Merge branch 'dev' of https://github.com/gpgpu-sim/gpgpu-sim_distribution into dev

13 files changed, 129 insertions, 41 deletions

diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index aaa4b00..7125b6b 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -383,7 +383,7 @@ protected:
     std::deque<simt_stack_entry> m_stack;
 };
 
-#define GLOBAL_HEAP_START 0x703E20000
+#define GLOBAL_HEAP_START 0xC0000000
    // start allocating from this address (lower values used for allocating globals in .ptx file)
 #define SHARED_MEM_SIZE_MAX (64*1024)
 #define LOCAL_MEM_SIZE_MAX (8*1024)
diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h
index 4721e8a..afac330 100644
--- a/src/cuda-sim/cuda-math.h
+++ b/src/cuda-sim/cuda-math.h
@@ -67,6 +67,8 @@
 #ifndef CUDA_MATH
 #define CUDA_MATH
 
+#include <cmath>
+
 // cuda math implementations
 #undef max
 #undef min
@@ -148,6 +150,7 @@ float __ll2float_rd(long long int a) {
 #include <device_types.h>
 #include <fenv.h>
 
+
 // 32-bit integer to float
 float __int2float_rn(int a) {
    int orig_rnd_mode = fegetround();
@@ -321,7 +324,7 @@ float __internal_accurate_fdividef(float a, float b)
 float __saturatef(float a)
 {
    float b; 
-   if (isnan(a)) b = 0.0f; 
+   if (std::isnan(a)) b = 0.0f; 
    else if (a >= 1.0f) b = 1.0f;
    else if (a <= 0.0f) b = 0.0f; 
    else b = a; 
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index d4ace76..a34b99b 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -1408,7 +1408,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id)
    if ( (g_ptx_sim_num_insn % 100000) == 0 ) {
       dim3 ctaid = get_ctaid();
       dim3 tid = get_tid();
-      printf("GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
+      DPRINTF(LIVENESS, "GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
              g_ptx_sim_num_insn, ctaid.x,ctaid.y,ctaid.z,tid.x,tid.y,tid.z );
       fflush(stdout);
    }
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 011c285..71286c9 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -1961,7 +1961,7 @@ ptx_reg_t d2d( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign,
       y.f64 = x.f64;
       break; 
    }
-   if (isnan(y.f64)) {
+   if (std::isnan(y.f64)) {
       y.u64 = 0xfff8000000000000ull;
    } else if (saturation_mode) {
       y.f64 = cuda_math::__saturatef(y.f64); 
@@ -2086,7 +2086,7 @@ void ptx_round(ptx_reg_t& data, int rounding_mode, int type)
       }
    }
    if ((type == F64_TYPE)||(type == FF64_TYPE)) {
-      if (isnan(data.f64)) {
+      if (std::isnan(data.f64)) {
          data.u64 = 0xfff8000000000000ull;
       }
    }
@@ -2648,12 +2648,12 @@ void mad_def( const ptx_instruction *pI, ptx_thread_info *thread, bool use_carry
 
 bool isNaN(float x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 bool isNaN(double x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 void max_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index e00aa4b..4edae5d 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -47,7 +47,6 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 %token  PTR_DIRECTIVE
 %token  ENTRY_DIRECTIVE
 %token  EXTERN_DIRECTIVE
-%token  WEAK_DIRECTIVE
 %token  FILE_DIRECTIVE
 %token  FUNC_DIRECTIVE
 %token  GLOBAL_DIRECTIVE
@@ -513,7 +512,8 @@ compare_spec:EQ_OPTION { add_option(EQ_OPTION); }
 	| NAN_OPTION { add_option(NAN_OPTION); } 
 	;
 
-operand_list: operand
+operand_list: /* empty*/
+    | operand
 	| operand COMMA operand_list;
 
 operand: IDENTIFIER  { add_scalar_operand( $1 ); }
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index baa3bcd..a180da9 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -433,7 +433,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
       assert( (num_bits%8) == 0  );
       addr = g_current_symbol_table->get_shared_next();
       addr_pad = pad_address(addr, num_bits/8, 128);
-      printf("from 0x%x to 0x%lx (shared memory space)\n",
+      printf("from 0x%llx to 0x%llx (shared memory space)\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8);
          fflush(stdout);
@@ -450,7 +450,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
          assert( (num_bits%8) == 0  ); 
          addr = g_current_symbol_table->get_global_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (global memory space) %u\n",
+         printf("from 0x%llx to 0x%llx (global memory space) %u\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8,
               g_const_alloc++);
@@ -471,7 +471,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
       assert( (num_bits%8) == 0  );
       addr = g_current_symbol_table->get_global_next();
       addr_pad = pad_address(addr, num_bits/8, 128);
-      printf("from 0x%x to 0x%lx (global memory space)\n",
+      printf("from 0x%llx to 0x%llx (global memory space)\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8);
       fflush(stdout);
@@ -488,7 +488,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
          assert( (num_bits%8) == 0  );
          addr = g_current_symbol_table->get_local_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (local memory space)\n",
+         printf("from 0x%llx to 0x%llx (local memory space)\n",
                  addr+addr_pad,
                  addr+addr_pad + num_bits/8);
          fflush(stdout);
@@ -501,7 +501,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
         assert( (num_bits%8) == 0 );
         addr = g_current_symbol_table->get_local_next();
         addr_pad = pad_address(addr, num_bits/8, 128);
-        printf("from 0x%x to 0x%lx\n",
+        printf("from 0x%llx to 0x%llx\n",
                 addr+addr_pad,
                 addr+addr_pad + num_bits/8);
         fflush(stdout);
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 "<cache_name>[<request_type>][<request_status>] = <stat_value>"
     ///
+    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
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 58a5d16..bd4c00c 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -32,10 +32,12 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <math.h>
+#include <signal.h>
 #include "zlib.h"
 
 
 #include "shader.h"
+#include "shader_trace.h"
 #include "dram.h"
 #include "mem_fetch.h"
 
@@ -87,6 +89,14 @@ 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  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;
 
 /* Clock Domains */
 
@@ -223,7 +233,7 @@ void shader_core_config::reg_options(class OptionParser * opp)
                    "per-shader L1 data cache config "
                    " {<nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>,<mshr>:<N>:<merge>,<mq> | none}",
                    "none" );
-    option_parser_register(opp, "-gpgpu_cache:dl1PreShared", OPT_CSTR, &m_L1D_config.m_config_stringPrefShared,
+    option_parser_register(opp, "-gpgpu_cache:dl1PrefShared", OPT_CSTR, &m_L1D_config.m_config_stringPrefShared,
                    "per-shader L1 data cache config "
                    " {<nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>,<mshr>:<N>:<merge>,<mq> | none}",
                    "none" );
@@ -744,6 +754,10 @@ 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;
+    gpu_sim_cycle_parition_util = 0;
 
     reinit_clock_domains();
     set_param_gpgpu_num_shaders(m_config.num_shader());
@@ -780,8 +794,16 @@ 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_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_sim_insn = 0;
     m_total_cta_launched = 0;
 }
@@ -965,6 +987,21 @@ 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("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);
+   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 );
@@ -1175,8 +1212,8 @@ bool shader_core_ctx::occupy_shader_resource_1block(kernel_info_t & k, bool occu
        m_occupied_regs += (padded_cta_size * ((kernel_info->regs+3)&~3));
        m_occupied_ctas++;
 
-       printf("GPGPU-Sim uArch: Shader %d occupied %d threads, %d shared mem, %d registers, %d ctas\n",
-            m_sid, m_occupied_n_threads, m_occupied_shmem, m_occupied_regs, m_occupied_ctas);  
+      SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Occupied %d threads, %d shared mem, %d registers, %d ctas\n",
+            m_occupied_n_threads, m_occupied_shmem, m_occupied_regs, m_occupied_ctas);  
    }
 
    return true;
@@ -1301,8 +1338,8 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel )
     m_n_active_cta++;
 
     shader_CTA_count_log(m_sid, 1);
-    printf("GPGPU-Sim uArch: core:%3d, cta:%2u, start_tid:%4u, end_tid:%4u, initialized @(%lld,%lld)\n", 
-        m_sid, free_cta_hw_id, start_thread, end_thread, gpu_sim_cycle, gpu_tot_sim_cycle );
+    SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: cta:%2u, start_tid:%4u, end_tid:%4u, initialized @(%lld,%lld)\n", 
+        free_cta_hw_id, start_thread, end_thread, gpu_sim_cycle, gpu_tot_sim_cycle );
 
 }
 
@@ -1366,6 +1403,7 @@ void gpgpu_sim::cycle()
       for (unsigned i=0;i<m_shader_config->n_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;i<m_memory_config->m_n_mem_sub_partition;i++) {
@@ -1378,6 +1416,7 @@ 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++;
                 }
@@ -1386,6 +1425,7 @@ void gpgpu_sim::cycle()
             }
         }
     }
+    partiton_replys_in_parallel += partiton_replys_in_parallel_per_cycle;
 
    if (clock_mask & DRAM) {
       for (unsigned i=0;i<m_memory_config->m_n_mem;i++){
@@ -1398,6 +1438,7 @@ 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;i<m_memory_config->m_n_mem_sub_partition;i++) {
@@ -1408,11 +1449,17 @@ 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;
+	   gpu_sim_cycle_parition_util++;
+   }
 
    if (clock_mask & ICNT) {
       icnt_transfer();
@@ -1440,7 +1487,7 @@ 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 ) 
@@ -1499,7 +1546,8 @@ void gpgpu_sim::cycle()
             hrs     = elapsed_time/3600 - 24*days;
             minutes = elapsed_time/60 - 60*(hrs + 24*days);
             sec = elapsed_time - 60*(minutes + 60*(hrs + 24*days));
-            printf("GPGPU-Sim uArch: cycles simulated: %lld  inst.: %lld (ipc=%4.1f) sim_rate=%u (inst/sec) elapsed = %u:%u:%02u:%02u / %s", 
+
+            DPRINTF(LIVENESS, "GPGPU-Sim uArch: cycles simulated: %lld  inst.: %lld (ipc=%4.1f) sim_rate=%u (inst/sec) elapsed = %u:%u:%02u:%02u / %s", 
                    gpu_tot_sim_cycle + gpu_sim_cycle, gpu_tot_sim_insn + gpu_sim_insn, 
                    (double)gpu_sim_insn/(double)gpu_sim_cycle,
                    (unsigned)((gpu_tot_sim_insn+gpu_sim_insn) / elapsed_time),
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index a2d1b9b..7d92c66 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -195,7 +195,7 @@ struct memory_config {
       for (i=0; nbkt>0; i++) {
           nbkt = nbkt>>1;
       }
-      bk_tag_length = i;
+      bk_tag_length = i-1;
       assert(nbkgrp>0 && "Number of bank groups cannot be zero");
       tRCDWR = tRCD-(WL+1);
       tRTW = (CL+(BL/data_command_freq_ratio)+2-WL);
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index d17e51d..4640d65 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -1969,12 +1969,12 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t
       m_barriers.deallocate_barrier(cta_num);
       shader_CTA_count_unlog(m_sid, 1);
 
-      printf("GPGPU-Sim uArch: Shader %d finished CTA #%d (%lld,%lld), %u CTAs running\n", m_sid, cta_num, gpu_sim_cycle, gpu_tot_sim_cycle,
-             m_n_active_cta );
+     SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Finished CTA #%d (%lld,%lld), %u CTAs running\n",
+        cta_num, gpu_sim_cycle, gpu_tot_sim_cycle, m_n_active_cta);
 
       if( m_n_active_cta == 0 ) {
-          printf("GPGPU-Sim uArch: Shader %u empty (last released kernel %u \'%s\').\n", m_sid, kernel->get_uid(),
-                 kernel->name().c_str() );
+        SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Empty (last released kernel %u \'%s\').\n",
+            kernel->get_uid(), kernel->name().c_str());
           fflush(stdout);
 
           //Shader can only be empty when no more cta are dispatched
@@ -1989,8 +1989,10 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t
       kernel->dec_running();
       if( !m_gpu->kernel_more_cta_left(kernel) ) {
           if( !kernel->running() ) {
-              printf("GPGPU-Sim uArch: GPU detected kernel %u \'%s\' finished on shader %u.\n", kernel->get_uid(),
-                kernel->name().c_str(), m_sid );
+              SHADER_DPRINTF(LIVENESS,
+                "GPGPU-Sim uArch: GPU detected kernel %u \'%s\' finished on shader %u.\n", kernel->get_uid(),
+                kernel->name().c_str(), m_sid);
+
               if(m_kernel == kernel)
                 m_kernel = NULL;
               m_gpu->set_kernel_done( kernel );
diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc
index 04845e7..ede9f20 100644
--- a/src/gpgpusim_entrypoint.cc
+++ b/src/gpgpusim_entrypoint.cc
@@ -91,9 +91,17 @@ void *gpgpu_sim_thread_sequential(void*)
 pthread_mutex_t g_sim_lock = PTHREAD_MUTEX_INITIALIZER;
 bool g_sim_active = false;
 bool g_sim_done = true;
+bool break_limit = false;
+
+static void termination_callback()
+{
+    printf("GPGPU-Sim: *** exit detected ***\n");
+    fflush(stdout);
+}
 
 void *gpgpu_sim_thread_concurrent(void*)
 {
+    atexit(termination_callback);
     // concurrent kernel execution simulation thread
     do {
         if(g_debug_execution >= 3) {
@@ -144,17 +152,19 @@ void *gpgpu_sim_thread_concurrent(void*)
                 if(g_the_gpu->cycle_insn_cta_max_hit()){
                     g_stream_manager->stop_all_running_kernels();
                     g_sim_done = true;
+                    break_limit = true;
                 }
             }
 
             active=g_the_gpu->active() || !g_stream_manager->empty_protected();
-        } while( active );
+
+        } while( active && !g_sim_done);
         if(g_debug_execution >= 3) {
            printf("GPGPU-Sim: ** STOP simulation thread (no work) **\n");
            fflush(stdout);
         }
-        g_the_gpu->print_stats();
         if(sim_cycles) {
+            g_the_gpu->print_stats();
             g_the_gpu->update_stats();
             print_simulation_time();
         }
@@ -162,10 +172,14 @@ void *gpgpu_sim_thread_concurrent(void*)
         g_sim_active = false;
         pthread_mutex_unlock(&g_sim_lock);
     } while( !g_sim_done );
-    if(g_debug_execution >= 3) {
-       printf("GPGPU-Sim: *** simulation thread exiting ***\n");
-       fflush(stdout);
+    printf("GPGPU-Sim: *** simulation thread exiting ***\n");
+    fflush(stdout);
+
+    if(break_limit) {
+    	printf("GPGPU-Sim: ** break due to reaching the maximum cycles (or instructions) **\n");
+    	exit(1);
     }
+
     sem_post(&g_sim_signal_exit);
     return NULL;
 }
@@ -179,7 +193,7 @@ void synchronize()
     bool done = false;
     do {
         pthread_mutex_lock(&g_sim_lock);
-        done = g_stream_manager->empty() && !g_sim_active;
+        done = ( g_stream_manager->empty() && !g_sim_active ) || g_sim_done;
         pthread_mutex_unlock(&g_sim_lock);
     } while (!done);
     printf("GPGPU-Sim: detected inactive GPU simulation thread\n");
diff --git a/src/intersim2/Makefile b/src/intersim2/Makefile
index ef948d6..bd42000 100644
--- a/src/intersim2/Makefile
+++ b/src/intersim2/Makefile
@@ -125,6 +125,14 @@ endif
 
 # rules to compile simulator
 
+$(OBJDIR)/Makefile.makedepend: depend
+
+ALL_SRCS = $(CPP_SRCS)
+ALL_SRCS += $(shell ls **/*.cpp)
+
+depend:
+	touch $(OBJDIR)/Makefile.makedepend
+	makedepend -f$(OBJDIR)/Makefile.makedepend -I$(INCPATH) -p$(OBJDIR)/ $(ALL_SRCS) 2> /dev/null
 
 ${LEX_OBJS}: $(OBJDIR)/lex.yy.c $(OBJDIR)/y.tab.h
 	$(CC) $(CPPFLAGS) -c $< -o $@
@@ -173,3 +181,5 @@ $(OBJDIR)/y.tab.c $(OBJDIR)/y.tab.h: config.y
 
 $(OBJDIR)/lex.yy.c: config.l
 	$(LEX) -o$@ $<
+
+include $(OBJDIR)/Makefile.makedepend
diff --git a/src/trace_streams.tup b/src/trace_streams.tup
index c41690e..fd68a16 100644
--- a/src/trace_streams.tup
+++ b/src/trace_streams.tup
@@ -29,5 +29,6 @@ TS_TUP_BEGIN( trace_streams_type )
     TS_TUP( WARP_SCHEDULER ),
     TS_TUP( SCOREBOARD ),
     TS_TUP( MEMORY_PARTITION_UNIT ),
+    TS_TUP( LIVENESS ),
     TS_TUP( NUM_TRACE_STREAMS )
 TS_TUP_END( trace_streams_type )