simulator "working" on ptxplus, simulated IPC "too low" by about a factor of up to 50%

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 7816]

6 files changed, 23 insertions, 14 deletions

diff --git a/configs/QuadroFX5800/gpgpusim.config b/configs/QuadroFX5800/gpgpusim.config
index 560796d..c0fb71b 100644
--- a/configs/QuadroFX5800/gpgpusim.config
+++ b/configs/QuadroFX5800/gpgpusim.config
@@ -13,7 +13,6 @@
 -gpgpu_shader_registers 16384
 -gpgpu_shader_core_pipeline 1024:32:32 
 -gpgpu_shader_cta 8
--gpgpu_pre_mem_stages 0
 -gpgpu_pdom_sched_type 8
 -gpgpu_simd_model 1 
 
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 02d58c2..15a7296 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -315,6 +315,7 @@ public:
             fprintf(fp, "%c", ((warp_active_mask[i])?'1':'0') );
     }
     bool active( unsigned thread ) const { return warp_active_mask.test(thread); }
+    unsigned active_count() const { return warp_active_mask.count(); }
     bool empty() const { return m_empty; }
     unsigned warp_id() const 
     { 
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 7fff9fd..96d5c62 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -113,13 +113,7 @@ unsigned freed_L1write_mfs = 0;
 unsigned freed_L2write_mfs = 0;
 unsigned freed_dummy_read_mfs = 0;
 unsigned long long  gpu_sim_cycle = 0;
-unsigned long long  gpu_sim_insn = 0;
-unsigned long long  gpu_sim_prev_insn = 0;
 unsigned long long  gpu_tot_sim_cycle = 0;
-unsigned long long  gpu_tot_sim_insn = 0;
-unsigned long long  gpu_last_sim_cycle = 0;
-unsigned long long  gpu_tot_issued_cta = 0;
-unsigned long long  gpu_tot_completed_thread = 0;
 
 unsigned int **concurrent_row_access; //concurrent_row_access[dram chip id][bank id]
 unsigned int **num_activates; //num_activates[dram chip id][bank id]
@@ -143,16 +137,13 @@ int gpu_runtime_stat_flag = 0;
 
 unsigned long long  gpu_max_cycle = 0;
 unsigned long long  gpu_max_insn = 0;
-int gpu_deadlock = 0;
 unsigned g_next_mf_request_uid = 1;
-static unsigned long long  last_gpu_sim_insn = 0;
 int g_total_cta_left;
 
 // GPGPU-Sim timing model options
 int   gpu_max_cycle_opt;
 int   gpu_max_insn_opt;
 int   gpu_max_cta_opt;
-bool  gpu_deadlock_detect;
 char *gpgpu_shader_core_pipeline_opt;
 int   gpgpu_dram_sched_queue_size; 
 bool  gpgpu_flush_cache;
@@ -466,6 +457,11 @@ gpgpu_sim::gpgpu_sim()
    m_shader_stats = (shader_core_stats*)calloc(1,sizeof(shader_core_stats));
    m_memory_config = (memory_config*)calloc(1,sizeof(memory_config));
    m_memory_stats = NULL;
+   gpu_sim_insn = 0;
+   gpu_tot_sim_insn = 0;
+   gpu_tot_issued_cta = 0;
+   gpu_tot_completed_thread = 0;
+   gpu_deadlock = false;
 }
 
 void set_ptx_warp_size(unsigned warp_size);
@@ -1397,7 +1393,7 @@ void gpgpu_sim::cycle()
       if (!(gpu_sim_cycle % 20000)) {
          // deadlock detection 
          if (gpu_deadlock_detect && gpu_sim_insn == last_gpu_sim_insn) {
-            gpu_deadlock = 1;
+            gpu_deadlock = true;
          } else {
             last_gpu_sim_insn = gpu_sim_insn;
          }
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index fab1ab8..ac03cbc 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -269,9 +269,21 @@ private:
 
    int m_pdom_sched_type;
 
+   // options
+   bool gpu_deadlock_detect;
+
+   // stats
    struct shader_core_stats  *m_shader_stats;
    class memory_stats_t      *m_memory_stats;
+   unsigned long long  gpu_tot_issued_cta;
+   unsigned long long  gpu_tot_completed_thread;
+   unsigned long long  last_gpu_sim_insn;
+
+   // debug
+   bool gpu_deadlock;
 public:
+   unsigned long long  gpu_sim_insn;
+   unsigned long long  gpu_tot_sim_insn;
    unsigned long long  gpu_sim_insn_last_update;
    unsigned gpu_sim_insn_last_update_sid;
 };
@@ -280,8 +292,6 @@ public:
 
 extern unsigned long long  gpu_sim_cycle;
 extern unsigned long long  gpu_tot_sim_cycle;
-extern unsigned long long  gpu_sim_insn;
-extern unsigned long long  gpu_tot_sim_insn;
 extern unsigned g_next_mf_request_uid;
 
 // stats 
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 0d17be4..061952a 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -1194,6 +1194,9 @@ void shader_core_ctx::writeback()
     if( !m_pipeline_reg[MM_WB]->empty() ) {
         m_scoreboard->releaseRegisters( m_pipeline_reg[MM_WB] );
         m_warp[m_pipeline_reg[MM_WB]->warp_id()].dec_inst_in_pipeline();
+        m_gpu->gpu_sim_insn_last_update_sid = m_sid;
+        m_gpu->gpu_sim_insn_last_update = gpu_sim_cycle;
+        m_gpu->gpu_sim_insn += m_pipeline_reg[MM_WB]->active_count();
     }
     move_warp(m_pipeline_reg[WB_RT],m_pipeline_reg[MM_WB]);
 }
diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc
index 4f5cffe..7947a8f 100644
--- a/src/gpgpusim_entrypoint.cc
+++ b/src/gpgpusim_entrypoint.cc
@@ -151,7 +151,7 @@ void print_simulation_time()
    fflush(stderr);
    printf("\n\ngpgpu_simulation_time = %u days, %u hrs, %u min, %u sec (%u sec)\n",
           (unsigned)d, (unsigned)h, (unsigned)m, (unsigned)s, (unsigned)difference );
-   printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(gpu_tot_sim_insn / difference) );
+   printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(g_the_gpu.gpu_tot_sim_insn / difference) );
    printf("gpgpu_simulation_rate = %u (cycle/sec)\n", (unsigned)(gpu_tot_sim_cycle / difference) );
    fflush(stdout);
 }