Skip to contents

Time the compile step on its own

Source: R/benchmark.R
benchmark_compile.Rd

No fit records how long it took to compile, so a benchmark times a compile of its own. force = TRUE is what makes the number a compile rather than a cache hit: under cmdstanr a second call to the same code reuses the executable in 0.02 s against 4.7 s for a compile (measured 2026-10-05), while rstan recompiles every time (23.3 s).

Usage

benchmark_compile(
  code,
  backend = "cmdstanr",
  force = TRUE,
  ...,
  .compiler = NULL
)

Arguments

code

Stan code, as bmm::stancode() or brms::make_stancode() returns it.

backend

"cmdstanr" or "rstan".

force

Recompile even when an executable is cached. Leaving it TRUE is the point; FALSE measures a reuse.

...

Passed to the backend's compiler.

.compiler

The compiling function, the backend's by default. Tests inject a stand-in so that nothing is compiled.

Value

A one-row tibble with compile_seconds, backend, forced, code_hash and exe.

See also

benchmark(), benchmark_metrics()

Examples

if (FALSE) { # \dontrun{
code <- bmm::stancode(formula, data, model, backend = "cmdstanr")
benchmark_compile(code, backend = "cmdstanr")
} # }