refactor: mostly finished getting rid of extern decl

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 7183]

17 files changed, 65 insertions, 113 deletions

diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h
index 7e05f7e..baf3b63 100644
--- a/src/cuda-sim/cuda-sim.h
+++ b/src/cuda-sim/cuda-sim.h
@@ -17,6 +17,10 @@ extern memory_space *g_global_mem;
 extern int g_debug_execution;
 extern int g_debug_thread_uid;
 extern std::map<std::string,function_info*> *g_kernel_name_to_function_lookup;
+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 void   gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key,
                                             struct gpgpu_ptx_sim_arg *args, 
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 1b1d864..ce68505 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -1194,6 +1194,7 @@ void copy_args_into_buffer_list( const ptx_instruction * pI,
 void copy_buffer_list_into_frame(ptx_thread_info * thread, arg_buffer_list_t &arg_values);
 void copy_buffer_to_frame(ptx_thread_info * thread, const arg_buffer_t &a);
 
+#if !defined(__CUDA_RUNTIME_API_H__)
 /*******************************/
 // These declarations should be identical to those in ./../../cuda-sim-dev/libcuda/texture_types.h
 enum cudaChannelFormatKind {
@@ -1241,8 +1242,8 @@ struct textureReference {
    enum cudaTextureAddressMode   addressMode[2];
    struct cudaChannelFormatDesc  channelDesc;
 };
-
 /**********************************/
+#endif
 
 struct textureInfo {
    unsigned int texel_size; //size in bytes, e.g. (channelDesc.x+y+z+w)/8
@@ -1274,5 +1275,7 @@ extern std::map<std::string,symbol_table*> g_sym_name_to_symbol_table;
 void gpgpu_ptx_assemble( std::string kname, void *kinfo );
 #include "../option_parser.h"
 void ptx_reg_options(option_parser_t opp);
+unsigned ptx_kernel_shmem_size( void *kernel_impl );
+unsigned ptx_kernel_nregs( void *kernel_impl );
 
 #endif
diff --git a/src/debug.h b/src/debug.h
index 463780a..a6f2b10 100644
--- a/src/debug.h
+++ b/src/debug.h
@@ -56,6 +56,8 @@ private:
    unsigned     m_value;
 };
 
+extern int gpgpu_ptx_instruction_classification ;
+
 class ptx_thread_info;
 class ptx_instruction;
 bool thread_at_brkpt( void *ptx_thd_info, const struct brk_pt &b );
diff --git a/src/gpgpu-sim/addrdec.h b/src/gpgpu-sim/addrdec.h
index ba28c05..8315bc9 100644
--- a/src/gpgpu-sim/addrdec.h
+++ b/src/gpgpu-sim/addrdec.h
@@ -92,5 +92,7 @@ void addrdec_tlx(unsigned long long int addr, addrdec_t *tlx);
 void addrdec_setnchip(unsigned int nchips);
 void addrdec_setoption(option_parser_t opp);
 void addrdec_parseoption(const char *option);
-
+extern unsigned long long int addrdec_packbits(unsigned long long int mask, 
+                                               unsigned long long int val,
+                                               unsigned char high, unsigned char low);
 #endif
diff --git a/src/gpgpu-sim/dwf.h b/src/gpgpu-sim/dwf.h
index 9771af5..75c06da 100644
--- a/src/gpgpu-sim/dwf.h
+++ b/src/gpgpu-sim/dwf.h
@@ -115,4 +115,7 @@ void dwf_set_accPC( int n_shaders, unsigned *acc_pc_count );
 
 void dwf_update_statistics( int shader_id );
 
+extern unsigned int gpgpu_dwf_heuristic;
+extern unsigned int gpgpu_dwf_regbk;
+
 #endif
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index 679d1d2..013ca54 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -67,6 +67,8 @@
 #include "gpu-cache.h"
 #include "gpu-misc.h"
 #include "addrdec.h"
+#include "stat-tool.h"
+#include "gpu-sim.h"
 #include <assert.h>
 #include <string.h>
 
@@ -141,19 +143,12 @@ void shd_cache_destroy( shd_cache_t* cp ) {
    free(cp);
 }
 
-extern void shader_cache_miss_log( int logger_id, int type );
 // hook up with shader core logger
 void shd_cache_bind_logger(shd_cache_t* cp, int core_id, int type_id) {
    cp->core_id = core_id; 
    cp->type_id = type_id;
 }
 
-extern unsigned long long int addrdec_packbits(unsigned long long int mask, 
-                                               unsigned long long int val,
-                                               unsigned char high, unsigned char low);
-extern void shader_cache_access_log( int logger_id, int type, int miss);
-extern void shader_cache_access_unlog( int logger_id, int type, int miss);
-
 shd_cache_line_t* shd_cache_access_internal( shd_cache_t *cp, 
                                              unsigned long long int addr, 
                                              unsigned int nbytes, 
@@ -211,7 +206,6 @@ shd_cache_line_t* shd_cache_access( shd_cache_t *cp,
    return shd_cache_access_internal(cp,addr,nbytes,write,sim_cycle,1/*this is a real access*/);
 }
 
-extern int gpgpu_cache_wt_through;
 shd_cache_t *test = NULL;
 enum cache_request_status shd_cache_access_wb( shd_cache_t *cp, 
                                     unsigned long long int addr, 
diff --git a/src/gpgpu-sim/gpu-cache.h b/src/gpgpu-sim/gpu-cache.h
index cdb3414..d07d1e1 100644
--- a/src/gpgpu-sim/gpu-cache.h
+++ b/src/gpgpu-sim/gpu-cache.h
@@ -186,6 +186,7 @@ unsigned long long int L2_shd_cache_fill( shd_cache_t *cp,
                                           unsigned int sim_cycle );
 
 void shd_cache_print( shd_cache_t *cp,  FILE *stream);
-
+float shd_cache_windowed_cache_miss_rate(shd_cache_t*, int);
+void shd_cache_new_window(shd_cache_t*);
 
 #endif
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index f0646d2..b716acc 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -96,6 +96,8 @@
 #include "../gpgpusim_entrypoint.h"
 #include "../cuda-sim/cuda-sim.h"
 
+#include "mem_latency_stat.h"
+
 #include <stdio.h>
 #include <string.h>
 #define MAX(a,b) (((a)>(b))?(a):(b))
@@ -146,17 +148,9 @@ unsigned int gpu_stall_by_MSHRwb = 0;
 //Note: it is accumulative for all shaders and is never reset
 //so it might increase 8 times in a cycle if we have 8 shaders
 unsigned int gpu_stall_sh2icnt = 0;        
-// performance counters to account for instruction distribution
-extern unsigned int gpgpu_n_load_insn;
-extern unsigned int gpgpu_n_store_insn;
-extern unsigned int gpgpu_n_shmem_insn;
-extern unsigned int gpgpu_n_tex_insn;
-extern unsigned int gpgpu_n_const_insn;
-extern unsigned int gpgpu_multi_unq_fetches;
 char *gpgpu_runtime_stat;
 int gpu_stat_sample_freq = 10000;
 int gpu_runtime_stat_flag = 0;
-extern int gpgpu_warpdistro_shader;
 
 // GPGPU options
 unsigned long long  gpu_max_cycle = 0;
@@ -174,12 +168,10 @@ char *gpgpu_cache_texl1_opt;
 char *gpgpu_cache_constl1_opt;
 char *gpgpu_cache_dl1_opt;
 char *gpgpu_cache_dl2_opt;
-extern int gpgpu_l2_readoverwrite;
 int gpgpu_partial_write_mask = 0;
 
 bool gpgpu_perfect_mem = false;
 char *gpgpu_shader_core_pipeline_opt;
-extern unsigned int *requests_by_warp;
 unsigned int gpgpu_dram_buswidth = 4;
 unsigned int gpgpu_dram_burst_length = 4;
 int gpgpu_dram_sched_queue_size = 0; 
@@ -187,23 +179,14 @@ char * gpgpu_dram_timing_opt;
 int gpgpu_flush_cache = 0;
 int gpgpu_mem_address_mask = 0;
 unsigned int recent_dram_util = 0;
-
 int gpgpu_cflog_interval = 0;
-
 unsigned int finished_trace = 0;
-
 unsigned g_next_request_uid = 1;
-
-extern struct regs_t regs;
-
-extern long int gpu_reads;
-
-void ptx_dump_regs( void *thd );
-
 int g_nthreads_issued;
 int g_total_cta_left;
 
 
+void ptx_dump_regs( void *thd );
 unsigned ptx_kernel_program_size();
 void visualizer_printstat();
 void time_vector_create(int ld_size,int st_size);
@@ -213,9 +196,6 @@ void node_req_hist_clear(void *p);
 void node_req_hist_dump(void *p);
 void node_req_hist_update(void * p,int node, long long cycle);
 
-/* functionally simulated memory */
-extern struct mem_t *mem;
-
 /* Defining Clock Domains
 basically just the ratio is important */
 
@@ -252,56 +232,25 @@ unsigned int gpu_n_mshr_per_shader;
 unsigned int gpu_n_thread_per_shader = 128;
 unsigned int gpu_n_warp_per_shader;
 unsigned int gpu_n_mshr_per_thread = 1;
-
-extern int gpgpu_interwarp_mshr_merge ;
-
-extern unsigned int gpgpu_shmem_size;
-extern unsigned int gpgpu_shader_registers;
-extern unsigned int gpgpu_shader_cta;
-extern int gpgpu_shmem_bkconflict;
-extern int gpgpu_cache_bkconflict;
-extern int gpgpu_n_cache_bank;
-extern unsigned int warp_size; 
-extern int pipe_simd_width;
-extern unsigned int gpgpu_dwf_heuristic;
-extern unsigned int gpgpu_dwf_regbk;
 bool gpgpu_reg_bankconflict = false;
-extern int gpgpu_shmem_port_per_bank;
-extern int gpgpu_cache_port_per_bank;
-extern int gpgpu_const_port_per_bank;
-extern int gpgpu_shmem_pipe_speedup;  
-extern int gpgpu_reg_bank_conflict_model;
-extern int gpgpu_reg_bank_use_warp_id;
 int gpgpu_operand_collector;
-extern int gpgpu_operand_collector_num_units;
-extern unsigned int gpgpu_num_reg_banks;
-
-extern unsigned int gpu_max_cta_per_shader;
-extern unsigned int gpu_padded_cta_size;
-extern int gpgpu_local_mem_map;
-
+int gpgpu_operand_collector_num_units = 4;
 unsigned int gpgpu_pre_mem_stages = 0;
 unsigned int gpgpu_no_divg_load = 0;
 char *gpgpu_dwf_hw_opt;
 unsigned int gpgpu_thread_swizzling = 0;
 unsigned int gpgpu_strict_simd_wrbk = 0;
-
 int pdom_sched_type = 0;
 int n_pdom_sc_orig_stat = 0; //the selected pdom schedular is used 
 int n_pdom_sc_single_stat = 0; //only a single warp is ready to go in that cycle.  
 int *num_warps_issuable;
 int *num_warps_issuable_pershader;
-
-// Thread Dispatching Unit option 
 int gpgpu_cuda_sim = 1;
 int gpgpu_spread_blocks_across_cores = 1;
-
-/* GPU uArch structures */
 shader_core_ctx_t **sc;
 dram_t **dram;
 unsigned int common_clock = 0;
 unsigned int more_thread = 1;
-extern unsigned int n_regconflict_stall;
 unsigned int warp_conflict_at_writeback = 0;
 unsigned int gpgpu_commit_pc_beyond_two = 0;
 int gpgpu_cache_wt_through = 0;
@@ -331,18 +280,11 @@ unsigned ptx_sim_grid_size();
 void icnt_init_grid();
 void interconnect_stats();
 void icnt_overal_stat();
-
 void gpu_sim_loop( int grid_num );
-
 void print_shader_cycle_distro( FILE *fout ) ;
 void find_reconvergence_points();
 void dwf_process_reconv_pts();
 
-extern int gpgpu_ptx_instruction_classification ;
-extern int g_ptx_sim_mode;
-
-extern int gpgpu_coalesce_arch;
-
 #define CREATELOG 111
 #define SAMPLELOG 222
 #define DUMPLOG 333
@@ -874,12 +816,6 @@ unsigned int run_gpu_sim(int grid_num)
    return gpu_sim_cycle;
 }
 
-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 unsigned get_max_mshr_used(shader_core_ctx_t* shader);
-
 void gpu_print_stat()
 {  
    unsigned i;
@@ -1160,8 +1096,6 @@ int issue_mf_from_fq(mem_fetch_t *mf){
    return 0;
 }
 
-extern void mshr_return_from_mem(shader_core_ctx_t * shader, mshr_entry_t* mshr);
-
 inline void fill_shd_L1_with_new_line(shader_core_ctx_t * sc, mem_fetch_t * mf) {
    unsigned long long int repl_addr = -1;
    // When the data arrives, it flags all the appropriate MSHR
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 338336e..a69a3ef 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -148,6 +148,7 @@ extern unsigned int gpu_stall_by_MSHRwb;
 extern unsigned int gpu_stall_shd_mem;
 extern unsigned int gpu_stall_sh2icnt;
 extern int gpgpu_operand_collector;
+extern int gpgpu_operand_collector_num_units;
 extern int gpu_runtime_stat_flag;
 extern unsigned int *max_return_queue_length;
 extern int gpgpu_partial_write_mask;
@@ -168,5 +169,7 @@ extern unsigned int gpu_n_mem_per_ctrlr;
 extern unsigned int recent_dram_util;
 extern unsigned int **concurrent_row_access; //concurrent_row_access[dram chip id][bank id]
 extern unsigned long long  gpu_tot_sim_insn;
+extern unsigned int gpgpu_n_sent_writes;
+extern unsigned int gpgpu_n_processed_writes;
 
 #endif
diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc
index 48dcda1..e9f4780 100644
--- a/src/gpgpu-sim/l2cache.cc
+++ b/src/gpgpu-sim/l2cache.cc
@@ -51,8 +51,6 @@ extern int gpgpu_dram_sched_queue_size;
 extern unsigned made_write_mfs;
 extern unsigned freed_L1write_mfs;
 extern unsigned freed_L2write_mfs;
-extern unsigned int gpgpu_n_sent_writes;
-extern unsigned int gpgpu_n_processed_writes;
 extern unsigned g_next_request_uid;
 
 void memlatstat_icnt2sh_push(mem_fetch_t *mf);
diff --git a/src/gpgpu-sim/l2cache.h b/src/gpgpu-sim/l2cache.h
index 11bef11..d245408 100644
--- a/src/gpgpu-sim/l2cache.h
+++ b/src/gpgpu-sim/l2cache.h
@@ -50,3 +50,4 @@ extern unsigned L2_write_miss;
 extern unsigned L2_write_hit;
 extern unsigned L2_read_hit;
 extern unsigned L2_read_miss;
+extern int gpgpu_l2_readoverwrite;
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 6cafc48..bba2b8b 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -141,7 +141,6 @@ int          gpgpu_shmem_pipe_speedup = 2;
 unsigned int gpu_max_cta_per_shader = 8;
 unsigned int gpu_padded_cta_size = 32;
 int          gpgpu_local_mem_map = 1;
-int          gpgpu_operand_collector_num_units = 4;
 
 /////////////////////////////////////////////////////////////////////////////
 /*-------------------------------------------------------------------------*/
@@ -468,8 +467,6 @@ int log2i(int n) {
    return lg;
 }
 
-extern unsigned int gpu_n_warp_per_shader;
-
 shader_core_ctx_t* shader_create( const char *name, int sid,
                                   unsigned int n_threads,
                                   unsigned int n_mshr,
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 1d871d9..0a90225 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -897,5 +897,36 @@ unsigned int max_cta_per_shader( shader_core_ctx_t *shader);
 #define ID_OC 7
 
 extern shader_core_ctx_t **sc;
+extern unsigned int gpgpu_n_load_insn;
+extern unsigned int gpgpu_n_store_insn;
+extern unsigned int gpgpu_n_shmem_insn;
+extern unsigned int gpgpu_n_tex_insn;
+extern unsigned int gpgpu_n_const_insn;
+extern unsigned int gpgpu_multi_unq_fetches;
+extern unsigned int gpgpu_n_cache_bkconflict;
+extern unsigned int gpgpu_n_shmem_bkconflict;
+extern int gpgpu_warpdistro_shader;
+extern unsigned int *shader_cycle_distro;
+extern int gpgpu_interwarp_mshr_merge;
+extern unsigned int gpgpu_shmem_size;
+extern unsigned int gpgpu_shader_registers;
+extern unsigned int gpgpu_shader_cta;
+extern int gpgpu_shmem_bkconflict;
+extern int gpgpu_cache_bkconflict;
+extern int gpgpu_n_cache_bank;
+extern int pipe_simd_width;
+extern int gpgpu_shmem_port_per_bank;
+extern int gpgpu_cache_port_per_bank;
+extern int gpgpu_const_port_per_bank;
+extern int gpgpu_shmem_pipe_speedup;  
+extern int gpgpu_reg_bank_conflict_model;
+extern unsigned int gpgpu_num_reg_banks;
+extern unsigned int gpu_max_cta_per_shader;
+extern unsigned int gpu_padded_cta_size;
+extern int gpgpu_local_mem_map;
+extern unsigned int n_regconflict_stall;
+extern int gpgpu_coalesce_arch;
+extern unsigned get_max_mshr_used(shader_core_ctx_t* shader);
+extern void mshr_return_from_mem(shader_core_ctx_t * shader, mshr_entry_t* mshr);
 
 #endif /* SHADER_H */
diff --git a/src/gpgpu-sim/stat-tool.h b/src/gpgpu-sim/stat-tool.h
index 79de754..d67daf5 100644
--- a/src/gpgpu-sim/stat-tool.h
+++ b/src/gpgpu-sim/stat-tool.h
@@ -68,6 +68,7 @@
 
 #include "../abstract_hardware_model.h"
 #include <stdio.h>
+#include <zlib.h>
 
 void try_snap_shot (unsigned long long  current_cycle);
 void set_spill_interval (unsigned long long  interval);
@@ -80,7 +81,7 @@ void cflog_snapshot( int logger_id, unsigned long long  cycle );
 void cflog_print(FILE *fout);
 void cflog_print_path_expression(FILE *fout);
 void cflog_visualizer_print(FILE *fout);
-
+void cflog_visualizer_gzprint(gzFile fout);
 
 void insn_warp_occ_create( int n_loggers, int simd_width, int n_insn );
 void insn_warp_occ_log( int logger_id, address_type pc, int warp_occ );
@@ -120,5 +121,6 @@ void shader_CTA_count_unlog( int shader_id, int nCTAdone );
 void shader_CTA_count_resetnow( );
 void shader_CTA_count_print( FILE *fout );
 void shader_CTA_count_visualizer_print( FILE *fout );
+void shader_CTA_count_visualizer_gzprint(gzFile fout);
 
 #endif /* CFLOGGER_H */
diff --git a/src/gpgpu-sim/visualizer.cc b/src/gpgpu-sim/visualizer.cc
index 6eec3be..4ed8157 100644
--- a/src/gpgpu-sim/visualizer.cc
+++ b/src/gpgpu-sim/visualizer.cc
@@ -67,32 +67,14 @@
 #include "../option_parser.h"
 #include "mem_latency_stat.h"
 #include "stat-tool.h"
+#include "gpu-cache.h"
 
 #include <time.h>
 #include <string.h>
 #include <zlib.h>
 
-extern unsigned int gpu_mem_n_bk;
-extern unsigned long long int mf_total_lat;
-extern unsigned num_mfs;
-extern unsigned long long  gpu_tot_sim_insn;
-extern unsigned int gpgpu_n_sent_writes;
-extern unsigned int gpgpu_n_processed_writes;
-extern unsigned int gpgpu_n_cache_bkconflict;
-extern unsigned int gpgpu_n_shmem_bkconflict;
-extern unsigned int *max_return_queue_length;
-extern int gpgpu_warpdistro_shader;
-extern unsigned ***mem_access_type_stats;
-
-extern unsigned int warp_size; 
-extern unsigned int *shader_cycle_distro;
-
 static void time_vector_print_interval2file(FILE *outfile);
 static void time_vector_print_interval2gzfile(gzFile outfile);
-void cflog_visualizer_gzprint(gzFile fout);
-void shader_CTA_count_visualizer_gzprint(gzFile fout);
-float shd_cache_windowed_cache_miss_rate(shd_cache_t*, int);
-void shd_cache_new_window(shd_cache_t*);
 
 int g_visualizer_enabled = 1;
 char *g_visualizer_filename = NULL;
diff --git a/src/gpgpu-sim/warp_tracker.cc b/src/gpgpu-sim/warp_tracker.cc
index 138011c..2096d93 100644
--- a/src/gpgpu-sim/warp_tracker.cc
+++ b/src/gpgpu-sim/warp_tracker.cc
@@ -65,12 +65,10 @@
 
 #include "warp_tracker.h"
 #include "gpu-sim.h"
+#include "shader.h"
 
 using namespace std;
 
-extern unsigned int warp_size;
-extern unsigned int gpu_n_shader;
-
 #include <set>
 
 class warp_tracker {
@@ -415,9 +413,6 @@ void free_commit_warp( int *commit_warp )
    free_commit_warp_q.push(commit_warp);
 }
 
-
-extern int pipe_simd_width;
-
 // uncomment to enable checking for warp consistency
 // #define CHECK_WARP_CONSISTENCY
 
diff --git a/src/intersim/interconnect_interface.cpp b/src/intersim/interconnect_interface.cpp
index 34fd39d..737e61e 100644
--- a/src/intersim/interconnect_interface.cpp
+++ b/src/intersim/interconnect_interface.cpp
@@ -18,10 +18,10 @@
 #include "interconnect_interface.h"
 #include "../gpgpu-sim/mem_fetch.h"
 #include "../gpgpu-sim/gpu-sim.h"
+#include "../gpgpu-sim/shader.h"
 #include <string.h>
 
 int _flit_size ;
-extern unsigned int warp_size;
 
 bool doub_net = false; //double networks disabled by default