Merge pull request #30 from abdallm/dev

Dev

11 files changed, 101 insertions, 45 deletions

diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h
index 4721e8a..afac330 100644
--- a/src/cuda-sim/cuda-math.h
+++ b/src/cuda-sim/cuda-math.h
@@ -67,6 +67,8 @@
 #ifndef CUDA_MATH
 #define CUDA_MATH
 
+#include <cmath>
+
 // cuda math implementations
 #undef max
 #undef min
@@ -148,6 +150,7 @@ float __ll2float_rd(long long int a) {
 #include <device_types.h>
 #include <fenv.h>
 
+
 // 32-bit integer to float
 float __int2float_rn(int a) {
    int orig_rnd_mode = fegetround();
@@ -321,7 +324,7 @@ float __internal_accurate_fdividef(float a, float b)
 float __saturatef(float a)
 {
    float b; 
-   if (isnan(a)) b = 0.0f; 
+   if (std::isnan(a)) b = 0.0f; 
    else if (a >= 1.0f) b = 1.0f;
    else if (a <= 0.0f) b = 0.0f; 
    else b = a; 
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index d4ace76..99e054d 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -62,8 +62,8 @@ addr_t g_debug_pc = 0xBEEF1518;
 unsigned g_ptx_sim_num_insn = 0;
 unsigned gpgpu_param_num_shaders = 0;
 
-char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp;
-char *opcode_initiation_int, *opcode_initiation_fp, *opcode_initiation_dp;
+char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu;
+char *opcode_initiation_int, *opcode_initiation_fp, *opcode_initiation_dp,*opcode_initiation_sfu;
 char *cdp_latency_str;
 unsigned cdp_latency[5];
 
@@ -80,6 +80,10 @@ void ptx_opcocde_latency_options (option_parser_t opp) {
 			"Opcode latencies for double precision floating points <ADD,MAX,MUL,MAD,DIV>"
 			"Default 8,8,8,8,335",
 			"8,8,8,8,335");
+	option_parser_register(opp, "-ptx_opcode_latency_sfu", OPT_CSTR, &opcode_latency_sfu,
+			"Opcode latencies for SFU instructions"
+			"Default 8",
+			"8");
 	option_parser_register(opp, "-ptx_opcode_initiation_int", OPT_CSTR, &opcode_initiation_int,
 			"Opcode initiation intervals for integers <ADD,MAX,MUL,MAD,DIV>"
 			"Default 1,1,4,4,32",
@@ -92,6 +96,10 @@ void ptx_opcocde_latency_options (option_parser_t opp) {
 			"Opcode initiation intervals for double precision floating points <ADD,MAX,MUL,MAD,DIV>"
 			"Default 8,8,8,8,130",
 			"8,8,8,8,130");
+	option_parser_register(opp, "-ptx_opcode_initiation_sfu", OPT_CSTR, &opcode_initiation_sfu,
+			"Opcode initiation intervals for sfu instructions"
+			"Default 8",
+			"8");
 	option_parser_register(opp, "-cdp_latency", OPT_CSTR, &cdp_latency_str,
 			"CDP API latency <cudaStreamCreateWithFlags, \
 cudaGetParameterBufferV2_init_perWarp, cudaGetParameterBufferV2_perKernel, \
@@ -393,6 +401,10 @@ void gpgpu_t::memcpy_to_gpu( size_t dst_start_addr, const void *src, size_t coun
    char *src_data = (char*)src;
    for (unsigned n=0; n < count; n ++ ) 
       m_global_mem->write(dst_start_addr+n,1, src_data+n,NULL,NULL);
+
+   // Copy into the performance model.
+   extern gpgpu_sim* g_the_gpu; 
+   g_the_gpu->perf_memcpy_to_gpu(dst_start_addr, count);
    if(g_debug_execution >= 3) {
       printf( " done.\n");
       fflush(stdout);
@@ -408,6 +420,10 @@ void gpgpu_t::memcpy_from_gpu( void *dst, size_t src_start_addr, size_t count )
    unsigned char *dst_data = (unsigned char*)dst;
    for (unsigned n=0; n < count; n ++ ) 
       m_global_mem->read(src_start_addr+n,1,dst_data+n);
+
+   // Copy into the performance model.
+   extern gpgpu_sim* g_the_gpu; 
+   g_the_gpu->perf_memcpy_to_gpu(src_start_addr, count);
    if(g_debug_execution >= 3) {
       printf( " done.\n");
       fflush(stdout);
@@ -589,9 +605,11 @@ void ptx_instruction::set_opcode_and_latency()
 	unsigned int_latency[5];
 	unsigned fp_latency[5];
 	unsigned dp_latency[5];
+	unsigned sfu_latency;
 	unsigned int_init[5];
 	unsigned fp_init[5];
 	unsigned dp_init[5];
+	unsigned sfu_init;
 	/*
 	 * [0] ADD,SUB
 	 * [1] MAX,Min
@@ -608,6 +626,8 @@ void ptx_instruction::set_opcode_and_latency()
 	sscanf(opcode_latency_dp, "%u,%u,%u,%u,%u",
 			&dp_latency[0],&dp_latency[1],&dp_latency[2],
 			&dp_latency[3],&dp_latency[4]);
+	sscanf(opcode_latency_sfu, "%u",
+			&sfu_latency);
 	sscanf(opcode_initiation_int, "%u,%u,%u,%u,%u",
 			&int_init[0],&int_init[1],&int_init[2],
 			&int_init[3],&int_init[4]);
@@ -617,8 +637,10 @@ void ptx_instruction::set_opcode_and_latency()
 	sscanf(opcode_initiation_dp, "%u,%u,%u,%u,%u",
 			&dp_init[0],&dp_init[1],&dp_init[2],
 			&dp_init[3],&dp_init[4]);
+	sscanf(opcode_initiation_sfu, "%u",
+			&sfu_init);
 	sscanf(cdp_latency_str, "%u,%u,%u,%u,%u",
-			&cdp_latency[0],&cdp_latency[1],&cdp_latency[2], 
+			&cdp_latency[0],&cdp_latency[1],&cdp_latency[2],
             &cdp_latency[3],&cdp_latency[4]);
 
 	if(!m_operands.empty()){
@@ -678,6 +700,7 @@ void ptx_instruction::set_opcode_and_latency()
 	   case FF64_TYPE:
 		   latency = dp_latency[0];
 		   initiation_interval = dp_init[0];
+		   op = DP_OP;
 		   break;
 	   case B32_TYPE:
 	   case U32_TYPE:
@@ -699,6 +722,7 @@ void ptx_instruction::set_opcode_and_latency()
 	   case FF64_TYPE:
 		   latency = dp_latency[1];
 		   initiation_interval = dp_init[1];
+		   op = DP_OP;
 		   break;
 	   case B32_TYPE:
 	   case U32_TYPE:
@@ -715,13 +739,12 @@ void ptx_instruction::set_opcode_and_latency()
 	   case F32_TYPE:
 		   latency = fp_latency[2];
 		   initiation_interval = fp_init[2];
-		   op = ALU_SFU_OP;
 		   break;
 	   case F64_TYPE:
 	   case FF64_TYPE:
 		   latency = dp_latency[2];
 		   initiation_interval = dp_init[2];
-		   op = ALU_SFU_OP;
+		   op = DP_OP;
 		   break;
 	   case B32_TYPE:
 	   case U32_TYPE:
@@ -744,6 +767,7 @@ void ptx_instruction::set_opcode_and_latency()
 	   case FF64_TYPE:
 		   latency = dp_latency[3];
 		   initiation_interval = dp_init[3];
+		   op = DP_OP;
 		   break;
 	   case B32_TYPE:
 	   case U32_TYPE:
@@ -779,13 +803,13 @@ void ptx_instruction::set_opcode_and_latency()
 	   break;
    case SQRT_OP: case SIN_OP: case COS_OP: case EX2_OP: case LG2_OP: case RSQRT_OP: case RCP_OP:
 	   //Using double to approximate those
-	  latency = dp_latency[2];
-	  initiation_interval = dp_init[2];
+	  latency = sfu_latency;
+	  initiation_interval = sfu_init;
       op = SFU_OP;
       break;
    case SHFL_OP:
 	   latency = 32;
-	   initiation_interval = 15;
+	   initiation_interval = 4;
 	   break;
    default: 
        break;
@@ -874,6 +898,7 @@ void ptx_instruction::pre_decode()
 
    switch( m_cache_option ) {
    case CA_OPTION: cache_op = CACHE_ALL; break;
+   case NC_OPTION: cache_op = CACHE_L1; break;
    case CG_OPTION: cache_op = CACHE_GLOBAL; break;
    case CS_OPTION: cache_op = CACHE_STREAMING; break;
    case LU_OPTION: cache_op = CACHE_LAST_USE; break;
@@ -1135,13 +1160,13 @@ void function_info::finalize( memory_space *param_mem )
       // copy the parameter over word-by-word so that parameter that crosses a memory page can be copied over
       //Jin: copy parameter using aligned rules
       const size_t word_size = 4; 
-      param_address = (param_address + size - 1) / size * size; //aligned with size 
+      //param_address = (param_address + size - 1) / size * size; //aligned with size TODO: align not correct
       for (size_t idx = 0; idx < size; idx += word_size) {
          const char *pdata = reinterpret_cast<const char*>(param_value.pdata) + idx; // cast to char * for ptr arithmetic
          param_mem->write(param_address + idx, word_size, pdata,NULL,NULL); 
       }
       unsigned offset = p.get_offset();
-      assert(offset == param_address);
+      //assert(offset == param_address);
       param->set_address(param_address);
       param_address += size;
    }
@@ -1408,7 +1433,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id)
    if ( (g_ptx_sim_num_insn % 100000) == 0 ) {
       dim3 ctaid = get_ctaid();
       dim3 tid = get_tid();
-      printf("GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
+      DPRINTF(LIVENESS, "GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
              g_ptx_sim_num_insn, ctaid.x,ctaid.y,ctaid.z,tid.x,tid.y,tid.z );
       fflush(stdout);
    }
@@ -2067,7 +2092,7 @@ struct rec_pts {
    int s_num_recon;
 };
 
-struct std::map<function_info*,rec_pts> g_rpts;
+class std::map<function_info*,rec_pts> g_rpts;
 
 struct rec_pts find_reconvergence_points( function_info *finfo )
 {
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 011c285..e3b8970 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -33,6 +33,7 @@
 #include "ptx.tab.h"
 #include <stdlib.h>
 #include <math.h>
+#include <cmath>
 #include <fenv.h>
 #include "cuda-math.h"
 #include "../abstract_hardware_model.h"
@@ -1961,7 +1962,7 @@ ptx_reg_t d2d( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign,
       y.f64 = x.f64;
       break; 
    }
-   if (isnan(y.f64)) {
+   if (std::isnan(y.f64)) {
       y.u64 = 0xfff8000000000000ull;
    } else if (saturation_mode) {
       y.f64 = cuda_math::__saturatef(y.f64); 
@@ -2086,7 +2087,7 @@ void ptx_round(ptx_reg_t& data, int rounding_mode, int type)
       }
    }
    if ((type == F64_TYPE)||(type == FF64_TYPE)) {
-      if (isnan(data.f64)) {
+      if (std::isnan(data.f64)) {
          data.u64 = 0xfff8000000000000ull;
       }
    }
@@ -2648,12 +2649,12 @@ void mad_def( const ptx_instruction *pI, ptx_thread_info *thread, bool use_carry
 
 bool isNaN(float x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 bool isNaN(double x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 void max_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
diff --git a/src/cuda-sim/ptx.l b/src/cuda-sim/ptx.l
index 5471d6f..af015c9 100644
--- a/src/cuda-sim/ptx.l
+++ b/src/cuda-sim/ptx.l
@@ -36,7 +36,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "ptx.tab.h"
 #include <string.h>
 
-char linebuf[1024];
+char linebuf[4096];
 unsigned col = 0;
 #define TC col+=strlen(ptx_text); 
 #define CHECK_UNSIGNED \
@@ -382,7 +382,7 @@ breakaddr  TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE;
 
 "//"[^\n]* TC;	// eat single
 
-\n.*  col=0; strncpy(linebuf, yytext + 1, 1024); yyless( 1 );
+\n.*  col=0; strncpy(linebuf, yytext + 1, sizeof(linebuf)); yyless( 1 );
 
 " " TC;
 "\t" TC;
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index e00aa4b..2783fc2 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -47,7 +47,6 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 %token  PTR_DIRECTIVE
 %token  ENTRY_DIRECTIVE
 %token  EXTERN_DIRECTIVE
-%token  WEAK_DIRECTIVE
 %token  FILE_DIRECTIVE
 %token  FUNC_DIRECTIVE
 %token  GLOBAL_DIRECTIVE
@@ -228,7 +227,8 @@ function_defn: function_decl { set_symtab($1); func_header(".skip"); } statement
 
 block_spec: MAXNTID_DIRECTIVE INT_OPERAND COMMA INT_OPERAND COMMA INT_OPERAND {func_header_info_int(".maxntid", $2);
 										func_header_info_int(",", $4);
-										func_header_info_int(",", $6); }
+										func_header_info_int(",", $6); 
+                                                                                maxnt_id($2, $4, $6);}
 	| MINNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".minnctapersm", $2); printf("GPGPU-Sim: Warning: .minnctapersm ignored. \n"); }
 	| MAXNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".maxnctapersm", $2); printf("GPGPU-Sim: Warning: .maxnctapersm ignored. \n"); }
 	;
@@ -270,6 +270,7 @@ ptr_spec: /*empty*/
 ptr_space_spec: GLOBAL_DIRECTIVE { add_ptr_spec(global_space); }
               | LOCAL_DIRECTIVE  { add_ptr_spec(local_space); }
               | SHARED_DIRECTIVE { add_ptr_spec(shared_space); }
+			  | CONST_DIRECTIVE { add_ptr_spec(global_space); }
 
 ptr_align_spec: ALIGN_DIRECTIVE INT_OPERAND
 
@@ -331,6 +332,7 @@ var_spec_list: var_spec
 var_spec: space_spec 
 	| type_spec
 	| align_spec
+	| VISIBLE_DIRECTIVE
 	| EXTERN_DIRECTIVE { add_extern_spec(); }
     | WEAK_DIRECTIVE
 	;
@@ -513,7 +515,8 @@ compare_spec:EQ_OPTION { add_option(EQ_OPTION); }
 	| NAN_OPTION { add_option(NAN_OPTION); } 
 	;
 
-operand_list: operand
+operand_list: /* empty*/
+    | operand
 	| operand COMMA operand_list;
 
 operand: IDENTIFIER  { add_scalar_operand( $1 ); }
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 8ebdcf8..ba6d7ed 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -222,6 +222,7 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
    } else {
       *func_info = new function_info(entry_point);
       (*func_info)->set_name(name);
+      (*func_info)->set_maxnt_id(0);
       m_function_info_lookup[key] = *func_info;
    }
 
@@ -1210,6 +1211,7 @@ ptx_instruction::ptx_instruction( int opcode,
 	  case EXTP_OPTION:
 		 break;
 	  case NC_OPTION:
+		 m_cache_option = last_ptx_inst_option;
 		 break;
 	  case UP_OPTION:
 	  case DOWN_OPTION:
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 9ad1571..f4d3641 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -105,7 +105,7 @@ private:
    int m_is_function;
    bool m_is_non_arch_reg;
 
-   friend class type_info_key_compare;
+   friend struct type_info_key_compare;
 };
 
 class symbol_table;
@@ -1245,6 +1245,7 @@ public:
 
    const struct gpgpu_ptx_sim_info* get_kernel_info () const
    {
+      assert (m_kernel_info.maxthreads == maxnt_id);
       return &m_kernel_info;
    }
 
@@ -1252,6 +1253,8 @@ public:
       m_kernel_info = info;
       m_kernel_info.ptx_version = 10*get_ptx_version().ver();
       m_kernel_info.sm_target = get_ptx_version().target();
+      // THIS DEPENDS ON ptxas being called after the PTX is parsed.
+      m_kernel_info.maxthreads = maxnt_id;
    }
    symbol_table *get_symtab()
    {
@@ -1275,7 +1278,11 @@ public:
    }
    bool is_entry_point() const { return m_entry_point; }
 
+   void set_maxnt_id(unsigned maxthreads) { maxnt_id = maxthreads;}
+   unsigned get_maxnt_id() { return maxnt_id;}
+
 private:
+   unsigned maxnt_id;
    unsigned m_uid;
    unsigned m_local_mem_framesize;
    bool m_entry_point;
diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc
index 6c1b595..11d9013 100644
--- a/src/cuda-sim/ptx_loader.cc
+++ b/src/cuda-sim/ptx_loader.cc
@@ -54,6 +54,7 @@ extern int ptxinfo_debug;
 extern FILE *ptxinfo_in;
 
 static bool g_save_embedded_ptx;
+static int g_occupancy_sm_number;
 bool g_keep_intermediate_files;
 bool m_ptx_save_converted_ptxplus;
 
@@ -71,6 +72,10 @@ void ptx_reg_options(option_parser_t opp)
                 &m_ptx_save_converted_ptxplus,
                 "Saved converted ptxplus to a file",
                 "0");
+   option_parser_register(opp, "-gpgpu_occupancy_sm_number", OPT_INT32, &g_occupancy_sm_number,
+                "The SM number to pass to ptxas when getting register usage for computing GPU occupancy. "
+                "This parameter is required in the config.",
+                "0");
 }
 
 void print_ptx_file( const char *p, unsigned source_num, const char *filename )
@@ -120,7 +125,7 @@ char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptxfilenam
 	fflush(stdout);
 	printf("GPGPU-Sim PTX: calling cuobjdump_to_ptxplus\ncommandline: %s\n", commandline);
 	result = system(commandline);
-	if(result){printf("GPGPU-Sim PTX: ERROR ** could not execute %s\n", commandline); exit(1);}
+	if(result){fprintf(stderr, "GPGPU-Sim PTX: ERROR ** could not execute %s\n", commandline); exit(1);}
 
 
 	// Get ptxplus from file
@@ -142,7 +147,7 @@ char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptxfilenam
 		printf("GPGPU-Sim PTX: removing temporary files using \"%s\"\n", rm_commandline);
 		int rm_result = system(rm_commandline);
 		if( rm_result != 0 ) {
-			printf("GPGPU-Sim PTX: ERROR ** while removing temporary files %d\n", rm_result);
+			fprintf(stderr, "GPGPU-Sim PTX: ERROR ** while removing temporary files %d\n", rm_result);
 			exit(1);
 		}
 	}
@@ -193,7 +198,7 @@ void fix_duplicate_errors(char fname2[1024]) {
 	printf("Running: %s\n", commandline);
 	int result = system(commandline);
 	if (result != 0) {
-		printf("GPGPU-Sim PTX: ERROR ** while changing filename from %s to %s", fname2, tempfile);
+		fprintf(stderr, "GPGPU-Sim PTX: ERROR ** while changing filename from %s to %s", fname2, tempfile);
 		exit(1);
 	}
 
@@ -282,12 +287,12 @@ void fix_duplicate_errors(char fname2[1024]) {
 	printf("Running: %s\n", commandline);
 	result = system(commandline);
 	if (result != 0) {
-		printf("GPGPU-Sim PTX: ERROR ** while deleting %s", tempfile);
+		fprintf(stderr, "GPGPU-Sim PTX: ERROR ** while deleting %s", tempfile);
 		exit(1);
 	}
 }
 
-void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version )
+void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num )
 {
     char fname[1024];
     snprintf(fname,1024,"_ptx_XXXXXX");
@@ -308,9 +313,9 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num
     printf("Running: %s\n", commandline2);
     int result = system(commandline2);
     if( result != 0 ) {
-       printf("GPGPU-Sim PTX: ERROR ** while loading PTX (a) %d\n", result);
-       printf("               Ensure you have write access to simulation directory\n");
-       printf("               and have \'cat\' and \'sed\' in your path.\n");
+       fprintf(stderr, "GPGPU-Sim PTX: ERROR ** while loading PTX (a) %d\n", result);
+       fprintf(stderr, "               Ensure you have write access to simulation directory\n");
+       fprintf(stderr, "               and have \'cat\' and \'sed\' in your path.\n");
        exit(1);
     }
 
@@ -321,15 +326,20 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num
     extra_flags[0]=0;
 
 #if CUDART_VERSION >= 3000
-    if (sm_version == 0) sm_version = 20;
+    if ( g_occupancy_sm_number == 0 ) {
+        fprintf( stderr, "gpgpusim.config must specify the sm version for the GPU that you use to compute occupancy \"-gpgpu_occupancy_sm_number XX\".\n"
+                         "The register file size is specifically tied to the sm version used to querry ptxas for register usage.\n"
+                         "A register size/SM mismatch may result in occupancy differences." );
+        exit(1);
+    }
     extern bool g_cdp_enabled;
     if(!g_cdp_enabled)
-        snprintf(extra_flags,1024,"--gpu-name=sm_%u",sm_version);
+        snprintf(extra_flags,1024,"--gpu-name=sm_%u", g_occupancy_sm_number);
     else
-        snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",sm_version);
+        snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",g_occupancy_sm_number);
 #endif
 
-    snprintf(commandline,1024,"$CUDA_INSTALL_PATH/bin/ptxas %s -v %s --output-file  /dev/null 2> %s",
+    snprintf(commandline,1024,"$PTXAS_CUDA_INSTALL_PATH/bin/ptxas %s -v %s --output-file  /dev/null 2> %s",
              extra_flags, fname2, tempfile_ptxinfo);
     printf("GPGPU-Sim PTX: generating ptxinfo using \"%s\"\n", commandline);
     result = system(commandline);
@@ -347,8 +357,8 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num
     		result = system(commandline);
 	}
 	if (result != 0) {
-		printf("GPGPU-Sim PTX: ERROR ** while loading PTX (b) %d\n", result);
-		printf("               Ensure ptxas is in your path.\n");
+		fprintf(stderr, "GPGPU-Sim PTX: ERROR ** while loading PTX (b) %d\n", result);
+		fprintf(stderr, "               Ensure ptxas is in your path.\n");
 		exit(1);
 	}
     }
@@ -362,7 +372,7 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num
         printf("GPGPU-Sim PTX: removing ptxinfo using \"%s\"\n", commandline);
         result = system(commandline);
         if( result != 0 ) {
-    	    printf("GPGPU-Sim PTX: ERROR ** while loading PTX (c) %d\n", result);
+    	    fprintf(stderr, "GPGPU-Sim PTX: ERROR ** while loading PTX (c) %d\n", result);
     	    exit(1);
         }
     }
diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h
index d3d0c92..cb02eda 100644
--- a/src/cuda-sim/ptx_loader.h
+++ b/src/cuda-sim/ptx_loader.h
@@ -32,7 +32,7 @@
 extern bool g_override_embedded_ptx;
  
 class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num );
-void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20 );
+void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num );
 char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str);
 bool keep_intermediate_files();
 
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index baa3bcd..6dc1f40 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -265,7 +265,7 @@ void parse_assert_impl( int test_value, const char *file, unsigned line, const c
       parse_error_impl(file,line, msg);
 }
 
-extern char linebuf[1024];
+extern char linebuf[4096];
 
 
 void set_return()
@@ -433,7 +433,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
       assert( (num_bits%8) == 0  );
       addr = g_current_symbol_table->get_shared_next();
       addr_pad = pad_address(addr, num_bits/8, 128);
-      printf("from 0x%x to 0x%lx (shared memory space)\n",
+      printf("from 0x%llx to 0x%llx (shared memory space)\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8);
          fflush(stdout);
@@ -450,7 +450,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
          assert( (num_bits%8) == 0  ); 
          addr = g_current_symbol_table->get_global_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (global memory space) %u\n",
+         printf("from 0x%llx to 0x%llx (global memory space) %u\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8,
               g_const_alloc++);
@@ -471,7 +471,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
       assert( (num_bits%8) == 0  );
       addr = g_current_symbol_table->get_global_next();
       addr_pad = pad_address(addr, num_bits/8, 128);
-      printf("from 0x%x to 0x%lx (global memory space)\n",
+      printf("from 0x%llx to 0x%llx (global memory space)\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8);
       fflush(stdout);
@@ -488,7 +488,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
          assert( (num_bits%8) == 0  );
          addr = g_current_symbol_table->get_local_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (local memory space)\n",
+         printf("from 0x%llx to 0x%llx (local memory space)\n",
                  addr+addr_pad,
                  addr+addr_pad + num_bits/8);
          fflush(stdout);
@@ -501,7 +501,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
         assert( (num_bits%8) == 0 );
         addr = g_current_symbol_table->get_local_next();
         addr_pad = pad_address(addr, num_bits/8, 128);
-        printf("from 0x%x to 0x%lx\n",
+        printf("from 0x%llx to 0x%llx\n",
                 addr+addr_pad,
                 addr+addr_pad + num_bits/8);
         fflush(stdout);
@@ -969,6 +969,10 @@ void target_header3(char* a, char* b, char* c)
    g_global_symbol_table->set_sm_target(a,b,c);
 }
 
+void maxnt_id(int x, int y, int z) {
+  g_func_info->set_maxnt_id(x * y * z);
+}
+
 void func_header(const char* a) {} //intentional dummy function
 void func_header_info(const char* a) {} //intentional dummy function
 void func_header_info_int(const char* a, int b) {} //intentional dummy function
diff --git a/src/cuda-sim/ptx_parser.h b/src/cuda-sim/ptx_parser.h
index 32f3903..13042e1 100644
--- a/src/cuda-sim/ptx_parser.h
+++ b/src/cuda-sim/ptx_parser.h
@@ -93,6 +93,7 @@ void change_double_operand_type( int addr_type );
 void change_operand_neg( );
 void set_immediate_operand_type( );
 void version_header(double a);
+void maxnt_id(int x, int y, int z);
 
 //Jin: handle instructino group for cdp
 void start_inst_group();