.__                                         .___       
  ____   ____ |__| _________.__.           ____  __ __  __| _/____   
 /    \ /  _ \|  |/  ___<   |  |  ______ _/ ___\|  |  \/ __ |\__  \  
|   |  (  <_> )  |\___ \ \___  | /_____/ \  \___|  |  / /_/ | / __ \_
|___|  /\____/|__/____  >/ ____|          \___  >____/\____ |(____  /
     \/               \/ \/                   \/           \/     \/ 

A high-performance Perlin noise generator developed for the "Accelerated Processing Systems - M" course at University of Bologna, with the professors Stefano Mattoccia and Fabio Tosi.

Features

• Multi-platform: CPU and CUDA GPU acceleration support
• Flexible output: Multiple output formats (PNG, RAW, CSV, PPM)
• Customizable parameters: Adjustable octaves, image size, and seed
• High performance: Optimized noise generation algorithms

Available Backends and Versions

Backend	Available Versions	Description	Dependencies
CPP	v1, v2	It is the naive C++ Implementation	C++ compiler
CUDA	v1, v2, v3	It uses CUDA implementation	CUDA
SIMD	-	(not implemented)	-

Other versions may be added in the future, but currently only the ones listed above are available.

Note: The latest keyword automatically selects the highest available version.

Building

This project supports both CPU and GPU acceleration. CUDA toolkit is required for GPU acceleration, but you can also build for CPU-only execution.

Build Tutorial

# Build CUDA backend
cmake -B build_cuda -DUSE_CUDA=ON -DCUDA_VERSION=v1
cmake --build build_cuda

# Build SIMD backend version v1
cmake -B build_simd -DUSE_SIMD=ON -DSIMD_VERSION=v1
cmake --build build_simd

# Build C++ backend (CPU only)
cmake -B build_cpp -DUSE_CPP=ON -DCPP_VERSION=v1
cmake --build build_cpp

Tip: Use separate build directories per backend to avoid conflicts and unnecessary recompilation.

Command-Line Options

Flag	Long Option	Argument	Default	Description
-h	--help	none	—	Show this help message and exit.
—	--version	none	—	Show program version and exit.
-o	--output	<filename>	perlin.<ext>	Output filename. Extension inferred from format.
-f	--format	<string>	png	Output format. Supported: png, raw, csv, ppm.
-s	--size	<WxH>	2048x2048	Output size in pixels (width x height).
-v	--verbose	none	false	Print processing steps and timings.
-n	--no-output	none	false	Disable output file generation.
-F	--frequency	<float>	1.0	Base frequency (scale factor).
-A	--amplitude	<float>	1.0	Base amplitude.
-L	--lacunarity	<float>	2.0	Frequency multiplier per octave.
-P	--persistence	<float>	0.5	Amplitude multiplier per octave.
-O	--offset	<x,y>	0,0	Offset applied to noise coordinates.
-C	--octaves	<int>	1	Number of octaves (>= 1).
positional	(seed)	<uint64>	—	Positional unsigned integer seed (e.g. ./perlin 13813).
-b	--benchmark	none	false	Generate CSV benchmark data.

Usage: ./perlin [seed] [OPTIONS]

Output formats

• PNG: Portable Network Graphics (lossless compression)
• RAW: Raw binary data
• CSV: Comma-separated values

Benchmarking notes

If the --benchmark flag is set, the program will output benchmark data in CSV format to standard output, that includes:

timestamp,width,height,pixels,octaves,frequency,wall_ms,cpu_s,ms_per_pixel,mem_bytes

Let's test it!

• All the output .csv files is in the directory /tests/benchmarks/outputs/csv.

• The Nsight Compute outputs are in /tests/benchmarks/outputs/cuda.

Note: Ensure you have built the backend to test before running the benchmark tests. If you have compiled the project in a different build directory, adjust the script accordingly.

C++: v1 vs. v2

To run the C++ benchmark test in v1 and v2, use the following command:

./tests/benchmarks/cpp_benchmark.sh

Cuda: GPU timer

Incremental size of matrix, grouped by incremental number of octaves:

# Example: Test CUDA version 3
./tests/benchmarks/cuda_benchmark.sh 3

Cuda: Nsight compute profiling

You need NVIDIA Nsight Compute, typically installed with the CUDA Toolkit. Installation (Arch Linux example): sudo pacman -S nsight-compute

The profiling script uses sudo due to the need to access low-level GPU hardware counters. This is strongly discouraged for security reasons. Instead, ensure your user has the necessary permissions to access the device for profiling without sudo.

Use the provided script, supplying the integer version of the CUDA code to be tested.

# Example: Profile CUDA version 3
./tests/start-cuda-profiling-test.sh 3

Plotting

• All the plotting Python scripts are in the directory /tests/benchmarks/plots. I was bored documenting them, feel free to explore and expand ;)

Examples

Below are four example terrains generated with Perlin noise using seed 1234.
The output images are located in docs/examples/png/.

Freq: 5.0 Amp: 1.0 Lac: 2.0 Pers: 0.5 Octaves: 1

Freq: 5.0 Amp: 1.0 Lac: 2.0 Pers: 0.5 Octaves: 3

Freq: 10.0 Amp: 1.0 Lac: 2.0 Pers: 0.5 Octaves: 6

Freq: 10.0 Amp: 2.0 Lac: 2.0 Pers: 0.5 Octaves: 6

License

This project is released under the Apache License 2.0. See LICENSE file for details.