better oc selecting when sub core enabled

2 files changed, 43 insertions, 5 deletions

diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 7cee40f..bcfda18 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -3997,6 +3997,9 @@ void opndcoll_rfu_t::init(unsigned num_banks, shader_core_ctx *shader) {
     m_cu[j]->init(j, num_banks, m_bank_warp_shift, shader->get_config(), this,
                   sub_core_model, reg_id, m_num_banks_per_sched);
   }
+  for (unsigned j = 0; j < m_dispatch_units.size(); j++) {
+    m_dispatch_units[j].init(sub_core_model,m_num_warp_scheds);
+  }
   m_initialized = true;
 }
 
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 9cb256a..f2fac12 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -950,13 +950,44 @@ class opndcoll_rfu_t {  // operand collector based register file unit
       m_num_collectors = (*cus).size();
       m_next_cu = 0;
     }
+    void init(bool sub_core_model, unsigned num_warp_scheds) {
+      m_sub_core_model = sub_core_model;
+      m_num_warp_scheds = num_warp_scheds;
+      if (m_sub_core_model) {
+        m_last_cu_set = new unsigned(m_num_warp_scheds);
+        for (unsigned i = 0; i < m_num_warp_scheds; i++)
+        {
+          m_last_cu_set[i] = i * m_num_collectors / m_num_warp_scheds;
+        }
+      }
+      
+    }
 
     collector_unit_t *find_ready() {
-      for (unsigned n = 0; n < m_num_collectors; n++) {
-        unsigned c = (m_last_cu + n + 1) % m_num_collectors;
-        if ((*m_collector_units)[c].ready()) {
-          m_last_cu = c;
-          return &((*m_collector_units)[c]);
+      if (m_sub_core_model) {
+        assert(m_num_collectors % m_num_warp_scheds == 0 &&
+                 m_num_collectors >= m_num_warp_scheds);
+        unsigned cusPerSched = m_num_collectors / m_num_warp_scheds;
+        for (unsigned i = 0; i < m_num_warp_scheds; i++) {
+          unsigned cuLowerBound = i * cusPerSched;
+          unsigned cuUpperBound = cuLowerBound + cusPerSched;
+          assert(0 <= cuLowerBound && cuUpperBound <= m_num_collectors);
+          assert(cuLowerBound <= m_last_cu_set[i] && m_last_cu_set[i] <= cuUpperBound);
+          for (unsigned j = cuLowerBound; j < cuUpperBound; j++) {
+            unsigned c = cuLowerBound + (m_last_cu_set[i] + j + 1) % cusPerSched;
+            if ((*m_collector_units)[c].ready()) {
+            m_last_cu_set[i] = c;
+            return &((*m_collector_units)[c]);
+            }
+          }
+        }
+      } else {
+        for (unsigned n = 0; n < m_num_collectors; n++) {
+          unsigned c = (m_last_cu + n + 1) % m_num_collectors;
+          if ((*m_collector_units)[c].ready()) {
+            m_last_cu = c;
+            return &((*m_collector_units)[c]);
+          }
         }
       }
       return NULL;
@@ -966,7 +997,11 @@ class opndcoll_rfu_t {  // operand collector based register file unit
     unsigned m_num_collectors;
     std::vector<collector_unit_t> *m_collector_units;
     unsigned m_last_cu;  // dispatch ready cu's rr
+    unsigned *m_last_cu_set;
     unsigned m_next_cu;  // for initialization
+
+    bool m_sub_core_model;
+    unsigned m_num_warp_scheds;
   };
 
   // opndcoll_rfu_t data members