Free memory allocated on the C++ side by MakeADFun.
FreeADFun(obj)An object returned by MakeADFun contains pointers to
structures allocated on the C++ side. These are managed by R's
garbage collector which for the most cases is sufficient. However,
because the garbage collector is unaware of the C++ object sizes,
it may fail to release memory to the system as frequently as
necessary. In such cases one can manually call
FreeADFun(obj) to release the resources.
This function is normally not needed.
Memory allocated on the C++ side by MakeADFun is
represented by external pointers. Each such pointer has an
associated 'finalizer' (see reg.finalizer) that deallocates
the external pointer when gc() decides the pointer is no
longer needed. Deallocated pointers are recognized on the R
side as external null pointers <pointer: (nil)>. This is
important as it provides a way to prevent the finalizers from
freeing pointers that have already been deallocated even if
the deallocation C-code has been unloaded.
The user DLL maintains a list of all external pointers on the C
side. Three events can reduce the list:
Garbage collection of an external pointer that is no longer needed (triggers corresponding finalizer).
Explicit deallocation of external pointers using FreeADFun() (corresponding finalizers are untriggered but harmless).
Unload/reload of the user's DLL deallocates all external pointers (corresponding finalizers are untriggered but harmless).
runExample("simple", thisR = TRUE) ## Create 'obj'
#> Building example simple
#> Note: Using Makevars in /data/user-homes/elizabethb/.R/Makevars
#> using C++ compiler: ‘g++ (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0’
#> Build time 46.115 seconds
#>
#> Running example simple
#>
#> > require(TMB)
#>
#> > dyn.load(dynlib("simple"))
#>
#> > set.seed(123)
#>
#> > y <- rep(1900:2010, each = 2)
#>
#> > year <- factor(y)
#>
#> > quarter <- factor(rep(1:4, length.out = length(year)))
#>
#> > period <- factor((y > mean(y)) + 1)
#>
#> > B <- model.matrix(~year + quarter - 1)
#>
#> > A <- model.matrix(~period - 1)
#>
#> > B <- as(B, "TsparseMatrix")
#>
#> > A <- as(A, "TsparseMatrix")
#>
#> > u <- rnorm(ncol(B))
#>
#> > beta <- rnorm(ncol(A)) * 100
#>
#> > eps <- rnorm(nrow(B), sd = 1)
#>
#> > x <- as.numeric(A %*% beta + B %*% u + eps)
#>
#> > obj <- MakeADFun(data = list(x = x, B = B, A = A),
#> + parameters = list(u = u * 0, beta = beta * 0, logsdu = 1,
#> + logsd0 = 1), random = .... [TRUNCATED]
#>
#> > opt <- nlminb(obj$par, obj$fn, obj$gr)
FreeADFun(obj) ## Free external pointers
#> NULL
obj$fn() ## Re-allocate external pointers
#> [1] 375.0672
#> attr(,"logarithm")
#> [1] TRUE