Move opcode_latency_int thus pass gpgpu_context into many classes

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

16 files changed, 62 insertions, 34 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 261d605..df0bbd7 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -51,6 +51,7 @@ typedef void * yyscan_t;
 #include "decuda_pred_table/decuda_pred_table.h"
 #include "../stream_manager.h"
 #include "cuda_device_runtime.h"
+#include "../../libcuda/gpgpu_context.h"
 
 int gpgpu_ptx_instruction_classification;
 void ** g_inst_classification_stat = NULL;
@@ -66,12 +67,12 @@ int cp_cta_resume;
 unsigned g_ptx_sim_num_insn = 0;
 unsigned gpgpu_param_num_shaders = 0;
 
-char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor;
+char *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor;
 char *opcode_initiation_int, *opcode_initiation_fp, *opcode_initiation_dp,*opcode_initiation_sfu,*opcode_initiation_tensor;
 char *cdp_latency_str;
 unsigned cdp_latency[5];
 
-void ptx_opcocde_latency_options (option_parser_t opp) {
+void cuda_sim::ptx_opcocde_latency_options (option_parser_t opp) {
 	option_parser_register(opp, "-ptx_opcode_latency_int", OPT_CSTR, &opcode_latency_int,
 			"Opcode latencies for integers <ADD,MAX,MUL,MAD,DIV>"
 			"Default 1,1,19,25,145",
@@ -667,7 +668,7 @@ void ptx_instruction::set_opcode_and_latency()
 	 * [3] MAD
 	 * [4] DIV
 	 */
-	sscanf(opcode_latency_int, "%u,%u,%u,%u,%u",
+	sscanf(gpgpu_ctx->func_sim->opcode_latency_int, "%u,%u,%u,%u,%u",
 			&int_latency[0],&int_latency[1],&int_latency[2],
 			&int_latency[3],&int_latency[4]);
 	sscanf(opcode_latency_fp, "%u,%u,%u,%u,%u",
diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h
index e690356..96d34f5 100644
--- a/src/cuda-sim/cuda-sim.h
+++ b/src/cuda-sim/cuda-sim.h
@@ -47,10 +47,9 @@ extern int g_debug_thread_uid;
 extern void ** g_inst_classification_stat;
 extern void ** g_inst_op_classification_stat;
 extern int g_ptx_kernel_count; // used for classification stat collection purposes 
-extern char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor;
+extern char *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor;
 
 
-void ptx_opcocde_latency_options (option_parser_t opp);
 extern class kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry,
                                             gpgpu_ptx_sim_arg_list_t args, 
                                             struct dim3 gridDim, 
@@ -134,4 +133,12 @@ void print_ptxinfo();
 void clear_ptxinfo();
 struct gpgpu_ptx_sim_info get_ptxinfo();
 
+class cuda_sim {
+    public:
+	//global variables
+	char *opcode_latency_int;
+	//global functions
+	void ptx_opcocde_latency_options (option_parser_t opp);
+};
+
 #endif
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 8cedf79..1bd409e 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -1095,8 +1095,10 @@ ptx_instruction::ptx_instruction( int opcode,
                                   const char *file, 
                                   unsigned line,
                                   const char *source,
-                                  const core_config *config ) : warp_inst_t(config)
+                                  const core_config *config,
+				  gpgpu_context* ctx ) : warp_inst_t(config)
 {
+   gpgpu_ctx = ctx;
    m_uid = ++g_num_ptx_inst_uid;
    m_PC = 0;
    m_opcode = opcode;
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index fd869c6..1604551 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -43,6 +43,8 @@
 
 #include "memory.h"
 
+class gpgpu_context;
+
 class type_info_key {
 public:
    type_info_key()
@@ -931,7 +933,8 @@ public:
                     const char *file, 
                     unsigned line,
                     const char *source,
-                    const core_config *config );
+                    const core_config *config,
+		    gpgpu_context* ctx);
 
    void print_insn() const;
    virtual void print_insn( FILE *fp ) const;
@@ -1187,6 +1190,8 @@ private:
    virtual void pre_decode();
    friend class function_info;
    static unsigned g_num_ptx_inst_uid;
+   // backward pointer
+   class gpgpu_context* gpgpu_ctx;
 };
 
 class param_info {
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index 05fc618..0139534 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -285,7 +285,8 @@ void ptx_recognizer::add_instruction()
                                              gpgpu_ctx->g_filename,
                                              ptx_get_lineno(scanner),
                                              linebuf,
-                                             g_shader_core_config );
+                                             g_shader_core_config,
+					     gpgpu_ctx );
    g_instructions.push_back(i);
    g_inst_lookup[gpgpu_ctx->g_filename][ptx_get_lineno(scanner)] = i;
    init_instruction_state();
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 2ff37d1..39acdd9 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -1826,7 +1826,7 @@ void gpgpu_sim::dump_pipeline( int mask, int s, int m ) const
    fflush(stdout);
 }
 
-const struct shader_core_config * gpgpu_sim::getShaderCoreConfig()
+const shader_core_config * gpgpu_sim::getShaderCoreConfig()
 {
    return m_shader_config;
 }
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index e2c913a..7eeb7dd 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -295,7 +295,10 @@ extern bool g_interactive_debugger_enabled;
 
 class gpgpu_sim_config : public power_config, public gpgpu_functional_sim_config {
 public:
-    gpgpu_sim_config() { m_valid = false; }
+    gpgpu_sim_config(gpgpu_context* ctx): m_shader_config(ctx) {
+	m_valid = false;
+	gpgpu_ctx = ctx;
+    }
     void reg_options(class OptionParser * opp);
     void init() 
     {
@@ -341,6 +344,8 @@ private:
     void init_clock_domains(void ); 
 
 
+    // backward pointer
+    class gpgpu_context* gpgpu_ctx;
     bool m_valid;
     shader_core_config m_shader_config;
     memory_config m_memory_config;
@@ -473,7 +478,7 @@ public:
    /*!
     * Returning the configuration of the shader core, used by the functional simulation only so far
     */
-   const struct shader_core_config * getShaderCoreConfig();
+   const shader_core_config * getShaderCoreConfig();
    
    
    //! Get shader core Memory Configuration
@@ -537,7 +542,7 @@ private:
    const gpgpu_sim_config &m_config;
   
    const struct cudaDeviceProp     *m_cuda_properties;
-   const struct shader_core_config *m_shader_config;
+   const shader_core_config *m_shader_config;
    const struct memory_config      *m_memory_config;
 
    // stats
diff --git a/src/gpgpu-sim/mem_latency_stat.cc b/src/gpgpu-sim/mem_latency_stat.cc
index 04dc75b..d08ba39 100644
--- a/src/gpgpu-sim/mem_latency_stat.cc
+++ b/src/gpgpu-sim/mem_latency_stat.cc
@@ -42,7 +42,7 @@
 #include <stdlib.h>
 #include <stdio.h>
 
-memory_stats_t::memory_stats_t( unsigned n_shader, const struct shader_core_config *shader_config, const struct memory_config *mem_config, const class gpgpu_sim* gpu )
+memory_stats_t::memory_stats_t( unsigned n_shader, const shader_core_config *shader_config, const struct memory_config *mem_config, const class gpgpu_sim* gpu )
 {
    assert( mem_config->m_valid );
    assert( shader_config->m_valid );
diff --git a/src/gpgpu-sim/mem_latency_stat.h b/src/gpgpu-sim/mem_latency_stat.h
index b9285c1..6ce568d 100644
--- a/src/gpgpu-sim/mem_latency_stat.h
+++ b/src/gpgpu-sim/mem_latency_stat.h
@@ -35,7 +35,7 @@
 class memory_stats_t {
 public:
    memory_stats_t( unsigned n_shader, 
-                   const struct shader_core_config *shader_config, 
+                   const class shader_core_config *shader_config, 
                    const struct memory_config *mem_config,
 				   const class gpgpu_sim* gpu);
 
@@ -53,7 +53,7 @@ public:
 
    unsigned m_n_shader;
 
-   const struct shader_core_config *m_shader_config;
+   const shader_core_config *m_shader_config;
    const struct memory_config *m_memory_config;
    const class gpgpu_sim* m_gpu;
 
diff --git a/src/gpgpu-sim/power_interface.cc b/src/gpgpu-sim/power_interface.cc
index 3861b6a..0272aa6 100644
--- a/src/gpgpu-sim/power_interface.cc
+++ b/src/gpgpu-sim/power_interface.cc
@@ -38,7 +38,7 @@ void init_mcpat(const gpgpu_sim_config &config, class gpgpu_sim_wrapper *wrapper
 
 }
 
-void mcpat_cycle(const gpgpu_sim_config &config, const struct shader_core_config *shdr_config, class gpgpu_sim_wrapper *wrapper, class power_stat_t *power_stats, unsigned stat_sample_freq, unsigned tot_cycle, unsigned cycle, unsigned tot_inst, unsigned inst){
+void mcpat_cycle(const gpgpu_sim_config &config, const shader_core_config *shdr_config, class gpgpu_sim_wrapper *wrapper, class power_stat_t *power_stats, unsigned stat_sample_freq, unsigned tot_cycle, unsigned cycle, unsigned tot_inst, unsigned inst){
 
 	static bool mcpat_init=true;
 
diff --git a/src/gpgpu-sim/power_interface.h b/src/gpgpu-sim/power_interface.h
index afac22b..a388c23 100644
--- a/src/gpgpu-sim/power_interface.h
+++ b/src/gpgpu-sim/power_interface.h
@@ -36,7 +36,7 @@
 #include "gpgpu_sim_wrapper.h"
 
 void init_mcpat(const gpgpu_sim_config &config, class gpgpu_sim_wrapper *wrapper, unsigned stat_sample_freq, unsigned tot_inst, unsigned inst);
-void mcpat_cycle(const gpgpu_sim_config &config, const struct shader_core_config *shdr_config, class gpgpu_sim_wrapper *wrapper, class power_stat_t *power_stats,
+void mcpat_cycle(const gpgpu_sim_config &config, const shader_core_config *shdr_config, class gpgpu_sim_wrapper *wrapper, class power_stat_t *power_stats,
         unsigned stat_sample_freq, unsigned tot_cycle, unsigned cycle, unsigned tot_inst, unsigned inst);
 void mcpat_reset_perf_count(class gpgpu_sim_wrapper *wrapper);
 
diff --git a/src/gpgpu-sim/power_stat.cc b/src/gpgpu-sim/power_stat.cc
index 4c995e9..007b4c6 100644
--- a/src/gpgpu-sim/power_stat.cc
+++ b/src/gpgpu-sim/power_stat.cc
@@ -42,7 +42,7 @@
 
 
 
-power_mem_stat_t::power_mem_stat_t(const struct memory_config *mem_config, const struct shader_core_config *shdr_config, memory_stats_t *mem_stats, shader_core_stats *shdr_stats){
+power_mem_stat_t::power_mem_stat_t(const struct memory_config *mem_config, const shader_core_config *shdr_config, memory_stats_t *mem_stats, shader_core_stats *shdr_stats){
 	   assert( mem_config->m_valid );
 	   m_mem_stats = mem_stats;
 	   m_config = mem_config;
@@ -125,7 +125,7 @@ void power_mem_stat_t::print (FILE *fout) const {
 }
 
 
-power_core_stat_t::power_core_stat_t( const struct shader_core_config *shader_config, shader_core_stats *core_stats )
+power_core_stat_t::power_core_stat_t( const shader_core_config *shader_config, shader_core_stats *core_stats )
 {
      	assert( shader_config->m_valid );
         m_config = shader_config;
@@ -266,7 +266,7 @@ for(unsigned i=0; i<m_config->num_shader(); ++i){
     }
 }
 
-power_stat_t::power_stat_t( const struct shader_core_config *shader_config,float * average_pipeline_duty_cycle,float *active_sms,shader_core_stats * shader_stats, const struct memory_config *mem_config,memory_stats_t * memory_stats)
+power_stat_t::power_stat_t( const shader_core_config *shader_config,float * average_pipeline_duty_cycle,float *active_sms,shader_core_stats * shader_stats, const struct memory_config *mem_config,memory_stats_t * memory_stats)
 {
 	assert( shader_config->m_valid );
 	assert( mem_config->m_valid );
diff --git a/src/gpgpu-sim/power_stat.h b/src/gpgpu-sim/power_stat.h
index 20af2e5..91fade9 100644
--- a/src/gpgpu-sim/power_stat.h
+++ b/src/gpgpu-sim/power_stat.h
@@ -73,7 +73,7 @@ struct shader_core_power_stats_pod {
 
 class power_core_stat_t : public shader_core_power_stats_pod {
 public:
-   power_core_stat_t(const struct shader_core_config *shader_config, shader_core_stats *core_stats);
+   power_core_stat_t(const shader_core_config *shader_config, shader_core_stats *core_stats);
    void visualizer_print( gzFile visualizer_file );
    void print (FILE *fout);
    void init();
@@ -113,7 +113,7 @@ struct mem_power_stats_pod{
 
 class power_mem_stat_t : public mem_power_stats_pod{
 public:
-   power_mem_stat_t(const struct memory_config *mem_config, const struct shader_core_config *shdr_config, memory_stats_t *mem_stats, shader_core_stats *shdr_stats);
+   power_mem_stat_t(const struct memory_config *mem_config, const shader_core_config *shdr_config, memory_stats_t *mem_stats, shader_core_stats *shdr_stats);
    void visualizer_print( gzFile visualizer_file );
    void print (FILE *fout) const;
    void init();
@@ -128,7 +128,7 @@ private:
 
 class power_stat_t {
 public:
-   power_stat_t( const struct shader_core_config *shader_config,float * average_pipeline_duty_cycle,float * active_sms,shader_core_stats * shader_stats, const struct memory_config *mem_config,memory_stats_t * memory_stats);
+   power_stat_t( const shader_core_config *shader_config,float * average_pipeline_duty_cycle,float * active_sms,shader_core_stats * shader_stats, const struct memory_config *mem_config,memory_stats_t * memory_stats);
    void visualizer_print( gzFile visualizer_file );
    void print (FILE *fout) const;
    void save_stats(){
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 69b619a..4d12068 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -46,6 +46,7 @@
 #include <limits.h>
 #include "traffic_breakdown.h"
 #include "shader_trace.h"
+#include "../../libcuda/gpgpu_context.h"
 
 #define PRIORITIZE_MSHR_OVER_WB 1
 #define MAX(a,b) (((a)>(b))?(a):(b))
@@ -69,7 +70,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
                                   class simt_core_cluster *cluster,
                                   unsigned shader_id,
                                   unsigned tpc_id,
-                                  const struct shader_core_config *config,
+                                  const shader_core_config *config,
                                   const struct memory_config *mem_config,
                                   shader_core_stats *stats )
    : core_t( gpu, NULL, config->warp_size, config->n_thread_per_shader ),
@@ -3018,7 +3019,7 @@ void shader_core_config::set_pipeline_latency() {
 			 * [3] MAD
 			 * [4] DIV
 			 */
-			sscanf(opcode_latency_int, "%u,%u,%u,%u,%u",
+			sscanf(gpgpu_ctx->func_sim->opcode_latency_int, "%u,%u,%u,%u,%u",
 					&int_latency[0],&int_latency[1],&int_latency[2],
 					&int_latency[3],&int_latency[4]);
 			sscanf(opcode_latency_fp, "%u,%u,%u,%u,%u",
@@ -3786,7 +3787,7 @@ void opndcoll_rfu_t::collector_unit_t::dispatch()
 
 simt_core_cluster::simt_core_cluster( class gpgpu_sim *gpu, 
                                       unsigned cluster_id, 
-                                      const struct shader_core_config *config, 
+                                      const shader_core_config *config, 
                                       const struct memory_config *mem_config,
                                       shader_core_stats *stats, 
                                       class memory_stats_t *mstats )
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 25b9607..e0cefac 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -69,6 +69,8 @@
 
 #define WRITE_MASK_SIZE 8
 
+class gpgpu_context;
+
 enum exec_unit_type_t
 {
   NONE = 0,
@@ -294,7 +296,7 @@ typedef std::bitset<WARP_PER_CTA_MAX> warp_set_t;
 int register_bank(int regnum, int wid, unsigned num_banks, unsigned bank_warp_shift, bool sub_core_model, unsigned banks_per_sched, unsigned sched_id );
 
 class shader_core_ctx;
-struct shader_core_config;
+class shader_core_config;
 class shader_core_stats;
 
 enum scheduler_prioritization_type
@@ -1032,7 +1034,7 @@ struct ifetch_buffer_t {
     unsigned m_warp_id;
 };
 
-struct shader_core_config;
+class shader_core_config;
 
 class simd_function_unit {
 public:
@@ -1362,10 +1364,12 @@ const char* const pipeline_stage_name_decode[] = {
     "N_PIPELINE_STAGES" 
 };
 
-struct shader_core_config : public core_config
+class shader_core_config : public core_config
 {
-    shader_core_config(){
+    public:
+    shader_core_config(gpgpu_context* ctx){
 	pipeline_widths_string = NULL;
+	gpgpu_ctx = ctx;
     }
 
     void init()
@@ -1425,6 +1429,8 @@ struct shader_core_config : public core_config
     unsigned cid_to_sid( unsigned cid, unsigned cluster_id ) const { return cluster_id*n_simt_cores_per_cluster + cid; }
     void set_pipeline_latency();
 
+    // backward pointer
+    class gpgpu_context* gpgpu_ctx;
 // data
     char *gpgpu_shader_core_pipeline_opt;
     bool gpgpu_perfect_mem;
@@ -1770,7 +1776,7 @@ public:
                      class simt_core_cluster *cluster,
                      unsigned shader_id,
                      unsigned tpc_id,
-                     const struct shader_core_config *config,
+                     const shader_core_config *config,
                      const struct memory_config *mem_config,
                      shader_core_stats *stats );
 
@@ -2065,7 +2071,7 @@ class simt_core_cluster {
 public:
     simt_core_cluster( class gpgpu_sim *gpu, 
                        unsigned cluster_id, 
-                       const struct shader_core_config *config, 
+                       const shader_core_config *config, 
                        const struct memory_config *mem_config,
                        shader_core_stats *stats,
                        memory_stats_t *mstats );
diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc
index b54f20c..5018305 100644
--- a/src/gpgpusim_entrypoint.cc
+++ b/src/gpgpusim_entrypoint.cc
@@ -218,10 +218,10 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf()
    option_parser_t opp = option_parser_create();
 
    ptx_reg_options(opp);
-   ptx_opcocde_latency_options(opp);
+   func_sim->ptx_opcocde_latency_options(opp);
 
    icnt_reg_options(opp);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config();
+   GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config(this);
    GPGPUsim_ctx_ptr()->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
 
    option_parser_cmdline(opp, sg_argc, sg_argv); // parse configuration options