1 files changed, 38 insertions, 4 deletions

diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 1aa468b..bdd8dbe 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -108,6 +108,10 @@ public:
         m_last_fetch=0;
         m_next=0;
         m_inst_at_barrier=NULL;
+
+        //Jin: cdp support
+        m_cdp_latency = 0;
+        m_cdp_dummy = false;
     }
     void init( address_type start_pc,
                unsigned cta_id,
@@ -124,6 +128,10 @@ public:
         n_completed   -= active.count(); // active threads are not yet completed
         m_active_threads = active;
         m_done_exit=false;
+
+        //Jin: cdp support
+        m_cdp_latency = 0;
+        m_cdp_dummy = false;
     }
 
     bool functional_done() const;
@@ -260,6 +268,11 @@ private:
 
     unsigned m_stores_outstanding; // number of store requests sent but not yet acknowledged
     unsigned m_inst_in_pipeline;
+
+    //Jin: cdp support
+public:
+    unsigned int m_cdp_latency;
+    bool m_cdp_dummy;
 };
 
 
@@ -267,7 +280,7 @@ private:
 inline unsigned hw_tid_from_wid(unsigned wid, unsigned warp_size, unsigned i){return wid * warp_size + i;};
 inline unsigned wid_from_hw_tid(unsigned tid, unsigned warp_size){return tid/warp_size;};
 
-const unsigned WARP_PER_CTA_MAX = 48;
+const unsigned WARP_PER_CTA_MAX = 64;
 typedef std::bitset<WARP_PER_CTA_MAX> warp_set_t;
 
 int register_bank(int regnum, int wid, unsigned num_banks, unsigned bank_warp_shift);
@@ -1327,11 +1340,14 @@ struct shader_core_config : public core_config
     int simt_core_sim_order; 
     
     unsigned mem2device(unsigned memid) const { return memid + n_simt_clusters; }
+
+    //Jin: concurrent kernel on sm
+    bool gpgpu_concurrent_kernel_sm;
 };
 
 struct shader_core_stats_pod {
 
-	void* shader_core_stats_pod_start[0]; // DO NOT MOVE FROM THE TOP - spaceless pointer to the start of this structure
+	void* shader_core_stats_pod_start[]; // DO NOT MOVE FROM THE TOP - spaceless pointer to the start of this structure
 	unsigned long long *shader_cycles;
     unsigned *m_num_sim_insn; // number of scalar thread instructions committed by this shader core
     unsigned *m_num_sim_winsn; // number of warp instructions committed by this shader core
@@ -1574,6 +1590,7 @@ public:
     void cycle();
     void reinit(unsigned start_thread, unsigned end_thread, bool reset_not_completed );
     void issue_block2core( class kernel_info_t &kernel );
+
     void cache_flush();
     void accept_fetch_response( mem_fetch *mf );
     void accept_ldst_unit_response( class mem_fetch * mf );
@@ -1582,7 +1599,7 @@ public:
     {
         assert(k);
         m_kernel=k; 
-        k->inc_running(); 
+//        k->inc_running(); 
         printf("GPGPU-Sim uArch: Shader %d bind to kernel %u \'%s\'\n", m_sid, m_kernel->get_uid(),
                  m_kernel->name().c_str() );
     }
@@ -1749,7 +1766,7 @@ public:
     virtual void checkExecutionStatusAndUpdate(warp_inst_t &inst, unsigned t, unsigned tid);
     address_type next_pc( int tid ) const;
     void fetch();
-    void register_cta_thread_exit( unsigned cta_num );
+    void register_cta_thread_exit(unsigned cta_num, kernel_info_t * kernel );
 
     void decode();
     
@@ -1831,6 +1848,22 @@ public:
     // is that the dynamic_warp_id is a running number unique to every warp
     // run on this shader, where the warp_id is the static warp slot.
     unsigned m_dynamic_warp_id;
+
+    //Jin: concurrent kernels on a sm
+public:
+    bool can_issue_1block(kernel_info_t & kernel);
+    bool occupy_shader_resource_1block(kernel_info_t & kernel, bool occupy);
+    void release_shader_resource_1block(unsigned hw_ctaid, kernel_info_t & kernel);
+    int find_available_hwtid(unsigned int cta_size, bool occupy);
+private:
+    unsigned int m_occupied_n_threads; 
+    unsigned int m_occupied_shmem; 
+    unsigned int m_occupied_regs;
+    unsigned int m_occupied_ctas;
+    std::bitset<MAX_THREAD_PER_SM> m_occupied_hwtid;
+    std::map<unsigned int, unsigned int> m_occupied_cta_to_hwtid; 
+
+
 };
 
 class simt_core_cluster {
@@ -1851,6 +1884,7 @@ public:
     bool icnt_injection_buffer_full(unsigned size, bool write);
     void icnt_inject_request_packet(class mem_fetch *mf);
 
+
     // for perfect memory interface
     bool response_queue_full() {
         return ( m_response_fifo.size() >= m_config->n_simt_ejection_buffer_size );