Adding configs for device limits and fixed other misc bugs

5 files changed, 32 insertions, 17 deletions

diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index 820f5fd..cebdb25 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -88,8 +88,6 @@ void checkpoint::load_global_mem(class memory_space *temp_mem, char * f1name)
       fclose ( fp2 );
 }
 
-
-
 void checkpoint::store_global_mem(class memory_space * mem, char *fname, char * format)
 {
 
@@ -195,6 +193,8 @@ gpgpu_t::gpgpu_t( const gpgpu_functional_sim_config &config )
    // initialize texture mappings to empty
    m_NameToTextureInfo.clear();
    m_NameToCudaArray.clear();
+   m_TextureRefToName.clear();
+   m_NameToAttribute.clear();
 
    if(m_function_model_config.get_ptx_inst_debug_to_file() != 0) 
       ptx_inst_debug_file = fopen(m_function_model_config.get_ptx_inst_debug_file(), "w");
@@ -694,8 +694,7 @@ unsigned kernel_info_t::m_next_uid = 1;
 /*A snapshot of the texture mappings needs to be stored in the kernel's info as 
 kernels should use the texture bindings seen at the time of launch and textures
  can be bound/unbound asynchronously with respect to streams. */
-
-kernel_info_t::kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map<std::string, const struct cudaArray*> nameToCudaArray, std::map<std::string, const struct textureInfo*> nameToTexureInfo)   
+kernel_info_t::kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map<std::string, const struct cudaArray*> nameToCudaArray, std::map<std::string, const struct textureInfo*> nameToTextureInfo)   
 {
     m_kernel_entry=entry;
     m_grid_dim=gridDim;
@@ -716,7 +715,7 @@ kernel_info_t::kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *
 
     volta_cache_config_set=false;
     m_NameToCudaArray = nameToCudaArray;
-    m_NameToTexureInfo = nameToTexureInfo;
+    m_NameToTextureInfo = nameToTextureInfo;
 }
 
 kernel_info_t::~kernel_info_t()
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 201b21f..e8716ab 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -212,7 +212,7 @@ public:
 //      m_num_cores_running=0;
 //      m_param_mem=NULL;
 //   }
-   kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map<std::string, const struct cudaArray*> nameToCudaArray, std::map<std::string, const struct textureInfo*> nameToTexureInfo);
+   kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map<std::string, const struct cudaArray*> nameToCudaArray, std::map<std::string, const struct textureInfo*> nameToTextureInfo);
    ~kernel_info_t();
 
    void inc_running() { m_num_cores_running++; }
@@ -287,8 +287,8 @@ public:
 
    const struct textureInfo* get_texinfo( const std::string &texname ) const
    {
-      std::map<std::string, const struct textureInfo*>::const_iterator t=m_NameToTexureInfo.find(texname);
-      assert(t != m_NameToTexureInfo.end());
+      std::map<std::string, const struct textureInfo*>::const_iterator t=m_NameToTextureInfo.find(texname);
+      assert(t != m_NameToTextureInfo.end());
       return t->second;
    }
 
@@ -303,7 +303,7 @@ private:
    
    //These maps contain the snapshot of the texture mappings at kernel launch
    std::map<std::string, const struct cudaArray*> m_NameToCudaArray;
-   std::map<std::string, const struct textureInfo*> m_NameToTexureInfo;
+   std::map<std::string, const struct textureInfo*> m_NameToTextureInfo;
 
    dim3 m_grid_dim;
    dim3 m_block_dim;
@@ -601,8 +601,8 @@ public:
 
     const struct textureInfo* get_texinfo( const std::string &texname ) const
     {
-        std::map<std::string, const struct textureInfo*>::const_iterator t=m_NameToTexureInfo.find(texname);
-        assert(t != m_NameToTexureInfo.end());
+        std::map<std::string, const struct textureInfo*>::const_iterator t=m_NameToTextureInfo.find(texname);
+        assert(t != m_NameToTextureInfo.end());
         return t->second;
     }
 
@@ -617,9 +617,8 @@ public:
     FILE* get_ptx_inst_debug_file() { return ptx_inst_debug_file; }
     
     //  These maps return the current texture mappings for the GPU at any given time.
-    
     std::map<std::string, const struct cudaArray*> getNameArrayMapping() {return m_NameToCudaArray;}
-    std::map<std::string, const struct textureInfo*> getNameInfoMapping() {return m_NameToTexureInfo;}
+    std::map<std::string, const struct textureInfo*> getNameInfoMapping() {return m_NameToTextureInfo;}
 
 protected:
     const gpgpu_functional_sim_config &m_function_model_config;
@@ -634,7 +633,7 @@ protected:
     std::map<std::string, std::set<const struct textureReference*> > m_NameToTextureRef;
     std::map<const struct textureReference*, std::string> m_TextureRefToName;
     std::map<std::string, const struct cudaArray*> m_NameToCudaArray;
-    std::map<std::string, const struct textureInfo*> m_NameToTexureInfo;
+    std::map<std::string, const struct textureInfo*> m_NameToTextureInfo;
     std::map<std::string, const struct textureReferenceAttr*> m_NameToAttribute;
 };
 
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 9d59411..f7bb9cc 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -218,7 +218,7 @@ void gpgpu_t::gpgpu_ptx_sim_bindTextureToArray(const struct textureReference* te
    texInfo->Ty_numbits = intLOGB2(Ty);
    texInfo->texel_size = texel_size;
    texInfo->texel_size_numbits = intLOGB2(texel_size);
-   m_NameToTexureInfo[texname] = texInfo;
+   m_NameToTextureInfo[texname] = texInfo;
 }
 
 void gpgpu_t::gpgpu_ptx_sim_unbindTexture(const struct textureReference* texref)
@@ -226,7 +226,7 @@ void gpgpu_t::gpgpu_ptx_sim_unbindTexture(const struct textureReference* texref)
    //assumes bind-use-unbind-bind-use-unbind pattern
    std::string texname = gpgpu_ptx_sim_findNamefromTexture(texref);
    m_NameToCudaArray.erase(texname);
-   m_NameToTexureInfo.erase(texname);
+   m_NameToTextureInfo.erase(texname);
 }
 
 unsigned g_assemble_code_next_pc=0; 
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index ec570bf..2a862a4 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -532,6 +532,14 @@ void gpgpu_sim_config::reg_options(option_parser_t opp)
    option_parser_register(opp, "-visualizer_zlevel", OPT_INT32,
                           &g_visualizer_zlevel, "Compression level of the visualizer output log (0=no comp, 9=highest)",
                           "6");
+   option_parser_register(opp, "-gpgpu_stack_size_limit", OPT_INT32, &stack_size_limit,
+                          "GPU thread stack size", "1024" );
+   option_parser_register(opp, "-gpgpu_heap_size_limit", OPT_INT32, &heap_size_limit,
+                          "GPU malloc heap size ", "8388608" );
+   option_parser_register(opp, "-gpgpu_runtime_sync_depth_limit", OPT_INT32, &runtime_sync_depth_limit,
+                          "GPU device runtime synchronize depth", "2" );
+   option_parser_register(opp, "-gpgpu_runtime_pending_launch_count_limit", OPT_INT32, &runtime_pending_launch_count_limit,
+                          "GPU device runtime pending launch count", "2048" );
     option_parser_register(opp, "-trace_enabled", OPT_BOOL, 
                           &Trace::enabled, "Turn on traces",
                           "0");
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 6ce5524..11ffe49 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -337,6 +337,11 @@ public:
     unsigned get_max_concurrent_kernel() const { return max_concurrent_kernel; }
     unsigned checkpoint_option;
 
+    size_t stack_limit() const {return stack_size_limit; }
+    size_t heap_limit() const {return heap_size_limit; }
+    size_t sync_depth_limit() const {return runtime_sync_depth_limit; }
+    size_t pending_launch_count_limit() const {return runtime_pending_launch_count_limit;}
+
 private:
     void init_clock_domains(void ); 
 
@@ -377,7 +382,11 @@ private:
     int gpu_stat_sample_freq;
     int gpu_runtime_stat_flag;
 
-
+    // Device Limits
+    size_t stack_size_limit;
+    size_t heap_size_limit;
+    size_t runtime_sync_depth_limit;
+    size_t runtime_pending_launch_count_limit;	
 
     unsigned long long liveness_message_freq;