bug fixing for lb router. Add physical mode to default node expanding settings

openfpga/src/repack/lb_router.cpp

@@ -109,7 +109,7 @@ bool LbRouter::route_has_conflict(const LbRRGraph& lb_rr_graph, t_trace* rt) con
   t_mode* cur_mode = nullptr;
   for (unsigned int i = 0; i < rt->next_nodes.size(); i++) {
     std::vector<LbRREdgeId> edges = lb_rr_graph.find_edge(rt->current_node, rt->next_nodes[i].current_node);
-    VTR_ASSERT(0 == edges.size());
+    VTR_ASSERT(1 == edges.size());
     t_mode* new_mode = lb_rr_graph.edge_mode(edges[0]);
     if (cur_mode != nullptr && cur_mode != new_mode) {
       return true;
@@ -165,6 +165,34 @@ void LbRouter::add_net_atom_pins(const NetId& net, const AtomPinId& src_pin, con
   lb_net_atom_pins_[net].insert(lb_net_atom_pins_[net].begin(), src_pin);
 }
 
+void LbRouter::set_physical_pb_modes(const LbRRGraph& lb_rr_graph,
+                                     const VprDeviceAnnotation& device_annotation) {
+  /* Go through each node in the routing resource graph
+   * Find the physical mode of each pb_graph_pin that is binded to the node
+   * For input pins, the physical mode is a mode of its parent pb_type
+   * For output pins, the physical mode is a mode of the parent pb_type of its parent
+   */
+  for (const LbRRNodeId& node : lb_rr_graph.nodes()) {
+    t_pb_graph_pin* pb_pin = lb_rr_graph.node_pb_graph_pin(node);
+    if (nullptr == pb_pin) {
+      routing_status_[node].mode = nullptr;
+    } else {
+      if (IN_PORT == pb_pin->port->type) {
+        routing_status_[node].mode = device_annotation.physical_mode(pb_pin->parent_node->pb_type);
+      } else {
+        VTR_ASSERT(OUT_PORT == pb_pin->port->type);
+        /* For top-level pb_graph node, the physical mode is nullptr */
+        if (true == pb_pin->parent_node->is_root()) {
+          routing_status_[node].mode = nullptr;
+        } else { 
+          routing_status_[node].mode = device_annotation.physical_mode(pb_pin->parent_node->parent_pb_graph_node->pb_type);
+          /* TODO: need to think about how to handle INOUT ports !!! */
+        }
+      }
+    }
+  }
+}
+
 bool LbRouter::try_route(const LbRRGraph& lb_rr_graph,
                          const AtomNetlist& atom_nlist,
                          const int& verbosity) {
@@ -263,8 +291,10 @@ bool LbRouter::try_route(const LbRRGraph& lb_rr_graph,
     } else {
       --inet;
       VTR_LOGV(verbosity < 3,
-               "Net '%s' is impossible to route within proposed %s cluster\n",
+               "Net %lu '%s' is impossible to route within proposed %s cluster\n",
-               atom_nlist.net_name(lb_net_atom_net_ids_[NetId(inet)]).c_str(), lb_type_->name);
+               inet,
+               atom_nlist.net_name(lb_net_atom_net_ids_[NetId(inet)]).c_str(),
+               lb_type_->name);
       VTR_LOGV(verbosity < 3,
                "\tNet source pin '%s'\n",
                lb_rr_graph.node_pb_graph_pin(lb_net_terminals_[NetId(inet)][0])->to_string().c_str()); 
@@ -573,8 +603,11 @@ void LbRouter::expand_edges(const LbRRGraph& lb_rr_graph,
     t_mode* next_mode = routing_status_[enode.node_index].mode;
     /* Assume first mode if a mode hasn't been forced. */
     if (nullptr == next_mode) {
+      /* If the node is mapped to a nullptr pb_graph_pin, this is a special SINK. Use nullptr mode */
+      if (nullptr == lb_rr_graph.node_pb_graph_pin(enode.node_index)) {
+        next_mode = nullptr;
+      } else if (true == is_primitive_pb_type(lb_rr_graph.node_pb_graph_pin(enode.node_index)->parent_node->pb_type)) {
       /* For primitive node, we give nullptr as default */
-      if (true == is_primitive_pb_type(lb_rr_graph.node_pb_graph_pin(enode.node_index)->parent_node->pb_type)) {
         next_mode = nullptr;
       } else {
         next_mode = &(lb_rr_graph.node_pb_graph_pin(enode.node_index)->parent_node->pb_type->modes[0]);
@@ -614,8 +647,14 @@ void LbRouter::expand_node(const LbRRGraph& lb_rr_graph,
   float cur_cost = exp_node.cost;
   t_mode* mode = routing_status_[cur_node].mode;
   if (nullptr == mode) {
+    if (nullptr == lb_rr_graph.node_pb_graph_pin(cur_node)) {
+      mode = nullptr;
+    } else if (true == is_primitive_pb_type(lb_rr_graph.node_pb_graph_pin(cur_node)->parent_node->pb_type)) {
+      mode = nullptr;
+    } else {
       mode = &(lb_rr_graph.node_pb_graph_pin(cur_node)->parent_node->pb_type->modes[0]);
     }
+  }
 
   expand_edges(lb_rr_graph, mode, cur_node, cur_cost, net_fanout);
 }

openfpga/src/repack/lb_router.h

@@ -15,6 +15,7 @@
 #include "physical_types.h"
 #include "vpr_context.h"
 
+#include "vpr_device_annotation.h"
 #include "lb_rr_graph.h"
 
 /********************************************************************
@@ -184,6 +185,17 @@ class LbRouter {
     void add_net_atom_net_id(const NetId& net, const AtomNetId& atom_net);
     void add_net_atom_pins(const NetId& net, const AtomPinId& src_pin, const std::vector<AtomPinId>& terminal_pins);
 
+    /* TODO: Initialize all the modes in routing status with the mode set in pb
+     * This is function used for general purpose packing
+     */
+
+    /* Set all the modes in routing status with the physical mode defined in device annotation
+     * This method is used only in repacking for physical logical blocks
+     * Do NOT use it during the general purpose packing
+     */ 
+    void set_physical_pb_modes(const LbRRGraph& lb_rr_graph,
+                               const VprDeviceAnnotation& device_annotation);
+
     /**
      * Perform routing algorithm on a given logical tile routing resource graph
      * Note: the lb_rr_graph must be the same as you initilized the router!!!

openfpga/src/repack/repack.cpp

@@ -269,6 +269,11 @@ void repack_cluster(const DeviceContext& device_ctx,
                      clustering_ctx, const_cast<const VprClusteringAnnotation&>(clustering_annotation),
                      block_id, verbose);
 
+  /* Initialize the modes to expand routing trees with the physical modes in device annotation
+   * This is a must-do before running the routeri in the purpose of repacking!!!
+   */
+  lb_router.set_physical_pb_modes(lb_rr_graph, device_annotation); 
+
   /* Run the router */
   bool route_success = lb_router.try_route(lb_rr_graph, atom_ctx.nlist, verbose);
 

openfpga/test_openfpga_arch/k6_frac_N10_40nm_openfpga.xml

@@ -226,7 +226,7 @@
       <port name="in" physical_mode_port="in[0:5]"/>
       <port name="out" physical_mode_port="lut6_out"/>
     </pb_type>
-    <pb_type name="clb.fle[n1_lut6].ble6.ff" physical_pb_type_name="clb.fle[physical].fabric.ff" physical_pb_type_index_factor="2" physical_pb_type_index_offset="1"/>
+    <pb_type name="clb.fle[n1_lut6].ble6.ff" physical_pb_type_name="clb.fle[physical].fabric.ff" physical_pb_type_index_factor="2" physical_pb_type_index_offset="0"/>
     <!-- End physical pb_type binding in complex block IO -->
   </pb_type_annotations>
 </openfpga_architecture>

openfpga/test_vpr_arch/k6_frac_N10_40nm.xml

@@ -237,8 +237,9 @@
               </pb_type>
               <interconnect>
                 <direct name="direct1" input="frac_logic.in" output="frac_lut6.in"/>
-                <direct name="direct2" input="frac_lut6.lut5_out[0]" output="frac_logic.out[0]"/>
+                <direct name="direct2" input="frac_lut6.lut5_out[1]" output="frac_logic.out[1]"/>
-                <mux name="mux1" input="frac_lut6.lut6_out frac_lut6.lut5_out[1]" output="frac_logic.out[1]"/>
+                <!-- Xifan Tang: I use out[0] because the output of lut6 in lut6 mode is wired to the out[0] -->
+                <mux name="mux1" input="frac_lut6.lut6_out frac_lut6.lut5_out[0]" output="frac_logic.out[0]"/>
               </interconnect>
             </pb_type>
             <!-- Define flip-flop -->
@@ -251,6 +252,7 @@
             </pb_type>
             <interconnect>
               <direct name="direct1" input="fabric.in" output="frac_logic.in"/>
+              <direct name="direct2" input="frac_logic.out[1:0]" output="ff[1:0].D"/>
               <complete name="direct3" input="fabric.clk" output="ff[1:0].clk"/>
               <mux name="mux1" input="ff[0].Q frac_logic.out[0]" output="fabric.out[0]">
                 <!-- LUT to output is faster than FF to output on a Stratix IV -->