Merge branch 'dev-purdue-integration' of https://github.rcac.purdue.edu/TimRogersGroup/gpgpu-sim_distribution into dev-purdue-integration

3 files changed, 15 insertions, 6 deletions

diff --git a/configs/Pascal-P100-HBM/gpgpusim.config b/configs/Pascal-P100-HBM/gpgpusim.config
index 533a865..bd868ac 100644
--- a/configs/Pascal-P100-HBM/gpgpusim.config
+++ b/configs/Pascal-P100-HBM/gpgpusim.config
@@ -56,7 +56,7 @@
 # Pascal GP100 has 64KB Shared memory
 -gpgpu_cache:dl1  S:64:128:6,L:L:f:N:H,A:256:8,16:0,32
 -gpgpu_shmem_size 65536
--gmem_skip_L1D 1
+-gmem_skip_L1D 0
 -icnt_flit_size 40
 -gpgpu_n_cluster_ejection_buffer_size 32
 
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index 0602e20..8191f62 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -962,7 +962,7 @@ void baseline_cache::send_read_request(new_addr_type addr, new_addr_type block_a
         m_mshrs.add(mshr_addr,mf);
         m_extra_mf_fields[mf] = extra_mf_fields(mshr_addr,mf->get_addr(),cache_index, mf->get_data_size(), m_config);
         mf->set_data_size( m_config.get_atom_sz() );
-        mf->set_addr( block_addr );
+        mf->set_addr( mshr_addr );
         m_miss_queue.push_back(mf);
         mf->set_status(m_miss_queue_status,time);
         if(!wa)
@@ -1432,7 +1432,7 @@ data_cache::process_tag_probe( bool wr,
             access_status = (this->*m_wr_hit)( addr,
                                       cache_index,
                                       mf, time, events, probe_status );
-        }else if ( probe_status != RESERVATION_FAIL ) {
+        }else if ( (probe_status != RESERVATION_FAIL) || (probe_status == RESERVATION_FAIL && m_config.m_write_alloc_policy == NO_WRITE_ALLOCATE) ) {
             access_status = (this->*m_wr_miss)( addr,
                                        cache_index,
                                        mf, time, events, probe_status );
diff --git a/src/gpgpu-sim/gpu-cache.h b/src/gpgpu-sim/gpu-cache.h
index 0d07878..76b07a2 100644
--- a/src/gpgpu-sim/gpu-cache.h
+++ b/src/gpgpu-sim/gpu-cache.h
@@ -39,7 +39,7 @@
 #include <iostream>
 
 enum cache_block_state {
-    INVALID,
+    INVALID=0,
     RESERVED,
     VALID,
     MODIFIED
@@ -125,6 +125,7 @@ struct cache_block_t {
     virtual unsigned get_modified_size() = 0;
     virtual void set_m_readable(bool readable, mem_access_sector_mask_t sector_mask)=0;
     virtual bool is_readable(mem_access_sector_mask_t sector_mask)=0;
+    virtual void print_status()=0;
     virtual ~cache_block_t() {}
 
 
@@ -144,7 +145,7 @@ struct line_cache_block: public cache_block_t  {
 	        m_set_modified_on_fill = false;
 	        m_readable = true;
 	    }
-	    void allocate( new_addr_type tag, new_addr_type block_addr, unsigned time, mem_access_sector_mask_t sector_mask )
+	    void allocate( new_addr_type tag, new_addr_type block_addr, unsigned time, mem_access_sector_mask_t sector_mask)
 	    {
 	        m_tag=tag;
 	        m_block_addr=block_addr;
@@ -220,6 +221,9 @@ struct line_cache_block: public cache_block_t  {
 		virtual bool is_readable(mem_access_sector_mask_t sector_mask) {
 			return m_readable;
 		}
+		virtual void print_status() {
+			 printf("m_block_addr is %u, status = %u\n", m_block_addr, m_status);
+		}
 
 
 private:
@@ -401,6 +405,11 @@ struct sector_cache_block : public cache_block_t {
 		return modified * SECTOR_SIZE;
 	}
 
+    virtual void print_status() {
+    	 printf("m_block_addr is %u, status = %u %u %u %u\n", m_block_addr, m_status[0], m_status[1], m_status[2], m_status[3]);
+    }
+
+
 private:
     unsigned m_sector_alloc_time[SECTOR_CHUNCK_SIZE];
     unsigned m_last_sector_access_time[SECTOR_CHUNCK_SIZE];
@@ -1030,7 +1039,7 @@ protected:
     std::string m_name;
     cache_config &m_config;
     tag_array*  m_tag_array;
-    mshr_table m_mshrs;
+	mshr_table m_mshrs;
     std::list<mem_fetch*> m_miss_queue;
     enum mem_fetch_status m_miss_queue_status;
     mem_fetch_interface *m_memport;