Merge pull request #28 from abdallm/dev-purdue-integration

Dev purdue integration

1 files changed, 29 insertions, 0 deletions

diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index bb8e4b6..8a84970 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -2760,6 +2760,35 @@ unsigned int shader_core_config::max_cta( const kernel_info_t &k ) const
        abort();
     }
 
+    if(adpative_volta_cache_config && !k.volta_cache_config_set) {
+    	//For Volta, we assign the remaining shared memory to L1 cache
+    	//For more info, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x
+    	unsigned total_shmed = kernel_info->smem * result;
+    	assert(total_shmed >=0 && total_shmed <= gpgpu_shmem_size);
+    	assert(gpgpu_shmem_size == 98304); //Volta has 96 KB shared
+    	assert(m_L1D_config.get_nset() == 4);  //Volta L1 has four sets
+    	if(total_shmed < gpgpu_shmem_size){
+    		if(total_shmed == 0)
+    			m_L1D_config.set_assoc(256);  //L1 is 128KB ans shd=0
+    		else if(total_shmed > 0 && total_shmed <= 8192)
+    			m_L1D_config.set_assoc(240);  //L1 is 120KB ans shd=8KB
+    		else if(total_shmed > 8192 && total_shmed <= 16384)
+    		    m_L1D_config.set_assoc(224);  //L1 is 112KB ans shd=16KB
+    		else if(total_shmed > 16384 && total_shmed <= 32768)
+    		    m_L1D_config.set_assoc(192);  //L1 is 96KB ans shd=32KB
+    		else if(total_shmed > 32768 && total_shmed <= 65536)
+    		    m_L1D_config.set_assoc(128);	//L1 is 64KB ans shd=64KB
+    		else if(total_shmed > 65536 && total_shmed <= gpgpu_shmem_size)
+    		     m_L1D_config.set_assoc(64); //L1 is 32KB and shd=96KB
+    		else
+    			assert(0);
+
+    		 printf ("GPGPU-Sim: Reconfigure L1 cache in Volta Archi to %uKB\n", m_L1D_config.get_total_size_inKB());
+    	}
+
+    	k.volta_cache_config_set = true;
+    }
+
     return result;
 }