Welcome to the Parrot Docs#

Parrot is a C++ library for fused array operations using CUDA/Thrust. It provides efficient GPU-accelerated operations with fused evaluation semantics, allowing for chaining of operations without unnecessary intermediate materializations.

Compare examples from the CUDA C++ library Thrust.

Explore the code on GitHub. If you would like to contribute, please refer to the contributing and building guides.

Features#

  • Implicit fusion¹ of array operations

  • GPU acceleration using CUDA/Thrust

  • Chainable operations with a clean API

¹ - Any operation that can be fused is fused.

Quick Start#

#include "parrot.hpp"

int main() {
    // Create a matrix
    auto matrix = parrot::range(10000).as<float>().reshape({100, 100});

    // Calculate the row-wise softmax of a matrix
    auto cols = matrix.ncols();
    auto z    = matrix - matrix.maxr<2>().replicate(cols);
    auto num  = z.exp();
    auto den  = num.sum<2>();
    (num / den.replicate(cols)).print();

    return 0;
}

Performance Comparison of Softmax:

For a full screen version, click here.

Find more examples here.

Contents#