move reg_id calc to cu object init

2 files changed, 19 insertions, 15 deletions

diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index c1bc495..7247616 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -3868,14 +3868,21 @@ void opndcoll_rfu_t::init(unsigned num_banks, shader_core_ctx *shader) {
 
   sub_core_model = shader->get_config()->sub_core_model;
   m_num_warp_scheds = shader->get_config()->gpgpu_num_sched_per_core;
-  if (sub_core_model)
+  unsigned reg_id;
+  if (sub_core_model) {
     assert(num_banks % shader->get_config()->gpgpu_num_sched_per_core == 0);
+    assert(m_num_warp_scheds >= m_cu.size() && m_cu.size() % m_num_warp_scheds == 0);
+  }
   m_num_banks_per_sched =
       num_banks / shader->get_config()->gpgpu_num_sched_per_core;
 
   for (unsigned j = 0; j < m_cu.size(); j++) {
+    if (sub_core_model) {
+      unsigned cusPerSched = m_cu.size() / m_num_warp_scheds;
+      reg_id = j / cusPerSched;
+    }
     m_cu[j]->init(j, num_banks, m_bank_warp_shift, shader->get_config(), this,
-                  sub_core_model, m_num_banks_per_sched);
+                  sub_core_model, reg_id, m_num_banks_per_sched);
   }
   m_initialized = true;
 }
@@ -3962,10 +3969,8 @@ void opndcoll_rfu_t::dispatch_ready_cu() {
         }
       }
       unsigned reg_id;
-      if (sub_core_model) {
-        unsigned cusPerSched = du.get_num_collectors() / m_num_warp_scheds;
-        reg_id = cu->get_id() / cusPerSched;
-      }
+      if (sub_core_model) 
+        reg_id = cu->get_reg_id();
       cu->dispatch(sub_core_model, reg_id);
     }
   }
@@ -4074,6 +4079,7 @@ void opndcoll_rfu_t::collector_unit_t::init(unsigned n, unsigned num_banks,
                                             const core_config *config,
                                             opndcoll_rfu_t *rfu,
                                             bool sub_core_model,
+                                            unsigned reg_id,
                                             unsigned banks_per_sched) {
   m_rfu = rfu;
   m_cuid = n;
@@ -4082,6 +4088,7 @@ void opndcoll_rfu_t::collector_unit_t::init(unsigned n, unsigned num_banks,
   m_warp = new warp_inst_t(config);
   m_bank_warp_shift = log2_warp_size;
   m_sub_core_model = sub_core_model;
+  m_reg_id = reg_id;
   m_num_banks_per_sched = banks_per_sched;
 }
 
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 0b96ec0..a5a8166 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -867,7 +867,7 @@ class opndcoll_rfu_t {  // operand collector based register file unit
       m_bank_warp_shift = 0;
     }
     // accessors
-    bool ready(bool sub_core_modle, unsigned reg_id) const;
+    bool ready(bool sub_core_model, unsigned reg_id) const;
     const op_t *get_operands() const { return m_src_op; }
     void dump(FILE *fp, const shader_core_ctx *shader) const;
 
@@ -878,11 +878,12 @@ class opndcoll_rfu_t {  // operand collector based register file unit
     }
     unsigned get_sp_op() const { return m_warp->sp_op; }
     unsigned get_id() const { return m_cuid; }  // returns CU hw id
+    unsigned get_reg_id() const { return m_reg_id; }
 
     // modifiers
     void init(unsigned n, unsigned num_banks, unsigned log2_warp_size,
               const core_config *config, opndcoll_rfu_t *rfu,
-              bool m_sub_core_model, unsigned num_banks_per_sched);
+              bool m_sub_core_model, unsigned reg_id, unsigned num_banks_per_sched);
     bool allocate(register_set *pipeline_reg, register_set *output_reg);
 
     void collect_operand(unsigned op) { m_not_ready.reset(op); }
@@ -906,6 +907,7 @@ class opndcoll_rfu_t {  // operand collector based register file unit
 
     unsigned m_num_banks_per_sched;
     bool m_sub_core_model;
+    unsigned m_reg_id; // if sub_core_model enabled, limit regs this cu can r/w
   };
 
   class dispatch_unit_t {
@@ -921,11 +923,8 @@ class opndcoll_rfu_t {  // operand collector based register file unit
       for (unsigned n = 0; n < m_num_collectors; n++) {
         unsigned c = (m_last_cu + n + 1) % m_num_collectors;
         unsigned reg_id;
-        if (sub_core_model) {
-          assert (m_num_collectors >= m_num_warp_scheds);
-          unsigned cusPerSched = m_num_collectors / m_num_warp_scheds;
-          reg_id = c / cusPerSched;
-        }
+        if (sub_core_model)
+          reg_id = (*m_collector_units)[c].get_reg_id();
         if ((*m_collector_units)[c].ready(sub_core_model, reg_id)) {
           m_last_cu = c;
           return &((*m_collector_units)[c]);
@@ -934,8 +933,6 @@ class opndcoll_rfu_t {  // operand collector based register file unit
       return NULL;
     }
 
-    unsigned get_num_collectors(){return m_num_collectors;}
-
    private:
     unsigned m_num_collectors;
     std::vector<collector_unit_t> *m_collector_units;