5 files changed, 150 insertions, 0 deletions

diff --git a/src/core/platform.h b/src/core/platform.h
index 48f1e1a7..0a0709d1 100644
--- a/src/core/platform.h
+++ b/src/core/platform.h
@@ -163,6 +163,15 @@ typedef struct nni_atomic_flag nni_atomic_flag;
 extern bool nni_atomic_flag_test_and_set(nni_atomic_flag *);
 extern void nni_atomic_flag_reset(nni_atomic_flag *);
 
+typedef struct nni_atomic_u64 nni_atomic_u64;
+
+extern void     nni_atomic_init64(nni_atomic_u64 *);
+extern void     nni_atomic_inc64(nni_atomic_u64 *, uint64_t);
+extern void     nni_atomic_dec64(nni_atomic_u64 *, uint64_t);
+extern uint64_t nni_atomic_get64(nni_atomic_u64 *);
+extern void     nni_atomic_set64(nni_atomic_u64 *, uint64_t);
+extern uint64_t nni_atomic_swap64(nni_atomic_u64 *, uint64_t);
+
 //
 // Clock Support
 //
diff --git a/src/platform/posix/posix_atomic.c b/src/platform/posix/posix_atomic.c
index a8d2579d..7f735ec0 100644
--- a/src/platform/posix/posix_atomic.c
+++ b/src/platform/posix/posix_atomic.c
@@ -28,6 +28,43 @@ nni_atomic_flag_reset(nni_atomic_flag *f)
 {
 	atomic_flag_clear(&f->f);
 }
+
+void
+nni_atomic_inc64(nni_atomic_u64 *v, uint64_t bump)
+{
+	(void) atomic_fetch_add_explicit(&v->v, bump, memory_order_relaxed);
+}
+
+void
+nni_atomic_dec64(nni_atomic_u64 *v, uint64_t bump)
+{
+	(void) atomic_fetch_sub_explicit(&v->v, bump, memory_order_relaxed);
+}
+
+uint64_t
+nni_atomic_get64(nni_atomic_u64 *v)
+{
+	return (atomic_load(&v->v));
+}
+
+void
+nni_atomic_set64(nni_atomic_u64 *v, uint64_t u)
+{
+	atomic_store(&v->v, u);
+}
+
+uint64_t
+nni_atomic_swap64(nni_atomic_u64 *v, uint64_t u)
+{
+	return (atomic_exchange(&v->v, u));
+}
+
+void
+nni_atomic_init64(nni_atomic_u64 *v)
+{
+	atomic_init(&v->v, 0);
+}
+
 #else
 
 #include <pthread.h>
@@ -52,6 +89,58 @@ nni_atomic_flag_reset(nni_atomic_flag *f)
 	f->f = false;
 	pthread_mutex_unlock(&plat_atomic_lock);
 }
+
+void
+nni_atomic_inc64(nni_atomic_u64 *v, uint64_t bump)
+{
+	pthread_mutex_lock(&plat_atomic_lock);
+	v += bump;
+	pthread_mutex_unlock(&plat_atomic_lock);
+}
+
+void
+nni_atomic_dec64(nni_atomic_u64 *v, uint64_t bump)
+{
+	pthread_mutex_lock(&plat_atomic_lock);
+	v -= bump;
+	pthread_mutex_unlock(&plat_atomic_lock);
+}
+
+uint64_t
+nni_atomic_get64(nni_atomic_u64 *v)
+{
+	uint64_t rv;
+	pthread_mutex_lock(&plat_atomic_lock);
+	rv = v->v;
+	pthread_mutex_unlock(&plat_atomic_lock);
+	return (rv);
+}
+
+void
+nni_atomic_set64(nni_atomic_u64 *v, uint64_t u)
+{
+	pthread_mutex_lock(&plat_atomic_lock);
+	v->v = u;
+	pthread_mutex_unlock(&plat_atomic_lock);
+}
+
+uint64_t
+nni_atomic_swap64(nni_atomic_u64 *v, uint64_t u)
+{
+	uint64_t rv;
+	pthread_mutex_lock(&plat_atomic_lock);
+	rv   = v->v;
+	v->v = u;
+	pthread_mutex_unlock(&plat_atomic_lock);
+	return (rv);
+}
+
+void
+nni_atomic_init64(nni_atomic_u64 *v)
+{
+	v->v = 0;
+}
+
 #endif
 
 #endif // NNG_PLATFORM_POSIX
diff --git a/src/platform/posix/posix_impl.h b/src/platform/posix/posix_impl.h
index 70a3615f..9f3b775b 100644
--- a/src/platform/posix/posix_impl.h
+++ b/src/platform/posix/posix_impl.h
@@ -82,10 +82,20 @@ struct nni_plat_flock {
 struct nni_atomic_flag {
 	atomic_flag f;
 };
+
+struct nni_atomic_u64 {
+	_Atomic unsigned long long v;
+};
+
 #else // NNG_HAVE_C11_ATOMIC
 struct nni_atomic_flag {
 	bool f;
 };
+
+struct nni_atomic_flag {
+	uint64_t v;
+};
+
 #endif
 
 #endif
diff --git a/src/platform/windows/win_impl.h b/src/platform/windows/win_impl.h
index 2cdff1c0..462e1707 100644
--- a/src/platform/windows/win_impl.h
+++ b/src/platform/windows/win_impl.h
@@ -52,6 +52,10 @@ struct nni_atomic_flag {
 	unsigned f;
 };
 
+struct nni_atomic_u64 {
+	LONGLONG v;
+};
+
 // nni_win_io is used with io completion ports.  This allows us to get
 // to a specific completion callback without requiring the poller (in the
 // completion port) to know anything about the event itself.
diff --git a/src/platform/windows/win_thread.c b/src/platform/windows/win_thread.c
index b9743c5e..f80c8e3f 100644
--- a/src/platform/windows/win_thread.c
+++ b/src/platform/windows/win_thread.c
@@ -121,6 +121,44 @@ nni_atomic_flag_reset(nni_atomic_flag *f)
 	InterlockedExchange(&f->f, 0);
 }
 
+void
+nni_atomic_inc64(nni_atomic_u64 *v, uint64_t bump)
+{
+	InterlockedAddNoFence64(&v->v, (LONGLONG) bump);
+}
+
+void
+nni_atomic_dec64(nni_atomic_u64 *v, uint64_t bump)
+{
+	// Windows lacks a sub, so we add the negative.
+	InterlockedAddNoFence64(&v->v, (0ll - (LONGLONG) bump));
+}
+
+uint64_t
+nni_atomic_get64(nni_atomic_u64 *v)
+{
+
+	return ((uint64_t)(InterlockedExchangeAdd64(&v->v, 0)));
+}
+
+void
+nni_atomic_set64(nni_atomic_u64 *v, uint64_t u)
+{
+	return (InterlockedExchange64(&v->v, (LONGLONG) u));
+}
+
+uint64_t
+nni_atomic_swap64(nni_atomic_u64 *v, uint64_t u)
+{
+	return ((uint64_t)(InterlockedExchange64(&v->v, (LONGLONG) u)));
+}
+
+void
+nni_atomic_init64(nni_atomic_u64 *v)
+{
+	InterlockedExchange64(&v->v, 0);
+}
+
 static unsigned int __stdcall nni_plat_thr_main(void *arg)
 {
 	nni_plat_thr *thr = arg;