Fixed incorrect counting of predicated instruction. (Bug #15 external)

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 13918]

3 files changed, 10 insertions, 4 deletions

diff --git a/CHANGES b/CHANGES
index 4653061..1a05fbc 100644
--- a/CHANGES
+++ b/CHANGES
@@ -23,6 +23,9 @@ Version 3.1.1+edits (development branch) versus 3.1.1
       and store instructions.
     - Fixed a bug that was causing inconsistency in local memory address
       calculations
+    - Fixed incorrect counting of predicated instructions.  Now instructions
+      are counted into gpu_sim_insn regardless of predication outcome.  (Bug
+      #15 external)
 
 Version 3.1.1 versus 3.1.0
 - Add checks to top level makefile to ensure setup_environment is run and checks to
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index fda4763..766dff1 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -689,7 +689,8 @@ public:
     }
     void issue( const active_mask_t &mask, unsigned warp_id, unsigned long long cycle ) 
     {
-        m_warp_active_mask=mask;
+        m_warp_active_mask = mask;
+        m_warp_issued_mask = mask; 
         m_uid = ++sm_next_uid;
         m_warp_id = warp_id;
         issue_cycle = cycle;
@@ -763,6 +764,7 @@ public:
     }
     bool active( unsigned thread ) const { return m_warp_active_mask.test(thread); }
     unsigned active_count() const { return m_warp_active_mask.count(); }
+    unsigned issued_count() const { assert(m_empty == false); return m_warp_issued_mask.count(); }  // for instruction counting 
     bool empty() const { return m_empty; }
     unsigned warp_id() const 
     { 
@@ -810,7 +812,8 @@ protected:
     bool m_is_printf;
     unsigned m_warp_id;
     const core_config *m_config; 
-    active_mask_t m_warp_active_mask;
+    active_mask_t m_warp_active_mask; // dynamic active mask for timing model (after predication)
+    active_mask_t m_warp_issued_mask; // active mask at issue (prior to predication test) -- for instruction counting
 
     struct per_thread_info {
         per_thread_info() {
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index ddb544a..b3e5fc5 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -872,9 +872,9 @@ void shader_core_ctx::warp_inst_complete(const warp_inst_t &inst)
       printf("[warp_inst_complete] uid=%u core=%u warp=%u pc=%#x @ time=%llu issued@%llu\n", 
              inst.get_uid(), m_sid, inst.warp_id(), inst.pc, gpu_tot_sim_cycle + gpu_sim_cycle, inst.get_issue_cycle()); 
    #endif
-   m_stats->m_num_sim_insn[m_sid] += inst.active_count();
+   m_stats->m_num_sim_insn[m_sid] += inst.issued_count();
    m_stats->m_num_sim_winsn[m_sid]++;
-   m_gpu->gpu_sim_insn += inst.active_count();
+   m_gpu->gpu_sim_insn += inst.issued_count();
    inst.completed(gpu_tot_sim_cycle + gpu_sim_cycle); 
 }