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

7 files changed, 60 insertions, 80 deletions

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);