PlaceReasoner-Bench

An end-to-end macro-placement benchmark built entirely from open-source RTL, an open-source EDA flow, and an open technology node. Eight designs are each evaluated at 60% core utilization under two core aspect ratios (1:1 and 2:1), giving 16 placement tasks.

Used by OpenLayout, a multi-agent physical-design flow on OpenROAD.

Designs

Design	Top module	#Macros	#Clusters	#Types	Clk (ns)
ariane133	ariane	133	14	1	4.0
ariane81	ariane	81	8	6	4.0
bp_be	bp_be_top	10	3	3	2.6
bp_fe	bp_fe_top	11	4	3	1.8
ethernet	eth_top	64	2	1	8.0
swerv_wrapper	swerv_wrapper	28	3	3	2.0
vga_lcd	vga_enh_top	62	12	3	2.0
VeriGPU	gpu_card	12	1	1	1.2
Total		401	47	--	--

#Types counts distinct macro types; Clk is the target clock period.

Top module is the design top cell name, carried by both the DESIGN statement in each DEF and the single (flattened) module in <design>.v — use it for link_design / -top arguments. Note it differs from the directory name for all but swerv_wrapper, and that ariane133 and ariane81 share the top name ariane, so they must be kept in separate run directories.

Layout

Each design directory is self-contained:

<design>/
  <design>.v                        gate-level netlist (flattened)
  constraint.sdc                    timing constraints
  lef/                              tech LEF + macro LEFs (Nangate45, FakeRAM 2.0)
  lib/                              timing libraries
  def/
    1_1_u60_floorplan.def           floorplan, aspect ratio 1:1 (ar1), 60% utilization
    2_1_floorplan.def               floorplan, aspect ratio 2:1 (ar2), 60% utilization
  <design>_macro_clusters.txt       macro cluster assignment

The two DEFs differ only in core geometry; both fix the die/core outline and IO pin placement, and leave macro locations to be determined. A placement method consumes the DEF plus the LEF/LIB/SDC and emits macro coordinates and orientations; everything downstream (standard-cell placement, CTS, routing, timing repair) is held identical across methods so that post-route differences are attributable to macro placement alone.

<design>_macro_clusters.txt gives a connectivity-based grouping of macros, used as a soft placement constraint that encourages dataflow locality.

Usage

pip install huggingface_hub

hf download fenggev566/PlaceReasoner-Bench \
  --repo-type dataset \
  --local-dir Benchmark/PlaceReasoner-Bench

Or from Python:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="fenggev566/PlaceReasoner-Bench",
    repo_type="dataset",
    local_dir="Benchmark/PlaceReasoner-Bench",
)

To fetch a single design:

hf download fenggev566/PlaceReasoner-Bench \
  --repo-type dataset --include "bp_fe/*" \
  --local-dir Benchmark/PlaceReasoner-Bench

Provenance and licensing

• Designs and flow scaffolding derive from the open-source MacroPlacement testcases.
• Technology: Nangate45 open cell library (NangateOpenCellLibrary.*).
• Memory macros: FakeRAM 2.0 (fakeram45_*).
• Reference flow: Yosys for synthesis, OpenROAD for place-and-route.

All components are open source and redistributable, so the benchmark is fully reproducible without commercial tools or proprietary PDKs. The constituent parts carry their own upstream licenses (the RTL designs, the Nangate45 library, and the FakeRAM macros each have separate terms), which is why this repository is tagged other rather than a single SPDX identifier — consult the respective upstream projects before redistribution.