KaMPIng-v2

Modern C++ MPI bindings: pass your containers, compose what you need.

Calling MPI from C++ means passing raw pointers, element counts, datatypes and displacement arrays by hand, and getting one of them wrong usually crashes or hangs. KaMPIng-v2 lets you pass the objects you already have, such as a std::vector, a std::span or a Kokkos view, and works out the pointer, count and datatype for you. When you need more, you add it with a | pipe in the style of std::ranges views: a receive buffer that resizes itself, per-rank counts that are exchanged for you, or a custom datatype.

v2::send(data, dest, comm);
v2::recv(buf | views::resize, source, comm);
auto [_, everything] = v2::allgatherv(mine, v2::auto_recv_v<int>(), comm);

KaMPIng-v2 is header-only and requires C++20. New here? Start with Getting started.

How it is built

KaMPIng-v2 is organized in layers on top of the MPI standard:

  • mpi-core (mpi::experimental) is a minimal C++ interface that expresses MPI’s own abstractions — data buffers, MPI objects and their handles — as C++20 concepts. It mirrors the MPI standard closely: one MPI call per function, MPI semantics unchanged, no hidden allocation or communication. It is designed as a reference for standardized C++ bindings, and it is the extension point for everything above it.

  • KaMPIng-v2 (kamping::v2) builds the convenience on top: composable views that attach MPI metadata to any buffer, receive buffers that size themselves, memory-safe non-blocking operations, and defaults for omitted arguments. It re-exports the names you need from mpi-core, so you rarely have to name mpi::experimental.

  • Ecosystem adapters make Kokkos views, Thrust and SYCL buffers and Cereal-serialized objects buffers, usable with every operation.

This documentation covers the kamping::v2 interface. Throughout, it follows the principle of you only pay for what you use: convenience that costs something, such as a probe or a count exchange, only happens where you ask for it at the call site.

KaMPIng-v2 also comes with dstl, a library of distributed algorithms and communication patterns built on it, such as sparse all-to-all exchanges and redistribution.

Guide

Reference

dstl

About