Cannot reproduce example

[cedounet]

Hello,

I have not managed to run the provided example using:

and github head:

❯ g lg | head -5
9e44674 Brian Benchoff  (HEAD -> main, origin/main, origin/HEAD) Merge pull request #11 from makuru-org/main
45258df Makuru  Fixed error
f205705 Brian Benchoff  found IPC API incompatability with Content and plugin manager; updated instructions.
c84ecf7 Brian Benchoff  fix OpenGL
1c5c21a Brian Benchoff  derive max iterations is actually rea

Install as per documentation, using the example provided I run

❯  python main.py plugin --no-gui

2025-12-22 17:55:44,201 - root - WARNING - [LOGGING] Console: WARNING only (full logs in logs/ directory)
2025-12-22 17:55:44,461 - orthoroute.algorithms.manhattan.unified_pathfinder - WARNING - [BOUNDS] No pads found via board.nets, falling back to board._kicad_bounds + 3.0mm
2025-12-22 17:55:44,461 - orthoroute.algorithms.manhattan.unified_pathfinder - ERROR - [BOUNDS] No bounds available, using default 100x100mm
2025-12-22 17:56:43,510 - orthoroute.presentation.plugin.kicad_plugin - ERROR - [STEP1] End-to-end routing failed with instance PF-44205 - no results returned
2025-12-22 17:56:43,510 - root - ERROR - Plugin execution failed

The log has more information (I did enable debug in .json file:

[...]
2025-12-22 17:56:15,398 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO -   ... and 296 more pairs (showing first 10 only)
2025-12-22 17:56:15,398 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - Finalizing CSR: 40,690,112 edges from pre-allocated array
2025-12-22 17:56:15,398 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - Sorting 40,690,112 edges by source node...
2025-12-22 17:56:15,398 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU-SORT] Using CuPy GPU radix sort (expected ~3-5s for 54M edges)
2025-12-22 17:56:18,000 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU-SORT] GPU sort completed in 2.6 seconds (15.6M edges/sec)
2025-12-22 17:56:26,246 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [LAYER-MAP] Built edge→layer mapping: 40690112 edges, 19 layers
2025-12-22 17:56:26,246 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [EDGE-KIND] Horizontal/Vertical=2133500, Via=38556612
2025-12-22 17:56:26,546 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - CSR: 1197019 nodes, 40690112 edges
2025-12-22 17:56:26,553 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [MANHATTAN-VALIDATION] Sampling 1000 edges to verify H/V discipline...
2025-12-22 17:56:27,344 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [MANHATTAN-VALIDATION] All 1000 sampled edges are legal ✓
2025-12-22 17:56:27,347 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [LAYER-BALANCE] Initialized for 19 layers
2025-12-22 17:56:27,349 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VIA-POOL] Column pooling enabled (GPU): capacity=4 per (x,y)
2025-12-22 17:56:27,350 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VIA-POOL] Segment pooling enabled (GPU): 17 segments (z=1..17), capacity=2 per segment
2025-12-22 17:56:27,351 - orthoroute.algorithms.manhattan.pathfinder.via_kernels - INFO - [VIA-KERNELS] CUDA kernels compiled successfully
2025-12-22 17:56:27,351 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VIA-KERNELS] Manager initialized (GPU=YES)
2025-12-22 17:56:27,351 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [NODE-OWNER] Initialized node ownership tracking for 1,197,019 nodes
2025-12-22 17:56:27,415 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU-INIT] config.use_gpu=True, GPU_AVAILABLE=True, CUDA_DIJKSTRA_AVAILABLE=True
2025-12-22 17:56:27,415 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU-INIT] use_gpu_solver=True
2025-12-22 17:56:27,481 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA-COMPILE] Compiling validate_parents kernel for debugging
2025-12-22 17:56:27,481 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA-COMPILE] Compiling backtrace_paths kernel with use_bitmap parameter
2025-12-22 17:56:27,481 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled parallel edge relaxation kernel
2025-12-22 17:56:27,481 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled FULLY PARALLEL wavefront expansion kernel
2025-12-22 17:56:27,481 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled ACTIVE-LIST kernel (2-3× faster than one-block-per-ROI!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled PROCEDURAL NEIGHBOR kernel (P1-8: ditches CSR, pure arithmetic!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled DELTA-STEPPING bucket assignment kernel (P1-7: replaces Python loop!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled PERSISTENT KERNEL (P1-6: device-side queues, eliminates launch overhead!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled COMPACTION KERNEL (P3: GPU-side frontier compaction, no host sync!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled ACCOUNTANT KERNEL (Phase 5: GPU-side history/present/cost updates!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.pathfinder.cuda_dijkstra - INFO - [CUDA] Compiled GPU BACKTRACE KERNEL (eliminates 256 MB parent/dist CPU transfers!)
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU] CUDA Near-Far Dijkstra enabled (ROI > 5K nodes) with lattice dims
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU] GPU Compute Capability: 70
2025-12-22 17:56:27,482 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU] GPU Memory: 32.4 GB free / 34.1 GB total
2025-12-22 17:56:38,206 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - Identified 38556612 via edges
2025-12-22 17:56:43,495 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VIA-METADATA] Built metadata for 38556612 via edges on GPU in 5.288s
2025-12-22 17:56:43,509 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - Mapped 0 pads (from ~0)
2025-12-22 17:56:43,509 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VERIFY] Sample pad IDs: []
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.pad_escape_planner - INFO - PadEscapePlanner initialized with random seed: 42
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - Portal planning delegated to PadEscapePlanner (column-based, seed=42)
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - === Init complete ===
2025-12-22 17:56:43,510 - orthoroute.presentation.plugin.kicad_plugin - INFO - [STEP1-CALL2] pf.map_all_pads(board) with instance PF-44205
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - map_all_pads: Already mapped 0 pads
2025-12-22 17:56:43,510 - orthoroute.presentation.plugin.kicad_plugin - INFO - [STEP1-CALL3] pf.route_multiple_nets(board.nets) with instance PF-44205
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - === Route 0 nets ===
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [GPU-THRESHOLD] GPU pathfinding enabled for ROIs with > 1000 nodes
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VERIFY] Parsed 0 tasks from 0 requests
2025-12-22 17:56:43,510 - orthoroute.algorithms.manhattan.unified_pathfinder - INFO - [VERIFY]   routable=0, trivial=0, unmapped=0, total_handled=0
2025-12-22 17:56:43,510 - orthoroute.presentation.plugin.kicad_plugin - ERROR - [STEP1] End-to-end routing failed with instance PF-44205 - no results returned
2025-12-22 17:56:43,510 - root - ERROR - Plugin execution failed
`

The only thing different I did was to use :

cupy-cuda12x                  13.6.0

because CUDA dropped support for the Volta GPUs, and that is all I have right now.

System :
ANSI_COLOR="0;31"
BUG_REPORT_URL="https://issues.redhat.com/"
CPE_NAME="cpe:/o:redhat:enterprise_linux:9::baseos"
cpe:/o:redhat:enterprise_linux:9::baseos
DOCUMENTATION_URL="https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/9"
HOME_URL="https://www.redhat.com/"
ID_LIKE="fedora"
ID="rhel"
LOGO="fedora-logo-icon"
NAME="Red Hat Enterprise Linux"
PLATFORM_ID="platform:el9"
PRETTY_NAME="Red Hat Enterprise Linux 9.7 (Plow)"
REDHAT_BUGZILLA_PRODUCT="Red Hat Enterprise Linux 9"
REDHAT_BUGZILLA_PRODUCT_VERSION=9.7
Red Hat Enterprise Linux release 9.7 (Plow)
REDHAT_SUPPORT_PRODUCT="Red Hat Enterprise Linux"
REDHAT_SUPPORT_PRODUCT_VERSION="9.7"
VERSION="9.7 (Plow)"
VERSION_ID="9.7"
Kernel : Linux 5.14.0-611.5.1.el9_7.x86_64
Date : 2025-12-22 18:07:33
Shell : Zsh 5.8 (PID = 3799027, 3 nests)
Term : /dev/pts/6 (xterm-256color), bauds, 166 x 36 cars
Locale : en_GB.UTF-8

Happy to provide more info/run more tests if that helps.

Thanks a lot,

C/

[bbenchoff]

Hey cedounet

Obviously I've let this issue go for a bit because it's the holidays. But looking at this I don't belive it's because you're using a volta GPU. I have a V100 at home, and I'll test this at the beginning of next week, but my initial instinct is that the V100 isn't the problem.

There's some weird stuff in your log, though:

[VERIFY] Sample pad IDs: []
and map_all_pads: Already mapped 0 pads

That... doesn't seem right, but it looks like you're actually loading a board, with the 296 via layer pairs and the column/row dimensions. You have 40690112 edges, 19 layers, which is a lot but /shrug it should work.

What board are you using for your test? What board is loaded into KiCad when you run the python main.py? Can you try with the TestBackplane included in /TestBoards/?

The only thing I can think of is that you're trying this on a board that's all through-holes instead of pads, which... yeah. I forgot that was possible. I didn't care about through-holes. So now I need to add that.

Try it out with the TestBackplane, let me know what board you're testing this with, and I'll see if I can figure anything out.

[cedounet]

Hello Brian,

Thanks for coming back to me!

Unless I did something stupid, I think I ran the test backplane you provided. At least that is the one I thought I had open in KiCad and have not changed any code (short of enabling debug in settings files)

I can check again next week that I ran what i thought I ran.

Best,

C/

PS: that being said the board I envisaged to use your code will indeed be through holes at source and at destination (high density connectors)…

[bbenchoff]

Ah, yeah, if what you're intending to use this for is all through holes.... in the words of our greatest philosophers, 'well there's your problem'.

Fixing the 'through holes really aren't supported' problem isn't that big of a lift, but I will need to write a bit of code and test it out. Gimme a week or two and I can come up with a board and some basic changes to enable this. I'll ping you here when it's done.

[cedounet]

Sorry I was unclear. Right now I cannot reporoduce the example in GitHub the backplane as you call it.

Through hole is step 2, which now I know won’t work :)

Thanks

Ced