Adding the ability to querry the WARPSZ flag from the ptx script.

   Also changed some initialization code when cores are created in both the funcational and perfromance simulator
   review:3001 lgtm:5

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 16506]

1 files changed, 34 insertions, 6 deletions

diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 9a29bc3..883e122 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -29,7 +29,9 @@
 #define ABSTRACT_HARDWARE_MODEL_INCLUDED
 
 
-
+// Forward declarations
+class gpgpu_sim;
+class kernel_info_t;
 
 enum _memory_space_t {
    undefined_space=0,
@@ -964,23 +966,49 @@ size_t get_kernel_code_size( class function_info *entry );
  */
 class core_t {
     public:
-        virtual ~core_t() {}
+        core_t( gpgpu_sim *gpu, 
+                kernel_info_t *kernel,
+                unsigned warp_size,
+                unsigned threads_per_shader )
+            : m_gpu( gpu ),
+              m_kernel( kernel ),
+              m_simt_stack( NULL ),
+              m_thread( NULL ),
+              m_warp_size( warp_size )
+        {
+            m_warp_count = threads_per_shader/m_warp_size;
+            // Handle the case where the number of threads is not a
+            // multiple of the warp size
+            if ( threads_per_shader % m_warp_size != 0 ) {
+                m_warp_count += 1;
+            }
+            assert( m_warp_count * m_warp_size > 0 );
+            m_thread = ( ptx_thread_info** )
+                     calloc( m_warp_count * m_warp_size,
+                             sizeof( ptx_thread_info* ) );
+            initilizeSIMTStack(m_warp_count,m_warp_size);
+        }
+        virtual ~core_t() { free(m_thread); }
         virtual void warp_exit( unsigned warp_id ) = 0;
         virtual bool warp_waiting_at_barrier( unsigned warp_id ) const = 0;
         virtual void checkExecutionStatusAndUpdate(warp_inst_t &inst, unsigned t, unsigned tid)=0;
         class gpgpu_sim * get_gpu() {return m_gpu;}
-        void execute_warp_inst_t(warp_inst_t &inst, unsigned warpSize, unsigned warpId =(unsigned)-1);
+        void execute_warp_inst_t(warp_inst_t &inst, unsigned warpId =(unsigned)-1);
         bool  ptx_thread_done( unsigned hw_thread_id ) const ;
-        void updateSIMTStack(unsigned warpId, unsigned warpSize, warp_inst_t * inst);
-        void initilizeSIMTStack(unsigned warps, unsigned warpsSize);
+        void updateSIMTStack(unsigned warpId, warp_inst_t * inst);
+        void initilizeSIMTStack(unsigned warp_count, unsigned warps_size);
+        void deleteSIMTStack();
         warp_inst_t getExecuteWarp(unsigned warpId);
         void get_pdom_stack_top_info( unsigned warpId, unsigned *pc, unsigned *rpc ) const;
         kernel_info_t * get_kernel_info(){ return m_kernel;}
+        unsigned get_warp_size() const { return m_warp_size; }
     protected:
         class gpgpu_sim *m_gpu;
         kernel_info_t *m_kernel;
         simt_stack  **m_simt_stack; // pdom based reconvergence context for each warp
-        class ptx_thread_info ** m_thread; 
+        class ptx_thread_info ** m_thread;
+        unsigned m_warp_size;
+        unsigned m_warp_count;
 };