| Title: | Minimal Rust SIMD Multiversioning Example for R |
|---|---|
| Description: | Template R package with minimal setup to use Rust code in R without hacks or frameworks, plus a Rust SIMD example selected by C runtime dispatch. Includes basic examples of importing cargo dependencies, spawning threads and passing numbers or strings from Rust to R. The SIMD demo keeps scalar and optional SSE2, AVX2, AVX-512, or NEON kernels in one Rust static library and lets C choose the best Rust function using Rust-reported compiled and CPU features. Cargo crates are automatically 'vendored' in the R source package to support offline installation. This project was first presented at 'Erum2018' to showcase R-Rust integration <https://jeroen.github.io/erum2018/>; for a real world use-case, see the 'gifski' package on 'CRAN'. |
| Authors: | Sounkou Toure [aut, cre], Jeroen Ooms [cph] (ORCID: <https://orcid.org/0000-0002-4035-0289>), Authors of the dependency Rust crates [aut] (see AUTHORS file) |
| Maintainer: | Sounkou Toure <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 1.3.0 |
| Built: | 2026-05-28 18:08:15 UTC |
| Source: | https://github.com/sounkou-bioinfo/hellorust-multiversion |
Minimal examples of calling rust functions in R via C.
hello() random() runthreads()hello() random() runthreads()
These functions call out to rust functions defined in the myrustlib cargo
crate which is embedded in this package. They return values generated in Rust,
such as a UTF-8 string or random number. In addition, runthreads is an
example of a multi-threaded rust function.
a value generated in Rust (a string, random number, and NULL respectively).
hello() random() runthreads()hello() random() runthreads()
This is a Rust/C/R multiversioning demo. The SIMD kernels and CPU feature probes live in Rust, while the C layer only caches a function pointer and dispatches to the best Rust backend whose compiled feature and runtime CPU feature are both available. The package still builds one R shared library; it does not create one shared library per SIMD backend.
"auto" selects the fastest available Rust backend from the intersection of
compiled features and runtime CPU features. "scalar" is always available and
is useful for correctness checks and scalar-vs-SIMD benchmarks.
simd_count_nonzero(x) simd_set_backend( backend = c("auto", "scalar", "sse2", "avx2", "avx512", "neon") ) simd_info()simd_count_nonzero(x) simd_set_backend( backend = c("auto", "scalar", "sse2", "avx2", "avx512", "neon") ) simd_info()
x |
A raw vector. |
backend |
One of |
The number of non-zero bytes in x.
The selected backend name, invisibly.
A named list with character-vector compiled and supported
backend sets, the runtime selected backend, and logical fields for
Rust-reported compiled and CPU features.
simd_count_nonzero(as.raw(c(0, 1, 2, 0, 3))) simd_set_backend("scalar") simd_set_backend("auto") simd_info()simd_count_nonzero(as.raw(c(0, 1, 2, 0, 3))) simd_set_backend("scalar") simd_set_backend("auto") simd_info()