Merge branch 'dev' of https://github.com/purdue-aalp/gpgpu-sim_distribution into dev-traces

23 files changed, 213 insertions, 254 deletions

diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index 9aa1d73..e8e4b51 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -808,14 +808,14 @@ void kernel_info_t::notify_parent_finished() {
    if(m_parent_kernel) {
        m_kernel_entry->gpgpu_ctx->device_runtime->g_total_param_size -= ((m_kernel_entry->get_args_aligned_size() + 255)/256*256);
        m_parent_kernel->remove_child(this);
-       g_stream_manager()->register_finished_kernel(m_parent_kernel->get_uid());
+       m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->register_finished_kernel(m_parent_kernel->get_uid());
    }
 }
 
 CUstream_st * kernel_info_t::create_stream_cta(dim3 ctaid) {
     assert(get_default_stream_cta(ctaid));
     CUstream_st * stream = new CUstream_st();
-    g_stream_manager()->add_stream(stream);
+    m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->add_stream(stream);
     assert(m_cta_streams.find(ctaid) != m_cta_streams.end());
     assert(m_cta_streams[ctaid].size() >= 1); //must have default stream
     m_cta_streams[ctaid].push_back(stream);
@@ -831,7 +831,7 @@ CUstream_st * kernel_info_t::get_default_stream_cta(dim3 ctaid) {
     else {
       m_cta_streams[ctaid] = std::list<CUstream_st *>();
       CUstream_st * stream = new CUstream_st();
-      g_stream_manager()->add_stream(stream);
+      m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->add_stream(stream);
       m_cta_streams[ctaid].push_back(stream);
       return stream;
     }
@@ -863,7 +863,7 @@ void kernel_info_t::destroy_cta_streams() {
      for(auto s = m_cta_streams.begin(); s != m_cta_streams.end(); s++) {
         stream_size += s->second.size();
         for(auto ss = s->second.begin(); ss != s->second.end(); ss++)
-        g_stream_manager()->destroy_stream(*ss);
+        m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->destroy_stream(*ss);
         s->second.clear();
      }
      printf("size %lu\n", stream_size);
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 60f7f68..341e44c 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -447,19 +447,25 @@ protected:
     class gpgpu_sim * m_gpu;
 };
 
-#define GLOBAL_HEAP_START 0xC0000000
-   // start allocating from this address (lower values used for allocating globals in .ptx file)
-#define SHARED_MEM_SIZE_MAX (96*1024)
-#define LOCAL_MEM_SIZE_MAX (16*1024)
-#define MAX_STREAMING_MULTIPROCESSORS 80   //scale it to Volta
-#define MAX_THREAD_PER_SM 2048
-#define MAX_WARP_PER_SM 64
-#define TOTAL_LOCAL_MEM_PER_SM (MAX_THREAD_PER_SM*LOCAL_MEM_SIZE_MAX)
-#define TOTAL_SHARED_MEM (MAX_STREAMING_MULTIPROCESSORS*SHARED_MEM_SIZE_MAX)
-#define TOTAL_LOCAL_MEM (MAX_STREAMING_MULTIPROCESSORS*MAX_THREAD_PER_SM*LOCAL_MEM_SIZE_MAX)
-#define SHARED_GENERIC_START (GLOBAL_HEAP_START-TOTAL_SHARED_MEM)
-#define LOCAL_GENERIC_START (SHARED_GENERIC_START-TOTAL_LOCAL_MEM)
-#define STATIC_ALLOC_LIMIT (GLOBAL_HEAP_START - (TOTAL_LOCAL_MEM+TOTAL_SHARED_MEM))
+// Let's just upgrade to C++11 so we can use constexpr here...
+// start allocating from this address (lower values used for allocating globals in .ptx file)
+const unsigned long long GLOBAL_HEAP_START = 0xC0000000;
+// Volta max shmem size is 96kB
+const unsigned long long SHARED_MEM_SIZE_MAX = 96 * (1 << 10);
+// Volta max local mem is 16kB
+const unsigned long long LOCAL_MEM_SIZE_MAX = 1 << 14;
+// Volta Titan V has 80 SMs
+const unsigned MAX_STREAMING_MULTIPROCESSORS = 80;
+// Max 2048 threads / SM
+const unsigned MAX_THREAD_PER_SM = 1 << 11;
+// MAX 64 warps / SM
+const unsigned MAX_WARP_PER_SM = 1 << 6;
+const unsigned long long TOTAL_LOCAL_MEM_PER_SM = MAX_THREAD_PER_SM * LOCAL_MEM_SIZE_MAX;
+const unsigned long long TOTAL_SHARED_MEM = MAX_STREAMING_MULTIPROCESSORS * SHARED_MEM_SIZE_MAX;
+const unsigned long long TOTAL_LOCAL_MEM = MAX_STREAMING_MULTIPROCESSORS * MAX_THREAD_PER_SM * LOCAL_MEM_SIZE_MAX;
+const unsigned long long SHARED_GENERIC_START = GLOBAL_HEAP_START - TOTAL_SHARED_MEM;
+const unsigned long long LOCAL_GENERIC_START = SHARED_GENERIC_START - TOTAL_LOCAL_MEM;
+const unsigned long long STATIC_ALLOC_LIMIT = GLOBAL_HEAP_START - (TOTAL_LOCAL_MEM + TOTAL_SHARED_MEM);
 
 #if !defined(__CUDA_RUNTIME_API_H__)
 
@@ -525,10 +531,10 @@ private:
     int checkpoint_option;
     int checkpoint_kernel;
     int checkpoint_CTA;
-    int resume_option;
-    int resume_kernel;
-    int resume_CTA;
-    int checkpoint_CTA_t;
+    unsigned resume_option;
+    unsigned resume_kernel;
+    unsigned resume_CTA;
+    unsigned checkpoint_CTA_t;
     int checkpoint_insn_Y;
     int   g_ptx_inst_debug_to_file;
     char* g_ptx_inst_debug_file;
@@ -546,10 +552,10 @@ public:
     int checkpoint_option;
     int checkpoint_kernel;
     int checkpoint_CTA;
-    int resume_option;
-    int resume_kernel;
-    int resume_CTA;
-    int checkpoint_CTA_t;
+    unsigned resume_option;
+    unsigned resume_kernel;
+    unsigned resume_CTA;
+    unsigned checkpoint_CTA_t;
     int checkpoint_insn_Y;
 
     //Move some cycle core stats here instead of being global
@@ -998,7 +1004,7 @@ public:
 			printf("Printing mem access generated\n");
 			std::list<mem_access_t>::iterator it;	
 			for (it = m_accessq.begin(); it != m_accessq.end(); ++it){
-   				 printf("MEM_TXN_GEN:%s:%x, Size:%d \n",mem_access_type_str(it->get_type()), it->get_addr(),it->get_size());
+   				 printf("MEM_TXN_GEN:%s:%llx, Size:%d \n",mem_access_type_str(it->get_type()), it->get_addr(),it->get_size());
 			}	
 		}
     }   
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 1b0e841..7a130ea 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -429,7 +429,7 @@ void gpgpu_t::memcpy_to_gpu( size_t dst_start_addr, const void *src, size_t coun
 
    // Copy into the performance model.
    //extern gpgpu_sim* g_the_gpu;
-   g_the_gpu()->perf_memcpy_to_gpu(dst_start_addr, count);
+   gpgpu_ctx->the_gpgpusim->g_the_gpu->perf_memcpy_to_gpu(dst_start_addr, count);
    if(g_debug_execution >= 3) {
       printf( " done.\n");
       fflush(stdout);
@@ -448,7 +448,7 @@ void gpgpu_t::memcpy_from_gpu( void *dst, size_t src_start_addr, size_t count )
 
    // Copy into the performance model.
    //extern gpgpu_sim* g_the_gpu;
-   g_the_gpu()->perf_memcpy_to_gpu(src_start_addr, count);
+   gpgpu_ctx->the_gpgpusim->g_the_gpu->perf_memcpy_to_gpu(src_start_addr, count);
    if(g_debug_execution >= 3) {
       printf( " done.\n");
       fflush(stdout);
@@ -1254,7 +1254,7 @@ void function_info::param_to_shared( memory_space *shared_mem, symbol_table *sym
 {
    // TODO: call this only for PTXPlus with GT200 models 
    //extern gpgpu_sim* g_the_gpu;
-   if (not g_the_gpu()->get_config().convert_to_ptxplus()) return;
+   if (not gpgpu_ctx->the_gpgpusim->g_the_gpu->get_config().convert_to_ptxplus()) return;
 
    // copies parameters into simulated shared memory
    for( std::map<unsigned,param_info>::iterator i=m_ptx_kernel_param_info.begin(); i!=m_ptx_kernel_param_info.end(); i++ ) {
@@ -1309,7 +1309,7 @@ void function_info::ptx_jit_config(std::map<unsigned long long, size_t> mallocPt
     char buff[1024];
     std::string filename_c(filename+"_c");
     snprintf(buff,1024,"c++filt %s > %s", get_name().c_str(), filename_c.c_str());
-    system(buff);
+    assert(system(buff) != NULL);
     FILE *fp = fopen(filename_c.c_str(), "r");
     fgets(buff, 1024, fp);
     fclose(fp);
@@ -1432,13 +1432,13 @@ void function_info::ptx_jit_config(std::map<unsigned long long, size_t> mallocPt
     fout = fopen(ptx_config_fn.c_str(), "a");
     assert(fout!=NULL);
     for (unsigned i = 0; i<line_number; i++){
-        fgets(buff, 1024, fin);
+        assert(fgets(buff, 1024, fin) != NULL);
         assert(!feof(fin));
     }
     fprintf(fout, "\n\n");
     do{
         fprintf(fout, "%s", buff);
-        fgets(buff, 1024, fin);
+        assert(fgets(buff, 1024, fin) != NULL);
         if(feof(fin)){
             break;
         }
@@ -1491,7 +1491,6 @@ static unsigned get_tex_datasize( const ptx_instruction *pI, ptx_thread_info *th
    const operand_info &src1 = pI->src1(); //the name of the texture
    std::string texname = src1.name();
 
-   gpgpu_t *gpu = thread->get_gpu();
    /*
      For programs with many streams, textures can be bound and unbound
      asynchronously.  This means we need to use the kernel's "snapshot" of
@@ -1577,7 +1576,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id)
       }
       
       //Tensorcore is warp synchronous operation. So these instructions needs to be executed only once. To make the simulation faster removing the redundant tensorcore operation
-      if(!tensorcore_op(inst_opcode)||(tensorcore_op(inst_opcode))&&(lane_id==0)){
+      if(!tensorcore_op(inst_opcode)||((tensorcore_op(inst_opcode))&&(lane_id==0))){
 	      switch ( inst_opcode ) {
 	#define OP_DEF(OP,FUNC,STR,DST,CLASSIFICATION) case OP: FUNC(pI,this); op_classification = CLASSIFICATION; break;
 	#define OP_W_DEF(OP,FUNC,STR,DST,CLASSIFICATION) case OP: FUNC(pI,get_core(),inst); op_classification = CLASSIFICATION; break;
@@ -2138,19 +2137,13 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
 	 kernel_func_info->set_pdom();
     }
 
-    unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu()->getShaderCoreConfig()->warp_size, g_the_gpu()->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu()->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu()->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu()->getShaderCoreConfig()->max_cta_per_core);
+    unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->warp_size, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->n_thread_per_shader, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->gpgpu_shmem_size, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->gpgpu_shader_registers, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->max_cta_per_core);
     printf("Max CTA : %d\n",max_cta_tot);
 
-    
-
-
-      
-    int inst_count=50;
-    int cp_op= g_the_gpu()->checkpoint_option;
-    int cp_CTA = g_the_gpu()->checkpoint_CTA;
-    int cp_kernel= g_the_gpu()->checkpoint_kernel;
-    cp_count= g_the_gpu()->checkpoint_insn_Y;
-    cp_cta_resume= g_the_gpu()->checkpoint_CTA_t;
+    int cp_op= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_option;
+    int cp_kernel= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_kernel;
+    cp_count= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_insn_Y;
+    cp_cta_resume= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_CTA_t;
     int cta_launched =0;
 
     //we excute the kernel one CTA (Block) at the time, as synchronization functions work block wise
@@ -2162,8 +2155,8 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
         {
            functionalCoreSim cta(
                &kernel,
-               g_the_gpu(),
-               g_the_gpu()->getShaderCoreConfig()->warp_size
+               gpgpu_ctx->the_gpgpusim->g_the_gpu,
+               gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->warp_size
            );
            cta.execute(cp_count,temp);
 
@@ -2184,7 +2177,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
 	{
       char f1name[2048];
       snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", kernel.get_uid() );
-      g_checkpoint->store_global_mem(g_the_gpu()->get_global_memory(), f1name , "%08x");
+      g_checkpoint->store_global_mem(gpgpu_ctx->the_gpgpusim->g_the_gpu->get_global_memory(), f1name , (char *)"%08x");
 	}
 
 
@@ -2195,7 +2188,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
    //openCL kernel simulation calls don't register the kernel so we don't register its exit
    if(!openCL) {
       //extern stream_manager *g_stream_manager;
-      g_stream_manager()->register_finished_kernel(kernel.get_uid());
+      gpgpu_ctx->the_gpgpusim->g_stream_manager->register_finished_kernel(kernel.get_uid());
    }
 
    //******PRINTING*******
@@ -2210,7 +2203,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
    //g_simulation_starttime is initilized by gpgpu_ptx_sim_init_perf() in gpgpusim_entrypoint.cc upon starting gpgpu-sim
    time_t end_time, elapsed_time, days, hrs, minutes, sec;
    end_time = time((time_t *)NULL);
-   elapsed_time = MAX(end_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1);
+   elapsed_time = MAX(end_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1);
 	
 
    //calculating and printing simulation time in terms of days, hours, minutes and seconds
@@ -2312,18 +2305,15 @@ void functionalCoreSim::execute(int inst_count, unsigned ctaid_cp)
     checkpoint *g_checkpoint;
     g_checkpoint = new checkpoint();
     
-    symbol * sym;
     ptx_reg_t regval;
     regval.u64= 123;
-    symbol_table * symtab= m_kernel->entry()->get_symtab();
-
 
     unsigned ctaid =m_kernel->get_next_cta_id_single();
     if(m_gpu->checkpoint_option==1 && (m_kernel->get_uid()==m_gpu->checkpoint_kernel) && (ctaid_cp>=m_gpu->checkpoint_CTA) && (ctaid_cp<m_gpu->checkpoint_CTA_t))
    {
        char fname[2048];
        snprintf(fname,2048,"checkpoint_files/shared_mem_%d.txt",ctaid-1 );
-       g_checkpoint->store_global_mem(m_thread[0]->m_shared_mem, fname , "%08x");
+       g_checkpoint->store_global_mem(m_thread[0]->m_shared_mem, fname , (char *)"%08x");
       for(int i=0; i<32*m_warp_count;i++)
       {
          char fname[2048];
@@ -2331,7 +2321,7 @@ void functionalCoreSim::execute(int inst_count, unsigned ctaid_cp)
           m_thread[i]->print_reg_thread(fname);
           char f1name[2048];
          snprintf(f1name,2048,"checkpoint_files/local_mem_thread_%d_%d_reg.txt",i,ctaid-1 );
-         g_checkpoint->store_global_mem(m_thread[i]->m_local_mem, f1name , "%08x");
+         g_checkpoint->store_global_mem(m_thread[i]->m_local_mem, f1name , (char *)"%08x");
          m_thread[i]->set_done();
          m_thread[i]->exitCore();
          m_thread[i]->registerExit();
diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc
index dc3adc3..4baced5 100644
--- a/src/cuda-sim/cuda_device_runtime.cc
+++ b/src/cuda-sim/cuda_device_runtime.cc
@@ -27,7 +27,6 @@
    }
 
 
-//extern stream_manager *g_stream_manager();
 
 //Handling device runtime api:
 //void * cudaGetParameterBufferV2(void *func, dim3 gridDimension, dim3 blockDimension, unsigned int sharedMemSize)
@@ -285,7 +284,7 @@ void cuda_device_runtime::launch_one_device_kernel() {
         device_launch_operation_t &op = g_cuda_device_launch_op.front();
 
         stream_operation stream_op = stream_operation(op.grid, gpgpu_ctx->func_sim->g_ptx_sim_mode, op.stream);
-        g_stream_manager()->push(stream_op);
+        gpgpu_ctx->the_gpgpusim->g_stream_manager->push(stream_op);
         g_cuda_device_launch_op.pop_front();
     }
 }
diff --git a/src/cuda-sim/half.h b/src/cuda-sim/half.h
index 8f1a8eb..9f74bb7 100644
--- a/src/cuda-sim/half.h
+++ b/src/cuda-sim/half.h
@@ -642,10 +642,10 @@ namespace half_float
 			if(exp > 16)

 			{

 				if(R == std::round_toward_infinity)

-					return hbits | 0x7C00 - (hbits>>15);

+					return hbits | (0x7C00 - (hbits>>15));

 				else if(R == std::round_toward_neg_infinity)

-					return hbits | 0x7BFF + (hbits>>15);

-				return hbits | 0x7BFF + (R!=std::round_toward_zero);

+					return hbits | (0x7BFF + (hbits>>15));

+				return hbits | (0x7BFF + (R!=std::round_toward_zero));

 			}

 			if(exp < -13)

 				value = std::ldexp(value, 24);

diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 58a077e..014e588 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -203,7 +203,7 @@ void ptx_thread_info::print_reg_thread(char * fname)
           const std::string &name = it->first->name();
           const std::string &dec= it->first->decl_location();
           unsigned size = it->first->get_size_in_bytes();
-          fprintf(fp,"%s %llu %s %d\n",name.c_str(),it->second, dec.c_str(),size );
+          fprintf(fp,"%s %llu %s %d\n", name.c_str(), it->second, dec.c_str(), size);
           
         }
    //m_regs.pop_back();     
@@ -224,7 +224,6 @@ void ptx_thread_info::resume_reg_thread(char * fname, symbol_table * symtab)
       {
           symbol *reg;
           char * pch;
-          unsigned size;
           pch = strtok (line," ");
           char * name =pch;
           reg= symtab->lookup(name);
@@ -232,11 +231,7 @@ void ptx_thread_info::resume_reg_thread(char * fname, symbol_table * symtab)
           pch = strtok (NULL," "); 
           data = atoi(pch);
           pch = strtok (NULL," ");
-          char * decl= pch;
           pch = strtok (NULL," ");
-          size = atoi(pch);
-
-
           m_regs.back()[reg] = data;
       }
       fclose ( fp2 );
@@ -1819,9 +1814,7 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
    	ptx_reg_t matrix_d[16][16];
    	ptx_reg_t src_data;
 	ptx_thread_info *thread;
-	int stride;
 
-	unsigned wmma_type = pI->get_wmma_type();
 	unsigned a_layout = pI->get_wmma_layout(0);
 	unsigned b_layout = pI->get_wmma_layout(1);
 	unsigned type = pI->get_type();
@@ -1833,7 +1826,6 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
          	tid= inst.warp_id_func()*core->get_warp_size();
         else
          	tid= inst.warp_id()*core->get_warp_size();
-	unsigned thread_group_index;
 	float temp;
 	half temp2; 	
 	
@@ -1847,9 +1839,9 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
          		ptx_reg_t v[8];
          		thread->get_vector_operand_values( src_a, v, nelem );
 			if(core->get_gpu()->gpgpu_ctx->debug_tensorcore){
-				printf("Thread%d_Iteration=%d\n:",thrd,operand_num);
-				for(k=0;k<nelem;k++){
-					printf("%x ",v[k].u64);
+				printf("Thread%d_Iteration=%d\n:", thrd, operand_num);
+				for(k = 0; k < nelem; k++){
+					printf("%llx ",v[k].u64);
 				}
 				printf("\n");
 			}
@@ -2027,7 +2019,7 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
 
 				printf("thread%d:",thrd);
 				for(k=0;k<8;k++){
-					printf("%x ",matrix_d[row_t[k]][col_t[k]].f16);
+					printf("%x ", (unsigned int)matrix_d[row_t[k]][col_t[k]].f16);
 				}
 				printf("\n");
 			}
@@ -2038,7 +2030,7 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
 			nw_data4.s64=((matrix_d[row_t[6]][col_t[6]].s64   & 0xffff))|((matrix_d[row_t[7]][col_t[7]].s64&0xffff)<<16);
    			thread->set_vector_operand_values(dst,nw_data1,nw_data2,nw_data3,nw_data4);
 			if(core->get_gpu()->gpgpu_ctx->debug_tensorcore)
-		 		printf("thread%d=%x,%x,%x,%x",thrd,nw_data1.s64,nw_data2.s64,nw_data3.s64,nw_data4.s64);
+		 		printf("thread%d=%llx,%llx,%llx,%llx", thrd, nw_data1.s64, nw_data2.s64, nw_data3.s64, nw_data4.s64);
 		
 		}
 		else{
@@ -2298,9 +2290,8 @@ unsigned int saturatei(unsigned int a, unsigned int max)
 
 ptx_reg_t f2x( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign, int rounding_mode, int saturation_mode )
 {
- half mytemp;
- float myfloat;
-  half_float::half tmp_h;
+   half mytemp;
+   half_float::half tmp_h;
    //assert( from_width == 32); 
 
    enum cudaRoundMode mode = cudaRoundZero;
@@ -3085,7 +3076,7 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
    size_t size;
    unsigned smid;
    int t;
-   int thrd,odd,inx,k;
+   int thrd, k;
    ptx_thread_info *thread;
    
    const operand_info &src  = pI->operand_lookup(1);
@@ -3105,15 +3096,13 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
    _memory_op_t insn_memory_op = pI->has_memory_read() ? memory_load : memory_store;
    for (thrd=0; thrd < core->get_warp_size(); thrd++) {
 	thread = core->get_thread_info()[tid+thrd];
-	odd=thrd%2;
-	inx=thrd/2;
-    	ptx_reg_t addr_reg = thread->get_operand_value(src1, src, type, thread, 1);
+    ptx_reg_t addr_reg = thread->get_operand_value(src1, src, type, thread, 1);
    	ptx_reg_t src2_data = thread->get_operand_value(src2, src, type, thread, 1);
 	const operand_info &src_a=  pI->operand_lookup(1);
         unsigned nelem = src_a.get_vect_nelem();
         ptx_reg_t* v= new ptx_reg_t[8]; 
        	thread->get_vector_operand_values( src_a, v, nelem );
-  	stride=src2_data.u32;
+  	stride = src2_data.u32;
  
    	memory_space_t space = pI->get_space();
 
@@ -3130,9 +3119,9 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 	}
    	decode_space(space,thread,src1,mem,addr);
 
-   	type_info_key::type_decode(type,size,t);
+   	type_info_key::type_decode(type, size, t);
    	if(core->get_gpu()->gpgpu_ctx->debug_tensorcore)
-		printf("mma_st: thrd=%d,addr=%x, fp(size=%d), stride=%d\n",thrd,addr_reg.u32,size,src2_data.u32);
+		printf("mma_st: thrd=%d, addr=%x, fp(size=%zu), stride=%d\n", thrd, addr_reg.u32, size, src2_data.u32);
 	addr_t new_addr = addr+thread_group_offset(thrd,wmma_type,wmma_layout,type,stride)*size/8;  
 	addr_t push_addr;
 
@@ -3152,7 +3141,7 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 			mem_txn_addr[num_mem_txn++]=push_addr;
 	
 			if(core->get_gpu()->gpgpu_ctx->debug_tensorcore){
-				printf("wmma:store:thread%d=%x,%x,%x,%x,%x,%x,%x,%x\n",thrd,v[0].s64,v[1].s64,v[2].s64,v[3].s64,v[4].s64,v[5].s64,v[6].s64,v[7].s64);   
+				printf("wmma:store:thread%d=%llx,%llx,%llx,%llx,%llx,%llx,%llx,%llx\n",thrd,v[0].s64,v[1].s64,v[2].s64,v[3].s64,v[4].s64,v[5].s64,v[6].s64,v[7].s64);   
 				float temp;
 				int l;
 				printf("thread=%d:",thrd);
@@ -3179,7 +3168,7 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 			}
 	
 			if(core->get_gpu()->gpgpu_ctx->debug_tensorcore)
-				printf("wmma:store:thread%d=%x,%x,%x,%x,%x,%x,%x,%x\n",thrd,nw_v[0].s64,nw_v[1].s64,nw_v[2].s64,nw_v[3].s64,nw_v[4].s64,nw_v[5].s64,nw_v[6].s64,nw_v[7].s64);   
+				printf("wmma:store:thread%d=%llx,%llx,%llx,%llx,%llx,%llx,%llx,%llx\n",thrd,nw_v[0].s64,nw_v[1].s64,nw_v[2].s64,nw_v[3].s64,nw_v[4].s64,nw_v[5].s64,nw_v[6].s64,nw_v[7].s64);   
 		}
 	}
    	
@@ -3238,11 +3227,11 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 	}
 
 	decode_space(space,thread,src1,mem,addr);
-   	type_info_key::type_decode(type,size,t);
+   	type_info_key::type_decode(type, size, t);
 	
 	ptx_reg_t data[16];
    	if(core->get_gpu()->gpgpu_ctx->debug_tensorcore)	
-		printf("mma_ld: thrd=%d,addr=%x, fpsize=%d, stride=%d\n",thrd,src1_data.u32,size,src2_data.u32);
+		printf("mma_ld: thrd=%d,addr=%x, fpsize=%zu, stride=%d\n", thrd, src1_data.u32, size, src2_data.u32);
 	
 	addr_t new_addr = addr+thread_group_offset(thrd,wmma_type,wmma_layout,type,stride)*size/8;  
 	addr_t fetch_addr;
@@ -3341,7 +3330,7 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 		if(type==F16_TYPE){
 			printf("\nmma_ld:thread%d= ",thrd);
 			for(i=0;i<16;i++){
-				printf("%x ",data[i].u64);
+				printf("%llx ",data[i].u64);
 			}
 			printf("\n");
 			
@@ -3361,7 +3350,7 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 			printf("\n");
 			printf("\nmma_ld:thread%d= ",thrd);
 			for(i=0;i<8;i++){
-				printf("%x ",data[i].u64);
+				printf("%llx ",data[i].u64);
 			}
 			printf("\n");
 		}
@@ -3388,15 +3377,15 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst )
 		else
    			thread->set_wmma_vector_operand_values(dst,nw_data[0],nw_data[1],nw_data[2],nw_data[3],nw_data[4],nw_data[5],nw_data[6],nw_data[7]);
 		if(core->get_gpu()->gpgpu_ctx->debug_tensorcore){	
-			printf("mma_ld:data[0].s64=%x,data[1].s64=%x,new_data[0].s64=%x\n",data[0].u64,data[1].u64,nw_data[0].u64);	
-			printf("mma_ld:data[2].s64=%x,data[3].s64=%x,new_data[1].s64=%x\n",data[2].u64,data[3].u64,nw_data[1].u64);	
-			printf("mma_ld:data[4].s64=%x,data[5].s64=%x,new_data[2].s64=%x\n",data[4].u64,data[5].u64,nw_data[2].u64);	
-			printf("mma_ld:data[6].s64=%x,data[7].s64=%x,new_data[3].s64=%x\n",data[6].u64,data[7].u64,nw_data[3].u64);	
+			printf("mma_ld:data[0].s64=%llx,data[1].s64=%llx,new_data[0].s64=%llx\n",data[0].u64,data[1].u64,nw_data[0].u64);	
+			printf("mma_ld:data[2].s64=%llx,data[3].s64=%llx,new_data[1].s64=%llx\n",data[2].u64,data[3].u64,nw_data[1].u64);	
+			printf("mma_ld:data[4].s64=%llx,data[5].s64=%llx,new_data[2].s64=%llx\n",data[4].u64,data[5].u64,nw_data[2].u64);	
+			printf("mma_ld:data[6].s64=%llx,data[7].s64=%llx,new_data[3].s64=%llx\n",data[6].u64,data[7].u64,nw_data[3].u64);	
 			if(wmma_type!=LOAD_C){
-			printf("mma_ld:data[8].s64=%x,data[9].s64=%x,new_data[4].s64=%x\n",data[8].u64,data[9].u64,nw_data[4].s64);	
-			printf("mma_ld:data[10].s64=%x,data[11].s64=%x,new_data[5].s64=%x\n",data[10].u64,data[11].u64,nw_data[5].u64);	
-			printf("mma_ld:data[12].s64=%x,data[13].s64=%x,new_data[6].s64=%x\n",data[12].u64,data[13].u64,nw_data[6].u64);	
-			printf("mma_ld:data[14].s64=%x,data[15].s64=%x,new_data[7].s64=%x\n",data[14].u64,data[15].u64,nw_data[3].u64);	
+			printf("mma_ld:data[8].s64=%llx,data[9].s64=%llx,new_data[4].s64=%llx\n",data[8].u64,data[9].u64,nw_data[4].s64);	
+			printf("mma_ld:data[10].s64=%llx,data[11].s64=%llx,new_data[5].s64=%llx\n",data[10].u64,data[11].u64,nw_data[5].u64);	
+			printf("mma_ld:data[12].s64=%llx,data[13].s64=%llx,new_data[6].s64=%llx\n",data[12].u64,data[13].u64,nw_data[6].u64);	
+			printf("mma_ld:data[14].s64=%llx,data[15].s64=%llx,new_data[7].s64=%llx\n",data[14].u64,data[15].u64,nw_data[3].u64);	
 			}
 		}
 	}
@@ -4132,9 +4121,9 @@ int prmt_mode_present(int mode)
 	}
 	return returnval;
 }
-int read_byte(int mode,int control,int d_sel_index,signed long long value){
+int read_byte(int mode, int control, int d_sel_index, signed long long value){
 
-	int returnval;
+	int returnval = 0;
 	int prmt_f4e_mode[4][4]={{0,1,2,3},{1,2,3,4},{2,3,4,5},{3,4,5,6}};
 	int prmt_b4e_mode[4][4]={{0,7,6,5},{1,0,7,6},{2,1,0,7},{3,2,1,0}};
 	int prmt_rc8_mode[4][4]={{0,0,0,0},{1,1,1,1},{2,2,2,2},{3,3,3,3}};
@@ -4157,11 +4146,12 @@ int read_byte(int mode,int control,int d_sel_index,signed long long value){
 			case PRMT_RC8_MODE:	returnval=prmt_rc8_mode[control][d_sel_index];break;
 			case PRMT_ECL_MODE:	returnval=prmt_ecl_mode[control][d_sel_index];break;
 			case PRMT_ECR_MODE:	returnval=prmt_ecr_mode[control][d_sel_index];break;
-			case PRMT_RC16_MODE:	returnval=prmt_rc16_mode[control][d_sel_index];break;
-			default: printf("ERROR\n");break;
+			case PRMT_RC16_MODE: returnval=prmt_rc16_mode[control][d_sel_index];break;
+            // Change the default from printing "ERROR" to just asserting
+			default: assert(false);
 		}
 	}	
-	return (returnval<<8*d_sel_index);
+	return (returnval << 8 * d_sel_index);
 }
 
 void prmt_impl( const ptx_instruction *pI, ptx_thread_info *thread ) { 
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 6978cc1..d8943d2 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -1415,7 +1415,7 @@ unsigned function_info::print_insn( unsigned pc, FILE * fp ) const
    snprintf(command,1024,"c++filt -p %s",m_name.c_str());
    FILE *p = popen(command,"r");
    buffer[0]=0;
-   fgets(buffer, 1023, p);
+   assert(fgets(buffer, 1023, p) != NULL);
    // Remove trailing "\n" in buffer
    char *c;
    if ((c=strchr(buffer, '\n')) != NULL) *c = '\0';
diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc
index 207fa1f..f06fefd 100644
--- a/src/cuda-sim/ptx_loader.cc
+++ b/src/cuda-sim/ptx_loader.cc
@@ -214,7 +214,8 @@ void fix_duplicate_errors(char fname2[1024]) {
 	long filesize = ftell(ptxsource);
 	rewind(ptxsource);
 	char *ptxdata = (char*)malloc((filesize+1)*sizeof(char));
-	fread(ptxdata, filesize, 1, ptxsource);
+    // Fail if we do not read the file
+	assert(fread(ptxdata, filesize, 1, ptxsource) == 1);
 	fclose(ptxsource);
 
 	FILE *ptxdest = fopen(fname2,"w");
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index 9e9d500..24af3ad 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -421,7 +421,7 @@ void ptx_recognizer::add_identifier( const char *identifier, int array_dim, unsi
          assert( (num_bits%8) == 0  );
          addr = g_current_symbol_table->get_sstarr_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (sstarr memory space)\n",
+         printf("from 0x%llx to 0x%llx (sstarr memory space)\n",
                  addr+addr_pad,
                  addr+addr_pad + num_bits/8);
             fflush(stdout);
diff --git a/src/gpgpu-sim/addrdec.cc b/src/gpgpu-sim/addrdec.cc
index 3262456..f225933 100644
--- a/src/gpgpu-sim/addrdec.cc
+++ b/src/gpgpu-sim/addrdec.cc
@@ -184,7 +184,7 @@ void linear_to_raw_address_translation::addrdec_tlx(new_addr_type addr, addrdec_
 			  }
 
 			  assert(tlx->chip < m_n_channel);
-			  assert(tlx->sub_partition < m_n_channel*m_n_sub_partition_in_channel);
+			  assert(tlx->sub_partition < m_n_channel * m_n_sub_partition_in_channel);
 			  return;
 			break;
 		}
diff --git a/src/gpgpu-sim/addrdec.h b/src/gpgpu-sim/addrdec.h
index a5333fb..c9a1420 100644
--- a/src/gpgpu-sim/addrdec.h
+++ b/src/gpgpu-sim/addrdec.h
@@ -92,7 +92,7 @@ private:
    new_addr_type sub_partition_id_mask; 
 
    unsigned int gap;
-   int m_n_channel;
+   unsigned m_n_channel;
    int m_n_sub_partition_in_channel; 
 
 };
diff --git a/src/gpgpu-sim/dram.cc b/src/gpgpu-sim/dram.cc
index d443d79..9c33822 100644
--- a/src/gpgpu-sim/dram.cc
+++ b/src/gpgpu-sim/dram.cc
@@ -482,7 +482,6 @@ void dram_t::cycle()
 
         bool memory_pending_rw_found=false;
         for (unsigned j=0;j<m_config->nbk;j++) {
-     	   unsigned grp = get_bankgrp_number(j);
      	   if (bk[j]->mrq && (((bk[j]->curr_row == bk[j]->mrq->row) &&
      		  (bk[j]->mrq->rw == READ)  &&
      		  (bk[j]->state == BANK_ACTIVE))
@@ -817,10 +816,10 @@ void dram_t::visualize() const
 
 void dram_t::print_stat( FILE* simFile ) 
 {
-   fprintf(simFile,"DRAM (%llu): n_cmd=%llu n_nop=%llu n_act=%llu n_pre=%llu n_ref=%llu n_req=%llu n_rd=%llu n_write=%llu bw_util=%.4g ",
+   fprintf(simFile,"DRAM (%u): n_cmd=%llu n_nop=%llu n_act=%llu n_pre=%llu n_ref=%llu n_req=%llu n_rd=%llu n_write=%llu bw_util=%.4g ",
            id, n_cmd, n_nop, n_act, n_pre, n_ref, n_req, n_rd, n_wr,
            (float)bwutil/n_cmd);
-   fprintf(simFile, "mrqq: %d %.4g mrqsmax=%d ", max_mrqs, (float)ave_mrqs/n_cmd, max_mrqs_temp);
+   fprintf(simFile, "mrqq: %d %.4g mrqsmax=%llu ", max_mrqs, (float)ave_mrqs/n_cmd, max_mrqs_temp);
    fprintf(simFile, "\n");
    fprintf(simFile, "dram_util_bins:");
    for (unsigned i=0;i<10;i++) fprintf(simFile, " %d", dram_util_bins[i]);
@@ -899,10 +898,10 @@ void dram_t::set_dram_power_stats(	unsigned &cmd,
 unsigned dram_t::get_bankgrp_number(unsigned i)
 {
 	if(m_config->dram_bnkgrp_indexing_policy == HIGHER_BITS) { //higher bits
-		return i>>m_config->bk_tag_length;
+		return i >> m_config->bk_tag_length;
 	}
 	else if (m_config->dram_bnkgrp_indexing_policy == LOWER_BITS) { //lower bits
-		return i&((m_config->nbkgrp-1));
+		return i & ((m_config->nbkgrp - 1));
 	}
 	else {
 		assert(1);
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index b7d0ac3..65f2f6d 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -790,7 +790,7 @@ void cache_stats::print_stats(FILE *fout, const char *cache_name) const{
     }
     for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) {
     	 if(total_access[type] > 0)
-    	  fprintf(fout, "\t%s[%s][%s] = %llu\n",
+    	  fprintf(fout, "\t%s[%s][%s] = %u\n",
 				m_cache_name.c_str(),
 				mem_access_type_str((enum mem_access_type)type),
 				"TOTAL_ACCESS",
@@ -803,7 +803,7 @@ void cache_stats::print_fail_stats(FILE *fout, const char *cache_name) const{
 	    for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) {
 	        for (unsigned fail = 0; fail < NUM_CACHE_RESERVATION_FAIL_STATUS; ++fail) {
 	            if(m_fail_stats[type][fail] > 0){
-	                fprintf(fout, "\t%s[%s][%s] = %u\n",
+	                fprintf(fout, "\t%s[%s][%s] = %llu\n",
 	                    m_cache_name.c_str(),
 	                    mem_access_type_str((enum mem_access_type)type),
 						cache_fail_status_str((enum cache_reservation_fail_reason)fail),
@@ -1430,8 +1430,6 @@ data_cache::wr_miss_wa_lazy_fetch_on_read( new_addr_type addr,
 {
 
 	    new_addr_type block_addr = m_config.block_addr(addr);
-	    new_addr_type mshr_addr = m_config.mshr_addr(mf->get_addr());
-
 
 		//if the request writes to the whole cache line/sector, then, write and set cache line Modified.
 		//and no need to send read request to memory or reserve mshr
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 234a8d6..a70185e 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -475,10 +475,10 @@ void gpgpu_sim_config::reg_options(option_parser_t opp)
     m_shader_config.reg_options(opp);
     m_memory_config.reg_options(opp);
     power_config::reg_options(opp);
-   option_parser_register(opp, "-gpgpu_max_cycle", OPT_INT32, &gpu_max_cycle_opt, 
+   option_parser_register(opp, "-gpgpu_max_cycle", OPT_INT64, &gpu_max_cycle_opt, 
                "terminates gpu simulation early (0 = no limit)",
                "0");
-   option_parser_register(opp, "-gpgpu_max_insn", OPT_INT32, &gpu_max_insn_opt, 
+   option_parser_register(opp, "-gpgpu_max_insn", OPT_INT64, &gpu_max_insn_opt, 
                "terminates gpu simulation early (0 = no limit)",
                "0");
    option_parser_register(opp, "-gpgpu_max_cta", OPT_INT32, &gpu_max_cta_opt, 
@@ -1126,8 +1126,8 @@ void gpgpu_sim::gpu_print_stat()
    printf("gpu_tot_sim_insn = %lld\n", gpu_tot_sim_insn+gpu_sim_insn);
    printf("gpu_tot_ipc = %12.4f\n", (float)(gpu_tot_sim_insn+gpu_sim_insn) / (gpu_tot_sim_cycle+gpu_sim_cycle));
    printf("gpu_tot_issued_cta = %lld\n", gpu_tot_issued_cta + m_total_cta_launched);
-   printf("gpu_occupancy = %.4f\% \n", gpu_occupancy.get_occ_fraction() * 100);
-   printf("gpu_tot_occupancy = %.4f\% \n", (gpu_occupancy + gpu_tot_occupancy).get_occ_fraction() * 100);
+   printf("gpu_occupancy = %.4f%% \n", gpu_occupancy.get_occ_fraction() * 100);
+   printf("gpu_tot_occupancy = %.4f%% \n", (gpu_occupancy + gpu_tot_occupancy).get_occ_fraction() * 100);
 
 
    fprintf(statfout, "max_total_param_size = %llu\n", gpgpu_ctx->device_runtime->g_max_total_param_size);
@@ -1153,7 +1153,7 @@ void gpgpu_sim::gpu_print_stat()
 
    time_t curr_time;
    time(&curr_time);
-   unsigned long long elapsed_time = MAX( curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1 );
+   unsigned long long elapsed_time = MAX( curr_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1 );
    printf( "gpu_total_sim_rate=%u\n", (unsigned)( ( gpu_tot_sim_insn + gpu_sim_insn ) / elapsed_time ) );
 
    //shader_print_l1_miss_stat( stdout );
@@ -1368,7 +1368,7 @@ bool shader_core_ctx::occupy_shader_resource_1block(kernel_info_t & k, bool occu
        m_occupied_regs += (padded_cta_size * ((kernel_info->regs+3)&~3));
        m_occupied_ctas++;
 
-      SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Occupied %d threads, %d shared mem, %d registers, %d ctas\n",
+      SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Occupied %u threads, %u shared mem, %u registers, %u ctas\n",
             m_occupied_n_threads, m_occupied_shmem, m_occupied_regs, m_occupied_ctas);  
    }
 
@@ -1485,7 +1485,7 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel )
         nthreads_in_block += ptx_sim_init_thread(kernel,&m_thread[i],m_sid,i,cta_size-(i-start_thread),m_config->n_thread_per_shader,this,free_cta_hw_id,warp_id,m_cluster->get_gpu());
         m_threadState[i].m_active = true; 
         // load thread local memory and register file
-        if(m_gpu->resume_option==1 && kernel.get_uid()==m_gpu->resume_kernel && ctaid>=m_gpu->resume_CTA && ctaid<m_gpu->checkpoint_CTA_t )
+        if(m_gpu->resume_option == 1 && kernel.get_uid() == m_gpu->resume_kernel && ctaid >= m_gpu->resume_CTA && ctaid < m_gpu->checkpoint_CTA_t )
         {
             char fname[2048];
             snprintf(fname,2048,"checkpoint_files/thread_%d_%d_reg.txt",i%cta_size,ctaid );
@@ -1500,7 +1500,7 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel )
     assert( nthreads_in_block > 0 && nthreads_in_block <= m_config->n_thread_per_shader); // should be at least one, but less than max
     m_cta_status[free_cta_hw_id]=nthreads_in_block;
 
-    if(m_gpu->resume_option==1 && kernel.get_uid()==m_gpu->resume_kernel && ctaid>=m_gpu->resume_CTA && ctaid<m_gpu->checkpoint_CTA_t )
+    if(m_gpu->resume_option == 1 && kernel.get_uid() == m_gpu->resume_kernel && ctaid >= m_gpu->resume_CTA && ctaid < m_gpu->checkpoint_CTA_t )
     {
         char f1name[2048];
         snprintf(f1name,2048,"checkpoint_files/shared_mem_%d.txt", ctaid);
@@ -1726,7 +1726,7 @@ void gpgpu_sim::cycle()
          time_t days, hrs, minutes, sec;
          time_t curr_time;
          time(&curr_time);
-         unsigned long long  elapsed_time = MAX(curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1);
+         unsigned long long  elapsed_time = MAX(curr_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1);
          if ( (elapsed_time - last_liveness_message_time) >= m_config.liveness_message_freq && DTRACE(LIVENESS) ) {
             days    = elapsed_time/(3600*24);
             hrs     = elapsed_time/3600 - 24*days;
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 6b57bd8..86d6206 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -371,8 +371,8 @@ private:
     double l2_period;
 
     // GPGPU-Sim timing model options
-    unsigned gpu_max_cycle_opt;
-    unsigned gpu_max_insn_opt;
+    unsigned long long gpu_max_cycle_opt;
+    unsigned long long gpu_max_insn_opt;
     unsigned gpu_max_cta_opt;
     char *gpgpu_runtime_stat;
     bool  gpgpu_flush_l1_cache;
diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc
index 39a5812..fb4ce32 100644
--- a/src/gpgpu-sim/l2cache.cc
+++ b/src/gpgpu-sim/l2cache.cc
@@ -82,7 +82,7 @@ void memory_partition_unit::handle_memcpy_to_gpu( size_t addr, unsigned global_s
     unsigned p = global_sub_partition_id_to_local_id(global_subpart_id);
     std::string mystring =
         mask.to_string<char,std::string::traits_type,std::string::allocator_type>();
-    MEMPART_DPRINTF("Copy Engine Request Received For Address=%llx, local_subpart=%u, global_subpart=%u, sector_mask=%s \n", addr, p, global_subpart_id, mystring.c_str()); 
+    MEMPART_DPRINTF("Copy Engine Request Received For Address=%zx, local_subpart=%u, global_subpart=%u, sector_mask=%s \n", addr, p, global_subpart_id, mystring.c_str()); 
     m_sub_partition[p]->force_l2_tag_update(addr,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle, mask);
 }
 
@@ -681,7 +681,7 @@ std::vector<mem_fetch*> memory_sub_partition::breakdown_request_to_sector_reques
 			}
 		} else
 			{
-			    printf("Invalid sector received, address = 0x%06x, sector mask = %s, data size = %d",
+			    printf("Invalid sector received, address = 0x%06llx, sector mask = %s, data size = %d",
 			    		mf->get_addr(), mf->get_access_sector_mask(), mf->get_data_size());
 				assert(0 && "Undefined sector mask is received");
 			}
@@ -716,7 +716,7 @@ std::vector<mem_fetch*> memory_sub_partition::breakdown_request_to_sector_reques
 			 byte_sector_mask <<= SECTOR_SIZE;
 		}
 	} else {
-		 printf("Invalid sector received, address = 0x%06x, sector mask = %d, byte mask = , data size = %d",
+		 printf("Invalid sector received, address = 0x%06llx, sector mask = %d, byte mask = , data size = %u",
 					    		mf->get_addr(), mf->get_access_sector_mask().count(), mf->get_data_size());
 		 assert(0 && "Undefined data size is received");
 	}
diff --git a/src/gpgpu-sim/local_interconnect.cc b/src/gpgpu-sim/local_interconnect.cc
index da8a65c..c70477c 100644
--- a/src/gpgpu-sim/local_interconnect.cc
+++ b/src/gpgpu-sim/local_interconnect.cc
@@ -253,7 +253,7 @@ LocalInterconnect::LocalInterconnect(const struct inct_config& m_localinct_confi
 }
 
 LocalInterconnect::~LocalInterconnect(){
-	for (int i=0; i<m_inct_config.subnets; ++i) {
+	for (unsigned i = 0; i < m_inct_config.subnets; ++i) {
 		delete net[i];
 	}
 }
diff --git a/src/gpgpu-sim/scoreboard.cc b/src/gpgpu-sim/scoreboard.cc
index 80f95c6..1017e75 100644
--- a/src/gpgpu-sim/scoreboard.cc
+++ b/src/gpgpu-sim/scoreboard.cc
@@ -140,10 +140,10 @@ bool Scoreboard::checkCollision( unsigned wid, const class inst_t *inst ) const
 	// Get list of all input and output registers
 	std::set<int> inst_regs;
 
-	for(int iii=0;iii<inst->outcount;iii++)
+	for(unsigned iii=0; iii < inst->outcount; iii++)
 		inst_regs.insert(inst->out[iii]);
 
-	for(int jjj=0;jjj<inst->incount;jjj++)
+	for(unsigned jjj=0;jjj<inst->incount;jjj++)
 		inst_regs.insert(inst->in[jjj]);
 
         if(inst->pred > 0) inst_regs.insert(inst->pred);
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 0514a77..86508f6 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -87,7 +87,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
                                   shader_core_stats *stats )
    : core_t( gpu, NULL, config->warp_size, config->n_thread_per_shader ),
      m_barriers( this, config->max_warps_per_shader, config->max_cta_per_core, config->max_barriers_per_cta, config->warp_size ),
-     m_dynamic_warp_id(0), m_active_warps(0)
+     m_active_warps(0), m_dynamic_warp_id(0)
 {
     m_cluster = cluster;
     m_config = config;
@@ -164,7 +164,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
                                          NUM_CONCRETE_SCHEDULERS;
     assert ( scheduler != NUM_CONCRETE_SCHEDULERS );
     
-    for (int i = 0; i < m_config->gpgpu_num_sched_per_core; i++) {
+    for (unsigned i = 0; i < m_config->gpgpu_num_sched_per_core; i++) {
         switch( scheduler )
         {
             case CONCRETE_SCHEDULER_LRR:
@@ -263,7 +263,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
         //distribute i's evenly though schedulers;
         schedulers[i%m_config->gpgpu_num_sched_per_core]->add_supervised_warp_id(i);
     }
-    for ( int i = 0; i < m_config->gpgpu_num_sched_per_core; ++i ) {
+    for ( unsigned i = 0; i < m_config->gpgpu_num_sched_per_core; ++i ) {
         schedulers[i]->done_adding_supervised_warps();
     }
     
@@ -474,7 +474,7 @@ void shader_core_ctx::init_warps( unsigned cta_id, unsigned start_thread, unsign
             }
             m_simt_stack[i]->launch(start_pc,active_threads);
 
-              if(m_gpu->resume_option==1 && kernel_id==m_gpu->resume_kernel && ctaid>=m_gpu->resume_CTA && ctaid<m_gpu->checkpoint_CTA_t )
+              if(m_gpu->resume_option == 1 && kernel_id == m_gpu->resume_kernel && ctaid >= m_gpu->resume_CTA && ctaid < m_gpu->checkpoint_CTA_t )
                { 
                 char fname[2048];
                 snprintf(fname,2048,"checkpoint_files/warp_%d_%d_simt.txt",i%warp_per_cta,ctaid );
@@ -868,7 +868,7 @@ void shader_core_ctx::func_exec_inst( warp_inst_t &inst )
 
 void shader_core_ctx::issue_warp( register_set& pipe_reg_set, const warp_inst_t* next_inst, const active_mask_t &active_mask, unsigned warp_id, unsigned sch_id )
 {
-	warp_inst_t** pipe_reg = pipe_reg = pipe_reg_set.get_free(m_config->sub_core_model, sch_id);
+	warp_inst_t** pipe_reg = pipe_reg_set.get_free(m_config->sub_core_model, sch_id);
     assert(pipe_reg);
 
     m_warp[warp_id].ibuffer_free();
@@ -2147,7 +2147,7 @@ ldst_unit::ldst_unit( mem_fetch_interface *icnt,
         l1_latency_queue.resize(m_config->m_L1D_config.l1_banks);
         assert(m_config->m_L1D_config.l1_latency > 0);
 
-		for(int j=0; j<m_config->m_L1D_config.l1_banks; j++ )
+		for(unsigned j = 0; j < m_config->m_L1D_config.l1_banks; j++ )
 			l1_latency_queue[j].resize(m_config->m_L1D_config.l1_latency,(mem_fetch*)NULL);
 
     }
@@ -2459,7 +2459,7 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t
       m_barriers.deallocate_barrier(cta_num);
       shader_CTA_count_unlog(m_sid, 1);
 
-     SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Finished CTA #%d (%lld,%lld), %u CTAs running\n",
+     SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Finished CTA #%u (%lld,%lld), %u CTAs running\n",
         cta_num, m_gpu->gpu_sim_cycle, m_gpu->gpu_tot_sim_cycle, m_n_active_cta);
 
       if( m_n_active_cta == 0 ) {
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 62e0e42..667cb2d 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1392,9 +1392,9 @@ class shader_core_config : public core_config
 	 	If we won't remove it, old regression will be broken.
 		So to support the legacy config files it's best to handle in this way.
          */  
-	int num_config_to_read=N_PIPELINE_STAGES-2*(!gpgpu_tensor_core_avail);
+	int num_config_to_read= N_PIPELINE_STAGES - 2 * (!gpgpu_tensor_core_avail);
 
-        for (unsigned i = 0; i <num_config_to_read; i++) { 
+        for (int i = 0; i < num_config_to_read; i++) { 
 	    assert(toks);
 	    ntok = sscanf(toks,"%d", &pipe_widths[i]);
 	    assert(ntok == 1); 
@@ -1455,7 +1455,7 @@ class shader_core_config : public core_config
 
     bool gpgpu_dwf_reg_bankconflict;
 
-    int gpgpu_num_sched_per_core;
+    unsigned gpgpu_num_sched_per_core;
     int gpgpu_max_insn_issue_per_warp;
     bool gpgpu_dual_issue_diff_exec_units;
 
diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc
index 4da74c0..69cc1b0 100644
--- a/src/gpgpusim_entrypoint.cc
+++ b/src/gpgpusim_entrypoint.cc
@@ -43,45 +43,28 @@ static int sg_argc = 3;
 static const char *sg_argv[] = {"", "-config","gpgpusim.config"};
 
 
-GPGPUsim_ctx* the_gpgpusim =  NULL;
-
-GPGPUsim_ctx* GPGPUsim_ctx_ptr(){
-	if(the_gpgpusim == NULL)
-		the_gpgpusim = GPGPU_Context()->the_gpgpusim;
-
-	return the_gpgpusim;
-}
-
-class gpgpu_sim* g_the_gpu() {
-	return GPGPUsim_ctx_ptr()->g_the_gpu;
-}
-
-class stream_manager* g_stream_manager()  {
-	return GPGPUsim_ctx_ptr()->g_stream_manager;
-}
-
-
-void *gpgpu_sim_thread_sequential(void*)
+void * gpgpu_sim_thread_sequential(void * ctx_ptr)
 {
+    gpgpu_context * ctx = (gpgpu_context *)ctx_ptr;
    // at most one kernel running at a time
    bool done;
    do {
-      sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_start));
+      sem_wait(&(ctx->the_gpgpusim->g_sim_signal_start));
       done = true;
-      if( GPGPUsim_ctx_ptr()->g_the_gpu->get_more_cta_left() ) {
+      if( ctx->the_gpgpusim->g_the_gpu->get_more_cta_left() ) {
           done = false;
-          GPGPUsim_ctx_ptr()->g_the_gpu->init();
-          while( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) {
-        	  GPGPUsim_ctx_ptr()->g_the_gpu->cycle();
-        	  GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check();
+          ctx->the_gpgpusim->g_the_gpu->init();
+          while( ctx->the_gpgpusim->g_the_gpu->active() ) {
+        	  ctx->the_gpgpusim->g_the_gpu->cycle();
+        	  ctx->the_gpgpusim->g_the_gpu->deadlock_check();
           }
-          GPGPUsim_ctx_ptr()->g_the_gpu->print_stats();
-          GPGPUsim_ctx_ptr()->g_the_gpu->update_stats();
-          print_simulation_time();
+          ctx->the_gpgpusim->g_the_gpu->print_stats();
+          ctx->the_gpgpusim->g_the_gpu->update_stats();
+          ctx->print_simulation_time();
       }
-      sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish));
+      sem_post(&(ctx->the_gpgpusim->g_sim_signal_finish));
    } while(!done);
-   sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit));
+   sem_post(&(ctx->the_gpgpusim->g_sim_signal_exit));
    return NULL;
 }
 
@@ -93,8 +76,9 @@ static void termination_callback()
     fflush(stdout);
 }
 
-void *gpgpu_sim_thread_concurrent(void*)
+void *gpgpu_sim_thread_concurrent(void * ctx_ptr)
 {
+    gpgpu_context * ctx = (gpgpu_context *)ctx_ptr;
     atexit(termination_callback);
     // concurrent kernel execution simulation thread
     do {
@@ -102,19 +86,19 @@ void *gpgpu_sim_thread_concurrent(void*)
            printf("GPGPU-Sim: *** simulation thread starting and spinning waiting for work ***\n");
            fflush(stdout);
         }
-        while( GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected() && !GPGPUsim_ctx_ptr()->g_sim_done )
+        while( ctx->the_gpgpusim->g_stream_manager->empty_protected() && !ctx->the_gpgpusim->g_sim_done )
             ;
         if(g_debug_execution >= 3) {
            printf("GPGPU-Sim: ** START simulation thread (detected work) **\n");
-           GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout);
+           ctx->the_gpgpusim->g_stream_manager->print(stdout);
            fflush(stdout);
         }
-        pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-        GPGPUsim_ctx_ptr()->g_sim_active = true;
-        pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
+        pthread_mutex_lock(&(ctx->the_gpgpusim->g_sim_lock));
+        ctx->the_gpgpusim->g_sim_active = true;
+        pthread_mutex_unlock(&(ctx->the_gpgpusim->g_sim_lock));
         bool active = false;
         bool sim_cycles = false;
-        GPGPUsim_ctx_ptr()->g_the_gpu->init();
+        ctx->the_gpgpusim->g_the_gpu->init();
         do {
             // check if a kernel has completed
             // launch operation on device if one is pending and can be run
@@ -126,82 +110,82 @@ void *gpgpu_sim_thread_concurrent(void*)
             // another kernel, the gpu is not re-initialized and the inter-kernel
             // behaviour may be incorrect. Check that a kernel has finished and
             // no other kernel is currently running.
-            if(GPGPUsim_ctx_ptr()->g_stream_manager->operation(&sim_cycles) && !GPGPUsim_ctx_ptr()->g_the_gpu->active())
+            if(ctx->the_gpgpusim->g_stream_manager->operation(&sim_cycles) && !ctx->the_gpgpusim->g_the_gpu->active())
                 break;
 
             //functional simulation
-            if( GPGPUsim_ctx_ptr()->g_the_gpu->is_functional_sim()) {
-                kernel_info_t * kernel = GPGPUsim_ctx_ptr()->g_the_gpu->get_functional_kernel();
+            if( ctx->the_gpgpusim->g_the_gpu->is_functional_sim()) {
+                kernel_info_t * kernel = ctx->the_gpgpusim->g_the_gpu->get_functional_kernel();
                 assert(kernel);
-                GPGPUsim_ctx_ptr()->gpgpu_ctx->func_sim->gpgpu_cuda_ptx_sim_main_func(*kernel);
-                GPGPUsim_ctx_ptr()->g_the_gpu->finish_functional_sim(kernel);
+                ctx->the_gpgpusim->gpgpu_ctx->func_sim->gpgpu_cuda_ptx_sim_main_func(*kernel);
+                ctx->the_gpgpusim->g_the_gpu->finish_functional_sim(kernel);
             }
 
             //performance simulation
-            if( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) {
-            	GPGPUsim_ctx_ptr()->g_the_gpu->cycle();
+            if( ctx->the_gpgpusim->g_the_gpu->active() ) {
+            	ctx->the_gpgpusim->g_the_gpu->cycle();
             	sim_cycles = true;
-            	GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check();
+            	ctx->the_gpgpusim->g_the_gpu->deadlock_check();
             }else {
-                if(GPGPUsim_ctx_ptr()->g_the_gpu->cycle_insn_cta_max_hit()){
-                	GPGPUsim_ctx_ptr()->g_stream_manager->stop_all_running_kernels();
-                	GPGPUsim_ctx_ptr()->g_sim_done = true;
-                	GPGPUsim_ctx_ptr()->break_limit = true;
+                if(ctx->the_gpgpusim->g_the_gpu->cycle_insn_cta_max_hit()){
+                	ctx->the_gpgpusim->g_stream_manager->stop_all_running_kernels();
+                	ctx->the_gpgpusim->g_sim_done = true;
+                	ctx->the_gpgpusim->break_limit = true;
                 }
             }
 
-            active=GPGPUsim_ctx_ptr()->g_the_gpu->active() || !(GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected());
+            active=ctx->the_gpgpusim->g_the_gpu->active() || !(ctx->the_gpgpusim->g_stream_manager->empty_protected());
 
-        } while( active && !GPGPUsim_ctx_ptr()->g_sim_done);
+        } while( active && !ctx->the_gpgpusim->g_sim_done);
         if(g_debug_execution >= 3) {
            printf("GPGPU-Sim: ** STOP simulation thread (no work) **\n");
            fflush(stdout);
         }
         if(sim_cycles) {
-        	GPGPUsim_ctx_ptr()->g_the_gpu->print_stats();
-        	GPGPUsim_ctx_ptr()->g_the_gpu->update_stats();
-            print_simulation_time();
+        	ctx->the_gpgpusim->g_the_gpu->print_stats();
+        	ctx->the_gpgpusim->g_the_gpu->update_stats();
+            ctx->print_simulation_time();
         }
-        pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-        GPGPUsim_ctx_ptr()->g_sim_active = false;
-        pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-    } while( !GPGPUsim_ctx_ptr()->g_sim_done );
+        pthread_mutex_lock(&(ctx->the_gpgpusim->g_sim_lock));
+        ctx->the_gpgpusim->g_sim_active = false;
+        pthread_mutex_unlock(&(ctx->the_gpgpusim->g_sim_lock));
+    } while( !ctx->the_gpgpusim->g_sim_done );
 
     printf("GPGPU-Sim: *** simulation thread exiting ***\n");
     fflush(stdout);
 
-    if(GPGPUsim_ctx_ptr()->break_limit) {
+    if(ctx->the_gpgpusim->break_limit) {
     	printf("GPGPU-Sim: ** break due to reaching the maximum cycles (or instructions) **\n");
     	exit(1);
     }
 
-    sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit));
+    sem_post(&(ctx->the_gpgpusim->g_sim_signal_exit));
     return NULL;
 }
 
-void synchronize()
+void gpgpu_context::synchronize()
 {
     printf("GPGPU-Sim: synchronize waiting for inactive GPU simulation\n");
-    GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout);
+    the_gpgpusim->g_stream_manager->print(stdout);
     fflush(stdout);
 //    sem_wait(&g_sim_signal_finish);
     bool done = false;
     do {
-        pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-        done = ( GPGPUsim_ctx_ptr()->g_stream_manager->empty() && !GPGPUsim_ctx_ptr()->g_sim_active ) || GPGPUsim_ctx_ptr()->g_sim_done;
-        pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
+        pthread_mutex_lock(&(the_gpgpusim->g_sim_lock));
+        done = ( the_gpgpusim->g_stream_manager->empty() && !the_gpgpusim->g_sim_active ) || the_gpgpusim->g_sim_done;
+        pthread_mutex_unlock(&(the_gpgpusim->g_sim_lock));
     } while (!done);
     printf("GPGPU-Sim: detected inactive GPU simulation thread\n");
     fflush(stdout);
 //    sem_post(&g_sim_signal_start);
 }
 
-void exit_simulation()
+void gpgpu_context::exit_simulation()
 {
-	GPGPUsim_ctx_ptr()->g_sim_done=true;
+	the_gpgpusim->g_sim_done=true;
     printf("GPGPU-Sim: exit_simulation called\n");
     fflush(stdout);
-    sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit));
+    sem_wait(&(the_gpgpusim->g_sim_signal_exit));
     printf("GPGPU-Sim: simulation thread signaled exit\n");
     fflush(stdout);
 }
@@ -218,8 +202,8 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf()
    func_sim->ptx_opcocde_latency_options(opp);
 
    icnt_reg_options(opp);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config(this);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
+   the_gpgpusim->g_the_gpu_config = new gpgpu_sim_config(this);
+   the_gpgpusim->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
 
    option_parser_cmdline(opp, sg_argc, sg_argv); // parse configuration options
    fprintf(stdout, "GPGPU-Sim: Configuration options:\n\n");
@@ -227,18 +211,18 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf()
    // Set the Numeric locale to a standard locale where a decimal point is a "dot" not a "comma"
    // so it does the parsing correctly independent of the system environment variables
    assert(setlocale(LC_NUMERIC,"C"));
-   GPGPUsim_ctx_ptr()->g_the_gpu_config->init();
+   the_gpgpusim->g_the_gpu_config->init();
 
-   GPGPUsim_ctx_ptr()->g_the_gpu = new gpgpu_sim(*(GPGPUsim_ctx_ptr()->g_the_gpu_config), this);
-   GPGPUsim_ctx_ptr()->g_stream_manager = new stream_manager((GPGPUsim_ctx_ptr()->g_the_gpu), func_sim->g_cuda_launch_blocking);
+   the_gpgpusim->g_the_gpu = new gpgpu_sim(*(the_gpgpusim->g_the_gpu_config), this);
+   the_gpgpusim->g_stream_manager = new stream_manager((the_gpgpusim->g_the_gpu), func_sim->g_cuda_launch_blocking);
 
-   GPGPUsim_ctx_ptr()->g_simulation_starttime = time((time_t *)NULL);
+   the_gpgpusim->g_simulation_starttime = time((time_t *)NULL);
 
-   sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_start),0,0);
-   sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish),0,0);
-   sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit),0,0);
+   sem_init(&(the_gpgpusim->g_sim_signal_start),0,0);
+   sem_init(&(the_gpgpusim->g_sim_signal_finish),0,0);
+   sem_init(&(the_gpgpusim->g_sim_signal_exit),0,0);
 
-   return GPGPUsim_ctx_ptr()->g_the_gpu;
+   return the_gpgpusim->g_the_gpu;
 }
 
 gpgpu_sim *gpgpu_context::gpgpu_trace_sim_init_perf(int argc, const char *argv[])
@@ -253,8 +237,8 @@ gpgpu_sim *gpgpu_context::gpgpu_trace_sim_init_perf(int argc, const char *argv[]
    func_sim->ptx_opcocde_latency_options(opp);
 
    icnt_reg_options(opp);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config(this);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
+   the_gpgpusim->g_the_gpu_config = new gpgpu_sim_config(this);
+   the_gpgpusim->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
 
    option_parser_cmdline(opp, argc, argv); // parse configuration options
    fprintf(stdout, "GPGPU-Sim: Configuration options:\n\n");
@@ -262,33 +246,33 @@ gpgpu_sim *gpgpu_context::gpgpu_trace_sim_init_perf(int argc, const char *argv[]
    // Set the Numeric locale to a standard locale where a decimal point is a "dot" not a "comma"
    // so it does the parsing correctly independent of the system environment variables
    assert(setlocale(LC_NUMERIC,"C"));
-   GPGPUsim_ctx_ptr()->g_the_gpu_config->init();
+   the_gpgpusim->g_the_gpu_config->init();
 
-   GPGPUsim_ctx_ptr()->g_the_gpu = new gpgpu_sim(*(GPGPUsim_ctx_ptr()->g_the_gpu_config), this);
-   GPGPUsim_ctx_ptr()->g_stream_manager = new stream_manager((GPGPUsim_ctx_ptr()->g_the_gpu), func_sim->g_cuda_launch_blocking);
+   the_gpgpusim->g_the_gpu = new gpgpu_sim(*(the_gpgpusim->g_the_gpu_config), this);
+   the_gpgpusim->g_stream_manager = new stream_manager((the_gpgpusim->g_the_gpu), func_sim->g_cuda_launch_blocking);
 
-   GPGPUsim_ctx_ptr()->g_simulation_starttime = time((time_t *)NULL);
+   the_gpgpusim->g_simulation_starttime = time((time_t *)NULL);
 
-   return GPGPUsim_ctx_ptr()->g_the_gpu;
+   return the_gpgpusim->g_the_gpu;
 }
 
-void start_sim_thread(int api)
+void gpgpu_context::start_sim_thread(int api)
 {
-    if( GPGPUsim_ctx_ptr()->g_sim_done ) {
-    	GPGPUsim_ctx_ptr()->g_sim_done = false;
+    if( the_gpgpusim->g_sim_done ) {
+    	the_gpgpusim->g_sim_done = false;
         if( api == 1 ) {
-           pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,NULL);
+           pthread_create(&(the_gpgpusim->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,(void *)this);
         } else {
-           pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,NULL);
+           pthread_create(&(the_gpgpusim->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,(void *)this);
         }
     }
 }
 
-void print_simulation_time()
+void gpgpu_context::print_simulation_time()
 {
    time_t current_time, difference, d, h, m, s;
    current_time = time((time_t *)NULL);
-   difference = MAX(current_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1);
+   difference = MAX(current_time - the_gpgpusim->g_simulation_starttime, 1);
 
    d = difference/(3600*24);
    h = difference/3600 - 24*d;
@@ -298,16 +282,18 @@ void print_simulation_time()
    fflush(stderr);
    printf("\n\ngpgpu_simulation_time = %u days, %u hrs, %u min, %u sec (%u sec)\n",
           (unsigned)d, (unsigned)h, (unsigned)m, (unsigned)s, (unsigned)difference );
-   printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_insn / difference) );
-   printf("gpgpu_simulation_rate = %u (cycle/sec)\n", (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_cycle / difference) );
+   printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(the_gpgpusim->g_the_gpu->gpu_tot_sim_insn / difference) );
+   const unsigned cycles_per_sec = (unsigned)(the_gpgpusim->g_the_gpu->gpu_tot_sim_cycle / difference);
+   printf("gpgpu_simulation_rate = %u (cycle/sec)\n", cycles_per_sec );
+   printf("gpgpu_silicon_slowdown = %ux\n", the_gpgpusim->g_the_gpu->shader_clock() * 1000 / cycles_per_sec);
    fflush(stdout);
 }
 
-int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid )
+int gpgpu_context::gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid )
 {
-   GPGPUsim_ctx_ptr()->g_the_gpu->launch(grid);
-   sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_start));
-   sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish));
+   the_gpgpusim->g_the_gpu->launch(grid);
+   sem_post(&(the_gpgpusim->g_sim_signal_start));
+   sem_wait(&(the_gpgpusim->g_sim_signal_finish));
    return 0;
 }
 
diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h
index 4a8d796..9f408df 100644
--- a/src/gpgpusim_entrypoint.h
+++ b/src/gpgpusim_entrypoint.h
@@ -76,14 +76,4 @@ class GPGPUsim_ctx {
 
 };
 
-void start_sim_thread(int api);
-
-class gpgpu_sim* g_the_gpu();
-struct GPGPUsim_ctx* GPGPUsim_ctx_ptr();
-class stream_manager* g_stream_manager();
-
-int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid );
-
-void print_simulation_time();
-
 #endif
diff --git a/src/trace-driven/gpgpusim_trace_driven_main.cc b/src/trace-driven/gpgpusim_trace_driven_main.cc
index 3b9851e..4f973cd 100644
--- a/src/trace-driven/gpgpusim_trace_driven_main.cc
+++ b/src/trace-driven/gpgpusim_trace_driven_main.cc
@@ -66,7 +66,7 @@ int main ( int argc, const char **argv )
 				m_gpgpu_sim->deadlock_check();
 			}else {
 				if(m_gpgpu_sim->cycle_insn_cta_max_hit()){
-					g_stream_manager()->stop_all_running_kernels();
+					m_gpgpu_context->the_gpgpusim->g_stream_manager->stop_all_running_kernels();
 					break_limit = true;
 				}
 			}
@@ -81,7 +81,7 @@ int main ( int argc, const char **argv )
 
 		if(sim_cycles) {
 			m_gpgpu_sim->update_stats();
-			print_simulation_time();
+			m_gpgpu_context->print_simulation_time();
 		}
 
 		if(break_limit) {