There are ways to control the nature of compiled code via the
declare
special form and
proclaim
function. See later in this chapter for fuller discussion of these two forms.
In particular there are a set of optimize qualities which take integral values from 0 to 3, in order to control the trade-offs between code size, speed, compilation time, debuggability of the resulting code, and the safety of the code (whether type checks are omitted). For example:
(proclaim '(optimize (speed 3) (safety 0) (debug 0)))
tells the compiler to concentrate on code speed rather than anything else, and
(proclaim '(optimize (safety 3)))
ensures that the compiler never takes liberties with Lisp semantics and produces code that checks for every kind of signallable error.
The important declarations to the compiler are type declarations and optimize declarations. To declare that the type of the value of a variable can be relied upon to be unchanging (and hence allow the compiler to omit various checks in the code), say:
(declare (type
the-type
variable
* )
Optimize declarations have various qualities, and these take values from 0 to 3. The names are
safety
,
fixnum-safety
,
float
,
sys:interruptable
,
debug
,
speed
,
compilation-speed
, and
space
.
Most of the qualities default to 1 (but
safety
and
fixnum-safety
default to 3 and
interruptable
defaults to 0). You can either associate an optimize quality with a new value (with local lexical scope if in
declare
, and global scope if
proclaim
), or just give it by itself, which implies the value 3 (taken to mean "maximum" in some loose sense).
Thus you ensure code is at maximum safety by:
(proclaim '(optimize (safety 3)))
(proclaim '(optimize safety))
and reduce debugging information to a minimum by
(proclaim '(optimize (debug 0)))
The
float
declaration allows generation of more efficient code using float numbers. It reduces allocation during float calculations. It is best used with
safety 0.
For example:
(progn
(setf a
(make-array 1000
:initial-element 1D0
:element-type 'double-float))
nil ; to avoid printing the large array
)
(compile
(defun test (a)
(declare (optimize (speed 3) (safety 0) (float 0)))
(declare (type (simple-array double-float (1000))
a))
(let ((sum 0D0))
(declare (type double-float sum))
(dotimes (i 1000)
(incf sum (the double-float (aref a i))))
sum)))
(time (test a))
=>
Timing the evaluation of (TEST A)
user time = 0.000
system time = 0.000
Elapsed time = 0:00:00
Allocation = 16 bytes standard / 0 bytes conses
0 Page faults
Normally code is interruptible, but when going for the extreme levels of speed and "undebuggability" this ceases to be the case unless you also ensure it thus:
(proclaim '(optimize (debug 0) (safety 0) (speed 3) interruptable))
The levels of
safety
have the following implications:
type
and
fixnum-safety
declarations to take effect. Array index bounds are checked for both reading and writing.
fixnum-safety
and
type
declarations from taking any effect.
The levels of
fixnum-safety
have the following implications:
The effects of combining these qualities is summarized below:
Ensure debugger knows values of args (and variable when source debugging is on) |
|
Fixnum-only arithmetic with errors for non fixnum arguments. |
|
The other optimize qualities are:
speed
-- the attention to fast code,
space
-- the degree of compactness,
compilation-speed
-- speed of compilation,
interruptable
-- whether code must be interruptible when unsafe.
Note that if you compile code with a low level of safety, you may get segmentation violations if the code is incorrect (for example, if type checking is turned off and you supply incorrect types). You can check this by interpreting the code rather than compiling it.
For example, the following function, compiled with safety = 2, does not check the type of its argument because it merely reads:
(defun foo (x)
(declare (optimize (safety 2)))
(car x))
However the following function, also compiled with safety = 2, does check the type of its argument because it writes:
(defun set-foo (x y)
(declare (optimize (safety 2)))
(setf (car x) y))