;;;; provided with absolutely no warranty. See the COPYING and CREDITS
;;;; files for more information.
-(in-package "SB!C")
+(in-package "SB!REGALLOC")
;;; for debugging: some parameters controlling which optimizations we
;;; attempt
(return t)))
(setq block (optimized-emit-saves-block block saves restores)))
(setq block (ir2-block-prev block)))))
-
+\f
;;; Iterate over the normal TNs, finding the cost of packing on the
-;;; stack in units of the number of references. We count all
-;;; references as +1, and subtract out REGISTER-SAVE-PENALTY for each
-;;; place where we would have to save a register.
+;;; stack in units of the number of references. We count all read
+;;; references as +1, write references as + *tn-write-cost*, and
+;;; subtract out REGISTER-SAVE-PENALTY for each place where we would
+;;; have to save a register.
+;;; The subtraction reflects the fact that having a value in a
+;;; register around a call means that code to spill and unspill must
+;;; be inserted.
+;;;
+;;; The costs also take into account the loop depth at which each
+;;; reference occurs: the penalty or cost is incremented by the depth
+;;; scaled by *tn-loop-depth-multiplier*. The default (NIL) is to let
+;;; this be one more than the max of the cost for reads (1), for write
+;;; references and for being live across a call.
+(defvar *tn-write-cost* 2)
+(defvar *tn-loop-depth-multiplier* nil)
+
(defun assign-tn-costs (component)
- (do-ir2-blocks (block component)
- (do ((vop (ir2-block-start-vop block) (vop-next vop)))
- ((null vop))
- (when (eq (vop-info-save-p (vop-info vop)) t)
- (do-live-tns (tn (vop-save-set vop) block)
- (decf (tn-cost tn) *backend-register-save-penalty*)))))
-
- (do ((tn (ir2-component-normal-tns (component-info component))
- (tn-next tn)))
- ((null tn))
- (let ((cost (tn-cost tn)))
- (declare (fixnum cost))
- (do ((ref (tn-reads tn) (tn-ref-next ref)))
- ((null ref))
- (incf cost))
- (do ((ref (tn-writes tn) (tn-ref-next ref)))
- ((null ref))
- (incf cost))
- (setf (tn-cost tn) cost))))
-
-;;; Iterate over the normal TNs, storing the depth of the deepest loop
-;;; that the TN is used in TN-LOOP-DEPTH.
-(defun assign-tn-depths (component)
+ (let* ((save-penalty *backend-register-save-penalty*)
+ (write-cost *tn-write-cost*)
+ (depth-scale (or *tn-loop-depth-multiplier*
+ (1+ (max 1 write-cost save-penalty)))))
+ (flet ((vop-depth-cost (vop)
+ (let ((loop (block-loop
+ (ir2-block-block
+ (vop-block vop)))))
+ (if loop
+ (* depth-scale (loop-depth loop))
+ 0))))
+ (do-ir2-blocks (block component)
+ (do ((vop (ir2-block-start-vop block) (vop-next vop)))
+ ((null vop))
+ (when (eq (vop-info-save-p (vop-info vop)) t)
+ (let ((penalty (+ save-penalty (vop-depth-cost vop))))
+ (do-live-tns (tn (vop-save-set vop) block)
+ (decf (tn-cost tn) penalty))))))
+
+ (do ((tn (ir2-component-normal-tns (component-info component))
+ (tn-next tn)))
+ ((null tn))
+ (let ((cost (tn-cost tn)))
+ (declare (fixnum cost))
+ (do ((ref (tn-reads tn) (tn-ref-next ref)))
+ ((null ref))
+ (incf cost (1+ (vop-depth-cost (tn-ref-vop ref)))))
+ (do ((ref (tn-writes tn) (tn-ref-next ref)))
+ ((null ref))
+ (incf cost (+ write-cost (vop-depth-cost (tn-ref-vop ref)))))
+ (setf (tn-cost tn) cost))))))
+
+;;; Iterate over the normal TNs, folding over the depth of the looops
+;;; that the TN is used in and storing the result in TN-LOOP-DEPTH.
+;;: reducer is the function used to join depth values together. #'max
+;;; gives the maximum depth, #'+ the sum.
+(defun assign-tn-depths (component &key (reducer #'max))
+ (declare (type function reducer))
(when *loop-analyze*
- (do-ir2-blocks (block component)
- (do ((vop (ir2-block-start-vop block)
- (vop-next vop)))
- ((null vop))
- (flet ((find-all-tns (head-fun)
- (collect ((tns))
- (do ((ref (funcall head-fun vop) (tn-ref-across ref)))
- ((null ref))
- (tns (tn-ref-tn ref)))
- (tns))))
- (dolist (tn (nconc (find-all-tns #'vop-args)
- (find-all-tns #'vop-results)
- (find-all-tns #'vop-temps)
- ;; What does "references in this VOP
- ;; mean"? Probably something that isn't
- ;; useful in this context, since these
- ;; TN-REFs are linked with TN-REF-NEXT
- ;; instead of TN-REF-ACROSS. --JES
- ;; 2004-09-11
- ;; (find-all-tns #'vop-refs)
- ))
- (setf (tn-loop-depth tn)
- (max (tn-loop-depth tn)
- (let* ((ir1-block (ir2-block-block (vop-block vop)))
- (loop (block-loop ir1-block)))
- (if loop
- (loop-depth loop)
- 0))))))))))
-
+ ;; We only use tn depth for normal TNs
+ (do ((tn (ir2-component-normal-tns (component-info component))
+ (tn-next tn)))
+ ((null tn))
+ (let ((depth 0))
+ (declare (type fixnum depth))
+ (flet ((frob (ref)
+ (declare (type (or null tn-ref) ref))
+ (do ((ref ref (tn-ref-next ref)))
+ ((null ref))
+ (let* ((vop (tn-ref-vop ref))
+ (block (ir2-block-block (vop-block vop)))
+ (loop (block-loop block)))
+ (setf depth (funcall reducer
+ depth
+ (if loop
+ (loop-depth loop)
+ 0)))))))
+ (frob (tn-reads tn))
+ (frob (tn-writes tn))
+ (setf (tn-loop-depth tn) depth))))))
\f
;;;; load TN packing
nil)))
;;; Scan along the target path from TN, looking at readers or writers.
-;;; When we find a packed TN, return CHECK-OK-TARGET of that TN. If
-;;; there is no target, or if the TN has multiple readers (writers),
-;;; then we return NIL. We also always return NIL after 10 iterations
-;;; to get around potential circularity problems.
+;;; When we find a TN, call CALLEE with that TN, and then resume
+;;; walking down that TN's target. As soon as there is no target, or
+;;; if the TN has multiple readers (writers), we stop walking the
+;;; targetting chain. We also always stop after 10 iterations to get
+;;; around potential circularity problems.
;;;
-;;; FIXME: (30 minutes of reverse engineering?) It'd be nice to
-;;; rewrite the header comment here to explain the interface and its
-;;; motivation, and move remarks about implementation details (like
-;;; 10!) inside.
-(defun find-ok-target-offset (tn sc)
- (declare (type tn tn) (type sc sc))
- (flet ((frob-slot (slot-fun)
- (declare (type function slot-fun))
- (let ((count 10)
+;;; Why the single-reader/writer constraint? As far as I can tell,
+;;; this is concerned with straight pipeline of data, e.g. CASTs. In
+;;; that case, limiting to chains of length 10 seems to be more than
+;;; enough.
+(declaim (inline %call-with-target-tns))
+(defun %call-with-target-tns (tn callee
+ &key (limit 10) (reads t) (writes t))
+ (declare (type tn tn) (type function callee) (type index limit))
+ (flet ((frob-slot (slot-function)
+ (declare (type function slot-function))
+ (let ((count limit)
(current tn))
(declare (type index count))
(loop
- (let ((refs (funcall slot-fun current)))
+ (let ((refs (funcall slot-function current)))
(unless (and (plusp count)
refs
(not (tn-ref-next refs)))
(let ((target (tn-ref-target refs)))
(unless target (return nil))
(setq current (tn-ref-tn target))
- (when (tn-offset current)
- (return (check-ok-target current tn sc)))
+ (funcall callee current)
(decf count)))))))
- (declare (inline frob-slot)) ; until DYNAMIC-EXTENT works
- (or (frob-slot #'tn-reads)
- (frob-slot #'tn-writes))))
+ (when reads
+ (frob-slot #'tn-reads))
+ (when writes
+ (frob-slot #'tn-writes))
+ nil))
+
+(defmacro do-target-tns ((target-variable source-tn
+ &rest keys &key limit reads writes)
+ &body body)
+ (declare (ignore limit reads writes))
+ (let ((callback (gensym "CALLBACK")))
+ `(flet ((,callback (,target-variable)
+ ,@body))
+ (declare (dynamic-extent #',callback))
+ (%call-with-target-tns ,source-tn #',callback ,@keys))))
+
+(defun find-ok-target-offset (tn sc)
+ (declare (type tn tn) (type sc sc))
+ (do-target-tns (target tn)
+ (awhen (and (tn-offset target)
+ (check-ok-target target tn sc))
+ (return-from find-ok-target-offset it))))
\f
;;;; location selection
(let ((locations (sc-locations sc)))
(when optimize
(setf locations
- (stable-sort (copy-list locations) #'>
- :key (lambda (location-offset)
- (loop for offset from location-offset
- repeat element-size
- maximize (svref
- (finite-sb-always-live-count sb)
- offset))))))
+ (schwartzian-stable-sort-list
+ locations '>
+ :key (lambda (location-offset)
+ (loop for offset from location-offset
+ repeat element-size
+ maximize (svref
+ (finite-sb-always-live-count sb)
+ offset))))))
(dolist (offset locations)
(when (or use-reserved-locs
(not (member offset
\f
;;;; pack interface
+;; Misc. utilities
+(declaim (inline unbounded-sc-p))
+(defun unbounded-sc-p (sc)
+ (eq (sb-kind (sc-sb sc)) :unbounded))
+
+(defun unbounded-tn-p (tn)
+ (unbounded-sc-p (tn-sc tn)))
+(declaim (notinline unbounded-sc-p))
+
;;; Attempt to pack TN in all possible SCs, first in the SC chosen by
;;; representation selection, then in the alternate SCs in the order
;;; they were specified in the SC definition. If the TN-COST is
;;; of allocating a new stack location.
(defun pack-tn (tn restricted optimize &key (allow-unbounded-sc t))
(declare (type tn tn))
+ (aver (not (tn-offset tn)))
(let* ((original (original-tn tn))
(fsc (tn-sc tn))
(alternates (unless restricted (sc-alternate-scs fsc)))
((null sc)
(failed-to-pack-error tn restricted))
(unless (or allow-unbounded-sc
- (neq (sb-kind (sc-sb sc)) :unbounded))
+ (not (unbounded-sc-p sc)))
(return nil))
(when (eq sc specified-save-sc)
(unless (tn-offset save)
(when (or restricted
(not (and (minusp (tn-cost tn)) (sc-save-p sc))))
(let ((loc (or (find-ok-target-offset original sc)
- (select-location original sc)
+ (select-location original sc :optimize optimize)
(and restricted
- (select-location original sc :use-reserved-locs t))
- (when (eq (sb-kind (sc-sb sc)) :unbounded)
+ (select-location original sc :use-reserved-locs t
+ :optimize optimize))
+ (when (unbounded-sc-p sc)
(grow-sc sc)
(or (select-location original sc)
(error "failed to pack after growing SC?"))))))
(or (= offset 0)
(= offset 1))))
(conflicts-in-sc original sc offset))
- (error "~S is wired to a location that it conflicts with." tn))
+ (error "~S is wired to location ~D in SC ~A of kind ~S that it conflicts with."
+ tn offset sc (tn-kind tn)))
(add-location-conflicts original sc offset optimize)))
(defun tn-lexical-depth (tn)
(let ((path t)) ; dummy initial value
(labels ((path (lambda)
- (nreverse (loop while lambda
- collect lambda
- do (setf lambda (lambda-parent lambda)))))
+ (do ((acc '())
+ (lambda lambda (lambda-parent lambda)))
+ ((null lambda) acc)
+ (push lambda acc)))
(register-scope (lambda)
(let ((new-path (path lambda)))
(setf path (if (eql path t)
0 (mismatch path new-path))))))
(walk-tn-refs (ref)
(do ((ref ref (tn-ref-next ref)))
- ((null ref))
- (binding* ((node (vop-node (tn-ref-vop ref))
- :exit-if-null))
- (register-scope (lexenv-lambda
- (node-lexenv node)))))))
+ ((or (null ref)
+ (null path)))
+ (awhen (vop-node (tn-ref-vop ref))
+ (register-scope (lexenv-lambda (node-lexenv it)))))))
(walk-tn-refs (tn-reads tn))
(walk-tn-refs (tn-writes tn))
(if (eql path t)
(when target-fun
(funcall target-fun vop)))))
- ;; Pack wired TNs first.
- (do ((tn (ir2-component-wired-tns 2comp) (tn-next tn)))
- ((null tn))
- (pack-wired-tn tn optimize))
-
- ;; Pack restricted component TNs.
- (do ((tn (ir2-component-restricted-tns 2comp) (tn-next tn)))
- ((null tn))
- (when (eq (tn-kind tn) :component)
- (pack-tn tn t optimize)))
-
- ;; Pack other restricted TNs.
- (do ((tn (ir2-component-restricted-tns 2comp) (tn-next tn)))
- ((null tn))
- (unless (tn-offset tn)
- (pack-tn tn t optimize)))
-
- ;; Assign costs to normal TNs so we know which ones should
- ;; always be packed on the stack.
+ ;; Assign costs to normal TNs so we know which ones should always
+ ;; be packed on the stack, and which are important not to spill.
(when *pack-assign-costs*
- (assign-tn-costs component)
- (assign-tn-depths component))
-
- ;; Allocate normal TNs, starting with the TNs that are used
- ;; in deep loops. Only allocate in finite SCs (i.e. not on
- ;; the stack).
- (collect ((tns))
- (do-ir2-blocks (block component)
- (let ((ltns (ir2-block-local-tns block)))
- (do ((i (1- (ir2-block-local-tn-count block)) (1- i)))
- ((minusp i))
- (declare (fixnum i))
- (let ((tn (svref ltns i)))
- (unless (or (null tn)
- (eq tn :more)
- (tn-offset tn))
- ;; If loop analysis has been disabled we might as
- ;; well revert to the old behaviour of just
- ;; packing TNs linearly as they appear.
- (unless *loop-analyze*
- (pack-tn tn nil optimize :allow-unbounded-sc nil))
- (tns tn))))))
- (dolist (tn (stable-sort (tns)
- (lambda (a b)
- (cond
- ((> (tn-loop-depth a)
- (tn-loop-depth b))
- t)
- ((= (tn-loop-depth a)
- (tn-loop-depth b))
- (> (tn-cost a) (tn-cost b)))
- (t nil)))))
- (unless (tn-offset tn)
- (pack-tn tn nil optimize :allow-unbounded-sc nil))))
-
- ;; Pack any leftover normal TNs that could not be allocated
- ;; to finite SCs, or TNs that do not appear in any local TN
- ;; map (e.g. :MORE TNs). Since we'll likely be allocating
- ;; on the stack, first allocate TNs that are associated with
- ;; code at shallow lexical depths: this will allocate long
- ;; live ranges (i.e. TNs with more conflicts) first, and
- ;; hopefully minimise stack fragmentation.
- ;;
- ;; Collect in reverse order to give priority to older TNs.
- (let ((contiguous-tns '())
+ (assign-tn-costs component))
+
+ ;; Actually allocate registers for most TNs. After this, only
+ ;; :normal tns may be left unallocated (or TNs :restricted to
+ ;; an unbounded SC).
+ (pack-greedy component 2comp optimize)
+
+ ;; Pack any leftover normal/restricted TN that is not already
+ ;; allocated to a finite SC, or TNs that do not appear in any
+ ;; local TN map (e.g. :MORE TNs). Since we'll likely be
+ ;; allocating on the stack, first allocate TNs that are
+ ;; associated with code at shallow lexical depths: this will
+ ;; allocate long live ranges (i.e. TNs with more conflicts)
+ ;; first, and hopefully minimise stack fragmentation.
+ ;; Component TNs are a degenerate case: they are always live.
+ (let ((component-tns '())
+ (contiguous-tns '())
(tns '()))
- (do ((tn (ir2-component-normal-tns 2comp) (tn-next tn)))
- ((null tn))
- (unless (tn-offset tn)
- (let ((key (cons tn (tn-lexical-depth tn))))
- (if (memq (tn-kind tn) '(:environment :debug-environment
- :component))
- (push key contiguous-tns)
- (push key tns)))))
- (flet ((pack-tns (tns)
- (dolist (tn (stable-sort tns #'< :key #'cdr))
- (let ((tn (car tn)))
- (unless (tn-offset tn)
- (pack-tn tn nil optimize))))))
- ;; first pack TNs that are known to have simple
- ;; live ranges (contiguous lexical scopes)
+ (flet ((register-tn (tn)
+ (unless (tn-offset tn)
+ (case (tn-kind tn)
+ (:component
+ (push tn component-tns))
+ ((:environment :debug-environment)
+ (push tn contiguous-tns))
+ (t
+ (push tn tns))))))
+ (do ((tn (ir2-component-restricted-tns 2comp) (tn-next tn)))
+ ((null tn))
+ ;; by this time, restricted TNs must either be
+ ;; allocated in the right SC or unbounded
+ (aver (or (tn-offset tn) (unbounded-tn-p tn)))
+ (register-tn tn))
+ (do ((tn (ir2-component-normal-tns 2comp) (tn-next tn)))
+ ((null tn))
+ (register-tn tn)))
+ (flet ((pack-tns (tns &optional in-order)
+ (dolist (tn (if in-order
+ tns
+ (schwartzian-stable-sort-list
+ tns #'< :key #'tn-lexical-depth)))
+ (unless (tn-offset tn)
+ (pack-tn tn nil optimize)))))
+ ;; first pack TNs that are known to have simple live
+ ;; ranges (contiguous lexical scopes)
+ (pack-tns component-tns t)
(pack-tns contiguous-tns)
(pack-tns tns)))
(values))
(clean-up-pack-structures)))
+
+(defun pack-greedy (component 2comp optimize)
+ (declare (type component component)
+ (type ir2-component 2comp))
+ ;; Pack wired TNs first.
+ (do ((tn (ir2-component-wired-tns 2comp) (tn-next tn)))
+ ((null tn))
+ (pack-wired-tn tn optimize))
+
+ ;; Then, pack restricted TNs, ones that are live over the whole
+ ;; component first (they cause no fragmentation). Sort by TN cost
+ ;; to help important TNs get good targeting.
+ (collect ((component)
+ (normal))
+ (do ((tn (ir2-component-restricted-tns 2comp) (tn-next tn)))
+ ((null tn))
+ (unless (or (tn-offset tn) (unbounded-tn-p tn))
+ (if (eq :component (tn-kind tn))
+ (component tn)
+ (normal tn))))
+ (flet ((pack-tns (tns)
+ (dolist (tn (stable-sort tns #'> :key #'tn-cost))
+ (pack-tn tn t optimize))))
+ (pack-tns (component))
+ (pack-tns (normal))))
+
+ (cond ((and *loop-analyze* *pack-assign-costs*)
+ ;; Allocate normal TNs, starting with the TNs that are
+ ;; heavily used in deep loops (which is taken into account in
+ ;; TN spill costs). Only allocate in finite SCs (i.e. not on
+ ;; the stack).
+ (collect ((tns))
+ (do ((tn (ir2-component-normal-tns 2comp) (tn-next tn)))
+ ((null tn))
+ (unless (or (tn-offset tn)
+ (eq (tn-kind tn) :more)
+ (unbounded-tn-p tn)
+ (and (sc-save-p (tn-sc tn)) ; SC caller-save, but TN
+ (minusp (tn-cost tn)))) ; lives over many calls
+ (tns tn)))
+ (dolist (tn (stable-sort (tns) #'> :key #'tn-cost))
+ (unless (tn-offset tn)
+ ;; if it can't fit in a bounded SC, the final pass will
+ ;; take care of stack packing.
+ (pack-tn tn nil optimize :allow-unbounded-sc nil)))))
+ (t
+ ;; If loop analysis has been disabled we might as well revert
+ ;; to the old behaviour of just packing TNs linearly as they
+ ;; appear.
+ (do-ir2-blocks (block component)
+ (let ((ltns (ir2-block-local-tns block)))
+ (do ((i (1- (ir2-block-local-tn-count block)) (1- i)))
+ ((minusp i))
+ (declare (fixnum i))
+ (let ((tn (svref ltns i)))
+ (unless (or (null tn)
+ (eq tn :more)
+ (tn-offset tn)
+ (unbounded-tn-p tn))
+ (pack-tn tn nil optimize :allow-unbounded-sc nil)))))))))