4 files changed, 26 insertions, 7 deletions

diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 121e079..9a28b55 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -668,7 +668,7 @@ gpgpu_sim::gpgpu_sim( const gpgpu_sim_config &config )
     m_memory_partition_unit = new memory_partition_unit*[m_memory_config->m_n_mem];
     m_memory_sub_partition = new memory_sub_partition*[m_memory_config->m_n_mem_sub_partition];
     for (unsigned i=0;i<m_memory_config->m_n_mem;i++) {
-        m_memory_partition_unit[i] = new memory_partition_unit(i, m_memory_config, m_memory_stats);
+        m_memory_partition_unit[i] = new memory_partition_unit(i, m_memory_config, m_memory_stats, this);
         for (unsigned p = 0; p < m_memory_config->m_n_sub_partition_per_memory_channel; p++) {
             unsigned submpid = i * m_memory_config->m_n_sub_partition_per_memory_channel + p; 
             m_memory_sub_partition[submpid] = m_memory_partition_unit[i]->get_sub_partition(p); 
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 1778008..f9b5dad 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -36,6 +36,7 @@
 #include <iostream>
 #include <fstream>
 #include <list>
+#include <unordered_set>
 #include <stdio.h>
 
 
@@ -428,6 +429,8 @@ public:
 
     void perf_memcpy_to_gpu( size_t dst_start_addr, size_t count );
 
+   std::unordered_set<unsigned long long> data_footprint_stats;
+
    //The next three functions added to be used by the functional simulation function
    
    //! Get shader core configuration
diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc
index b1465a8..a4aaf28 100644
--- a/src/gpgpu-sim/l2cache.cc
+++ b/src/gpgpu-sim/l2cache.cc
@@ -62,15 +62,16 @@ mem_fetch * partition_mf_allocator::alloc(new_addr_type addr, mem_access_type ty
 
 memory_partition_unit::memory_partition_unit( unsigned partition_id, 
                                               const struct memory_config *config,
-                                              class memory_stats_t *stats )
-: m_id(partition_id), m_config(config), m_stats(stats), m_arbitration_metadata(config) 
+                                              class memory_stats_t *stats,
+                                              class gpgpu_sim *gpu)
+: m_id(partition_id), m_config(config), m_stats(stats), m_arbitration_metadata(config), m_gpu(gpu)
 {
     m_dram = new dram_t(m_id,m_config,m_stats,this);
 
     m_sub_partition = new memory_sub_partition*[m_config->m_n_sub_partition_per_memory_channel]; 
     for (unsigned p = 0; p < m_config->m_n_sub_partition_per_memory_channel; p++) {
         unsigned sub_partition_id = m_id * m_config->m_n_sub_partition_per_memory_channel + p; 
-        m_sub_partition[p] = new memory_sub_partition(sub_partition_id, m_config, stats); 
+        m_sub_partition[p] = new memory_sub_partition(sub_partition_id, m_config, stats, m_gpu); 
     }
 }
 
@@ -310,11 +311,13 @@ void memory_partition_unit::print( FILE *fp ) const
 
 memory_sub_partition::memory_sub_partition( unsigned sub_partition_id, 
                                             const struct memory_config *config,
-                                            class memory_stats_t *stats )
+                                            class memory_stats_t *stats ,
+                                            class gpgpu_sim *gpu)
 {
     m_id = sub_partition_id;
     m_config=config;
     m_stats=stats;
+    m_gpu=gpu;
     m_memcpy_cycle_offset = 0;
 
     assert(m_id < m_config->m_n_mem_sub_partition); 
@@ -410,6 +413,15 @@ void memory_sub_partition::cache_cycle( unsigned cycle )
                 bool read_sent = was_read_sent(events);
                 MEM_SUBPART_DPRINTF("Probing L2 cache Address=%llx, status=%u\n", mf->get_addr(), status); 
 
+                if ( (mf->get_access_type() == GLOBAL_ACC_R) ||
+                     (mf->get_access_type() == GLOBAL_ACC_W) ||
+                     (mf->get_access_type() == LOCAL_ACC_R)  ||
+                     (mf->get_access_type() == LOCAL_ACC_W)  ||
+                     (mf->get_access_type() == CONST_ACC_R) ) {
+                    if (!m_gpu->data_footprint_stats.count(mf->get_addr()))
+                        m_gpu->data_footprint_stats.insert(mf->get_addr());
+                }
+
                 if ( status == HIT ) {
                     if( !write_sent ) {
                         // L2 cache replies
diff --git a/src/gpgpu-sim/l2cache.h b/src/gpgpu-sim/l2cache.h
index 2d13918..6aa7505 100644
--- a/src/gpgpu-sim/l2cache.h
+++ b/src/gpgpu-sim/l2cache.h
@@ -58,7 +58,7 @@ private:
 class memory_partition_unit
 {
 public: 
-   memory_partition_unit( unsigned partition_id, const struct memory_config *config, class memory_stats_t *stats );
+   memory_partition_unit( unsigned partition_id, const struct memory_config *config, class memory_stats_t *stats , class gpgpu_sim *gpu);
    ~memory_partition_unit(); 
 
    bool busy() const;
@@ -93,6 +93,8 @@ public:
 
    unsigned get_mpid() const { return m_id; }
 
+   gpgpu_sim *m_gpu;
+
 private: 
 
    unsigned m_id;
@@ -145,7 +147,7 @@ private:
 class memory_sub_partition
 {
 public:
-   memory_sub_partition( unsigned sub_partition_id, const struct memory_config *config, class memory_stats_t *stats );
+   memory_sub_partition( unsigned sub_partition_id, const struct memory_config *config, class memory_stats_t *stats, class gpgpu_sim *gpu);
    ~memory_sub_partition(); 
 
    unsigned get_id() const { return m_id; } 
@@ -185,6 +187,8 @@ public:
         m_memcpy_cycle_offset += 1;
    }
 
+   gpgpu_sim *m_gpu;
+
 private:
 // data
    unsigned m_id;  //< the global sub partition ID