Merge branch 'master' into temp_place_est_chan_util

Author: soheilshahrouz

CMakeLists.txt

@@ -42,9 +42,6 @@ option(VTR_ENABLE_CAPNPROTO "Enable capnproto binary serialization support in VP
 #Allow the user to decide whether to compile the server module
 option(VPR_USE_SERVER "Specify whether vpr enables the server mode" ON)
 
-#Allow the user to enable/disable VPR analytic placement
-#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)
 

README.developers.md

@@ -179,6 +179,29 @@ For large scale reformatting (should only be performed by VTR maintainers) the s
 
 Python files are automatically checked using `pylint` to ensure they follow established Python conventions.  You can run `pylint` on the entire repository by running `./dev/pylint_check.py`.  Certain files which were created before we adopted Python lint checking are grandfathered and are not checked.  To check *all* files, provide the `--check_grandfathered` argument.  You can also manually check individual files using `./dev/pylint_check.py <path_to_file1> <path_to_file2> ...`.
 
+# Sanitizing Includes
+
+You can use include-what-you-use or the clangd language server to make sure includes are correct and you don't have missing or unused includes.
+
+## include-what-you-use
+
+First, install include-what-you-use. Ubuntu/Debian users can run `sudo apt install iwyu` and Fedora/RHEL users can run `sudo dnf install iwyu`. You can then compile VTR with include-what-you-use enabled to get diagnostic messages about includes in all files with the following command:
+
+```
+make CMAKE_PARAMS="-DCMAKE_CXX_INCLUDE_WHAT_YOU_USE=include-what-you-use"
+```
+
+Note that this method checks all source files and the diagnostic messages can be very long.
+
+## clangd language server
+
+Alternatively, if your editor supports clangd, you can use it to get diagnostic messages for the specific file you are working with. Visual Studio Code users can use the clangd extension to use clangd instead of Microsoft's C/C++ extension. To enable include diagnostics, create a file named `.clangd` in VTR root directory and add the following lines to it:
+```
+Diagnostics:
+  UnusedIncludes: Strict
+  MissingIncludes: Strict
+```
+
 # Running Tests
 
 VTR has a variety of tests which are used to check for correctness, performance and Quality of Result (QoR).

doc/src/vpr/command_line_usage.rst

@@ -223,6 +223,12 @@ General Options
 
     If this option is not specified it may be set from the ``VPR_NUM_WORKERS`` environment variable; otherwise the default is used.
 
+    If this option is set to something other than 1, the following algorithms can be run in parallel:
+    
+    * Timing Analysis
+    * Routing (If routing algorithm is set to parallel or parallel_decomp; See :option:`--router_algorithm`)
+    * Portions of analytical placement (If using the analytical placement flow and compiled VPR with Eigen enabled; See :option:`--analytical_place`)
+
     .. note:: To compile VPR to allow the usage of parallel workers, ``libtbb-dev`` must be installed in the system.
 
     **Default:** ``1``
@@ -1517,6 +1523,35 @@ VPR uses a negotiated congestion algorithm (based on Pathfinder) to perform rout
     * `swns` - setup Worst Negative Slack (sWNS) [ns]
     * `stns` - Setup Total Negative Slack (sTNS) [ns]
 
+
+.. option:: --generate_net_timing_report {on | off}
+
+    Generates a report that lists the bounding box, slack, and delay of every routed connection in a design in CSV format (``report_net_timing.csv``). Each row in the CSV corresponds to a single net.
+
+    The report can later be used by other tools to enable further optimizations. For example, the Synopsys synthesis tool (Synplify) can use this information to re-synthesize the design and improve the Quality of Results (QoR).
+
+    Fields in the report are:
+
+    .. code-block:: none
+        
+        netname         : The name assigned to the net in the atom netlist
+        Fanout          : Net's fanout (number of sinks)
+        bb_xmin         : X coordinate of the net's bounding box's bottom-left corner
+        bb_ymin         : Y coordinate of the net's bounding box's bottom-left corner
+        bb_layer_min    : Lowest layer number of the net's bounding box
+        bb_xmax         : X coordinate of the net's bounding box's top-right corner
+        bb_ymax         : Y coordinate of the net's bounding box's top-right corner
+        bb_layer_max    : Highest layer number of the net's bounding box
+        src_pin_name    : Name of the net's source pin
+        src_pin_slack   : Setup slack of the net's source pin
+        sinks           : A semicolon-separated list of sink pin entries, each in the format:
+                          <sink_pin_name>,<sink_pin_slack>,<sink_pin_delay>
+
+    Example value for the ``sinks`` field:
+    ``"U2.B,0.12,0.5;U3.C,0.10,0.6;U4.D,0.08,0.7"``
+
+    **Default:** ``off``
+
 .. option:: --route_verbosity <int>
 
     Controls the verbosity of routing output.

libs/libarchfpga/src/arch_types.h

@@ -16,7 +16,7 @@
 #define TOKENS " \t\n"
 
 /* Value for UNDEFINED data */
-constexpr int UNDEFINED = -1;
+constexpr int ARCH_FPGA_UNDEFINED_VAL = -1;
 
 /* Maximum value for minimum channel width to avoid overflows of short data type.               */
 constexpr int MAX_CHANNEL_WIDTH = 8000;

libs/libarchfpga/src/arch_util.cpp

@@ -444,7 +444,7 @@ t_physical_tile_type get_empty_physical_type(const char* name /*= EMPTY_BLOCK_NA
     type.capacity = 0;
     type.num_drivers = 0;
     type.num_receivers = 0;
-    type.area = UNDEFINED;
+    type.area = ARCH_FPGA_UNDEFINED_VAL;
     type.switchblock_locations = vtr::Matrix<e_sb_type>({{size_t(type.width), size_t(type.height)}}, e_sb_type::FULL);
     type.switchblock_switch_overrides = vtr::Matrix<int>({{size_t(type.width), size_t(type.height)}}, DEFAULT_SWITCH);
     type.is_input_type = false;

libs/libarchfpga/src/read_xml_arch_file.cpp

@@ -2887,7 +2887,7 @@ static void ProcessChanWidthDistrDir(pugi::xml_node Node, t_chan* chan, const pu
                        "Unknown property %s for chan_width_distr x\n", Prop);
     }
 
-    chan->peak = get_attribute(Node, "peak", loc_data).as_float(UNDEFINED);
+    chan->peak = get_attribute(Node, "peak", loc_data).as_float(ARCH_FPGA_UNDEFINED_VAL);
     chan->width = get_attribute(Node, "width", loc_data, hasWidth).as_float(0);
     chan->xpeak = get_attribute(Node, "xpeak", loc_data, hasXpeak).as_float(0);
     chan->dc = get_attribute(Node, "dc", loc_data, hasDc).as_float(0);
@@ -2994,7 +2994,7 @@ static void ProcessTileProps(pugi::xml_node Node,
     /* Load properties */
     PhysicalTileType->width = get_attribute(Node, "width", loc_data, ReqOpt::OPTIONAL).as_uint(1);
     PhysicalTileType->height = get_attribute(Node, "height", loc_data, ReqOpt::OPTIONAL).as_uint(1);
-    PhysicalTileType->area = get_attribute(Node, "area", loc_data, ReqOpt::OPTIONAL).as_float(UNDEFINED);
+    PhysicalTileType->area = get_attribute(Node, "area", loc_data, ReqOpt::OPTIONAL).as_float(ARCH_FPGA_UNDEFINED_VAL);
 
     if (atof(Prop) < 0) {
         archfpga_throw(loc_data.filename_c_str(), loc_data.line(Node),
@@ -4264,8 +4264,8 @@ static std::vector<t_arch_switch_inf> ProcessSwitches(pugi::xml_node Parent,
 static void ProcessSwitchTdel(pugi::xml_node Node, const bool timing_enabled, t_arch_switch_inf& arch_switch, const pugiutil::loc_data& loc_data) {
     /* check if switch node has the Tdel property */
     bool has_Tdel_prop = false;
-    float Tdel_prop_value = get_attribute(Node, "Tdel", loc_data, ReqOpt::OPTIONAL).as_float(UNDEFINED);
-    if (Tdel_prop_value != UNDEFINED) {
+    float Tdel_prop_value = get_attribute(Node, "Tdel", loc_data, ReqOpt::OPTIONAL).as_float(ARCH_FPGA_UNDEFINED_VAL);
+    if (Tdel_prop_value != ARCH_FPGA_UNDEFINED_VAL) {
         has_Tdel_prop = true;
     }
 

libs/librrgraph/src/base/rr_graph_builder.cpp

@@ -31,31 +31,29 @@ void RRGraphBuilder::add_node_to_all_locs(RRNodeId node) {
     short node_layer = node_storage_.node_layer(node);
     short node_twist = node_storage_.node_ptc_twist(node);
     int node_offset = 0;
+
     for (int ix = node_storage_.node_xlow(node); ix <= node_storage_.node_xhigh(node); ix++) {
         for (int iy = node_storage_.node_ylow(node); iy <= node_storage_.node_yhigh(node); iy++) {
             node_ptc_num += node_twist * node_offset;
             node_offset++;
+
             switch (node_type) {
                 case e_rr_type::SOURCE:
                 case e_rr_type::SINK:
                 case e_rr_type::CHANY:
-                    node_lookup_.add_node(node, node_layer, ix, iy, node_type, node_ptc_num, TOTAL_2D_SIDES[0]);
-                    break;
                 case e_rr_type::CHANX:
-                    /* Currently need to swap x and y for CHANX because of chan, seg convention 
-                     * TODO: Once the builders is reworked for use consistent (x, y) convention,
-                     * the following swapping can be removed
-                     */
-                    node_lookup_.add_node(node, node_layer, iy, ix, node_type, node_ptc_num, TOTAL_2D_SIDES[0]);
+                    node_lookup_.add_node(node, node_layer, ix, iy, node_type, node_ptc_num, TOTAL_2D_SIDES[0]);
                     break;
+
                 case e_rr_type::OPIN:
                 case e_rr_type::IPIN:
-                    for (const e_side& side : TOTAL_2D_SIDES) {
+                    for (const e_side side : TOTAL_2D_SIDES) {
                         if (node_storage_.is_node_on_specific_side(node, side)) {
                             node_lookup_.add_node(node,node_layer, ix, iy, node_type, node_ptc_num, side);
                         }
                     }
                     break;
+
                 default:
                     VTR_LOG_ERROR("Invalid node type for node '%lu' in the routing resource graph file", size_t(node));
                     break;

libs/librrgraph/src/base/rr_graph_storage.cpp

@@ -464,7 +464,7 @@ size_t t_rr_graph_storage::count_rr_switches(
 
             if (arch_switch_inf[iswitch].fixed_Tdel()) {
                 //If delay is independent of fanin drop the unique fanin info
-                fanin = UNDEFINED;
+                fanin = ARCH_FPGA_UNDEFINED_VAL;
             }
 
             if (arch_switch_fanins[iswitch].count(fanin) == 0) {        //New fanin for this switch
@@ -482,7 +482,7 @@ size_t t_rr_graph_storage::count_rr_switches(
     for(size_t iswitch = 0; iswitch < arch_switch_counts.size(); ++iswitch) {
         if(arch_switch_fanins[iswitch].empty()){
             if(arch_switch_inf[iswitch].fixed_Tdel()){
-                arch_switch_fanins[iswitch][UNDEFINED] = num_rr_switches++;
+                arch_switch_fanins[iswitch][ARCH_FPGA_UNDEFINED_VAL] = num_rr_switches++;
             }
         }
     }
@@ -504,8 +504,8 @@ void t_rr_graph_storage::remap_rr_node_switch_indices(const t_arch_switch_fanin&
         int switch_index = edge_switch_[edge];
         int fanin = node_fan_in_[to_node];
 
-        if (switch_fanin[switch_index].count(UNDEFINED) == 1) {
-            fanin = UNDEFINED;
+        if (switch_fanin[switch_index].count(ARCH_FPGA_UNDEFINED_VAL) == 1) {
+            fanin = ARCH_FPGA_UNDEFINED_VAL;
         }
 
         auto itr = switch_fanin[switch_index].find(fanin);

libs/librrgraph/src/base/rr_graph_view.h

@@ -351,7 +351,7 @@ class RRGraphView {
 
                 start_x = " (" + std::to_string(node_xhigh(node)) + ","; //start coordinates have large value
                 start_y = std::to_string(node_yhigh(node)) + ",";
-                start_layer_str = std::to_string(node_layer_num);
+                start_layer_str = std::to_string(node_layer_num) + ")";
                 end_x = " (" + std::to_string(node_xlow(node)) + ","; //end coordinates have smaller value
                 end_y = std::to_string(node_ylow(node)) + ",";
                 end_layer_str = std::to_string(node_layer_num) + ")";
@@ -360,7 +360,7 @@ class RRGraphView {
             else {                                                      // signal travels in increasing direction, stays at same point, or can travel both directions
                 start_x = " (" + std::to_string(node_xlow(node)) + ","; //start coordinates have smaller value
                 start_y = std::to_string(node_ylow(node)) + ",";
-                start_layer_str = std::to_string(node_layer_num);
+                start_layer_str = std::to_string(node_layer_num) + ")";
                 end_x = " (" + std::to_string(node_xhigh(node)) + ","; //end coordinates have larger value
                 end_y = std::to_string(node_yhigh(node)) + ",";
                 end_layer_str = std::to_string(node_layer_num) + ")"; //layer number

libs/librrgraph/src/base/rr_spatial_lookup.cpp

@@ -1,9 +1,6 @@
 #include "vtr_assert.h"
 #include "rr_spatial_lookup.h"
 
-RRSpatialLookup::RRSpatialLookup() {
-}
-
 RRNodeId RRSpatialLookup::find_node(int layer,
                                     int x,
                                     int y,
@@ -33,18 +30,6 @@ RRNodeId RRSpatialLookup::find_node(int layer,
         return RRNodeId::INVALID();
     }
 
-    /* Currently need to swap x and y for CHANX because of chan, seg convention 
-     * This is due to that the fast look-up builders uses (y, x) coordinate when
-     * registering a CHANX node in the look-up
-     * TODO: Once the builders is reworked for use consistent (x, y) convention,
-     * the following swapping can be removed
-     */
-    size_t node_x = x;
-    size_t node_y = y;
-    if (type == e_rr_type::CHANX) {
-        std::swap(node_x, node_y);
-    }
-
     VTR_ASSERT_SAFE(4 == rr_node_indices_[type].ndims());
 
     /* Sanity check to ensure the layer, x, y, side and ptc are in range
@@ -59,23 +44,23 @@ RRNodeId RRSpatialLookup::find_node(int layer,
         return RRNodeId::INVALID();
     }
 
-    if (node_x >= rr_node_indices_[type].dim_size(1)) {
+    if (size_t(x) >= rr_node_indices_[type].dim_size(1)) {
         return RRNodeId::INVALID();
     }
 
-    if(node_y >= rr_node_indices_[type].dim_size(2)){
+    if (size_t(y) >= rr_node_indices_[type].dim_size(2)){
         return RRNodeId::INVALID();
     }
 
     if (node_side >= rr_node_indices_[type].dim_size(3)) {
         return RRNodeId::INVALID();
     }
 
-    if (size_t(ptc) >= rr_node_indices_[type][layer][node_x][node_y][node_side].size()) {
+    if (size_t(ptc) >= rr_node_indices_[type][layer][x][y][node_side].size()) {
         return RRNodeId::INVALID();
     }
 
-    return rr_node_indices_[type][layer][node_x][node_y][node_side][ptc];
+    return rr_node_indices_[type][layer][x][y][node_side][ptc];
 }
 
 std::vector<RRNodeId> RRSpatialLookup::find_nodes_in_range(int layer,
@@ -114,18 +99,6 @@ std::vector<RRNodeId> RRSpatialLookup::find_nodes(int layer,
         return nodes;
     }
 
-    /* Currently need to swap x and y for CHANX because of chan, seg convention 
-     * This is due to that the fast look-up builders uses (y, x) coordinate when
-     * registering a CHANX node in the look-up
-     * TODO: Once the builders is reworked for use consistent (x, y) convention,
-     * the following swapping can be removed
-     */
-    size_t node_x = x;
-    size_t node_y = y;
-    if (type == e_rr_type::CHANX) {
-        std::swap(node_x, node_y);
-    }
-
     VTR_ASSERT_SAFE(4 == rr_node_indices_[type].ndims());
 
     /* Sanity check to ensure the x, y, side are in range 
@@ -140,11 +113,11 @@ std::vector<RRNodeId> RRSpatialLookup::find_nodes(int layer,
         return nodes;
     }
 
-    if (node_x >= rr_node_indices_[type].dim_size(1)) {
+    if (size_t(x) >= rr_node_indices_[type].dim_size(1)) {
         return nodes;
     }
 
-    if(node_y >= rr_node_indices_[type].dim_size(2)){
+    if (size_t(y) >= rr_node_indices_[type].dim_size(2)){
         return nodes;
     }
 
@@ -154,14 +127,14 @@ std::vector<RRNodeId> RRSpatialLookup::find_nodes(int layer,
 
     /* Reserve space to avoid memory fragmentation */
     size_t num_nodes = 0;
-    for (const auto& node : rr_node_indices_[type][layer][node_x][node_y][side]) {
+    for (RRNodeId node : rr_node_indices_[type][layer][x][y][side]) {
         if (node.is_valid()) {
             num_nodes++;
         }
     }
 
     nodes.reserve(num_nodes);
-    for (const auto& node : rr_node_indices_[type][layer][node_x][node_y][side]) {
+    for (RRNodeId node : rr_node_indices_[type][layer][x][y][side]) {
         if (node.is_valid()) {
             nodes.emplace_back(node);
         }
@@ -338,7 +311,7 @@ void RRSpatialLookup::resize_nodes(int layer,
         || (x >= int(rr_node_indices_[type].dim_size(1)))
         || (y >= int(rr_node_indices_[type].dim_size(2)))
         || (size_t(side) >= rr_node_indices_[type].dim_size(3))) {
-        rr_node_indices_[type].resize({std::max(rr_node_indices_[type].dim_size(0),size_t(layer)+1),
+        rr_node_indices_[type].resize({std::max(rr_node_indices_[type].dim_size(0), size_t(layer)+1),
                                        std::max(rr_node_indices_[type].dim_size(1), size_t(x) + 1),
                                        std::max(rr_node_indices_[type].dim_size(2), size_t(y) + 1),
                                        std::max(rr_node_indices_[type].dim_size(3), size_t(side) + 1)});
@@ -352,7 +325,7 @@ void RRSpatialLookup::reorder(const vtr::vector<RRNodeId, RRNodeId>& dest_order)
             for (size_t x = 0; x < grid.dim_size(1); x++) {
                 for (size_t y = 0; y < grid.dim_size(2); y++) {
                     for (size_t s = 0; s < grid.dim_size(3); s++) {
-                        for (auto &node: grid[l][x][y][s]) {
+                        for (RRNodeId &node: grid[l][x][y][s]) {
                             if (node.is_valid()) {
                                 node = dest_order[node];
                             }

libs/librrgraph/src/base/rr_spatial_lookup.h

@@ -1,5 +1,4 @@
-#ifndef RR_SPATIAL_LOOKUP_H
-#define RR_SPATIAL_LOOKUP_H
+#pragma once
 
 /** 
  * @file
@@ -25,7 +24,7 @@ class RRSpatialLookup {
     /* -- Constructors -- */
   public:
     /* Explicitly define the only way to create an object */
-    explicit RRSpatialLookup();
+    explicit RRSpatialLookup() = default;
 
     /* Disable copy constructors and copy assignment operator
      * This is to avoid accidental copy because it could be an expensive operation considering that the 
@@ -293,5 +292,3 @@ class RRSpatialLookup {
     /* Fast look-up: TODO: Should rework the data type. Currently it is based on a 3-dimensional array mater where some dimensions must always be accessed with a specific index. Such limitation should be overcome */
     t_rr_node_indices rr_node_indices_;
 };
-
-#endif

libs/librrgraph/src/io/rr_graph_uxsdcxx_serializer.h

@@ -1848,16 +1848,11 @@ class RrGraphSerializer final : public uxsd::RrGraphBase<RrGraphContextTypes> {
 
         /* Alloc the lookup table */
         for (e_rr_type rr_type : RR_TYPES) {
-            if (rr_type == e_rr_type::CHANX) {
-                rr_graph_builder.node_lookup().resize_nodes(grid_.get_num_layers(), grid_.height(), grid_.width(), rr_type, NUM_2D_SIDES);
-            } else {
-                rr_graph_builder.node_lookup().resize_nodes(grid_.get_num_layers(), grid_.width(), grid_.height(), rr_type, NUM_2D_SIDES);
-            }
+            rr_graph_builder.node_lookup().resize_nodes(grid_.get_num_layers(), grid_.width(), grid_.height(), rr_type, NUM_2D_SIDES);
         }
 
         /* Add the correct node into the vector */
-        for (size_t inode = 0; inode < rr_nodes_->size(); inode++) {
-            auto node = (*rr_nodes_)[inode];
+        for (const t_rr_node& node : *rr_nodes_) {
             rr_graph_builder.add_node_to_all_locs(node.id());
         }
     }