integrate bug fix (passes fast regression)

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 8310]

4 files changed, 21 insertions, 6 deletions

diff --git a/src/gpgpu-sim/dram.cc b/src/gpgpu-sim/dram.cc
index bd6758b..288baaa 100644
--- a/src/gpgpu-sim/dram.cc
+++ b/src/gpgpu-sim/dram.cc
@@ -70,6 +70,7 @@
 #include "mem_latency_stat.h"
 #include "dram_sched.h"
 #include "mem_fetch.h"
+#include "l2cache.h"
 
 #ifdef DRAM_VERIFY
 int PRINT_CYCLE = 0;
@@ -78,9 +79,11 @@ int PRINT_CYCLE = 0;
 template class fifo_pipeline<mem_fetch>;
 template class fifo_pipeline<dram_req_t>;
 
-dram_t::dram_t( unsigned int partition_id, const struct memory_config *config, memory_stats_t *stats )
+dram_t::dram_t( unsigned int partition_id, const struct memory_config *config, memory_stats_t *stats,
+                memory_partition_unit *mp )
 {
    id = partition_id;
+   m_memory_partition_unit = mp;
    m_stats = stats;
    m_config = config;
 
@@ -237,8 +240,13 @@ void dram_t::cycle()
            if (cmd->dqbytes >= cmd->nbytes) {
               mem_fetch *data = cmd->data; 
               data->set_status(IN_PARTITION_MC_RETURNQ,gpu_sim_cycle+gpu_tot_sim_cycle); 
-              data->set_reply();
-              returnq->push(data);
+              if( data->get_access_type() != L1_WRBK_ACC && data->get_access_type() != L2_WRBK_ACC ) {
+                 data->set_reply();
+                 returnq->push(data);
+              } else {
+                 m_memory_partition_unit->set_done(data);
+                 delete data;
+              }
               delete cmd;
            }
 #ifdef DRAM_VIEWCMD 
diff --git a/src/gpgpu-sim/dram.h b/src/gpgpu-sim/dram.h
index 1a684c2..d4e8255 100644
--- a/src/gpgpu-sim/dram.h
+++ b/src/gpgpu-sim/dram.h
@@ -123,7 +123,8 @@ struct mem_fetch;
 class dram_t 
 {
 public:
-   dram_t( unsigned int parition_id, const struct memory_config *config, class memory_stats_t *stats );
+   dram_t( unsigned int parition_id, const struct memory_config *config, class memory_stats_t *stats, 
+           class memory_partition_unit *mp );
 
    bool full() const;
    void print( FILE* simFile ) const;
@@ -139,7 +140,7 @@ public:
    void cycle();
    void dram_log (int task);
 
-   struct memory_partition_unit *m_memory_partition_unit;
+   class memory_partition_unit *m_memory_partition_unit;
    unsigned int id;
 
 private:
diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc
index bb13cbc..5c19c32 100644
--- a/src/gpgpu-sim/l2cache.cc
+++ b/src/gpgpu-sim/l2cache.cc
@@ -105,7 +105,7 @@ memory_partition_unit::memory_partition_unit( unsigned partition_id,
     m_id = partition_id;
     m_config=config;
     m_stats=stats;
-    m_dram = new dram_t(m_id,m_config,m_stats);
+    m_dram = new dram_t(m_id,m_config,m_stats,this);
 
     char L2c_name[32];
     snprintf(L2c_name, 32, "L2_bank_%03d", m_id);
@@ -317,6 +317,11 @@ mem_fetch* memory_partition_unit::top()
     return mf;
 }
 
+void memory_partition_unit::set_done( mem_fetch *mf )
+{
+    m_request_tracker.erase(mf);
+}
+
 void memory_partition_unit::dram_cycle() 
 { 
     // pop completed memory request from dram and push it to dram-to-L2 queue 
diff --git a/src/gpgpu-sim/l2cache.h b/src/gpgpu-sim/l2cache.h
index 761a220..fb0c6e2 100644
--- a/src/gpgpu-sim/l2cache.h
+++ b/src/gpgpu-sim/l2cache.h
@@ -105,6 +105,7 @@ public:
    void push( class mem_fetch* mf, unsigned long long clock_cycle );
    class mem_fetch* pop(); 
    class mem_fetch* top();
+   void set_done( mem_fetch *mf );
 
    unsigned flushL2();