Cleaning up interconnection network memory partition to core statistics. Now counting the number of flits arriving at the cores instead of the number of flits leaving each memory partition.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 15746]

6 files changed, 18 insertions, 29 deletions

diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 07b06a8..0abe14a 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -824,19 +824,18 @@ void gpgpu_sim::gpu_print_stat()
    // Interconnect power stat print
    unsigned total_mem_to_simt=0;
    unsigned total_simt_to_mem=0;
-   for (unsigned i=0;i<m_memory_config->m_n_mem;i++){
-      unsigned temp=0;
-      m_memory_partition_unit[i]->set_icnt_power_stats(temp);
-      total_mem_to_simt += temp;
-   }
+
    for(unsigned i=0; i<m_config.num_cluster(); i++){
 	   unsigned temp=0;
-	   m_cluster[i]->set_icnt_stats(temp);
+	   unsigned temp2 = 0;
+	   m_cluster[i]->set_icnt_stats(temp, temp2);
 	   total_simt_to_mem += temp;
+	   total_mem_to_simt += temp2;
    }
    printf("\nicnt_total_pkts_mem_to_simt=%u\n", total_mem_to_simt);
    printf("icnt_total_pkts_simt_to_mem=%u\n\n", total_simt_to_mem);
 
+
    time_vector_print();
    fflush(stdout);
 
@@ -1053,8 +1052,6 @@ void gpgpu_sim::cycle()
           m_memory_partition_unit[i]->cache_cycle(gpu_sim_cycle+gpu_tot_sim_cycle);
           m_memory_partition_unit[i]->set_L2cache_power_stats(m_power_stats->pwr_mem_stat->n_l2_read_access[0][i], m_power_stats->pwr_mem_stat->n_l2_read_miss[0][i],
           m_power_stats->pwr_mem_stat->n_l2_write_access[0][i], m_power_stats->pwr_mem_stat->n_l2_write_miss[0][i]);
-
-          m_memory_partition_unit[i]->set_icnt_power_stats(m_power_stats->pwr_mem_stat->n_mem_to_simt[0][i]);
        }
    }
 
@@ -1070,7 +1067,8 @@ void gpgpu_sim::cycle()
                *active_sms+=m_cluster[i]->get_n_active_sms();
 
                // Interconnect power stats: SIMT->MEM
-               m_cluster[i]->set_icnt_stats(m_power_stats->pwr_mem_stat->n_simt_to_mem[0][i]);
+               m_cluster[i]->set_icnt_stats(m_power_stats->pwr_mem_stat->n_simt_to_mem[0][i],
+            		   m_power_stats->pwr_mem_stat->n_mem_to_simt[0][i]);
          }
       }
       float temp=0;
diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc
index 21f837d..0a3d56f 100644
--- a/src/gpgpu-sim/l2cache.cc
+++ b/src/gpgpu-sim/l2cache.cc
@@ -76,7 +76,6 @@ memory_partition_unit::memory_partition_unit( unsigned partition_id,
     if(!m_config->m_L2_config.disabled())
        m_L2cache = new l2_cache(L2c_name,m_config->m_L2_config,-1,-1,m_L2interface,m_mf_allocator,IN_PARTITION_L2_MISS_QUEUE);
 
-    n_mem_to_simt=0;
     unsigned int icnt_L2;
     unsigned int L2_dram;
     unsigned int dram_L2;
@@ -109,7 +108,6 @@ void memory_partition_unit::cache_cycle( unsigned cycle )
 				mf->set_reply();
 				mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle);
 				m_L2_icnt_queue->push(mf);
-				n_mem_to_simt+=mf->get_num_flits(false); // Interconnect power stats (# of flits sent to the SMs)
            }else{
 				m_request_tracker.erase(mf);
 				delete mf;
@@ -128,7 +126,6 @@ void memory_partition_unit::cache_cycle( unsigned cycle )
             mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle);
             m_L2_icnt_queue->push(mf);
             m_dram_L2_queue->pop();
-            n_mem_to_simt+=mf->get_num_flits(false); // Interconnect power stats (# of flits sent to the SMs)
         }
     }
 
@@ -160,7 +157,6 @@ void memory_partition_unit::cache_cycle( unsigned cycle )
                             mf->set_reply();
                             mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle);
                             m_L2_icnt_queue->push(mf);
-                            n_mem_to_simt+=mf->get_num_flits(false); // Interconnect power stats (# of flits sent to the SMs)
                         }
                         m_icnt_L2_queue->pop();
                     } else {
@@ -413,7 +409,3 @@ void memory_partition_unit::set_L2cache_power_stats(unsigned &n_read_access,
 												unsigned &n_write_miss) const{
 	m_L2cache->get_data_stats(n_read_access,n_read_miss,n_write_access,n_write_miss);
 }
-
-void memory_partition_unit::set_icnt_power_stats(unsigned &mem_to_simt) const{
-	mem_to_simt = n_mem_to_simt;
-}
diff --git a/src/gpgpu-sim/l2cache.h b/src/gpgpu-sim/l2cache.h
index 9317c18..31b9c4a 100644
--- a/src/gpgpu-sim/l2cache.h
+++ b/src/gpgpu-sim/l2cache.h
@@ -93,8 +93,6 @@ public:
 			unsigned &n_write_access,
 			unsigned &n_write_miss) const;
 
-   void set_icnt_power_stats(unsigned &n_mem_to_simt) const;
-
 private:
 // data
    unsigned m_id;
@@ -134,9 +132,6 @@ private:
    std::set<mem_fetch*> m_request_tracker;
 
    friend class L2interface;
-
-   // interconnect power stats
-   unsigned n_mem_to_simt;
 };
 
 class L2interface : public mem_fetch_interface {
diff --git a/src/gpgpu-sim/power_stat.cc b/src/gpgpu-sim/power_stat.cc
index 12c9574..eb4cb31 100644
--- a/src/gpgpu-sim/power_stat.cc
+++ b/src/gpgpu-sim/power_stat.cc
@@ -95,7 +95,8 @@ void power_mem_stat_t::init(){
 		n_req[i] = (unsigned *)calloc(m_config->m_n_mem,sizeof(unsigned));
 
 	    // Interconnect stats
-	    n_mem_to_simt[i] = (unsigned *)calloc(m_config->m_n_mem,sizeof(unsigned)); // Counted at memory partition
+	    //n_mem_to_simt[i] = (unsigned *)calloc(m_config->m_n_mem,sizeof(unsigned)); // Counted at memory partition
+		n_mem_to_simt[i] = (unsigned *)calloc(m_core_config->n_simt_clusters*m_core_config->n_simt_cores_per_cluster,sizeof(unsigned)); // Counted at SM
 	    n_simt_to_mem[i] = (unsigned *)calloc(m_core_config->n_simt_clusters*m_core_config->n_simt_cores_per_cluster,sizeof(unsigned)); // Counted at SM
     }
 }
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index d38f300..ba44c01 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -2621,8 +2621,9 @@ void shader_core_ctx::get_cache_stats(unsigned &read_accesses, unsigned &write_a
    m_ldst_unit->get_cache_stats(read_accesses, write_accesses, read_misses, write_misses, cache_type);
 }
 
-void shader_core_ctx::set_icnt_power_stats(unsigned &n_simt_to_mem) const{
+void shader_core_ctx::set_icnt_power_stats(unsigned &n_simt_to_mem, unsigned &n_mem_to_simt) const{
 	n_simt_to_mem += m_stats->n_simt_to_mem[m_sid];
+	n_mem_to_simt += m_stats->n_mem_to_simt[m_sid];
 }
 
 bool shd_warp_t::functional_done() const
@@ -3109,6 +3110,7 @@ void simt_core_cluster::icnt_cycle()
         mf->set_status(IN_CLUSTER_TO_SHADER_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle);
         //m_memory_stats->memlatstat_read_done(mf,m_shader_config->max_warps_per_shader);
         m_response_fifo.push_back(mf);
+        m_stats->n_mem_to_simt[m_config->sid_to_cid(mf->get_sid())] += mf->get_num_flits(false);
     }
 }
 
@@ -3143,9 +3145,9 @@ void simt_core_cluster::get_cache_stats(unsigned &read_accesses, unsigned &write
    }
 }
 
-void simt_core_cluster::set_icnt_stats(unsigned &n_simt_to_mem) const {
+void simt_core_cluster::set_icnt_stats(unsigned &n_simt_to_mem, unsigned &n_mem_to_simt) const {
 	for ( unsigned i = 0; i < m_config->n_simt_cores_per_cluster; ++i ) {
-		m_core[i]->set_icnt_power_stats(n_simt_to_mem);
+		m_core[i]->set_icnt_power_stats(n_simt_to_mem, n_mem_to_simt);
 	}
 }
 
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index c3420e6..ba63634 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1376,7 +1376,7 @@ struct shader_core_stats_pod {
     unsigned *l1d_write_miss;		// L1 Data cache write miss
 
     unsigned *n_simt_to_mem; // Interconnect power stats
-
+    unsigned *n_mem_to_simt;
 };
 
 class shader_core_stats : public shader_core_stats_pod {
@@ -1439,6 +1439,7 @@ public:
         l1d_write_miss = (unsigned *)calloc(config->num_shader(), sizeof(unsigned));
 
         n_simt_to_mem = (unsigned *)calloc(config->num_shader(), sizeof(unsigned));
+        n_mem_to_simt = (unsigned *)calloc(config->num_shader(), sizeof(unsigned));
 
         gpgpu_n_shmem_bank_access = (unsigned *)calloc(config->num_shader(), sizeof(unsigned));
 
@@ -1585,7 +1586,7 @@ public:
     void print_cache_stats( FILE *fp, unsigned& dl1_accesses, unsigned& dl1_misses );
     void get_cache_stats(unsigned &read_accesses, unsigned &write_accesses, unsigned &read_misses, unsigned &write_misses, unsigned cache_type);
 
-    void set_icnt_power_stats(unsigned &n_simt_to_mem) const;
+    void set_icnt_power_stats(unsigned &n_simt_to_mem, unsigned &n_mem_to_simt) const;
 
 // debug:
     void display_simt_state(FILE *fout, int mask ) const;
@@ -1829,7 +1830,7 @@ public:
     void print_cache_stats( FILE *fp, unsigned& dl1_accesses, unsigned& dl1_misses ) const;
     void get_cache_stats(unsigned &read_accesses, unsigned &write_accesses, unsigned &read_misses, unsigned &write_misses, unsigned cache_type) const;
 
-    void set_icnt_stats(unsigned &n_simt_to_mem) const;
+    void set_icnt_stats(unsigned &n_simt_to_mem, unsigned &n_mem_to_simt) const;
 
 private:
     unsigned m_cluster_id;