Reverted the previous commit to add a cleaner way of getting NUM_THREADS. Now, sst_impl doesn't functionally execute on the last indexed element of an array, but instead on the actual last thread that executes

1 files changed, 16 insertions, 0 deletions

diff --git a/src/cuda-sim/ptx_sim.cc b/src/cuda-sim/ptx_sim.cc
index 511e8d6..f48115b 100644
--- a/src/cuda-sim/ptx_sim.cc
+++ b/src/cuda-sim/ptx_sim.cc
@@ -44,6 +44,7 @@ ptx_cta_info::ptx_cta_info( unsigned sm_idx )
 
    m_sm_idx = sm_idx;
    m_uid = g_ptx_cta_info_uid++;
+   m_bar_threads = 0;
 }
 
 void ptx_cta_info::add_thread( ptx_thread_info *thd )
@@ -128,6 +129,21 @@ unsigned ptx_cta_info::get_sm_idx() const
    return m_sm_idx;
 }
 
+unsigned ptx_cta_info::get_bar_threads() const
+{
+   return m_bar_threads;
+}
+
+void ptx_cta_info::inc_bar_threads()
+{
+	m_bar_threads++;
+}
+
+void ptx_cta_info::reset_bar_threads()
+{
+	m_bar_threads = 0;
+}
+
 unsigned g_ptx_thread_info_uid_next=1;
 unsigned g_ptx_thread_info_delete_count=0;