A much easier way to attempt to fix the problem targeted by CL12362 is to just print the stats whenever a kernel is done.

This requires decoupling updating the stats from printing them and modifying the printing code to accomodate this change.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 12366]

2 files changed, 9 insertions, 5 deletions

diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index c96ed9e..ee8f99a 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -554,11 +554,14 @@ void gpgpu_sim::init()
        icnt_init_grid(); 
 }
 
-void gpgpu_sim::print_stats()
-{
+void gpgpu_sim::update_stats() {
     m_memory_stats->memlatstat_lat_pw();
     gpu_tot_sim_cycle += gpu_sim_cycle;
     gpu_tot_sim_insn += gpu_sim_insn;
+}
+
+void gpgpu_sim::print_stats()
+{
 
     ptx_file_line_stats_write_file();
     gpu_print_stat();
@@ -616,9 +619,9 @@ void gpgpu_sim::gpu_print_stat() const
    printf("gpu_sim_cycle = %lld\n", gpu_sim_cycle);
    printf("gpu_sim_insn = %lld\n", gpu_sim_insn);
    printf("gpu_ipc = %12.4f\n", (float)gpu_sim_insn / gpu_sim_cycle);
-   printf("gpu_tot_sim_cycle = %lld\n", gpu_tot_sim_cycle);
-   printf("gpu_tot_sim_insn = %lld\n", gpu_tot_sim_insn);
-   printf("gpu_tot_ipc = %12.4f\n", (float)gpu_tot_sim_insn / gpu_tot_sim_cycle);
+   printf("gpu_tot_sim_cycle = %lld\n", gpu_tot_sim_cycle+gpu_sim_cycle);
+   printf("gpu_tot_sim_insn = %lld\n", gpu_tot_sim_insn+gpu_sim_insn);
+   printf("gpu_tot_ipc = %12.4f\n", (float)(gpu_tot_sim_insn+gpu_sim_cycle) / (gpu_tot_sim_cycle+gpu_sim_cycle));
    printf("gpu_tot_issued_cta = %lld\n", gpu_tot_issued_cta);
 
    // performance counter for stalls due to congestion.
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 208cfb8..e77f859 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -237,6 +237,7 @@ public:
    void cycle();
    bool active(); 
    void print_stats();
+   void update_stats();
    void deadlock_check();
 
    void get_pdom_stack_top_info( unsigned sid, unsigned tid, unsigned *pc, unsigned *rpc );