knygren.r-universe.dev

[knygren] nmathopencl 0.2.0

kjell.a.nygren@gmail.com (Kjell Nygren) — Thu, 04 Jun 2026 21:25:58 GMT

Ships statistical and mathematical routines from R internal 'nmath' ('Mathlib') as 'OpenCL' C sources under directory 'inst/cl/', with R wrappers that use the GPU when 'OpenCL' is available at compile time and fall back to 'stats' equivalents otherwise. Aimed at package developers building custom kernels (for example Bayesian GLMs via suggested package 'glmbayes') using opencltools kernel loaders and related helpers. Contains translated shims, an illustrative GLM-related kernel subsystem, vignettes, and optional GPU acceleration.

[knygren] glmbayes 0.9.6

kjell.a.nygren@gmail.com (Kjell Nygren) — Thu, 04 Jun 2026 02:16:39 GMT

Provides Bayesian linear and generalized linear model fitting with independent and identically distributed (iid) posterior samples. The main functions mirror R's lm() and glm() interfaces while adding prior family specifications for Gaussian, Poisson, binomial, and Gamma models with log-concave likelihoods. Sampling for supported non-conjugate models uses accept-reject methods based on likelihood subgradients as in Nygren and Nygren (2006) . The package also includes tools for prior setup, posterior summaries, prediction, diagnostics, simulation, vignettes, and optional 'OpenCL' acceleration for larger models.

[knygren] opencltools 0.8.1

kjell.a.nygren@gmail.com (Kjell Nygren) — Wed, 03 Jun 2026 23:00:15 GMT

Runtime 'OpenCL' support for R package developers: probe hardware and drivers, load and concatenate kernel sources, and manage dependency-annotated '.cl' libraries, so packages like 'nmathopencl' and other ported libraries can offer GPU acceleration without each re-implementing the same plumbing. Vignettes use the 'glmbayes' envelope-gradient example and likelihood subgradient methodology (Nygren and Nygren, 2006, ).