Code Noodling

Prime Generation + Physical Dice Simulation Suite

An experimental C/C++/CUDA project exploring:

• Optimized prime sieving (CPU + multi-GPU CUDA)
• Accurate dice randomness for tabletop and RPG mechanics
• Physically simulated dice rolls using NVIDIA PhysX
• Auto-tuned performance across CPU/GPU hardware
• Live visualization with the PhysX Visual Debugger (PVD)

---

Features

Component	Description
OSE / CUDA Sieve	Optimized Sieve of Eratosthenes segmented, odd-only, GPU-parallel, JSON export
Dice Engine (CPU)	Pure math engine for NdM+K expressions unbiased, optional prime-seeded randomness, chi-square validation
PhysX Dice Simulator	Fully physical D6, D8, D12, D20 dice with realistic collisions, ramp chute, multi-threading, and GPU acceleration
Visualization	Optional live streaming via --pvd to PhysX Visual Debugger or NVIDIA Omniverse View
Data Outputs	JSON and CSV result logging, with chi-square test of fairness and probability distribution graphs

---

Build Instructions

Prerequisites

• CMake 3.20
• C++17 compiler (clang / gcc / MSVC)
• (Optional) CUDA Toolkit (for GPU sieve)
• (Optional) NVIDIA PhysX SDK (for physics dice)
• (Optional) PhysX Visual Debugger (PVD) to visualize rolls live

Build All Targets

mkdir build && cd build
cmake .. -DBUILD_CUDA_SIEVE_MGPU=ON -DBUILD_PHYSX_DICE=ON
cmake --build . -j

Tesla K80 users: CUDA architecture defaults to sm_37.
Override with:
-DCMAKE_CUDA_ARCHITECTURES=37 (or 80 for A100s, etc.)

---

CPU Dice Engine

# Fair D6, 10k rolls, chi-square test
./dice_cpu --faces 6 --count 10000 --chi

# D20, 20k rolls, prime-seeded, CSV output
./dice_cpu --faces 20 --count 20000 --use-prime-seeds primes_50M.json --csv d20.csv --chi

# RPG Expression Example
./dice_cpu --spec "3d6+2" --count 5000 --use-prime-seeds primes_50M.json --log-json rolls.json --chi

Flags

• --spec NdM+K Dice expression (can repeat)
• --count N Number of rolls per set
• --chi Enable chi-square fairness check
• --csv / --log-json โ�� Export results

---

CUDA Multi-GPU Prime Generator

./cuda_sieve_mgpu 50000000 --gpus 4 --seg 256M --json primes_50M.json

Generates all primes 50 M using up to four Tesla K80 GPUs, writing them to primes_50M.json.

---

PhysX Dice Simulator (D6/D8/D12/D20)

# Basic D6 simulation with 50k rolls and chi-square test
./physx_dice_multi --spec 1d6 --trials 50000 --chi

# Complex run with multiple dice, chute ramp, PVD visualization, and JSON/CSV output
./physx_dice_multi   --spec 3d6+2 --spec 1d8 --spec 1d12 --spec 1d20   --trials 20000   --use-prime-seeds primes_50M.json   --chute   --pvd 127.0.0.1:5425   --json physx_runs.json --csv physx_counts.csv --chi

Flags

• --chute Adds an angled ramp, side walls, and backstop for realism
• --pvd [host:port] Stream live to PhysX Visual Debugger (default: 127.0.0.1 : 5425)
• --use-prime-seeds Deterministic seed list from JSON primes
• --json / --csv Write results to disk

Make sure PVD is running before launching your sim to see live dice tumbling!

---

##��� Repository Layout

code-noodling/
��CMakeLists.txt
��.gitignore
��cuda_sieve_mgpu.cu
dice_cpu.cpp
��physx_dice_multi.cpp
��die_mesh.h
��OSE.c                      (optional legacy)
OSE_CUDA.cc                (optional legacy)
��README.md
��primes_50M.json            (generated at runtime)

---

Technical Notes

• Hardware autodetect: automatically uses all CPU threads and available CUDA devices.
• Deterministic seeding: optional uses primes for stable pseudo-random sequences.
• PhysX cooking: convex meshes generated internally for D8, D12, D20 via die_mesh.h.
• Chute geometry: procedural ramp (30° tilt) + sidewalls + backstop.
• Visualization: controlled via --pvd; debug draw enabled for collision shapes & contacts.
• Data outputs: JSON + CSV + chi-square test summary.

---

Example Workflow

# Step 1 โ�� Generate prime seeds
./cuda_sieve_mgpu 50000000 --json primes_50M.json

# Step 2 โ�� Simulate physical dice using those seeds
./physx_dice_multi --spec 2d6+3 --trials 10000 --use-prime-seeds primes_50M.json --chute --pvd --chi

---

Author

Donovan Worrell
Seattle, WA, USA
donovan.worrell@gmail.com

---

License

MIT License ยฉ 2025 Donovan Worrell
Permission is granted to use, modify, and distribute this software with attribution.