Merge branch 'master' into ueqri-enhanced-heap-for-connection-router

ueqri · web-flow · commit d7ea38a9823f · 2024-10-18T14:33:03.000-04:00
diff --git a/doc/src/vpr/command_line_usage.rst b/doc/src/vpr/command_line_usage.rst
@@ -200,12 +200,14 @@ General Options
 
 .. option:: --device <string>
 
-    Specifies which device layout/floorplan to use from the architecture file.
+    Specifies which device layout/floorplan to use from the architecture file.  Valid values are:
 
-    ``auto`` uses the smallest device satisfying the circuit's resource requirements.
-    Other values are assumed to be the names of device layouts defined in the :ref:`arch_grid_layout` section of the architecture file.
+    * ``auto`` VPR uses the smallest device satisfying the circuit's resource requirements.  This option will use the ``<auto_layout>`` tag if it is present in the architecture file in order to construct the smallest FPGA that has sufficient resources to fit the design. If the ``<auto_layout>`` tag is not present, the ``auto`` option chooses the smallest device amongst all the architecture file's ``<fixed_layout>`` specifications into which the design can be packed.
+    * Any string matching ``name`` attribute of a device layout defined with a ``<fixed_layout>`` tag in the :ref:`arch_grid_layout` section of the architecture file.
 
-    .. note:: If the architecture contains both ``<auto_layout>`` and ``<fixed_layout>`` specifications, specifying an ``auto`` device will use the ``<auto_layout>``.
+    If the value specified is neither ``auto`` nor matches the ``name`` attribute value of a ``<fixed_layout>`` tag, VPR issues an error.
+       
+    .. note:: If the only layout in the architecture file is a single device specified using ``<fixed_layout>``, it is recommended to always specify the ``--device`` option; this prevents the value ``--device auto`` from interfering with operations supported only for ``<fixed_layout>`` grids.
 
     **Default:** ``auto``
 
diff --git a/libs/libarchfpga/src/physical_types.h b/libs/libarchfpga/src/physical_types.h
@@ -1382,7 +1382,7 @@ class t_pb_graph_pin {
     float tco_max = std::numeric_limits<float>::quiet_NaN(); /* For sequential logic elements the maximum clock to output time */
     t_pb_graph_pin* associated_clock_pin = nullptr;          /* For sequentail elements, the associated clock */
 
-    /* This member is used when flat-routing and has_choking_spot are enabled.
+    /* This member is used when flat-routing and router_opt_choke_points are enabled.
      * It is used to identify choke points.
      * This is only valid for IPINs, and it only contain the pins that are reachable to the pin by a forwarding path.
      * It doesn't take into account feed-back connection.
diff --git a/vpr/src/base/SetupVPR.cpp b/vpr/src/base/SetupVPR.cpp
@@ -318,7 +318,7 @@ void SetupVPR(const t_options* options,
         vtr::ScopedStartFinishTimer timer("Allocate intra-cluster resources");
         // The following two functions should be called when the data structured related to t_pb_graph_node, t_pb_type,
         // and t_pb_graph_edge are initialized
-        alloc_and_load_intra_cluster_resources(routerOpts->has_choking_spot);
+        alloc_and_load_intra_cluster_resources(routerOpts->has_choke_point);
         add_intra_tile_switches();
     }
 
@@ -510,7 +510,7 @@ static void SetupRouterOpts(const t_options& Options, t_router_opts* RouterOpts)
     RouterOpts->max_logged_overused_rr_nodes = Options.max_logged_overused_rr_nodes;
     RouterOpts->generate_rr_node_overuse_report = Options.generate_rr_node_overuse_report;
     RouterOpts->flat_routing = Options.flat_routing;
-    RouterOpts->has_choking_spot = Options.has_choking_spot;
+    RouterOpts->has_choke_point = Options.router_opt_choke_points;
     RouterOpts->custom_3d_sb_fanin_fanout = Options.custom_3d_sb_fanin_fanout;
     RouterOpts->with_timing_analysis = Options.timing_analysis;
 }
diff --git a/vpr/src/base/ShowSetup.cpp b/vpr/src/base/ShowSetup.cpp
@@ -255,11 +255,11 @@ static void ShowRouterOpts(const t_router_opts& RouterOpts) {
         VTR_LOG("false\n");
     }
 
-    VTR_LOG("RouterOpts.has_choking_spot: ");
-    if (RouterOpts.has_choking_spot) {
-        VTR_LOG("true\n");
+    VTR_LOG("RouterOpts.choke_points: ");
+    if (RouterOpts.has_choke_point) {
+        VTR_LOG("on\n");
     } else {
-        VTR_LOG("false\n");
+        VTR_LOG("off\n");
     }
 
     VTR_ASSERT(GLOBAL == RouterOpts.route_type || DETAILED == RouterOpts.route_type);
diff --git a/vpr/src/base/place_and_route.cpp b/vpr/src/base/place_and_route.cpp
@@ -398,7 +398,7 @@ int binary_search_place_and_route(const Netlist<>& placement_net_list,
 
     init_route_structs(router_net_list,
                        router_opts.bb_factor,
-                       router_opts.has_choking_spot,
+                       router_opts.has_choke_point,
                        is_flat);
 
     restore_routing(best_routing,
diff --git a/vpr/src/base/read_options.cpp b/vpr/src/base/read_options.cpp
@@ -2487,13 +2487,13 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
         .default_value("off")
         .show_in(argparse::ShowIn::HELP_ONLY);
 
-    route_grp.add_argument(args.has_choking_spot, "--has_choking_spot")
+    route_grp.add_argument<bool, ParseOnOff>(args.router_opt_choke_points, "--router_opt_choke_points")
         .help(
             ""
-            "Some FPGA architectures, due to the lack of full connectivity inside the cluster, may have"
-            " a choking spot inside the cluster. Thus, if routing doesn't converge, enabling this option may"
-            " help it.")
-        .default_value("false")
+            "Some FPGA architectures with limited fan-out options within a cluster (e.g. fracturable LUTs with shared pins) do" 
+            " not converge well in routing unless these fan-out choke points are discovered and optimized for during net routing." 
+            " This option helps router convergence for such architectures.")
+        .default_value("on")
         .show_in(argparse::ShowIn::HELP_ONLY);
 
 
diff --git a/vpr/src/base/read_options.h b/vpr/src/base/read_options.h
@@ -218,7 +218,7 @@ struct t_options {
     argparse::ArgValue<int> reorder_rr_graph_nodes_threshold;
     argparse::ArgValue<int> reorder_rr_graph_nodes_seed;
     argparse::ArgValue<bool> flat_routing;
-    argparse::ArgValue<bool> has_choking_spot;
+    argparse::ArgValue<bool> router_opt_choke_points;
     argparse::ArgValue<int> route_verbosity;
     argparse::ArgValue<int> custom_3d_sb_fanin_fanout;
 
diff --git a/vpr/src/base/read_route.cpp b/vpr/src/base/read_route.cpp
@@ -107,7 +107,7 @@ bool read_route(const char* route_file, const t_router_opts& router_opts, bool v
     const Netlist<>& router_net_list = (flat_router) ? (const Netlist<>&)g_vpr_ctx.atom().nlist : (const Netlist<>&)g_vpr_ctx.clustering().clb_nlist;
     init_route_structs(router_net_list,
                        router_opts.bb_factor,
-                       router_opts.has_choking_spot,
+                       router_opts.has_choke_point,
                        flat_router);
 
     /*Check dimensions*/
diff --git a/vpr/src/base/vpr_types.h b/vpr/src/base/vpr_types.h
@@ -1339,7 +1339,7 @@ struct t_router_opts {
     bool generate_rr_node_overuse_report;
 
     bool flat_routing;
-    bool has_choking_spot;
+    bool has_choke_point;
 
     int custom_3d_sb_fanin_fanout = 1;
 
diff --git a/vpr/src/place/net_cost_handler.cpp b/vpr/src/place/net_cost_handler.cpp
@@ -31,8 +31,9 @@
 #include "placer_state.h"
 #include "move_utils.h"
 #include "place_timing_update.h"
-#include "noc_place_utils.h"
 #include "vtr_math.h"
+#include "vtr_ndmatrix.h"
+#include "vtr_ndoffsetmatrix.h"
 
 #include <array>
 
@@ -53,9 +54,6 @@ constexpr std::array<float, MAX_FANOUT_CROSSING_COUNT> cross_count = {1.0000, 1.
                                                                       2.5610, 2.5864, 2.6117, 2.6371, 2.6625, 2.6887, 2.7148, 2.7410,
                                                                       2.7671, 2.7933};
 
-
-
-
 /**
  * @brief If the moving pin is of type type SINK, update bb_pin_sink_count_new which stores the number of sink pins on each layer of "net_id"
  * @param pin_old_loc Old location of the moving pin
@@ -229,6 +227,70 @@ void NetCostHandler::alloc_and_load_chan_w_factors_for_place_cost_(float place_c
             chany_place_cost_fac_[high][low] = pow((double)chany_place_cost_fac_[high][low], (double)place_cost_exp);
         }
     }
+    
+    if (device_ctx.grid.get_num_layers() > 1) {
+        alloc_and_load_for_fast_vertical_cost_update_(place_cost_exp);
+    }
+}
+
+void NetCostHandler::alloc_and_load_for_fast_vertical_cost_update_(float place_cost_exp) {
+    const auto& device_ctx = g_vpr_ctx.device();
+    const auto& rr_graph = device_ctx.rr_graph;
+    
+    const size_t grid_height = device_ctx.grid.height();
+    const size_t grid_width = device_ctx.grid.width();
+
+
+    chanz_place_cost_fac_ = vtr::NdMatrix<float, 4>({grid_width, grid_height, grid_width, grid_height}, 0.);
+
+    vtr::NdMatrix<float, 2> tile_num_inter_die_conn({grid_width, grid_height}, 0.);                           
+
+    for (const auto& src_rr_node : rr_graph.nodes()) {
+        for (const auto& rr_edge_idx : rr_graph.configurable_edges(src_rr_node)) {
+            const auto& sink_rr_node = rr_graph.edge_sink_node(src_rr_node, rr_edge_idx);
+            if (rr_graph.node_layer(src_rr_node) != rr_graph.node_layer(sink_rr_node)) {
+                // We assume that the nodes driving the inter-layer connection or being driven by it
+                // are not streched across multiple tiles
+                int src_x = rr_graph.node_xhigh(src_rr_node);
+                int src_y = rr_graph.node_yhigh(src_rr_node);
+                VTR_ASSERT(rr_graph.node_xlow(src_rr_node) == src_x && rr_graph.node_ylow(src_rr_node) == src_y);
+
+                tile_num_inter_die_conn[src_x][src_y]++;
+            }
+        }
+
+        for (const auto& rr_edge_idx : rr_graph.non_configurable_edges(src_rr_node)) {
+            const auto& sink_rr_node = rr_graph.edge_sink_node(src_rr_node, rr_edge_idx);
+            if (rr_graph.node_layer(src_rr_node) != rr_graph.node_layer(sink_rr_node)) {
+                int src_x = rr_graph.node_xhigh(src_rr_node);
+                VTR_ASSERT(rr_graph.node_xlow(src_rr_node) == src_x && rr_graph.node_xlow(src_rr_node) == src_x);
+                int src_y = rr_graph.node_yhigh(src_rr_node);
+                VTR_ASSERT(rr_graph.node_ylow(src_rr_node) == src_y && rr_graph.node_ylow(src_rr_node) == src_y);
+                tile_num_inter_die_conn[src_x][src_y]++;
+            }
+        }
+    }
+
+    for (int x_high = 0; x_high < (int)device_ctx.grid.width(); x_high++) {
+        for (int y_high = 0; y_high < (int)device_ctx.grid.height(); y_high++) {
+            for (int x_low = 0; x_low <= x_high; x_low++) {
+                for (int y_low = 0; y_low <= y_high; y_low++) {
+                    int num_inter_die_conn = 0;
+                    for (int x = x_low; x <= x_high; x++) {
+                        for (int y = y_low; y <= y_high; y++) {
+                            num_inter_die_conn += tile_num_inter_die_conn[x][y];
+                        }
+                    }
+                    int seen_num_tiles = (x_high - x_low + 1) * (y_high - y_low + 1);
+                    chanz_place_cost_fac_[x_high][y_high][x_low][y_low] = seen_num_tiles / static_cast<float>(num_inter_die_conn);
+
+                    chanz_place_cost_fac_[x_high][y_high][x_low][y_low] = pow(
+                        (double)chanz_place_cost_fac_[x_high][y_high][x_low][y_low],
+                        (double)place_cost_exp);
+                }
+            }
+        }
+    }
 }
 
 double NetCostHandler::comp_bb_cost(e_cost_methods method) {
@@ -1395,6 +1457,8 @@ double NetCostHandler::get_net_cube_bb_cost_(ClusterNetId net_id, bool use_ts) {
 
     const t_bb& bb = use_ts ? ts_bb_coord_new_[net_id] : placer_state_.move().bb_coords[net_id];
 
+    const bool is_multi_layer = (g_vpr_ctx.device().grid.get_num_layers() > 1);
+
     double crossing = wirelength_crossing_count(cluster_ctx.clb_nlist.net_pins(net_id).size());
 
     /* Could insert a check for xmin == xmax.  In that case, assume  *
@@ -1413,6 +1477,9 @@ double NetCostHandler::get_net_cube_bb_cost_(ClusterNetId net_id, bool use_ts) {
     double ncost;
     ncost = (bb.xmax - bb.xmin + 1) * crossing * chanx_place_cost_fac_[bb.ymax][bb.ymin - 1];
     ncost += (bb.ymax - bb.ymin + 1) * crossing * chany_place_cost_fac_[bb.xmax][bb.xmin - 1];
+    if (is_multi_layer) {
+        ncost += (bb.layer_max - bb.layer_min) * crossing * chanz_place_cost_fac_[bb.xmax][bb.ymax][bb.xmin][bb.ymin];
+    }
 
     return ncost;
 }
diff --git a/vpr/src/place/net_cost_handler.h b/vpr/src/place/net_cost_handler.h
@@ -195,6 +195,13 @@ class NetCostHandler {
      */
     vtr::NdOffsetMatrix<float, 2> chanx_place_cost_fac_; // [-1...device_ctx.grid.width()-1]
     vtr::NdOffsetMatrix<float, 2> chany_place_cost_fac_; // [-1...device_ctx.grid.height()-1]
+    /**
+      @brief This data structure functions similarly to the matrices described above 
+      but is applied to 3D connections linking different FPGA layers. It is used in the 
+      placement cost function calculation, where the height of the bounding box is divided 
+      by the average number of inter-die connections within the bounding box.
+     */
+    vtr::NdMatrix<float, 4> chanz_place_cost_fac_; // [0...device_ctx.grid.width()-1][0...device_ctx.grid.height()-1][0...device_ctx.grid.width()-1][0...device_ctx.grid.height()-1]
 
 
   private:
@@ -249,6 +256,18 @@ class NetCostHandler {
      */
     void alloc_and_load_chan_w_factors_for_place_cost_(float place_cost_exp);
 
+    /**
+    * @brief Allocates and loads the chanz_place_cost_fac array with the inverse of
+    * the average number of inter-die connections between [subhigh] and [sublow].
+    *
+    * @details This is only useful for multi-die FPGAs. The place_cost_exp factor specifies to
+    * what power the average number of inter-die connections should be take -- larger numbers make narrower channels more expensive.
+    *
+    * @param place_cost_exp It is an exponent to which you take the average number of inter-die connections;
+    * a higher value would favour areas with more inter-die connections over areas with less of those during placement (usually we use 1).
+    */
+    void alloc_and_load_for_fast_vertical_cost_update_(float place_cost_exp);
+
     /**
      * @brief Calculate the new connection delay and timing cost of all the
      * sink pins affected by moving a specific pin to a new location. Also
diff --git a/vpr/src/route/connection_router.cpp b/vpr/src/route/connection_router.cpp
@@ -764,7 +764,7 @@ void ConnectionRouter<Heap>::evaluate_timing_driven_node_costs(RTExploredNode* t
         //cost.
         cong_cost = 0.;
     }
-    if (conn_params_->has_choking_spot_ && is_flat_ && rr_graph_->node_type(to->index) == IPIN) {
+    if (conn_params_->router_opt_choke_points_ && is_flat_ && rr_graph_->node_type(to->index) == IPIN) {
         auto find_res = conn_params_->connection_choking_spots_.find(to->index);
         if (find_res != conn_params_->connection_choking_spots_.end()) {
             cong_cost = cong_cost / pow(2, (float)find_res->second);
diff --git a/vpr/src/route/route.cpp b/vpr/src/route/route.cpp
@@ -71,7 +71,7 @@ bool route(const Netlist<>& net_list,
 
     init_route_structs(net_list,
                        router_opts.bb_factor,
-                       router_opts.has_choking_spot,
+                       router_opts.has_choke_point,
                        is_flat);
 
     IntraLbPbPinLookup intra_lb_pb_pin_lookup(device_ctx.logical_block_types);
@@ -80,7 +80,7 @@ bool route(const Netlist<>& net_list,
     auto choking_spots = set_nets_choking_spots(net_list,
                                                 route_ctx.net_terminal_groups,
                                                 route_ctx.net_terminal_group_num,
-                                                router_opts.has_choking_spot,
+                                                router_opts.has_choke_point,
                                                 is_flat);
 
     //Initially, the router runs normally trying to reduce congestion while
diff --git a/vpr/src/route/route_net.tpp b/vpr/src/route/route_net.tpp
@@ -436,8 +436,8 @@ inline NetResultFlags route_sink(ConnectionRouter& router,
     bool sink_critical = (cost_params.criticality > HIGH_FANOUT_CRITICALITY_THRESHOLD);
     bool net_is_clock = route_ctx.is_clock_net[net_id] != 0;
 
-    bool has_choking_spot = ((int)choking_spots[target_pin].size() != 0) && router_opts.has_choking_spot;
-    ConnectionParameters conn_params(net_id, target_pin, has_choking_spot, choking_spots[target_pin]);
+    bool router_opt_choke_points = ((int)choking_spots[target_pin].size() != 0) && router_opts.has_choke_point;
+    ConnectionParameters conn_params(net_id, target_pin, router_opt_choke_points, choking_spots[target_pin]);
 
     //We normally route high fanout nets by only adding spatially close-by routing to the heap (reduces run-time).
     //However, if the current sink is 'critical' from a timing perspective, we put the entire route tree back onto
diff --git a/vpr/src/route/route_utils.cpp b/vpr/src/route/route_utils.cpp
@@ -391,21 +391,18 @@ vtr::vector<ParentNetId, std::vector<std::unordered_map<RRNodeId, int>>> set_net
                                                                                                                   std::vector<std::vector<int>>>& net_terminal_groups,
                                                                                                 const vtr::vector<ParentNetId,
                                                                                                                   std::vector<int>>& net_terminal_group_num,
-                                                                                                bool has_choking_spot,
+                                                                                                bool router_opt_choke_points,
                                                                                                 bool is_flat) {
     vtr::vector<ParentNetId, std::vector<std::unordered_map<RRNodeId, int>>> choking_spots(net_list.nets().size());
     for (const auto& net_id : net_list.nets()) {
         choking_spots[net_id].resize(net_list.net_pins(net_id).size());
     }
 
-    // Return if the architecture doesn't have any potential choke points
-    if (!has_choking_spot) {
+    // Return if the architecture doesn't have any potential choke points or flat router is not enabled
+    if (!router_opt_choke_points || !is_flat) {
         return choking_spots;
     }
 
-    // We only identify choke points if flat_routing is enabled.
-    VTR_ASSERT(is_flat);
-
     const auto& device_ctx = g_vpr_ctx.device();
     const auto& rr_graph = device_ctx.rr_graph;
     const auto& route_ctx = g_vpr_ctx.routing();
diff --git a/vpr/src/route/route_utils.h b/vpr/src/route/route_utils.h
diff --git a/vpr/src/route/router_stats.h b/vpr/src/route/router_stats.h
diff --git a/vtr_flow/tasks/regression_tests/vtr_reg_nightly_test7/vtr_reg_qor_large_depop_run_flat/config/config.txt b/vtr_flow/tasks/regression_tests/vtr_reg_nightly_test7/vtr_reg_qor_large_depop_run_flat/config/config.txt

Original file line number	Diff line number	Diff line change
`@@ -255,11 +255,11 @@ static void ShowRouterOpts(const t_router_opts& RouterOpts) {`
`255`	`255`	`VTR_LOG("false\n");`
`256`	`256`	`}`
`257`	`257`
`258`		`- VTR_LOG("RouterOpts.has_choking_spot: ");`
`259`		`- if (RouterOpts.has_choking_spot) {`
`260`		`- VTR_LOG("true\n");`
	`258`	`+ VTR_LOG("RouterOpts.choke_points: ");`
	`259`	`+ if (RouterOpts.has_choke_point) {`
	`260`	`+ VTR_LOG("on\n");`
`261`	`261`	`} else {`
`262`		`- VTR_LOG("false\n");`
	`262`	`+ VTR_LOG("off\n");`
`263`	`263`	`}`
`264`	`264`
`265`	`265`	`VTR_ASSERT(GLOBAL == RouterOpts.route_type \|\| DETAILED == RouterOpts.route_type);`