make sure L1 cache is flushed at a configuration change between kernels, even if flushing L1 cache between kernels option is not set

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 15834]

6 files changed, 30 insertions, 19 deletions

diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 5d74651..d1acff6 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -47,6 +47,15 @@ enum _memory_space_t {
    instruction_space
 };
 
+
+enum FuncCache
+{
+  FuncCachePreferNone = 0,
+  FuncCachePreferShared = 1,
+  FuncCachePreferL1 = 2
+};
+
+
 #ifdef __cplusplus
 
 #include <string.h>
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index 3d8d911..28fcd5c 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -48,7 +48,7 @@ const char * cache_request_status_str(enum cache_request_status status)
 }
 
 void l2_cache_config::init(linear_to_raw_address_translation *address_mapping){
-	cache_config::init(m_config_string);
+	cache_config::init(m_config_string,FuncCachePreferNone);
 	m_address_mapping = address_mapping;
 }
 
diff --git a/src/gpgpu-sim/gpu-cache.h b/src/gpgpu-sim/gpu-cache.h
index dd56d9f..59e56bc 100644
--- a/src/gpgpu-sim/gpu-cache.h
+++ b/src/gpgpu-sim/gpu-cache.h
@@ -134,8 +134,9 @@ public:
         m_config_stringPrefL1 = NULL;
         m_config_stringPrefShared = NULL;
     }
-    void init(char * config)
+    void init(char * config, FuncCache status)
     {
+    	cache_status= status;
         assert( config );
         char rp, wp, ap, mshr_type, wap;
 
@@ -235,10 +236,11 @@ public:
     {
         return addr & ~(m_line_sz-1);
     }
-
+    FuncCache get_cache_status() {return cache_status;}
     char *m_config_string;
     char *m_config_stringPrefL1;
     char *m_config_stringPrefShared;
+    FuncCache cache_status;
 
 protected:
     void exit_parse_error()
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 2778077..39216f9 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -811,20 +811,27 @@ void gpgpu_sim::set_cache_config(std::string kernel_name)
 
 void gpgpu_sim::change_cache_config(FuncCache cache_config)
 {
+	if(cache_config != m_shader_config->m_L1D_config.get_cache_status()){
+		printf("FLUSH L1 Cache at configuration change between kernels\n");
+		for (unsigned i=0;i<m_shader_config->n_simt_clusters;i++) {
+			m_cluster[i]->cache_flush();
+	    }
+	}
+
 	switch(cache_config){
 	case FuncCachePreferNone:
-		m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_string);
+		m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_string, FuncCachePreferNone);
 		m_shader_config->gpgpu_shmem_size=m_shader_config->gpgpu_shmem_sizeDefault;
 		break;
 	case FuncCachePreferL1:
 		if((m_shader_config->m_L1D_config.m_config_stringPrefL1 == NULL) || (m_shader_config->gpgpu_shmem_sizePrefL1 == (unsigned)-1))
 		{
 			printf("WARNING: missing Preferred L1 configuration\n");
-			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_string);
+			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_string, FuncCachePreferNone);
 			m_shader_config->gpgpu_shmem_size=m_shader_config->gpgpu_shmem_sizeDefault;
 
 		}else{
-			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_stringPrefL1);
+			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_stringPrefL1, FuncCachePreferL1);
 			m_shader_config->gpgpu_shmem_size=m_shader_config->gpgpu_shmem_sizePrefL1;
 		}
 		break;
@@ -832,10 +839,10 @@ void gpgpu_sim::change_cache_config(FuncCache cache_config)
 		if((m_shader_config->m_L1D_config.m_config_stringPrefShared == NULL) || (m_shader_config->gpgpu_shmem_sizePrefShared == (unsigned)-1))
 		{
 			printf("WARNING: missing Preferred L1 configuration\n");
-			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_string);
+			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_string, FuncCachePreferNone);
 			m_shader_config->gpgpu_shmem_size=m_shader_config->gpgpu_shmem_sizeDefault;
 		}else{
-			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_stringPrefShared);
+			m_shader_config->m_L1D_config.init(m_shader_config->m_L1D_config.m_config_stringPrefShared, FuncCachePreferShared);
 			m_shader_config->gpgpu_shmem_size=m_shader_config->gpgpu_shmem_sizePrefShared;
 		}
 		break;
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 8616fd4..cadddac 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -71,13 +71,6 @@ enum dram_ctrl_t {
    DRAM_FRFCFS=1
 };
 
-enum FuncCache
-{
-  FuncCachePreferNone = 0,
-  FuncCachePreferShared = 1,
-  FuncCachePreferL1 = 2
-};
-
 
 
 struct power_config {
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 4f5342f..f2d38a0 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1208,10 +1208,10 @@ struct shader_core_config : public core_config
         assert( !(n_thread_per_shader % warp_size) );
         max_sfu_latency = 512;
         max_sp_latency = 32;
-        m_L1I_config.init(m_L1I_config.m_config_string);
-        m_L1T_config.init(m_L1T_config.m_config_string);
-        m_L1C_config.init(m_L1C_config.m_config_string);
-        m_L1D_config.init(m_L1D_config.m_config_string);
+        m_L1I_config.init(m_L1I_config.m_config_string,FuncCachePreferNone);
+        m_L1T_config.init(m_L1T_config.m_config_string,FuncCachePreferNone);
+        m_L1C_config.init(m_L1C_config.m_config_string,FuncCachePreferNone);
+        m_L1D_config.init(m_L1D_config.m_config_string,FuncCachePreferNone);
         gpgpu_cache_texl1_linesize = m_L1T_config.get_line_sz();
         gpgpu_cache_constl1_linesize = m_L1C_config.get_line_sz();
         m_valid = true;