The R Project SVN R

        get(name, compilerOptions)

        get(name, compilerOptions)

}

}

@ %def getCompilerOption

@ %def getCompilerOption

The [[suppressAll]] function determines whether a context has its

The [[suppressAll]] function determines whether a context has its

<<[[suppressAll]] function>>=

<<[[suppressAll]] function>>=

suppressAll <- function(cntxt)

suppressAll <- function(cntxt)

    identical(cntxt$suppressAll, TRUE)

    identical(cntxt$suppressAll, TRUE)

@ %def suppressAll

@ %def suppressAll

The [[suppressNoSuperAssignVar]] function determines whether a context has

The [[suppressNoSuperAssignVar]] function determines whether a context has

would significantly complicate the compiler and the compiled code, and

would significantly complicate the compiler and the compiled code, and

carry at least some runtime penalty. The approach taken here is

carry at least some runtime penalty. The approach taken here is

therefore to permit the compiler to inline some functions when they

therefore to permit the compiler to inline some functions when they

are not visibly shadowed in the compiled code.  What the compiler is

are not visibly shadowed in the compiled code.  What the compiler is

permitted to do is determined by the setting of an optimization level.

permitted to do is determined by the setting of an optimization level.

For the compiler to be able to decide whether is can inline a function

For the compiler to be able to decide whether is can inline a function

it needs to be able to determine whether there are any local variable

it needs to be able to determine whether there are any local variable

that might shadow a variable from a base package. This requires adding

that might shadow a variable from a base package. This requires adding

environment information to the compilation process.

environment information to the compilation process.

[[quote]].  These might be shadowed by local definitions in a

[[quote]].  These might be shadowed by local definitions in a

surrounding function.  To allow for this we can add an optional

surrounding function.  To allow for this we can add an optional

variable [[shadowed]] for providing a character vector of names of

variable [[shadowed]] for providing a character vector of names of

variables with shadowing local definitions.

variables with shadowing local definitions.

recursive calls. [[findLocals1]] now, instead of searching through the full

recursive calls. [[findLocals1]] now, instead of searching through the full

transitive closure of language objects, only searches from the first, but

transitive closure of language objects, only searches from the first, but

returns what remains to be searched. The variables found are stored into an

returns what remains to be searched. The variables found are stored into an

environment, which avoids some extra calls and assures that each variable is

environment, which avoids some extra calls and assures that each variable is

listed at most once.

listed at most once.

not.  [[ENDLOOPCNTXT]] terminates the context established by

not.  [[ENDLOOPCNTXT]] terminates the context established by

[[STARTLOOPCNTXT]] and pops it off the context stack.  The context

[[STARTLOOPCNTXT]] and pops it off the context stack.  The context

data is stored on the byte code interpreter stack; in the case of a

data is stored on the byte code interpreter stack; in the case of a

[[for]] loop some loop state information is duplicated on the stack by

[[for]] loop some loop state information is duplicated on the stack by

[[STARTLOOPCNTXT]] and removed again by [[ENDLOOPCNTXT]]. The byte

[[STARTLOOPCNTXT]] and removed again by [[ENDLOOPCNTXT]]. The byte

structure so it is available after a [[longjmp]] triggered by a

structure so it is available after a [[longjmp]] triggered by a

[[break]] for retrieving the label on the [[ENDLOOPCNTXT]]

[[break]] for retrieving the label on the [[ENDLOOPCNTXT]]

instruction.  An alternative would be to add separate

instruction.  An alternative would be to add separate

[[STARTFORLOOPCNTXT]] and [[ENDFORLOOPCNTXT]] instructions. Then the

[[STARTFORLOOPCNTXT]] and [[ENDFORLOOPCNTXT]] instructions. Then the

[[pc]] or the label could be stored on the note stack.

[[pc]] or the label could be stored on the note stack.

    }

    }

}

}

@ %def cmpPrim2

@ %def cmpPrim2

The [[INCLNK]] and [[DECLNK]] instructions are used to protect

The [[INCLNK]] and [[DECLNK]] instructions are used to protect

of subsequent arguments. These instructions can be omitted if the

of subsequent arguments. These instructions can be omitted if the

subsequent argument evaluations cannot modify values on the stack.

subsequent argument evaluations cannot modify values on the stack.

With changes to stack protection this is mechanism is no longer

With changes to stack protection this is mechanism is no longer

needed, so the check just returns [[FALSE]].

needed, so the check just returns [[FALSE]].

<<[[checkNeedsInc]] function>>=

<<[[checkNeedsInc]] function>>=

})

})

@ 

@ 

A number of one argument math functions are handled by the interpreter

A number of one argument math functions are handled by the interpreter

using the function [[math1]] in [[arithmetic.c]]. The [[MATH1.OP]]

using the function [[math1]] in [[arithmetic.c]]. The [[MATH1.OP]]

operands, an index for the call expression in the constant table, and

operands, an index for the call expression in the constant table, and

an index for the function to be called in a table of function

an index for the function to be called in a table of function

pointers. The table of names in the byte code compiler has to match

pointers. The table of names in the byte code compiler has to match

the function pointer array in the byte code interpreter.  It would

the function pointer array in the byte code interpreter.  It would

have been possible to use the same indices as the offset values used

have been possible to use the same indices as the offset values used

  it at this time. But it would be good to be able to change it in the

  it at this time. But it would be good to be able to change it in the

  future.

  future.

\end{itemize}

\end{itemize}

Because of these issues the compiler implements slightly different

Because of these issues the compiler implements slightly different

\emph{Evaluation} semantics should be identical; the difference arises

\emph{Evaluation} semantics should be identical; the difference arises

in how intermediate values are managed and has some effect on results

in how intermediate values are managed and has some effect on results

produced by [[substitute]].  In particular, no intermediate [[*tmp*]]

produced by [[substitute]].  In particular, no intermediate [[*tmp*]]

value is used and therefore no cleanup frame is needed.  This does

value is used and therefore no cleanup frame is needed.  This does

mean that uses of the form

mean that uses of the form

}

}

@ %def .onLoad

@ %def .onLoad

When [[enableJIT]] is set to 3, loops should be compiled before executing. 

When [[enableJIT]] is set to 3, loops should be compiled before executing. 

However, if the [[optimize]] option is set to 0 or 1, a compiled loop will

However, if the [[optimize]] option is set to 0 or 1, a compiled loop will

[[do_for]]), and the compilation would run into infinite recursion. 

[[do_for]]), and the compilation would run into infinite recursion. 

[[checkCompilerOptions]] will detect invalid combinations of [[enableJIT]]

[[checkCompilerOptions]] will detect invalid combinations of [[enableJIT]]

and [[optimize]] and report a warning.

and [[optimize]] and report a warning.

%% **** could also change the interface and atomically set enableJIT and optimize

%% **** could also change the interface and atomically set enableJIT and optimize

<<[[checkCompilerOptions]] function>>=

<<[[checkCompilerOptions]] function>>=

\appendix

\appendix

\section{General utility functions}

\section{General utility functions}

This appendix provides a few general utility functions.

This appendix provides a few general utility functions.

The utility function [[pasteExpr]] is used in the error messages.

The utility function [[pasteExpr]] is used in the error messages.

%% **** use error context or catch errors?

%% **** use error context or catch errors?

%% **** maybe don't need expression if we catch errors?

%% **** maybe don't need expression if we catch errors?

<<[[pasteExpr]] function>>=

<<[[pasteExpr]] function>>=

pasteExpr <- function(e, prefix = "\n    ") {

pasteExpr <- function(e, prefix = "\n    ") {

    de <- deparse(e)

    de <- deparse(e)

@ %def nsName

@ %def nsName

\section{Experimental utilities}

\section{Experimental utilities}

not exported. The first is a simple byte code profiler. This requires

not exported. The first is a simple byte code profiler. This requires

that the file [[eval.c]] be compiled with [[BC_PROFILING]] enabled,

that the file [[eval.c]] be compiled with [[BC_PROFILING]] enabled,

which on [[gcc]]-compatible compilers will disable threaded code. The

which on [[gcc]]-compatible compilers will disable threaded code. The

byte code profiler uses the profile timer to record the active byte

byte code profiler uses the profile timer to record the active byte

code instruction at interrupt time. The function [[bcprof]] runs the

code instruction at interrupt time. The function [[bcprof]] runs the