Revived PC-Histogram in AerialVision.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 11308]

3 files changed, 18 insertions, 1 deletions

diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index cc47da9..ddee6eb 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -514,7 +514,9 @@ void gpgpu_sim::init()
        m_cluster[i]->reinit();
     m_shader_stats->new_grid();
     // initialize the control-flow, memory access, memory latency logger
-    create_thread_CFlogger( m_config.num_shader(), m_shader_config->n_thread_per_shader, 0, m_config.gpgpu_cflog_interval );
+    if (m_config.g_visualizer_enabled) {
+        create_thread_CFlogger( m_config.num_shader(), m_shader_config->n_thread_per_shader, 0, m_config.gpgpu_cflog_interval );
+    }
     shader_CTA_count_create( m_config.num_shader(), m_config.gpgpu_cflog_interval);
     if (m_config.gpgpu_cflog_interval != 0) {
        insn_warp_occ_create( m_config.num_shader(), m_shader_config->warp_size );
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 3d37688..0275757 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -2204,4 +2204,14 @@ void shader_core_ctx::checkExecutionStatusAndUpdate(warp_inst_t &inst, unsigned
             m_warp[inst.warp_id()].set_completed(t);
             m_warp[inst.warp_id()].ibuffer_flush();
         }
+
+    // PC-Histogram Update 
+    unsigned warp_id = inst.warp_id(); 
+    unsigned pc = inst.pc; 
+    for (unsigned t = 0; t < m_config->warp_size; t++) {
+        if (inst.active(t)) {
+            int tid = warp_id * m_config->warp_size + t; 
+            cflog_update_thread_pc(m_sid, tid, pc);  
+        }
+    }
 }
diff --git a/src/gpgpu-sim/stat-tool.cc b/src/gpgpu-sim/stat-tool.cc
index 77e4aca..94998f2 100644
--- a/src/gpgpu-sim/stat-tool.cc
+++ b/src/gpgpu-sim/stat-tool.cc
@@ -149,10 +149,12 @@ void destroy_thread_CFlogger( )
 
 void cflog_update_thread_pc( int logger_id, int thread_id, address_type pc ) 
 {
+   if (thread_CFlogger == NULL) return;  // this means no visualizer output 
    if (thread_id < 0) return;
    thread_CFlogger[logger_id]->update_thread_pc(thread_id, pc);
 }
 
+// deprecated 
 void cflog_snapshot( int logger_id, unsigned long long  cycle ) 
 {
    thread_CFlogger[logger_id]->snap_shot(cycle);
@@ -160,6 +162,7 @@ void cflog_snapshot( int logger_id, unsigned long long  cycle )
 
 void cflog_print(FILE *fout) 
 {
+   if (thread_CFlogger == NULL) return;  // this means no visualizer output 
    for (int i = 0; i < n_thread_CFloggers; i++) {
       thread_CFlogger[i]->print_histo(fout);
    }
@@ -167,6 +170,7 @@ void cflog_print(FILE *fout)
 
 void cflog_visualizer_print(FILE *fout) 
 {
+   if (thread_CFlogger == NULL) return;  // this means no visualizer output 
    for (int i = 0; i < n_thread_CFloggers; i++) {
       thread_CFlogger[i]->print_visualizer(fout);
    }
@@ -174,6 +178,7 @@ void cflog_visualizer_print(FILE *fout)
 
 void cflog_visualizer_gzprint(gzFile fout) 
 {
+   if (thread_CFlogger == NULL) return;  // this means no visualizer output 
    for (int i = 0; i < n_thread_CFloggers; i++) {
       thread_CFlogger[i]->print_visualizer(fout);
    }