cgvR

Interactive 3D visualization of large Cayley and state-space graphs via Vulkan.

https://github.com/user-attachments/assets/9af70289-d865-44db-b0a0-e67269fdd120

Overview

Cayley graphs of permutation puzzles (TopSpin, Rubik’s cube, etc.) and game state graphs can have millions of nodes — far too many to render at once. cgvR provides GPU-accelerated 3D rendering on top of the Datoviz Vulkan engine, plus force-directed layout and camera/path animation utilities for exploring such graphs interactively.

Key features

Vulkan rendering of nodes (points) and edges (segments), driven by Datoviz.
Layout algorithms: Fruchterman-Reingold in pure R (cgv_layout_fr) and a Barnes-Hut O(n log n) version in C (cgv_layout_fr_bh) for large graphs.
Camera control: fly (WASD + mouse), orbit, arcball; cgv_fly_to to a node and cgv_fly_path along a Catmull-Rom spline of waypoints.
Path highlighting with custom colors, node scale and edge width.
Video recording via ffmpeg pipe (cgv_record_start / cgv_record_stop).
Headless mode (cgv_viewer(..., offscreen = TRUE)) for tests and scripted offline rendering.

Dependencies

Package	Role
cayleyR	Optional — Cayley graph construction, BFS, pathfinding

System requirements

Supported operating systems (matches Datoviz upstream):

Linux x86_64 — Ubuntu 22.04 or later (glibc 2.34+).
macOS 12 or later — x86_64 and arm64 (Apple silicon M1–M4).
Windows 10 or later — x86_64.

Vulkan support is auto-detected at install time. If the dependencies are missing, the package falls back to a stub build (see below).

Ubuntu / Debian — to enable full rendering:

sudo apt install libvulkan-dev libglfw3-dev pkg-config build-essential \
                 mesa-vulkan-drivers
# Optional, only for cgv_record_*:
sudo apt install ffmpeg

Shaders and cglm headers are bundled — no need for glslc or cmake. Tested on Ubuntu 22.04+ (matches upstream Datoviz: glibc 2.34+).

Windows — install Rtools and the LunarG Vulkan SDK. Make sure the VULKAN_SDK environment variable is set before R CMD INSTALL. Internet access during install is required — configure.win downloads the prebuilt datoviz.dll from GitHub releases (cached in inst/lib/ afterwards). Optional: ffmpeg.exe on PATH for cgv_record_*.

macOS 12 or later (x86_64 and arm64) — install dependencies via Homebrew plus the LunarG Vulkan SDK (which ships MoltenVK):

brew install glfw pkg-config
# Then install LunarG Vulkan SDK from
#   https://vulkan.lunarg.com/sdk/home#mac
# and 'export VULKAN_SDK=...' before R CMD INSTALL.

configure auto-detects Darwin (sets OS_MACOS=1, links against the system MoltenVK loader, applies -mmacosx-version-min=12.0). GLFW pulls the required Cocoa / IOKit / CoreFoundation frameworks via its own pkg-config --libs glfw3.

A working Vulkan GPU driver is required at runtime regardless of platform (Mesa / NVIDIA / AMD on Linux, vendor driver on Windows, MoltenVK on macOS).

Build options

Force or skip the native build:

install.packages("cgvR", configure.args = "--with-vulkan")     # require Vulkan; error if missing
install.packages("cgvR", configure.args = "--without-vulkan")  # always stub build

Enable SIMD acceleration for fpng PNG screenshots (SSE4.1 + PCLMUL on x86):

install.packages("cgvR", configure.args = "--with-simd")

Combine flags as needed:

install.packages("cgvR", configure.args = "--with-vulkan --with-simd")

Installation

# install.packages("remotes")
remotes::install_github("Zabis13/cgvR")

# Optional: enable SIMD acceleration for fpng (PNG screenshots).
# Disabled by default for portability. Requires SSE4.1 + PCLMUL.
remotes::install_github("Zabis13/cgvR", configure.args = "--with-simd")

Quick start

library(cgvR)

# Build a small graph
n <- 100L
edges <- cbind(sample.int(n, 200, replace = TRUE),
               sample.int(n, 200, replace = TRUE))

# Force-directed 3D layout
pos <- cgv_layout_fr(n, edges, n_iter = 200L, seed = 1L)

# Open a viewer and upload the graph
v <- cgv_viewer(1280, 720, "cgvR demo")
cgv_background(v, "black")
cgv_set_graph(v, 1:n, edges,
              positions   = pos,
              node_values = as.double(seq_len(n)),
              node_sizes  = rep(8, n))

# Highlight a path
cgv_highlight_path(v, c(1, 5, 17, 42), color = "#FF2200",
                   node_scale = 1.6, edge_width = 3.0)

# Camera + run loop (right-click drag = rotate, scroll = zoom)
cgv_camera(v, position = c(20, 16, 24), target = c(0, 0, 0))
cgv_run(v)

More examples in inst/examples/:

demo_small_graph.R — random 100-node graph with FR layout.
demo_cycles_bh.R — TopSpin cycles, Barnes-Hut FR layout.
demo_tictactoe.R — Tic-Tac-Toe game graph, color = move number.
demo_record.R, demo_cycles_bh_record.R — video recording.

Architecture

R API  →  .Call  →  C layer  →  Datoviz (Vulkan visuals)
                                  ↕
                              cayleyR (optional graph ops)

Datoviz, cglm and GLFW are compiled from sources bundled in src/ and linked statically into cgvR.so. The only external runtime dependency is libvulkan.so.

Stub build (no Vulkan)

If Vulkan or GLFW are missing at install time (or you pass --without-vulkan), cgvR falls back to a stub build: only a tiny C file is compiled, the package installs cleanly, and all rendering APIs (cgv_viewer, cgv_run, …) raise an informative error when called. Pure-R helpers like cgv_layout_fr() keep working. Use cgv_is_stub_build() to detect this mode at runtime.

License

MIT