Merge pull request #2396 from verilog-to-routing/placement_move_primitive

Placement Refactoring

Author: vaughnbetz

libs/libarchfpga/src/physical_types.h

@@ -924,10 +924,10 @@ struct t_logical_block_type {
     std::vector<t_physical_tile_type_ptr> equivalent_tiles; ///>List of physical tiles at which one could
                                                             ///>place this type of netlist block.
 
-    std::unordered_map<int, t_pb_graph_pin*> pin_logical_num_to_pb_pin_mapping;                   /* pin_logical_num_to_pb_pin_mapping[pin logical number] -> pb_graph_pin ptr} */
-    std::unordered_map<const t_pb_graph_pin*, int> primitive_pb_pin_to_logical_class_num_mapping; /* primitive_pb_pin_to_logical_class_num_mapping[pb_graph_pin ptr] -> class logical number */
-    std::vector<t_class> primitive_logical_class_inf;                                             /* primitive_logical_class_inf[class_logical_number] -> class */
-    std::unordered_map<const t_pb_graph_node*, t_class_range> pb_graph_node_class_range;
+    std::unordered_map<int, t_pb_graph_pin*> pin_logical_num_to_pb_pin_mapping;                    /* pin_logical_num_to_pb_pin_mapping[pin logical number] -> pb_graph_pin ptr} */
+    std::unordered_map<const t_pb_graph_pin*, int> primitive_pb_pin_to_logical_class_num_mapping;  /* primitive_pb_pin_to_logical_class_num_mapping[pb_graph_pin ptr] -> class logical number */
+    std::vector<t_class> primitive_logical_class_inf;                                              /* primitive_logical_class_inf[class_logical_number] -> class */
+    std::unordered_map<const t_pb_graph_node*, t_class_range> primitive_pb_graph_node_class_range; /* primitive_pb_graph_node_class_range[primitive_pb_graph_node ptr] -> class range for that primitive*/
 
     // Is this t_logical_block_type empty?
     bool is_empty() const;
@@ -1239,6 +1239,12 @@ class t_pb_graph_node {
 
     int placement_index;
 
+    /*
+     * There is a root-level pb_graph_node assigned to each logical type. Each logical type can contain multiple primitives.
+     * If this pb_graph_node is associated with a primitive, a unique number is assigned to it within the logical block level.
+     */
+    int primitive_num = OPEN;
+
     /* Contains a collection of mode indices that cannot be used as they produce conflicts during VPR packing stage
      *
      * Illegal modes do arise when children of a graph_node do have inconsistent `edge_modes` with respect to

libs/libarchfpga/src/physical_types_util.cpp

@@ -965,7 +965,7 @@ t_class_range get_pb_graph_node_class_physical_range(t_physical_tile_type_ptr /*
                                                      const t_pb_graph_node* pb_graph_node) {
     VTR_ASSERT(pb_graph_node->is_primitive());
 
-    t_class_range class_range = logical_block->pb_graph_node_class_range.at(pb_graph_node);
+    t_class_range class_range = logical_block->primitive_pb_graph_node_class_range.at(pb_graph_node);
     int logical_block_class_offset = sub_tile->primitive_class_range[sub_tile_relative_cap].at(logical_block).low;
 
     class_range.low += logical_block_class_offset;

libs/libvtrutil/src/vtr_vec_id_set.h

@@ -2,6 +2,7 @@
 #define VTR_SET_H
 
 #include <vector>
+#include <algorithm>
 
 namespace vtr {
 

vpr/src/base/vpr_types.h

@@ -1919,4 +1919,10 @@ void free_pack_molecules(t_pack_molecule* list_of_pack_molecules);
  */
 void free_cluster_placement_stats(t_cluster_placement_stats* cluster_placement_stats);
 
+struct pair_hash {
+    std::size_t operator()(const std::pair<ClusterBlockId, ClusterBlockId>& p) const noexcept {
+        return std::hash<ClusterBlockId>()(p.first) ^ (std::hash<ClusterBlockId>()(p.second) << 1);
+    }
+};
+
 #endif

vpr/src/pack/pb_type_graph.cpp

@@ -50,7 +50,8 @@ static void alloc_and_load_pb_graph(t_pb_graph_node* pb_graph_node,
                                     const int index,
                                     const int flat_index,
                                     bool load_power_structures,
-                                    int& pin_count_in_cluster);
+                                    int& pin_count_in_cluster,
+                                    int& primitive_num);
 
 static void alloc_and_load_pb_graph_pin_sinks(t_pb_graph_node* pb_graph_node);
 
@@ -153,13 +154,15 @@ void alloc_and_load_all_pb_graphs(bool load_power_structures, bool is_flat) {
         if (type.pb_type) {
             type.pb_graph_head = new t_pb_graph_node();
             int pin_count_in_cluster = 0;
+            int primitive_num = 0;
             alloc_and_load_pb_graph(type.pb_graph_head,
                                     nullptr,
                                     type.pb_type,
                                     0,
                                     0,
                                     load_power_structures,
-                                    pin_count_in_cluster);
+                                    pin_count_in_cluster,
+                                    primitive_num);
             type.pb_graph_head->total_pb_pins = pin_count_in_cluster;
             load_pin_classes_in_pb_graph_head(type.pb_graph_head);
             if (is_flat) {
@@ -233,7 +236,8 @@ static void alloc_and_load_pb_graph(t_pb_graph_node* pb_graph_node,
                                     const int index,
                                     const int flat_index,
                                     bool load_power_structures,
-                                    int& pin_count_in_cluster) {
+                                    int& pin_count_in_cluster,
+                                    int& primitive_num) {
     int i, j, k, i_input, i_output, i_clockport;
 
     pb_graph_node->placement_index = index;
@@ -350,6 +354,11 @@ static void alloc_and_load_pb_graph(t_pb_graph_node* pb_graph_node,
         pb_graph_node->pb_node_power->transistor_cnt_pb_children = 0.;
     }
 
+    if (pb_graph_node->is_primitive()) {
+        pb_graph_node->primitive_num = primitive_num;
+        primitive_num++;
+    }
+
     /* Allocate and load child nodes for each mode and create interconnect in each mode */
 
     pb_graph_node->child_pb_graph_nodes = (t_pb_graph_node***)vtr::calloc(pb_type->num_modes, sizeof(t_pb_graph_node**));
@@ -368,7 +377,8 @@ static void alloc_and_load_pb_graph(t_pb_graph_node* pb_graph_node,
                                         k,
                                         child_flat_index,
                                         load_power_structures,
-                                        pin_count_in_cluster);
+                                        pin_count_in_cluster,
+                                        primitive_num);
             }
         }
     }
@@ -384,6 +394,7 @@ static void alloc_and_load_pb_graph(t_pb_graph_node* pb_graph_node,
                                          load_power_structures);
     }
 
+
     // update the total number of primitives of that type
     if (pb_graph_node->is_primitive()) {
         int total_count = 1;
@@ -549,8 +560,8 @@ static void add_primitive_logical_classes(t_logical_block_type* logical_block) {
             }
             num_added_classes += add_port_logical_classes(logical_block, pb_graph_pins, num_ports, num_pins);
         }
-        logical_block->pb_graph_node_class_range.insert(std::make_pair(pb_graph_node, t_class_range(first_class_num,
-                                                                                                    first_class_num + num_added_classes - 1)));
+        logical_block->primitive_pb_graph_node_class_range.insert(std::make_pair(pb_graph_node, t_class_range(first_class_num,
+                                                                                                              first_class_num + num_added_classes - 1)));
     }
 }
 

vpr/src/pack/re_cluster.cpp

@@ -200,9 +200,9 @@ bool swap_two_molecules(t_pack_molecule* molecule_1,
     }
 
     t_pb* clb_pb_1 = cluster_ctx.clb_nlist.block_pb(clb_1);
-    std::string clb_pb_1_name = (std::string)clb_pb_1->name;
+    std::string clb_pb_1_name = static_cast<std::string>(clb_pb_1->name);
     t_pb* clb_pb_2 = cluster_ctx.clb_nlist.block_pb(clb_2);
-    std::string clb_pb_2_name = (std::string)clb_pb_2->name;
+    std::string clb_pb_2_name = static_cast<std::string>(clb_pb_2->name);
 
     //remove the molecule from its current cluster
     remove_mol_from_cluster(molecule_1, molecule_1_size, clb_1, clb_1_atoms, false, old_1_router_data);

vpr/src/pack/re_cluster_util.cpp

@@ -1,5 +1,4 @@
 #include "re_cluster_util.h"
-
 #include "clustered_netlist_utils.h"
 #include "cluster_util.h"
 #include "cluster_router.h"
@@ -8,7 +7,6 @@
 #include "initial_placement.h"
 #include "read_netlist.h"
 
-
 // The name suffix of the new block (if exists)
 // This suffix is useful in preventing duplicate high-level cluster block names
 const char* name_suffix = "_m";

vpr/src/place/move_transactions.cpp

@@ -128,6 +128,7 @@ void clear_move_blocks(t_pl_blocks_to_be_moved& blocks_affected) {
 
     //For run-time, we just reset num_moved_blocks to zero, but do not free the blocks_affected
     //array to avoid memory allocation
+
     blocks_affected.num_moved_blocks = 0;
 
     blocks_affected.affected_pins.clear();

vpr/src/place/move_transactions.h

@@ -9,12 +9,17 @@
  * old_loc: the location the block is moved from                *
  * new_loc: the location the block is moved to                  */
 struct t_pl_moved_block {
+    t_pl_moved_block() = default;
+    t_pl_moved_block(ClusterBlockId block_num_, const t_pl_loc& old_loc_, const t_pl_loc& new_loc_)
+        : block_num(block_num_)
+        , old_loc(old_loc_)
+        , new_loc(new_loc_) {}
     ClusterBlockId block_num;
     t_pl_loc old_loc;
     t_pl_loc new_loc;
 };
 
-/* Stores the list of blocks to be moved in a swap during       *
+/* Stores the list of cluster blocks to be moved in a swap during       *
  * placement.                                                   *
  * Store the information on the blocks to be moved in a swap during     *
  * placement, in the form of array of structs instead of struct with    *
@@ -29,7 +34,7 @@ struct t_pl_moved_block {
  *                incrementally invalidate parts of the timing  *
  *                graph.                                        */
 struct t_pl_blocks_to_be_moved {
-    t_pl_blocks_to_be_moved(size_t max_blocks)
+    explicit t_pl_blocks_to_be_moved(size_t max_blocks)
         : moved_blocks(max_blocks) {}
 
     int num_moved_blocks = 0;

vpr/src/place/move_utils.h

@@ -95,6 +95,14 @@ void report_aborted_moves();
 
 e_create_move create_move(t_pl_blocks_to_be_moved& blocks_affected, ClusterBlockId b_from, t_pl_loc to);
 
+/**
+ * @brief Find the blocks that will be affected by a move of b_from to to_loc
+ * @param blocks_affected Loaded by this routine and returned via reference; it lists the blocks etc. moved
+ * @param b_from Id of the cluster-level block to be moved
+ * @param to Where b_from will be moved to
+ * @return e_block_move_result ABORT if either of the the moving blocks are already stored, or either of the blocks are fixed, to location is not
+ * compatible, etc. INVERT if the "from" block is a single block and the "to" block is a macro. VALID otherwise.
+ */
 e_block_move_result find_affected_blocks(t_pl_blocks_to_be_moved& blocks_affected, ClusterBlockId b_from, t_pl_loc to);
 
 e_block_move_result record_single_block_swap(t_pl_blocks_to_be_moved& blocks_affected, ClusterBlockId b_from, t_pl_loc to);
@@ -109,6 +117,12 @@ e_block_move_result record_macro_move(t_pl_blocks_to_be_moved& blocks_affected,
 e_block_move_result identify_macro_self_swap_affected_macros(std::vector<int>& macros, const int imacro, t_pl_offset swap_offset);
 e_block_move_result record_macro_self_swaps(t_pl_blocks_to_be_moved& blocks_affected, const int imacro, t_pl_offset swap_offset);
 
+/**
+ * @brief Check whether the "to" location is legal for the given "blk"
+ * @param blk
+ * @param to
+ * @return True if this would be a legal move, false otherwise
+ */
 bool is_legal_swap_to_location(ClusterBlockId blk, t_pl_loc to);
 
 std::set<t_pl_loc> determine_locations_emptied_by_move(t_pl_blocks_to_be_moved& blocks_affected);