Merge remote-tracking branch 'localpub/dev' into dev

1 files changed, 39 insertions, 34 deletions

diff --git a/src/stream_manager.h b/src/stream_manager.h
index d543e68..afcbb0e 100644
--- a/src/stream_manager.h
+++ b/src/stream_manager.h
@@ -44,6 +44,43 @@
 //    unsigned m_pending_streams;
 //};
 
+struct CUevent_st {
+ public:
+  CUevent_st(bool blocking) {
+    m_uid = ++m_next_event_uid;
+    m_blocking = blocking;
+    m_updates = 0;
+    m_wallclock = 0;
+    m_gpu_tot_sim_cycle = 0;
+    m_issued = 0;
+    m_done = false;
+  }
+  void update(double cycle, time_t clk) {
+    m_updates++;
+    m_wallclock = clk;
+    m_gpu_tot_sim_cycle = cycle;
+    m_done = true;
+  }
+  // void set_done() { assert(!m_done); m_done=true; }
+  int get_uid() const { return m_uid; }
+  unsigned num_updates() const { return m_updates; }
+  bool done() const { return m_updates == m_issued; }
+  time_t clock() const { return m_wallclock; }
+  void issue() { m_issued++; }
+  unsigned int num_issued() const { return m_issued; }
+
+ private:
+  int m_uid;
+  bool m_blocking;
+  bool m_done;
+  int m_updates;
+  unsigned int m_issued;
+  time_t m_wallclock;
+  double m_gpu_tot_sim_cycle;
+
+  static int m_next_event_uid;
+};
+
 enum stream_operation_type {
   stream_no_op,
   stream_memcpy_host_to_device,
@@ -106,6 +143,7 @@ class stream_operation {
     m_kernel = NULL;
     m_type = stream_wait_event;
     m_event = e;
+    m_cnt = m_event->num_issued();
     m_stream = stream;
     m_done = false;
   }
@@ -184,40 +222,6 @@ class stream_operation {
   kernel_info_t *m_kernel;
   struct CUevent_st *m_event;
 };
-
-struct CUevent_st {
- public:
-  CUevent_st(bool blocking) {
-    m_uid = ++m_next_event_uid;
-    m_blocking = blocking;
-    m_updates = 0;
-    m_wallclock = 0;
-    m_gpu_tot_sim_cycle = 0;
-    m_done = false;
-  }
-  void update(double cycle, time_t clk) {
-    m_updates++;
-    m_wallclock = clk;
-    m_gpu_tot_sim_cycle = cycle;
-    m_done = true;
-  }
-  // void set_done() { assert(!m_done); m_done=true; }
-  int get_uid() const { return m_uid; }
-  unsigned num_updates() const { return m_updates; }
-  bool done() const { return m_done; }
-  time_t clock() const { return m_wallclock; }
-
- private:
-  int m_uid;
-  bool m_blocking;
-  bool m_done;
-  int m_updates;
-  time_t m_wallclock;
-  double m_gpu_tot_sim_cycle;
-
-  static int m_next_event_uid;
-};
-
 struct CUstream_st {
  public:
   CUstream_st();
@@ -272,6 +276,7 @@ class stream_manager {
   CUstream_st m_stream_zero;
   bool m_service_stream_zero;
   pthread_mutex_t m_lock;
+  std::list<struct CUstream_st *>::iterator m_last_stream;
 };
 
 #endif