Sitta

A lightweight, dynamic multi-dimensional array / view for C++20. Pure container: shape, strides, memory ownership and zero-copy views, nothing else. Optional, opt-in interop with xtensor, Kokkos, HDF5, TBB, Boost.MultiArray, and Python.

#include <sitta/sitta.hpp>
using namespace sitta;

Installation

Header-only with zero dependencies for the core functionality.

cmake -B build
cmake --build build
ctest --test-dir build

Interop features (xtensor, Kokkos, HDF5, TBB, Boost.MultiArray, Python) are auto-detected and built only if their dependencies are found; pass -DSITTA_ENABLE_<NAME>=OFF to skip one even when present.

Usage example

Two flavors: static rank (SittaArray<T, N>, rank known at compile time, tightest codegen) and dynamic rank (SittaArray<T>, rank known at runtime). Both share the same API.

// Owning allocation, static rank 2.
auto a = SittaArray<double, 2>::from_shape({3, 4});
a(1, 2) = 3.14;

// Views share the buffer (refcounted): no copy, no allocation.
auto t = a.transpose();                                        // shape (4, 3)
auto s = a.slice(std::array<Range,2>{Range{0, 2, 1}, Range{1, 3, 1}});
s(0, 0) = 42;   // writes through to `a`

// Wrap external memory without taking ownership.
std::vector<float> buf(6);
auto w = SittaArray<float, 2>::wrap(buf.data(), {2, 3});

See the Getting Started guide for the full walkthrough: element access, reshape/squeeze, the rank-reducing drop_axis/subview, static/dynamic rank conversion, inspecting an array, and the element-wise for_each/transform algorithms.

Interop

• xtensor: zero-copy view for host math (sitta::xt_interop::as_xtensor)
• Kokkos: wraps a Kokkos::View's refcount for host/device buffers
• HDF5: read/write full datasets or hyperslabs, including user structs
• TBB: parallel for_each/transform
• Boost.MultiArray: zero-copy or copying conversion, static rank only
• Python: nanobind module with NumPy and DLPack zero-copy interop

Full usage for each adapter is in docs/interop.rst.

What this class is not

No arithmetic, no broadcasting, no iterators, no I/O beyond the HDF5 adapter, no automatic host/device mirroring. It's a metadata + buffer-lifetime layer: build numeric operations on top via the xtensor/Kokkos/Python adapters.

Documentation

Full API reference (generated from source via clang-doc) and detailed guides build with Sphinx:

cmake -B build -DSITTA_ENABLE_DOCS=ON
cmake --build build --target docs

See docs/README.md for the doc-build pipeline, or browse the .rst sources directly: docs/getting_started.rst, docs/interop.rst.

License

MIT. See LICENSE.