;;;; This file contains macro-like source transformations which
;;;; convert uses of certain functions into the canonical form desired
-;;;; within the compiler. ### and other IR1 transforms and stuff.
+;;;; within the compiler. FIXME: and other IR1 transforms and stuff.
;;;; This software is part of the SBCL system. See the README file for
;;;; more information.
(define-source-transform identity (x) `(prog1 ,x))
(define-source-transform values (x) `(prog1 ,x))
-;;; Bind the values and make a closure that returns them.
+;;; Bind the value and make a closure that returns it.
(define-source-transform constantly (value)
- (let ((rest (gensym "CONSTANTLY-REST-")))
- `(lambda (&rest ,rest)
- (declare (ignore ,rest))
- ,value)))
+ (let ((rest (gensym "CONSTANTLY-REST-"))
+ (n-value (gensym "CONSTANTLY-VALUE-")))
+ `(let ((,n-value ,value))
+ (lambda (&rest ,rest)
+ (declare (ignore ,rest))
+ ,n-value))))
;;; If the function has a known number of arguments, then return a
;;; lambda with the appropriate fixed number of args. If the
;;; Apply the function F to a bound X. If X is an open bound, then
;;; the result will be open. IF X is NIL, the result is NIL.
(defun bound-func (f x)
+ (declare (type function f))
(and x
(with-float-traps-masked (:underflow :overflow :inexact :divide-by-zero)
;; With these traps masked, we might get things like infinity
;;; result makes sense. It will if F is monotonic increasing (or
;;; non-decreasing).
(defun interval-func (f x)
- (declare (type interval x))
+ (declare (type function f)
+ (type interval x))
(let ((lo (bound-func f (interval-low x)))
(hi (bound-func f (interval-high x))))
(make-interval :low lo :high hi)))
;;; positive. If we didn't do this, we wouldn't be able to tell.
(defun two-arg-derive-type (arg1 arg2 derive-fcn fcn
&optional (convert-type t))
+ (declare (type function derive-fcn fcn))
#!+negative-zero-is-not-zero
(declare (ignore convert-type))
(flet (#!-negative-zero-is-not-zero
(or result 0)))
;;; If arg is a constant power of two, turn FLOOR into a shift and
-;;; mask. If CEILING, add in (1- (ABS Y)) and then do FLOOR.
+;;; mask. If CEILING, add in (1- (ABS Y)), do FLOOR and correct a
+;;; remainder.
(flet ((frob (y ceil-p)
(unless (constant-continuation-p y)
(give-up-ir1-transform))
(unless (= y-abs (ash 1 len))
(give-up-ir1-transform))
(let ((shift (- len))
- (mask (1- y-abs)))
- `(let ,(when ceil-p `((x (+ x ,(1- y-abs)))))
+ (mask (1- y-abs))
+ (delta (if ceil-p (* (signum y) (1- y-abs)) 0)))
+ `(let ((x (+ x ,delta)))
,(if (minusp y)
`(values (ash (- x) ,shift)
- (- (logand (- x) ,mask)))
+ (- (- (logand (- x) ,mask)) ,delta))
`(values (ash x ,shift)
- (logand x ,mask))))))))
+ (- (logand x ,mask) ,delta))))))))
(deftransform floor ((x y) (integer integer) *)
"convert division by 2^k to shift"
(frob y nil))
;;; Expand MAX and MIN into the obvious comparisons.
(define-source-transform max (arg0 &rest rest)
(once-only ((arg0 arg0))
- ;; ANSI says MAX should signal TYPE-ERROR if any arg isn't a REAL.
- ;;
- ;; KLUDGE: This UNLESS hackery is a workaround for bug 194.
- ;; Better, when that bug is fixed, would be (THE REAL ,ARG0).
- ;; -- WHN 2002-09-02
- `(progn
- (unless (realp ,arg0)
- (error-not-a-real ,arg0))
- ,(if (null rest)
- arg0
- `(let ((maxrest (max ,@rest)))
- (if (> ,arg0 maxrest) ,arg0 maxrest))))))
+ (if (null rest)
+ `(values (the real ,arg0))
+ `(let ((maxrest (max ,@rest)))
+ (if (> ,arg0 maxrest) ,arg0 maxrest)))))
(define-source-transform min (arg0 &rest rest)
(once-only ((arg0 arg0))
- ;; ANSI says MIN should signal TYPE-ERROR if any arg isn't a REAL.
- ;;
- ;; KLUDGE: This UNLESS hackery is a workaround for bug 194.
- ;; Better, when that bug is fixed, would be (THE REAL ,ARG0).
- ;; -- WHN 2002-09-02
- `(progn
- (unless (realp ,arg0)
- (error-not-a-real ,arg0))
- ,(if (null rest)
- arg0
- `(let ((minrest (min ,@rest)))
- (if (< ,arg0 minrest) ,arg0 minrest))))))
+ (if (null rest)
+ `(values (the real ,arg0))
+ `(let ((minrest (min ,@rest)))
+ (if (< ,arg0 minrest) ,arg0 minrest)))))
\f
;;;; converting N-arg arithmetic functions
;;;;
;;; Do source transformations for transitive functions such as +.
;;; One-arg cases are replaced with the arg and zero arg cases with
-;;; the identity. If LEAF-FUN is true, then replace two-arg calls with
-;;; a call to that function.
-(defun source-transform-transitive (fun args identity &optional leaf-fun)
+;;; the identity. ONE-ARG-RESULT-TYPE is, if non-NIL, the type to
+;;; ensure (with THE) that the argument in one-argument calls is.
+(defun source-transform-transitive (fun args identity
+ &optional one-arg-result-type)
(declare (symbol fun leaf-fun) (list args))
(case (length args)
(0 identity)
- (1 `(values ,(first args)))
- (2 (if leaf-fun
- `(,leaf-fun ,(first args) ,(second args))
- (values nil t)))
+ (1 (if one-arg-result-type
+ `(values (the ,one-arg-result-type ,(first args)))
+ `(values ,(first args))))
+ (2 (values nil t))
(t
(associate-args fun (first args) (rest args)))))
(define-source-transform + (&rest args)
- (source-transform-transitive '+ args 0))
+ (source-transform-transitive '+ args 0 'number))
(define-source-transform * (&rest args)
- (source-transform-transitive '* args 1))
+ (source-transform-transitive '* args 1 'number))
(define-source-transform logior (&rest args)
- (source-transform-transitive 'logior args 0))
+ (source-transform-transitive 'logior args 0 'integer))
(define-source-transform logxor (&rest args)
- (source-transform-transitive 'logxor args 0))
+ (source-transform-transitive 'logxor args 0 'integer))
(define-source-transform logand (&rest args)
- (source-transform-transitive 'logand args -1))
+ (source-transform-transitive 'logand args -1 'integer))
(define-source-transform logeqv (&rest args)
(if (evenp (length args))
;;; Do source transformations for intransitive n-arg functions such as
;;; /. With one arg, we form the inverse. With two args we pass.
;;; Otherwise we associate into two-arg calls.
-(declaim (ftype (function (symbol list t) list) source-transform-intransitive))
+(declaim (ftype (function (symbol list t)
+ (values list &optional (member nil t)))
+ source-transform-intransitive))
(defun source-transform-intransitive (function args inverse)
(case (length args)
((0 2) (values nil t))
nil)))
(defoptimizer (coerce derive-type) ((value type))
- (let ((value-type (continuation-type value))
- (type-type (continuation-type type)))
- (labels
- ((good-cons-type-p (cons-type)
- ;; Make sure the cons-type we're looking at is something
- ;; we're prepared to handle which is basically something
- ;; that array-element-type can return.
- (or (and (member-type-p cons-type)
- (null (rest (member-type-members cons-type)))
- (null (first (member-type-members cons-type))))
- (let ((car-type (cons-type-car-type cons-type)))
- (and (member-type-p car-type)
- (null (rest (member-type-members car-type)))
- (or (symbolp (first (member-type-members car-type)))
- (numberp (first (member-type-members car-type)))
- (and (listp (first (member-type-members car-type)))
- (numberp (first (first (member-type-members
- car-type))))))
- (good-cons-type-p (cons-type-cdr-type cons-type))))))
- (unconsify-type (good-cons-type)
- ;; Convert the "printed" respresentation of a cons
- ;; specifier into a type specifier. That is, the specifier
- ;; (cons (eql signed-byte) (cons (eql 16) null)) is
- ;; converted to (signed-byte 16).
- (cond ((or (null good-cons-type)
- (eq good-cons-type 'null))
- nil)
- ((and (eq (first good-cons-type) 'cons)
- (eq (first (second good-cons-type)) 'member))
- `(,(second (second good-cons-type))
- ,@(unconsify-type (caddr good-cons-type))))))
- (coerceable-p (c-type)
- ;; Can the value be coerced to the given type? Coerce is
- ;; complicated, so we don't handle every possible case
- ;; here---just the most common and easiest cases:
- ;;
- ;; o Any real can be coerced to a float type.
- ;; o Any number can be coerced to a complex single/double-float.
- ;; o An integer can be coerced to an integer.
- (let ((coerced-type c-type))
- (or (and (subtypep coerced-type 'float)
- (csubtypep value-type (specifier-type 'real)))
- (and (subtypep coerced-type
- '(or (complex single-float)
- (complex double-float)))
- (csubtypep value-type (specifier-type 'number)))
- (and (subtypep coerced-type 'integer)
- (csubtypep value-type (specifier-type 'integer))))))
- (process-types (type)
- ;; FIXME:
- ;; This needs some work because we should be able to derive
- ;; the resulting type better than just the type arg of
- ;; coerce. That is, if x is (integer 10 20), the (coerce x
- ;; 'double-float) should say (double-float 10d0 20d0)
- ;; instead of just double-float.
- (cond ((member-type-p type)
- (let ((members (member-type-members type)))
- (if (every #'coerceable-p members)
- (specifier-type `(or ,@members))
- *universal-type*)))
- ((and (cons-type-p type)
- (good-cons-type-p type))
- (let ((c-type (unconsify-type (type-specifier type))))
- (if (coerceable-p c-type)
- (specifier-type c-type)
- *universal-type*)))
- (t
- *universal-type*))))
- (cond ((union-type-p type-type)
- (apply #'type-union (mapcar #'process-types
- (union-type-types type-type))))
- ((or (member-type-p type-type)
- (cons-type-p type-type))
- (process-types type-type))
- (t
- *universal-type*)))))
+ (cond
+ ((constant-continuation-p type)
+ ;; This branch is essentially (RESULT-TYPE-SPECIFIER-NTH-ARG 2),
+ ;; but dealing with the niggle that complex canonicalization gets
+ ;; in the way: (COERCE 1 'COMPLEX) returns 1, which is not of
+ ;; type COMPLEX.
+ (let* ((specifier (continuation-value type))
+ (result-typeoid (careful-specifier-type specifier)))
+ (cond
+ ((null result-typeoid) nil)
+ ((csubtypep result-typeoid (specifier-type 'number))
+ ;; the difficult case: we have to cope with ANSI 12.1.5.3
+ ;; Rule of Canonical Representation for Complex Rationals,
+ ;; which is a truly nasty delivery to field.
+ (cond
+ ((csubtypep result-typeoid (specifier-type 'real))
+ ;; cleverness required here: it would be nice to deduce
+ ;; that something of type (INTEGER 2 3) coerced to type
+ ;; DOUBLE-FLOAT should return (DOUBLE-FLOAT 2.0d0 3.0d0).
+ ;; FLOAT gets its own clause because it's implemented as
+ ;; a UNION-TYPE, so we don't catch it in the NUMERIC-TYPE
+ ;; logic below.
+ result-typeoid)
+ ((and (numeric-type-p result-typeoid)
+ (eq (numeric-type-complexp result-typeoid) :real))
+ ;; FIXME: is this clause (a) necessary or (b) useful?
+ result-typeoid)
+ ((or (csubtypep result-typeoid
+ (specifier-type '(complex single-float)))
+ (csubtypep result-typeoid
+ (specifier-type '(complex double-float)))
+ #!+long-float
+ (csubtypep result-typeoid
+ (specifier-type '(complex long-float))))
+ ;; float complex types are never canonicalized.
+ result-typeoid)
+ (t
+ ;; if it's not a REAL, or a COMPLEX FLOAToid, it's
+ ;; probably just a COMPLEX or equivalent. So, in that
+ ;; case, we will return a complex or an object of the
+ ;; provided type if it's rational:
+ (type-union result-typeoid
+ (type-intersection (continuation-type value)
+ (specifier-type 'rational))))))
+ (t result-typeoid))))
+ (t
+ ;; OK, the result-type argument isn't constant. However, there
+ ;; are common uses where we can still do better than just
+ ;; *UNIVERSAL-TYPE*: e.g. (COERCE X (ARRAY-ELEMENT-TYPE Y)),
+ ;; where Y is of a known type. See messages on cmucl-imp
+ ;; 2001-02-14 and sbcl-devel 2002-12-12. We only worry here
+ ;; about types that can be returned by (ARRAY-ELEMENT-TYPE Y), on
+ ;; the basis that it's unlikely that other uses are both
+ ;; time-critical and get to this branch of the COND (non-constant
+ ;; second argument to COERCE). -- CSR, 2002-12-16
+ (let ((value-type (continuation-type value))
+ (type-type (continuation-type type)))
+ (labels
+ ((good-cons-type-p (cons-type)
+ ;; Make sure the cons-type we're looking at is something
+ ;; we're prepared to handle which is basically something
+ ;; that array-element-type can return.
+ (or (and (member-type-p cons-type)
+ (null (rest (member-type-members cons-type)))
+ (null (first (member-type-members cons-type))))
+ (let ((car-type (cons-type-car-type cons-type)))
+ (and (member-type-p car-type)
+ (null (rest (member-type-members car-type)))
+ (or (symbolp (first (member-type-members car-type)))
+ (numberp (first (member-type-members car-type)))
+ (and (listp (first (member-type-members
+ car-type)))
+ (numberp (first (first (member-type-members
+ car-type))))))
+ (good-cons-type-p (cons-type-cdr-type cons-type))))))
+ (unconsify-type (good-cons-type)
+ ;; Convert the "printed" respresentation of a cons
+ ;; specifier into a type specifier. That is, the
+ ;; specifier (CONS (EQL SIGNED-BYTE) (CONS (EQL 16)
+ ;; NULL)) is converted to (SIGNED-BYTE 16).
+ (cond ((or (null good-cons-type)
+ (eq good-cons-type 'null))
+ nil)
+ ((and (eq (first good-cons-type) 'cons)
+ (eq (first (second good-cons-type)) 'member))
+ `(,(second (second good-cons-type))
+ ,@(unconsify-type (caddr good-cons-type))))))
+ (coerceable-p (c-type)
+ ;; Can the value be coerced to the given type? Coerce is
+ ;; complicated, so we don't handle every possible case
+ ;; here---just the most common and easiest cases:
+ ;;
+ ;; * Any REAL can be coerced to a FLOAT type.
+ ;; * Any NUMBER can be coerced to a (COMPLEX
+ ;; SINGLE/DOUBLE-FLOAT).
+ ;;
+ ;; FIXME I: we should also be able to deal with characters
+ ;; here.
+ ;;
+ ;; FIXME II: I'm not sure that anything is necessary
+ ;; here, at least while COMPLEX is not a specialized
+ ;; array element type in the system. Reasoning: if
+ ;; something cannot be coerced to the requested type, an
+ ;; error will be raised (and so any downstream compiled
+ ;; code on the assumption of the returned type is
+ ;; unreachable). If something can, then it will be of
+ ;; the requested type, because (by assumption) COMPLEX
+ ;; (and other difficult types like (COMPLEX INTEGER)
+ ;; aren't specialized types.
+ (let ((coerced-type c-type))
+ (or (and (subtypep coerced-type 'float)
+ (csubtypep value-type (specifier-type 'real)))
+ (and (subtypep coerced-type
+ '(or (complex single-float)
+ (complex double-float)))
+ (csubtypep value-type (specifier-type 'number))))))
+ (process-types (type)
+ ;; FIXME: This needs some work because we should be able
+ ;; to derive the resulting type better than just the
+ ;; type arg of coerce. That is, if X is (INTEGER 10
+ ;; 20), then (COERCE X 'DOUBLE-FLOAT) should say
+ ;; (DOUBLE-FLOAT 10d0 20d0) instead of just
+ ;; double-float.
+ (cond ((member-type-p type)
+ (let ((members (member-type-members type)))
+ (if (every #'coerceable-p members)
+ (specifier-type `(or ,@members))
+ *universal-type*)))
+ ((and (cons-type-p type)
+ (good-cons-type-p type))
+ (let ((c-type (unconsify-type (type-specifier type))))
+ (if (coerceable-p c-type)
+ (specifier-type c-type)
+ *universal-type*)))
+ (t
+ *universal-type*))))
+ (cond ((union-type-p type-type)
+ (apply #'type-union (mapcar #'process-types
+ (union-type-types type-type))))
+ ((or (member-type-p type-type)
+ (cons-type-p type-type))
+ (process-types type-type))
+ (t
+ *universal-type*)))))))
+(defoptimizer (compile derive-type) ((nameoid function))
+ (when (csubtypep (continuation-type nameoid)
+ (specifier-type 'null))
+ (values-specifier-type '(values function boolean boolean))))
+
+;;; FIXME: Maybe also STREAM-ELEMENT-TYPE should be given some loving
+;;; treatment along these lines? (See discussion in COERCE DERIVE-TYPE
+;;; optimizer, above).
(defoptimizer (array-element-type derive-type) ((array))
(let ((array-type (continuation-type array)))
(labels ((consify (list)
(mapcar #'get-element-type (union-type-types array-type))))
(t
*universal-type*)))))
+
+(define-source-transform sb!impl::sort-vector (vector start end predicate key)
+ `(macrolet ((%index (x) `(truly-the index ,x))
+ (%parent (i) `(ash ,i -1))
+ (%left (i) `(%index (ash ,i 1)))
+ (%right (i) `(%index (1+ (ash ,i 1))))
+ (%heapify (i)
+ `(do* ((i ,i)
+ (left (%left i) (%left i)))
+ ((> left current-heap-size))
+ (declare (type index i left))
+ (let* ((i-elt (%elt i))
+ (i-key (funcall keyfun i-elt))
+ (left-elt (%elt left))
+ (left-key (funcall keyfun left-elt)))
+ (multiple-value-bind (large large-elt large-key)
+ (if (funcall ,',predicate i-key left-key)
+ (values left left-elt left-key)
+ (values i i-elt i-key))
+ (let ((right (%right i)))
+ (multiple-value-bind (largest largest-elt)
+ (if (> right current-heap-size)
+ (values large large-elt)
+ (let* ((right-elt (%elt right))
+ (right-key (funcall keyfun right-elt)))
+ (if (funcall ,',predicate large-key right-key)
+ (values right right-elt)
+ (values large large-elt))))
+ (cond ((= largest i)
+ (return))
+ (t
+ (setf (%elt i) largest-elt
+ (%elt largest) i-elt
+ i largest)))))))))
+ (%sort-vector (keyfun &optional (vtype 'vector))
+ `(macrolet (;; KLUDGE: In SBCL ca. 0.6.10, I had trouble getting
+ ;; type inference to propagate all the way
+ ;; through this tangled mess of
+ ;; inlining. The TRULY-THE here works
+ ;; around that. -- WHN
+ (%elt (i)
+ `(aref (truly-the ,',vtype ,',',vector)
+ (%index (+ (%index ,i) start-1)))))
+ (let ((start-1 (1- ,',start)) ; Heaps prefer 1-based addressing.
+ (current-heap-size (- ,',end ,',start))
+ (keyfun ,keyfun))
+ (declare (type (integer -1 #.(1- most-positive-fixnum))
+ start-1))
+ (declare (type index current-heap-size))
+ (declare (type function keyfun))
+ (loop for i of-type index
+ from (ash current-heap-size -1) downto 1 do
+ (%heapify i))
+ (loop
+ (when (< current-heap-size 2)
+ (return))
+ (rotatef (%elt 1) (%elt current-heap-size))
+ (decf current-heap-size)
+ (%heapify 1))))))
+ (if (typep ,vector 'simple-vector)
+ ;; (VECTOR T) is worth optimizing for, and SIMPLE-VECTOR is
+ ;; what we get from (VECTOR T) inside WITH-ARRAY-DATA.
+ (if (null ,key)
+ ;; Special-casing the KEY=NIL case lets us avoid some
+ ;; function calls.
+ (%sort-vector #'identity simple-vector)
+ (%sort-vector ,key simple-vector))
+ ;; It's hard to anticipate many speed-critical applications for
+ ;; sorting vector types other than (VECTOR T), so we just lump
+ ;; them all together in one slow dynamically typed mess.
+ (locally
+ (declare (optimize (speed 2) (space 2) (inhibit-warnings 3)))
+ (%sort-vector (or ,key #'identity))))))
\f
;;;; debuggers' little helpers