curaster 0.4.0

CUDA-accelerated Raster Algebra Calculator

cuRaster

cuRaster is a high-performance Python module written in C++ and CUDA. It is designed to compute general raster algebra expressions on massive raster datasets using GPU acceleration.

Installation

Install the precompiled wheel via pip:

pip install curaster

Requirements (for building from source)

To build and run this module from source, your system needs:

• NVIDIA GPU with CUDA Toolkit installed (nvcc)
• GDAL C++ library (libgdal-dev)
• CMake 3.18+
• pybind11 (pip install pybind11)
• A C++20-compatible compiler

Usage

Once installed, simply import it into Python:

import curaster

curaster.compute(
    input_file="input.tif",
    output_file="output.tif",
    expression="(B4-B1)/(B4+B1)",
    verbose=True
)

API Reference

curaster.compute(...)

Computes raster algebra from a string expression on the GPU and saves the result to a new single-band Float32 GeoTIFF.

Argument	Type	Default	Description
input_file	str	Required	Path to the input multispectral raster
output_file	str	Required	Path where the output raster will be saved
expression	str	Required	The algebraic formula (e.g., "(B4-B1)/(B4+B1)"). Bands are referenced as B1, B2, etc. (1-indexed)
verbose	bool	True	If true, prints memory telemetry and a progress bar to the terminal

Building from Source

To build from source using CMake:

mkdir build
cd build
cmake .. -Dpybind11_DIR=$(python -c "import pybind11; print(pybind11.get_cmake_dir())")
make

Alternatively, you can compile directly with nvcc:

nvcc -O3 -shared -std=c++20 -Xcompiler -fPIC -Xcompiler -fopenmp --use_fast_math \
    -o curaster$(python3-config --extension-suffix) src/raster.cu \
    $(python3 -m pybind11 --includes) \
    $(gdal-config --cflags) $(gdal-config --libs)

License

See LICENSE file for details.