📑 in case u need official docs:
- https://mc-stan.org/docs/cmdstan-guide/cmdstan-installation.html
- https://github.com/bstatcomp/stan_gpu_install_docs/blob/master/install_docs/INSTALL_CMDSTAN.md
- https://blog.mc-stan.org/2022/08/03/options-for-improving-stan-sampling-speed/
prepare at least 2 GiB disk space
tested version: CmdStan v2.36.0: https://github.com/stan-dev/cmdstan/releases/download/v2.36.0/cmdstan-2.36.0.tar.gz
untar the file then create a file named local inside the make folder with:
STAN_CPP_OPTIMS=true
CXXFLAGS+=-mtune=native -march=native
CFLAGS+=-mtune=native -march=native
CXXFLAGS_OPTIM_SUNDIALS+=-mtune=native -march=native
CPPFLAGS_OPTIM_SUNDIALS+=-mtune=native -march=native
CXXFLAGS_OPTIM_TBB+=-mtune=native -march=native
STAN_THREADS=true
STAN_OPENCL=true
OPENCL_PLATFORM_ID=0
OPENCL_DEVICE_ID=0
LDFLAGS_OPENCL=-L"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6/lib/x64" -lOpenCL
need Rtools Msys2 console setup with mingw32-make + g++
mingw32-make -j build
additional step: export PATH=$PATH:███/cmdstan-2.36.0/stan/lib/stan_math/lib/tbb
R script for installation
install.packages("cmdstanr", repos = c("https://stan-dev.r-universe.dev", getOption("repos")))
library(cmdstanr)
check_cmdstan_toolchain()
set_cmdstan_path("<PATH_TO_INSTALL_CMDSTAN>")
cmdstan_make_local(cpp_options = list(
"STAN_CPP_OPTIMS" = TRUE,
"CXXFLAGS+=-mtune=native -march=native",
"CFLAGS+=-mtune=native -march=native",
"CXXFLAGS_OPTIM_SUNDIALS+=-mtune=native -march=native",
"CPPFLAGS_OPTIM_SUNDIALS+=-mtune=native -march=native",
"CXXFLAGS_OPTIM_TBB+=-mtune=native -march=native",
"STAN_THREADS" = TRUE,
"STAN_OPENCL" = TRUE,
"OPENCL_PLATFORM_ID" = 0,
"OPENCL_DEVICE_ID" = 0,
"LDFLAGS_OPENCL" = sprintf('-L"%s/lib/x64" -lOpenCL', gsub("\\", "/", Sys.getenv("CUDA_PATH"), fixed = TRUE))
))
install_cmdstan(cores = parallel::detectCores(logical = FALSE), overwrite = TRUE)run test
get my test files and put in the examples folder:
- stan file: https://github.com/phineas-pta/Bayesian-Methods-for-Hackers-using-PyStan/blob/main/stan%20files/chap2ex2.stan
- data file: https://github.com/phineas-pta/Bayesian-Methods-for-Hackers-using-PyStan/blob/main/stan%20files/chap2ex2.data.json
test:
mingw32-make -j examples/chap2ex2.exe
examples/chap2ex2.exe optimize data file=examples/chap2ex2.data.json output file=examples/chap2ex2_optim.csv
examples/chap2ex2.exe sample num_chains=4 num_samples=50000 num_warmup=10000 thin=5 data file=examples/chap2ex2.data.json output file=examples/chap2ex2_fit.csv diagnostic_file=examples/chap2ex2_dia.csv refresh=5000 num_threads=4
bin/stansummary.exe examples/chap2ex2_fit_*.csv
examples/chap2ex2.exe variational data file=examples/chap2ex2.data.json output file=examples/chap2ex2_advi.csv
examples/chap2ex2.exe diagnose data file=examples/chap2ex2.data.json output file=examples/chap2ex2_diag.csv