adding 1st level data cache

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 8153]

1 files changed, 96 insertions, 79 deletions

diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index c746da7..594e683 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -70,25 +70,24 @@
 
 tag_array::~tag_array() 
 {
-   delete m_lines;
+    delete m_lines;
 }
 
 tag_array::tag_array( const cache_config &config, int core_id, int type_id ) 
-    : m_config(config)
+: m_config(config)
 {
-   assert( m_config.m_write_policy == READ_ONLY );
-   m_lines = new cache_block_t[ config.get_num_lines()];
+    assert( m_config.m_write_policy == READ_ONLY );
+    m_lines = new cache_block_t[ config.get_num_lines()];
 
-   // initialize snapshot counters for visualizer
-   m_prev_snapshot_access = 0;
-   m_prev_snapshot_miss = 0;
-   m_prev_snapshot_pending_hit = 0;
-   m_core_id = core_id; 
-   m_type_id = type_id;
+    // initialize snapshot counters for visualizer
+    m_prev_snapshot_access = 0;
+    m_prev_snapshot_miss = 0;
+    m_prev_snapshot_pending_hit = 0;
+    m_core_id = core_id; 
+    m_type_id = type_id;
 }
 
-enum cache_request_status tag_array::probe( new_addr_type addr, unsigned &idx ) const
-{
+enum cache_request_status tag_array::probe( new_addr_type addr, unsigned &idx ) const {
     assert( m_config.m_write_policy == READ_ONLY );
     unsigned set_index = m_config.set_index(addr);
     new_addr_type tag = m_config.tag(addr);
@@ -98,57 +97,70 @@ enum cache_request_status tag_array::probe( new_addr_type addr, unsigned &idx )
     unsigned valid_timestamp = (unsigned)-1;
 
     bool all_reserved = true;
-    
+
     // check for hit or pending hit
     for (unsigned way=0; way<m_config.m_assoc; way++) {
-	unsigned index = set_index*m_config.m_assoc+way;
-	cache_block_t *line = &m_lines[index];
-	if (line->m_tag == tag) {
-	    if( line->m_status == RESERVED ) {
-		idx = index;
-		return HIT_RESERVED;
-	    } else if( line->m_status == VALID ) {
-		idx = index;
-		return HIT;
-	    } else {
-		assert( line->m_status == INVALID );
-	    }
-	} 
-	if(line->m_status != RESERVED) {
-	    all_reserved = false;
-	    if(line->m_status == INVALID) {
-		invalid_line = index;
-	    } else {
-		// valid line : keep track of most appropriate replacement candidate
-		if( m_config.m_replacement_policy == LRU ) {
-		    if( line->m_last_access_time < valid_timestamp ) {
-			valid_timestamp = line->m_last_access_time;
-			valid_line = index;
-		    }
-		} else if( m_config.m_replacement_policy == FIFO ) {
-		    if( line->m_alloc_time < valid_timestamp ) {
-			valid_timestamp = line->m_alloc_time;
-			valid_line = index;
-		    }
-		}
-	    }
-	}
-   }
-   if( all_reserved ) {
-       assert( m_config.m_alloc_policy == ON_MISS ); 
-       return RESERVATION_FAIL; // miss and not enough space in cache to allocate on miss
-   }
+        unsigned index = set_index*m_config.m_assoc+way;
+        cache_block_t *line = &m_lines[index];
+        if (line->m_tag == tag) {
+            if ( line->m_status == RESERVED ) {
+                idx = index;
+                return HIT_RESERVED;
+            } else if ( line->m_status == VALID ) {
+                idx = index;
+                return HIT;
+            } else if ( line->m_status == MODIFIED ) {
+                idx = index;
+                return HIT;
+            } else {
+                assert( line->m_status == INVALID );
+            }
+        }
+        if (line->m_status != RESERVED) {
+            all_reserved = false;
+            if (line->m_status == INVALID) {
+                invalid_line = index;
+            } else {
+                // valid line : keep track of most appropriate replacement candidate
+                if ( m_config.m_replacement_policy == LRU ) {
+                    if ( line->m_last_access_time < valid_timestamp ) {
+                        valid_timestamp = line->m_last_access_time;
+                        valid_line = index;
+                    }
+                } else if ( m_config.m_replacement_policy == FIFO ) {
+                    if ( line->m_alloc_time < valid_timestamp ) {
+                        valid_timestamp = line->m_alloc_time;
+                        valid_line = index;
+                    }
+                }
+            }
+        }
+    }
+    if ( all_reserved ) {
+        assert( m_config.m_alloc_policy == ON_MISS ); 
+        return RESERVATION_FAIL; // miss and not enough space in cache to allocate on miss
+    }
 
-   if( invalid_line != (unsigned)-1 ) {
-       idx = invalid_line;
-   } else if( valid_line != (unsigned)-1) {
-       idx = valid_line;
-   } else abort(); // if an unreserved block exists, it is either invalid or replaceable 
+    if ( invalid_line != (unsigned)-1 ) {
+        idx = invalid_line;
+    } else if ( valid_line != (unsigned)-1) {
+        idx = valid_line;
+    } else abort(); // if an unreserved block exists, it is either invalid or replaceable 
 
-   return MISS;
+    return MISS;
 }
 
-enum cache_request_status tag_array::access( new_addr_type addr, unsigned time, unsigned &idx ) {
+enum cache_request_status tag_array::access( new_addr_type addr, unsigned time, unsigned &idx )
+{
+    bool wb=false;
+    cache_block_t evicted;
+    enum cache_request_status result = access(addr,time,idx,wb,evicted);
+    assert(!wb);
+    return result;
+}
+
+enum cache_request_status tag_array::access( new_addr_type addr, unsigned time, unsigned &idx, bool &wb, cache_block_t &evicted ) 
+{
     m_access++;
     shader_cache_access_log(m_core_id, m_type_id, 0); // log accesses to cache
     enum cache_request_status status = probe(addr,idx);
@@ -161,8 +173,13 @@ enum cache_request_status tag_array::access( new_addr_type addr, unsigned time,
     case MISS:
         m_miss++;
         shader_cache_access_log(m_core_id, m_type_id, 1); // log cache misses
-        if ( m_config.m_alloc_policy == ON_MISS )
+        if ( m_config.m_alloc_policy == ON_MISS ) {
+            if( m_lines[idx].m_status == MODIFIED ) {
+                wb = true;
+                evicted = m_lines[idx];
+            }
             m_lines[idx].allocate( m_config.tag(addr), m_config.block_addr(addr), time );
+        }
         break;
     case RESERVATION_FAIL:
         m_miss++;
@@ -190,37 +207,37 @@ void tag_array::fill( unsigned index, unsigned time )
 
 void tag_array::flush() 
 {
-   for (unsigned i=0; i < m_config.get_num_lines(); i++) 
-      m_lines[i].m_status = INVALID;
+    for (unsigned i=0; i < m_config.get_num_lines(); i++)
+        m_lines[i].m_status = INVALID;
 }
 
 float tag_array::windowed_miss_rate( bool minus_pending_hit ) const
 {
-   unsigned n_access    = m_access - m_prev_snapshot_access;
-   unsigned n_miss      = m_miss - m_prev_snapshot_miss;
-   unsigned n_pending_hit = m_pending_hit - m_prev_snapshot_pending_hit;
-   
-   if (minus_pending_hit) 
-      n_miss -= n_pending_hit;
-   float missrate = 0.0f;
-   if (n_access != 0) 
-      missrate = (float) n_miss / n_access;
-   return missrate;
+    unsigned n_access    = m_access - m_prev_snapshot_access;
+    unsigned n_miss      = m_miss - m_prev_snapshot_miss;
+    unsigned n_pending_hit = m_pending_hit - m_prev_snapshot_pending_hit;
+
+    if (minus_pending_hit)
+        n_miss -= n_pending_hit;
+    float missrate = 0.0f;
+    if (n_access != 0)
+        missrate = (float) n_miss / n_access;
+    return missrate;
 }
 
 void tag_array::new_window()
 {
-   m_prev_snapshot_access = m_access;
-   m_prev_snapshot_miss = m_miss;
-   m_prev_snapshot_pending_hit = m_pending_hit;
+    m_prev_snapshot_access = m_access;
+    m_prev_snapshot_miss = m_miss;
+    m_prev_snapshot_pending_hit = m_pending_hit;
 }
 
 void tag_array::print( FILE *stream, unsigned &total_access, unsigned &total_misses ) const
 {
-   m_config.print(stream);
-   fprintf( stream, "\t\tAccess = %d, Miss = %d (%.3g), -MgHts = %d (%.3g)\n", 
-            m_access, m_miss, (float) m_miss / m_access, 
-            m_miss - m_pending_hit, (float) (m_miss - m_pending_hit) / m_access);
-   total_misses+=m_miss;
-   total_access+=m_access;
+    m_config.print(stream);
+    fprintf( stream, "\t\tAccess = %d, Miss = %d (%.3g), -MgHts = %d (%.3g)\n", 
+             m_access, m_miss, (float) m_miss / m_access, 
+             m_miss - m_pending_hit, (float) (m_miss - m_pending_hit) / m_access);
+    total_misses+=m_miss;
+    total_access+=m_access;
 }