(in-package "SB!C")
-;;; Convert into an IF so that IF optimizations will eliminate redundant
-;;; negations.
-(define-source-transform not (x) `(if ,x nil t))
-(define-source-transform null (x) `(if ,x nil t))
-
-;;; ENDP is just NULL with a LIST assertion. The assertion will be
-;;; optimized away when SAFETY optimization is low; hopefully that
-;;; is consistent with ANSI's "should return an error".
-(define-source-transform endp (x) `(null (the list ,x)))
-
;;; We turn IDENTITY into PROG1 so that it is obvious that it just
;;; returns the first value of its argument. Ditto for VALUES with one
;;; arg.
(define-source-transform identity (x) `(prog1 ,x))
(define-source-transform values (x) `(prog1 ,x))
-;;; Bind the value and make a closure that returns it.
-(define-source-transform constantly (value)
- (with-unique-names (rest n-value)
- `(let ((,n-value ,value))
- (lambda (&rest ,rest)
- (declare (ignore ,rest))
- ,n-value))))
+;;; CONSTANTLY is pretty much never worth transforming, but it's good to get the type.
+(defoptimizer (constantly derive-type) ((value))
+ (specifier-type
+ `(function (&rest t) (values ,(type-specifier (lvar-type value)) &optional))))
;;; If the function has a known number of arguments, then return a
;;; lambda with the appropriate fixed number of args. If the
(give-up-ir1-transform
"The function doesn't have a fixed argument count.")))))
\f
+;;;; SYMBOL-VALUE &co
+(defun derive-symbol-value-type (lvar node)
+ (if (constant-lvar-p lvar)
+ (let* ((sym (lvar-value lvar))
+ (var (maybe-find-free-var sym))
+ (local-type (when var
+ (let ((*lexenv* (node-lexenv node)))
+ (lexenv-find var type-restrictions))))
+ (global-type (info :variable :type sym)))
+ (if local-type
+ (type-intersection local-type global-type)
+ global-type))
+ *universal-type*))
+
+(defoptimizer (symbol-value derive-type) ((symbol) node)
+ (derive-symbol-value-type symbol node))
+
+(defoptimizer (symbol-global-value derive-type) ((symbol) node)
+ (derive-symbol-value-type symbol node))
+\f
;;;; list hackery
;;; Translate CxR into CAR/CDR combos.
;;; Make source transforms to turn CxR forms into combinations of CAR
;;; and CDR. ANSI specifies that everything up to 4 A/D operations is
;;; defined.
+;;; Don't transform CAD*R, they are treated specially for &more args
+;;; optimizations
+
(/show0 "about to set CxR source transforms")
(loop for i of-type index from 2 upto 4 do
;; Iterate over BUF = all names CxR where x = an I-element
(declare (type index k))
(setf (aref buf (1+ k))
(if (logbitp k j) #\A #\D)))
- (setf (info :function :source-transform (intern buf))
- #'source-transform-cxr))))
+ (unless (member buf '("CADR" "CADDR" "CADDDR")
+ :test #'equal)
+ (setf (info :function :source-transform (intern buf))
+ #'source-transform-cxr)))))
(/show0 "done setting CxR source transforms")
;;; Turn FIRST..FOURTH and REST into the obvious synonym, assuming
;;; whatever is right for them is right for us. FIFTH..TENTH turn into
;;; Nth, which can be expanded into a CAR/CDR later on if policy
;;; favors it.
-(define-source-transform first (x) `(car ,x))
(define-source-transform rest (x) `(cdr ,x))
+(define-source-transform first (x) `(car ,x))
(define-source-transform second (x) `(cadr ,x))
(define-source-transform third (x) `(caddr ,x))
(define-source-transform fourth (x) `(cadddr ,x))
(setf (cdr ,n-x) ,y)
,n-x)))
-(define-source-transform nth (n l) `(car (nthcdr ,n ,l)))
-
(deftransform last ((list &optional n) (t &optional t))
(let ((c (constant-lvar-p n)))
(cond ((or (not n)
(defun set-bound (x open-p)
(if (and x open-p) (list x) x))
-;;; 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)
+;;; Apply the function F to a bound X. If X is an open bound and the
+;;; function is declared strictly monotonic, then the result will be
+;;; open. IF X is NIL, the result is NIL.
+(defun bound-func (f x strict)
(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
- ;; or negative infinity returned. Check for this and return
- ;; NIL to indicate unbounded.
- (let ((y (funcall f (type-bound-number x))))
- (if (and (floatp y)
- (float-infinity-p y))
- nil
- (set-bound y (consp x)))))))
+ (handler-case
+ (with-float-traps-masked (:underflow :overflow :inexact :divide-by-zero)
+ ;; With these traps masked, we might get things like infinity
+ ;; or negative infinity returned. Check for this and return
+ ;; NIL to indicate unbounded.
+ (let ((y (funcall f (type-bound-number x))))
+ (if (and (floatp y)
+ (float-infinity-p y))
+ nil
+ (set-bound y (and strict (consp x))))))
+ ;; Some numerical operations will signal SIMPLE-TYPE-ERROR, e.g.
+ ;; in the course of converting a bignum to a float. Default to
+ ;; NIL in that case.
+ (simple-type-error ()))))
(defun safe-double-coercion-p (x)
(or (typep x 'double-float)
(defun safe-single-coercion-p (x)
(or (typep x 'single-float)
- ;; Fix for bug 420, and related issues: during type derivation we often
- ;; end up deriving types for both
- ;;
- ;; (some-op <int> <single>)
- ;; and
- ;; (some-op (coerce <int> 'single-float) <single>)
- ;;
- ;; or other equivalent transformed forms. The problem with this is that
- ;; on some platforms like x86 (+ <int> <single>) is on the machine level
- ;; equivalent of
- ;;
- ;; (coerce (+ (coerce <int> 'double-float)
- ;; (coerce <single> 'double-float))
- ;; 'single-float)
- ;;
- ;; so if the result of (coerce <int> 'single-float) is not exact, the
- ;; derived types for the transformed forms will have an empty
- ;; intersection -- which in turn means that the compiler will conclude
- ;; that the call never returns, and all hell breaks lose when it *does*
- ;; return at runtime. (This affects not just +, but other operators are
- ;; well.)
- (and (not (typep x `(or (integer * (,most-negative-exactly-single-float-fixnum))
- (integer (,most-positive-exactly-single-float-fixnum) *))))
- (<= most-negative-single-float x most-positive-single-float))))
+ (and
+ ;; Fix for bug 420, and related issues: during type derivation we often
+ ;; end up deriving types for both
+ ;;
+ ;; (some-op <int> <single>)
+ ;; and
+ ;; (some-op (coerce <int> 'single-float) <single>)
+ ;;
+ ;; or other equivalent transformed forms. The problem with this
+ ;; is that on x86 (+ <int> <single>) is on the machine level
+ ;; equivalent of
+ ;;
+ ;; (coerce (+ (coerce <int> 'double-float)
+ ;; (coerce <single> 'double-float))
+ ;; 'single-float)
+ ;;
+ ;; so if the result of (coerce <int> 'single-float) is not exact, the
+ ;; derived types for the transformed forms will have an empty
+ ;; intersection -- which in turn means that the compiler will conclude
+ ;; that the call never returns, and all hell breaks lose when it *does*
+ ;; return at runtime. (This affects not just +, but other operators are
+ ;; well.)
+ ;;
+ ;; See also: SAFE-CTYPE-FOR-SINGLE-COERCION-P
+ ;;
+ ;; FIXME: If we ever add SSE-support for x86, this conditional needs to
+ ;; change.
+ #!+x86
+ (not (typep x `(or (integer * (,most-negative-exactly-single-float-fixnum))
+ (integer (,most-positive-exactly-single-float-fixnum) *))))
+ (<= most-negative-single-float x most-positive-single-float))))
;;; Apply a binary operator OP to two bounds X and Y. The result is
;;; NIL if either is NIL. Otherwise bound is computed and the result
(t (,op ,x ,y))))
(defmacro bound-binop (op x y)
- `(and ,x ,y
- (with-float-traps-masked (:underflow :overflow :inexact :divide-by-zero)
- (set-bound (safely-binop ,op (type-bound-number ,x)
- (type-bound-number ,y))
- (or (consp ,x) (consp ,y))))))
+ (with-unique-names (xb yb res)
+ `(and ,x ,y
+ (with-float-traps-masked (:underflow :overflow :inexact :divide-by-zero)
+ (let* ((,xb (type-bound-number ,x))
+ (,yb (type-bound-number ,y))
+ (,res (safely-binop ,op ,xb ,yb)))
+ (set-bound ,res
+ (and (or (consp ,x) (consp ,y))
+ ;; Open bounds can very easily be messed up
+ ;; by FP rounding, so take care here.
+ ,(case op
+ (*
+ ;; Multiplying a greater-than-zero with
+ ;; less than one can round to zero.
+ `(or (not (fp-zero-p ,res))
+ (cond ((and (consp ,x) (fp-zero-p ,xb))
+ (>= (abs ,yb) 1))
+ ((and (consp ,y) (fp-zero-p ,yb))
+ (>= (abs ,xb) 1)))))
+ (/
+ ;; Dividing a greater-than-zero with
+ ;; greater than one can round to zero.
+ `(or (not (fp-zero-p ,res))
+ (cond ((and (consp ,x) (fp-zero-p ,xb))
+ (<= (abs ,yb) 1))
+ ((and (consp ,y) (fp-zero-p ,yb))
+ (<= (abs ,xb) 1)))))
+ ((+ -)
+ ;; Adding or subtracting greater-than-zero
+ ;; can end up with identity.
+ `(and (not (fp-zero-p ,xb))
+ (not (fp-zero-p ,yb))))))))))))
+
+(defun coercion-loses-precision-p (val type)
+ (typecase val
+ (single-float)
+ (double-float (subtypep type 'single-float))
+ (rational (subtypep type 'float))
+ (t (bug "Unexpected arguments to bounds coercion: ~S ~S" val type))))
(defun coerce-for-bound (val type)
(if (consp val)
- (list (coerce-for-bound (car val) type))
+ (let ((xbound (coerce-for-bound (car val) type)))
+ (if (coercion-loses-precision-p (car val) type)
+ xbound
+ (list xbound)))
(cond
((subtypep type 'double-float)
(if (<= most-negative-double-float val most-positive-double-float)
(defun coerce-and-truncate-floats (val type)
(when val
(if (consp val)
- (list (coerce-and-truncate-floats (car val) type))
+ (let ((xbound (coerce-for-bound (car val) type)))
+ (if (coercion-loses-precision-p (car val) type)
+ xbound
+ (list xbound)))
(cond
((subtypep type 'double-float)
(if (<= most-negative-double-float val most-positive-double-float)
:high (copy-interval-limit (interval-high x))))
;;; Given a point P contained in the interval X, split X into two
-;;; interval at the point P. If CLOSE-LOWER is T, then the left
+;;; intervals at the point P. If CLOSE-LOWER is T, then the left
;;; interval contains P. If CLOSE-UPPER is T, the right interval
;;; contains P. You can specify both to be T or NIL.
(defun interval-split (p x &optional close-lower close-upper)
;;; the negative of an interval
(defun interval-neg (x)
(declare (type interval x))
- (make-interval :low (bound-func #'- (interval-high x))
- :high (bound-func #'- (interval-low x))))
+ (make-interval :low (bound-func #'- (interval-high x) t)
+ :high (bound-func #'- (interval-low x) t)))
;;; Add two intervals.
(defun interval-add (x y)
((zerop (type-bound-number y))
;; Divide by zero means result is infinity
nil)
- ((and (numberp x) (zerop x))
- ;; Zero divided by anything is zero.
- x)
(t
(bound-binop / x y)))))
(let ((top-range (interval-range-info top))
;;; Apply the function F to the interval X. If X = [a, b], then the
;;; result is [f(a), f(b)]. It is up to the user to make sure the
-;;; result makes sense. It will if F is monotonic increasing (or
-;;; non-decreasing).
-(defun interval-func (f x)
+;;; result makes sense. It will if F is monotonic increasing (or, if
+;;; the interval is closed, non-decreasing).
+;;;
+;;; (Actually most uses of INTERVAL-FUNC are coercions to float types,
+;;; which are not monotonic increasing, so default to calling
+;;; BOUND-FUNC with a non-strict argument).
+(defun interval-func (f x &optional increasing)
(declare (type function f)
(type interval x))
- (let ((lo (bound-func f (interval-low x)))
- (hi (bound-func f (interval-high x))))
+ (let ((lo (bound-func f (interval-low x) increasing))
+ (hi (bound-func f (interval-high x) increasing)))
(make-interval :low lo :high hi)))
;;; Return T if X < Y. That is every number in the interval X is
;;; Compute the square of an interval.
(defun interval-sqr (x)
(declare (type interval x))
- (interval-func (lambda (x) (* x x))
- (interval-abs x)))
+ (interval-func (lambda (x) (* x x)) (interval-abs x)))
\f
;;;; numeric DERIVE-TYPE methods
;;; a utility for defining derive-type methods of integer operations. If
;;; the types of both X and Y are integer types, then we compute a new
-;;; integer type with bounds determined Fun when applied to X and Y.
+;;; integer type with bounds determined by FUN when applied to X and Y.
;;; Otherwise, we use NUMERIC-CONTAGION.
(defun derive-integer-type-aux (x y fun)
(declare (type function fun))
(t
type-list)))
-;;; FIXME: MAKE-CANONICAL-UNION-TYPE and CONVERT-MEMBER-TYPE probably
-;;; belong in the kernel's type logic, invoked always, instead of in
-;;; the compiler, invoked only during some type optimizations. (In
-;;; fact, as of 0.pre8.100 or so they probably are, under
-;;; MAKE-MEMBER-TYPE, so probably this code can be deleted)
-
;;; Take a list of types and return a canonical type specifier,
;;; combining any MEMBER types together. If both positive and negative
;;; MEMBER types are present they are converted to a float type.
;;; XXX This would be far simpler if the type-union methods could handle
;;; member/number unions.
-(defun make-canonical-union-type (type-list)
+;;;
+;;; If we're about to generate an overly complex union of numeric types, start
+;;; collapse the ranges together.
+;;;
+;;; FIXME: The MEMBER canonicalization parts of MAKE-DERIVED-UNION-TYPE and
+;;; entire CONVERT-MEMBER-TYPE probably belong in the kernel's type logic,
+;;; invoked always, instead of in the compiler, invoked only during some type
+;;; optimizations.
+(defvar *derived-numeric-union-complexity-limit* 6)
+
+(defun make-derived-union-type (type-list)
(let ((xset (alloc-xset))
(fp-zeroes '())
- (misc-types '()))
+ (misc-types '())
+ (numeric-type *empty-type*))
(dolist (type type-list)
(cond ((member-type-p type)
(mapc-member-type-members
(pushnew member fp-zeroes))
(add-to-xset member xset)))
type))
+ ((numeric-type-p type)
+ (let ((*approximate-numeric-unions*
+ (when (and (union-type-p numeric-type)
+ (nthcdr *derived-numeric-union-complexity-limit*
+ (union-type-types numeric-type)))
+ t)))
+ (setf numeric-type (type-union type numeric-type))))
(t
(push type misc-types))))
(if (and (xset-empty-p xset) (not fp-zeroes))
- (apply #'type-union misc-types)
- (apply #'type-union (make-member-type :xset xset :fp-zeroes fp-zeroes) misc-types))))
+ (apply #'type-union numeric-type misc-types)
+ (apply #'type-union (make-member-type :xset xset :fp-zeroes fp-zeroes)
+ numeric-type misc-types))))
;;; Convert a member type with a single member to a numeric type.
(defun convert-member-type (arg)
(setf results (append results result))
(push result results))))
(if (rest results)
- (make-canonical-union-type results)
+ (make-derived-union-type results)
(first results)))))))
;;; Same as ONE-ARG-DERIVE-TYPE, except we assume the function takes
(setf results (append results result))
(push result results))))))
(if (rest results)
- (make-canonical-union-type results)
+ (make-derived-union-type results)
(first results)))))))
\f
#+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
#'%unary-truncate-derive-type-aux
#'%unary-truncate))
+(defoptimizer (%unary-truncate/single-float derive-type) ((number))
+ (one-arg-derive-type number
+ #'%unary-truncate-derive-type-aux
+ #'%unary-truncate))
+
+(defoptimizer (%unary-truncate/double-float derive-type) ((number))
+ (one-arg-derive-type number
+ #'%unary-truncate-derive-type-aux
+ #'%unary-truncate))
+
(defoptimizer (%unary-ftruncate derive-type) ((number))
(let ((divisor (specifier-type '(integer 1 1))))
(one-arg-derive-type number
(ftruncate-derive-type-quot-aux n divisor nil))
#'%unary-ftruncate)))
+(defoptimizer (%unary-round derive-type) ((number))
+ (one-arg-derive-type number
+ (lambda (n)
+ (block nil
+ (unless (numeric-type-real-p n)
+ (return *empty-type*))
+ (let* ((interval (numeric-type->interval n))
+ (low (interval-low interval))
+ (high (interval-high interval)))
+ (when (consp low)
+ (setf low (car low)))
+ (when (consp high)
+ (setf high (car high)))
+ (specifier-type
+ `(integer ,(if low
+ (round low)
+ '*)
+ ,(if high
+ (round high)
+ '*))))))
+ #'%unary-round))
+
;;; Define optimizers for FLOOR and CEILING.
(macrolet
((def (name q-name r-name)
(if (and divisor-low divisor-high)
;; We know the range of the divisor, and the remainder must be
;; smaller than the divisor. We can tell the sign of the
- ;; remainer if we know the sign of the number.
+ ;; remainder if we know the sign of the number.
(let ((divisor-max (1- (max (abs divisor-low) (abs divisor-high)))))
`(integer ,(if (or (null number-low)
(minusp number-low))
divisor-max
0)))
;; The divisor is potentially either very positive or very
- ;; negative. Therefore, the remainer is unbounded, but we might
+ ;; negative. Therefore, the remainder is unbounded, but we might
;; be able to tell something about the sign from the number.
`(integer ,(if (and number-low (not (minusp number-low)))
;; The number we are dividing is positive.
(reoptimize-component (node-component node) :maybe))
(cut-node (node &aux did-something)
(when (and (not (block-delete-p (node-block node)))
+ (ref-p node)
+ (constant-p (ref-leaf node)))
+ (let* ((constant-value (constant-value (ref-leaf node)))
+ (new-value (if signedp
+ (mask-signed-field width constant-value)
+ (ldb (byte width 0) constant-value))))
+ (unless (= constant-value new-value)
+ (change-ref-leaf node (make-constant new-value))
+ (let ((lvar (node-lvar node)))
+ (setf (lvar-%derived-type lvar)
+ (and (lvar-has-single-use-p lvar)
+ (make-values-type :required (list (ctype-of new-value))))))
+ (setf (block-reoptimize (node-block node)) t)
+ (reoptimize-component (node-component node) :maybe)
+ (return-from cut-node t))))
+ (when (and (not (block-delete-p (node-block node)))
(combination-p node)
(eq (basic-combination-kind node) :known))
(let* ((fun-ref (lvar-use (combination-fun node)))
(best-modular-version width nil)
(when w
;; FIXME: This should be (CUT-TO-WIDTH NODE KIND WIDTH SIGNEDP).
- (cut-to-width x kind width signedp)
- (cut-to-width y kind width signedp)
- nil ; After fixing above, replace with T.
+ ;;
+ ;; FIXME: I think the FIXME (which is from APD) above
+ ;; implies that CUT-TO-WIDTH should do /everything/
+ ;; that's required, including reoptimizing things
+ ;; itself that it knows are necessary. At the moment,
+ ;; CUT-TO-WIDTH sets up some new calls with
+ ;; combination-type :FULL, which later get noticed as
+ ;; known functions and properly converted.
+ ;;
+ ;; We cut to W not WIDTH if SIGNEDP is true, because
+ ;; signed constant replacement needs to know which bit
+ ;; in the field is the signed bit.
+ (let ((xact (cut-to-width x kind (if signedp w width) signedp))
+ (yact (cut-to-width y kind (if signedp w width) signedp)))
+ (declare (ignore xact yact))
+ nil) ; After fixing above, replace with T, meaning
+ ; "don't reoptimize this (LOGAND) node any more".
))))))))
(defoptimizer (mask-signed-field optimizer) ((width x) node)
(when (and (numberp low) (numberp high))
(let ((width (max (integer-length high) (integer-length low))))
(multiple-value-bind (w kind)
- (best-modular-version width t)
+ (best-modular-version (1+ width) t)
(when w
- ;; FIXME: This should be (CUT-TO-WIDTH NODE KIND WIDTH T).
- (cut-to-width x kind width t)
+ ;; FIXME: This should be (CUT-TO-WIDTH NODE KIND W T).
+ ;; [ see comment above in LOGAND optimizer ]
+ (cut-to-width x kind w t)
nil ; After fixing above, replace with T.
))))))))
\f
`(- (ash x ,len))
`(ash x ,len))))
+;;; These must come before the ones below, so that they are tried
+;;; first. Since %FLOOR and %CEILING are inlined, this allows
+;;; the general case to be handled by TRUNCATE transforms.
+(deftransform floor ((x y))
+ `(%floor x y))
+
+(deftransform ceiling ((x y))
+ `(%ceiling x y))
+
;;; If arg is a constant power of two, turn FLOOR into a shift and
;;; mask. If CEILING, add in (1- (ABS Y)), do FLOOR and correct a
;;; remainder.
`(if (minusp x)
(- (logand (- x) ,mask))
(logand x ,mask)))))
+
+;;; Return an expression to calculate the integer quotient of X and
+;;; constant Y, using multiplication, shift and add/sub instead of
+;;; division. Both arguments must be unsigned, fit in a machine word and
+;;; Y must neither be zero nor a power of two. The quotient is rounded
+;;; towards zero.
+;;; The algorithm is taken from the paper "Division by Invariant
+;;; Integers using Multiplication", 1994 by Torbj\"{o}rn Granlund and
+;;; Peter L. Montgomery, Figures 4.2 and 6.2, modified to exclude the
+;;; case of division by powers of two.
+;;; The algorithm includes an adaptive precision argument. Use it, since
+;;; we often have sub-word value ranges. Careful, in this case, we need
+;;; p s.t 2^p > n, not the ceiling of the binary log.
+;;; Also, for some reason, the paper prefers shifting to masking. Mask
+;;; instead. Masking is equivalent to shifting right, then left again;
+;;; all the intermediate values are still words, so we just have to shift
+;;; right a bit more to compensate, at the end.
+;;;
+;;; The following two examples show an average case and the worst case
+;;; with respect to the complexity of the generated expression, under
+;;; a word size of 64 bits:
+;;;
+;;; (UNSIGNED-DIV-TRANSFORMER 10 MOST-POSITIVE-WORD) ->
+;;; (ASH (%MULTIPLY (LOGANDC2 X 0) 14757395258967641293) -3)
+;;;
+;;; (UNSIGNED-DIV-TRANSFORMER 7 MOST-POSITIVE-WORD) ->
+;;; (LET* ((NUM X)
+;;; (T1 (%MULTIPLY NUM 2635249153387078803)))
+;;; (ASH (LDB (BYTE 64 0)
+;;; (+ T1 (ASH (LDB (BYTE 64 0)
+;;; (- NUM T1))
+;;; -1)))
+;;; -2))
+;;;
+(defun gen-unsigned-div-by-constant-expr (y max-x)
+ (declare (type (integer 3 #.most-positive-word) y)
+ (type word max-x))
+ (aver (not (zerop (logand y (1- y)))))
+ (labels ((ld (x)
+ ;; the floor of the binary logarithm of (positive) X
+ (integer-length (1- x)))
+ (choose-multiplier (y precision)
+ (do* ((l (ld y))
+ (shift l (1- shift))
+ (expt-2-n+l (expt 2 (+ sb!vm:n-word-bits l)))
+ (m-low (truncate expt-2-n+l y) (ash m-low -1))
+ (m-high (truncate (+ expt-2-n+l
+ (ash expt-2-n+l (- precision)))
+ y)
+ (ash m-high -1)))
+ ((not (and (< (ash m-low -1) (ash m-high -1))
+ (> shift 0)))
+ (values m-high shift)))))
+ (let ((n (expt 2 sb!vm:n-word-bits))
+ (precision (integer-length max-x))
+ (shift1 0))
+ (multiple-value-bind (m shift2)
+ (choose-multiplier y precision)
+ (when (and (>= m n) (evenp y))
+ (setq shift1 (ld (logand y (- y))))
+ (multiple-value-setq (m shift2)
+ (choose-multiplier (/ y (ash 1 shift1))
+ (- precision shift1))))
+ (cond ((>= m n)
+ (flet ((word (x)
+ `(truly-the word ,x)))
+ `(let* ((num x)
+ (t1 (%multiply-high num ,(- m n))))
+ (ash ,(word `(+ t1 (ash ,(word `(- num t1))
+ -1)))
+ ,(- 1 shift2)))))
+ ((and (zerop shift1) (zerop shift2))
+ (let ((max (truncate max-x y)))
+ ;; Explicit TRULY-THE needed to get the FIXNUM=>FIXNUM
+ ;; VOP.
+ `(truly-the (integer 0 ,max)
+ (%multiply-high x ,m))))
+ (t
+ `(ash (%multiply-high (logandc2 x ,(1- (ash 1 shift1))) ,m)
+ ,(- (+ shift1 shift2)))))))))
+
+;;; If the divisor is constant and both args are positive and fit in a
+;;; machine word, replace the division by a multiplication and possibly
+;;; some shifts and an addition. Calculate the remainder by a second
+;;; multiplication and a subtraction. Dead code elimination will
+;;; suppress the latter part if only the quotient is needed. If the type
+;;; of the dividend allows to derive that the quotient will always have
+;;; the same value, emit much simpler code to handle that. (This case
+;;; may be rare but it's easy to detect and the compiler doesn't find
+;;; this optimization on its own.)
+(deftransform truncate ((x y) (word (constant-arg word))
+ *
+ :policy (and (> speed compilation-speed)
+ (> speed space)))
+ "convert integer division to multiplication"
+ (let* ((y (lvar-value y))
+ (x-type (lvar-type x))
+ (max-x (or (and (numeric-type-p x-type)
+ (numeric-type-high x-type))
+ most-positive-word)))
+ ;; Division by zero, one or powers of two is handled elsewhere.
+ (when (zerop (logand y (1- y)))
+ (give-up-ir1-transform))
+ `(let* ((quot ,(gen-unsigned-div-by-constant-expr y max-x))
+ (rem (ldb (byte #.sb!vm:n-word-bits 0)
+ (- x (* quot ,y)))))
+ (values quot rem))))
\f
;;;; arithmetic and logical identity operation elimination
(values (type= (numeric-contagion x y)
(numeric-contagion y y)))))))
+(def!type exact-number ()
+ '(or rational (complex rational)))
+
;;; Fold (+ x 0).
;;;
-;;; If y is not constant, not zerop, or is contagious, or a positive
-;;; float +0.0 then give up.
-(deftransform + ((x y) (t (constant-arg t)) *)
+;;; Only safely applicable for exact numbers. For floating-point
+;;; x, one would have to first show that neither x or y are signed
+;;; 0s, and that x isn't an SNaN.
+(deftransform + ((x y) (exact-number (constant-arg (eql 0))) *)
"fold zero arg"
- (let ((val (lvar-value y)))
- (unless (and (zerop val)
- (not (and (floatp val) (plusp (float-sign val))))
- (not-more-contagious y x))
- (give-up-ir1-transform)))
'x)
;;; Fold (- x 0).
-;;;
-;;; If y is not constant, not zerop, or is contagious, or a negative
-;;; float -0.0 then give up.
-(deftransform - ((x y) (t (constant-arg t)) *)
+(deftransform - ((x y) (exact-number (constant-arg (eql 0))) *)
"fold zero arg"
- (let ((val (lvar-value y)))
- (unless (and (zerop val)
- (not (and (floatp val) (minusp (float-sign val))))
- (not-more-contagious y x))
- (give-up-ir1-transform)))
'x)
;;; Fold (OP x +/-1)
-(macrolet ((def (name result minus-result)
- `(deftransform ,name ((x y) (t (constant-arg real)) *)
- "fold identity operations"
- (let ((val (lvar-value y)))
- (unless (and (= (abs val) 1)
- (not-more-contagious y x))
- (give-up-ir1-transform))
- (if (minusp val) ',minus-result ',result)))))
+;;;
+;;; %NEGATE might not always signal correctly.
+(macrolet
+ ((def (name result minus-result)
+ `(deftransform ,name ((x y)
+ (exact-number (constant-arg (member 1 -1))))
+ "fold identity operations"
+ (if (minusp (lvar-value y)) ',minus-result ',result))))
(def * x (%negate x))
(def / x (%negate x))
(def expt x (/ 1 x)))
(def round)
(def floor)
(def ceiling))
+
+(macrolet ((def (name &optional float)
+ (let ((x (if float '(float x) 'x)))
+ `(deftransform ,name ((x y) (integer (constant-arg (member 1 -1)))
+ *)
+ "fold division by 1"
+ `(values ,(if (minusp (lvar-value y))
+ '(%negate ,x)
+ ',x) 0)))))
+ (def truncate)
+ (def round)
+ (def floor)
+ (def ceiling)
+ (def ftruncate t)
+ (def fround t)
+ (def ffloor t)
+ (def fceiling t))
+
\f
;;;; character operations
(define-source-transform > (&rest args) (multi-compare '> args nil 'real))
;;; We cannot do the inversion for >= and <= here, since both
;;; (< NaN X) and (> NaN X)
-;;; are false, and we don't have type-inforation available yet. The
+;;; are false, and we don't have type-information available yet. The
;;; deftransforms for two-argument versions of >= and <= takes care of
;;; the inversion to > and < when possible.
(define-source-transform <= (&rest args) (multi-compare '<= args nil 'real))
;;;; versions, and degenerate cases are flushed.
;;; Left-associate FIRST-ARG and MORE-ARGS using FUNCTION.
-(declaim (ftype (function (symbol t list) list) associate-args))
-(defun associate-args (function first-arg more-args)
+(declaim (ftype (sfunction (symbol t list t) list) associate-args))
+(defun associate-args (fun first-arg more-args identity)
(let ((next (rest more-args))
(arg (first more-args)))
(if (null next)
- `(,function ,first-arg ,arg)
- (associate-args function `(,function ,first-arg ,arg) next))))
+ `(,fun ,first-arg ,(if arg arg identity))
+ (associate-args fun `(,fun ,first-arg ,arg) next identity))))
+
+;;; Reduce constants in ARGS list.
+(declaim (ftype (sfunction (symbol list t symbol) list) reduce-constants))
+(defun reduce-constants (fun args identity one-arg-result-type)
+ (let ((one-arg-constant-p (ecase one-arg-result-type
+ (number #'numberp)
+ (integer #'integerp)))
+ (reduced-value identity)
+ (reduced-p nil))
+ (collect ((not-constants))
+ (dolist (arg args)
+ (if (funcall one-arg-constant-p arg)
+ (setf reduced-value (funcall fun reduced-value arg)
+ reduced-p t)
+ (not-constants arg)))
+ ;; It is tempting to drop constants reduced to identity here,
+ ;; but if X is SNaN in (* X 1), we cannot drop the 1.
+ (if (not-constants)
+ (if reduced-p
+ `(,reduced-value ,@(not-constants))
+ (not-constants))
+ `(,reduced-value)))))
;;; Do source transformations for transitive functions such as +.
;;; One-arg cases are replaced with the arg and zero arg cases with
-;;; the identity. ONE-ARG-RESULT-TYPE is, if non-NIL, the type to
-;;; ensure (with THE) that the argument in one-argument calls is.
+;;; the identity. ONE-ARG-RESULT-TYPE is the type to ensure (with THE)
+;;; that the argument in one-argument calls is.
+(declaim (ftype (function (symbol list t &optional symbol list)
+ (values t &optional (member nil t)))
+ source-transform-transitive))
(defun source-transform-transitive (fun args identity
- &optional one-arg-result-type)
- (declare (symbol fun) (list args))
+ &optional (one-arg-result-type 'number)
+ (one-arg-prefixes '(values)))
(case (length args)
(0 identity)
- (1 (if one-arg-result-type
- `(values (the ,one-arg-result-type ,(first args)))
- `(values ,(first args))))
+ (1 `(,@one-arg-prefixes (the ,one-arg-result-type ,(first args))))
(2 (values nil t))
- (t
- (associate-args fun (first args) (rest args)))))
+ (t (let ((reduced-args (reduce-constants fun args identity one-arg-result-type)))
+ (associate-args fun (first reduced-args) (rest reduced-args) identity)))))
(define-source-transform + (&rest args)
- (source-transform-transitive '+ args 0 'number))
+ (source-transform-transitive '+ args 0))
(define-source-transform * (&rest args)
- (source-transform-transitive '* args 1 'number))
+ (source-transform-transitive '* args 1))
(define-source-transform logior (&rest args)
(source-transform-transitive 'logior args 0 'integer))
(define-source-transform logxor (&rest args)
(source-transform-transitive 'logand args -1 'integer))
(define-source-transform logeqv (&rest args)
(source-transform-transitive 'logeqv args -1 'integer))
-
-;;; Note: we can't use SOURCE-TRANSFORM-TRANSITIVE for GCD and LCM
-;;; because when they are given one argument, they return its absolute
-;;; value.
-
(define-source-transform gcd (&rest args)
- (case (length args)
- (0 0)
- (1 `(abs (the integer ,(first args))))
- (2 (values nil t))
- (t (associate-args 'gcd (first args) (rest args)))))
-
+ (source-transform-transitive 'gcd args 0 'integer '(abs)))
(define-source-transform lcm (&rest args)
- (case (length args)
- (0 1)
- (1 `(abs (the integer ,(first args))))
- (2 (values nil t))
- (t (associate-args 'lcm (first args) (rest args)))))
+ (source-transform-transitive 'lcm args 1 'integer '(abs)))
;;; 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)
+(declaim (ftype (function (symbol symbol list t list &optional symbol)
(values list &optional (member nil t)))
source-transform-intransitive))
-(defun source-transform-intransitive (function args inverse)
+(defun source-transform-intransitive (fun fun* args identity one-arg-prefixes
+ &optional (one-arg-result-type 'number))
(case (length args)
((0 2) (values nil t))
- (1 `(,@inverse ,(first args)))
- (t (associate-args function (first args) (rest args)))))
+ (1 `(,@one-arg-prefixes (the ,one-arg-result-type ,(first args))))
+ (t (let ((reduced-args
+ (reduce-constants fun* (rest args) identity one-arg-result-type)))
+ (associate-args fun (first args) reduced-args identity)))))
(define-source-transform - (&rest args)
- (source-transform-intransitive '- args '(%negate)))
+ (source-transform-intransitive '- '+ args 0 '(%negate)))
(define-source-transform / (&rest args)
- (source-transform-intransitive '/ args '(/ 1)))
+ (source-transform-intransitive '/ '* args 1 '(/ 1)))
\f
;;;; transforming APPLY
(define-source-transform apply (fun arg &rest more-args)
(let ((args (cons arg more-args)))
`(multiple-value-call ,fun
- ,@(mapcar (lambda (x)
- `(values ,x))
- (butlast args))
+ ,@(mapcar (lambda (x) `(values ,x)) (butlast args))
(values-list ,(car (last args))))))
+
+;;;; transforming references to &REST argument
+
+;;; We add magical &MORE arguments to all functions with &REST. If ARG names
+;;; the &REST argument, this returns the lambda-vars for the context and
+;;; count.
+(defun possible-rest-arg-context (arg)
+ (when (symbolp arg)
+ (let* ((var (lexenv-find arg vars))
+ (info (when (lambda-var-p var)
+ (lambda-var-arg-info var))))
+ (when (and info
+ (eq :rest (arg-info-kind info))
+ (consp (arg-info-default info)))
+ (values-list (arg-info-default info))))))
+
+(defun mark-more-context-used (rest-var)
+ (let ((info (lambda-var-arg-info rest-var)))
+ (aver (eq :rest (arg-info-kind info)))
+ (destructuring-bind (context count &optional used) (arg-info-default info)
+ (unless used
+ (setf (arg-info-default info) (list context count t))))))
+
+(defun mark-more-context-invalid (rest-var)
+ (let ((info (lambda-var-arg-info rest-var)))
+ (aver (eq :rest (arg-info-kind info)))
+ (setf (arg-info-default info) t)))
+
+;;; This determines of we the REF to a &REST variable is headed towards
+;;; parts unknown, or if we can really use the context.
+(defun rest-var-more-context-ok (lvar)
+ (let* ((use (lvar-use lvar))
+ (var (when (ref-p use) (ref-leaf use)))
+ (home (when (lambda-var-p var) (lambda-var-home var)))
+ (info (when (lambda-var-p var) (lambda-var-arg-info var)))
+ (restp (when info (eq :rest (arg-info-kind info)))))
+ (flet ((ref-good-for-more-context-p (ref)
+ (let ((dest (principal-lvar-end (node-lvar ref))))
+ (and (combination-p dest)
+ ;; If the destination is to anything but these, we're going to
+ ;; actually need the rest list -- and since other operations
+ ;; might modify the list destructively, the using the context
+ ;; isn't good anywhere else either.
+ (lvar-fun-is (combination-fun dest)
+ '(%rest-values %rest-ref %rest-length
+ %rest-null %rest-true))
+ ;; If the home lambda is different and isn't DX, it might
+ ;; escape -- in which case using the more context isn't safe.
+ (let ((clambda (node-home-lambda dest)))
+ (or (eq home clambda)
+ (leaf-dynamic-extent clambda)))))))
+ (let ((ok (and restp
+ (consp (arg-info-default info))
+ (not (lambda-var-specvar var))
+ (not (lambda-var-sets var))
+ (every #'ref-good-for-more-context-p (lambda-var-refs var)))))
+ (if ok
+ (mark-more-context-used var)
+ (when restp
+ (mark-more-context-invalid var)))
+ ok))))
+
+;;; VALUES-LIST -> %REST-VALUES
+(define-source-transform values-list (list)
+ (multiple-value-bind (context count) (possible-rest-arg-context list)
+ (if context
+ `(%rest-values ,list ,context ,count)
+ (values nil t))))
+
+;;; NTH -> %REST-REF
+(define-source-transform nth (n list)
+ (multiple-value-bind (context count) (possible-rest-arg-context list)
+ (if context
+ `(%rest-ref ,n ,list ,context ,count)
+ `(car (nthcdr ,n ,list)))))
+
+(define-source-transform elt (seq n)
+ (if (policy *lexenv* (= safety 3))
+ (values nil t)
+ (multiple-value-bind (context count) (possible-rest-arg-context seq)
+ (if context
+ `(%rest-ref ,n ,seq ,context ,count)
+ (values nil t)))))
+
+;;; CAxR -> %REST-REF
+(defun source-transform-car (list nth)
+ (multiple-value-bind (context count) (possible-rest-arg-context list)
+ (if context
+ `(%rest-ref ,nth ,list ,context ,count)
+ (values nil t))))
+
+(define-source-transform car (list)
+ (source-transform-car list 0))
+
+(define-source-transform cadr (list)
+ (or (source-transform-car list 1)
+ `(car (cdr ,list))))
+
+(define-source-transform caddr (list)
+ (or (source-transform-car list 2)
+ `(car (cdr (cdr ,list)))))
+
+(define-source-transform cadddr (list)
+ (or (source-transform-car list 3)
+ `(car (cdr (cdr (cdr ,list))))))
+
+;;; LENGTH -> %REST-LENGTH
+(defun source-transform-length (list)
+ (multiple-value-bind (context count) (possible-rest-arg-context list)
+ (if context
+ `(%rest-length ,list ,context ,count)
+ (values nil t))))
+(define-source-transform length (list) (source-transform-length list))
+(define-source-transform list-length (list) (source-transform-length list))
+
+;;; ENDP, NULL and NOT -> %REST-NULL
+;;;
+;;; Outside &REST convert into an IF so that IF optimizations will eliminate
+;;; redundant negations.
+(defun source-transform-null (x op)
+ (multiple-value-bind (context count) (possible-rest-arg-context x)
+ (cond (context
+ `(%rest-null ',op ,x ,context ,count))
+ ((eq 'endp op)
+ `(if (the list ,x) nil t))
+ (t
+ `(if ,x nil t)))))
+(define-source-transform not (x) (source-transform-null x 'not))
+(define-source-transform null (x) (source-transform-null x 'null))
+(define-source-transform endp (x) (source-transform-null x 'endp))
+
+(deftransform %rest-values ((list context count))
+ (if (rest-var-more-context-ok list)
+ `(%more-arg-values context 0 count)
+ `(values-list list)))
+
+(deftransform %rest-ref ((n list context count))
+ (cond ((rest-var-more-context-ok list)
+ `(and (< (the index n) count)
+ (%more-arg context n)))
+ ((and (constant-lvar-p n) (zerop (lvar-value n)))
+ `(car list))
+ (t
+ `(nth n list))))
+
+(deftransform %rest-length ((list context count))
+ (if (rest-var-more-context-ok list)
+ 'count
+ `(length list)))
+
+(deftransform %rest-null ((op list context count))
+ (aver (constant-lvar-p op))
+ (if (rest-var-more-context-ok list)
+ `(eql 0 count)
+ `(,(lvar-value op) list)))
+
+(deftransform %rest-true ((list context count))
+ (if (rest-var-more-context-ok list)
+ `(not (eql 0 count))
+ `list))
\f
;;;; transforming FORMAT
;;;;
:format-arguments
(list nargs 'cerror y x (max max1 max2))))))))))))))
-(defoptimizer (coerce derive-type) ((value type))
+(defoptimizer (coerce derive-type) ((value type) node)
(cond
((constant-lvar-p type)
;; This branch is essentially (RESULT-TYPE-SPECIFIER-NTH-ARG 2),
(type-union result-typeoid
(type-intersection (lvar-type value)
(specifier-type 'rational))))))
- (t result-typeoid))))
+ ((and (policy node (zerop safety))
+ (csubtypep result-typeoid (specifier-type '(array * (*)))))
+ ;; At zero safety the deftransform for COERCE can elide dimension
+ ;; checks for the things like (COERCE X '(SIMPLE-VECTOR 5)) -- so we
+ ;; need to simplify the type to drop the dimension information.
+ (let ((vtype (simplify-vector-type result-typeoid)))
+ (if vtype
+ (specifier-type vtype)
+ result-typeoid)))
+ (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