NoC SAT Routing

.gitmodules

@@ -3,5 +3,4 @@
 	url = https://github.com/catchorg/Catch2.git
 [submodule "libs/EXTERNAL/sockpp"]
 	path = libs/EXTERNAL/sockpp
-	#url = [email protected]:fpagliughi/sockpp.git
-	url = [email protected]:w0lek/sockpp.git # fork where in branch v1.0.0_no_complication_warnings there are compilation warnings fixes for upstream tag v1.0.0 of sockpp
+	url = https://github.com/w0lek/sockpp.git # fork where in branch v1.0.0_no_complication_warnings there are compilation warnings fixes for upstream tag v1.0.0 of sockpp

CMakeLists.txt

@@ -46,6 +46,7 @@ option(VPR_USE_SERVER "Specify whether vpr enables the server mode" ON)
 #VPR option --enable_analytic_placer is also required for Analytic Placement
 option(VPR_ANALYTIC_PLACE "Enable analytic placement in VPR." ON)
 option(VPR_ENABLE_INTERCHANGE "Enable FPGA interchange." ON)
+option(VPR_ENABLE_NOC_SAT_ROUTING "Enable NoC SAT routing." OFF)
 
 option(WITH_BLIFEXPLORER "Enable build with blifexplorer" OFF)
 
@@ -220,12 +221,12 @@ endforeach()
 set(FLEX_BISON_WARN_SUPPRESS_FLAGS "")
 set(FLEX_BISON_WARN_SUPPRESS_FLAGS_TO_CHECK
     "-Wno-redundant-decls"  #Flex/bison generate code with redundant declarations
-    "-Wno-switch-default"   #Flex/bison generate switch statments w/o default cases
+    "-Wno-switch-default"   #Flex/bison generate switch statements w/o default cases
     "-Wno-unused-parameter" #Flex produces functions with unused params in re-entrant mode
     "-Wno-missing-declarations" #Flex misses some declarations in re-entrant mode
     "-Wimplicit-fallthrough=0" #Bison produces some cases with explicit
     "-Wno-sign-compare" #Flex generates code which performs some signed/unsigned comparison
-    "-Wno-null-dereference" #Bison produces some cases with potenetial null derefs
+    "-Wno-null-dereference" #Bison produces some cases with potential null derefs
     )
 foreach(flag ${FLEX_BISON_WARN_SUPPRESS_FLAGS_TO_CHECK})
     CHECK_CXX_COMPILER_FLAG(${flag} CXX_COMPILER_SUPPORTS_${flag})

libs/EXTERNAL/CMakeLists.txt

@@ -9,6 +9,8 @@ add_subdirectory(libsdcparse)
 add_subdirectory(libblifparse)
 add_subdirectory(libtatum)
 add_subdirectory(libcatch2)
+
+#SET(BUILD_DEPS ON CACHE BOOL "Build libfoo shared library")
 #add_subdirectory(parmys)
 
 #VPR_USE_SERVER is initialized in the root CMakeLists

libs/libvtrutil/src/vtr_time.cpp

@@ -3,6 +3,8 @@
 #include "vtr_log.h"
 #include "vtr_rusage.h"
 
+#include <utility>
+
 namespace vtr {
 
 int f_timer_depth = 0;
@@ -30,7 +32,7 @@ float Timer::delta_max_rss_mib() const {
 
 ///@brief Constructor
 ScopedActionTimer::ScopedActionTimer(std::string action_str)
-    : action_(action_str)
+    : action_(std::move(action_str))
     , depth_(f_timer_depth++) {
 }
 
@@ -69,7 +71,7 @@ int ScopedActionTimer::depth() const {
 
 ///@brief Constructor
 ScopedFinishTimer::ScopedFinishTimer(std::string action_str)
-    : ScopedActionTimer(action_str) {
+    : ScopedActionTimer(std::move(action_str)) {
 }
 
 ///@brief Destructor
@@ -83,7 +85,7 @@ ScopedFinishTimer::~ScopedFinishTimer() {
 
 ///@brief Constructor
 ScopedStartFinishTimer::ScopedStartFinishTimer(std::string action_str)
-    : ScopedActionTimer(action_str) {
+    : ScopedActionTimer(std::move(action_str)) {
     vtr::printf_info("%s%s\n", pad().c_str(), action().c_str());
 }
 

libs/libvtrutil/src/vtr_time.h

@@ -39,7 +39,7 @@ class Timer {
 ///@brief Scoped time class which prints the time elapsed for the specifid action
 class ScopedActionTimer : public Timer {
   public:
-    ScopedActionTimer(const std::string action);
+    ScopedActionTimer(std::string action);
     ~ScopedActionTimer();
 
     void quiet(bool value);
@@ -71,7 +71,7 @@ class ScopedActionTimer : public Timer {
  */
 class ScopedFinishTimer : public ScopedActionTimer {
   public:
-    ScopedFinishTimer(const std::string action);
+    ScopedFinishTimer(std::string action);
     ~ScopedFinishTimer();
 };
 
@@ -91,7 +91,7 @@ class ScopedFinishTimer : public ScopedActionTimer {
  */
 class ScopedStartFinishTimer : public ScopedActionTimer {
   public:
-    ScopedStartFinishTimer(const std::string action);
+    ScopedStartFinishTimer(std::string action);
     ~ScopedStartFinishTimer();
 };
 } // namespace vtr

vpr/CMakeLists.txt

@@ -92,6 +92,20 @@ if(${VPR_ANALYTIC_PLACE})
 	endif(TARGET Eigen3::Eigen)
 endif()
 
+if (${VPR_ENABLE_NOC_SAT_ROUTING})
+    message(STATUS "VPR NoC SAT Routing: Requested")
+    find_package(ortools CONFIG REQUIRED)
+    if (TARGET ortools::ortools)
+        message(STATUS "VPR NoC SAT Routing dependency (or-tools): Found")
+        message(STATUS "VPR NoC SAT Routing: Enabled")
+        target_link_libraries(libvpr ortools::ortools)
+        target_compile_definitions(libvpr PUBLIC -DENABLE_NOC_SAT_ROUTING)
+    else ()
+        message(STATUS "VPR NoC SAT Routing dependency (or-tools): Not Found (You may need to set CMAKE_PREFIX_PATH in order for CMake to find your OR-Tools installation)")
+        message(STATUS "VPR NoC SAT Routing: Disabled")
+    endif (TARGET ortools::ortools)
+endif ()
+
 set_target_properties(libvpr PROPERTIES PREFIX "") #Avoid extra 'lib' prefix
 
 #Specify link-time dependencies
@@ -161,7 +175,7 @@ add_executable(vpr ${EXEC_SOURCES})
 target_link_libraries(vpr libvpr)
 
 
-#Supress IPO link warnings if IPO is enabled
+#Suppress IPO link warnings if IPO is enabled
 get_target_property(VPR_USES_IPO vpr INTERPROCEDURAL_OPTIMIZATION)
 if (VPR_USES_IPO)
     set_property(TARGET vpr APPEND PROPERTY LINK_FLAGS ${IPO_LINK_WARN_SUPRESS_FLAGS})
@@ -292,7 +306,7 @@ target_link_libraries(test_vpr
                         Catch2::Catch2WithMain
                         libvpr)
 
-#Supress IPO link warnings if IPO is enabled
+#Suppress IPO link warnings if IPO is enabled
 get_target_property(TEST_VPR_USES_IPO vpr INTERPROCEDURAL_OPTIMIZATION)
 if (TEST_VPR_USES_IPO)
     set_property(TARGET test_vpr APPEND PROPERTY LINK_FLAGS ${IPO_LINK_WARN_SUPRESS_FLAGS})

vpr/src/base/SetupVPR.cpp

@@ -744,8 +744,13 @@ static void SetupNocOpts(const t_options& Options, t_noc_opts* NocOpts) {
     NocOpts->noc_latency_constraints_weighting = Options.noc_latency_constraints_weighting;
     NocOpts->noc_latency_weighting = Options.noc_latency_weighting;
     NocOpts->noc_congestion_weighting = Options.noc_congestion_weighting;
-    NocOpts->noc_swap_percentage = Options.noc_swap_percentage;
     NocOpts->noc_centroid_weight = Options.noc_centroid_weight;
+    NocOpts->noc_swap_percentage = Options.noc_swap_percentage;
+    NocOpts->noc_sat_routing_bandwidth_resolution = Options.noc_sat_routing_bandwidth_resolution;
+    NocOpts->noc_sat_routing_latency_overrun_weighting = Options.noc_sat_routing_latency_overrun_weighting_factor;
+    NocOpts->noc_sat_routing_congestion_weighting = Options.noc_sat_routing_congestion_weighting_factor;
+    NocOpts->noc_sat_routing_num_workers = Options.noc_sat_routing_num_workers;
+    NocOpts->noc_sat_routing_log_search_progress = Options.noc_sat_routing_log_search_progress;
     NocOpts->noc_placement_file_name = Options.noc_placement_file_name;
 }
 

vpr/src/base/ShowSetup.cpp

@@ -802,6 +802,10 @@ static void ShowNocOpts(const t_noc_opts& NocOpts) {
     VTR_LOG("NocOpts.noc_latency_weighting: %f\n", NocOpts.noc_latency_weighting);
     VTR_LOG("NocOpts.noc_congestion_weighting: %f\n", NocOpts.noc_congestion_weighting);
     VTR_LOG("NocOpts.noc_swap_percentage: %d%%\n", NocOpts.noc_swap_percentage);
+    VTR_LOG("NocOpts.noc_sat_routing_bandwidth_resolution: %d\n", NocOpts.noc_sat_routing_bandwidth_resolution);
+    VTR_LOG("NocOpts.noc_sat_routing_latency_overrun_weighting: %d\n", NocOpts.noc_sat_routing_latency_overrun_weighting);
+    VTR_LOG("NocOpts.noc_sat_routing_congestion_weighting: %d\n", NocOpts.noc_sat_routing_congestion_weighting);
+    VTR_LOG("NocOpts.noc_sat_routing_num_workers: %d\n", NocOpts.noc_sat_routing_num_workers);
     VTR_LOG("NocOpts.noc_routing_algorithm: %s\n", NocOpts.noc_placement_file_name.c_str());
     VTR_LOG("\n");
 }

vpr/src/base/place_and_route.cpp

@@ -65,8 +65,8 @@ int binary_search_place_and_route(const Netlist<>& placement_net_list,
                                   t_det_routing_arch* det_routing_arch,
                                   std::vector<t_segment_inf>& segment_inf,
                                   NetPinsMatrix<float>& net_delay,
-                                  std::shared_ptr<SetupHoldTimingInfo> timing_info,
-                                  std::shared_ptr<RoutingDelayCalculator> delay_calc,
+                                  const std::shared_ptr<SetupHoldTimingInfo>& timing_info,
+                                  const std::shared_ptr<RoutingDelayCalculator>& delay_calc,
                                   bool is_flat) {
     vtr::vector<ParentNetId, vtr::optional<RouteTree>> best_routing; /* Saves the best routing found so far. */
     int current, low, high, final;
@@ -101,7 +101,7 @@ int binary_search_place_and_route(const Netlist<>& placement_net_list,
         graph_directionality = (det_routing_arch->directionality == BI_DIRECTIONAL ? GRAPH_BIDIR : GRAPH_UNIDIR);
     }
 
-    VTR_ASSERT(net_delay.size());
+    VTR_ASSERT(!net_delay.empty());
 
     if (det_routing_arch->directionality == BI_DIRECTIONAL)
         udsd_multiplier = 1;
@@ -298,7 +298,7 @@ int binary_search_place_and_route(const Netlist<>& placement_net_list,
         current = current + current % udsd_multiplier;
     }
 
-    /* The binary search above occassionally does not find the minimum    *
+    /* The binary search above occasionally does not find the minimum    *
      * routeable channel width.  Sometimes a circuit that will not route  *
      * in 19 channels will route in 18, due to router flukiness.  If      *
      * verify_binary_search is set, the code below will ensure that FPGAs *

vpr/src/base/place_and_route.h

@@ -36,8 +36,8 @@ int binary_search_place_and_route(const Netlist<>& placement_net_list,
                                   t_det_routing_arch* det_routing_arch,
                                   std::vector<t_segment_inf>& segment_inf,
                                   NetPinsMatrix<float>& net_delay,
-                                  std::shared_ptr<SetupHoldTimingInfo> timing_info,
-                                  std::shared_ptr<RoutingDelayCalculator> delay_calc,
+                                  const std::shared_ptr<SetupHoldTimingInfo>& timing_info,
+                                  const std::shared_ptr<RoutingDelayCalculator>& delay_calc,
                                   bool is_flat);
 
 t_chan_width init_chan(int cfactor,

vpr/src/base/read_options.cpp

@@ -2788,8 +2788,8 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
 
     noc_grp.add_argument<bool, ParseOnOff>(args.noc, "--noc")
         .help(
-            "Enables a NoC-driven placer that optimizes the placement of routers on the NoC."
-            "Also enables an option in the graphical display that can be used to display the NoC on the FPGA."
+            "Enables a NoC-driven placer that optimizes the placement of routers on the NoC. "
+            "Also enables an option in the graphical display that can be used to display the NoC on the FPGA. "
             "This should be on only when the FPGA device contains a NoC and the provided netlist connects to the NoC.")
         .default_value("off")
         .show_in(argparse::ShowIn::HELP_ONLY);
@@ -2805,7 +2805,7 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
         .help(
             "Controls the algorithm used by the NoC to route packets.\n"
             "* xy_routing: Uses the direction oriented routing algorithm. This is recommended to be used with mesh NoC topologies.\n"
-            "* bfs_routing: Uses the breadth first search algorithm. The objective is to find a route that uses a minimum number of links."
+            "* bfs_routing: Uses the breadth first search algorithm. The objective is to find a route that uses a minimum number of links. "
             " This algorithm is not guaranteed to generate deadlock-free traffic flow routes, but can be used with any NoC topology\n"
             "* west_first_routing: Uses the west-first routing algorithm. This is recommended to be used with mesh NoC topologies.\n"
             "* north_last_routing: Uses the north-last routing algorithm. This is recommended to be used with mesh NoC topologies.\n"
@@ -2818,9 +2818,9 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
 
     noc_grp.add_argument<double>(args.noc_placement_weighting, "--noc_placement_weighting")
         .help(
-            "Controls the importance of the NoC placement parameters relative to timing and wirelength of the design."
-            "This value can be >=0, where 0 would mean the placement is based solely on timing and wirelength."
-            "A value of 1 would mean noc placement is considered equal to timing and wirelength"
+            "Controls the importance of the NoC placement parameters relative to timing and wirelength of the design. "
+            "This value can be >=0, where 0 would mean the placement is based solely on timing and wirelength. "
+            "A value of 1 would mean noc placement is considered equal to timing and wirelength "
             "A value greater than 1 would mean the placement is increasingly dominated by NoC parameters.")
         .default_value("5.0")
         .show_in(argparse::ShowIn::HELP_ONLY);
@@ -2830,7 +2830,7 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
             "Controls the importance of minimizing the NoC aggregate bandwidth.\n"
             "This value can be >=0, where 0 would mean the aggregate bandwidth has no relevance to placement.\n"
             "Other positive numbers specify the importance of minimizing the NoC aggregate bandwidth to other NoC-related cost terms.\n"
-            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and"
+            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and "
             "only their relative ratios determine the importance of each cost term.")
         .default_value("0.38")
         .show_in(argparse::ShowIn::HELP_ONLY);
@@ -2840,7 +2840,7 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
             "Controls the importance of meeting all the NoC traffic flow latency constraints.\n"
             "This value can be >=0, where 0 would mean the latency constraints have no relevance to placement.\n"
             "Other positive numbers specify the importance of meeting latency constraints to other NoC-related cost terms.\n"
-            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and"
+            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and "
             "only their relative ratios determine the importance of each cost term.")
         .default_value("0.6")
         .show_in(argparse::ShowIn::HELP_ONLY);
@@ -2850,7 +2850,7 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
             "Controls the importance of reducing the latencies of the NoC traffic flows.\n"
             "This value can be >=0, where 0 would mean the latencies have no relevance to placement.\n"
             "Other positive numbers specify the importance of minimizing aggregate latency to other NoC-related cost terms.\n"
-            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and"
+            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and "
             "only their relative ratios determine the importance of each cost term.")
         .default_value("0.02")
         .show_in(argparse::ShowIn::HELP_ONLY);
@@ -2860,25 +2860,58 @@ argparse::ArgumentParser create_arg_parser(const std::string& prog_name, t_optio
             "Controls the importance of reducing the congestion of the NoC links.\n"
             "This value can be >=0, where 0 would mean the congestion has no relevance to placement.\n"
             "Other positive numbers specify the importance of minimizing congestion to other NoC-related cost terms.\n"
-            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and"
+            "Weighting factors for NoC-related cost terms are normalized internally. Therefore, their absolute values are not important, and "
             "only their relative ratios determine the importance of each cost term.")
         .default_value("0.25")
         .show_in(argparse::ShowIn::HELP_ONLY);
 
-    noc_grp.add_argument<double>(args.noc_swap_percentage, "--noc_swap_percentage")
+	noc_grp.add_argument<double>(args.noc_centroid_weight, "--noc_centroid_weight")
         .help(
             "Sets the minimum fraction of swaps attempted by the placer that are NoC blocks."
             "This value is an integer ranging from 0-100. 0 means NoC blocks will be moved at the same rate as other blocks. 100 means all swaps attempted by the placer are NoC router blocks.")
         .default_value("0")
         .show_in(argparse::ShowIn::HELP_ONLY);
-
-    noc_grp.add_argument<double>(args.noc_centroid_weight, "--noc_centroid_weight")
+        
+    noc_grp.add_argument<double>(args.noc_swap_percentage, "--noc_swap_percentage")
         .help(
-            "Sets the minimum fraction of swaps attempted by the placer that are NoC blocks."
+            "Sets the minimum fraction of swaps attempted by the placer that are NoC blocks. "
             "This value is an integer ranging from 0-100. 0 means NoC blocks will be moved at the same rate as other blocks. 100 means all swaps attempted by the placer are NoC router blocks.")
         .default_value("0")
         .show_in(argparse::ShowIn::HELP_ONLY);
 
+    noc_grp.add_argument<int>(args.noc_sat_routing_bandwidth_resolution, "--noc_sat_routing_bandwidth_resolution")
+        .help(
+            "Specifies the resolution by which traffic flow bandwidths are converted into integers in SAT routing algorithm.\n"
+            "The higher this number is, the more accurate the congestion estimation and aggregate bandwidth minimization is.\n"
+            "Higher resolution for bandwidth conversion increases the number of variables in the SAT formulation.")
+        .default_value("128")
+        .show_in(argparse::ShowIn::HELP_ONLY);
+
+    noc_grp.add_argument<int>(args.noc_sat_routing_latency_overrun_weighting_factor, "--noc_sat_routing_latency_overrun_weighting_factor")
+        .help(
+            "Controls the importance of reducing traffic flow latency overrun in SAT routing.")
+        .default_value("1024")
+        .show_in(argparse::ShowIn::HELP_ONLY);
+
+    noc_grp.add_argument<int>(args.noc_sat_routing_congestion_weighting_factor, "--noc_sat_routing_congestion_weighting_factor")
+        .help(
+            "Controls the importance of reducing the number of congested NoC links in SAT routing.")
+        .default_value("16384")
+        .show_in(argparse::ShowIn::HELP_ONLY);
+
+    noc_grp.add_argument<int>(args.noc_sat_routing_num_workers, "--noc_sat_routing_num_workers")
+        .help(
+            "The maximum number of parallel threads that the SAT solver can use to explore the solution space.\n"
+            "When set to 0, the number of parallel workers is set automatically to maximize parallelism.")
+        .default_value("0")
+        .show_in(argparse::ShowIn::HELP_ONLY);
+
+    noc_grp.add_argument<bool, ParseOnOff>(args.noc_sat_routing_log_search_progress, "--noc_sat_routing_log_search_progress")
+        .help(
+            "Print the detailed log of the SAT solver's search progress.")
+        .default_value("off")
+        .show_in(argparse::ShowIn::HELP_ONLY);
+
     noc_grp.add_argument<std::string>(args.noc_placement_file_name, "--noc_placement_file_name")
         .help(
             "Name of the output file that contains the NoC placement information."