WIP adding support for PTX JIT and dumping params to cudaLaunches

5 files changed, 93 insertions, 0 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 34368ce..6875edd 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -1226,6 +1226,87 @@ void function_info::list_param( FILE *fout ) const
    fflush(fout);
 }
 
+void function_info::debug_param( ) const
+{
+   char filename[] = "params.txt";
+   char buff[1024];
+   snprintf(buff,1024,"c++filt %s > %s", get_name().c_str(), filename);
+   system(buff);
+   FILE *fp = fopen(filename, "r");
+   fgets(buff, 1024, fp);
+   fclose(fp);
+
+   std::string fn(buff);
+   size_t pos1, pos2;
+   pos1 = fn.find("(");
+   pos2 = fn.find(")");
+   assert(pos2>pos1&&pos1>0);
+   strcpy(buff, fn.substr(pos1 + 1, pos2 - pos1 - 1).c_str());
+   printf("params: %s\n", buff);
+   char *tok;
+   std::vector<std::string> params;
+   tok = strtok(buff, ",");
+   while(tok!=NULL){
+       std::string param(tok);
+       param.erase(0, param.find_first_not_of(" "));
+       param.erase(param.find_last_not_of(" ")+1);
+       params.push_back(param);
+       tok = strtok(NULL, ",");
+   }
+   for (auto const& it : params){
+      std::cout<<it<<std::endl;
+   }
+
+   FILE *fout  = fopen (filename, "w");
+   fprintf(fout, "Name of function:%s\n", fn.c_str());
+
+   for( std::map<unsigned,param_info>::const_iterator i=m_ptx_kernel_param_info.begin(); i!=m_ptx_kernel_param_info.end(); i++ ) {
+      const param_info &p = i->second;
+      std::string name = p.get_name();
+      param_t param_value = p.get_value();
+      if(params[i->first].find("const")!=std::string::npos){
+         fprintf(fout, "Input: ");
+      } else {
+         fprintf(fout, "Input/output: ");
+      }
+
+      symbol *param = m_symtab->lookup(name.c_str());
+      addr_t param_addr = param->get_address();
+      fprintf(fout, "%s: %#08x, ", name.c_str(), param_addr);
+
+      if(params[i->first].find("int")!=std::string::npos){
+         size_t len = param_value.size/sizeof(int);
+         int val[len];
+         memcpy((void*) val, param_value.pdata+param_value.offset, param_value.size);
+         fprintf(fout, "val (int) = ");
+         for (unsigned i = 0; i<len; i++){
+             fprintf(fout, "%d ", val[i]);
+         }
+         fprintf(fout, "\n");
+      } else if(params[i->first].find("float")!=std::string::npos){
+         size_t len = param_value.size/sizeof(float);
+         float val[len];
+         memcpy((void*) val, param_value.pdata+param_value.offset, param_value.size);
+         fprintf(fout, "val (float) = ");
+         for (unsigned i = 0; i<len; i++){
+             fprintf(fout, "%f ", val[i]);
+         }
+         fprintf(fout, "\n");
+      }else{
+         size_t len = param_value.size/sizeof(char);
+         char val[len];
+         memcpy((void*) val, param_value.pdata+param_value.offset, param_value.size);
+         fprintf(fout, "val (char) = ");
+         for (unsigned i = 0; i<len; i++){
+             fprintf(fout, "%c ", val[i]);
+         }
+         fprintf(fout, "\n");
+      }
+   }
+   fflush(fout);
+   fclose(fout);
+}
+
 template<int activate_level> 
 bool ptx_debug_exec_dump_cond(int thd_uid, addr_t pc)
 {
diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h
index 958daba..9049a84 100644
--- a/src/cuda-sim/cuda-sim.h
+++ b/src/cuda-sim/cuda-sim.h
@@ -32,6 +32,7 @@
 #include"../gpgpu-sim/shader.h"
 #include <stdlib.h>
 #include <map>
+#include <vector>
 #include <string>
 #include"ptx_sim.h"
 
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index c77e4da..034a7b9 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -2753,6 +2753,7 @@ void mov_impl( const ptx_instruction *pI, ptx_thread_info *thread )
    const operand_info &dst  = pI->dst();
    const operand_info &src1 = pI->src1();
    unsigned i_type = pI->get_type();
+   assert( src1.is_param_local() == 0 );
 
    if( (src1.is_vector() || dst.is_vector()) && (i_type != BB64_TYPE) && (i_type != BB128_TYPE) && (i_type != FF64_TYPE) ) {
       // pack or unpack operation
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 016c600..482b9e0 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -257,6 +257,14 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
    return prior_decl;
 }
 
+function_info *symbol_table::lookup_function( std::string name )
+{
+   std::string key = std::string(name);
+   std::map<std::string,function_info*>::iterator it = m_function_info_lookup.find(key);
+   assert ( it != m_function_info_lookup.end() );
+   return it->second;
+}
+
 type_info *symbol_table::add_type( memory_space_t space_spec, int scalar_type_spec, int vector_spec, int alignment_spec, int extern_spec )
 {
    if( space_spec == param_space_unclassified ) 
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 5b68fcf..341f9b7 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -313,6 +313,7 @@ public:
    symbol *add_variable( const char *identifier, const type_info *type, unsigned size, const char *filename, unsigned line );
    void add_function( function_info *func, const char *filename, unsigned linenumber );
    bool add_function_decl( const char *name, int entry_point, function_info **func_info, symbol_table **symbol_table );
+   function_info *lookup_function(std::string name);
    type_info *add_type( memory_space_t space_spec, int scalar_type_spec, int vector_spec, int alignment_spec, int extern_spec );
    type_info *add_type( function_info *func );
    type_info *get_array_type( type_info *base_type, unsigned array_dim ); 
@@ -1256,6 +1257,7 @@ public:
    void finalize( memory_space *param_mem );
    void param_to_shared( memory_space *shared_mem, symbol_table *symtab ); 
    void list_param( FILE *fout ) const;
+   void debug_param() const;
 
    const struct gpgpu_ptx_sim_info* get_kernel_info () const
    {