Actually fixing the atomic bug

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 10756]

4 files changed, 21 insertions, 18 deletions

diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index 802547e..90f45d9 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -125,6 +125,23 @@ void warp_inst_t::set_active( const active_mask_t &active ) {
    }
 }
 
+void warp_inst_t::do_atomic() {
+    do_atomic( m_warp_active_mask );
+}
+
+void warp_inst_t::do_atomic( const active_mask_t& access_mask ) {
+    assert( m_isatomic && !m_empty );
+    for( unsigned i=0; i < m_config->warp_size; i++ )
+    {
+        if( access_mask.test(i) )
+        {
+            dram_callback_t &cb = m_per_scalar_thread[i].callback;
+            if( cb.thread )
+                cb.function(cb.instruction, cb.thread);
+        }
+    }
+}
+
 void warp_inst_t::generate_mem_accesses()
 {
     if( empty() || op == MEMORY_BARRIER_OP || m_mem_accesses_created ) 
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index a314300..4942110 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -654,23 +654,8 @@ public:
     }
 
     // modifiers
-    void do_atomic() {
-        do_atomic( m_warp_active_mask );
-    }
-    void do_atomic( const active_mask_t& access_mask )
-    {
-        assert( m_isatomic && !m_empty );
-	for( unsigned i=0; i < m_config->warp_size; i++ )
-		{
-			if( access_mask.test(i) )
-			{
-				dram_callback_t &cb = m_per_scalar_thread[i].callback;
-				if( cb.thread )
-					cb.function(cb.instruction, cb.thread);
-			}
-		}
-
-    }
+    void do_atomic();
+    void do_atomic( const active_mask_t& access_mask );
     void clear() 
     { 
         m_empty=true; 
diff --git a/src/gpgpu-sim/mem_fetch.h b/src/gpgpu-sim/mem_fetch.h
index 369ea33..cc3147c 100644
--- a/src/gpgpu-sim/mem_fetch.h
+++ b/src/gpgpu-sim/mem_fetch.h
@@ -118,6 +118,7 @@ public:
    unsigned get_icnt_receive_time() const { return m_icnt_receive_time; }
 
    enum mem_access_type get_access_type() const { return m_access.get_type(); }
+   const active_mask_t& get_access_warp_mask() const { return m_access.get_warp_mask(); }
    address_type get_pc() const { return m_inst.empty()?-1:m_inst.pc; }
    const warp_inst_t &get_inst() { return m_inst; }
    enum mem_fetch_status get_status() const { return m_status; }
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 50ad5ea..4ce7536 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -1132,7 +1132,7 @@ void ldst_unit::writeback()
             if( m_next_global ) {
                 m_next_wb = m_next_global->get_inst();
                 if( m_next_global->isatomic() ) 
-                    m_core->decrement_atomic_count(m_next_global->get_wid(),m_next_wb.active_count());
+                    m_core->decrement_atomic_count(m_next_global->get_wid(),m_next_global->get_access_warp_mask().count());
                 delete m_next_global;
                 m_next_global = NULL;
             }