From 6b7d14e294d8cb73e2281b7e18a4083e3271a85e Mon Sep 17 00:00:00 2001 From: Nathan Conrad Date: Mon, 26 Mar 2018 19:20:24 -0400 Subject: Provide portable (non-x86) breakpoint method which should work on all linuxes. Tested on PowerPC. --- src/gpgpu-sim/gpu-sim.cc | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'src/gpgpu-sim') diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 58a5d16..3829861 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -32,6 +32,7 @@ #include #include #include +#include #include "zlib.h" @@ -1440,7 +1441,7 @@ void gpgpu_sim::cycle() if( g_single_step && ((gpu_sim_cycle+gpu_tot_sim_cycle) >= g_single_step) ) { - asm("int $03"); + raise(SIGTRAP); // Debug breakpoint } gpu_sim_cycle++; if( g_interactive_debugger_enabled ) -- cgit v1.3 From c5da0034dd329976c77e2951e2f879095ea8cdac Mon Sep 17 00:00:00 2001 From: tgrogers Date: Mon, 2 Apr 2018 15:23:44 -0400 Subject: Adding in Mahmoud's changes to the print cache stats --- src/gpgpu-sim/gpu-cache.cc | 24 +++++++++++++++++------- 1 file changed, 17 insertions(+), 7 deletions(-) (limited to 'src/gpgpu-sim') diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc index 5ea4190..8886398 100644 --- a/src/gpgpu-sim/gpu-cache.cc +++ b/src/gpgpu-sim/gpu-cache.cc @@ -517,18 +517,28 @@ void cache_stats::print_stats(FILE *fout, const char *cache_name) const{ /// the provided name is used. /// The printed format is "[][] = " /// + std::vector< unsigned > total_access; + total_access.resize(NUM_MEM_ACCESS_TYPE, 0); std::string m_cache_name = cache_name; for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) { for (unsigned status = 0; status < NUM_CACHE_REQUEST_STATUS; ++status) { - if(m_stats[type][status] > 0){ - fprintf(fout, "\t%s[%s][%s] = %u\n", - m_cache_name.c_str(), - mem_access_type_str((enum mem_access_type)type), - cache_request_status_str((enum cache_request_status)status), - m_stats[type][status]); - } + fprintf(fout, "\t%s[%s][%s] = %u\n", + m_cache_name.c_str(), + mem_access_type_str((enum mem_access_type)type), + cache_request_status_str((enum cache_request_status)status), + m_stats[type][status]); + if(status != RESERVATION_FAIL) + total_access[type]+= m_stats[type][status]; } } + for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) { + if(total_access[type] > 0) + fprintf(fout, "\t%s[%s][%s] = %u\n", + m_cache_name.c_str(), + mem_access_type_str((enum mem_access_type)type), + "TOTAL_ACCESS", + total_access[type]); + } } void cache_sub_stats::print_port_stats(FILE *fout, const char *cache_name) const -- cgit v1.3 From 60a32420caa0e79c7d5133ff67a55c02333b1dc2 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Wed, 28 Mar 2018 14:00:16 -0400 Subject: adding new counters for parition level para and L2 BW --- src/gpgpu-sim/gpu-sim.cc | 36 ++++++++++++++++++++++++++++++++++++ 1 file changed, 36 insertions(+) (limited to 'src/gpgpu-sim') diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 0ef267d..95951a8 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -88,6 +88,12 @@ unsigned long long gpu_tot_sim_cycle = 0; // performance counter for stalls due to congestion. unsigned int gpu_stall_dramfull = 0; unsigned int gpu_stall_icnt2sh = 0; +unsigned long long partiton_reqs_in_parallel = 0; +unsigned long long partiton_reqs_in_parallel_total = 0; +unsigned long long partiton_reqs_in_parallel_util = 0; +unsigned long long partiton_reqs_in_parallel_util_total = 0; +unsigned long long partiton_replys_in_parallel = 0; +unsigned long long partiton_replys_in_parallel_total = 0; /* Clock Domains */ @@ -745,6 +751,9 @@ void gpgpu_sim::init() gpu_sim_insn = 0; last_gpu_sim_insn = 0; m_total_cta_launched=0; + partiton_reqs_in_parallel = 0; + partiton_replys_in_parallel = 0; + partiton_reqs_in_parallel_util = 0; reinit_clock_domains(); set_param_gpgpu_num_shaders(m_config.num_shader()); @@ -781,8 +790,14 @@ void gpgpu_sim::update_stats() { gpu_tot_sim_cycle += gpu_sim_cycle; gpu_tot_sim_insn += gpu_sim_insn; gpu_tot_issued_cta += m_total_cta_launched; + partiton_reqs_in_parallel_total += partiton_reqs_in_parallel; + partiton_replys_in_parallel_total += partiton_replys_in_parallel; + partiton_reqs_in_parallel_util_total += partiton_reqs_in_parallel_util; gpu_sim_cycle = 0; + partiton_reqs_in_parallel = 0; + partiton_replys_in_parallel = 0; + partiton_reqs_in_parallel_util = 0; gpu_sim_insn = 0; m_total_cta_launched = 0; } @@ -966,6 +981,19 @@ void gpgpu_sim::gpu_print_stat() printf("gpu_stall_dramfull = %d\n", gpu_stall_dramfull); printf("gpu_stall_icnt2sh = %d\n", gpu_stall_icnt2sh ); + printf("partiton_reqs_in_parallel = %lld\n", partiton_reqs_in_parallel); + printf("partiton_reqs_in_parallel_total = %lld\n", partiton_reqs_in_parallel_total ); + printf("partiton_level_parallism = %12.4f\n", (float)partiton_reqs_in_parallel / gpu_sim_cycle); + printf("partiton_level_parallism_total = %12.4f\n", (float)(partiton_reqs_in_parallel+partiton_reqs_in_parallel_total) / (gpu_tot_sim_cycle+gpu_sim_cycle) ); + printf("partiton_reqs_in_parallel_util = %lld\n", partiton_reqs_in_parallel_util); + printf("partiton_reqs_in_parallel_util_total = %lld\n", partiton_reqs_in_parallel_util_total ); + printf("partiton_level_parallism_util = %12.4f\n", (float)partiton_reqs_in_parallel_util / gpu_sim_cycle); + printf("partiton_level_parallism_util_total = %12.4f\n", (float)(partiton_reqs_in_parallel_util+partiton_reqs_in_parallel_util_total) / (gpu_tot_sim_cycle+gpu_sim_cycle) ); + printf("partiton_replys_in_parallel = %lld\n", partiton_replys_in_parallel); + printf("partiton_replys_in_parallel_total = %lld\n", partiton_replys_in_parallel_total ); + printf("L2_BW = %12.4f GB/Sec\n", ((float)(partiton_replys_in_parallel * 32) / (gpu_sim_cycle * m_config.icnt_period)) / 1000000000); + printf("L2_BW_total = %12.4f GB/Sec\n", ((float)((partiton_replys_in_parallel+partiton_replys_in_parallel_total) * 32) / ((gpu_tot_sim_cycle+gpu_sim_cycle) * m_config.icnt_period)) / 1000000000 ); + time_t curr_time; time(&curr_time); unsigned long long elapsed_time = MAX( curr_time - g_simulation_starttime, 1 ); @@ -1367,6 +1395,7 @@ void gpgpu_sim::cycle() for (unsigned i=0;in_simt_clusters;i++) m_cluster[i]->icnt_cycle(); } + unsigned partiton_replys_in_parallel_per_cycle = 0; if (clock_mask & ICNT) { // pop from memory controller to interconnect for (unsigned i=0;im_n_mem_sub_partition;i++) { @@ -1379,6 +1408,7 @@ void gpgpu_sim::cycle() mf->set_status(IN_ICNT_TO_SHADER,gpu_sim_cycle+gpu_tot_sim_cycle); ::icnt_push( m_shader_config->mem2device(i), mf->get_tpc(), mf, response_size ); m_memory_sub_partition[i]->pop(); + partiton_replys_in_parallel_per_cycle++; } else { gpu_stall_icnt2sh++; } @@ -1387,6 +1417,7 @@ void gpgpu_sim::cycle() } } } + partiton_replys_in_parallel += partiton_replys_in_parallel_per_cycle; if (clock_mask & DRAM) { for (unsigned i=0;im_n_mem;i++){ @@ -1399,6 +1430,7 @@ void gpgpu_sim::cycle() } // L2 operations follow L2 clock domain + unsigned partiton_reqs_in_parallel_per_cycle = 0; if (clock_mask & L2) { m_power_stats->pwr_mem_stat->l2_cache_stats[CURRENT_STAT_IDX].clear(); for (unsigned i=0;im_n_mem_sub_partition;i++) { @@ -1409,11 +1441,15 @@ void gpgpu_sim::cycle() } else { mem_fetch* mf = (mem_fetch*) icnt_pop( m_shader_config->mem2device(i) ); m_memory_sub_partition[i]->push( mf, gpu_sim_cycle + gpu_tot_sim_cycle ); + partiton_reqs_in_parallel_per_cycle++; } m_memory_sub_partition[i]->cache_cycle(gpu_sim_cycle+gpu_tot_sim_cycle); m_memory_sub_partition[i]->accumulate_L2cache_stats(m_power_stats->pwr_mem_stat->l2_cache_stats[CURRENT_STAT_IDX]); } } + partiton_reqs_in_parallel += partiton_reqs_in_parallel_per_cycle; + if(partiton_reqs_in_parallel_per_cycle > 0) + partiton_reqs_in_parallel_util += partiton_reqs_in_parallel_per_cycle; if (clock_mask & ICNT) { icnt_transfer(); -- cgit v1.3 From 038df10118351d44f8950065ddeb698ae3e390a1 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Wed, 28 Mar 2018 15:22:33 -0400 Subject: adding new stats counter --- src/gpgpu-sim/gpu-sim.cc | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) (limited to 'src/gpgpu-sim') diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 95951a8..da35dc3 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -92,6 +92,8 @@ unsigned long long partiton_reqs_in_parallel = 0; unsigned long long partiton_reqs_in_parallel_total = 0; unsigned long long partiton_reqs_in_parallel_util = 0; unsigned long long partiton_reqs_in_parallel_util_total = 0; +unsigned long long gpu_sim_cycle_parition_util = 0; +unsigned long long gpu_tot_sim_cycle_parition_util = 0; unsigned long long partiton_replys_in_parallel = 0; unsigned long long partiton_replys_in_parallel_total = 0; @@ -754,6 +756,7 @@ void gpgpu_sim::init() partiton_reqs_in_parallel = 0; partiton_replys_in_parallel = 0; partiton_reqs_in_parallel_util = 0; + gpu_sim_cycle_parition_util = 0; reinit_clock_domains(); set_param_gpgpu_num_shaders(m_config.num_shader()); @@ -793,11 +796,14 @@ void gpgpu_sim::update_stats() { partiton_reqs_in_parallel_total += partiton_reqs_in_parallel; partiton_replys_in_parallel_total += partiton_replys_in_parallel; partiton_reqs_in_parallel_util_total += partiton_reqs_in_parallel_util; + gpu_tot_sim_cycle_parition_util += gpu_sim_cycle_parition_util ; gpu_sim_cycle = 0; partiton_reqs_in_parallel = 0; partiton_replys_in_parallel = 0; partiton_reqs_in_parallel_util = 0; + gpu_sim_cycle_parition_util = 0; + gpu_tot_sim_cycle = 0; gpu_sim_insn = 0; m_total_cta_launched = 0; } @@ -987,8 +993,10 @@ void gpgpu_sim::gpu_print_stat() printf("partiton_level_parallism_total = %12.4f\n", (float)(partiton_reqs_in_parallel+partiton_reqs_in_parallel_total) / (gpu_tot_sim_cycle+gpu_sim_cycle) ); printf("partiton_reqs_in_parallel_util = %lld\n", partiton_reqs_in_parallel_util); printf("partiton_reqs_in_parallel_util_total = %lld\n", partiton_reqs_in_parallel_util_total ); - printf("partiton_level_parallism_util = %12.4f\n", (float)partiton_reqs_in_parallel_util / gpu_sim_cycle); - printf("partiton_level_parallism_util_total = %12.4f\n", (float)(partiton_reqs_in_parallel_util+partiton_reqs_in_parallel_util_total) / (gpu_tot_sim_cycle+gpu_sim_cycle) ); + printf("gpu_sim_cycle_parition_util = %lld\n", gpu_sim_cycle_parition_util); + printf("gpu_tot_sim_cycle_parition_util = %lld\n", gpu_tot_sim_cycle_parition_util ); + printf("partiton_level_parallism_util = %12.4f\n", (float)partiton_reqs_in_parallel_util / gpu_sim_cycle_parition_util); + printf("partiton_level_parallism_util_total = %12.4f\n", (float)(partiton_reqs_in_parallel_util+partiton_reqs_in_parallel_util_total) / (gpu_sim_cycle_parition_util+gpu_tot_sim_cycle_parition_util) ); printf("partiton_replys_in_parallel = %lld\n", partiton_replys_in_parallel); printf("partiton_replys_in_parallel_total = %lld\n", partiton_replys_in_parallel_total ); printf("L2_BW = %12.4f GB/Sec\n", ((float)(partiton_replys_in_parallel * 32) / (gpu_sim_cycle * m_config.icnt_period)) / 1000000000); @@ -1448,8 +1456,10 @@ void gpgpu_sim::cycle() } } partiton_reqs_in_parallel += partiton_reqs_in_parallel_per_cycle; - if(partiton_reqs_in_parallel_per_cycle > 0) + if(partiton_reqs_in_parallel_per_cycle > 0){ partiton_reqs_in_parallel_util += partiton_reqs_in_parallel_per_cycle; + gpu_sim_cycle_parition_util++; + } if (clock_mask & ICNT) { icnt_transfer(); -- cgit v1.3 From cfab001e7c65541eaf8f7573bbf4a2e5d69f296e Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Fri, 30 Mar 2018 11:27:57 -0400 Subject: fixing gpu-tot-cycle bug --- src/gpgpu-sim/gpu-sim.cc | 1 - 1 file changed, 1 deletion(-) (limited to 'src/gpgpu-sim') diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index da35dc3..550c550 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -803,7 +803,6 @@ void gpgpu_sim::update_stats() { partiton_replys_in_parallel = 0; partiton_reqs_in_parallel_util = 0; gpu_sim_cycle_parition_util = 0; - gpu_tot_sim_cycle = 0; gpu_sim_insn = 0; m_total_cta_launched = 0; } -- cgit v1.3 From 373d64290239f3ed74d98b20494383f03fe189b6 Mon Sep 17 00:00:00 2001 From: Nathan Conrad Date: Sun, 22 Apr 2018 23:31:42 -0400 Subject: Some classes were referred to as a class and a struct (reported as clang warnings). This makes these consistent. --- libcuda/cuda_runtime_api.cc | 2 +- src/cuda-sim/cuda-sim.cc | 2 +- src/cuda-sim/ptx_ir.h | 2 +- src/debug.cc | 2 +- src/debug.h | 2 +- src/gpgpu-sim/dram.h | 4 ++-- src/gpgpu-sim/mem_fetch.cc | 2 +- src/gpgpu-sim/mem_fetch.h | 4 ++-- src/gpgpu-sim/shader.h | 4 ++-- src/stream_manager.h | 6 +++--- 10 files changed, 15 insertions(+), 15 deletions(-) (limited to 'src/gpgpu-sim') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 9bdb993..5ef6115 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -324,7 +324,7 @@ private: gpgpu_ptx_sim_arg_list_t m_args; }; -class _cuda_device_id *GPGPUSim_Init() +struct _cuda_device_id *GPGPUSim_Init() { static _cuda_device_id *the_device = NULL; if( !the_device ) { diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index a34b99b..9f24c69 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -2067,7 +2067,7 @@ struct rec_pts { int s_num_recon; }; -struct std::map g_rpts; +class std::map g_rpts; struct rec_pts find_reconvergence_points( function_info *finfo ) { diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h index 9ad1571..8750187 100644 --- a/src/cuda-sim/ptx_ir.h +++ b/src/cuda-sim/ptx_ir.h @@ -105,7 +105,7 @@ private: int m_is_function; bool m_is_non_arch_reg; - friend class type_info_key_compare; + friend struct type_info_key_compare; }; class symbol_table; diff --git a/src/debug.cc b/src/debug.cc index cfd7bb0..ae15760 100644 --- a/src/debug.cc +++ b/src/debug.cc @@ -222,7 +222,7 @@ void gpgpu_sim::gpgpu_debug() } } -bool thread_at_brkpt( ptx_thread_info *thread, const struct brk_pt &b ) +bool thread_at_brkpt( ptx_thread_info *thread, const class brk_pt &b ) { return b.is_equal(thread->get_location(),thread->get_uid()); } diff --git a/src/debug.h b/src/debug.h index 7c79f1e..1277494 100644 --- a/src/debug.h +++ b/src/debug.h @@ -87,7 +87,7 @@ extern int gpgpu_ptx_instruction_classification ; class ptx_thread_info; class ptx_instruction; -bool thread_at_brkpt( ptx_thread_info *thd_info, const struct brk_pt &b ); +bool thread_at_brkpt( ptx_thread_info *thd_info, const class brk_pt &b ); void hit_watchpoint( unsigned watchpoint_num, ptx_thread_info *thd, const ptx_instruction *pI ); #endif diff --git a/src/gpgpu-sim/dram.h b/src/gpgpu-sim/dram.h index a8bff14..15c63e7 100644 --- a/src/gpgpu-sim/dram.h +++ b/src/gpgpu-sim/dram.h @@ -87,7 +87,7 @@ struct bank_t unsigned int bkgrpindex; }; -struct mem_fetch; +class mem_fetch; class dram_t { @@ -178,7 +178,7 @@ private: unsigned int ave_mrqs_partial; unsigned int bwutil_partial; - struct memory_stats_t *m_stats; + class memory_stats_t *m_stats; class Stats* mrqq_Dist; //memory request queue inside DRAM friend class frfcfs_scheduler; diff --git a/src/gpgpu-sim/mem_fetch.cc b/src/gpgpu-sim/mem_fetch.cc index 580c051..729636d 100644 --- a/src/gpgpu-sim/mem_fetch.cc +++ b/src/gpgpu-sim/mem_fetch.cc @@ -39,7 +39,7 @@ mem_fetch::mem_fetch( const mem_access_t &access, unsigned wid, unsigned sid, unsigned tpc, - const class memory_config *config ) + const struct memory_config *config ) { m_request_uid = sm_next_mf_request_uid++; m_access = access; diff --git a/src/gpgpu-sim/mem_fetch.h b/src/gpgpu-sim/mem_fetch.h index c89edbb..de98748 100644 --- a/src/gpgpu-sim/mem_fetch.h +++ b/src/gpgpu-sim/mem_fetch.h @@ -55,7 +55,7 @@ public: unsigned wid, unsigned sid, unsigned tpc, - const class memory_config *config ); + const struct memory_config *config ); ~mem_fetch(); void set_status( enum mem_fetch_status status, unsigned long long cycle ); @@ -141,7 +141,7 @@ private: static unsigned sm_next_mf_request_uid; - const class memory_config *m_mem_config; + const struct memory_config *m_mem_config; unsigned icnt_flit_size; }; diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index bdd8dbe..ea8c019 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -286,7 +286,7 @@ typedef std::bitset warp_set_t; int register_bank(int regnum, int wid, unsigned num_banks, unsigned bank_warp_shift); class shader_core_ctx; -class shader_core_config; +struct shader_core_config; class shader_core_stats; enum scheduler_prioritization_type @@ -967,7 +967,7 @@ struct ifetch_buffer_t { unsigned m_warp_id; }; -class shader_core_config; +struct shader_core_config; class simd_function_unit { public: diff --git a/src/stream_manager.h b/src/stream_manager.h index 222a1b2..d3a804f 100644 --- a/src/stream_manager.h +++ b/src/stream_manager.h @@ -93,7 +93,7 @@ public: m_stream=stream; m_done=false; } - stream_operation( class CUevent_st *e, struct CUstream_st *stream ) + stream_operation( struct CUevent_st *e, struct CUstream_st *stream ) { m_kernel=NULL; m_type=stream_event; @@ -172,10 +172,10 @@ private: bool m_sim_mode; kernel_info_t *m_kernel; - class CUevent_st *m_event; + struct CUevent_st *m_event; }; -class CUevent_st { +struct CUevent_st { public: CUevent_st( bool blocking ) { -- cgit v1.3 From c246c12dffa6683a1fe61ad3a311cce8b1c1cbbf Mon Sep 17 00:00:00 2001 From: Nathan Conrad Date: Sun, 22 Apr 2018 23:35:57 -0400 Subject: Clang disallows empty bracks to show array indices. I believe that inserting a zero is the intent. --- src/gpgpu-sim/shader.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/gpgpu-sim') diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index ea8c019..53a10e0 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -1347,7 +1347,7 @@ struct shader_core_config : public core_config struct shader_core_stats_pod { - void* shader_core_stats_pod_start[]; // DO NOT MOVE FROM THE TOP - spaceless pointer to the start of this structure + void* shader_core_stats_pod_start[0]; // DO NOT MOVE FROM THE TOP - spaceless pointer to the start of this structure unsigned long long *shader_cycles; unsigned *m_num_sim_insn; // number of scalar thread instructions committed by this shader core unsigned *m_num_sim_winsn; // number of warp instructions committed by this shader core -- cgit v1.3