From a48d1fd9fbd52aaf082152077ac3e8e3914c1e47 Mon Sep 17 00:00:00 2001 From: Rich FitzJohn Date: Mon, 14 Oct 2024 09:01:53 +0100 Subject: [PATCH] Drop reference to dust1 --- vignettes/design.Rmd | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) diff --git a/vignettes/design.Rmd b/vignettes/design.Rmd index 6c40741f..67ffb7a2 100644 --- a/vignettes/design.Rmd +++ b/vignettes/design.Rmd @@ -74,11 +74,7 @@ To initialise our system with a potentially very large number of particles we ta * First, we seed the first stream using the `splitmix64` RNG, following the xoshiro docs. This expands a single 64-bit integer into the 256-bits of RNG state, while ensuring that the resulting full random number seed does not contain all zeros. * Then, for each subsequent chain we take a "jump" in the sequence. This is a special move implemented by the RNG that is equivalent to a very large number of draws from the generator (e.g., about 2^128 for the default generator used for double-precision models) ensuring that each particles state occupies a non-overlapping section of the underling random number stream (see `vignette("rng")` for details). -With this setup we are free to parallelise the system as each realisation is completely independent of each other; the problem has become "[embarrassingly parallel](https://en.wikipedia.org/wiki/Embarrassingly_parallel)". In practice we do this using [OpenMP](https://www.openmp.org/) where available as this is well supported from R and gracefully falls back on serial operation where not available. See `dust::dust_openmp_support` for information on your system's OpenMP configuration as seen by R: - -```{r} -dust::dust_openmp_support() -``` +With this setup we are free to parallelise the system as each realisation is completely independent of each other; the problem has become "[embarrassingly parallel](https://en.wikipedia.org/wiki/Embarrassingly_parallel)". In practice we do this using [OpenMP](https://www.openmp.org/) where available as this is well supported from R and gracefully falls back on serial operation where not available. As the number of threads changes, the results will not change; the same calculations will be carried out and the same random numbers drawn.