Adding collective modeling functions in model-net (currently only...

Adding collective modeling functions in model-net (currently only dragonfly/torus models support it)

codes/model-net-method.h

@@ -32,6 +32,8 @@ struct model_net_method
     int (*mn_get_msg_sz)();
     void (*mn_report_stats)();
     tw_lpid (*model_net_method_find_local_device)(tw_lp *sender);
+    void (*mn_collective_call)(char* category, int message_size, int remote_event_size, const void* remote_event, tw_lp* sender);
+    void (*mn_collective_call_rc)(int message_size, tw_lp* sender);    
 };
 
 extern struct model_net_method * method_array[];

codes/model-net.h

@@ -153,6 +153,22 @@ int model_net_setup(char* net_name, uint64_t packet_size, const void* net_params
 
 /* utility function to get the modelnet ID post-setup */
 int model_net_get_id(char *net_name);
+
+/* This event does a collective operation call for model-net */
+void model_net_event_collective(
+    int net_id,
+    char* category,
+    int message_size,
+    int remote_event_size,
+    const void* remote_event,
+    tw_lp *sender);
+
+/* reverse event of the collective operation call */
+void model_net_event_collective_rc(
+        int net_id,
+        int message_size,
+        tw_lp *sender);
+
 /* allocate and transmit a new event that will pass through model_net to 
  * arrive at its destination:
  *

codes/net/dragonfly.h

@@ -44,15 +44,24 @@ struct terminal_message
    int output_chan;
     int is_pull;
     uint64_t pull_size;
-   
+
+   /* for reverse computation */   
    tw_stime saved_available_time;
    tw_stime saved_credit_time;
+   tw_stime saved_collective_init_time;  
 
    int intm_group_id;
    short chunk_id;
    uint64_t packet_size;
    int remote_event_size_bytes;
    int local_event_size_bytes;
+
+   /* for reverse computation of a node's fan in*/
+   int saved_fan_nodes;
+   tw_lpid sender_svr;
+
+  /* LP ID of the sending node, has to be a network node in the dragonfly */
+   tw_lpid sender_node;
 };
 
 #endif /* end of include guard: DRAGONFLY_H */

codes/net/torus.h

@@ -17,6 +17,9 @@ enum nodes_event_t
   ARRIVAL, 
   SEND,
   CREDIT,
+  T_COLLECTIVE_INIT,
+  T_COLLECTIVE_FAN_IN,
+  T_COLLECTIVE_FAN_OUT  
 };
 
 struct nodes_message
@@ -28,6 +31,9 @@ struct nodes_message
   /* for reverse event computation*/
   tw_stime saved_available_time;
 
+  /* message saved collective time */
+  tw_stime saved_collective_init_time;
+  
   /* packet ID */
   unsigned long long packet_ID;
   /* event type of the message */
@@ -44,8 +50,12 @@ struct nodes_message
   tw_lpid final_dest_gid;
   /* destination torus node of the message */
   tw_lpid dest_lp;
-  /* LP ID of the sender, comes from codes, can be a server or any other I/O LP type */
-  tw_lpid sender_lp;
+  /* LP ID of the sender, comes from codes, can be a server or any other I/O LP type. Should not change
+     during network operations. */
+  tw_lpid sender_svr;
+
+  /* LP ID of the sending node, has to be a network node in the torus */
+  tw_lpid sender_node;
 
   /* number of hops traversed by the packet */
   int my_N_hop;
@@ -57,6 +67,10 @@ struct nodes_message
   int next_stop;
   /* size of the torus packet */
   uint64_t packet_size;
+
+ /* for reverse computation of a node's fan in*/
+  int saved_fan_nodes;
+
   /* chunk id of the flit (distinguishes flits) */
   short chunk_id;
 

src/models/networks/model-net/loggp.c

@@ -67,6 +67,12 @@ static int loggp_get_msg_sz(void);
 /* Returns the loggp magic number */
 static int loggp_get_magic();
 
+/* collective network calls */
+static void loggp_collective();
+
+/* collective network calls-- rc */
+static void loggp_collective_rc();
+
 /* allocate a new event that will pass through loggp to arriave at its
  * destination:
  *
@@ -118,6 +124,8 @@ struct model_net_method loggp_method =
     .mn_get_msg_sz = loggp_get_msg_sz,
     .mn_report_stats = loggp_report_stats,
     .model_net_method_find_local_device = loggp_find_local_device,
+    .mn_collective_call = loggp_collective,
+    .mn_collective_call_rc = loggp_collective_rc
 };
 
 static void loggp_init(
@@ -187,6 +195,20 @@ static void loggp_report_stats()
    /* TODO: Do we have some loggp statistics to report like we have for torus and dragonfly? */
    return;
 }
+
+/* collective network calls */
+static void loggp_collective()
+{
+/* collectives not supported */
+    return;
+}
+
+static void loggp_collective_rc()
+{
+/* collectives not supported */
+   return;
+}
+
 static void loggp_init(
     loggp_state * ns,
     tw_lp * lp)

src/models/networks/model-net/model-net.c

@@ -522,6 +522,28 @@ uint64_t model_net_get_packet_size(int net_id)
   return method_array[net_id]->packet_size; // TODO: where to set the packet size?
 }
 
+/* This event does a collective operation call for model-net */
+void model_net_event_collective(int net_id, char* category, int message_size, int remote_event_size, const void* remote_event, tw_lp* sender)
+{
+  if(net_id < 0 || net_id > MAX_NETS)
+     {
+       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
+       exit(-1);
+     }
+  return method_array[net_id]->mn_collective_call(category, message_size, remote_event_size, remote_event, sender);
+}
+
+/* reverse event of the collective operation call */
+void model_net_event_collective_rc(int net_id, int message_size, tw_lp* sender)
+{
+  if(net_id < 0 || net_id > MAX_NETS)
+     {
+       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
+       exit(-1);
+     }
+  return method_array[net_id]->mn_collective_call_rc(message_size, sender);
+}
+
 /* returns lp type for modelnet */
 const tw_lptype* model_net_get_lp_type(int net_id)
 {

src/models/networks/model-net/simplenet-upd.c

@@ -62,6 +62,12 @@ static int sn_get_magic();
 
 static void print_msg(sn_message *m);
 
+/* collective network calls */
+static void simple_net_collective();
+
+/* collective network calls-- rc */
+static void simple_net_collective_rc();
+
 /* allocate a new event that will pass through simplenet to arriave at its
  * destination:
  *
@@ -111,6 +117,8 @@ struct model_net_method simplenet_method =
     .mn_get_msg_sz = sn_get_msg_sz,
     .mn_report_stats = sn_report_stats,
     .model_net_method_find_local_device = sn_find_local_device,
+    .mn_collective_call = simple_net_collective,	
+    .mn_collective_call_rc = simple_net_collective_rc
 };
 
 static void sn_init(
@@ -174,6 +182,19 @@ static int sn_get_msg_sz(void)
 {
     return(sizeof(sn_message));
 }
+/* collective network calls */
+static void simple_net_collective()
+{
+/* collectives not supported */
+    return;
+}
+
+/* collective network call -- rc*/
+static void simple_net_collective_rc()
+{
+/* collectives not supported */
+   return;
+}
 
 /* lets caller specify model parameters to use */
 static void sn_set_params(double net_startup_ns, double net_bw_mbs)

src/models/networks/model-net/simplewan.c

@@ -96,6 +96,12 @@ static double sw_get_startup(int from_id, int to_id);
 static category_idles* sw_get_category_idles(
         char * category, category_idles *idles);
 
+/* collective network calls */
+static void simple_wan_collective();
+
+/* collective network calls-- rc */
+static void simple_wan_collective_rc();
+
 /* allocate a new event that will pass through simplewan to arriave at its
  * destination:
  *
@@ -144,7 +150,9 @@ struct model_net_method simplewan_method =
     .mn_get_lp_type = sw_get_lp_type,
     .mn_get_msg_sz = sw_get_msg_sz,
     .mn_report_stats = sw_report_stats,
-    .model_net_method_find_local_device = sw_find_local_device,
+      .model_net_method_find_local_device = sw_find_local_device,
+    .mn_collective_call = simple_wan_collective,
+    .mn_collective_call_rc = simple_wan_collective_rc  
 };
 
 static void sw_init(
@@ -195,6 +203,19 @@ static void handle_msg_start_event(
     sw_message * m,
     tw_lp * lp);
 
+/* collective network calls */
+static void simple_wan_collective()
+{
+/* collectives not supported */
+    return;
+}
+
+static void simple_wan_collective_rc()
+{
+/* collectives not supported */
+   return;
+}
+
 /* returns pointer to LP information for simplewan module */
 static const tw_lptype* sw_get_lp_type()
 {

tests/Makefile.subdir

 
 check_PROGRAMS += tests/modelnet-test tests/modelnet-p2p-bw \
-				  tests/concurrent-msg-recv tests/modelnet-simplewan-test
+				  tests/concurrent-msg-recv tests/modelnet-simplewan-test \		
+				  tests/modelnet-test-collective
 
 TESTS += tests/modelnet-test.sh \
 	 tests/modelnet-test-torus.sh \
@@ -31,3 +32,8 @@ tests_concurrent_msg_recv_LDFLAGS = ${CODES_BASE_LDFLAGS}
 tests_concurrent_msg_recv_LDADD = $(testlib) ${CODES_BASE_LIBS} 
 tests_concurrent_msg_recv_CFLAGS = ${CODES_BASE_CFLAGS} 
 tests_concurrent_msg_recv_SOURCES = tests/concurrent-msg-recv.c
+
+tests_modelnet_test_collective_LDADD = $(testlib) ${CODES_BASE_LIBS}
+tests_modelnet_test_collective_CFLAGS = ${CODES_BASE_CFLAGS}
+tests_modelnet_test_collective_SOURCES = tests/modelnet-test-collective.c
+