Merge branch 'master' into temp_remove_redundant_strong_odin_tests and resolve conflicts

Author: soheilshahrouz

vpr/src/base/stats.cpp

@@ -24,18 +24,6 @@ void routing_stats(const Netlist<>& net_list,
                    RRSwitchId wire_to_ipin_switch,
                    bool is_flat);
 
-/**
- * @brief Calculates the routing channel width at each grid location.
- *
- * Iterates through all RR nodes and counts how many wires pass through each (x, y) location
- * for both horizontal (CHANX) and vertical (CHANY) channels.
- *
- * @return A pair of 3D matrices:
- *         - First: CHANX width per [layer][x][y]
- *         - Second: CHANY width per [layer][x][y]
- */
-std::pair<vtr::NdMatrix<int, 3>, vtr::NdMatrix<int, 3>> calculate_channel_width();
-
 void print_wirelen_prob_dist(bool is_flat);
 
 void print_lambda();

vpr/src/base/stats.h

@@ -263,6 +263,16 @@ struct DeviceContext : public Context {
 
     int delayless_switch_idx = UNDEFINED;
 
+    /// Stores the number of CHANX wire segments in each routing channel segment at [layer][x][y]
+    vtr::NdMatrix<int, 3> rr_chanx_segment_width;
+    /// Stores the number of CHANY wire segments in each routing channel segment at [layer][x][y]
+    vtr::NdMatrix<int, 3> rr_chany_segment_width;
+
+    /// Stores the maximum channel segment width in each horizontal channel
+    std::vector<int> rr_chanx_width;
+    /// Stores the maximum channel segment width in each vertical channel
+    std::vector<int> rr_chany_width;
+
     bool rr_graph_is_flat = false;
 
     /*

vpr/src/base/vpr_context.h

@@ -437,16 +437,18 @@ void init_draw_coords(float clb_width, const BlkLocRegistry& blk_loc_registry) {
     t_draw_state* draw_state = get_draw_state_vars();
     t_draw_coords* draw_coords = get_draw_coords_vars();
     const DeviceContext& device_ctx = g_vpr_ctx.device();
+    const DeviceGrid& grid = device_ctx.grid;
     const RRGraphView& rr_graph = device_ctx.rr_graph;
 
     /* Store a reference to block location variables so that other drawing
      * functions can access block location information without accessing
      * the global placement state, which is inaccessible during placement.*/
     draw_state->set_graphics_blk_loc_registry_ref(blk_loc_registry);
 
-    if (!draw_state->show_graphics && !draw_state->save_graphics
-        && draw_state->graphics_commands.empty())
-        return; //do not initialize only if --disp off and --save_graphics off
+    // do not initialize only if --disp off and --save_graphics off
+    if (!draw_state->show_graphics && !draw_state->save_graphics && draw_state->graphics_commands.empty()) {
+        return;
+    }
 
     /* Each time routing is on screen, need to reallocate the color of each *
      * rr_node, as the number of rr_nodes may change.						*/
@@ -468,32 +470,28 @@ void init_draw_coords(float clb_width, const BlkLocRegistry& blk_loc_registry) {
     }
 
     size_t j = 0;
-    for (size_t i = 0; i < (device_ctx.grid.width() - 1); i++) {
+    for (size_t i = 0; i < grid.width() - 1; i++) {
         draw_coords->tile_x[i] = (i * draw_coords->get_tile_width()) + j;
-        j += device_ctx.chan_width.y_list[i] + 1; /* N wires need N+1 units of space */
+        j += device_ctx.rr_chany_width[i] + 1; // N wires need N+1 units of space
     }
-    draw_coords->tile_x[device_ctx.grid.width() - 1] = ((device_ctx.grid.width()
-                                                         - 1)
-                                                        * draw_coords->get_tile_width())
-                                                       + j;
+    draw_coords->tile_x[grid.width() - 1] = (grid.width() - 1) * draw_coords->get_tile_width() + j;
+
     j = 0;
-    for (size_t i = 0; i < (device_ctx.grid.height() - 1); ++i) {
+    for (size_t i = 0; i < device_ctx.grid.height() - 1; ++i) {
         draw_coords->tile_y[i] = (i * draw_coords->get_tile_width()) + j;
-        j += device_ctx.chan_width.x_list[i] + 1;
+        j += device_ctx.rr_chanx_width[i] + 1;
     }
-    draw_coords->tile_y[device_ctx.grid.height() - 1] = ((device_ctx.grid.height() - 1) * draw_coords->get_tile_width())
-                                                        + j;
+    draw_coords->tile_y[grid.height() - 1] = (grid.height() - 1) * draw_coords->get_tile_width() + j;
+
     /* Load coordinates of sub-blocks inside the clbs */
     draw_internal_init_blk();
     //Margin beyond edge of the drawn device to extend the visible world
     //Setting this to > 0.0 means 'Zoom Fit' leave some fraction of white
     //space around the device edges
     constexpr float VISIBLE_MARGIN = 0.01;
 
-    float draw_width = draw_coords->tile_x[device_ctx.grid.width() - 1]
-                       + draw_coords->get_tile_width();
-    float draw_height = draw_coords->tile_y[device_ctx.grid.height() - 1]
-                        + draw_coords->get_tile_width();
+    float draw_width = draw_coords->tile_x[grid.width() - 1] + draw_coords->get_tile_width();
+    float draw_height = draw_coords->tile_y[grid.height() - 1] + draw_coords->get_tile_width();
 
     initial_world = ezgl::rectangle(
         {-VISIBLE_MARGIN * draw_width, -VISIBLE_MARGIN * draw_height},

vpr/src/draw/draw.cpp

@@ -143,33 +143,34 @@ void NetCostHandler::alloc_and_load_chan_w_factors_for_place_cost_() {
     const size_t grid_height = device_ctx.grid.height();
     const size_t grid_width = device_ctx.grid.width();
 
-    /* These arrays contain accumulative channel width between channel zero and
-     * the channel specified by the given index. The accumulated channel width
-     * is inclusive, meaning that it includes both channel zero and channel `idx`.
-     * To compute the total channel width between channels 'low' and 'high', use the
-     * following formula:
-     *      acc_chan?_width_[high] - acc_chan?_width_[low - 1]
-     * This returns the total number of tracks between channels 'low' and 'high',
-     * including tracks in these channels.
-     */
+    // These arrays contain accumulative channel width between channel zero and
+    // the channel specified by the given index. The accumulated channel width
+    // is inclusive, meaning that it includes both channel zero and channel `idx`.
+    // To compute the total channel width between channels 'low' and 'high', use the
+    // following formula:
+    //      acc_chan?_width_[high] - acc_chan?_width_[low - 1]
+    // This returns the total number of tracks between channels 'low' and 'high',
+    // including tracks in these channels.
     acc_chanx_width_ = vtr::PrefixSum1D<int>(grid_height, [&](size_t y) noexcept {
-        int chan_x_width = device_ctx.chan_width.x_list[y];
+        int chan_x_width = device_ctx.rr_chanx_width[y];
 
-        /* If the number of tracks in a channel is zero, two consecutive elements take the same
-         * value. This can lead to a division by zero in get_chanxy_cost_fac_(). To avoid this
-         * potential issue, we assume that the channel width is at least 1.
-         */
-        if (chan_x_width == 0)
+        // If the number of tracks in a channel is zero, two consecutive elements take the same
+        // value. This can lead to a division by zero in get_chanxy_cost_fac_(). To avoid this
+        // potential issue, we assume that the channel width is at least 1.
+        if (chan_x_width == 0) {
             return 1;
+        }
 
         return chan_x_width;
     });
+
     acc_chany_width_ = vtr::PrefixSum1D<int>(grid_width, [&](size_t x) noexcept {
-        int chan_y_width = device_ctx.chan_width.y_list[x];
+        int chan_y_width = device_ctx.rr_chany_width[x];
 
         // to avoid a division by zero
-        if (chan_y_width == 0)
+        if (chan_y_width == 0) {
             return 1;
+        }
 
         return chan_y_width;
     });
@@ -1819,7 +1820,8 @@ std::pair<vtr::NdMatrix<double, 3>, vtr::NdMatrix<double, 3>> NetCostHandler::es
         }
     }
 
-    const auto [chanx_width, chany_width] = calculate_channel_width();
+    const vtr::NdMatrix<int, 3>& chanx_width = device_ctx.rr_chanx_segment_width;
+    const vtr::NdMatrix<int, 3>& chany_width = device_ctx.rr_chany_segment_width;
 
     VTR_ASSERT(chanx_util.size() == chany_util.size());
     VTR_ASSERT(chanx_util.ndims() == chany_util.ndims());

vpr/src/place/net_cost_handler.cpp

@@ -376,6 +376,14 @@ static void build_rr_graph(e_graph_type graph_type,
 static int get_delayless_switch_id(const t_det_routing_arch& det_routing_arch,
                                    bool load_rr_graph);
 
+/**
+ * @brief Calculates the routing channel width at each grid location.
+ *
+ * Iterates through all RR nodes and counts how many wires pass through each (layer, x, y) location
+ * for both horizontal (CHANX) and vertical (CHANY) channels.
+ */
+static void alloc_and_init_channel_width();
+
 /******************* Subroutine definitions *******************************/
 
 void create_rr_graph(e_graph_type graph_type,
@@ -533,6 +541,8 @@ void create_rr_graph(e_graph_type graph_type,
                            device_ctx.rr_graph.rr_nodes(),
                            is_flat);
 
+    alloc_and_init_channel_width();
+
     print_rr_graph_stats();
 
     // Write out rr graph file if needed - Currently, writing the flat rr-graph is not supported since loading from a flat rr-graph is not supported.
@@ -1116,6 +1126,53 @@ static int get_delayless_switch_id(const t_det_routing_arch& det_routing_arch,
     return delayless_switch;
 }
 
+static void alloc_and_init_channel_width() {
+    DeviceContext& mutable_device_ctx = g_vpr_ctx.mutable_device();
+    const DeviceGrid& grid = mutable_device_ctx.grid;
+    const auto& rr_graph = mutable_device_ctx.rr_graph;
+
+    vtr::NdMatrix<int, 3>& chanx_width = mutable_device_ctx.rr_chanx_segment_width;
+    vtr::NdMatrix<int, 3>& chany_width = mutable_device_ctx.rr_chany_segment_width;
+
+    chanx_width.resize({grid.get_num_layers(), grid.width(), grid.height()});
+    chany_width.resize({grid.get_num_layers(), grid.width(), grid.height()});
+
+    chanx_width.fill(0);
+    chany_width.fill(0);
+
+    for (RRNodeId node_id : rr_graph.nodes()) {
+        e_rr_type rr_type = rr_graph.node_type(node_id);
+
+        if (rr_type == e_rr_type::CHANX) {
+            int y = rr_graph.node_ylow(node_id);
+            int layer = rr_graph.node_layer_low(node_id);
+            for (int x = rr_graph.node_xlow(node_id); x <= rr_graph.node_xhigh(node_id); x++) {
+                chanx_width[layer][x][y] += rr_graph.node_capacity(node_id);
+            }
+        } else if (rr_type == e_rr_type::CHANY) {
+            int x = rr_graph.node_xlow(node_id);
+            int layer = rr_graph.node_layer_low(node_id);
+            for (int y = rr_graph.node_ylow(node_id); y <= rr_graph.node_yhigh(node_id); y++) {
+                chany_width[layer][x][y] += rr_graph.node_capacity(node_id);
+            }
+        }
+    }
+
+    std::vector<int>& chanx_width_list = mutable_device_ctx.rr_chanx_width;
+    std::vector<int>& chany_width_list = mutable_device_ctx.rr_chany_width;
+
+    chanx_width_list.resize(grid.height());
+    chany_width_list.resize(grid.width());
+
+    std::ranges::fill(chanx_width_list, 0);
+    std::ranges::fill(chany_width_list, 0);
+
+    for (t_physical_tile_loc loc : grid.all_locations()) {
+        chanx_width_list[loc.y] = std::max(chanx_width[loc.layer_num][loc.x][loc.y], chanx_width_list[loc.y]);
+        chany_width_list[loc.x] = std::max(chany_width[loc.layer_num][loc.x][loc.y], chany_width_list[loc.x]);
+    }
+}
+
 void build_tile_rr_graph(RRGraphBuilder& rr_graph_builder,
                          const t_det_routing_arch& det_routing_arch,
                          t_physical_tile_type_ptr physical_tile,

vpr/src/route/rr_graph_generation/rr_graph.cpp

@@ -1,5 +1,5 @@
 arch	circuit	script_params	vtr_flow_elapsed_time	vtr_max_mem_stage	vtr_max_mem	error	odin_synth_time	max_odin_mem	parmys_synth_time	max_parmys_mem	abc_depth	abc_synth_time	abc_cec_time	abc_sec_time	max_abc_mem	ace_time	max_ace_mem	num_clb	num_io	num_memories	num_mult	vpr_status	vpr_revision	vpr_build_info	vpr_compiler	vpr_compiled	hostname	rundir	max_vpr_mem	num_primary_inputs	num_primary_outputs	num_pre_packed_nets	num_pre_packed_blocks	num_netlist_clocks	num_post_packed_nets	num_post_packed_blocks	device_width	device_height	device_grid_tiles	device_limiting_resources	device_name	pack_mem	pack_time	initial_placed_wirelength_est	placed_wirelength_est	total_swap	accepted_swap	rejected_swap	aborted_swap	place_mem	place_time	place_quench_time	min_chan_width	routed_wirelength	min_chan_width_route_success_iteration	logic_block_area_total	logic_block_area_used	min_chan_width_routing_area_total	min_chan_width_routing_area_per_tile	min_chan_width_route_time	
-k4_N10_memSize16384_memData64.xml	ch_intrinsics.v	common	1.07	vpr	64.88 MiB		-1	-1	0.15	28236	3	0.06	-1	-1	36544	-1	-1	72	99	1	0	success	v8.0.0-12799-g50a644d78	release IPO VTR_ASSERT_LEVEL=2	GNU 11.4.0 on Linux-6.8.0-60-generic x86_64	2025-06-10T17:21:16	llavign1-OptiPlex-7070	/home/llavign1/Gits/vtr-clone/vtr_flow/tasks	66436	99	130	353	483	1	222	302	13	13	169	clb	auto	25.3 MiB	0.03	1748.73	707	29650	4654	11713	13283	64.9 MiB	0.02	0.00	26	1506	9	3.33e+06	2.28e+06	360896.	2135.48	0.38	
-k4_N10_memSize16384_memData64.xml	diffeq1.v	common	2.66	vpr	67.86 MiB		-1	-1	0.19	32844	23	0.23	-1	-1	37316	-1	-1	74	162	0	5	success	v8.0.0-12799-g50a644d78	release IPO VTR_ASSERT_LEVEL=2	GNU 11.4.0 on Linux-6.8.0-60-generic x86_64	2025-06-10T17:21:16	llavign1-OptiPlex-7070	/home/llavign1/Gits/vtr-clone/vtr_flow/tasks	69484	162	96	1186	1127	1	667	337	13	13	169	clb	auto	28.2 MiB	0.10	7906.16	4859	81205	21212	54650	5343	67.9 MiB	0.08	0.00	50	9091	14	3.33e+06	2.67e+06	641417.	3795.37	1.37	
-k4_N10_memSize16384_memData64.xml	single_wire.v	common	0.31	vpr	62.98 MiB		-1	-1	0.05	25804	1	0.01	-1	-1	33044	-1	-1	0	1	0	0	success	v8.0.0-12799-g50a644d78	release IPO VTR_ASSERT_LEVEL=2	GNU 11.4.0 on Linux-6.8.0-60-generic x86_64	2025-06-10T17:21:16	llavign1-OptiPlex-7070	/home/llavign1/Gits/vtr-clone/vtr_flow/tasks	64496	1	1	1	2	0	1	2	3	3	9	-1	auto	24.5 MiB	0.00	2	2	3	0	3	0	63.0 MiB	0.00	0.00	2	1	1	30000	0	1489.46	165.495	0.00	
-k4_N10_memSize16384_memData64.xml	single_ff.v	common	0.41	vpr	63.02 MiB		-1	-1	0.08	26064	1	0.01	-1	-1	33064	-1	-1	1	2	0	0	success	v8.0.0-12799-g50a644d78	release IPO VTR_ASSERT_LEVEL=2	GNU 11.4.0 on Linux-6.8.0-60-generic x86_64	2025-06-10T17:21:16	llavign1-OptiPlex-7070	/home/llavign1/Gits/vtr-clone/vtr_flow/tasks	64532	2	1	3	4	1	3	4	3	3	9	-1	auto	24.7 MiB	0.00	6	6	9	6	0	3	63.0 MiB	0.00	0.00	16	5	1	30000	30000	2550.78	283.420	0.00	
+k4_N10_memSize16384_memData64.xml	ch_intrinsics.v	common	1.64	vpr	64.26 MiB		-1	-1	0.18	21064	3	0.07	-1	-1	32716	-1	-1	72	99	1	0	success	v8.0.0-14178-g4818739e3-dirty	release IPO VTR_ASSERT_LEVEL=2	GNU 13.3.0 on Linux-6.8.0-71-generic x86_64	2025-10-15T12:13:40	betzgrp-wintermute	/home/gholam39/vpr/vtr-verilog-to-routing/vtr_flow	65800	99	130	353	483	1	222	302	13	13	169	clb	auto	24.4 MiB	0.03	1748.73	1183	124778	46879	23081	54818	64.3 MiB	0.15	0.00	22	2035	37	3.33e+06	2.28e+06	311708.	1844.43	0.53	
+k4_N10_memSize16384_memData64.xml	diffeq1.v	common	2.81	vpr	67.24 MiB		-1	-1	0.22	25288	23	0.24	-1	-1	33564	-1	-1	74	162	0	5	success	v8.0.0-14178-g4818739e3-dirty	release IPO VTR_ASSERT_LEVEL=2	GNU 13.3.0 on Linux-6.8.0-71-generic x86_64	2025-10-15T12:13:40	betzgrp-wintermute	/home/gholam39/vpr/vtr-verilog-to-routing/vtr_flow	68852	162	96	1186	1127	1	667	337	13	13	169	clb	auto	27.5 MiB	0.12	7906.16	4910	96441	37729	58151	561	67.2 MiB	0.12	0.00	50	9566	16	3.33e+06	2.67e+06	641417.	3795.37	1.17	
+k4_N10_memSize16384_memData64.xml	single_wire.v	common	0.52	vpr	61.97 MiB		-1	-1	0.06	19028	1	0.02	-1	-1	29568	-1	-1	0	1	0	0	success	v8.0.0-14178-g4818739e3-dirty	release IPO VTR_ASSERT_LEVEL=2	GNU 13.3.0 on Linux-6.8.0-71-generic x86_64	2025-10-15T12:13:40	betzgrp-wintermute	/home/gholam39/vpr/vtr-verilog-to-routing/vtr_flow	63456	1	1	1	2	0	1	2	3	3	9	-1	auto	23.7 MiB	0.00	2	2	3	0	3	0	62.0 MiB	0.00	0.00	2	1	1	30000	0	1489.46	165.495	0.00	
+k4_N10_memSize16384_memData64.xml	single_ff.v	common	0.49	vpr	62.35 MiB		-1	-1	0.06	19528	1	0.02	-1	-1	29612	-1	-1	1	2	0	0	success	v8.0.0-14178-g4818739e3-dirty	release IPO VTR_ASSERT_LEVEL=2	GNU 13.3.0 on Linux-6.8.0-71-generic x86_64	2025-10-15T12:13:40	betzgrp-wintermute	/home/gholam39/vpr/vtr-verilog-to-routing/vtr_flow	63844	2	1	3	4	1	3	4	3	3	9	-1	auto	23.7 MiB	0.00	6	6	9	3	3	3	62.3 MiB	0.00	0.00	26	15	1	30000	30000	4706.78	522.975	0.01