUN_IFELSE(
@@ -794,7 +814,7 @@ return(r < 0);
               )],
               [libzmq_cv_have_poller_epoll="yes" ; $1],
               [libzmq_cv_have_poller_epoll="no" ; $2])
-        
+
         ])
 }])
 

configure.ac

@@ -479,6 +479,15 @@ AM_CONDITIONAL(ON_CYGWIN, test "x$libzmq_on_cygwin" = "xyes")
 AM_CONDITIONAL(ON_ANDROID, test "x$libzmq_on_android" = "xyes")
 AM_CONDITIONAL(ON_LINUX, test "x$libzmq_on_linux" = "xyes")
 
+# Check for __atomic_Xxx compiler intrinsics
+AC_LANG_PUSH([C++])
+LIBZMQ_CHECK_ATOMIC_INTRINSICS([
+    AC_DEFINE([ZMQ_HAVE_ATOMIC_INTRINSICS],
+              [1],
+              [Whether compiler has __atomic_Xxx intrinsics.])
+    ])
+AC_LANG_POP([C++])
+
 # Checks for library functions.
 AC_TYPE_SIGNAL
 AC_CHECK_FUNCS(perror gettimeofday clock_gettime memset socket getifaddrs freeifaddrs fork posix_memalign)

src/atomic_counter.hpp

@@ -25,6 +25,8 @@
 
 #if defined ZMQ_FORCE_MUTEXES
 #define ZMQ_ATOMIC_COUNTER_MUTEX
+#elif defined ZMQ_HAVE_ATOMIC_INTRINSICS
+#define ZMQ_ATOMIC_INTRINSIC
 #elif (defined __i386__ || defined __x86_64__) && defined __GNUC__
 #define ZMQ_ATOMIC_COUNTER_X86
 #elif defined __ARM_ARCH_7A__ && defined __GNUC__
@@ -83,11 +85,13 @@ namespace zmq
 
 #if defined ZMQ_ATOMIC_COUNTER_WINDOWS
             old_value = InterlockedExchangeAdd ((LONG*) &value, increment_);
+#elif defined ZMQ_ATOMIC_INTRINSIC
+            old_value = __atomic_fetch_add(&value, increment_, __ATOMIC_ACQ_REL);
 #elif defined ZMQ_ATOMIC_COUNTER_ATOMIC_H
             integer_t new_value = atomic_add_32_nv (&value, increment_);
             old_value = new_value - increment_;
 #elif defined ZMQ_ATOMIC_COUNTER_TILE
-	    old_value = arch_atomic_add (&value, increment_);
+            old_value = arch_atomic_add (&value, increment_);
 #elif defined ZMQ_ATOMIC_COUNTER_X86
             __asm__ volatile (
                 "lock; xadd %0, %1 \n\t"
@@ -125,6 +129,9 @@ namespace zmq
             LONG delta = - ((LONG) decrement);
             integer_t old = InterlockedExchangeAdd ((LONG*) &value, delta);
             return old - decrement != 0;
+#elif defined ZMQ_ATOMIC_INTRINSIC
+            integer_t nv = __atomic_sub_fetch(&value, decrement, __ATOMIC_ACQ_REL);
+            return nv != 0;
 #elif defined ZMQ_ATOMIC_COUNTER_ATOMIC_H
             int32_t delta = - ((int32_t) decrement);
             integer_t nv = atomic_add_32_nv (&value, delta);

src/atomic_ptr.hpp

@@ -24,6 +24,8 @@
 
 #if defined ZMQ_FORCE_MUTEXES
 #define ZMQ_ATOMIC_PTR_MUTEX
+#elif defined ZMQ_HAVE_ATOMIC_INTRINSICS
+#define ZMQ_ATOMIC_PTR_INTRINSIC
 #elif (defined __i386__ || defined __x86_64__) && defined __GNUC__
 #define ZMQ_ATOMIC_PTR_X86
 #elif defined __ARM_ARCH_7A__ && defined __GNUC__
@@ -82,6 +84,8 @@ namespace zmq
         {
 #if defined ZMQ_ATOMIC_PTR_WINDOWS
             return (T*) InterlockedExchangePointer ((PVOID*) &ptr, val_);
+#elif defined ZMQ_ATOMIC_PTR_INTRINSIC
+            return (T*) __atomic_exchange_n (&ptr, val_, __ATOMIC_ACQ_REL);
 #elif defined ZMQ_ATOMIC_PTR_ATOMIC_H
             return (T*) atomic_swap_ptr (&ptr, val_);
 #elif defined ZMQ_ATOMIC_PTR_TILE
@@ -127,6 +131,11 @@ namespace zmq
 #if defined ZMQ_ATOMIC_PTR_WINDOWS
             return (T*) InterlockedCompareExchangePointer (
                 (volatile PVOID*) &ptr, val_, cmp_);
+#elif defined ZMQ_ATOMIC_PTR_INTRINSIC
+            T *old = cmp_;
+            __atomic_compare_exchange_n (&ptr, (volatile T**) &old, val_, false,
+                    __ATOMIC_RELEASE, __ATOMIC_ACQUIRE);
+            return old;
 #elif defined ZMQ_ATOMIC_PTR_ATOMIC_H
             return (T*) atomic_cas_ptr (&ptr, cmp_, val_);
 #elif defined ZMQ_ATOMIC_PTR_TILE