1 ;;;; This file contains macro-like source transformations which
2 ;;;; convert uses of certain functions into the canonical form desired
3 ;;;; within the compiler. ### and other IR1 transforms and stuff.
5 ;;;; This software is part of the SBCL system. See the README file for
8 ;;;; This software is derived from the CMU CL system, which was
9 ;;;; written at Carnegie Mellon University and released into the
10 ;;;; public domain. The software is in the public domain and is
11 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
12 ;;;; files for more information.
16 ;;; Convert into an IF so that IF optimizations will eliminate redundant
18 (define-source-transform not (x) `(if ,x nil t))
19 (define-source-transform null (x) `(if ,x nil t))
21 ;;; ENDP is just NULL with a LIST assertion. The assertion will be
22 ;;; optimized away when SAFETY optimization is low; hopefully that
23 ;;; is consistent with ANSI's "should return an error".
24 (define-source-transform endp (x) `(null (the list ,x)))
26 ;;; We turn IDENTITY into PROG1 so that it is obvious that it just
27 ;;; returns the first value of its argument. Ditto for VALUES with one
29 (define-source-transform identity (x) `(prog1 ,x))
30 (define-source-transform values (x) `(prog1 ,x))
32 ;;; Bind the values and make a closure that returns them.
33 (define-source-transform constantly (value)
34 (let ((rest (gensym "CONSTANTLY-REST-")))
35 `(lambda (&rest ,rest)
36 (declare (ignore ,rest))
39 ;;; If the function has a known number of arguments, then return a
40 ;;; lambda with the appropriate fixed number of args. If the
41 ;;; destination is a FUNCALL, then do the &REST APPLY thing, and let
42 ;;; MV optimization figure things out.
43 (deftransform complement ((fun) * * :node node :when :both)
45 (multiple-value-bind (min max)
46 (fun-type-nargs (continuation-type fun))
48 ((and min (eql min max))
49 (let ((dums (make-gensym-list min)))
50 `#'(lambda ,dums (not (funcall fun ,@dums)))))
51 ((let* ((cont (node-cont node))
52 (dest (continuation-dest cont)))
53 (and (combination-p dest)
54 (eq (combination-fun dest) cont)))
55 '#'(lambda (&rest args)
56 (not (apply fun args))))
58 (give-up-ir1-transform
59 "The function doesn't have a fixed argument count.")))))
63 ;;; Translate CxR into CAR/CDR combos.
64 (defun source-transform-cxr (form)
65 (if (/= (length form) 2)
67 (let ((name (symbol-name (car form))))
68 (do ((i (- (length name) 2) (1- i))
70 `(,(ecase (char name i)
76 ;;; Make source transforms to turn CxR forms into combinations of CAR
77 ;;; and CDR. ANSI specifies that everything up to 4 A/D operations is
79 (/show0 "about to set CxR source transforms")
80 (loop for i of-type index from 2 upto 4 do
81 ;; Iterate over BUF = all names CxR where x = an I-element
82 ;; string of #\A or #\D characters.
83 (let ((buf (make-string (+ 2 i))))
84 (setf (aref buf 0) #\C
85 (aref buf (1+ i)) #\R)
86 (dotimes (j (ash 2 i))
87 (declare (type index j))
89 (declare (type index k))
90 (setf (aref buf (1+ k))
91 (if (logbitp k j) #\A #\D)))
92 (setf (info :function :source-transform (intern buf))
93 #'source-transform-cxr))))
94 (/show0 "done setting CxR source transforms")
96 ;;; Turn FIRST..FOURTH and REST into the obvious synonym, assuming
97 ;;; whatever is right for them is right for us. FIFTH..TENTH turn into
98 ;;; Nth, which can be expanded into a CAR/CDR later on if policy
100 (define-source-transform first (x) `(car ,x))
101 (define-source-transform rest (x) `(cdr ,x))
102 (define-source-transform second (x) `(cadr ,x))
103 (define-source-transform third (x) `(caddr ,x))
104 (define-source-transform fourth (x) `(cadddr ,x))
105 (define-source-transform fifth (x) `(nth 4 ,x))
106 (define-source-transform sixth (x) `(nth 5 ,x))
107 (define-source-transform seventh (x) `(nth 6 ,x))
108 (define-source-transform eighth (x) `(nth 7 ,x))
109 (define-source-transform ninth (x) `(nth 8 ,x))
110 (define-source-transform tenth (x) `(nth 9 ,x))
112 ;;; Translate RPLACx to LET and SETF.
113 (define-source-transform rplaca (x y)
118 (define-source-transform rplacd (x y)
124 (define-source-transform nth (n l) `(car (nthcdr ,n ,l)))
126 (defvar *default-nthcdr-open-code-limit* 6)
127 (defvar *extreme-nthcdr-open-code-limit* 20)
129 (deftransform nthcdr ((n l) (unsigned-byte t) * :node node)
130 "convert NTHCDR to CAxxR"
131 (unless (constant-continuation-p n)
132 (give-up-ir1-transform))
133 (let ((n (continuation-value n)))
135 (if (policy node (and (= speed 3) (= space 0)))
136 *extreme-nthcdr-open-code-limit*
137 *default-nthcdr-open-code-limit*))
138 (give-up-ir1-transform))
143 `(cdr ,(frob (1- n))))))
146 ;;;; arithmetic and numerology
148 (define-source-transform plusp (x) `(> ,x 0))
149 (define-source-transform minusp (x) `(< ,x 0))
150 (define-source-transform zerop (x) `(= ,x 0))
152 (define-source-transform 1+ (x) `(+ ,x 1))
153 (define-source-transform 1- (x) `(- ,x 1))
155 (define-source-transform oddp (x) `(not (zerop (logand ,x 1))))
156 (define-source-transform evenp (x) `(zerop (logand ,x 1)))
158 ;;; Note that all the integer division functions are available for
159 ;;; inline expansion.
161 (macrolet ((deffrob (fun)
162 `(define-source-transform ,fun (x &optional (y nil y-p))
169 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
171 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
174 (define-source-transform lognand (x y) `(lognot (logand ,x ,y)))
175 (define-source-transform lognor (x y) `(lognot (logior ,x ,y)))
176 (define-source-transform logandc1 (x y) `(logand (lognot ,x) ,y))
177 (define-source-transform logandc2 (x y) `(logand ,x (lognot ,y)))
178 (define-source-transform logorc1 (x y) `(logior (lognot ,x) ,y))
179 (define-source-transform logorc2 (x y) `(logior ,x (lognot ,y)))
180 (define-source-transform logtest (x y) `(not (zerop (logand ,x ,y))))
181 (define-source-transform logbitp (index integer)
182 `(not (zerop (logand (ash 1 ,index) ,integer))))
183 (define-source-transform byte (size position) `(cons ,size ,position))
184 (define-source-transform byte-size (spec) `(car ,spec))
185 (define-source-transform byte-position (spec) `(cdr ,spec))
186 (define-source-transform ldb-test (bytespec integer)
187 `(not (zerop (mask-field ,bytespec ,integer))))
189 ;;; With the ratio and complex accessors, we pick off the "identity"
190 ;;; case, and use a primitive to handle the cell access case.
191 (define-source-transform numerator (num)
192 (once-only ((n-num `(the rational ,num)))
196 (define-source-transform denominator (num)
197 (once-only ((n-num `(the rational ,num)))
199 (%denominator ,n-num)
202 ;;;; interval arithmetic for computing bounds
204 ;;;; This is a set of routines for operating on intervals. It
205 ;;;; implements a simple interval arithmetic package. Although SBCL
206 ;;;; has an interval type in NUMERIC-TYPE, we choose to use our own
207 ;;;; for two reasons:
209 ;;;; 1. This package is simpler than NUMERIC-TYPE.
211 ;;;; 2. It makes debugging much easier because you can just strip
212 ;;;; out these routines and test them independently of SBCL. (This is a
215 ;;;; One disadvantage is a probable increase in consing because we
216 ;;;; have to create these new interval structures even though
217 ;;;; numeric-type has everything we want to know. Reason 2 wins for
220 ;;; The basic interval type. It can handle open and closed intervals.
221 ;;; A bound is open if it is a list containing a number, just like
222 ;;; Lisp says. NIL means unbounded.
223 (defstruct (interval (:constructor %make-interval)
227 (defun make-interval (&key low high)
228 (labels ((normalize-bound (val)
229 (cond ((and (floatp val)
230 (float-infinity-p val))
231 ;; Handle infinities.
235 ;; Handle any closed bounds.
238 ;; We have an open bound. Normalize the numeric
239 ;; bound. If the normalized bound is still a number
240 ;; (not nil), keep the bound open. Otherwise, the
241 ;; bound is really unbounded, so drop the openness.
242 (let ((new-val (normalize-bound (first val))))
244 ;; The bound exists, so keep it open still.
247 (error "unknown bound type in MAKE-INTERVAL")))))
248 (%make-interval :low (normalize-bound low)
249 :high (normalize-bound high))))
251 ;;; Given a number X, create a form suitable as a bound for an
252 ;;; interval. Make the bound open if OPEN-P is T. NIL remains NIL.
253 #!-sb-fluid (declaim (inline set-bound))
254 (defun set-bound (x open-p)
255 (if (and x open-p) (list x) x))
257 ;;; Apply the function F to a bound X. If X is an open bound, then
258 ;;; the result will be open. IF X is NIL, the result is NIL.
259 (defun bound-func (f x)
261 (with-float-traps-masked (:underflow :overflow :inexact :divide-by-zero)
262 ;; With these traps masked, we might get things like infinity
263 ;; or negative infinity returned. Check for this and return
264 ;; NIL to indicate unbounded.
265 (let ((y (funcall f (type-bound-number x))))
267 (float-infinity-p y))
269 (set-bound (funcall f (type-bound-number x)) (consp x)))))))
271 ;;; Apply a binary operator OP to two bounds X and Y. The result is
272 ;;; NIL if either is NIL. Otherwise bound is computed and the result
273 ;;; is open if either X or Y is open.
275 ;;; FIXME: only used in this file, not needed in target runtime
276 (defmacro bound-binop (op x y)
278 (with-float-traps-masked (:underflow :overflow :inexact :divide-by-zero)
279 (set-bound (,op (type-bound-number ,x)
280 (type-bound-number ,y))
281 (or (consp ,x) (consp ,y))))))
283 ;;; Convert a numeric-type object to an interval object.
284 (defun numeric-type->interval (x)
285 (declare (type numeric-type x))
286 (make-interval :low (numeric-type-low x)
287 :high (numeric-type-high x)))
289 (defun copy-interval-limit (limit)
294 (defun copy-interval (x)
295 (declare (type interval x))
296 (make-interval :low (copy-interval-limit (interval-low x))
297 :high (copy-interval-limit (interval-high x))))
299 ;;; Given a point P contained in the interval X, split X into two
300 ;;; interval at the point P. If CLOSE-LOWER is T, then the left
301 ;;; interval contains P. If CLOSE-UPPER is T, the right interval
302 ;;; contains P. You can specify both to be T or NIL.
303 (defun interval-split (p x &optional close-lower close-upper)
304 (declare (type number p)
306 (list (make-interval :low (copy-interval-limit (interval-low x))
307 :high (if close-lower p (list p)))
308 (make-interval :low (if close-upper (list p) p)
309 :high (copy-interval-limit (interval-high x)))))
311 ;;; Return the closure of the interval. That is, convert open bounds
312 ;;; to closed bounds.
313 (defun interval-closure (x)
314 (declare (type interval x))
315 (make-interval :low (type-bound-number (interval-low x))
316 :high (type-bound-number (interval-high x))))
318 (defun signed-zero->= (x y)
322 (>= (float-sign (float x))
323 (float-sign (float y))))))
325 ;;; For an interval X, if X >= POINT, return '+. If X <= POINT, return
326 ;;; '-. Otherwise return NIL.
328 (defun interval-range-info (x &optional (point 0))
329 (declare (type interval x))
330 (let ((lo (interval-low x))
331 (hi (interval-high x)))
332 (cond ((and lo (signed-zero->= (type-bound-number lo) point))
334 ((and hi (signed-zero->= point (type-bound-number hi)))
338 (defun interval-range-info (x &optional (point 0))
339 (declare (type interval x))
340 (labels ((signed->= (x y)
341 (if (and (zerop x) (zerop y) (floatp x) (floatp y))
342 (>= (float-sign x) (float-sign y))
344 (let ((lo (interval-low x))
345 (hi (interval-high x)))
346 (cond ((and lo (signed->= (type-bound-number lo) point))
348 ((and hi (signed->= point (type-bound-number hi)))
353 ;;; Test to see whether the interval X is bounded. HOW determines the
354 ;;; test, and should be either ABOVE, BELOW, or BOTH.
355 (defun interval-bounded-p (x how)
356 (declare (type interval x))
363 (and (interval-low x) (interval-high x)))))
365 ;;; signed zero comparison functions. Use these functions if we need
366 ;;; to distinguish between signed zeroes.
367 (defun signed-zero-< (x y)
371 (< (float-sign (float x))
372 (float-sign (float y))))))
373 (defun signed-zero-> (x y)
377 (> (float-sign (float x))
378 (float-sign (float y))))))
379 (defun signed-zero-= (x y)
382 (= (float-sign (float x))
383 (float-sign (float y)))))
384 (defun signed-zero-<= (x y)
388 (<= (float-sign (float x))
389 (float-sign (float y))))))
391 ;;; See whether the interval X contains the number P, taking into
392 ;;; account that the interval might not be closed.
393 (defun interval-contains-p (p x)
394 (declare (type number p)
396 ;; Does the interval X contain the number P? This would be a lot
397 ;; easier if all intervals were closed!
398 (let ((lo (interval-low x))
399 (hi (interval-high x)))
401 ;; The interval is bounded
402 (if (and (signed-zero-<= (type-bound-number lo) p)
403 (signed-zero-<= p (type-bound-number hi)))
404 ;; P is definitely in the closure of the interval.
405 ;; We just need to check the end points now.
406 (cond ((signed-zero-= p (type-bound-number lo))
408 ((signed-zero-= p (type-bound-number hi))
413 ;; Interval with upper bound
414 (if (signed-zero-< p (type-bound-number hi))
416 (and (numberp hi) (signed-zero-= p hi))))
418 ;; Interval with lower bound
419 (if (signed-zero-> p (type-bound-number lo))
421 (and (numberp lo) (signed-zero-= p lo))))
423 ;; Interval with no bounds
426 ;;; Determine whether two intervals X and Y intersect. Return T if so.
427 ;;; If CLOSED-INTERVALS-P is T, the treat the intervals as if they
428 ;;; were closed. Otherwise the intervals are treated as they are.
430 ;;; Thus if X = [0, 1) and Y = (1, 2), then they do not intersect
431 ;;; because no element in X is in Y. However, if CLOSED-INTERVALS-P
432 ;;; is T, then they do intersect because we use the closure of X = [0,
433 ;;; 1] and Y = [1, 2] to determine intersection.
434 (defun interval-intersect-p (x y &optional closed-intervals-p)
435 (declare (type interval x y))
436 (multiple-value-bind (intersect diff)
437 (interval-intersection/difference (if closed-intervals-p
440 (if closed-intervals-p
443 (declare (ignore diff))
446 ;;; Are the two intervals adjacent? That is, is there a number
447 ;;; between the two intervals that is not an element of either
448 ;;; interval? If so, they are not adjacent. For example [0, 1) and
449 ;;; [1, 2] are adjacent but [0, 1) and (1, 2] are not because 1 lies
450 ;;; between both intervals.
451 (defun interval-adjacent-p (x y)
452 (declare (type interval x y))
453 (flet ((adjacent (lo hi)
454 ;; Check to see whether lo and hi are adjacent. If either is
455 ;; nil, they can't be adjacent.
456 (when (and lo hi (= (type-bound-number lo) (type-bound-number hi)))
457 ;; The bounds are equal. They are adjacent if one of
458 ;; them is closed (a number). If both are open (consp),
459 ;; then there is a number that lies between them.
460 (or (numberp lo) (numberp hi)))))
461 (or (adjacent (interval-low y) (interval-high x))
462 (adjacent (interval-low x) (interval-high y)))))
464 ;;; Compute the intersection and difference between two intervals.
465 ;;; Two values are returned: the intersection and the difference.
467 ;;; Let the two intervals be X and Y, and let I and D be the two
468 ;;; values returned by this function. Then I = X intersect Y. If I
469 ;;; is NIL (the empty set), then D is X union Y, represented as the
470 ;;; list of X and Y. If I is not the empty set, then D is (X union Y)
471 ;;; - I, which is a list of two intervals.
473 ;;; For example, let X = [1,5] and Y = [-1,3). Then I = [1,3) and D =
474 ;;; [-1,1) union [3,5], which is returned as a list of two intervals.
475 (defun interval-intersection/difference (x y)
476 (declare (type interval x y))
477 (let ((x-lo (interval-low x))
478 (x-hi (interval-high x))
479 (y-lo (interval-low y))
480 (y-hi (interval-high y)))
483 ;; If p is an open bound, make it closed. If p is a closed
484 ;; bound, make it open.
489 ;; Test whether P is in the interval.
490 (when (interval-contains-p (type-bound-number p)
491 (interval-closure int))
492 (let ((lo (interval-low int))
493 (hi (interval-high int)))
494 ;; Check for endpoints.
495 (cond ((and lo (= (type-bound-number p) (type-bound-number lo)))
496 (not (and (consp p) (numberp lo))))
497 ((and hi (= (type-bound-number p) (type-bound-number hi)))
498 (not (and (numberp p) (consp hi))))
500 (test-lower-bound (p int)
501 ;; P is a lower bound of an interval.
504 (not (interval-bounded-p int 'below))))
505 (test-upper-bound (p int)
506 ;; P is an upper bound of an interval.
509 (not (interval-bounded-p int 'above)))))
510 (let ((x-lo-in-y (test-lower-bound x-lo y))
511 (x-hi-in-y (test-upper-bound x-hi y))
512 (y-lo-in-x (test-lower-bound y-lo x))
513 (y-hi-in-x (test-upper-bound y-hi x)))
514 (cond ((or x-lo-in-y x-hi-in-y y-lo-in-x y-hi-in-x)
515 ;; Intervals intersect. Let's compute the intersection
516 ;; and the difference.
517 (multiple-value-bind (lo left-lo left-hi)
518 (cond (x-lo-in-y (values x-lo y-lo (opposite-bound x-lo)))
519 (y-lo-in-x (values y-lo x-lo (opposite-bound y-lo))))
520 (multiple-value-bind (hi right-lo right-hi)
522 (values x-hi (opposite-bound x-hi) y-hi))
524 (values y-hi (opposite-bound y-hi) x-hi)))
525 (values (make-interval :low lo :high hi)
526 (list (make-interval :low left-lo
528 (make-interval :low right-lo
531 (values nil (list x y))))))))
533 ;;; If intervals X and Y intersect, return a new interval that is the
534 ;;; union of the two. If they do not intersect, return NIL.
535 (defun interval-merge-pair (x y)
536 (declare (type interval x y))
537 ;; If x and y intersect or are adjacent, create the union.
538 ;; Otherwise return nil
539 (when (or (interval-intersect-p x y)
540 (interval-adjacent-p x y))
541 (flet ((select-bound (x1 x2 min-op max-op)
542 (let ((x1-val (type-bound-number x1))
543 (x2-val (type-bound-number x2)))
545 ;; Both bounds are finite. Select the right one.
546 (cond ((funcall min-op x1-val x2-val)
547 ;; x1 is definitely better.
549 ((funcall max-op x1-val x2-val)
550 ;; x2 is definitely better.
553 ;; Bounds are equal. Select either
554 ;; value and make it open only if
556 (set-bound x1-val (and (consp x1) (consp x2))))))
558 ;; At least one bound is not finite. The
559 ;; non-finite bound always wins.
561 (let* ((x-lo (copy-interval-limit (interval-low x)))
562 (x-hi (copy-interval-limit (interval-high x)))
563 (y-lo (copy-interval-limit (interval-low y)))
564 (y-hi (copy-interval-limit (interval-high y))))
565 (make-interval :low (select-bound x-lo y-lo #'< #'>)
566 :high (select-bound x-hi y-hi #'> #'<))))))
568 ;;; basic arithmetic operations on intervals. We probably should do
569 ;;; true interval arithmetic here, but it's complicated because we
570 ;;; have float and integer types and bounds can be open or closed.
572 ;;; the negative of an interval
573 (defun interval-neg (x)
574 (declare (type interval x))
575 (make-interval :low (bound-func #'- (interval-high x))
576 :high (bound-func #'- (interval-low x))))
578 ;;; Add two intervals.
579 (defun interval-add (x y)
580 (declare (type interval x y))
581 (make-interval :low (bound-binop + (interval-low x) (interval-low y))
582 :high (bound-binop + (interval-high x) (interval-high y))))
584 ;;; Subtract two intervals.
585 (defun interval-sub (x y)
586 (declare (type interval x y))
587 (make-interval :low (bound-binop - (interval-low x) (interval-high y))
588 :high (bound-binop - (interval-high x) (interval-low y))))
590 ;;; Multiply two intervals.
591 (defun interval-mul (x y)
592 (declare (type interval x y))
593 (flet ((bound-mul (x y)
594 (cond ((or (null x) (null y))
595 ;; Multiply by infinity is infinity
597 ((or (and (numberp x) (zerop x))
598 (and (numberp y) (zerop y)))
599 ;; Multiply by closed zero is special. The result
600 ;; is always a closed bound. But don't replace this
601 ;; with zero; we want the multiplication to produce
602 ;; the correct signed zero, if needed.
603 (* (type-bound-number x) (type-bound-number y)))
604 ((or (and (floatp x) (float-infinity-p x))
605 (and (floatp y) (float-infinity-p y)))
606 ;; Infinity times anything is infinity
609 ;; General multiply. The result is open if either is open.
610 (bound-binop * x y)))))
611 (let ((x-range (interval-range-info x))
612 (y-range (interval-range-info y)))
613 (cond ((null x-range)
614 ;; Split x into two and multiply each separately
615 (destructuring-bind (x- x+) (interval-split 0 x t t)
616 (interval-merge-pair (interval-mul x- y)
617 (interval-mul x+ y))))
619 ;; Split y into two and multiply each separately
620 (destructuring-bind (y- y+) (interval-split 0 y t t)
621 (interval-merge-pair (interval-mul x y-)
622 (interval-mul x y+))))
624 (interval-neg (interval-mul (interval-neg x) y)))
626 (interval-neg (interval-mul x (interval-neg y))))
627 ((and (eq x-range '+) (eq y-range '+))
628 ;; If we are here, X and Y are both positive.
630 :low (bound-mul (interval-low x) (interval-low y))
631 :high (bound-mul (interval-high x) (interval-high y))))
633 (error "internal error in INTERVAL-MUL"))))))
635 ;;; Divide two intervals.
636 (defun interval-div (top bot)
637 (declare (type interval top bot))
638 (flet ((bound-div (x y y-low-p)
641 ;; Divide by infinity means result is 0. However,
642 ;; we need to watch out for the sign of the result,
643 ;; to correctly handle signed zeros. We also need
644 ;; to watch out for positive or negative infinity.
645 (if (floatp (type-bound-number x))
647 (- (float-sign (type-bound-number x) 0.0))
648 (float-sign (type-bound-number x) 0.0))
650 ((zerop (type-bound-number y))
651 ;; Divide by zero means result is infinity
653 ((and (numberp x) (zerop x))
654 ;; Zero divided by anything is zero.
657 (bound-binop / x y)))))
658 (let ((top-range (interval-range-info top))
659 (bot-range (interval-range-info bot)))
660 (cond ((null bot-range)
661 ;; The denominator contains zero, so anything goes!
662 (make-interval :low nil :high nil))
664 ;; Denominator is negative so flip the sign, compute the
665 ;; result, and flip it back.
666 (interval-neg (interval-div top (interval-neg bot))))
668 ;; Split top into two positive and negative parts, and
669 ;; divide each separately
670 (destructuring-bind (top- top+) (interval-split 0 top t t)
671 (interval-merge-pair (interval-div top- bot)
672 (interval-div top+ bot))))
674 ;; Top is negative so flip the sign, divide, and flip the
675 ;; sign of the result.
676 (interval-neg (interval-div (interval-neg top) bot)))
677 ((and (eq top-range '+) (eq bot-range '+))
680 :low (bound-div (interval-low top) (interval-high bot) t)
681 :high (bound-div (interval-high top) (interval-low bot) nil)))
683 (error "internal error in INTERVAL-DIV"))))))
685 ;;; Apply the function F to the interval X. If X = [a, b], then the
686 ;;; result is [f(a), f(b)]. It is up to the user to make sure the
687 ;;; result makes sense. It will if F is monotonic increasing (or
689 (defun interval-func (f x)
690 (declare (type interval x))
691 (let ((lo (bound-func f (interval-low x)))
692 (hi (bound-func f (interval-high x))))
693 (make-interval :low lo :high hi)))
695 ;;; Return T if X < Y. That is every number in the interval X is
696 ;;; always less than any number in the interval Y.
697 (defun interval-< (x y)
698 (declare (type interval x y))
699 ;; X < Y only if X is bounded above, Y is bounded below, and they
701 (when (and (interval-bounded-p x 'above)
702 (interval-bounded-p y 'below))
703 ;; Intervals are bounded in the appropriate way. Make sure they
705 (let ((left (interval-high x))
706 (right (interval-low y)))
707 (cond ((> (type-bound-number left)
708 (type-bound-number right))
709 ;; The intervals definitely overlap, so result is NIL.
711 ((< (type-bound-number left)
712 (type-bound-number right))
713 ;; The intervals definitely don't touch, so result is T.
716 ;; Limits are equal. Check for open or closed bounds.
717 ;; Don't overlap if one or the other are open.
718 (or (consp left) (consp right)))))))
720 ;;; Return T if X >= Y. That is, every number in the interval X is
721 ;;; always greater than any number in the interval Y.
722 (defun interval->= (x y)
723 (declare (type interval x y))
724 ;; X >= Y if lower bound of X >= upper bound of Y
725 (when (and (interval-bounded-p x 'below)
726 (interval-bounded-p y 'above))
727 (>= (type-bound-number (interval-low x))
728 (type-bound-number (interval-high y)))))
730 ;;; Return an interval that is the absolute value of X. Thus, if
731 ;;; X = [-1 10], the result is [0, 10].
732 (defun interval-abs (x)
733 (declare (type interval x))
734 (case (interval-range-info x)
740 (destructuring-bind (x- x+) (interval-split 0 x t t)
741 (interval-merge-pair (interval-neg x-) x+)))))
743 ;;; Compute the square of an interval.
744 (defun interval-sqr (x)
745 (declare (type interval x))
746 (interval-func (lambda (x) (* x x))
749 ;;;; numeric DERIVE-TYPE methods
751 ;;; a utility for defining derive-type methods of integer operations. If
752 ;;; the types of both X and Y are integer types, then we compute a new
753 ;;; integer type with bounds determined Fun when applied to X and Y.
754 ;;; Otherwise, we use Numeric-Contagion.
755 (defun derive-integer-type (x y fun)
756 (declare (type continuation x y) (type function fun))
757 (let ((x (continuation-type x))
758 (y (continuation-type y)))
759 (if (and (numeric-type-p x) (numeric-type-p y)
760 (eq (numeric-type-class x) 'integer)
761 (eq (numeric-type-class y) 'integer)
762 (eq (numeric-type-complexp x) :real)
763 (eq (numeric-type-complexp y) :real))
764 (multiple-value-bind (low high) (funcall fun x y)
765 (make-numeric-type :class 'integer
769 (numeric-contagion x y))))
771 ;;; simple utility to flatten a list
772 (defun flatten-list (x)
773 (labels ((flatten-helper (x r);; 'r' is the stuff to the 'right'.
777 (t (flatten-helper (car x)
778 (flatten-helper (cdr x) r))))))
779 (flatten-helper x nil)))
781 ;;; Take some type of continuation and massage it so that we get a
782 ;;; list of the constituent types. If ARG is *EMPTY-TYPE*, return NIL
783 ;;; to indicate failure.
784 (defun prepare-arg-for-derive-type (arg)
785 (flet ((listify (arg)
790 (union-type-types arg))
793 (unless (eq arg *empty-type*)
794 ;; Make sure all args are some type of numeric-type. For member
795 ;; types, convert the list of members into a union of equivalent
796 ;; single-element member-type's.
797 (let ((new-args nil))
798 (dolist (arg (listify arg))
799 (if (member-type-p arg)
800 ;; Run down the list of members and convert to a list of
802 (dolist (member (member-type-members arg))
803 (push (if (numberp member)
804 (make-member-type :members (list member))
807 (push arg new-args)))
808 (unless (member *empty-type* new-args)
811 ;;; Convert from the standard type convention for which -0.0 and 0.0
812 ;;; are equal to an intermediate convention for which they are
813 ;;; considered different which is more natural for some of the
815 #!-negative-zero-is-not-zero
816 (defun convert-numeric-type (type)
817 (declare (type numeric-type type))
818 ;;; Only convert real float interval delimiters types.
819 (if (eq (numeric-type-complexp type) :real)
820 (let* ((lo (numeric-type-low type))
821 (lo-val (type-bound-number lo))
822 (lo-float-zero-p (and lo (floatp lo-val) (= lo-val 0.0)))
823 (hi (numeric-type-high type))
824 (hi-val (type-bound-number hi))
825 (hi-float-zero-p (and hi (floatp hi-val) (= hi-val 0.0))))
826 (if (or lo-float-zero-p hi-float-zero-p)
828 :class (numeric-type-class type)
829 :format (numeric-type-format type)
831 :low (if lo-float-zero-p
833 (list (float 0.0 lo-val))
836 :high (if hi-float-zero-p
838 (list (float -0.0 hi-val))
845 ;;; Convert back from the intermediate convention for which -0.0 and
846 ;;; 0.0 are considered different to the standard type convention for
848 #!-negative-zero-is-not-zero
849 (defun convert-back-numeric-type (type)
850 (declare (type numeric-type type))
851 ;;; Only convert real float interval delimiters types.
852 (if (eq (numeric-type-complexp type) :real)
853 (let* ((lo (numeric-type-low type))
854 (lo-val (type-bound-number lo))
856 (and lo (floatp lo-val) (= lo-val 0.0)
857 (float-sign lo-val)))
858 (hi (numeric-type-high type))
859 (hi-val (type-bound-number hi))
861 (and hi (floatp hi-val) (= hi-val 0.0)
862 (float-sign hi-val))))
864 ;; (float +0.0 +0.0) => (member 0.0)
865 ;; (float -0.0 -0.0) => (member -0.0)
866 ((and lo-float-zero-p hi-float-zero-p)
867 ;; shouldn't have exclusive bounds here..
868 (aver (and (not (consp lo)) (not (consp hi))))
869 (if (= lo-float-zero-p hi-float-zero-p)
870 ;; (float +0.0 +0.0) => (member 0.0)
871 ;; (float -0.0 -0.0) => (member -0.0)
872 (specifier-type `(member ,lo-val))
873 ;; (float -0.0 +0.0) => (float 0.0 0.0)
874 ;; (float +0.0 -0.0) => (float 0.0 0.0)
875 (make-numeric-type :class (numeric-type-class type)
876 :format (numeric-type-format type)
882 ;; (float -0.0 x) => (float 0.0 x)
883 ((and (not (consp lo)) (minusp lo-float-zero-p))
884 (make-numeric-type :class (numeric-type-class type)
885 :format (numeric-type-format type)
887 :low (float 0.0 lo-val)
889 ;; (float (+0.0) x) => (float (0.0) x)
890 ((and (consp lo) (plusp lo-float-zero-p))
891 (make-numeric-type :class (numeric-type-class type)
892 :format (numeric-type-format type)
894 :low (list (float 0.0 lo-val))
897 ;; (float +0.0 x) => (or (member 0.0) (float (0.0) x))
898 ;; (float (-0.0) x) => (or (member 0.0) (float (0.0) x))
899 (list (make-member-type :members (list (float 0.0 lo-val)))
900 (make-numeric-type :class (numeric-type-class type)
901 :format (numeric-type-format type)
903 :low (list (float 0.0 lo-val))
907 ;; (float x +0.0) => (float x 0.0)
908 ((and (not (consp hi)) (plusp hi-float-zero-p))
909 (make-numeric-type :class (numeric-type-class type)
910 :format (numeric-type-format type)
913 :high (float 0.0 hi-val)))
914 ;; (float x (-0.0)) => (float x (0.0))
915 ((and (consp hi) (minusp hi-float-zero-p))
916 (make-numeric-type :class (numeric-type-class type)
917 :format (numeric-type-format type)
920 :high (list (float 0.0 hi-val))))
922 ;; (float x (+0.0)) => (or (member -0.0) (float x (0.0)))
923 ;; (float x -0.0) => (or (member -0.0) (float x (0.0)))
924 (list (make-member-type :members (list (float -0.0 hi-val)))
925 (make-numeric-type :class (numeric-type-class type)
926 :format (numeric-type-format type)
929 :high (list (float 0.0 hi-val)))))))
935 ;;; Convert back a possible list of numeric types.
936 #!-negative-zero-is-not-zero
937 (defun convert-back-numeric-type-list (type-list)
941 (dolist (type type-list)
942 (if (numeric-type-p type)
943 (let ((result (convert-back-numeric-type type)))
945 (setf results (append results result))
946 (push result results)))
947 (push type results)))
950 (convert-back-numeric-type type-list))
952 (convert-back-numeric-type-list (union-type-types type-list)))
956 ;;; FIXME: MAKE-CANONICAL-UNION-TYPE and CONVERT-MEMBER-TYPE probably
957 ;;; belong in the kernel's type logic, invoked always, instead of in
958 ;;; the compiler, invoked only during some type optimizations.
960 ;;; Take a list of types and return a canonical type specifier,
961 ;;; combining any MEMBER types together. If both positive and negative
962 ;;; MEMBER types are present they are converted to a float type.
963 ;;; XXX This would be far simpler if the type-union methods could handle
964 ;;; member/number unions.
965 (defun make-canonical-union-type (type-list)
968 (dolist (type type-list)
969 (if (member-type-p type)
970 (setf members (union members (member-type-members type)))
971 (push type misc-types)))
973 (when (null (set-difference '(-0l0 0l0) members))
974 #!-negative-zero-is-not-zero
975 (push (specifier-type '(long-float 0l0 0l0)) misc-types)
976 #!+negative-zero-is-not-zero
977 (push (specifier-type '(long-float -0l0 0l0)) misc-types)
978 (setf members (set-difference members '(-0l0 0l0))))
979 (when (null (set-difference '(-0d0 0d0) members))
980 #!-negative-zero-is-not-zero
981 (push (specifier-type '(double-float 0d0 0d0)) misc-types)
982 #!+negative-zero-is-not-zero
983 (push (specifier-type '(double-float -0d0 0d0)) misc-types)
984 (setf members (set-difference members '(-0d0 0d0))))
985 (when (null (set-difference '(-0f0 0f0) members))
986 #!-negative-zero-is-not-zero
987 (push (specifier-type '(single-float 0f0 0f0)) misc-types)
988 #!+negative-zero-is-not-zero
989 (push (specifier-type '(single-float -0f0 0f0)) misc-types)
990 (setf members (set-difference members '(-0f0 0f0))))
992 (apply #'type-union (make-member-type :members members) misc-types)
993 (apply #'type-union misc-types))))
995 ;;; Convert a member type with a single member to a numeric type.
996 (defun convert-member-type (arg)
997 (let* ((members (member-type-members arg))
998 (member (first members))
999 (member-type (type-of member)))
1000 (aver (not (rest members)))
1001 (specifier-type `(,(if (subtypep member-type 'integer)
1006 ;;; This is used in defoptimizers for computing the resulting type of
1009 ;;; Given the continuation ARG, derive the resulting type using the
1010 ;;; DERIVE-FCN. DERIVE-FCN takes exactly one argument which is some
1011 ;;; "atomic" continuation type like numeric-type or member-type
1012 ;;; (containing just one element). It should return the resulting
1013 ;;; type, which can be a list of types.
1015 ;;; For the case of member types, if a member-fcn is given it is
1016 ;;; called to compute the result otherwise the member type is first
1017 ;;; converted to a numeric type and the derive-fcn is call.
1018 (defun one-arg-derive-type (arg derive-fcn member-fcn
1019 &optional (convert-type t))
1020 (declare (type function derive-fcn)
1021 (type (or null function) member-fcn)
1022 #!+negative-zero-is-not-zero (ignore convert-type))
1023 (let ((arg-list (prepare-arg-for-derive-type (continuation-type arg))))
1029 (with-float-traps-masked
1030 (:underflow :overflow :divide-by-zero)
1034 (first (member-type-members x))))))
1035 ;; Otherwise convert to a numeric type.
1036 (let ((result-type-list
1037 (funcall derive-fcn (convert-member-type x))))
1038 #!-negative-zero-is-not-zero
1040 (convert-back-numeric-type-list result-type-list)
1042 #!+negative-zero-is-not-zero
1045 #!-negative-zero-is-not-zero
1047 (convert-back-numeric-type-list
1048 (funcall derive-fcn (convert-numeric-type x)))
1049 (funcall derive-fcn x))
1050 #!+negative-zero-is-not-zero
1051 (funcall derive-fcn x))
1053 *universal-type*))))
1054 ;; Run down the list of args and derive the type of each one,
1055 ;; saving all of the results in a list.
1056 (let ((results nil))
1057 (dolist (arg arg-list)
1058 (let ((result (deriver arg)))
1060 (setf results (append results result))
1061 (push result results))))
1063 (make-canonical-union-type results)
1064 (first results)))))))
1066 ;;; Same as ONE-ARG-DERIVE-TYPE, except we assume the function takes
1067 ;;; two arguments. DERIVE-FCN takes 3 args in this case: the two
1068 ;;; original args and a third which is T to indicate if the two args
1069 ;;; really represent the same continuation. This is useful for
1070 ;;; deriving the type of things like (* x x), which should always be
1071 ;;; positive. If we didn't do this, we wouldn't be able to tell.
1072 (defun two-arg-derive-type (arg1 arg2 derive-fcn fcn
1073 &optional (convert-type t))
1074 #!+negative-zero-is-not-zero
1075 (declare (ignore convert-type))
1076 (flet (#!-negative-zero-is-not-zero
1077 (deriver (x y same-arg)
1078 (cond ((and (member-type-p x) (member-type-p y))
1079 (let* ((x (first (member-type-members x)))
1080 (y (first (member-type-members y)))
1081 (result (with-float-traps-masked
1082 (:underflow :overflow :divide-by-zero
1084 (funcall fcn x y))))
1085 (cond ((null result))
1086 ((and (floatp result) (float-nan-p result))
1087 (make-numeric-type :class 'float
1088 :format (type-of result)
1091 (make-member-type :members (list result))))))
1092 ((and (member-type-p x) (numeric-type-p y))
1093 (let* ((x (convert-member-type x))
1094 (y (if convert-type (convert-numeric-type y) y))
1095 (result (funcall derive-fcn x y same-arg)))
1097 (convert-back-numeric-type-list result)
1099 ((and (numeric-type-p x) (member-type-p y))
1100 (let* ((x (if convert-type (convert-numeric-type x) x))
1101 (y (convert-member-type y))
1102 (result (funcall derive-fcn x y same-arg)))
1104 (convert-back-numeric-type-list result)
1106 ((and (numeric-type-p x) (numeric-type-p y))
1107 (let* ((x (if convert-type (convert-numeric-type x) x))
1108 (y (if convert-type (convert-numeric-type y) y))
1109 (result (funcall derive-fcn x y same-arg)))
1111 (convert-back-numeric-type-list result)
1115 #!+negative-zero-is-not-zero
1116 (deriver (x y same-arg)
1117 (cond ((and (member-type-p x) (member-type-p y))
1118 (let* ((x (first (member-type-members x)))
1119 (y (first (member-type-members y)))
1120 (result (with-float-traps-masked
1121 (:underflow :overflow :divide-by-zero)
1122 (funcall fcn x y))))
1124 (make-member-type :members (list result)))))
1125 ((and (member-type-p x) (numeric-type-p y))
1126 (let ((x (convert-member-type x)))
1127 (funcall derive-fcn x y same-arg)))
1128 ((and (numeric-type-p x) (member-type-p y))
1129 (let ((y (convert-member-type y)))
1130 (funcall derive-fcn x y same-arg)))
1131 ((and (numeric-type-p x) (numeric-type-p y))
1132 (funcall derive-fcn x y same-arg))
1134 *universal-type*))))
1135 (let ((same-arg (same-leaf-ref-p arg1 arg2))
1136 (a1 (prepare-arg-for-derive-type (continuation-type arg1)))
1137 (a2 (prepare-arg-for-derive-type (continuation-type arg2))))
1139 (let ((results nil))
1141 ;; Since the args are the same continuation, just run
1144 (let ((result (deriver x x same-arg)))
1146 (setf results (append results result))
1147 (push result results))))
1148 ;; Try all pairwise combinations.
1151 (let ((result (or (deriver x y same-arg)
1152 (numeric-contagion x y))))
1154 (setf results (append results result))
1155 (push result results))))))
1157 (make-canonical-union-type results)
1158 (first results)))))))
1160 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1162 (defoptimizer (+ derive-type) ((x y))
1163 (derive-integer-type
1170 (values (frob (numeric-type-low x) (numeric-type-low y))
1171 (frob (numeric-type-high x) (numeric-type-high y)))))))
1173 (defoptimizer (- derive-type) ((x y))
1174 (derive-integer-type
1181 (values (frob (numeric-type-low x) (numeric-type-high y))
1182 (frob (numeric-type-high x) (numeric-type-low y)))))))
1184 (defoptimizer (* derive-type) ((x y))
1185 (derive-integer-type
1188 (let ((x-low (numeric-type-low x))
1189 (x-high (numeric-type-high x))
1190 (y-low (numeric-type-low y))
1191 (y-high (numeric-type-high y)))
1192 (cond ((not (and x-low y-low))
1194 ((or (minusp x-low) (minusp y-low))
1195 (if (and x-high y-high)
1196 (let ((max (* (max (abs x-low) (abs x-high))
1197 (max (abs y-low) (abs y-high)))))
1198 (values (- max) max))
1201 (values (* x-low y-low)
1202 (if (and x-high y-high)
1206 (defoptimizer (/ derive-type) ((x y))
1207 (numeric-contagion (continuation-type x) (continuation-type y)))
1211 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1213 (defun +-derive-type-aux (x y same-arg)
1214 (if (and (numeric-type-real-p x)
1215 (numeric-type-real-p y))
1218 (let ((x-int (numeric-type->interval x)))
1219 (interval-add x-int x-int))
1220 (interval-add (numeric-type->interval x)
1221 (numeric-type->interval y))))
1222 (result-type (numeric-contagion x y)))
1223 ;; If the result type is a float, we need to be sure to coerce
1224 ;; the bounds into the correct type.
1225 (when (eq (numeric-type-class result-type) 'float)
1226 (setf result (interval-func
1228 (coerce x (or (numeric-type-format result-type)
1232 :class (if (and (eq (numeric-type-class x) 'integer)
1233 (eq (numeric-type-class y) 'integer))
1234 ;; The sum of integers is always an integer.
1236 (numeric-type-class result-type))
1237 :format (numeric-type-format result-type)
1238 :low (interval-low result)
1239 :high (interval-high result)))
1240 ;; general contagion
1241 (numeric-contagion x y)))
1243 (defoptimizer (+ derive-type) ((x y))
1244 (two-arg-derive-type x y #'+-derive-type-aux #'+))
1246 (defun --derive-type-aux (x y same-arg)
1247 (if (and (numeric-type-real-p x)
1248 (numeric-type-real-p y))
1250 ;; (- X X) is always 0.
1252 (make-interval :low 0 :high 0)
1253 (interval-sub (numeric-type->interval x)
1254 (numeric-type->interval y))))
1255 (result-type (numeric-contagion x y)))
1256 ;; If the result type is a float, we need to be sure to coerce
1257 ;; the bounds into the correct type.
1258 (when (eq (numeric-type-class result-type) 'float)
1259 (setf result (interval-func
1261 (coerce x (or (numeric-type-format result-type)
1265 :class (if (and (eq (numeric-type-class x) 'integer)
1266 (eq (numeric-type-class y) 'integer))
1267 ;; The difference of integers is always an integer.
1269 (numeric-type-class result-type))
1270 :format (numeric-type-format result-type)
1271 :low (interval-low result)
1272 :high (interval-high result)))
1273 ;; general contagion
1274 (numeric-contagion x y)))
1276 (defoptimizer (- derive-type) ((x y))
1277 (two-arg-derive-type x y #'--derive-type-aux #'-))
1279 (defun *-derive-type-aux (x y same-arg)
1280 (if (and (numeric-type-real-p x)
1281 (numeric-type-real-p y))
1283 ;; (* X X) is always positive, so take care to do it right.
1285 (interval-sqr (numeric-type->interval x))
1286 (interval-mul (numeric-type->interval x)
1287 (numeric-type->interval y))))
1288 (result-type (numeric-contagion x y)))
1289 ;; If the result type is a float, we need to be sure to coerce
1290 ;; the bounds into the correct type.
1291 (when (eq (numeric-type-class result-type) 'float)
1292 (setf result (interval-func
1294 (coerce x (or (numeric-type-format result-type)
1298 :class (if (and (eq (numeric-type-class x) 'integer)
1299 (eq (numeric-type-class y) 'integer))
1300 ;; The product of integers is always an integer.
1302 (numeric-type-class result-type))
1303 :format (numeric-type-format result-type)
1304 :low (interval-low result)
1305 :high (interval-high result)))
1306 (numeric-contagion x y)))
1308 (defoptimizer (* derive-type) ((x y))
1309 (two-arg-derive-type x y #'*-derive-type-aux #'*))
1311 (defun /-derive-type-aux (x y same-arg)
1312 (if (and (numeric-type-real-p x)
1313 (numeric-type-real-p y))
1315 ;; (/ X X) is always 1, except if X can contain 0. In
1316 ;; that case, we shouldn't optimize the division away
1317 ;; because we want 0/0 to signal an error.
1319 (not (interval-contains-p
1320 0 (interval-closure (numeric-type->interval y)))))
1321 (make-interval :low 1 :high 1)
1322 (interval-div (numeric-type->interval x)
1323 (numeric-type->interval y))))
1324 (result-type (numeric-contagion x y)))
1325 ;; If the result type is a float, we need to be sure to coerce
1326 ;; the bounds into the correct type.
1327 (when (eq (numeric-type-class result-type) 'float)
1328 (setf result (interval-func
1330 (coerce x (or (numeric-type-format result-type)
1333 (make-numeric-type :class (numeric-type-class result-type)
1334 :format (numeric-type-format result-type)
1335 :low (interval-low result)
1336 :high (interval-high result)))
1337 (numeric-contagion x y)))
1339 (defoptimizer (/ derive-type) ((x y))
1340 (two-arg-derive-type x y #'/-derive-type-aux #'/))
1345 ;;; KLUDGE: All this ASH optimization is suppressed under CMU CL
1346 ;;; because as of version 2.4.6 for Debian, CMU CL blows up on (ASH
1347 ;;; 1000000000 -100000000000) (i.e. ASH of two bignums yielding zero)
1348 ;;; and it's hard to avoid that calculation in here.
1349 #-(and cmu sb-xc-host)
1352 (defun ash-derive-type-aux (n-type shift same-arg)
1353 (declare (ignore same-arg))
1354 (flet ((ash-outer (n s)
1355 (when (and (fixnump s)
1357 (> s sb!vm:*target-most-negative-fixnum*))
1359 ;; KLUDGE: The bare 64's here should be related to
1360 ;; symbolic machine word size values somehow.
1363 (if (and (fixnump s)
1364 (> s sb!vm:*target-most-negative-fixnum*))
1366 (if (minusp n) -1 0))))
1367 (or (and (csubtypep n-type (specifier-type 'integer))
1368 (csubtypep shift (specifier-type 'integer))
1369 (let ((n-low (numeric-type-low n-type))
1370 (n-high (numeric-type-high n-type))
1371 (s-low (numeric-type-low shift))
1372 (s-high (numeric-type-high shift)))
1373 (make-numeric-type :class 'integer :complexp :real
1376 (ash-outer n-low s-high)
1377 (ash-inner n-low s-low)))
1380 (ash-inner n-high s-low)
1381 (ash-outer n-high s-high))))))
1384 (defoptimizer (ash derive-type) ((n shift))
1385 (two-arg-derive-type n shift #'ash-derive-type-aux #'ash))
1388 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1389 (macrolet ((frob (fun)
1390 `#'(lambda (type type2)
1391 (declare (ignore type2))
1392 (let ((lo (numeric-type-low type))
1393 (hi (numeric-type-high type)))
1394 (values (if hi (,fun hi) nil) (if lo (,fun lo) nil))))))
1396 (defoptimizer (%negate derive-type) ((num))
1397 (derive-integer-type num num (frob -))))
1399 (defoptimizer (lognot derive-type) ((int))
1400 (derive-integer-type int int
1401 (lambda (type type2)
1402 (declare (ignore type2))
1403 (let ((lo (numeric-type-low type))
1404 (hi (numeric-type-high type)))
1405 (values (if hi (lognot hi) nil)
1406 (if lo (lognot lo) nil)
1407 (numeric-type-class type)
1408 (numeric-type-format type))))))
1410 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1411 (defoptimizer (%negate derive-type) ((num))
1412 (flet ((negate-bound (b)
1414 (set-bound (- (type-bound-number b))
1416 (one-arg-derive-type num
1418 (modified-numeric-type
1420 :low (negate-bound (numeric-type-high type))
1421 :high (negate-bound (numeric-type-low type))))
1424 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1425 (defoptimizer (abs derive-type) ((num))
1426 (let ((type (continuation-type num)))
1427 (if (and (numeric-type-p type)
1428 (eq (numeric-type-class type) 'integer)
1429 (eq (numeric-type-complexp type) :real))
1430 (let ((lo (numeric-type-low type))
1431 (hi (numeric-type-high type)))
1432 (make-numeric-type :class 'integer :complexp :real
1433 :low (cond ((and hi (minusp hi))
1439 :high (if (and hi lo)
1440 (max (abs hi) (abs lo))
1442 (numeric-contagion type type))))
1444 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1445 (defun abs-derive-type-aux (type)
1446 (cond ((eq (numeric-type-complexp type) :complex)
1447 ;; The absolute value of a complex number is always a
1448 ;; non-negative float.
1449 (let* ((format (case (numeric-type-class type)
1450 ((integer rational) 'single-float)
1451 (t (numeric-type-format type))))
1452 (bound-format (or format 'float)))
1453 (make-numeric-type :class 'float
1456 :low (coerce 0 bound-format)
1459 ;; The absolute value of a real number is a non-negative real
1460 ;; of the same type.
1461 (let* ((abs-bnd (interval-abs (numeric-type->interval type)))
1462 (class (numeric-type-class type))
1463 (format (numeric-type-format type))
1464 (bound-type (or format class 'real)))
1469 :low (coerce-numeric-bound (interval-low abs-bnd) bound-type)
1470 :high (coerce-numeric-bound
1471 (interval-high abs-bnd) bound-type))))))
1473 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1474 (defoptimizer (abs derive-type) ((num))
1475 (one-arg-derive-type num #'abs-derive-type-aux #'abs))
1477 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1478 (defoptimizer (truncate derive-type) ((number divisor))
1479 (let ((number-type (continuation-type number))
1480 (divisor-type (continuation-type divisor))
1481 (integer-type (specifier-type 'integer)))
1482 (if (and (numeric-type-p number-type)
1483 (csubtypep number-type integer-type)
1484 (numeric-type-p divisor-type)
1485 (csubtypep divisor-type integer-type))
1486 (let ((number-low (numeric-type-low number-type))
1487 (number-high (numeric-type-high number-type))
1488 (divisor-low (numeric-type-low divisor-type))
1489 (divisor-high (numeric-type-high divisor-type)))
1490 (values-specifier-type
1491 `(values ,(integer-truncate-derive-type number-low number-high
1492 divisor-low divisor-high)
1493 ,(integer-rem-derive-type number-low number-high
1494 divisor-low divisor-high))))
1497 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
1500 (defun rem-result-type (number-type divisor-type)
1501 ;; Figure out what the remainder type is. The remainder is an
1502 ;; integer if both args are integers; a rational if both args are
1503 ;; rational; and a float otherwise.
1504 (cond ((and (csubtypep number-type (specifier-type 'integer))
1505 (csubtypep divisor-type (specifier-type 'integer)))
1507 ((and (csubtypep number-type (specifier-type 'rational))
1508 (csubtypep divisor-type (specifier-type 'rational)))
1510 ((and (csubtypep number-type (specifier-type 'float))
1511 (csubtypep divisor-type (specifier-type 'float)))
1512 ;; Both are floats so the result is also a float, of
1513 ;; the largest type.
1514 (or (float-format-max (numeric-type-format number-type)
1515 (numeric-type-format divisor-type))
1517 ((and (csubtypep number-type (specifier-type 'float))
1518 (csubtypep divisor-type (specifier-type 'rational)))
1519 ;; One of the arguments is a float and the other is a
1520 ;; rational. The remainder is a float of the same
1522 (or (numeric-type-format number-type) 'float))
1523 ((and (csubtypep divisor-type (specifier-type 'float))
1524 (csubtypep number-type (specifier-type 'rational)))
1525 ;; One of the arguments is a float and the other is a
1526 ;; rational. The remainder is a float of the same
1528 (or (numeric-type-format divisor-type) 'float))
1530 ;; Some unhandled combination. This usually means both args
1531 ;; are REAL so the result is a REAL.
1534 (defun truncate-derive-type-quot (number-type divisor-type)
1535 (let* ((rem-type (rem-result-type number-type divisor-type))
1536 (number-interval (numeric-type->interval number-type))
1537 (divisor-interval (numeric-type->interval divisor-type)))
1538 ;;(declare (type (member '(integer rational float)) rem-type))
1539 ;; We have real numbers now.
1540 (cond ((eq rem-type 'integer)
1541 ;; Since the remainder type is INTEGER, both args are
1543 (let* ((res (integer-truncate-derive-type
1544 (interval-low number-interval)
1545 (interval-high number-interval)
1546 (interval-low divisor-interval)
1547 (interval-high divisor-interval))))
1548 (specifier-type (if (listp res) res 'integer))))
1550 (let ((quot (truncate-quotient-bound
1551 (interval-div number-interval
1552 divisor-interval))))
1553 (specifier-type `(integer ,(or (interval-low quot) '*)
1554 ,(or (interval-high quot) '*))))))))
1556 (defun truncate-derive-type-rem (number-type divisor-type)
1557 (let* ((rem-type (rem-result-type number-type divisor-type))
1558 (number-interval (numeric-type->interval number-type))
1559 (divisor-interval (numeric-type->interval divisor-type))
1560 (rem (truncate-rem-bound number-interval divisor-interval)))
1561 ;;(declare (type (member '(integer rational float)) rem-type))
1562 ;; We have real numbers now.
1563 (cond ((eq rem-type 'integer)
1564 ;; Since the remainder type is INTEGER, both args are
1566 (specifier-type `(,rem-type ,(or (interval-low rem) '*)
1567 ,(or (interval-high rem) '*))))
1569 (multiple-value-bind (class format)
1572 (values 'integer nil))
1574 (values 'rational nil))
1575 ((or single-float double-float #!+long-float long-float)
1576 (values 'float rem-type))
1578 (values 'float nil))
1581 (when (member rem-type '(float single-float double-float
1582 #!+long-float long-float))
1583 (setf rem (interval-func #'(lambda (x)
1584 (coerce x rem-type))
1586 (make-numeric-type :class class
1588 :low (interval-low rem)
1589 :high (interval-high rem)))))))
1591 (defun truncate-derive-type-quot-aux (num div same-arg)
1592 (declare (ignore same-arg))
1593 (if (and (numeric-type-real-p num)
1594 (numeric-type-real-p div))
1595 (truncate-derive-type-quot num div)
1598 (defun truncate-derive-type-rem-aux (num div same-arg)
1599 (declare (ignore same-arg))
1600 (if (and (numeric-type-real-p num)
1601 (numeric-type-real-p div))
1602 (truncate-derive-type-rem num div)
1605 (defoptimizer (truncate derive-type) ((number divisor))
1606 (let ((quot (two-arg-derive-type number divisor
1607 #'truncate-derive-type-quot-aux #'truncate))
1608 (rem (two-arg-derive-type number divisor
1609 #'truncate-derive-type-rem-aux #'rem)))
1610 (when (and quot rem)
1611 (make-values-type :required (list quot rem)))))
1613 (defun ftruncate-derive-type-quot (number-type divisor-type)
1614 ;; The bounds are the same as for truncate. However, the first
1615 ;; result is a float of some type. We need to determine what that
1616 ;; type is. Basically it's the more contagious of the two types.
1617 (let ((q-type (truncate-derive-type-quot number-type divisor-type))
1618 (res-type (numeric-contagion number-type divisor-type)))
1619 (make-numeric-type :class 'float
1620 :format (numeric-type-format res-type)
1621 :low (numeric-type-low q-type)
1622 :high (numeric-type-high q-type))))
1624 (defun ftruncate-derive-type-quot-aux (n d same-arg)
1625 (declare (ignore same-arg))
1626 (if (and (numeric-type-real-p n)
1627 (numeric-type-real-p d))
1628 (ftruncate-derive-type-quot n d)
1631 (defoptimizer (ftruncate derive-type) ((number divisor))
1633 (two-arg-derive-type number divisor
1634 #'ftruncate-derive-type-quot-aux #'ftruncate))
1635 (rem (two-arg-derive-type number divisor
1636 #'truncate-derive-type-rem-aux #'rem)))
1637 (when (and quot rem)
1638 (make-values-type :required (list quot rem)))))
1640 (defun %unary-truncate-derive-type-aux (number)
1641 (truncate-derive-type-quot number (specifier-type '(integer 1 1))))
1643 (defoptimizer (%unary-truncate derive-type) ((number))
1644 (one-arg-derive-type number
1645 #'%unary-truncate-derive-type-aux
1648 ;;; Define optimizers for FLOOR and CEILING.
1650 ((frob-opt (name q-name r-name)
1651 (let ((q-aux (symbolicate q-name "-AUX"))
1652 (r-aux (symbolicate r-name "-AUX")))
1654 ;; Compute type of quotient (first) result.
1655 (defun ,q-aux (number-type divisor-type)
1656 (let* ((number-interval
1657 (numeric-type->interval number-type))
1659 (numeric-type->interval divisor-type))
1660 (quot (,q-name (interval-div number-interval
1661 divisor-interval))))
1662 (specifier-type `(integer ,(or (interval-low quot) '*)
1663 ,(or (interval-high quot) '*)))))
1664 ;; Compute type of remainder.
1665 (defun ,r-aux (number-type divisor-type)
1666 (let* ((divisor-interval
1667 (numeric-type->interval divisor-type))
1668 (rem (,r-name divisor-interval))
1669 (result-type (rem-result-type number-type divisor-type)))
1670 (multiple-value-bind (class format)
1673 (values 'integer nil))
1675 (values 'rational nil))
1676 ((or single-float double-float #!+long-float long-float)
1677 (values 'float result-type))
1679 (values 'float nil))
1682 (when (member result-type '(float single-float double-float
1683 #!+long-float long-float))
1684 ;; Make sure that the limits on the interval have
1686 (setf rem (interval-func (lambda (x)
1687 (coerce x result-type))
1689 (make-numeric-type :class class
1691 :low (interval-low rem)
1692 :high (interval-high rem)))))
1693 ;; the optimizer itself
1694 (defoptimizer (,name derive-type) ((number divisor))
1695 (flet ((derive-q (n d same-arg)
1696 (declare (ignore same-arg))
1697 (if (and (numeric-type-real-p n)
1698 (numeric-type-real-p d))
1701 (derive-r (n d same-arg)
1702 (declare (ignore same-arg))
1703 (if (and (numeric-type-real-p n)
1704 (numeric-type-real-p d))
1707 (let ((quot (two-arg-derive-type
1708 number divisor #'derive-q #',name))
1709 (rem (two-arg-derive-type
1710 number divisor #'derive-r #'mod)))
1711 (when (and quot rem)
1712 (make-values-type :required (list quot rem))))))))))
1714 ;; FIXME: DEF-FROB-OPT, not just FROB-OPT
1715 (frob-opt floor floor-quotient-bound floor-rem-bound)
1716 (frob-opt ceiling ceiling-quotient-bound ceiling-rem-bound))
1718 ;;; Define optimizers for FFLOOR and FCEILING
1720 ((frob-opt (name q-name r-name)
1721 (let ((q-aux (symbolicate "F" q-name "-AUX"))
1722 (r-aux (symbolicate r-name "-AUX")))
1724 ;; Compute type of quotient (first) result.
1725 (defun ,q-aux (number-type divisor-type)
1726 (let* ((number-interval
1727 (numeric-type->interval number-type))
1729 (numeric-type->interval divisor-type))
1730 (quot (,q-name (interval-div number-interval
1732 (res-type (numeric-contagion number-type divisor-type)))
1734 :class (numeric-type-class res-type)
1735 :format (numeric-type-format res-type)
1736 :low (interval-low quot)
1737 :high (interval-high quot))))
1739 (defoptimizer (,name derive-type) ((number divisor))
1740 (flet ((derive-q (n d same-arg)
1741 (declare (ignore same-arg))
1742 (if (and (numeric-type-real-p n)
1743 (numeric-type-real-p d))
1746 (derive-r (n d same-arg)
1747 (declare (ignore same-arg))
1748 (if (and (numeric-type-real-p n)
1749 (numeric-type-real-p d))
1752 (let ((quot (two-arg-derive-type
1753 number divisor #'derive-q #',name))
1754 (rem (two-arg-derive-type
1755 number divisor #'derive-r #'mod)))
1756 (when (and quot rem)
1757 (make-values-type :required (list quot rem))))))))))
1759 ;; FIXME: DEF-FROB-OPT, not just FROB-OPT
1760 (frob-opt ffloor floor-quotient-bound floor-rem-bound)
1761 (frob-opt fceiling ceiling-quotient-bound ceiling-rem-bound))
1763 ;;; functions to compute the bounds on the quotient and remainder for
1764 ;;; the FLOOR function
1765 (defun floor-quotient-bound (quot)
1766 ;; Take the floor of the quotient and then massage it into what we
1768 (let ((lo (interval-low quot))
1769 (hi (interval-high quot)))
1770 ;; Take the floor of the lower bound. The result is always a
1771 ;; closed lower bound.
1773 (floor (type-bound-number lo))
1775 ;; For the upper bound, we need to be careful.
1778 ;; An open bound. We need to be careful here because
1779 ;; the floor of '(10.0) is 9, but the floor of
1781 (multiple-value-bind (q r) (floor (first hi))
1786 ;; A closed bound, so the answer is obvious.
1790 (make-interval :low lo :high hi)))
1791 (defun floor-rem-bound (div)
1792 ;; The remainder depends only on the divisor. Try to get the
1793 ;; correct sign for the remainder if we can.
1794 (case (interval-range-info div)
1796 ;; The divisor is always positive.
1797 (let ((rem (interval-abs div)))
1798 (setf (interval-low rem) 0)
1799 (when (and (numberp (interval-high rem))
1800 (not (zerop (interval-high rem))))
1801 ;; The remainder never contains the upper bound. However,
1802 ;; watch out for the case where the high limit is zero!
1803 (setf (interval-high rem) (list (interval-high rem))))
1806 ;; The divisor is always negative.
1807 (let ((rem (interval-neg (interval-abs div))))
1808 (setf (interval-high rem) 0)
1809 (when (numberp (interval-low rem))
1810 ;; The remainder never contains the lower bound.
1811 (setf (interval-low rem) (list (interval-low rem))))
1814 ;; The divisor can be positive or negative. All bets off. The
1815 ;; magnitude of remainder is the maximum value of the divisor.
1816 (let ((limit (type-bound-number (interval-high (interval-abs div)))))
1817 ;; The bound never reaches the limit, so make the interval open.
1818 (make-interval :low (if limit
1821 :high (list limit))))))
1823 (floor-quotient-bound (make-interval :low 0.3 :high 10.3))
1824 => #S(INTERVAL :LOW 0 :HIGH 10)
1825 (floor-quotient-bound (make-interval :low 0.3 :high '(10.3)))
1826 => #S(INTERVAL :LOW 0 :HIGH 10)
1827 (floor-quotient-bound (make-interval :low 0.3 :high 10))
1828 => #S(INTERVAL :LOW 0 :HIGH 10)
1829 (floor-quotient-bound (make-interval :low 0.3 :high '(10)))
1830 => #S(INTERVAL :LOW 0 :HIGH 9)
1831 (floor-quotient-bound (make-interval :low '(0.3) :high 10.3))
1832 => #S(INTERVAL :LOW 0 :HIGH 10)
1833 (floor-quotient-bound (make-interval :low '(0.0) :high 10.3))
1834 => #S(INTERVAL :LOW 0 :HIGH 10)
1835 (floor-quotient-bound (make-interval :low '(-1.3) :high 10.3))
1836 => #S(INTERVAL :LOW -2 :HIGH 10)
1837 (floor-quotient-bound (make-interval :low '(-1.0) :high 10.3))
1838 => #S(INTERVAL :LOW -1 :HIGH 10)
1839 (floor-quotient-bound (make-interval :low -1.0 :high 10.3))
1840 => #S(INTERVAL :LOW -1 :HIGH 10)
1842 (floor-rem-bound (make-interval :low 0.3 :high 10.3))
1843 => #S(INTERVAL :LOW 0 :HIGH '(10.3))
1844 (floor-rem-bound (make-interval :low 0.3 :high '(10.3)))
1845 => #S(INTERVAL :LOW 0 :HIGH '(10.3))
1846 (floor-rem-bound (make-interval :low -10 :high -2.3))
1847 #S(INTERVAL :LOW (-10) :HIGH 0)
1848 (floor-rem-bound (make-interval :low 0.3 :high 10))
1849 => #S(INTERVAL :LOW 0 :HIGH '(10))
1850 (floor-rem-bound (make-interval :low '(-1.3) :high 10.3))
1851 => #S(INTERVAL :LOW '(-10.3) :HIGH '(10.3))
1852 (floor-rem-bound (make-interval :low '(-20.3) :high 10.3))
1853 => #S(INTERVAL :LOW (-20.3) :HIGH (20.3))
1856 ;;; same functions for CEILING
1857 (defun ceiling-quotient-bound (quot)
1858 ;; Take the ceiling of the quotient and then massage it into what we
1860 (let ((lo (interval-low quot))
1861 (hi (interval-high quot)))
1862 ;; Take the ceiling of the upper bound. The result is always a
1863 ;; closed upper bound.
1865 (ceiling (type-bound-number hi))
1867 ;; For the lower bound, we need to be careful.
1870 ;; An open bound. We need to be careful here because
1871 ;; the ceiling of '(10.0) is 11, but the ceiling of
1873 (multiple-value-bind (q r) (ceiling (first lo))
1878 ;; A closed bound, so the answer is obvious.
1882 (make-interval :low lo :high hi)))
1883 (defun ceiling-rem-bound (div)
1884 ;; The remainder depends only on the divisor. Try to get the
1885 ;; correct sign for the remainder if we can.
1886 (case (interval-range-info div)
1888 ;; Divisor is always positive. The remainder is negative.
1889 (let ((rem (interval-neg (interval-abs div))))
1890 (setf (interval-high rem) 0)
1891 (when (and (numberp (interval-low rem))
1892 (not (zerop (interval-low rem))))
1893 ;; The remainder never contains the upper bound. However,
1894 ;; watch out for the case when the upper bound is zero!
1895 (setf (interval-low rem) (list (interval-low rem))))
1898 ;; Divisor is always negative. The remainder is positive
1899 (let ((rem (interval-abs div)))
1900 (setf (interval-low rem) 0)
1901 (when (numberp (interval-high rem))
1902 ;; The remainder never contains the lower bound.
1903 (setf (interval-high rem) (list (interval-high rem))))
1906 ;; The divisor can be positive or negative. All bets off. The
1907 ;; magnitude of remainder is the maximum value of the divisor.
1908 (let ((limit (type-bound-number (interval-high (interval-abs div)))))
1909 ;; The bound never reaches the limit, so make the interval open.
1910 (make-interval :low (if limit
1913 :high (list limit))))))
1916 (ceiling-quotient-bound (make-interval :low 0.3 :high 10.3))
1917 => #S(INTERVAL :LOW 1 :HIGH 11)
1918 (ceiling-quotient-bound (make-interval :low 0.3 :high '(10.3)))
1919 => #S(INTERVAL :LOW 1 :HIGH 11)
1920 (ceiling-quotient-bound (make-interval :low 0.3 :high 10))
1921 => #S(INTERVAL :LOW 1 :HIGH 10)
1922 (ceiling-quotient-bound (make-interval :low 0.3 :high '(10)))
1923 => #S(INTERVAL :LOW 1 :HIGH 10)
1924 (ceiling-quotient-bound (make-interval :low '(0.3) :high 10.3))
1925 => #S(INTERVAL :LOW 1 :HIGH 11)
1926 (ceiling-quotient-bound (make-interval :low '(0.0) :high 10.3))
1927 => #S(INTERVAL :LOW 1 :HIGH 11)
1928 (ceiling-quotient-bound (make-interval :low '(-1.3) :high 10.3))
1929 => #S(INTERVAL :LOW -1 :HIGH 11)
1930 (ceiling-quotient-bound (make-interval :low '(-1.0) :high 10.3))
1931 => #S(INTERVAL :LOW 0 :HIGH 11)
1932 (ceiling-quotient-bound (make-interval :low -1.0 :high 10.3))
1933 => #S(INTERVAL :LOW -1 :HIGH 11)
1935 (ceiling-rem-bound (make-interval :low 0.3 :high 10.3))
1936 => #S(INTERVAL :LOW (-10.3) :HIGH 0)
1937 (ceiling-rem-bound (make-interval :low 0.3 :high '(10.3)))
1938 => #S(INTERVAL :LOW 0 :HIGH '(10.3))
1939 (ceiling-rem-bound (make-interval :low -10 :high -2.3))
1940 => #S(INTERVAL :LOW 0 :HIGH (10))
1941 (ceiling-rem-bound (make-interval :low 0.3 :high 10))
1942 => #S(INTERVAL :LOW (-10) :HIGH 0)
1943 (ceiling-rem-bound (make-interval :low '(-1.3) :high 10.3))
1944 => #S(INTERVAL :LOW (-10.3) :HIGH (10.3))
1945 (ceiling-rem-bound (make-interval :low '(-20.3) :high 10.3))
1946 => #S(INTERVAL :LOW (-20.3) :HIGH (20.3))
1949 (defun truncate-quotient-bound (quot)
1950 ;; For positive quotients, truncate is exactly like floor. For
1951 ;; negative quotients, truncate is exactly like ceiling. Otherwise,
1952 ;; it's the union of the two pieces.
1953 (case (interval-range-info quot)
1956 (floor-quotient-bound quot))
1958 ;; just like CEILING
1959 (ceiling-quotient-bound quot))
1961 ;; Split the interval into positive and negative pieces, compute
1962 ;; the result for each piece and put them back together.
1963 (destructuring-bind (neg pos) (interval-split 0 quot t t)
1964 (interval-merge-pair (ceiling-quotient-bound neg)
1965 (floor-quotient-bound pos))))))
1967 (defun truncate-rem-bound (num div)
1968 ;; This is significantly more complicated than FLOOR or CEILING. We
1969 ;; need both the number and the divisor to determine the range. The
1970 ;; basic idea is to split the ranges of NUM and DEN into positive
1971 ;; and negative pieces and deal with each of the four possibilities
1973 (case (interval-range-info num)
1975 (case (interval-range-info div)
1977 (floor-rem-bound div))
1979 (ceiling-rem-bound div))
1981 (destructuring-bind (neg pos) (interval-split 0 div t t)
1982 (interval-merge-pair (truncate-rem-bound num neg)
1983 (truncate-rem-bound num pos))))))
1985 (case (interval-range-info div)
1987 (ceiling-rem-bound div))
1989 (floor-rem-bound div))
1991 (destructuring-bind (neg pos) (interval-split 0 div t t)
1992 (interval-merge-pair (truncate-rem-bound num neg)
1993 (truncate-rem-bound num pos))))))
1995 (destructuring-bind (neg pos) (interval-split 0 num t t)
1996 (interval-merge-pair (truncate-rem-bound neg div)
1997 (truncate-rem-bound pos div))))))
2000 ;;; Derive useful information about the range. Returns three values:
2001 ;;; - '+ if its positive, '- negative, or nil if it overlaps 0.
2002 ;;; - The abs of the minimal value (i.e. closest to 0) in the range.
2003 ;;; - The abs of the maximal value if there is one, or nil if it is
2005 (defun numeric-range-info (low high)
2006 (cond ((and low (not (minusp low)))
2007 (values '+ low high))
2008 ((and high (not (plusp high)))
2009 (values '- (- high) (if low (- low) nil)))
2011 (values nil 0 (and low high (max (- low) high))))))
2013 (defun integer-truncate-derive-type
2014 (number-low number-high divisor-low divisor-high)
2015 ;; The result cannot be larger in magnitude than the number, but the
2016 ;; sign might change. If we can determine the sign of either the
2017 ;; number or the divisor, we can eliminate some of the cases.
2018 (multiple-value-bind (number-sign number-min number-max)
2019 (numeric-range-info number-low number-high)
2020 (multiple-value-bind (divisor-sign divisor-min divisor-max)
2021 (numeric-range-info divisor-low divisor-high)
2022 (when (and divisor-max (zerop divisor-max))
2023 ;; We've got a problem: guaranteed division by zero.
2024 (return-from integer-truncate-derive-type t))
2025 (when (zerop divisor-min)
2026 ;; We'll assume that they aren't going to divide by zero.
2028 (cond ((and number-sign divisor-sign)
2029 ;; We know the sign of both.
2030 (if (eq number-sign divisor-sign)
2031 ;; Same sign, so the result will be positive.
2032 `(integer ,(if divisor-max
2033 (truncate number-min divisor-max)
2036 (truncate number-max divisor-min)
2038 ;; Different signs, the result will be negative.
2039 `(integer ,(if number-max
2040 (- (truncate number-max divisor-min))
2043 (- (truncate number-min divisor-max))
2045 ((eq divisor-sign '+)
2046 ;; The divisor is positive. Therefore, the number will just
2047 ;; become closer to zero.
2048 `(integer ,(if number-low
2049 (truncate number-low divisor-min)
2052 (truncate number-high divisor-min)
2054 ((eq divisor-sign '-)
2055 ;; The divisor is negative. Therefore, the absolute value of
2056 ;; the number will become closer to zero, but the sign will also
2058 `(integer ,(if number-high
2059 (- (truncate number-high divisor-min))
2062 (- (truncate number-low divisor-min))
2064 ;; The divisor could be either positive or negative.
2066 ;; The number we are dividing has a bound. Divide that by the
2067 ;; smallest posible divisor.
2068 (let ((bound (truncate number-max divisor-min)))
2069 `(integer ,(- bound) ,bound)))
2071 ;; The number we are dividing is unbounded, so we can't tell
2072 ;; anything about the result.
2075 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2076 (defun integer-rem-derive-type
2077 (number-low number-high divisor-low divisor-high)
2078 (if (and divisor-low divisor-high)
2079 ;; We know the range of the divisor, and the remainder must be
2080 ;; smaller than the divisor. We can tell the sign of the
2081 ;; remainer if we know the sign of the number.
2082 (let ((divisor-max (1- (max (abs divisor-low) (abs divisor-high)))))
2083 `(integer ,(if (or (null number-low)
2084 (minusp number-low))
2087 ,(if (or (null number-high)
2088 (plusp number-high))
2091 ;; The divisor is potentially either very positive or very
2092 ;; negative. Therefore, the remainer is unbounded, but we might
2093 ;; be able to tell something about the sign from the number.
2094 `(integer ,(if (and number-low (not (minusp number-low)))
2095 ;; The number we are dividing is positive.
2096 ;; Therefore, the remainder must be positive.
2099 ,(if (and number-high (not (plusp number-high)))
2100 ;; The number we are dividing is negative.
2101 ;; Therefore, the remainder must be negative.
2105 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2106 (defoptimizer (random derive-type) ((bound &optional state))
2107 (let ((type (continuation-type bound)))
2108 (when (numeric-type-p type)
2109 (let ((class (numeric-type-class type))
2110 (high (numeric-type-high type))
2111 (format (numeric-type-format type)))
2115 :low (coerce 0 (or format class 'real))
2116 :high (cond ((not high) nil)
2117 ((eq class 'integer) (max (1- high) 0))
2118 ((or (consp high) (zerop high)) high)
2121 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2122 (defun random-derive-type-aux (type)
2123 (let ((class (numeric-type-class type))
2124 (high (numeric-type-high type))
2125 (format (numeric-type-format type)))
2129 :low (coerce 0 (or format class 'real))
2130 :high (cond ((not high) nil)
2131 ((eq class 'integer) (max (1- high) 0))
2132 ((or (consp high) (zerop high)) high)
2135 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2136 (defoptimizer (random derive-type) ((bound &optional state))
2137 (one-arg-derive-type bound #'random-derive-type-aux nil))
2139 ;;;; DERIVE-TYPE methods for LOGAND, LOGIOR, and friends
2141 ;;; Return the maximum number of bits an integer of the supplied type
2142 ;;; can take up, or NIL if it is unbounded. The second (third) value
2143 ;;; is T if the integer can be positive (negative) and NIL if not.
2144 ;;; Zero counts as positive.
2145 (defun integer-type-length (type)
2146 (if (numeric-type-p type)
2147 (let ((min (numeric-type-low type))
2148 (max (numeric-type-high type)))
2149 (values (and min max (max (integer-length min) (integer-length max)))
2150 (or (null max) (not (minusp max)))
2151 (or (null min) (minusp min))))
2154 (defun logand-derive-type-aux (x y &optional same-leaf)
2155 (declare (ignore same-leaf))
2156 (multiple-value-bind (x-len x-pos x-neg) (integer-type-length x)
2157 (declare (ignore x-pos))
2158 (multiple-value-bind (y-len y-pos y-neg) (integer-type-length y)
2159 (declare (ignore y-pos))
2161 ;; X must be positive.
2163 ;; They must both be positive.
2164 (cond ((or (null x-len) (null y-len))
2165 (specifier-type 'unsigned-byte))
2166 ((or (zerop x-len) (zerop y-len))
2167 (specifier-type '(integer 0 0)))
2169 (specifier-type `(unsigned-byte ,(min x-len y-len)))))
2170 ;; X is positive, but Y might be negative.
2172 (specifier-type 'unsigned-byte))
2174 (specifier-type '(integer 0 0)))
2176 (specifier-type `(unsigned-byte ,x-len)))))
2177 ;; X might be negative.
2179 ;; Y must be positive.
2181 (specifier-type 'unsigned-byte))
2183 (specifier-type '(integer 0 0)))
2186 `(unsigned-byte ,y-len))))
2187 ;; Either might be negative.
2188 (if (and x-len y-len)
2189 ;; The result is bounded.
2190 (specifier-type `(signed-byte ,(1+ (max x-len y-len))))
2191 ;; We can't tell squat about the result.
2192 (specifier-type 'integer)))))))
2194 (defun logior-derive-type-aux (x y &optional same-leaf)
2195 (declare (ignore same-leaf))
2196 (multiple-value-bind (x-len x-pos x-neg) (integer-type-length x)
2197 (multiple-value-bind (y-len y-pos y-neg) (integer-type-length y)
2199 ((and (not x-neg) (not y-neg))
2200 ;; Both are positive.
2201 (if (and x-len y-len (zerop x-len) (zerop y-len))
2202 (specifier-type '(integer 0 0))
2203 (specifier-type `(unsigned-byte ,(if (and x-len y-len)
2207 ;; X must be negative.
2209 ;; Both are negative. The result is going to be negative
2210 ;; and be the same length or shorter than the smaller.
2211 (if (and x-len y-len)
2213 (specifier-type `(integer ,(ash -1 (min x-len y-len)) -1))
2215 (specifier-type '(integer * -1)))
2216 ;; X is negative, but we don't know about Y. The result
2217 ;; will be negative, but no more negative than X.
2219 `(integer ,(or (numeric-type-low x) '*)
2222 ;; X might be either positive or negative.
2224 ;; But Y is negative. The result will be negative.
2226 `(integer ,(or (numeric-type-low y) '*)
2228 ;; We don't know squat about either. It won't get any bigger.
2229 (if (and x-len y-len)
2231 (specifier-type `(signed-byte ,(1+ (max x-len y-len))))
2233 (specifier-type 'integer))))))))
2235 (defun logxor-derive-type-aux (x y &optional same-leaf)
2236 (declare (ignore same-leaf))
2237 (multiple-value-bind (x-len x-pos x-neg) (integer-type-length x)
2238 (multiple-value-bind (y-len y-pos y-neg) (integer-type-length y)
2240 ((or (and (not x-neg) (not y-neg))
2241 (and (not x-pos) (not y-pos)))
2242 ;; Either both are negative or both are positive. The result
2243 ;; will be positive, and as long as the longer.
2244 (if (and x-len y-len (zerop x-len) (zerop y-len))
2245 (specifier-type '(integer 0 0))
2246 (specifier-type `(unsigned-byte ,(if (and x-len y-len)
2249 ((or (and (not x-pos) (not y-neg))
2250 (and (not y-neg) (not y-pos)))
2251 ;; Either X is negative and Y is positive of vice-versa. The
2252 ;; result will be negative.
2253 (specifier-type `(integer ,(if (and x-len y-len)
2254 (ash -1 (max x-len y-len))
2257 ;; We can't tell what the sign of the result is going to be.
2258 ;; All we know is that we don't create new bits.
2260 (specifier-type `(signed-byte ,(1+ (max x-len y-len)))))
2262 (specifier-type 'integer))))))
2264 (macrolet ((deffrob (logfcn)
2265 (let ((fcn-aux (symbolicate logfcn "-DERIVE-TYPE-AUX")))
2266 `(defoptimizer (,logfcn derive-type) ((x y))
2267 (two-arg-derive-type x y #',fcn-aux #',logfcn)))))
2272 ;;;; miscellaneous derive-type methods
2274 (defoptimizer (integer-length derive-type) ((x))
2275 (let ((x-type (continuation-type x)))
2276 (when (and (numeric-type-p x-type)
2277 (csubtypep x-type (specifier-type 'integer)))
2278 ;; If the X is of type (INTEGER LO HI), then the INTEGER-LENGTH
2279 ;; of X is (INTEGER (MIN lo hi) (MAX lo hi), basically. Be
2280 ;; careful about LO or HI being NIL, though. Also, if 0 is
2281 ;; contained in X, the lower bound is obviously 0.
2282 (flet ((null-or-min (a b)
2283 (and a b (min (integer-length a)
2284 (integer-length b))))
2286 (and a b (max (integer-length a)
2287 (integer-length b)))))
2288 (let* ((min (numeric-type-low x-type))
2289 (max (numeric-type-high x-type))
2290 (min-len (null-or-min min max))
2291 (max-len (null-or-max min max)))
2292 (when (ctypep 0 x-type)
2294 (specifier-type `(integer ,(or min-len '*) ,(or max-len '*))))))))
2296 (defoptimizer (code-char derive-type) ((code))
2297 (specifier-type 'base-char))
2299 (defoptimizer (values derive-type) ((&rest values))
2300 (values-specifier-type
2301 `(values ,@(mapcar (lambda (x)
2302 (type-specifier (continuation-type x)))
2305 ;;;; byte operations
2307 ;;;; We try to turn byte operations into simple logical operations.
2308 ;;;; First, we convert byte specifiers into separate size and position
2309 ;;;; arguments passed to internal %FOO functions. We then attempt to
2310 ;;;; transform the %FOO functions into boolean operations when the
2311 ;;;; size and position are constant and the operands are fixnums.
2313 (macrolet (;; Evaluate body with SIZE-VAR and POS-VAR bound to
2314 ;; expressions that evaluate to the SIZE and POSITION of
2315 ;; the byte-specifier form SPEC. We may wrap a let around
2316 ;; the result of the body to bind some variables.
2318 ;; If the spec is a BYTE form, then bind the vars to the
2319 ;; subforms. otherwise, evaluate SPEC and use the BYTE-SIZE
2320 ;; and BYTE-POSITION. The goal of this transformation is to
2321 ;; avoid consing up byte specifiers and then immediately
2322 ;; throwing them away.
2323 (with-byte-specifier ((size-var pos-var spec) &body body)
2324 (once-only ((spec `(macroexpand ,spec))
2326 `(if (and (consp ,spec)
2327 (eq (car ,spec) 'byte)
2328 (= (length ,spec) 3))
2329 (let ((,size-var (second ,spec))
2330 (,pos-var (third ,spec)))
2332 (let ((,size-var `(byte-size ,,temp))
2333 (,pos-var `(byte-position ,,temp)))
2334 `(let ((,,temp ,,spec))
2337 (define-source-transform ldb (spec int)
2338 (with-byte-specifier (size pos spec)
2339 `(%ldb ,size ,pos ,int)))
2341 (define-source-transform dpb (newbyte spec int)
2342 (with-byte-specifier (size pos spec)
2343 `(%dpb ,newbyte ,size ,pos ,int)))
2345 (define-source-transform mask-field (spec int)
2346 (with-byte-specifier (size pos spec)
2347 `(%mask-field ,size ,pos ,int)))
2349 (define-source-transform deposit-field (newbyte spec int)
2350 (with-byte-specifier (size pos spec)
2351 `(%deposit-field ,newbyte ,size ,pos ,int))))
2353 (defoptimizer (%ldb derive-type) ((size posn num))
2354 (let ((size (continuation-type size)))
2355 (if (and (numeric-type-p size)
2356 (csubtypep size (specifier-type 'integer)))
2357 (let ((size-high (numeric-type-high size)))
2358 (if (and size-high (<= size-high sb!vm:n-word-bits))
2359 (specifier-type `(unsigned-byte ,size-high))
2360 (specifier-type 'unsigned-byte)))
2363 (defoptimizer (%mask-field derive-type) ((size posn num))
2364 (let ((size (continuation-type size))
2365 (posn (continuation-type posn)))
2366 (if (and (numeric-type-p size)
2367 (csubtypep size (specifier-type 'integer))
2368 (numeric-type-p posn)
2369 (csubtypep posn (specifier-type 'integer)))
2370 (let ((size-high (numeric-type-high size))
2371 (posn-high (numeric-type-high posn)))
2372 (if (and size-high posn-high
2373 (<= (+ size-high posn-high) sb!vm:n-word-bits))
2374 (specifier-type `(unsigned-byte ,(+ size-high posn-high)))
2375 (specifier-type 'unsigned-byte)))
2378 (defoptimizer (%dpb derive-type) ((newbyte size posn int))
2379 (let ((size (continuation-type size))
2380 (posn (continuation-type posn))
2381 (int (continuation-type int)))
2382 (if (and (numeric-type-p size)
2383 (csubtypep size (specifier-type 'integer))
2384 (numeric-type-p posn)
2385 (csubtypep posn (specifier-type 'integer))
2386 (numeric-type-p int)
2387 (csubtypep int (specifier-type 'integer)))
2388 (let ((size-high (numeric-type-high size))
2389 (posn-high (numeric-type-high posn))
2390 (high (numeric-type-high int))
2391 (low (numeric-type-low int)))
2392 (if (and size-high posn-high high low
2393 (<= (+ size-high posn-high) sb!vm:n-word-bits))
2395 (list (if (minusp low) 'signed-byte 'unsigned-byte)
2396 (max (integer-length high)
2397 (integer-length low)
2398 (+ size-high posn-high))))
2402 (defoptimizer (%deposit-field derive-type) ((newbyte size posn int))
2403 (let ((size (continuation-type size))
2404 (posn (continuation-type posn))
2405 (int (continuation-type int)))
2406 (if (and (numeric-type-p size)
2407 (csubtypep size (specifier-type 'integer))
2408 (numeric-type-p posn)
2409 (csubtypep posn (specifier-type 'integer))
2410 (numeric-type-p int)
2411 (csubtypep int (specifier-type 'integer)))
2412 (let ((size-high (numeric-type-high size))
2413 (posn-high (numeric-type-high posn))
2414 (high (numeric-type-high int))
2415 (low (numeric-type-low int)))
2416 (if (and size-high posn-high high low
2417 (<= (+ size-high posn-high) sb!vm:n-word-bits))
2419 (list (if (minusp low) 'signed-byte 'unsigned-byte)
2420 (max (integer-length high)
2421 (integer-length low)
2422 (+ size-high posn-high))))
2426 (deftransform %ldb ((size posn int)
2427 (fixnum fixnum integer)
2428 (unsigned-byte #.sb!vm:n-word-bits))
2429 "convert to inline logical operations"
2430 `(logand (ash int (- posn))
2431 (ash ,(1- (ash 1 sb!vm:n-word-bits))
2432 (- size ,sb!vm:n-word-bits))))
2434 (deftransform %mask-field ((size posn int)
2435 (fixnum fixnum integer)
2436 (unsigned-byte #.sb!vm:n-word-bits))
2437 "convert to inline logical operations"
2439 (ash (ash ,(1- (ash 1 sb!vm:n-word-bits))
2440 (- size ,sb!vm:n-word-bits))
2443 ;;; Note: for %DPB and %DEPOSIT-FIELD, we can't use
2444 ;;; (OR (SIGNED-BYTE N) (UNSIGNED-BYTE N))
2445 ;;; as the result type, as that would allow result types that cover
2446 ;;; the range -2^(n-1) .. 1-2^n, instead of allowing result types of
2447 ;;; (UNSIGNED-BYTE N) and result types of (SIGNED-BYTE N).
2449 (deftransform %dpb ((new size posn int)
2451 (unsigned-byte #.sb!vm:n-word-bits))
2452 "convert to inline logical operations"
2453 `(let ((mask (ldb (byte size 0) -1)))
2454 (logior (ash (logand new mask) posn)
2455 (logand int (lognot (ash mask posn))))))
2457 (deftransform %dpb ((new size posn int)
2459 (signed-byte #.sb!vm:n-word-bits))
2460 "convert to inline logical operations"
2461 `(let ((mask (ldb (byte size 0) -1)))
2462 (logior (ash (logand new mask) posn)
2463 (logand int (lognot (ash mask posn))))))
2465 (deftransform %deposit-field ((new size posn int)
2467 (unsigned-byte #.sb!vm:n-word-bits))
2468 "convert to inline logical operations"
2469 `(let ((mask (ash (ldb (byte size 0) -1) posn)))
2470 (logior (logand new mask)
2471 (logand int (lognot mask)))))
2473 (deftransform %deposit-field ((new size posn int)
2475 (signed-byte #.sb!vm:n-word-bits))
2476 "convert to inline logical operations"
2477 `(let ((mask (ash (ldb (byte size 0) -1) posn)))
2478 (logior (logand new mask)
2479 (logand int (lognot mask)))))
2481 ;;; miscellanous numeric transforms
2483 ;;; If a constant appears as the first arg, swap the args.
2484 (deftransform commutative-arg-swap ((x y) * * :defun-only t :node node)
2485 (if (and (constant-continuation-p x)
2486 (not (constant-continuation-p y)))
2487 `(,(continuation-fun-name (basic-combination-fun node))
2489 ,(continuation-value x))
2490 (give-up-ir1-transform)))
2492 (dolist (x '(= char= + * logior logand logxor))
2493 (%deftransform x '(function * *) #'commutative-arg-swap
2494 "place constant arg last"))
2496 ;;; Handle the case of a constant BOOLE-CODE.
2497 (deftransform boole ((op x y) * * :when :both)
2498 "convert to inline logical operations"
2499 (unless (constant-continuation-p op)
2500 (give-up-ir1-transform "BOOLE code is not a constant."))
2501 (let ((control (continuation-value op)))
2507 (#.boole-c1 '(lognot x))
2508 (#.boole-c2 '(lognot y))
2509 (#.boole-and '(logand x y))
2510 (#.boole-ior '(logior x y))
2511 (#.boole-xor '(logxor x y))
2512 (#.boole-eqv '(logeqv x y))
2513 (#.boole-nand '(lognand x y))
2514 (#.boole-nor '(lognor x y))
2515 (#.boole-andc1 '(logandc1 x y))
2516 (#.boole-andc2 '(logandc2 x y))
2517 (#.boole-orc1 '(logorc1 x y))
2518 (#.boole-orc2 '(logorc2 x y))
2520 (abort-ir1-transform "~S is an illegal control arg to BOOLE."
2523 ;;;; converting special case multiply/divide to shifts
2525 ;;; If arg is a constant power of two, turn * into a shift.
2526 (deftransform * ((x y) (integer integer) * :when :both)
2527 "convert x*2^k to shift"
2528 (unless (constant-continuation-p y)
2529 (give-up-ir1-transform))
2530 (let* ((y (continuation-value y))
2532 (len (1- (integer-length y-abs))))
2533 (unless (= y-abs (ash 1 len))
2534 (give-up-ir1-transform))
2539 ;;; If both arguments and the result are (UNSIGNED-BYTE 32), try to
2540 ;;; come up with a ``better'' multiplication using multiplier
2541 ;;; recoding. There are two different ways the multiplier can be
2542 ;;; recoded. The more obvious is to shift X by the correct amount for
2543 ;;; each bit set in Y and to sum the results. But if there is a string
2544 ;;; of bits that are all set, you can add X shifted by one more then
2545 ;;; the bit position of the first set bit and subtract X shifted by
2546 ;;; the bit position of the last set bit. We can't use this second
2547 ;;; method when the high order bit is bit 31 because shifting by 32
2548 ;;; doesn't work too well.
2549 (deftransform * ((x y)
2550 ((unsigned-byte 32) (unsigned-byte 32))
2552 "recode as shift and add"
2553 (unless (constant-continuation-p y)
2554 (give-up-ir1-transform))
2555 (let ((y (continuation-value y))
2558 (labels ((tub32 (x) `(truly-the (unsigned-byte 32) ,x))
2563 `(+ ,result ,(tub32 next-factor))
2565 (declare (inline add))
2566 (dotimes (bitpos 32)
2568 (when (not (logbitp bitpos y))
2569 (add (if (= (1+ first-one) bitpos)
2570 ;; There is only a single bit in the string.
2572 ;; There are at least two.
2573 `(- ,(tub32 `(ash x ,bitpos))
2574 ,(tub32 `(ash x ,first-one)))))
2575 (setf first-one nil))
2576 (when (logbitp bitpos y)
2577 (setf first-one bitpos))))
2579 (cond ((= first-one 31))
2583 (add `(- ,(tub32 '(ash x 31)) ,(tub32 `(ash x ,first-one))))))
2587 ;;; If arg is a constant power of two, turn FLOOR into a shift and
2588 ;;; mask. If CEILING, add in (1- (ABS Y)) and then do FLOOR.
2589 (flet ((frob (y ceil-p)
2590 (unless (constant-continuation-p y)
2591 (give-up-ir1-transform))
2592 (let* ((y (continuation-value y))
2594 (len (1- (integer-length y-abs))))
2595 (unless (= y-abs (ash 1 len))
2596 (give-up-ir1-transform))
2597 (let ((shift (- len))
2599 `(let ,(when ceil-p `((x (+ x ,(1- y-abs)))))
2601 `(values (ash (- x) ,shift)
2602 (- (logand (- x) ,mask)))
2603 `(values (ash x ,shift)
2604 (logand x ,mask))))))))
2605 (deftransform floor ((x y) (integer integer) *)
2606 "convert division by 2^k to shift"
2608 (deftransform ceiling ((x y) (integer integer) *)
2609 "convert division by 2^k to shift"
2612 ;;; Do the same for MOD.
2613 (deftransform mod ((x y) (integer integer) * :when :both)
2614 "convert remainder mod 2^k to LOGAND"
2615 (unless (constant-continuation-p y)
2616 (give-up-ir1-transform))
2617 (let* ((y (continuation-value y))
2619 (len (1- (integer-length y-abs))))
2620 (unless (= y-abs (ash 1 len))
2621 (give-up-ir1-transform))
2622 (let ((mask (1- y-abs)))
2624 `(- (logand (- x) ,mask))
2625 `(logand x ,mask)))))
2627 ;;; If arg is a constant power of two, turn TRUNCATE into a shift and mask.
2628 (deftransform truncate ((x y) (integer integer))
2629 "convert division by 2^k to shift"
2630 (unless (constant-continuation-p y)
2631 (give-up-ir1-transform))
2632 (let* ((y (continuation-value y))
2634 (len (1- (integer-length y-abs))))
2635 (unless (= y-abs (ash 1 len))
2636 (give-up-ir1-transform))
2637 (let* ((shift (- len))
2640 (values ,(if (minusp y)
2642 `(- (ash (- x) ,shift)))
2643 (- (logand (- x) ,mask)))
2644 (values ,(if (minusp y)
2645 `(- (ash (- x) ,shift))
2647 (logand x ,mask))))))
2649 ;;; And the same for REM.
2650 (deftransform rem ((x y) (integer integer) * :when :both)
2651 "convert remainder mod 2^k to LOGAND"
2652 (unless (constant-continuation-p y)
2653 (give-up-ir1-transform))
2654 (let* ((y (continuation-value y))
2656 (len (1- (integer-length y-abs))))
2657 (unless (= y-abs (ash 1 len))
2658 (give-up-ir1-transform))
2659 (let ((mask (1- y-abs)))
2661 (- (logand (- x) ,mask))
2662 (logand x ,mask)))))
2664 ;;;; arithmetic and logical identity operation elimination
2666 ;;; Flush calls to various arith functions that convert to the
2667 ;;; identity function or a constant.
2668 (macrolet ((def-frob (name identity result)
2669 `(deftransform ,name ((x y) (* (constant-arg (member ,identity)))
2671 "fold identity operations"
2674 (def-frob logand -1 x)
2675 (def-frob logand 0 0)
2676 (def-frob logior 0 x)
2677 (def-frob logior -1 -1)
2678 (def-frob logxor -1 (lognot x))
2679 (def-frob logxor 0 x))
2681 ;;; These are restricted to rationals, because (- 0 0.0) is 0.0, not -0.0, and
2682 ;;; (* 0 -4.0) is -0.0.
2683 (deftransform - ((x y) ((constant-arg (member 0)) rational) *
2685 "convert (- 0 x) to negate"
2687 (deftransform * ((x y) (rational (constant-arg (member 0))) *
2689 "convert (* x 0) to 0"
2692 ;;; Return T if in an arithmetic op including continuations X and Y,
2693 ;;; the result type is not affected by the type of X. That is, Y is at
2694 ;;; least as contagious as X.
2696 (defun not-more-contagious (x y)
2697 (declare (type continuation x y))
2698 (let ((x (continuation-type x))
2699 (y (continuation-type y)))
2700 (values (type= (numeric-contagion x y)
2701 (numeric-contagion y y)))))
2702 ;;; Patched version by Raymond Toy. dtc: Should be safer although it
2703 ;;; XXX needs more work as valid transforms are missed; some cases are
2704 ;;; specific to particular transform functions so the use of this
2705 ;;; function may need a re-think.
2706 (defun not-more-contagious (x y)
2707 (declare (type continuation x y))
2708 (flet ((simple-numeric-type (num)
2709 (and (numeric-type-p num)
2710 ;; Return non-NIL if NUM is integer, rational, or a float
2711 ;; of some type (but not FLOAT)
2712 (case (numeric-type-class num)
2716 (numeric-type-format num))
2719 (let ((x (continuation-type x))
2720 (y (continuation-type y)))
2721 (if (and (simple-numeric-type x)
2722 (simple-numeric-type y))
2723 (values (type= (numeric-contagion x y)
2724 (numeric-contagion y y)))))))
2728 ;;; If y is not constant, not zerop, or is contagious, or a positive
2729 ;;; float +0.0 then give up.
2730 (deftransform + ((x y) (t (constant-arg t)) * :when :both)
2732 (let ((val (continuation-value y)))
2733 (unless (and (zerop val)
2734 (not (and (floatp val) (plusp (float-sign val))))
2735 (not-more-contagious y x))
2736 (give-up-ir1-transform)))
2741 ;;; If y is not constant, not zerop, or is contagious, or a negative
2742 ;;; float -0.0 then give up.
2743 (deftransform - ((x y) (t (constant-arg t)) * :when :both)
2745 (let ((val (continuation-value y)))
2746 (unless (and (zerop val)
2747 (not (and (floatp val) (minusp (float-sign val))))
2748 (not-more-contagious y x))
2749 (give-up-ir1-transform)))
2752 ;;; Fold (OP x +/-1)
2753 (macrolet ((def-frob (name result minus-result)
2754 `(deftransform ,name ((x y) (t (constant-arg real))
2756 "fold identity operations"
2757 (let ((val (continuation-value y)))
2758 (unless (and (= (abs val) 1)
2759 (not-more-contagious y x))
2760 (give-up-ir1-transform))
2761 (if (minusp val) ',minus-result ',result)))))
2762 (def-frob * x (%negate x))
2763 (def-frob / x (%negate x))
2764 (def-frob expt x (/ 1 x)))
2766 ;;; Fold (expt x n) into multiplications for small integral values of
2767 ;;; N; convert (expt x 1/2) to sqrt.
2768 (deftransform expt ((x y) (t (constant-arg real)) *)
2769 "recode as multiplication or sqrt"
2770 (let ((val (continuation-value y)))
2771 ;; If Y would cause the result to be promoted to the same type as
2772 ;; Y, we give up. If not, then the result will be the same type
2773 ;; as X, so we can replace the exponentiation with simple
2774 ;; multiplication and division for small integral powers.
2775 (unless (not-more-contagious y x)
2776 (give-up-ir1-transform))
2777 (cond ((zerop val) '(float 1 x))
2778 ((= val 2) '(* x x))
2779 ((= val -2) '(/ (* x x)))
2780 ((= val 3) '(* x x x))
2781 ((= val -3) '(/ (* x x x)))
2782 ((= val 1/2) '(sqrt x))
2783 ((= val -1/2) '(/ (sqrt x)))
2784 (t (give-up-ir1-transform)))))
2786 ;;; KLUDGE: Shouldn't (/ 0.0 0.0), etc. cause exceptions in these
2787 ;;; transformations?
2788 ;;; Perhaps we should have to prove that the denominator is nonzero before
2789 ;;; doing them? -- WHN 19990917
2790 (macrolet ((def-frob (name)
2791 `(deftransform ,name ((x y) ((constant-arg (integer 0 0)) integer)
2798 (macrolet ((def-frob (name)
2799 `(deftransform ,name ((x y) ((constant-arg (integer 0 0)) integer)
2809 ;;;; character operations
2811 (deftransform char-equal ((a b) (base-char base-char))
2813 '(let* ((ac (char-code a))
2815 (sum (logxor ac bc)))
2817 (when (eql sum #x20)
2818 (let ((sum (+ ac bc)))
2819 (and (> sum 161) (< sum 213)))))))
2821 (deftransform char-upcase ((x) (base-char))
2823 '(let ((n-code (char-code x)))
2824 (if (and (> n-code #o140) ; Octal 141 is #\a.
2825 (< n-code #o173)) ; Octal 172 is #\z.
2826 (code-char (logxor #x20 n-code))
2829 (deftransform char-downcase ((x) (base-char))
2831 '(let ((n-code (char-code x)))
2832 (if (and (> n-code 64) ; 65 is #\A.
2833 (< n-code 91)) ; 90 is #\Z.
2834 (code-char (logxor #x20 n-code))
2837 ;;;; equality predicate transforms
2839 ;;; Return true if X and Y are continuations whose only use is a
2840 ;;; reference to the same leaf, and the value of the leaf cannot
2842 (defun same-leaf-ref-p (x y)
2843 (declare (type continuation x y))
2844 (let ((x-use (continuation-use x))
2845 (y-use (continuation-use y)))
2848 (eq (ref-leaf x-use) (ref-leaf y-use))
2849 (constant-reference-p x-use))))
2851 ;;; If X and Y are the same leaf, then the result is true. Otherwise,
2852 ;;; if there is no intersection between the types of the arguments,
2853 ;;; then the result is definitely false.
2854 (deftransform simple-equality-transform ((x y) * *
2857 (cond ((same-leaf-ref-p x y)
2859 ((not (types-equal-or-intersect (continuation-type x)
2860 (continuation-type y)))
2863 (give-up-ir1-transform))))
2865 (macrolet ((def-frob (x)
2866 `(%deftransform ',x '(function * *) #'simple-equality-transform)))
2871 ;;; This is similar to SIMPLE-EQUALITY-PREDICATE, except that we also
2872 ;;; try to convert to a type-specific predicate or EQ:
2873 ;;; -- If both args are characters, convert to CHAR=. This is better than
2874 ;;; just converting to EQ, since CHAR= may have special compilation
2875 ;;; strategies for non-standard representations, etc.
2876 ;;; -- If either arg is definitely not a number, then we can compare
2878 ;;; -- Otherwise, we try to put the arg we know more about second. If X
2879 ;;; is constant then we put it second. If X is a subtype of Y, we put
2880 ;;; it second. These rules make it easier for the back end to match
2881 ;;; these interesting cases.
2882 ;;; -- If Y is a fixnum, then we quietly pass because the back end can
2883 ;;; handle that case, otherwise give an efficiency note.
2884 (deftransform eql ((x y) * * :when :both)
2885 "convert to simpler equality predicate"
2886 (let ((x-type (continuation-type x))
2887 (y-type (continuation-type y))
2888 (char-type (specifier-type 'character))
2889 (number-type (specifier-type 'number)))
2890 (cond ((same-leaf-ref-p x y)
2892 ((not (types-equal-or-intersect x-type y-type))
2894 ((and (csubtypep x-type char-type)
2895 (csubtypep y-type char-type))
2897 ((or (not (types-equal-or-intersect x-type number-type))
2898 (not (types-equal-or-intersect y-type number-type)))
2900 ((and (not (constant-continuation-p y))
2901 (or (constant-continuation-p x)
2902 (and (csubtypep x-type y-type)
2903 (not (csubtypep y-type x-type)))))
2906 (give-up-ir1-transform)))))
2908 ;;; Convert to EQL if both args are rational and complexp is specified
2909 ;;; and the same for both.
2910 (deftransform = ((x y) * * :when :both)
2912 (let ((x-type (continuation-type x))
2913 (y-type (continuation-type y)))
2914 (if (and (csubtypep x-type (specifier-type 'number))
2915 (csubtypep y-type (specifier-type 'number)))
2916 (cond ((or (and (csubtypep x-type (specifier-type 'float))
2917 (csubtypep y-type (specifier-type 'float)))
2918 (and (csubtypep x-type (specifier-type '(complex float)))
2919 (csubtypep y-type (specifier-type '(complex float)))))
2920 ;; They are both floats. Leave as = so that -0.0 is
2921 ;; handled correctly.
2922 (give-up-ir1-transform))
2923 ((or (and (csubtypep x-type (specifier-type 'rational))
2924 (csubtypep y-type (specifier-type 'rational)))
2925 (and (csubtypep x-type
2926 (specifier-type '(complex rational)))
2928 (specifier-type '(complex rational)))))
2929 ;; They are both rationals and complexp is the same.
2933 (give-up-ir1-transform
2934 "The operands might not be the same type.")))
2935 (give-up-ir1-transform
2936 "The operands might not be the same type."))))
2938 ;;; If CONT's type is a numeric type, then return the type, otherwise
2939 ;;; GIVE-UP-IR1-TRANSFORM.
2940 (defun numeric-type-or-lose (cont)
2941 (declare (type continuation cont))
2942 (let ((res (continuation-type cont)))
2943 (unless (numeric-type-p res) (give-up-ir1-transform))
2946 ;;; See whether we can statically determine (< X Y) using type
2947 ;;; information. If X's high bound is < Y's low, then X < Y.
2948 ;;; Similarly, if X's low is >= to Y's high, the X >= Y (so return
2949 ;;; NIL). If not, at least make sure any constant arg is second.
2951 ;;; FIXME: Why should constant argument be second? It would be nice to
2952 ;;; find out and explain.
2953 #+sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2954 (defun ir1-transform-< (x y first second inverse)
2955 (if (same-leaf-ref-p x y)
2957 (let* ((x-type (numeric-type-or-lose x))
2958 (x-lo (numeric-type-low x-type))
2959 (x-hi (numeric-type-high x-type))
2960 (y-type (numeric-type-or-lose y))
2961 (y-lo (numeric-type-low y-type))
2962 (y-hi (numeric-type-high y-type)))
2963 (cond ((and x-hi y-lo (< x-hi y-lo))
2965 ((and y-hi x-lo (>= x-lo y-hi))
2967 ((and (constant-continuation-p first)
2968 (not (constant-continuation-p second)))
2971 (give-up-ir1-transform))))))
2972 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2973 (defun ir1-transform-< (x y first second inverse)
2974 (if (same-leaf-ref-p x y)
2976 (let ((xi (numeric-type->interval (numeric-type-or-lose x)))
2977 (yi (numeric-type->interval (numeric-type-or-lose y))))
2978 (cond ((interval-< xi yi)
2980 ((interval->= xi yi)
2982 ((and (constant-continuation-p first)
2983 (not (constant-continuation-p second)))
2986 (give-up-ir1-transform))))))
2988 (deftransform < ((x y) (integer integer) * :when :both)
2989 (ir1-transform-< x y x y '>))
2991 (deftransform > ((x y) (integer integer) * :when :both)
2992 (ir1-transform-< y x x y '<))
2994 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2995 (deftransform < ((x y) (float float) * :when :both)
2996 (ir1-transform-< x y x y '>))
2998 #-sb-xc-host ; (See CROSS-FLOAT-INFINITY-KLUDGE.)
2999 (deftransform > ((x y) (float float) * :when :both)
3000 (ir1-transform-< y x x y '<))
3002 ;;;; converting N-arg comparisons
3004 ;;;; We convert calls to N-arg comparison functions such as < into
3005 ;;;; two-arg calls. This transformation is enabled for all such
3006 ;;;; comparisons in this file. If any of these predicates are not
3007 ;;;; open-coded, then the transformation should be removed at some
3008 ;;;; point to avoid pessimization.
3010 ;;; This function is used for source transformation of N-arg
3011 ;;; comparison functions other than inequality. We deal both with
3012 ;;; converting to two-arg calls and inverting the sense of the test,
3013 ;;; if necessary. If the call has two args, then we pass or return a
3014 ;;; negated test as appropriate. If it is a degenerate one-arg call,
3015 ;;; then we transform to code that returns true. Otherwise, we bind
3016 ;;; all the arguments and expand into a bunch of IFs.
3017 (declaim (ftype (function (symbol list boolean) *) multi-compare))
3018 (defun multi-compare (predicate args not-p)
3019 (let ((nargs (length args)))
3020 (cond ((< nargs 1) (values nil t))
3021 ((= nargs 1) `(progn ,@args t))
3024 `(if (,predicate ,(first args) ,(second args)) nil t)
3027 (do* ((i (1- nargs) (1- i))
3029 (current (gensym) (gensym))
3030 (vars (list current) (cons current vars))
3032 `(if (,predicate ,current ,last)
3034 `(if (,predicate ,current ,last)
3037 `((lambda ,vars ,result) . ,args)))))))
3039 (define-source-transform = (&rest args) (multi-compare '= args nil))
3040 (define-source-transform < (&rest args) (multi-compare '< args nil))
3041 (define-source-transform > (&rest args) (multi-compare '> args nil))
3042 (define-source-transform <= (&rest args) (multi-compare '> args t))
3043 (define-source-transform >= (&rest args) (multi-compare '< args t))
3045 (define-source-transform char= (&rest args) (multi-compare 'char= args nil))
3046 (define-source-transform char< (&rest args) (multi-compare 'char< args nil))
3047 (define-source-transform char> (&rest args) (multi-compare 'char> args nil))
3048 (define-source-transform char<= (&rest args) (multi-compare 'char> args t))
3049 (define-source-transform char>= (&rest args) (multi-compare 'char< args t))
3051 (define-source-transform char-equal (&rest args)
3052 (multi-compare 'char-equal args nil))
3053 (define-source-transform char-lessp (&rest args)
3054 (multi-compare 'char-lessp args nil))
3055 (define-source-transform char-greaterp (&rest args)
3056 (multi-compare 'char-greaterp args nil))
3057 (define-source-transform char-not-greaterp (&rest args)
3058 (multi-compare 'char-greaterp args t))
3059 (define-source-transform char-not-lessp (&rest args)
3060 (multi-compare 'char-lessp args t))
3062 ;;; This function does source transformation of N-arg inequality
3063 ;;; functions such as /=. This is similar to Multi-Compare in the <3
3064 ;;; arg cases. If there are more than two args, then we expand into
3065 ;;; the appropriate n^2 comparisons only when speed is important.
3066 (declaim (ftype (function (symbol list) *) multi-not-equal))
3067 (defun multi-not-equal (predicate args)
3068 (let ((nargs (length args)))
3069 (cond ((< nargs 1) (values nil t))
3070 ((= nargs 1) `(progn ,@args t))
3072 `(if (,predicate ,(first args) ,(second args)) nil t))
3073 ((not (policy *lexenv*
3074 (and (>= speed space)
3075 (>= speed compilation-speed))))
3078 (let ((vars (make-gensym-list nargs)))
3079 (do ((var vars next)
3080 (next (cdr vars) (cdr next))
3083 `((lambda ,vars ,result) . ,args))
3084 (let ((v1 (first var)))
3086 (setq result `(if (,predicate ,v1 ,v2) nil ,result))))))))))
3088 (define-source-transform /= (&rest args) (multi-not-equal '= args))
3089 (define-source-transform char/= (&rest args) (multi-not-equal 'char= args))
3090 (define-source-transform char-not-equal (&rest args)
3091 (multi-not-equal 'char-equal args))
3093 ;;; Expand MAX and MIN into the obvious comparisons.
3094 (define-source-transform max (arg &rest more-args)
3095 (if (null more-args)
3097 (once-only ((arg1 arg)
3098 (arg2 `(max ,@more-args)))
3099 `(if (> ,arg1 ,arg2)
3101 (define-source-transform min (arg &rest more-args)
3102 (if (null more-args)
3104 (once-only ((arg1 arg)
3105 (arg2 `(min ,@more-args)))
3106 `(if (< ,arg1 ,arg2)
3109 ;;;; converting N-arg arithmetic functions
3111 ;;;; N-arg arithmetic and logic functions are associated into two-arg
3112 ;;;; versions, and degenerate cases are flushed.
3114 ;;; Left-associate FIRST-ARG and MORE-ARGS using FUNCTION.
3115 (declaim (ftype (function (symbol t list) list) associate-args))
3116 (defun associate-args (function first-arg more-args)
3117 (let ((next (rest more-args))
3118 (arg (first more-args)))
3120 `(,function ,first-arg ,arg)
3121 (associate-args function `(,function ,first-arg ,arg) next))))
3123 ;;; Do source transformations for transitive functions such as +.
3124 ;;; One-arg cases are replaced with the arg and zero arg cases with
3125 ;;; the identity. If LEAF-FUN is true, then replace two-arg calls with
3126 ;;; a call to that function.
3127 (defun source-transform-transitive (fun args identity &optional leaf-fun)
3128 (declare (symbol fun leaf-fun) (list args))
3131 (1 `(values ,(first args)))
3133 `(,leaf-fun ,(first args) ,(second args))
3136 (associate-args fun (first args) (rest args)))))
3138 (define-source-transform + (&rest args)
3139 (source-transform-transitive '+ args 0))
3140 (define-source-transform * (&rest args)
3141 (source-transform-transitive '* args 1))
3142 (define-source-transform logior (&rest args)
3143 (source-transform-transitive 'logior args 0))
3144 (define-source-transform logxor (&rest args)
3145 (source-transform-transitive 'logxor args 0))
3146 (define-source-transform logand (&rest args)
3147 (source-transform-transitive 'logand args -1))
3149 (define-source-transform logeqv (&rest args)
3150 (if (evenp (length args))
3151 `(lognot (logxor ,@args))
3154 ;;; Note: we can't use SOURCE-TRANSFORM-TRANSITIVE for GCD and LCM
3155 ;;; because when they are given one argument, they return its absolute
3158 (define-source-transform gcd (&rest args)
3161 (1 `(abs (the integer ,(first args))))
3163 (t (associate-args 'gcd (first args) (rest args)))))
3165 (define-source-transform lcm (&rest args)
3168 (1 `(abs (the integer ,(first args))))
3170 (t (associate-args 'lcm (first args) (rest args)))))
3172 ;;; Do source transformations for intransitive n-arg functions such as
3173 ;;; /. With one arg, we form the inverse. With two args we pass.
3174 ;;; Otherwise we associate into two-arg calls.
3175 (declaim (ftype (function (symbol list t) list) source-transform-intransitive))
3176 (defun source-transform-intransitive (function args inverse)
3178 ((0 2) (values nil t))
3179 (1 `(,@inverse ,(first args)))
3180 (t (associate-args function (first args) (rest args)))))
3182 (define-source-transform - (&rest args)
3183 (source-transform-intransitive '- args '(%negate)))
3184 (define-source-transform / (&rest args)
3185 (source-transform-intransitive '/ args '(/ 1)))
3187 ;;;; transforming APPLY
3189 ;;; We convert APPLY into MULTIPLE-VALUE-CALL so that the compiler
3190 ;;; only needs to understand one kind of variable-argument call. It is
3191 ;;; more efficient to convert APPLY to MV-CALL than MV-CALL to APPLY.
3192 (define-source-transform apply (fun arg &rest more-args)
3193 (let ((args (cons arg more-args)))
3194 `(multiple-value-call ,fun
3195 ,@(mapcar (lambda (x)
3198 (values-list ,(car (last args))))))
3200 ;;;; transforming FORMAT
3202 ;;;; If the control string is a compile-time constant, then replace it
3203 ;;;; with a use of the FORMATTER macro so that the control string is
3204 ;;;; ``compiled.'' Furthermore, if the destination is either a stream
3205 ;;;; or T and the control string is a function (i.e. FORMATTER), then
3206 ;;;; convert the call to FORMAT to just a FUNCALL of that function.
3208 (deftransform format ((dest control &rest args) (t simple-string &rest t) *
3209 :policy (> speed space))
3210 (unless (constant-continuation-p control)
3211 (give-up-ir1-transform "The control string is not a constant."))
3212 (let ((arg-names (make-gensym-list (length args))))
3213 `(lambda (dest control ,@arg-names)
3214 (declare (ignore control))
3215 (format dest (formatter ,(continuation-value control)) ,@arg-names))))
3217 (deftransform format ((stream control &rest args) (stream function &rest t) *
3218 :policy (> speed space))
3219 (let ((arg-names (make-gensym-list (length args))))
3220 `(lambda (stream control ,@arg-names)
3221 (funcall control stream ,@arg-names)
3224 (deftransform format ((tee control &rest args) ((member t) function &rest t) *
3225 :policy (> speed space))
3226 (let ((arg-names (make-gensym-list (length args))))
3227 `(lambda (tee control ,@arg-names)
3228 (declare (ignore tee))
3229 (funcall control *standard-output* ,@arg-names)
3232 (defoptimizer (coerce derive-type) ((value type))
3233 (let ((value-type (continuation-type value))
3234 (type-type (continuation-type type)))
3236 ((good-cons-type-p (cons-type)
3237 ;; Make sure the cons-type we're looking at is something
3238 ;; we're prepared to handle which is basically something
3239 ;; that array-element-type can return.
3240 (or (and (member-type-p cons-type)
3241 (null (rest (member-type-members cons-type)))
3242 (null (first (member-type-members cons-type))))
3243 (let ((car-type (cons-type-car-type cons-type)))
3244 (and (member-type-p car-type)
3245 (null (rest (member-type-members car-type)))
3246 (or (symbolp (first (member-type-members car-type)))
3247 (numberp (first (member-type-members car-type)))
3248 (and (listp (first (member-type-members car-type)))
3249 (numberp (first (first (member-type-members
3251 (good-cons-type-p (cons-type-cdr-type cons-type))))))
3252 (unconsify-type (good-cons-type)
3253 ;; Convert the "printed" respresentation of a cons
3254 ;; specifier into a type specifier. That is, the specifier
3255 ;; (cons (eql signed-byte) (cons (eql 16) null)) is
3256 ;; converted to (signed-byte 16).
3257 (cond ((or (null good-cons-type)
3258 (eq good-cons-type 'null))
3260 ((and (eq (first good-cons-type) 'cons)
3261 (eq (first (second good-cons-type)) 'member))
3262 `(,(second (second good-cons-type))
3263 ,@(unconsify-type (caddr good-cons-type))))))
3264 (coerceable-p (c-type)
3265 ;; Can the value be coerced to the given type? Coerce is
3266 ;; complicated, so we don't handle every possible case
3267 ;; here---just the most common and easiest cases:
3269 ;; o Any real can be coerced to a float type.
3270 ;; o Any number can be coerced to a complex single/double-float.
3271 ;; o An integer can be coerced to an integer.
3272 (let ((coerced-type c-type))
3273 (or (and (subtypep coerced-type 'float)
3274 (csubtypep value-type (specifier-type 'real)))
3275 (and (subtypep coerced-type
3276 '(or (complex single-float)
3277 (complex double-float)))
3278 (csubtypep value-type (specifier-type 'number)))
3279 (and (subtypep coerced-type 'integer)
3280 (csubtypep value-type (specifier-type 'integer))))))
3281 (process-types (type)
3283 ;; This needs some work because we should be able to derive
3284 ;; the resulting type better than just the type arg of
3285 ;; coerce. That is, if x is (integer 10 20), the (coerce x
3286 ;; 'double-float) should say (double-float 10d0 20d0)
3287 ;; instead of just double-float.
3288 (cond ((member-type-p type)
3289 (let ((members (member-type-members type)))
3290 (if (every #'coerceable-p members)
3291 (specifier-type `(or ,@members))
3293 ((and (cons-type-p type)
3294 (good-cons-type-p type))
3295 (let ((c-type (unconsify-type (type-specifier type))))
3296 (if (coerceable-p c-type)
3297 (specifier-type c-type)
3300 *universal-type*))))
3301 (cond ((union-type-p type-type)
3302 (apply #'type-union (mapcar #'process-types
3303 (union-type-types type-type))))
3304 ((or (member-type-p type-type)
3305 (cons-type-p type-type))
3306 (process-types type-type))
3308 *universal-type*)))))
3310 (defoptimizer (array-element-type derive-type) ((array))
3311 (let* ((array-type (continuation-type array)))
3312 (labels ((consify (list)
3315 `(cons (eql ,(car list)) ,(consify (rest list)))))
3316 (get-element-type (a)
3318 (type-specifier (array-type-specialized-element-type a))))
3319 (cond ((eq element-type '*)
3320 (specifier-type 'type-specifier))
3321 ((symbolp element-type)
3322 (make-member-type :members (list element-type)))
3323 ((consp element-type)
3324 (specifier-type (consify element-type)))
3326 (error "can't understand type ~S~%" element-type))))))
3327 (cond ((array-type-p array-type)
3328 (get-element-type array-type))
3329 ((union-type-p array-type)
3331 (mapcar #'get-element-type (union-type-types array-type))))
3333 *universal-type*)))))
3335 ;;;; debuggers' little helpers
3337 ;;; for debugging when transforms are behaving mysteriously,
3338 ;;; e.g. when debugging a problem with an ASH transform
3339 ;;; (defun foo (&optional s)
3340 ;;; (sb-c::/report-continuation s "S outside WHEN")
3341 ;;; (when (and (integerp s) (> s 3))
3342 ;;; (sb-c::/report-continuation s "S inside WHEN")
3343 ;;; (let ((bound (ash 1 (1- s))))
3344 ;;; (sb-c::/report-continuation bound "BOUND")
3345 ;;; (let ((x (- bound))
3347 ;;; (sb-c::/report-continuation x "X")
3348 ;;; (sb-c::/report-continuation x "Y"))
3349 ;;; `(integer ,(- bound) ,(1- bound)))))
3350 ;;; (The DEFTRANSFORM doesn't do anything but report at compile time,
3351 ;;; and the function doesn't do anything at all.)
3354 (defknown /report-continuation (t t) null)
3355 (deftransform /report-continuation ((x message) (t t))
3356 (format t "~%/in /REPORT-CONTINUATION~%")
3357 (format t "/(CONTINUATION-TYPE X)=~S~%" (continuation-type x))
3358 (when (constant-continuation-p x)
3359 (format t "/(CONTINUATION-VALUE X)=~S~%" (continuation-value x)))
3360 (format t "/MESSAGE=~S~%" (continuation-value message))
3361 (give-up-ir1-transform "not a real transform"))
3362 (defun /report-continuation (&rest rest)
3363 (declare (ignore rest))))