Move g_pc_to_finfo

Signed-off-by: Mengchi Zhang <[email protected]>

9 files changed, 34 insertions, 23 deletions

diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index 248e7a5..ef09051 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -39,6 +39,7 @@
 #include <sys/stat.h>
 #include <sstream>
 #include <iostream>
+#include "../libcuda/gpgpu_context.h"
 
 unsigned mem_access_t::sm_next_access_uid = 0;   
 unsigned warp_inst_t::sm_next_uid = 0;
@@ -945,7 +946,7 @@ void simt_stack::print (FILE *fout) const
         } else {
             fprintf(fout," " );
         }
-        ptx_print_insn( stack_entry.m_pc, fout );
+        m_gpu->gpgpu_ctx->func_sim->ptx_print_insn( stack_entry.m_pc, fout );
         fprintf(fout,"\n");
     }
 
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index da29a11..8ef8376 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -347,9 +347,11 @@ public:
    mutable bool volta_cache_config_set;
 };
 
-struct core_config {
-    core_config() 
-    { 
+class core_config {
+    public:
+    core_config(gpgpu_context* ctx)
+    {
+	gpgpu_ctx = ctx;
         m_valid = false; 
         num_shmem_bank=16; 
         shmem_limited_broadcast = false; 
@@ -361,6 +363,8 @@ struct core_config {
 
     bool m_valid;
     unsigned warp_size;
+    // backward pointer
+    class gpgpu_context* gpgpu_ctx;
 
     // off-chip memory request architecture parameters
     int gpgpu_coalesce_arch;
@@ -934,7 +938,7 @@ public:
         m_empty=true; 
         m_config=NULL; 
     }
-    warp_inst_t( const core_config *config ) 
+    warp_inst_t( const core_config *config )
     { 
         m_uid=0;
         assert(config->warp_size<=MAX_WARP_SIZE); 
@@ -1105,7 +1109,6 @@ public:
     unsigned get_uid() const { return m_uid; }
     unsigned get_schd_id() const { return m_scheduler_id; }
 
-
 protected:
 
     unsigned m_uid;
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 3d1da62..fb9bc9e 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -220,7 +220,6 @@ void gpgpu_t::gpgpu_ptx_sim_unbindTexture(const struct textureReference* texref)
 }
 
 unsigned g_assemble_code_next_pc=0; 
-std::map<unsigned,function_info*> g_pc_to_finfo;
 std::vector<ptx_instruction*> function_info::s_g_pc_to_insn;
 
 #define MAX_INST_SIZE 8 /*bytes*/
@@ -257,7 +256,7 @@ void function_info::ptx_assemble()
          const symbol *l = pI->get_label();
          labels[l->name()] = n;
       } else {
-         g_pc_to_finfo[PC] = this;
+         gpgpu_ctx->func_sim->g_pc_to_finfo[PC] = this;
          m_instr_mem[n] = pI;
          s_g_pc_to_insn.push_back(pI);
          assert(pI == s_g_pc_to_insn[PC]);
@@ -497,7 +496,7 @@ void gpgpu_t::gpu_memset( size_t dst_start_addr, int c, size_t count )
    }
 }
 
-void ptx_print_insn( address_type pc, FILE *fp )
+void cuda_sim::ptx_print_insn( address_type pc, FILE *fp )
 {
    std::map<unsigned,function_info*>::iterator f = g_pc_to_finfo.find(pc);
    if( f == g_pc_to_finfo.end() ) {
@@ -509,7 +508,7 @@ void ptx_print_insn( address_type pc, FILE *fp )
    finfo->print_insn(pc,fp);
 }
 
-std::string ptx_get_insn_str( address_type pc )
+std::string cuda_sim::ptx_get_insn_str( address_type pc )
 {
    std::map<unsigned,function_info*>::iterator f = g_pc_to_finfo.find(pc);
    if( f == g_pc_to_finfo.end() ) {
diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h
index 5c95100..c578524 100644
--- a/src/cuda-sim/cuda-sim.h
+++ b/src/cuda-sim/cuda-sim.h
@@ -61,8 +61,6 @@ unsigned ptx_sim_init_thread( kernel_info_t &kernel,
                               bool functionalSimulationMode = false);
 const warp_inst_t *ptx_fetch_inst( address_type pc );
 const struct gpgpu_ptx_sim_info* ptx_sim_kernel_info(const class function_info *kernel);
-void ptx_print_insn( address_type pc, FILE *fp );
-std::string ptx_get_insn_str( address_type pc );
 
 
 /*!
@@ -162,6 +160,7 @@ class cuda_sim {
 	void ** g_inst_op_classification_stat;
 	std::set<std::string>   g_globals;
 	std::set<std::string>   g_constants;
+	std::map<unsigned,function_info*> g_pc_to_finfo;
 	// backward pointer
 	class gpgpu_context* gpgpu_ctx;
 	//global functions
@@ -181,6 +180,8 @@ class cuda_sim {
 	struct rec_pts find_reconvergence_points( function_info *finfo );
 	address_type get_converge_point( address_type pc );
 	void   gpgpu_ptx_sim_memcpy_symbol(const char *hostVar, const void *src, size_t count, size_t offset, int to, gpgpu_t *gpu );
+	void ptx_print_insn( address_type pc, FILE *fp );
+	std::string ptx_get_insn_str( address_type pc );
 };
 
 #endif
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 1bd409e..c537091 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -83,8 +83,9 @@ symbol_table::symbol_table()
    assert(0); 
 }
 
-symbol_table::symbol_table( const char *scope_name, unsigned entry_point, symbol_table *parent )
+symbol_table::symbol_table( const char *scope_name, unsigned entry_point, symbol_table *parent, gpgpu_context* ctx )
 {
+   gpgpu_ctx = ctx;
    m_scope_name = std::string(scope_name);
    m_reg_allocator=0;
    m_shared_next = 0;
@@ -183,7 +184,7 @@ symbol_table* symbol_table::start_inst_group() {
 
    //previous added
    assert(m_inst_group_symtab.find(std::string(inst_group_name)) == m_inst_group_symtab.end());
-   symbol_table *sym_table = new symbol_table(inst_group_name, 3/*inst group*/, this );
+   symbol_table *sym_table = new symbol_table(inst_group_name, 3/*inst group*/, this, gpgpu_ctx );
  
    sym_table->m_global_next = m_global_next;
    sym_table->m_shared_next = m_shared_next;
@@ -221,7 +222,7 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
       *func_info = m_function_info_lookup[key];
       prior_decl = true;
    } else {
-      *func_info = new function_info(entry_point);
+      *func_info = new function_info(entry_point, gpgpu_ctx);
       (*func_info)->set_name(name);
       (*func_info)->set_maxnt_id(0);
       m_function_info_lookup[key] = *func_info;
@@ -232,7 +233,7 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
       *sym_table = m_function_symtab_lookup[key];
    } else {
       assert( !prior_decl );
-      *sym_table = new symbol_table( "", entry_point, this );
+      *sym_table = new symbol_table( "", entry_point, this, gpgpu_ctx );
       
       // Initial setup code to support a register represented as "_".
       // This register is used when an instruction operand is
@@ -1373,8 +1374,9 @@ std::string ptx_instruction::to_string() const
 
 unsigned function_info::sm_next_uid = 1;
 
-function_info::function_info(int entry_point ) 
+function_info::function_info(int entry_point, gpgpu_context* ctx )
 {
+   gpgpu_ctx = ctx;
    m_uid = sm_next_uid++;
    m_entry_point = (entry_point==1)?true:false;
    m_extern = (entry_point==2)?true:false;
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 1604551..babd54b 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -306,7 +306,7 @@ private:
 class symbol_table {
 public:
    symbol_table();
-   symbol_table( const char *scope_name, unsigned entry_point, symbol_table *parent );
+   symbol_table( const char *scope_name, unsigned entry_point, symbol_table *parent, gpgpu_context* ctx);
    void set_name( const char *name );
    const ptx_version &get_ptx_version() const;
    unsigned get_sm_target() const;
@@ -348,6 +348,9 @@ public:
    symbol_table* start_inst_group();
    symbol_table* end_inst_group();
 
+   // backward pointer
+   class gpgpu_context* gpgpu_ctx;
+
 private:
    unsigned m_reg_allocator;
    unsigned m_shared_next;
@@ -1233,7 +1236,7 @@ private:
 
 class function_info {
 public:
-   function_info(int entry_point );
+   function_info(int entry_point, gpgpu_context* ctx );
    const ptx_version &get_ptx_version() const { return m_symtab->get_ptx_version(); }
    unsigned get_sm_target() const { return m_symtab->get_sm_target(); }
    bool is_extern() const { return m_extern; }
@@ -1403,6 +1406,8 @@ public:
 
    void set_maxnt_id(unsigned maxthreads) { maxnt_id = maxthreads;}
    unsigned get_maxnt_id() { return maxnt_id;}
+   // backward pointer
+   class gpgpu_context* gpgpu_ctx;
 
 private:
    unsigned maxnt_id;
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index 269ec4d..5a94679 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -112,7 +112,7 @@ symbol_table * gpgpu_context::init_parser( const char *ptx_filename )
 {
    g_filename = strdup(ptx_filename);
    if  (g_global_allfiles_symbol_table == NULL) {
-       g_global_allfiles_symbol_table = new symbol_table("global_allfiles", 0, NULL);
+       g_global_allfiles_symbol_table = new symbol_table("global_allfiles", 0, NULL, this);
        ptx_parser->g_global_symbol_table = ptx_parser->g_current_symbol_table = g_global_allfiles_symbol_table;
    }
    /*else {
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 69790fc..6cd6d8f 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -1025,7 +1025,7 @@ void scheduler_unit::cycle()
             m_simt_stack[warp_id]->get_pdom_stack_top_info(&pc,&rpc);
             SCHED_DPRINTF( "Warp (warp_id %u, dynamic_warp_id %u) has valid instruction (%s)\n",
                            (*iter)->get_warp_id(), (*iter)->get_dynamic_warp_id(),
-                           ptx_get_insn_str( pc).c_str() );
+                           m_shader->m_config->gpgpu_ctx->func_sim->ptx_get_insn_str( pc).c_str() );
             if( pI ) {
                 assert(valid);
                 if( pc != pI->pc ) {
@@ -2690,7 +2690,7 @@ void warp_inst_t::print( FILE *fout ) const
     for (unsigned j=0; j<m_config->warp_size; j++)
         fprintf(fout, "%c", (active(j)?'1':'0') );
     fprintf(fout, "]: ");
-    ptx_print_insn( pc, fout );
+    m_config->gpgpu_ctx->func_sim->ptx_print_insn( pc, fout );
     fprintf(fout, "\n");
 }
 void shader_core_ctx::incexecstat(warp_inst_t *&inst)
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index e0cefac..2837f1b 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1367,7 +1367,7 @@ const char* const pipeline_stage_name_decode[] = {
 class shader_core_config : public core_config
 {
     public:
-    shader_core_config(gpgpu_context* ctx){
+    shader_core_config(gpgpu_context* ctx):core_config(ctx){
 	pipeline_widths_string = NULL;
 	gpgpu_ctx = ctx;
     }