- template runs/passes with CUDA 3.1, but requires that CUDA 2.3

is installed for math intrinsics used in instructions.cc (since
CUDA 3.1 does not seem to define __uint2float).
- currently cuda-math.h has hardcoded paths to my own install of
CUDA 2.3 (need to determine whether we can build simulator with
CUDA 2.3 and run CUDA 3.1 / compute_2x PTX, or need to make 
current setup more portable)
- modified ptxinfo.l to print out meaningful error message if 
error occurs on first line
- modified ptxinfo.l,y to understand "for 'compute_NN'" syntax
- updated template to 3.1 SDK version  
- updated cuda_runtime_api.cc to select highest compute capability
version when multiple versions are present.
- updated template/BlackScholes Makefiles to generate a "DLL build"
instead of static linked build (still confusing... requires one to
copy config files... probably need to modify config scripts)

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 6837]

4 files changed, 29 insertions, 20 deletions

diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h
index ea7f5fd..5abf4f3 100644
--- a/src/cuda-sim/cuda-math.h
+++ b/src/cuda-sim/cuda-math.h
@@ -47,7 +47,6 @@ namespace cuda_math {
 #define __attribute__(a) // to remove warnings inside math_functions.h
 #undef INT_MAX
 
-#if CUDART_VERSION < 3000
 // DEVICE_BUILTIN
    struct int4 {
       int x, y, z, w;
@@ -70,27 +69,21 @@ namespace cuda_math {
    typedef struct float2 float2;
 
 extern float rsqrtf(float); // CUDA 2.3 beta
-#else
-#define UINT_MAX ((unsigned int)-1)
-#endif
 
+#if CUDART_VERSION < 3000
 #define CUDA_FLOAT_MATH_FUNCTIONS
 #include <device_types.h>
 #define __CUDA_INTERNAL_COMPILATION__
-#if CUDART_VERSION >= 3000
-#define __CUDACC__
-#include <device_functions.h>
-#else
 #include <math_functions.h>
-#endif
 #undef __CUDA_INTERNAL_COMPILATION__
 #undef __attribute__
-
-#if CUDART_VERSION >= 3000
-#define __internal_float2int float2int
-#define __internal_float2uint float2uint
-#include <math.h>
-#define __internal_accurate_fdividef fdividef
+#else
+#define CUDA_FLOAT_MATH_FUNCTIONS
+#include "/home/taamodt/nvcuda/2.3/cuda/include/device_types.h"
+#define __CUDA_INTERNAL_COMPILATION__
+#include "/home/taamodt/nvcuda/2.3/cuda/include/math_functions.h"
+#undef __CUDA_INTERNAL_COMPILATION__
+#undef __attribute__
 #endif
 
 }
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index d691a64..4480221 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -1617,6 +1617,7 @@ static FILE *open_ptxinfo (const char* ptx_filename)
 
 struct ptx_info_t {
     char *str;
+    char *fname;
     ptx_info_t *next;
 };
 
@@ -1624,11 +1625,12 @@ ptx_info_t *g_ptx_source_array = NULL;
 unsigned g_used_embedded_ptx_files;
 extern double g_ptx_version;
 
-void gpgpu_ptx_sim_add_ptxstring( const char *ptx_string )
+void gpgpu_ptx_sim_add_ptxstring( const char *ptx_string, const char *sourcefname )
 {
     ptx_info_t *t = new ptx_info_t;
     t->next = NULL;
     t->str = strdup(ptx_string);
+    t->fname = strdup(sourcefname);
 
     // put ptx source into a fifo
     if (g_ptx_source_array == NULL) {
@@ -1732,7 +1734,13 @@ void gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num )
     char tempfile_ptxinfo[1024];
     snprintf(tempfile_ptxinfo,1024,"%sinfo",fname);
     char commandline[1024];
-    snprintf(commandline,1024,"ptxas -v %s --output-file /dev/null 2> %s", fname2, tempfile_ptxinfo);
+    char extra_flags[1024];
+    extra_flags[0]=0;
+#if CUDART_VERSION >= 3000
+    snprintf(extra_flags,1024,"--gpu-name=sm_20");
+#endif
+    snprintf(commandline,1024,"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);
     if( result != 0 ) {
diff --git a/src/cuda-sim/ptxinfo.l b/src/cuda-sim/ptxinfo.l
index 3a2ab52..bec382b 100644
--- a/src/cuda-sim/ptxinfo.l
+++ b/src/cuda-sim/ptxinfo.l
@@ -67,9 +67,13 @@
 #include "ptxinfo.tab.h"
 #include <string.h>
 
-char ptxinfo_linebuf[1024];
+#define LINEBUF_SIZE 1024
+char ptxinfo_linebuf[LINEBUF_SIZE];
 unsigned ptxinfo_col = 0;
-#define TC ptxinfo_col+=strlen(ptxinfo_text); 
+#define TC if( (ptxinfo_lineno == 1) && ((ptxinfo_col + strlen(ptxinfo_text)) < LINEBUF_SIZE) ) { \
+		strncpy(ptxinfo_linebuf+ptxinfo_col,ptxinfo_text,strlen(ptxinfo_text)); \
+	   } \
+	   ptxinfo_col+=strlen(ptxinfo_text); 
 %}
 
 %%
@@ -85,6 +89,7 @@ unsigned ptxinfo_col = 0;
 "smem"		TC; return SMEM;
 "cmem"		TC; return CMEM;
 "line"		TC; return LINE;
+"for"		TC; return FOR;
 
 [_A-Za-z$%][_0-9A-Za-z$]*  TC; ptxinfo_lval.string_value = strdup(yytext); return IDENTIFIER;
 [-]{0,1}[0-9]+	 TC; ptxinfo_lval.int_value =  atoi(yytext); return INT_OPERAND;
@@ -112,8 +117,9 @@ int ptxinfo_error( const char *s )
 	int i;
 	g_ptxinfo_error_detected = 1;
 	fflush(stdout);
+	printf("GPGPU-Sim: ERROR while parsing output of ptxas (used to capture resource usage information)\n");
 	if( s != NULL )
-		printf("ptxas -v %s (%s:%u) Syntax error:\n\n", g_filename, g_ptxinfo_filename, ptxinfo_lineno );
+		printf("GPGPU-Sim:     %s (%s:%u) Syntax error:\n\n", g_filename, g_ptxinfo_filename, ptxinfo_lineno );
 	printf("   %s\n", ptxinfo_linebuf );
 	printf("   ");
 	for( i=0; i < ptxinfo_col-1; i++ ) {
diff --git a/src/cuda-sim/ptxinfo.y b/src/cuda-sim/ptxinfo.y
index c9c84b8..6d2f7a7 100644
--- a/src/cuda-sim/ptxinfo.y
+++ b/src/cuda-sim/ptxinfo.y
@@ -86,6 +86,7 @@
 %token QUOTE
 %token LINE
 %token WARNING
+%token FOR
 
 %{
 	#include <stdlib.h>
@@ -118,6 +119,7 @@ line_info: function_name
 	;
 
 function_name: FUNC QUOTE IDENTIFIER QUOTE { ptxinfo_function($3); }
+	|  FUNC QUOTE IDENTIFIER QUOTE FOR QUOTE IDENTIFIER QUOTE { ptxinfo_function($3); }
 
 function_info: info
 	| function_info COMMA info