dpl: introduce new legalization algorithm - negotiation based

src/cts/test/array.ok

@@ -132,7 +132,9 @@ Dummys used:
 [INFO RSZ-0058] Using max wire length 693um.
 [INFO RSZ-0047] Found 40 long wires.
 [INFO RSZ-0048] Inserted 88 buffers in 40 nets.
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 24399648.00 um^2, Instances area: 20261654.79 um^2, Utilization: 83.0%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                    2902

src/cts/test/array_ins_delay.ok

@@ -122,7 +122,9 @@
 [INFO RSZ-0058] Using max wire length 693um.
 [INFO RSZ-0047] Found 40 long wires.
 [INFO RSZ-0048] Inserted 88 buffers in 40 nets.
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 24399648.00 um^2, Instances area: 20261654.79 um^2, Utilization: 83.0%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                    2902

src/cts/test/array_max_wl.ok

@@ -572,7 +572,9 @@ TritonCTS forced slew degradation on these wires.
 [INFO RSZ-0058] Using max wire length 693um.
 [INFO RSZ-0047] Found 41 long wires.
 [INFO RSZ-0048] Inserted 87 buffers in 41 nets.
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 24399648.00 um^2, Instances area: 20261996.33 um^2, Utilization: 83.0%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                    3088

src/cts/test/array_no_blockages.ok

@@ -121,7 +121,9 @@
 [INFO RSZ-0058] Using max wire length 693um.
 [INFO RSZ-0047] Found 42 long wires.
 [INFO RSZ-0048] Inserted 92 buffers in 42 nets.
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 24399648.00 um^2, Instances area: 20261660.91 um^2, Utilization: 83.0%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                    2907

src/cts/test/array_repair_clock_nets.ok

@@ -136,7 +136,9 @@ Dummys used:
 [INFO RSZ-0058] Using max wire length 693um.
 [INFO RSZ-0047] Found 5 long wires.
 [INFO RSZ-0048] Inserted 5 buffers in 5 nets.
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 24399648.00 um^2, Instances area: 20261659.31 um^2, Utilization: 83.0%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                    2901

src/cts/test/gated_clock4.ok

@@ -48,7 +48,9 @@ Using 2 tracks default min distance between IO pins.
 [InitialPlace]  Iter: 3 conjugate gradient residual: 0.00000008 HPWL: 61445
 [InitialPlace]  Iter: 4 conjugate gradient residual: 0.00000011 HPWL: 61447
 [InitialPlace]  Iter: 5 conjugate gradient residual: 0.00000011 HPWL: 61445
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 999570.01 um^2, Instances area: 1274.41 um^2, Utilization: 0.1%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                     283

src/cts/test/hier_insertion_delay.ok

@@ -50,7 +50,9 @@ Using 2 tracks default min distance between IO pins.
 [InitialPlace]  Iter: 3 conjugate gradient residual: 0.00000012 HPWL: 1229790
 [InitialPlace]  Iter: 4 conjugate gradient residual: 0.00000012 HPWL: 1229790
 [InitialPlace]  Iter: 5 conjugate gradient residual: 0.00000012 HPWL: 1229790
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 47872.02 um^2, Instances area: 1357.12 um^2, Utilization: 2.8%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                     278

src/cts/test/simple_test_hier.ok

@@ -48,7 +48,9 @@ Using 2 tracks default min distance between IO pins.
 [InitialPlace]  Iter: 3 conjugate gradient residual: 0.00000010 HPWL: 9260
 [InitialPlace]  Iter: 4 conjugate gradient residual: 0.00000010 HPWL: 9260
 [InitialPlace]  Iter: 5 conjugate gradient residual: 0.00000010 HPWL: 9260
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 47872.02 um^2, Instances area: 72.35 um^2, Utilization: 0.2%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                      16

src/dbSta/test/escape_slash.ok

@@ -47,7 +47,9 @@ Using 2 tracks default min distance between IO pins.
 [InitialPlace]  Iter: 3 conjugate gradient residual: 0.00000010 HPWL: 61447
 [InitialPlace]  Iter: 4 conjugate gradient residual: 0.00000008 HPWL: 61444
 [InitialPlace]  Iter: 5 conjugate gradient residual: 0.00000009 HPWL: 61447
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 999570.01 um^2, Instances area: 1274.41 um^2, Utilization: 0.1%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                     283

src/dbSta/test/escape_slash_hier.ok

@@ -48,7 +48,9 @@ Using 2 tracks default min distance between IO pins.
 [InitialPlace]  Iter: 3 conjugate gradient residual: 0.00000008 HPWL: 61445
 [InitialPlace]  Iter: 4 conjugate gradient residual: 0.00000011 HPWL: 61447
 [InitialPlace]  Iter: 5 conjugate gradient residual: 0.00000011 HPWL: 61445
-[INFO DPL-0005] Max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-0006] Core area: 999570.01 um^2, Instances area: 1274.41 um^2, Utilization: 0.1%
+[INFO DPL-0005] Diamond search max displacement: +/- 500 sites horizontally, +/- 100 rows vertically.
+[INFO DPL-1101] Legalizing using diamond search.
 Movements Summary
 ---------------------------------------
 Total cells:                     283

src/dpl/BUILD

@@ -18,6 +18,9 @@ cc_library(
         "src/CheckPlacement.cpp",
         "src/DecapPlacement.cpp",
         "src/FillerPlacement.cpp",
+        "src/NegotiationLegalizer.cpp",
+        "src/NegotiationLegalizer.h",
+        "src/NegotiationLegalizerPass.cpp",
         "src/Opendp.cpp",
         "src/OptMirror.cpp",
         "src/Optdp.cpp",

src/dpl/CMakeLists.txt

@@ -26,7 +26,8 @@ add_library(dpl_lib
   src/DecapPlacement.cpp
   src/OptMirror.cpp
   src/PlacementDRC.cpp
-
+  src/NegotiationLegalizer.cpp
+  src/NegotiationLegalizerPass.cpp
   src/Optdp.cpp
   src/infrastructure/architecture.cxx
   src/util/color.cxx

src/dpl/README.md

@@ -1,12 +1,60 @@
 # Detailed Placement
 
-The detailed placement module in OpenROAD (`dpl`) is based on OpenDP, or 
-Open-Source Detailed Placement Engine. Its key features are: 
+The detailed placement module in OpenROAD (`dpl`) is based on OpenDP, or
+Open-Source Detailed Placement Engine. Its key features are:
 
--   Fence region.
--   Fragmented ROWs.
+-   Fence region support
+-   Fragmented row support
+-   Mixed-cell-height (1x–4x) legalization
+-   Two placement engines selectable at runtime
 
-## Commands
+## Placement Engines
+
+#### Diamond Search
+
+The default engine performs a BFS-style diamond search from each cell's
+global placement position, expanding outward in Manhattan order until a
+legal site is found.
+
+#### NegotiationLegalizer
+
+An optional two-pass legalizer based on the NBLG paper. Enabled with
+`-use_negotiation` on the `detailed_placement` command.
+
+```
+Global Placement result
+        │
+        ▼
+┌───────────────────┐
+│   Abacus Pass     │  Fast DP sweep, row-by-row.
+│   (Skipped)       │  Near-optimal for uncongested cells.
+│                   │  Mixed-cell-height via row assignment.
+│  Handles:         │  Power-rail alignment enforced.
+│  - 1x/2x/3x/4x    │  Fence violations → skipped (→ negotiation).
+│  - Fence regions  │
+└────────┬──────────┘
+         │ illegal cells (overlap / fence violated)
+         ▼
+┌───────────────────┐
+│ Negotiation Pass  │  Iterative rip-up & replace (from NBLG paper).
+│                   │  Illegal cells + spatial neighbors compete for
+│  NBLG components: │  grid resources.  History cost penalises
+│  - Adaptive pf    │  persistent congestion.  Isolation point skips
+│  - Isolation pt   │  already-legal cells in phase 2.
+│  - History cost   │
+└────────┬──────────┘
+         │
+         ▼
+┌───────────────────┐
+│ Post-optimisation │  Greedy displacement improvement (5 passes).
+│     (Skipped)     │  Cell swap via bipartite matching within groups.
+│                   │  
+└────────┬──────────┘
+         │
+         ▼
+   Legal placement written back to OpenDB
+```
+# Commands
 
 ```{note}
 - Parameters in square brackets `[-param param]` are optional.
@@ -23,30 +71,35 @@ detailed_placement
     [-max_displacement disp|{disp_x disp_y}]
     [-disallow_one_site_gaps]
     [-report_file_name filename]
+    [-use_negotiation]
+    [-abacus]
 ```
 
 #### Options
 
-| Switch Name | Description | 
+| Switch Name | Description |
 | ----- | ----- |
 | `-max_displacement` | Max distance that an instance can be moved (in microns) when finding a site where it can be placed. Either set one value for both directions or set `{disp_x disp_y}` for individual directions. The default values are `{0, 0}`, and the allowed values within are integers `[0, MAX_INT]`. |
 | `-disallow_one_site_gaps` | Option is deprecated. |
 | `-report_file_name` | File name for saving the report to (e.g. `report.json`.) |
 | `-incremental` | By default DPL initiates with all instances unplaced. With this flag DPL will check for already legalized instances and set them as placed. |
+| `-report_file_name` | File name for saving the report to (e.g. `report.json`.) |
+| `-use_negotiation` | Use the NegotiationLegalizer instead of the default diamond search engine. |
+| `-abacus` | Enable the Abacus pre-pass within the NegotiationLegalizer. Only effective when `-use_negotiation` is set. |
 
 ### Set Placement Padding
 
 The `set_placement_padding` command sets left and right padding in multiples
 of the row site width. Use the `set_placement_padding` command before
 legalizing placement to leave room for routing. Use the `-global` flag
-for padding that applies to all instances. Use  `-instances`
+for padding that applies to all instances. Use `-instances`
 for instance-specific padding.  The instances `insts` can be a list of instance
 names, or an instance object returned by the SDC `get_cells` command. To
 specify padding for all instances of a common master, use the `-filter`
 "ref_name == <name>" option to `get_cells`.
 
 ```tcl
-set_placement_padding   
+set_placement_padding
     -global|-masters masters|-instances insts
     [-right site_count]
     [-left site_count]
@@ -59,7 +112,7 @@ Either one of these flags must be set: `-global | -masters | -instances`.
 The order of preference is `global > masters > instances`
 ```
 
-| Switch Name | Description | 
+| Switch Name | Description |
 | ----- | ----- |
 | `-global` | Set padding globally using `left` and `right` values. |
 | `-masters` |  Set padding only for these masters using `left` and `right` values. | 
@@ -77,7 +130,7 @@ is supported, so `FILL*` will match, e.g., `FILLCELL_X1 FILLCELL_X16 FILLCELL_X2
 FILLCELL_X32 FILLCELL_X4 FILLCELL_X8`.  To specify a different naming prefix
 from `FILLER_` use `-prefix <new prefix>`.
 
-```tcl 
+```tcl
 filler_placement
     [-prefix prefix]
     [-verbose]
@@ -90,14 +143,14 @@ filler_placement
 | ----- | ----- |
 | `-prefix` | Prefix to name the filler cells. The default value is `FILLER_`. |
 | `-verbose` | Print the filler cell usage. |
-| `filler_masters` | Filler master cells. | 
+| `filler_masters` | Filler master cells. |
 
 ### Remove Fillers
 
 This command removes all filler cells.
 
 ```tcl
-remove_fillers 
+remove_fillers
 ```
 
 ### Check Placement
@@ -128,6 +181,7 @@ a weak attempt to reduce the total half-perimeter wirelength (HPWL).
 ```tcl
 optimize_mirroring
 ```
+
 ### Improve Placement
 
 The `improve_placement` command optimizes a given placed design.
@@ -150,7 +204,7 @@ If you are a developer, you might find these useful. More details can be found i
 | `get_inst_bbox` | Get bounding box of an instance. |
 | `get_inst_grid_bbox` | Get grid bounding box of an instance. |
 | `format_grid` | Format grid (takes in length `x` and site width `w` as inputs). |
-| `get_row_site` | Get row site name.
+| `get_row_site` | Get row site name. |
 
 ## Example scripts
 
@@ -164,14 +218,33 @@ Examples scripts demonstrating how to run `dpl` on a sample design of `aes` as f
 
 There are a set of regression tests in `./test`. Refer to this [section](../../README.md#regression-tests) for more information.
 
-Simply run the following script: 
+Simply run the following script:
 
 ```shell
 ./test/regression
 ```
 
 ## Limitations
 
+The following limitations apply when using the NegotiationLegalizer (`-use_negotiation`):
+
+1. **Abacus cluster chain**: The current Abacus implementation uses a
+   simplified cluster structure. A production version should maintain an
+   explicit doubly-linked list of cells within each cluster, as in the
+   original Spindler et al. paper.
+
+2. **Multithreading**: The negotiation pass is single-threaded.
+   Extend with the inter-region parallelism from NBLG (Algorithm 2, dynamic
+   region adjustment) using OpenMP or std::thread.
+
+3. **Fence region R-tree**: Replace linear scan in `FenceRegion::nearestRect()`
+   with a spatial index (Boost.Geometry rtree or OpenROAD's existing RTree)
+   for large designs with many fence sub-rectangles.
+
+4. **Row rail inference**: Currently uses row-index parity as a proxy for
+   VDD/VSS. Replace with actual LEF pg_pin parsing once available in the
+   build context.
+
 ## FAQs
 
 Check out [GitHub discussion](https://github.com/The-OpenROAD-Project/OpenROAD/discussions/categories/q-a?discussions_q=category%3AQ%26A+opendp+in%3Atitle)
@@ -184,7 +257,11 @@ about this tool.
 -   Rewrite and port to OpenDB/OpenROAD by James Cherry, Parallax Software
 
 ## References
+
 1. Do, S., Woo, M., & Kang, S. (2019, May). Fence-region-aware mixed-height standard cell legalization. In Proceedings of the 2019 on Great Lakes Symposium on VLSI (pp. 259-262). [(.pdf)](https://dl.acm.org/doi/10.1145/3299874.3318012)
+2. P. Spindler et al., "Abacus: Fast legalization of standard cell circuits with minimal movement," ISPD 2008.
+3. J. Chen et al., "NBLG: A Robust Legalizer for Mixed-Cell-Height Modern Design," IEEE TCAD, vol. 41, no. 11, 2022.
+4. L. McMurchie and C. Ebeling, "PathFinder: A negotiation-based performance-driven router for FPGAs," 1995.
 
 ## License
 

src/dpl/include/dpl/Opendp.h

@@ -107,7 +107,9 @@ class Opendp
   void detailedPlacement(int max_displacement_x,
                          int max_displacement_y,
                          const std::string& report_file_name = std::string(""),
-                         bool incremental = false);
+                         bool incremental = false,
+                         bool use_negotiation = false,
+                         bool run_abacus = false);
   void reportLegalizationStats() const;
 
   void setPaddingGlobal(int left, int right);
@@ -192,6 +194,7 @@ class Opendp
   friend class OpendpTest_IsPlaced_Test;
   friend class Graphics;
   friend class CellPlaceOrderLess;
+  friend class NegotiationLegalizer;
   void findDisplacementStats();
   DbuPt pointOffMacro(const Node& cell);
   void convertDbToCell(odb::dbInst* db_inst, Node& cell);
@@ -214,7 +217,7 @@ class Opendp
   void initPlacementDRC();
 
   std::string printBgBox(const boost::geometry::model::box<bgPoint>& queryBox);
-  void detailedPlacement();
+  void diamondDPL();
   DbuPt nearestPt(const Node* cell, const DbuRect& rect) const;
   int distToRect(const Node* cell, const odb::Rect& rect) const;
   static bool checkOverlap(const odb::Rect& cell, const odb::Rect& box);
@@ -394,6 +397,7 @@ class Opendp
   bool iterative_debug_ = false;
   bool deep_iterative_debug_ = false;
   bool incremental_ = false;
+  bool use_negotiation_ = false;
 
   // Magic numbers
   static constexpr double group_refine_percent_ = .05;

src/dpl/src/CheckPlacement.cpp

@@ -133,7 +133,8 @@ void Opendp::checkPlacement(const bool verbose,
           + region_placement_failures.size() + edge_spacing_failures.size()
           + blocked_layers_failures.size()
       > 0) {
-    logger_->error(DPL, 33, "detailed placement checks failed.");
+    logger_->error(
+        DPL, 33, "detailed placement checks failed during check placement.");
   }
 }