(in-package "SB!IMPL")
-;;; Like CMU CL, we use HEAPSORT. However, other than that, this code
-;;; isn't really related to the CMU CL code, since instead of trying
-;;; to generalize the CMU CL code to allow START and END values, this
-;;; code has been written from scratch following Chapter 7 of
-;;; _Introduction to Algorithms_ by Corman, Rivest, and Shamir.
-(defun sort-vector (vector start end predicate key)
- (sort-vector vector start end predicate key))
+(defun sort-vector (vector start end predicate-fun key-fun-or-nil)
+ (sort-vector vector start end predicate-fun key-fun-or-nil))
;;; This is MAYBE-INLINE because it's not too hard to have an
;;; application where sorting is a major bottleneck, and inlining it
;;; allows the compiler to make enough optimizations that it might be
;;; worth the (large) cost in space.
(declaim (maybe-inline sort))
-(defun sort (sequence predicate &key key)
+(defun sort (sequence predicate &rest args &key key)
#!+sb-doc
"Destructively sort SEQUENCE. PREDICATE should return non-NIL if
ARG1 is to precede ARG2."
- (let ((predicate-function (%coerce-callable-to-fun predicate))
- (key-function (and key (%coerce-callable-to-fun key))))
- (typecase sequence
- (list (stable-sort-list sequence predicate-function key-function))
- (vector
- (with-array-data ((vector (the vector sequence))
- (start 0)
- (end (length sequence)))
- (sort-vector vector start end predicate-function key-function))
- sequence)
- (t
- (error 'simple-type-error
- :datum sequence
- :expected-type 'sequence
- :format-control "~S is not a sequence."
- :format-arguments (list sequence))))))
+ (declare (dynamic-extent args))
+ (let ((predicate-fun (%coerce-callable-to-fun predicate)))
+ (seq-dispatch sequence
+ (stable-sort-list sequence
+ predicate-fun
+ (if key (%coerce-callable-to-fun key) #'identity))
+ (let ((key-fun-or-nil (and key (%coerce-callable-to-fun key))))
+ (with-array-data ((vector (the vector sequence))
+ (start 0)
+ (end (length sequence)))
+ (sort-vector vector start end predicate-fun key-fun-or-nil))
+ sequence)
+ (apply #'sb!sequence:sort sequence predicate args))))
\f
;;;; stable sorting
-
-(defun stable-sort (sequence predicate &key key)
+(defun stable-sort (sequence predicate &rest args &key key)
#!+sb-doc
"Destructively sort SEQUENCE. PREDICATE should return non-NIL if
ARG1 is to precede ARG2."
- (typecase sequence
- (simple-vector
- (stable-sort-simple-vector sequence predicate key))
- (list
- (stable-sort-list sequence predicate key))
- (vector
- (stable-sort-vector sequence predicate key))
- (t
- (error 'simple-type-error
- :datum sequence
- :expected-type 'sequence
- :format-control "~S is not a sequence."
- :format-arguments (list sequence)))))
+ (declare (dynamic-extent args))
+ (let ((predicate-fun (%coerce-callable-to-fun predicate)))
+ (seq-dispatch sequence
+ (stable-sort-list sequence
+ predicate-fun
+ (if key (%coerce-callable-to-fun key) #'identity))
+ (if (typep sequence 'simple-vector)
+ (stable-sort-simple-vector sequence
+ predicate-fun
+ (and key (%coerce-callable-to-fun key)))
+ (stable-sort-vector sequence
+ predicate-fun
+ (and key (%coerce-callable-to-fun key))))
+ (apply #'sb!sequence:stable-sort sequence predicate args))))
\f
-;;; APPLY-KEYED-PRED saves us a function call sometimes.
+;;; FUNCALL-USING-KEY saves us a function call sometimes.
(eval-when (:compile-toplevel :execute)
- (sb!xc:defmacro apply-keyed-pred (one two pred key)
+ (sb!xc:defmacro funcall2-using-key (pred key one two)
`(if ,key
- (funcall ,pred (funcall ,key ,one)
- (funcall ,key ,two))
- (funcall ,pred ,one ,two)))
+ (funcall ,pred (funcall ,key ,one)
+ (funcall ,key ,two))
+ (funcall ,pred ,one ,two)))
) ; EVAL-WHEN
\f
;;;; stable sort of lists
(defun last-cons-of (list)
(loop (let ((rest (rest list)))
- (if rest
- (setf list rest)
- (return list)))))
+ (if rest
+ (setf list rest)
+ (return list)))))
;;; Destructively merge LIST-1 with LIST-2 (given that they're already
;;; sorted w.r.t. PRED-FUN on KEY-FUN, giving output sorted the same
;; Now maybe we're done.
(if (endp ,list-i)
(return (values (nreconc
- reversed-result-so-far
- ,other-list)
- (last-cons-of
- ,other-list)))
+ reversed-result-so-far
+ ,other-list)
+ (last-cons-of
+ ,other-list)))
(setf ,key-i
- (funcall key-fun (car ,list-i)))))))
+ (funcall key-fun (car ,list-i)))))))
;; Note that by making KEY-2 the first arg to
;; PRED-FUN, we arrange that if PRED-FUN is a function
;; in the #'< style, the outcome is stably sorted.
;;; that is, there are only two runs that can be merged, the first run
;;; starting at the beginning of the list, and the second being the
;;; remaining elements.
-(defun stable-sort-list (list pred key)
+(defun stable-sort-list (list pred-fun key-fun)
(let ((head (cons :header list)) ; head holds on to everything
- (n 1) ; bottom-up size of lists to be merged
- unsorted ; unsorted is the remaining list to be
- ; broken into n size lists and merged
- list-1 ; list-1 is one length n list to be merged
- last ; last points to the last visited cell
- (pred-fun (%coerce-callable-to-fun pred))
- (key-fun (if key
- (%coerce-callable-to-fun key)
- #'identity)))
- (declare (fixnum n))
+ (n 1) ; bottom-up size of lists to be merged
+ unsorted ; unsorted is the remaining list to be
+ ; broken into n size lists and merged
+ list-1 ; list-1 is one length n list to be merged
+ last) ; last points to the last visited cell
+ (declare (type function pred-fun key-fun)
+ (type fixnum n))
(loop
;; Start collecting runs of N at the first element.
(setf unsorted (cdr head))
(let ((n-1 (1- n)))
(declare (fixnum n-1))
(loop
- (setf list-1 unsorted)
- (let ((temp (nthcdr n-1 list-1))
- list-2)
- (cond (temp
- ;; There are enough elements for a second run.
- (setf list-2 (cdr temp))
- (setf (cdr temp) nil)
- (setf temp (nthcdr n-1 list-2))
- (cond (temp
- (setf unsorted (cdr temp))
- (setf (cdr temp) nil))
- ;; The second run goes off the end of the list.
- (t (setf unsorted nil)))
- (multiple-value-bind (merged-head merged-last)
- (merge-lists* list-1 list-2 pred-fun key-fun)
- (setf (cdr last) merged-head
- last merged-last))
- (if (null unsorted) (return)))
- ;; If there is only one run, then tack it on to the end.
- (t (setf (cdr last) list-1)
- (return)))))
+ (setf list-1 unsorted)
+ (let ((temp (nthcdr n-1 list-1))
+ list-2)
+ (cond (temp
+ ;; There are enough elements for a second run.
+ (setf list-2 (cdr temp))
+ (setf (cdr temp) nil)
+ (setf temp (nthcdr n-1 list-2))
+ (cond (temp
+ (setf unsorted (cdr temp))
+ (setf (cdr temp) nil))
+ ;; The second run goes off the end of the list.
+ (t (setf unsorted nil)))
+ (multiple-value-bind (merged-head merged-last)
+ (merge-lists* list-1 list-2 pred-fun key-fun)
+ (setf (cdr last) merged-head
+ last merged-last))
+ (if (null unsorted) (return)))
+ ;; If there is only one run, then tack it on to the end.
+ (t (setf (cdr last) list-1)
+ (return)))))
(setf n (ash n 1)) ; (+ n n)
;; If the inner loop only executed once, then there were only
;; enough elements for two runs given n, so all the elements
;; have been merged into one list. This may waste one outer
;; iteration to realize.
(if (eq list-1 (cdr head))
- (return list-1))))))
+ (return list-1))))))
\f
;;;; stable sort of vectors
;;; and merges them into a target vector starting at index start-1.
(sb!xc:defmacro stable-sort-merge-vectors* (source target start-1 end-1 end-2
- pred key source-ref
- target-ref)
+ pred key source-ref
+ target-ref)
(let ((i (gensym))
- (j (gensym))
- (target-i (gensym)))
+ (j (gensym))
+ (target-i (gensym)))
`(let ((,i ,start-1)
- (,j ,end-1) ; start-2
- (,target-i ,start-1))
+ (,j ,end-1) ; start-2
+ (,target-i ,start-1))
(declare (fixnum ,i ,j ,target-i))
(loop
- (cond ((= ,i ,end-1)
- (loop (if (= ,j ,end-2) (return))
- (setf (,target-ref ,target ,target-i)
- (,source-ref ,source ,j))
- (incf ,target-i)
- (incf ,j))
- (return))
- ((= ,j ,end-2)
- (loop (if (= ,i ,end-1) (return))
- (setf (,target-ref ,target ,target-i)
- (,source-ref ,source ,i))
- (incf ,target-i)
- (incf ,i))
- (return))
- ((apply-keyed-pred (,source-ref ,source ,j)
- (,source-ref ,source ,i)
- ,pred ,key)
- (setf (,target-ref ,target ,target-i)
- (,source-ref ,source ,j))
- (incf ,j))
- (t (setf (,target-ref ,target ,target-i)
- (,source-ref ,source ,i))
- (incf ,i)))
- (incf ,target-i)))))
+ (cond ((= ,i ,end-1)
+ (loop (if (= ,j ,end-2) (return))
+ (setf (,target-ref ,target ,target-i)
+ (,source-ref ,source ,j))
+ (incf ,target-i)
+ (incf ,j))
+ (return))
+ ((= ,j ,end-2)
+ (loop (if (= ,i ,end-1) (return))
+ (setf (,target-ref ,target ,target-i)
+ (,source-ref ,source ,i))
+ (incf ,target-i)
+ (incf ,i))
+ (return))
+ ((funcall2-using-key ,pred ,key
+ (,source-ref ,source ,j)
+ (,source-ref ,source ,i))
+ (setf (,target-ref ,target ,target-i)
+ (,source-ref ,source ,j))
+ (incf ,j))
+ (t (setf (,target-ref ,target ,target-i)
+ (,source-ref ,source ,i))
+ (incf ,i)))
+ (incf ,target-i)))))
;;; VECTOR-MERGE-SORT is the same algorithm used to stable sort lists,
;;; but it uses a temporary vector. DIRECTION determines whether we
;;; are merging into the temporary (T) or back into the given vector
;;; (NIL).
(sb!xc:defmacro vector-merge-sort (vector pred key vector-ref)
- (let ((vector-len (gensym)) (n (gensym))
- (direction (gensym)) (unsorted (gensym))
- (start-1 (gensym)) (end-1 (gensym))
- (end-2 (gensym)) (temp-len (gensym))
- (i (gensym)))
- `(let ((,vector-len (length (the vector ,vector)))
- (,n 1) ; bottom-up size of contiguous runs to be merged
- (,direction t) ; t vector --> temp nil temp --> vector
- (,temp-len (length (the simple-vector *merge-sort-temp-vector*)))
- (,unsorted 0) ; unsorted..vector-len are the elements that need
- ; to be merged for a given n
- (,start-1 0)) ; one n-len subsequence to be merged with the next
- (declare (fixnum ,vector-len ,n ,temp-len ,unsorted ,start-1))
+ (with-unique-names
+ (vector-len n direction unsorted start-1 end-1 end-2 temp temp-len i)
+ `(let* ((,vector-len (length (the vector ,vector)))
+ (,n 1) ; bottom-up size of contiguous runs to be merged
+ (,direction t) ; t vector --> temp nil temp --> vector
+ (,temp *merge-sort-temp-vector*)
+ (,temp-len (length ,temp))
+ (,unsorted 0) ; unsorted..vector-len are the elements that need
+ ; to be merged for a given n
+ (,start-1 0)) ; one n-len subsequence to be merged with the next
+ (declare (fixnum ,vector-len ,n ,temp-len ,unsorted ,start-1)
+ (simple-vector ,temp))
(if (> ,vector-len ,temp-len)
- (setf *merge-sort-temp-vector*
- (make-array (max ,vector-len (+ ,temp-len ,temp-len)))))
- (loop
- ;; for each n, we start taking n-runs from the start of the vector
- (setf ,unsorted 0)
- (loop
- (setf ,start-1 ,unsorted)
- (let ((,end-1 (+ ,start-1 ,n)))
- (declare (fixnum ,end-1))
- (cond ((< ,end-1 ,vector-len)
- ;; there are enough elements for a second run
- (let ((,end-2 (+ ,end-1 ,n)))
- (declare (fixnum ,end-2))
- (if (> ,end-2 ,vector-len) (setf ,end-2 ,vector-len))
- (setf ,unsorted ,end-2)
- (if ,direction
- (stable-sort-merge-vectors*
- ,vector *merge-sort-temp-vector*
- ,start-1 ,end-1 ,end-2 ,pred ,key ,vector-ref svref)
- (stable-sort-merge-vectors*
- *merge-sort-temp-vector* ,vector
- ,start-1 ,end-1 ,end-2 ,pred ,key svref ,vector-ref))
- (if (= ,unsorted ,vector-len) (return))))
- ;; if there is only one run, copy those elements to the end
- (t (if ,direction
- (do ((,i ,start-1 (1+ ,i)))
- ((= ,i ,vector-len))
- (declare (fixnum ,i))
- (setf (svref *merge-sort-temp-vector* ,i)
- (,vector-ref ,vector ,i)))
- (do ((,i ,start-1 (1+ ,i)))
- ((= ,i ,vector-len))
- (declare (fixnum ,i))
- (setf (,vector-ref ,vector ,i)
- (svref *merge-sort-temp-vector* ,i))))
- (return)))))
- ;; If the inner loop only executed once, then there were only enough
- ;; elements for two subsequences given n, so all the elements have
- ;; been merged into one list. Start-1 will have remained 0 upon exit.
- (when (zerop ,start-1)
- (if ,direction
- ;; if we just merged into the temporary, copy it all back
- ;; to the given vector.
- (dotimes (,i ,vector-len)
- (setf (,vector-ref ,vector ,i)
- (svref *merge-sort-temp-vector* ,i))))
- (return ,vector))
- (setf ,n (ash ,n 1)) ; (* 2 n)
- (setf ,direction (not ,direction))))))
+ (setf ,temp (make-array (max ,vector-len
+ (min most-positive-fixnum
+ (+ ,temp-len ,temp-len))))
+ *merge-sort-temp-vector* ,temp))
+ ;; rebind, in case PRED or KEY calls STABLE-SORT
+ (let ((*merge-sort-temp-vector* (vector)))
+ (loop
+ ;; for each n, we start taking n-runs from the start of the vector
+ (setf ,unsorted 0)
+ (loop
+ (setf ,start-1 ,unsorted)
+ (let ((,end-1 (+ ,start-1 ,n)))
+ (declare (fixnum ,end-1))
+ (cond ((< ,end-1 ,vector-len)
+ ;; there are enough elements for a second run
+ (let ((,end-2 (+ ,end-1 ,n)))
+ (declare (fixnum ,end-2))
+ (if (> ,end-2 ,vector-len) (setf ,end-2 ,vector-len))
+ (setf ,unsorted ,end-2)
+ (if ,direction
+ (stable-sort-merge-vectors*
+ ,vector ,temp
+ ,start-1 ,end-1 ,end-2 ,pred ,key ,vector-ref svref)
+ (stable-sort-merge-vectors*
+ ,temp ,vector
+ ,start-1 ,end-1 ,end-2 ,pred ,key svref ,vector-ref))
+ (if (= ,unsorted ,vector-len) (return))))
+ ;; if there is only one run, copy those elements to the end
+ (t (if ,direction
+ (do ((,i ,start-1 (1+ ,i)))
+ ((= ,i ,vector-len))
+ (declare (fixnum ,i))
+ (setf (svref ,temp ,i)
+ (,vector-ref ,vector ,i)))
+ (do ((,i ,start-1 (1+ ,i)))
+ ((= ,i ,vector-len))
+ (declare (fixnum ,i))
+ (setf (,vector-ref ,vector ,i)
+ (svref ,temp ,i))))
+ (return)))))
+ ;; If the inner loop only executed once, then there were only enough
+ ;; elements for two subsequences given n, so all the elements have
+ ;; been merged into one list. Start-1 will have remained 0 upon exit.
+ (when (zerop ,start-1)
+ (if ,direction
+ ;; if we just merged into the temporary, copy it all back
+ ;; to the given vector.
+ (dotimes (,i ,vector-len)
+ (setf (,vector-ref ,vector ,i)
+ (svref ,temp ,i))))
+ (return ,vector))
+ (setf ,n (ash ,n 1)) ; (* 2 n)
+ (setf ,direction (not ,direction)))))))
) ; EVAL-when
-;;; temporary vector for stable sorting vectors
-(defvar *merge-sort-temp-vector*
- (make-array 50))
+;;; temporary vector for stable sorting vectors, allocated for each new thread
+(defvar *merge-sort-temp-vector* (make-array 50))
(declaim (simple-vector *merge-sort-temp-vector*))
(defun stable-sort-simple-vector (vector pred key)
- (declare (simple-vector vector))
+ (declare (type simple-vector vector)
+ (type function pred)
+ (type (or null function) key))
(vector-merge-sort vector pred key svref))
(defun stable-sort-vector (vector pred key)
+ (declare (type function pred)
+ (type (or null function) key))
(vector-merge-sort vector pred key aref))
\f
;;;; merging
;;; are chosen only if they are strictly less than elements of
;;; VECTOR-1, (PRED ELT-2 ELT-1), as specified in the manual.
(sb!xc:defmacro merge-vectors (vector-1 length-1 vector-2 length-2
- result-vector pred key access)
+ result-vector pred key access)
(let ((result-i (gensym))
- (i (gensym))
- (j (gensym)))
+ (i (gensym))
+ (j (gensym)))
`(let* ((,result-i 0)
- (,i 0)
- (,j 0))
+ (,i 0)
+ (,j 0))
(declare (fixnum ,result-i ,i ,j))
(loop
- (cond ((= ,i ,length-1)
- (loop (if (= ,j ,length-2) (return))
- (setf (,access ,result-vector ,result-i)
- (,access ,vector-2 ,j))
- (incf ,result-i)
- (incf ,j))
- (return ,result-vector))
- ((= ,j ,length-2)
- (loop (if (= ,i ,length-1) (return))
- (setf (,access ,result-vector ,result-i)
- (,access ,vector-1 ,i))
- (incf ,result-i)
- (incf ,i))
- (return ,result-vector))
- ((apply-keyed-pred (,access ,vector-2 ,j) (,access ,vector-1 ,i)
- ,pred ,key)
- (setf (,access ,result-vector ,result-i)
- (,access ,vector-2 ,j))
- (incf ,j))
- (t (setf (,access ,result-vector ,result-i)
- (,access ,vector-1 ,i))
- (incf ,i)))
- (incf ,result-i)))))
+ (cond ((= ,i ,length-1)
+ (loop (if (= ,j ,length-2) (return))
+ (setf (,access ,result-vector ,result-i)
+ (,access ,vector-2 ,j))
+ (incf ,result-i)
+ (incf ,j))
+ (return ,result-vector))
+ ((= ,j ,length-2)
+ (loop (if (= ,i ,length-1) (return))
+ (setf (,access ,result-vector ,result-i)
+ (,access ,vector-1 ,i))
+ (incf ,result-i)
+ (incf ,i))
+ (return ,result-vector))
+ ((funcall2-using-key ,pred ,key
+ (,access ,vector-2 ,j) (,access ,vector-1 ,i))
+ (setf (,access ,result-vector ,result-i)
+ (,access ,vector-2 ,j))
+ (incf ,j))
+ (t (setf (,access ,result-vector ,result-i)
+ (,access ,vector-1 ,i))
+ (incf ,i)))
+ (incf ,result-i)))))
) ; EVAL-WHEN
;; reimplementing everything, we can't do the same for the LIST
;; case, so do relevant length checking here:
(let ((s1 (coerce sequence1 'list))
- (s2 (coerce sequence2 'list))
- (pred-fun (%coerce-callable-to-fun predicate))
- (key-fun (if key
- (%coerce-callable-to-fun key)
- #'identity)))
- (when (type= type (specifier-type 'list))
- (return-from merge (values (merge-lists* s1 s2 pred-fun key-fun))))
- (when (eq type *empty-type*)
- (bad-sequence-type-error nil))
- (when (type= type (specifier-type 'null))
- (if (and (null s1) (null s2))
- (return-from merge 'nil)
- ;; FIXME: This will break on circular lists (as,
- ;; indeed, will the whole MERGE function).
- (sequence-type-length-mismatch-error type
- (+ (length s1)
- (length s2)))))
- (if (csubtypep (specifier-type '(cons nil t)) type)
- (if (and (null s1) (null s2))
- (sequence-type-length-mismatch-error type 0)
- (values (merge-lists* s1 s2 pred-fun key-fun)))
- (sequence-type-too-hairy result-type))))
+ (s2 (coerce sequence2 'list))
+ (pred-fun (%coerce-callable-to-fun predicate))
+ (key-fun (if key
+ (%coerce-callable-to-fun key)
+ #'identity)))
+ (when (type= type (specifier-type 'list))
+ (return-from merge (values (merge-lists* s1 s2 pred-fun key-fun))))
+ (when (eq type *empty-type*)
+ (bad-sequence-type-error nil))
+ (when (type= type (specifier-type 'null))
+ (if (and (null s1) (null s2))
+ (return-from merge 'nil)
+ ;; FIXME: This will break on circular lists (as,
+ ;; indeed, will the whole MERGE function).
+ (sequence-type-length-mismatch-error type
+ (+ (length s1)
+ (length s2)))))
+ (if (cons-type-p type)
+ (multiple-value-bind (min exactp)
+ (sb!kernel::cons-type-length-info type)
+ (let ((length (+ (length s1) (length s2))))
+ (if exactp
+ (unless (= length min)
+ (sequence-type-length-mismatch-error type length))
+ (unless (>= length min)
+ (sequence-type-length-mismatch-error type length)))
+ (values (merge-lists* s1 s2 pred-fun key-fun))))
+ (sequence-type-too-hairy result-type))))
((csubtypep type (specifier-type 'vector))
(let* ((vector-1 (coerce sequence1 'vector))
- (vector-2 (coerce sequence2 'vector))
- (length-1 (length vector-1))
- (length-2 (length vector-2))
- (result (make-sequence result-type
- (+ length-1 length-2))))
- (declare (vector vector-1 vector-2)
- (fixnum length-1 length-2))
- (if (and (simple-vector-p result)
- (simple-vector-p vector-1)
- (simple-vector-p vector-2))
- (merge-vectors vector-1 length-1 vector-2 length-2
- result predicate key svref)
- (merge-vectors vector-1 length-1 vector-2 length-2
- result predicate key aref))))
+ (vector-2 (coerce sequence2 'vector))
+ (length-1 (length vector-1))
+ (length-2 (length vector-2))
+ (result (make-sequence result-type (+ length-1 length-2))))
+ (declare (vector vector-1 vector-2)
+ (fixnum length-1 length-2))
+ (if (and (simple-vector-p result)
+ (simple-vector-p vector-1)
+ (simple-vector-p vector-2))
+ (merge-vectors vector-1 length-1 vector-2 length-2
+ result predicate key svref)
+ (merge-vectors vector-1 length-1 vector-2 length-2
+ result predicate key aref))))
+ ((and (csubtypep type (specifier-type 'sequence))
+ (find-class result-type nil))
+ (let* ((vector-1 (coerce sequence1 'vector))
+ (vector-2 (coerce sequence2 'vector))
+ (length-1 (length vector-1))
+ (length-2 (length vector-2))
+ (temp (make-array (+ length-1 length-2)))
+ (result (make-sequence result-type (+ length-1 length-2))))
+ (declare (vector vector-1 vector-2) (fixnum length-1 length-2))
+ (merge-vectors vector-1 length-1 vector-2 length-2
+ temp predicate key aref)
+ (replace result temp)
+ result))
(t (bad-sequence-type-error result-type)))))