The R Project SVN R

% File src/library/compiler/man/compile.Rd
% Part of the R package, https://www.R-project.org
% Copyright 2011-2018 R Core Team
% Distributed under GPL 2 or later

\name{compile}
\alias{compile}
\alias{cmpfun}
\alias{cmpfile}
\alias{loadcmp}
\alias{disassemble}
\alias{enableJIT}
\alias{compilePKGS}
\alias{getCompilerOption}
\alias{setCompilerOptions}
\title{Byte Code Compiler}
\usage{
cmpfun(f, options = NULL)
compile(e, env = .GlobalEnv, options = NULL, srcref = NULL)
cmpfile(infile, outfile, ascii = FALSE, env = .GlobalEnv,
        verbose = FALSE, options = NULL, version = NULL)
loadcmp(file, envir = .GlobalEnv, chdir = FALSE)
disassemble(code)
enableJIT(level)
compilePKGS(enable)
getCompilerOption(name, options)
setCompilerOptions(...)
}
\arguments{
  \item{f}{a closure.}
  \item{options}{list of named compiler options: see \sQuote{Details}.}
  \item{env}{the top level environment for the compiling.}
  \item{srcref}{initial source reference for the expression.}
  \item{file,infile,outfile}{pathnames; outfile defaults to infile with
    a \file{.Rc} extension in place of any existing extension.}
  \item{ascii}{logical; should the compiled file be saved in ascii format?}
  \item{verbose}{logical; should the compiler show what is being compiled?}
  \item{version}{the workspace format version to use.  \code{NULL}
    specifies the current default format (3).  Version 1 was the default
    from \R 0.99.0 to \R 1.3.1 and version 2 from \R 1.4.0 to 3.5.0.
    Version 3 is supported from \R 3.5.0.}
  \item{envir}{environment to evaluate loaded expressions in.}
  \item{chdir}{logical; change directory before evaluation?}
  \item{code}{byte code expression or compiled closure}
  \item{e}{expression to compile.}
  \item{level}{integer; the JIT level to use (\code{0} to \code{3}, or
    negative to \emph{return} it).}
  \item{enable}{logical; enable compiling packages if \code{TRUE}.}
  \item{name}{character string; name of option to return.}
  \item{...}{named compiler options to set.}
}
\description{
  These functions provide an interface to a byte code compiler for \R.
}
\details{
  The function \code{cmpfun} compiles the body of a closure and
  returns a new closure with the same formals and the body replaced by
  the compiled body expression.

  \code{compile} compiles an expression into a byte code object; the
  object can then be evaluated with \code{eval}.

  \code{cmpfile} parses the expressions in \code{infile}, compiles
  them, and writes the compiled expressions to \code{outfile}.  If
  \code{outfile} is not provided, it is formed from \code{infile} by
  replacing or appending a \code{.Rc} suffix.

  \code{loadcmp} is used to load compiled files.  It is similar to
  \code{sys.source}, except that its default loading environment is the
  global environment rather than the base environment.

  \code{disassemble} produces a printed representation of the code
  that may be useful to give a hint of what is going on.

  \code{enableJIT} enables or disables just-in-time (JIT)
  compilation. JIT is disabled if the argument is 0.  If \code{level} is
  1 then larger closures are compiled before their first use.  If
  \code{level} is 2, then some small closures are also compiled before
  their second use.  If \code{level} is 3 then in addition
  all top level loops are compiled before they are executed.  JIT level
  3 requires the compiler option \code{optimize} to be 2 or 3.  The JIT
  level can also be selected by starting \R with the environment
  variable \code{R_ENABLE_JIT} set to one of these values. Calling
  \code{enableJIT} with a negative argument returns the current JIT
  level. The default JIT level is \code{3}.

  \code{compilePKGS} enables or disables compiling packages when they
  are installed.  This requires that the package uses lazy loading as
  compilation occurs as functions are written to the lazy loading data
  base. This can also be enabled by starting \R with the environment
  variable \code{_R_COMPILE_PKGS_} set to a positive integer value.
  This should not be enabled outside package installation, because it
  causes any serialized function to be compiled, which comes with
  time and space overhead. \code{R_COMPILE_PKGS} can be used, instead,
  to instruct \code{INSTALL} to enable/disable compilation of packages
  during installation.

  Currently the compiler warns about a variety of things.  It does
  this by using \code{cat} to print messages.  Eventually this should
  use the condition handling mechanism.

  The \code{options} argument can be used to control compiler operation. 
  There are currently four options: \code{optimize}, \code{suppressAll},
  \code{suppressUndefined}, and \code{suppressNoSuperAssignVar}. 
  \code{optimize} specifies the optimization level, an integer from \code{0}
  to \code{3} (the current out-of-the-box default is \code{2}). 
  \code{suppressAll} should be a scalar logical; if \code{TRUE} no messages
  will be shown (this is the default). \code{suppressUndefined} can be
  \code{TRUE} to suppress all messages about undefined variables, or it can
  be a character vector of the names of variables for which messages should
  not be shown.  \code{suppressNoSuperAssignVar} can be \code{TRUE} to
  suppress messages about super assignments to a variable for which no
  binding is visible at compile time.  During compilation of packages,
  \code{suppressAll} is currently \code{FALSE}, \code{suppressUndefined} is
  \code{TRUE} and \code{suppressNoSuperAssignVar} is \code{TRUE}.

  \code{getCompilerOption} returns the value of the specified option.
  The default value is returned unless a value is supplied in the
  \code{options} argument; the \code{options} argument is primarily for
  internal use.  \code{setCompilerOption} sets the default option
  values. Options to set are identified by argument names, e.g.
  \code{setCompilerOptions(suppressAll = TRUE, optimize = 3)}.
  It returns a named list of the previous values.

  Calling the compiler a byte code compiler is actually a bit of a
  misnomer: the external representation of code objects currently uses
  \code{int} operands, and when compiled with \code{gcc} the internal
  representation is actually threaded code rather than byte code.
}
\author{Luke Tierney}

\examples{
oldJIT <- enableJIT(0)
# a simple example
f <- function(x) x+1
fc <- cmpfun(f)
fc(2)
disassemble(fc)

# old R version of lapply
la1 <- function(X, FUN, ...) {
    FUN <- match.fun(FUN)
    if (!is.list(X))
    X <- as.list(X)
    rval <- vector("list", length(X))
    for(i in seq_along(X))
    rval[i] <- list(FUN(X[[i]], ...))
    names(rval) <- names(X)       # keep `names' !
    return(rval)
}
# a small variation
la2 <- function(X, FUN, ...) {
    FUN <- match.fun(FUN)
    if (!is.list(X))
    X <- as.list(X)
    rval <- vector("list", length(X))
    for(i in seq_along(X)) {
        v <- FUN(X[[i]], ...)
        if (is.null(v)) rval[i] <- list(v)
        else rval[[i]] <- v
    }
    names(rval) <- names(X)       # keep `names' !
    return(rval)
}
# Compiled versions
la1c <- cmpfun(la1)
la2c <- cmpfun(la2)
# some timings
x <- 1:10
y <- 1:100
\donttest{
system.time(for (i in 1:10000) lapply(x, is.null))
system.time(for (i in 1:10000) la1(x, is.null))
system.time(for (i in 1:10000) la1c(x, is.null))
system.time(for (i in 1:10000) la2(x, is.null))
system.time(for (i in 1:10000) la2c(x, is.null))
system.time(for (i in 1:1000) lapply(y, is.null))
system.time(for (i in 1:1000) la1(y, is.null))
system.time(for (i in 1:1000) la1c(y, is.null))
system.time(for (i in 1:1000) la2(y, is.null))
system.time(for (i in 1:1000) la2c(y, is.null))
}
\testonly{
for (i in 1:10000) lapply(x, is.null)
for (i in 1:10000) la1(x, is.null)
for (i in 1:10000) la1c(x, is.null)
for (i in 1:10000) la2(x, is.null)
for (i in 1:10000) la2c(x, is.null)
for (i in 1:1000) lapply(y, is.null)
for (i in 1:1000) la1(y, is.null)
for (i in 1:1000) la1c(y, is.null)
for (i in 1:1000) la2(y, is.null)
for (i in 1:1000) la2c(y, is.null)
}
enableJIT(oldJIT)}
\keyword{programming}