diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/gpgpu-sim/gpu-sim.cc | 49 | ||||
| -rw-r--r-- | src/gpgpu-sim/shader.cc | 132 |
2 files changed, 97 insertions, 84 deletions
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 33393ee..dcbdfe0 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -909,32 +909,35 @@ void gpgpu_sim::gpu_print_stat() m_memory_partition_unit[i]->print(stdout); // L2 cache stats - cache_stats l2_stats; - struct cache_sub_stats l2_css; - struct cache_sub_stats total_l2_css; - l2_stats.clear(); - l2_css.clear(); - total_l2_css.clear(); + if(!m_memory_config->m_L2_config.disabled()){ + cache_stats l2_stats; + struct cache_sub_stats l2_css; + struct cache_sub_stats total_l2_css; + l2_stats.clear(); + l2_css.clear(); + total_l2_css.clear(); - printf("\n========= L2 cache stats =========\n"); - for (unsigned i=0;i<m_memory_config->m_n_mem;i++){ - m_memory_partition_unit[i]->accumulate_L2cache_stats(l2_stats); - m_memory_partition_unit[i]->get_L2cache_sub_stats(l2_css); + printf("\n========= L2 cache stats =========\n"); + for (unsigned i=0;i<m_memory_config->m_n_mem;i++){ + m_memory_partition_unit[i]->accumulate_L2cache_stats(l2_stats); + m_memory_partition_unit[i]->get_L2cache_sub_stats(l2_css); - fprintf( stdout, "L2_cache_bank[%d]: Access = %u, Miss = %u, Miss_rate = %.3lf, Pending_hits = %u, Reservation_fails = %u\n", - i, l2_css.accesses, l2_css.misses, (double)l2_css.misses / (double)l2_css.accesses, l2_css.pending_hits, l2_css.res_fails); + fprintf( stdout, "L2_cache_bank[%d]: Access = %u, Miss = %u, Miss_rate = %.3lf, Pending_hits = %u, Reservation_fails = %u\n", + i, l2_css.accesses, l2_css.misses, (double)l2_css.misses / (double)l2_css.accesses, l2_css.pending_hits, l2_css.res_fails); - total_l2_css += l2_css; - } - if (!m_memory_config->m_L2_config.disabled() && m_memory_config->m_L2_config.get_num_lines()) { - //L2c_print_cache_stat(); - printf("L2_total_cache_accesses = %u\n", total_l2_css.accesses); - printf("L2_total_cache_misses = %u\n", total_l2_css.misses); - printf("L2_total_cache_miss_rate = %.4lf\n", (double)total_l2_css.misses/(double)total_l2_css.accesses); - printf("L2_total_cache_pending_hits = %u\n", total_l2_css.pending_hits); - printf("L2_total_cache_reservation_fails = %u\n", total_l2_css.res_fails); - printf("L2_total_cache_breakdown:\n"); - l2_stats.print_stats(stdout, "L2_cache_stats_breakdown"); + total_l2_css += l2_css; + } + if (!m_memory_config->m_L2_config.disabled() && m_memory_config->m_L2_config.get_num_lines()) { + //L2c_print_cache_stat(); + printf("L2_total_cache_accesses = %u\n", total_l2_css.accesses); + printf("L2_total_cache_misses = %u\n", total_l2_css.misses); + if(total_l2_css.accesses > 0) + printf("L2_total_cache_miss_rate = %.4lf\n", (double)total_l2_css.misses/(double)total_l2_css.accesses); + printf("L2_total_cache_pending_hits = %u\n", total_l2_css.pending_hits); + printf("L2_total_cache_reservation_fails = %u\n", total_l2_css.res_fails); + printf("L2_total_cache_breakdown:\n"); + l2_stats.print_stats(stdout, "L2_cache_stats_breakdown"); + } } if (m_config.gpgpu_cflog_interval != 0) { diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index e108963..ba79721 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -1110,13 +1110,16 @@ void ldst_unit::get_cache_stats(cache_stats &cs) { } void ldst_unit::get_L1D_sub_stats(struct cache_sub_stats &css) const{ - m_L1D->get_sub_stats(css); + if(m_L1D) + m_L1D->get_sub_stats(css); } void ldst_unit::get_L1C_sub_stats(struct cache_sub_stats &css) const{ - m_L1C->get_sub_stats(css); + if(m_L1C) + m_L1C->get_sub_stats(css); } void ldst_unit::get_L1T_sub_stats(struct cache_sub_stats &css) const{ - m_L1T->get_sub_stats(css); + if(m_L1T) + m_L1T->get_sub_stats(css); } void shader_core_ctx::warp_inst_complete(const warp_inst_t &inst) @@ -1878,75 +1881,81 @@ void gpgpu_sim::shader_print_cache_stats( FILE *fout ) const{ struct cache_sub_stats total_css; struct cache_sub_stats css; - total_css.clear(); - css.clear(); - - fprintf(fout, "\n========= Core cache stats =========\n"); - fprintf(fout, "L1I_cache:\n"); - for ( unsigned i = 0; i < m_shader_config->n_simt_clusters; ++i ) { - m_cluster[i]->get_L1I_sub_stats(css); - total_css += css; - } - fprintf(fout, "\tL1I_total_cache_accesses: %u\n", total_css.accesses); - fprintf(fout, "\tL1I_total_cache_misses: %u\n", total_css.misses); - if(total_css.accesses > 0){ - fprintf(fout, "\tL1I_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + if(!m_shader_config->m_L1I_config.disabled()){ + total_css.clear(); + css.clear(); + fprintf(fout, "\n========= Core cache stats =========\n"); + fprintf(fout, "L1I_cache:\n"); + for ( unsigned i = 0; i < m_shader_config->n_simt_clusters; ++i ) { + m_cluster[i]->get_L1I_sub_stats(css); + total_css += css; + } + fprintf(fout, "\tL1I_total_cache_accesses: %u\n", total_css.accesses); + fprintf(fout, "\tL1I_total_cache_misses: %u\n", total_css.misses); + if(total_css.accesses > 0){ + fprintf(fout, "\tL1I_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + } + fprintf(fout, "\tL1I_total_cache_pending_hits: %u\n", total_css.pending_hits); + fprintf(fout, "\tL1I_total_cache_reservation_fails: %u\n", total_css.res_fails); } - fprintf(fout, "\tL1I_total_cache_pending_hits: %u\n", total_css.pending_hits); - fprintf(fout, "\tL1I_total_cache_reservation_fails: %u\n", total_css.res_fails); // L1D - total_css.clear(); - css.clear(); - fprintf(fout, "L1D_cache:\n"); - for (unsigned i=0;i<m_shader_config->n_simt_clusters;i++){ - m_cluster[i]->get_L1D_sub_stats(css); + if(!m_shader_config->m_L1D_config.disabled()){ + total_css.clear(); + css.clear(); + fprintf(fout, "L1D_cache:\n"); + for (unsigned i=0;i<m_shader_config->n_simt_clusters;i++){ + m_cluster[i]->get_L1D_sub_stats(css); - fprintf( stdout, "\tL1D_cache_core[%d]: Access = %d, Miss = %d, Miss_rate = %.3lf, Pending_hits = %u, Reservation_fails = %u\n", - i, css.accesses, css.misses, (double)css.misses / (double)css.accesses, css.pending_hits, css.res_fails); + fprintf( stdout, "\tL1D_cache_core[%d]: Access = %d, Miss = %d, Miss_rate = %.3lf, Pending_hits = %u, Reservation_fails = %u\n", + i, css.accesses, css.misses, (double)css.misses / (double)css.accesses, css.pending_hits, css.res_fails); - total_css += css; - } - fprintf(fout, "\tL1D_total_cache_accesses: %u\n", total_css.accesses); - fprintf(fout, "\tL1D_total_cache_misses: %u\n", total_css.misses); - if(total_css.accesses > 0){ - fprintf(fout, "\tL1D_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + total_css += css; + } + fprintf(fout, "\tL1D_total_cache_accesses: %u\n", total_css.accesses); + fprintf(fout, "\tL1D_total_cache_misses: %u\n", total_css.misses); + if(total_css.accesses > 0){ + fprintf(fout, "\tL1D_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + } + fprintf(fout, "\tL1D_total_cache_pending_hits: %u\n", total_css.pending_hits); + fprintf(fout, "\tL1D_total_cache_reservation_fails: %u\n", total_css.res_fails); } - fprintf(fout, "\tL1D_total_cache_pending_hits: %u\n", total_css.pending_hits); - fprintf(fout, "\tL1D_total_cache_reservation_fails: %u\n", total_css.res_fails); - // L1C - total_css.clear(); - css.clear(); - fprintf(fout, "L1C_cache:\n"); - for ( unsigned i = 0; i < m_shader_config->n_simt_clusters; ++i ) { - m_cluster[i]->get_L1C_sub_stats(css); - total_css += css; - } - fprintf(fout, "\tL1C_total_cache_accesses: %u\n", total_css.accesses); - fprintf(fout, "\tL1C_total_cache_misses: %u\n", total_css.misses); - if(total_css.accesses > 0){ - fprintf(fout, "\tL1C_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + if(!m_shader_config->m_L1C_config.disabled()){ + total_css.clear(); + css.clear(); + fprintf(fout, "L1C_cache:\n"); + for ( unsigned i = 0; i < m_shader_config->n_simt_clusters; ++i ) { + m_cluster[i]->get_L1C_sub_stats(css); + total_css += css; + } + fprintf(fout, "\tL1C_total_cache_accesses: %u\n", total_css.accesses); + fprintf(fout, "\tL1C_total_cache_misses: %u\n", total_css.misses); + if(total_css.accesses > 0){ + fprintf(fout, "\tL1C_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + } + fprintf(fout, "\tL1C_total_cache_pending_hits: %u\n", total_css.pending_hits); + fprintf(fout, "\tL1C_total_cache_reservation_fails: %u\n", total_css.res_fails); } - fprintf(fout, "\tL1C_total_cache_pending_hits: %u\n", total_css.pending_hits); - fprintf(fout, "\tL1C_total_cache_reservation_fails: %u\n", total_css.res_fails); // L1T - total_css.clear(); - css.clear(); - fprintf(fout, "L1T_cache:\n"); - for ( unsigned i = 0; i < m_shader_config->n_simt_clusters; ++i ) { - m_cluster[i]->get_L1T_sub_stats(css); - total_css += css; - } - fprintf(fout, "\tL1T_total_cache_accesses: %u\n", total_css.accesses); - fprintf(fout, "\tL1T_total_cache_misses: %u\n", total_css.misses); - if(total_css.accesses > 0){ - fprintf(fout, "\tL1T_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + if(!m_shader_config->m_L1T_config.disabled()){ + total_css.clear(); + css.clear(); + fprintf(fout, "L1T_cache:\n"); + for ( unsigned i = 0; i < m_shader_config->n_simt_clusters; ++i ) { + m_cluster[i]->get_L1T_sub_stats(css); + total_css += css; + } + fprintf(fout, "\tL1T_total_cache_accesses: %u\n", total_css.accesses); + fprintf(fout, "\tL1T_total_cache_misses: %u\n", total_css.misses); + if(total_css.accesses > 0){ + fprintf(fout, "\tL1T_total_cache_miss_rate: %.4lf\n", (double)total_css.misses / (double)total_css.accesses); + } + fprintf(fout, "\tL1T_total_cache_pending_hits: %u\n", total_css.pending_hits); + fprintf(fout, "\tL1T_total_cache_reservation_fails: %u\n", total_css.res_fails); } - fprintf(fout, "\tL1T_total_cache_pending_hits: %u\n", total_css.pending_hits); - fprintf(fout, "\tL1T_total_cache_reservation_fails: %u\n", total_css.res_fails); } void gpgpu_sim::shader_print_l1_miss_stat( FILE *fout ) const @@ -2613,7 +2622,8 @@ void shader_core_ctx::get_cache_stats(cache_stats &cs){ } void shader_core_ctx::get_L1I_sub_stats(struct cache_sub_stats &css) const{ - m_L1I->get_sub_stats(css); + if(m_L1I) + m_L1I->get_sub_stats(css); } void shader_core_ctx::get_L1D_sub_stats(struct cache_sub_stats &css) const{ m_ldst_unit->get_L1D_sub_stats(css); |