rr_graph: implement more robust delay normalization calculation

libs/libvtrutil/src/vtr_math.cpp

@@ -1,4 +1,5 @@
 #include <map>
+#include <algorithm>
 
 #include "vtr_assert.h"
 #include "vtr_error.h"
@@ -20,6 +21,19 @@ int ipow(int base, int exp) {
     return result;
 }
 
+float median(std::vector<float> vector) {
+    VTR_ASSERT(vector.size() > 0);
+
+    std::sort(vector.begin(), vector.end());
+
+    auto size = vector.size();
+    if (size % 2 == 0) {
+        return (float)(vector[size / 2 - 1] + vector[size / 2]) / 2;
+    }
+
+    return (float)vector[size / 2];
+}
+
 /* Performs linear interpolation or extrapolation on the set of (x,y) values specified by the xy_map.
  * A requested x value is passed in, and we return the interpolated/extrapolated y value at this requested value of x.
  * Meant for maps where both key and element are numbers.

libs/libvtrutil/src/vtr_math.h

@@ -2,6 +2,7 @@
 #define VTR_MATH_H
 
 #include <map>
+#include <vector>
 #include <cmath>
 
 #include "vtr_assert.h"
@@ -10,6 +11,9 @@ namespace vtr {
 /*********************** Math operations *************************************/
 int ipow(int base, int exp);
 
+//Returns the median of an input vector.
+float median(std::vector<float> vector);
+
 template<typename X, typename Y>
 Y linear_interpolate_or_extrapolate(const std::map<X, Y>* xy_map, X requested_x);
 

vpr/src/base/SetupVPR.cpp

@@ -330,6 +330,7 @@ static void SetupRoutingArch(const t_arch& Arch,
 static void SetupRouterOpts(const t_options& Options, t_router_opts* RouterOpts) {
     RouterOpts->do_check_rr_graph = Options.check_rr_graph;
     RouterOpts->astar_fac = Options.astar_fac;
+    RouterOpts->router_profiler_astar_fac = Options.router_profiler_astar_fac;
     RouterOpts->bb_factor = Options.bb_factor;
     RouterOpts->criticality_exp = Options.criticality_exp;
     RouterOpts->max_criticality = Options.max_criticality;

vpr/src/base/ShowSetup.cpp

@@ -270,6 +270,7 @@ static void ShowRouterOpts(const t_router_opts& RouterOpts) {
 
         if (TIMING_DRIVEN == RouterOpts.router_algorithm) {
             VTR_LOG("RouterOpts.astar_fac: %f\n", RouterOpts.astar_fac);
+            VTR_LOG("RouterOpts.router_profiler_astar_fac: %f\n", RouterOpts.router_profiler_astar_fac);
             VTR_LOG("RouterOpts.criticality_exp: %f\n", RouterOpts.criticality_exp);
             VTR_LOG("RouterOpts.max_criticality: %f\n", RouterOpts.max_criticality);
             VTR_LOG("RouterOpts.init_wirelength_abort_threshold: %f\n", RouterOpts.init_wirelength_abort_threshold);
@@ -412,6 +413,7 @@ static void ShowRouterOpts(const t_router_opts& RouterOpts) {
         VTR_LOG("RouterOpts.exit_after_first_routing_iteration: %s\n", RouterOpts.exit_after_first_routing_iteration ? "true" : "false");
         if (TIMING_DRIVEN == RouterOpts.router_algorithm) {
             VTR_LOG("RouterOpts.astar_fac: %f\n", RouterOpts.astar_fac);
+            VTR_LOG("RouterOpts.router_profiler_astar_fac: %f\n", RouterOpts.router_profiler_astar_fac);
             VTR_LOG("RouterOpts.criticality_exp: %f\n", RouterOpts.criticality_exp);
             VTR_LOG("RouterOpts.max_criticality: %f\n", RouterOpts.max_criticality);
             VTR_LOG("RouterOpts.init_wirelength_abort_threshold: %f\n", RouterOpts.init_wirelength_abort_threshold);

vpr/src/base/echo_files.cpp

@@ -115,6 +115,7 @@ void alloc_and_load_echo_file_info() {
     setEchoFileName(E_ECHO_SBLOCK_PATTERN, "sblock_pattern.txt");
     setEchoFileName(E_ECHO_ENDPOINT_TIMING, "endpoint_timing.echo.json");
     setEchoFileName(E_ECHO_LOOKAHEAD_MAP, "lookahead_map.echo");
+    setEchoFileName(E_ECHO_RR_GRAPH_INDEXED_DATA, "rr_indexed_data.echo");
 }
 
 void free_echo_file_info() {

vpr/src/base/echo_files.h

@@ -47,6 +47,7 @@ enum e_echo_files {
     E_ECHO_SBLOCK_PATTERN,
     E_ECHO_ENDPOINT_TIMING,
     E_ECHO_LOOKAHEAD_MAP,
+    E_ECHO_RR_GRAPH_INDEXED_DATA,
 
     //Timing Graphs
     E_ECHO_PRE_PACKING_TIMING_GRAPH,

vpr/src/base/read_options.cpp

@@ -2000,7 +2000,16 @@ argparse::ArgumentParser create_arg_parser(std::string prog_name, t_options& arg
 
     route_timing_grp.add_argument(args.astar_fac, "--astar_fac")
         .help(
-            "Controls the directedness of the the timing-driven router's exploration."
+            "Controls the directedness of the timing-driven router's exploration."
+            " Values between 1 and 2 are resonable; higher values trade some quality for reduced run-time")
+        .default_value("1.2")
+        .show_in(argparse::ShowIn::HELP_ONLY);
+
+    route_timing_grp.add_argument(args.router_profiler_astar_fac, "--router_profiler_astar_fac")
+        .help(
+            "Controls the directedness of the timing-driven router's exploration"
+            " when doing router delay profiling."
+            " The router delay profiling step is currently used to calculate the place delay matrix lookup."
             " Values between 1 and 2 are resonable; higher values trade some quality for reduced run-time")
         .default_value("1.2")
         .show_in(argparse::ShowIn::HELP_ONLY);

vpr/src/base/read_options.h

@@ -158,6 +158,7 @@ struct t_options {
 
     /* Timing-driven router options only */
     argparse::ArgValue<float> astar_fac;
+    argparse::ArgValue<float> router_profiler_astar_fac;
     argparse::ArgValue<float> max_criticality;
     argparse::ArgValue<float> criticality_exp;
     argparse::ArgValue<float> router_init_wirelength_abort_threshold;

vpr/src/base/vpr_types.h

@@ -1190,6 +1190,7 @@ struct t_router_opts {
     enum e_router_algorithm router_algorithm;
     enum e_base_cost_type base_cost_type;
     float astar_fac;
+    float router_profiler_astar_fac;
     float max_criticality;
     float criticality_exp;
     float init_wirelength_abort_threshold;

vpr/src/route/router_delay_profiling.cpp

@@ -54,7 +54,7 @@ bool RouterDelayProfiler::calculate_delay(int source_node, int sink_node, const
 
     t_conn_cost_params cost_params;
     cost_params.criticality = 1.;
-    cost_params.astar_fac = router_opts.astar_fac;
+    cost_params.astar_fac = router_opts.router_profiler_astar_fac;
     cost_params.bend_cost = router_opts.bend_cost;
 
     route_budgets budgeting_inf;

vpr/src/route/rr_graph.cpp

@@ -257,7 +257,6 @@ static void load_rr_switch_inf(const int num_arch_switches, const float R_minW_n
 static void alloc_rr_switch_inf(t_arch_switch_fanin& switch_fanin);
 
 static void rr_graph_externals(const std::vector<t_segment_inf>& segment_inf,
-                               int max_chan_width,
                                int wire_to_rr_ipin_switch,
                                enum e_base_cost_type base_cost_type);
 
@@ -740,8 +739,7 @@ static void build_rr_graph(const t_graph_type graph_type,
     //Save the channel widths for the newly constructed graph
     device_ctx.chan_width = nodes_per_chan;
 
-    rr_graph_externals(segment_inf, max_chan_width,
-                       *wire_to_rr_ipin_switch, base_cost_type);
+    rr_graph_externals(segment_inf, *wire_to_rr_ipin_switch, base_cost_type);
 
     check_rr_graph(graph_type, grid, types);
 
@@ -907,14 +905,12 @@ static void remap_rr_node_switch_indices(const t_arch_switch_fanin& switch_fanin
 }
 
 static void rr_graph_externals(const std::vector<t_segment_inf>& segment_inf,
-                               int max_chan_width,
                                int wire_to_rr_ipin_switch,
                                enum e_base_cost_type base_cost_type) {
     auto& device_ctx = g_vpr_ctx.device();
 
     add_rr_graph_C_from_switches(device_ctx.rr_switch_inf[wire_to_rr_ipin_switch].Cin);
-    alloc_and_load_rr_indexed_data(segment_inf, device_ctx.rr_node_indices,
-                                   max_chan_width, wire_to_rr_ipin_switch, base_cost_type);
+    alloc_and_load_rr_indexed_data(segment_inf, wire_to_rr_ipin_switch, base_cost_type);
     load_rr_index_segments(segment_inf.size());
 }