diff --git a/CMakeLists.txt b/CMakeLists.txt
index 95ca8e085..114a86eb6 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -10,6 +10,7 @@ project(GPGPU-Sim
 # Specify the C++ standard
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED True)
+string(REPLACE "-DNDEBUG" "" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
 
 # GPGPU-Sim build option
 option(GPGPUSIM_ENABLE_TRACE "Whether to enable GPGPU-Sim debug tracing" ON)
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index fb61e1c89..30e718795 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -1577,7 +1577,6 @@ void gpgpu_sim::gpu_print_stat(unsigned long long streamID) {
     kernel_cycle +=
         kernel.second.elapsed() + gpgpu_ctx->device_runtime->g_kernel_launch_latency;
   }
-  assert(0 && "fix kernel_cycle");
 
   std::string kernel_info_str = executed_kernel_info_string();
   fprintf(statfout, "%s", kernel_info_str.c_str());
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 7d68e7b7d..00352fcaa 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -1977,11 +1977,11 @@ void shader_core_ctx::warp_inst_complete(const warp_inst_t &inst) {
 
   m_stats->m_num_sim_winsn[m_sid]++;
   m_gpu->gpu_sim_insn += inst.active_count();
-  unsigned stream_id = inst.get_streamID();
+  unsigned long long stream_id = inst.get_streamID();
   if (m_gpu->gpu_sim_insn_per_stream.find(stream_id) ==
       m_gpu->gpu_sim_insn_per_stream.end()) {
     m_gpu->gpu_sim_insn_per_stream.insert(
-        std::pair<unsigned, unsigned>(stream_id, 0));
+        std::pair<unsigned long long, unsigned>(stream_id, 0));
   }
   m_gpu->gpu_sim_insn_per_stream.at(stream_id) += inst.active_count();
   shader_inst += inst.active_count();