| blob | 007aa92940406e988f0c7db8454b926ec78e23f2 |
1 ;;;; This file implements local call analysis. A local call is a
2 ;;;; function call between functions being compiled at the same time.
3 ;;;; If we can tell at compile time that such a call is legal, then we
4 ;;;; change the combination to call the correct lambda, mark it as
5 ;;;; local, and add this link to our call graph. Once a call is local,
6 ;;;; it is then eligible for let conversion, which places the body of
7 ;;;; the function inline.
8 ;;;;
9 ;;;; We cannot always do a local call even when we do have the
10 ;;;; function being called. Calls that cannot be shown to have legal
11 ;;;; arg counts are not converted.
13 ;;;; This software is part of the SBCL system. See the README file for
14 ;;;; more information.
15 ;;;;
16 ;;;; This software is derived from the CMU CL system, which was
17 ;;;; written at Carnegie Mellon University and released into the
18 ;;;; public domain. The software is in the public domain and is
19 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
20 ;;;; files for more information.
24 ;;; This function propagates information from the variables in the
25 ;;; function FUN to the actual arguments in CALL. This is also called
26 ;;; by the VALUES IR1 optimizer when it sleazily converts MV-BINDs to
27 ;;; LETs.
28 ;;;
29 ;;; We flush all arguments to CALL that correspond to unreferenced
30 ;;; variables in FUN. We leave NILs in the COMBINATION-ARGS so that
31 ;;; the remaining args still match up with their vars.
32 ;;;
33 ;;; We also apply the declared variable type assertion to the argument
34 ;;; lvars.
54 ;; The block may end up being deleted due to cast optimization
55 ;; caused by USE-GOOD-FOR-DX-P
58 ;; Stack analysis requires that the CALL ends the block, so
59 ;; that MAP-BLOCK-NLXES sees the cleanup we insert here.
64 :mess-up entry
76 ;;; This function handles merging the tail sets if CALL is potentially
77 ;;; tail-recursive, and is a call to a function with a different
78 ;;; TAIL-SET than CALL's FUN. This must be called whenever we alter
79 ;;; IR1 so as to place a local call in what might be a tail-recursive
80 ;;; context. Note that any call which returns its value to a RETURN is
81 ;;; considered potentially tail-recursive, since any implicit MV-PROG1
82 ;;; might be optimized away.
83 ;;;
84 ;;; We destructively modify the set for the calling function to
85 ;;; represent both, and then change all the functions in callee's set
86 ;;; to reference the first. If we do merge, we reoptimize the
87 ;;; RETURN-RESULT lvar to cause IR1-OPTIMIZE-RETURN to recompute the
88 ;;; tail set type.
102 t)))))
104 ;;; Convert a combination into a local call. We PROPAGATE-TO-ARGS, set
105 ;;; the combination kind to :LOCAL, add FUN to the CALLS of the
106 ;;; function that the call is in, call MERGE-TAIL-SETS, then replace
107 ;;; the function in the REF node with the new function.
108 ;;;
109 ;;; We change the REF last, since changing the reference can trigger
110 ;;; LET conversion of the new function, but will only do so if the
111 ;;; call is local. Note that the replacement may trigger LET
112 ;;; conversion or other changes in IR1. We must call MERGE-TAIL-SETS
113 ;;; with NEW-FUN before the substitution, since after the substitution
114 ;;; (and LET conversion), the call may no longer be recognizable as
115 ;;; tail-recursive.
129 \f
130 ;;;; external entry point creation
132 ;;; Return a LAMBDA form that can be used as the definition of the XEP
133 ;;; for FUN.
134 ;;;
135 ;;; If FUN is a LAMBDA, then we check the number of arguments
136 ;;; (conditional on policy) and call FUN with all the arguments.
137 ;;;
138 ;;; If FUN is an OPTIONAL-DISPATCH, then we dispatch off of the number
139 ;;; of supplied arguments by doing do an = test for each entry-point,
140 ;;; calling the entry with the appropriate prefix of the passed
141 ;;; arguments.
142 ;;;
143 ;;; If there is a &MORE arg, then there are a couple of optimizations
144 ;;; that we make (more for space than anything else):
145 ;;; -- If MIN-ARGS is 0, then we make the more entry a T clause, since
146 ;;; no argument count error is possible.
147 ;;; -- We can omit the = clause for the last entry-point, allowing the
148 ;;; case of 0 more args to fall through to the more entry.
149 ;;;
150 ;;; We don't bother to policy conditionalize wrong arg errors in
151 ;;; optional dispatches, since the additional overhead is negligible
152 ;;; compared to the cost of everything else going on.
153 ;;;
154 ;;; Note that if policy indicates it, argument type declarations in
155 ;;; FUN will be verified. Since nothing is known about the type of the
156 ;;; XEP arg vars, type checks will be emitted when the XEP's arg vars
157 ;;; are passed to the actual function.
165 #!+precise-arg-count-error
170 #!-precise-arg-count-error
184 ;; Force convertion of all entries
187 and n from min
190 #!+precise-arg-count-error
196 ;; Arg-checking is performed before this step,
197 ;; arranged by INIT-XEP-ENVIRONMENT,
198 ;; perform the last action unconditionally,
199 ;; and without this the function derived type will be bad.
206 #!-precise-arg-count-error
212 ;; KLUDGE: (NOT (< ...)) instead of >= avoids one round of
213 ;; deftransforms and lambda-conversion.
222 ;;; Make an external entry point (XEP) for FUN and return it. We
223 ;;; convert the result of MAKE-XEP-LAMBDA in the correct environment,
224 ;;; then associate this lambda with FUN as its XEP. After the
225 ;;; conversion, we iterate over the function's associated lambdas,
226 ;;; redoing local call analysis so that the XEP calls will get
227 ;;; converted.
228 ;;;
229 ;;; We set REANALYZE and REOPTIMIZE in the component, just in case we
230 ;;; discover an XEP after the initial local call analyze pass.
246 xep)))
259 ;;; Notice a REF that is not in a local-call context. If the REF is
260 ;;; already to an XEP, then do nothing, otherwise change it to the
261 ;;; XEP, making an XEP if necessary.
262 ;;;
263 ;;; If REF is to a special :CLEANUP or :ESCAPE function, then we treat
264 ;;; it as though it was not an XEP reference (i.e. leave it alone).
272 \f
273 ;;; Attempt to convert all references to FUN to local calls. The
274 ;;; reference must be the function for a call, and the function lvar
275 ;;; must be used only once, since otherwise we cannot be sure what
276 ;;; function is to be called. The call lvar would be multiply used if
277 ;;; there is hairy stuff such as conditionals in the expression that
278 ;;; computes the function.
279 ;;;
280 ;;; If we cannot convert a reference, then we mark the referenced
281 ;;; function as an entry-point, creating a new XEP if necessary. We
282 ;;; don't try to convert calls that are in error (:ERROR kind.)
283 ;;;
284 ;;; This is broken off from LOCALL-ANALYZE-COMPONENT so that people
285 ;;; can force analysis of newly introduced calls. Note that we don't
286 ;;; do LET conversion here.
311 ;;; We examine all NEW-FUNCTIONALS in COMPONENT, attempting to convert
312 ;;; calls into local calls when it is legal. We also attempt to
313 ;;; convert each LAMBDA to a LET. LET conversion is also triggered by
314 ;;; deletion of a function reference, but functions that start out
315 ;;; eligible for conversion must be noticed sometime.
316 ;;;
317 ;;; Note that there is a lot of action going on behind the scenes
318 ;;; here, triggered by reference deletion. In particular, the
319 ;;; COMPONENT-LAMBDAS are being hacked to remove newly deleted and LET
320 ;;; converted LAMBDAs, so it is important that the LAMBDA is added to
321 ;;; the COMPONENT-LAMBDAS when it is. Also, the
322 ;;; COMPONENT-NEW-FUNCTIONALS may contain all sorts of drivel, since
323 ;;; it is not updated when we delete functions, etc. Only
324 ;;; COMPONENT-LAMBDAS is updated.
325 ;;;
326 ;;; COMPONENT-REANALYZE-FUNCTIONALS is treated similarly to
327 ;;; COMPONENT-NEW-FUNCTIONALS, but we don't add lambdas to the
328 ;;; LAMBDAS.
346 ;; Fix/check FUNCTIONAL's relationship to COMPONENT-LAMDBAS.
348 ;; Since FUNCTIONAL isn't a LAMBDA, this doesn't
349 ;; apply: no-op.
352 ; NEW-FUNCTIONALS, hence is new.
353 ;; FUNCTIONAL becomes part of COMPONENT-LAMBDAS now.
358 ;; FUNCTIONAL should be in COMPONENT-LAMBDAS already.
360 component)))))
371 ;; The original CMU CL code seemed to implicitly assume that
372 ;; COMPONENT is the only one here. Let's make that explicit.
383 ;;; If policy is auspicious and CALL is not in an XEP and we don't seem
384 ;;; to be in an infinite recursive loop, then change the reference to
385 ;;; reference a fresh copy. We return whichever function we decide to
386 ;;; reference.
404 original-functional)))))))
409 calls this LET-converted local function:~
410 ~% ~S"
414 expand because expansion refers to ~
415 the optimized away object ~S."
416 losing-local-object)))
423 original-functional)
426 converted-lambda))))
427 original-functional))
429 ;;; Dispatch to the appropriate function to attempt to convert a call.
430 ;;; REF must be a reference to a FUNCTIONAL. This is called in IR1
431 ;;; optimization as well as in local call analysis. If the call is is
432 ;;; already :LOCAL, we do nothing. If the call is already scheduled
433 ;;; for deletion, also do nothing (in addition to saving time, this
434 ;;; also avoids some problems with optimizing collections of functions
435 ;;; that are partially deleted.)
436 ;;;
437 ;;; This is called both before and after FIND-INITIAL-DFO runs. When
438 ;;; called on a :INITIAL component, we don't care whether the caller
439 ;;; and callee are in the same component. Afterward, we must stick
440 ;;; with whatever component division we have chosen.
441 ;;;
442 ;;; Before attempting to convert a call, we see whether the function
443 ;;; is supposed to be inline expanded. Call conversion proceeds as
444 ;;; before after any expansion.
445 ;;;
446 ;;; We bind *COMPILER-ERROR-CONTEXT* to the node for the call so that
447 ;;; warnings will get the right context.
462 component))))
465 original-fun))
483 ;;; Attempt to convert a multiple-value call. The only interesting
484 ;;; case is a call to a function that LOOKS-LIKE-AN-MV-BIND, has
485 ;;; exactly one reference and no XEP, and is called with one values
486 ;;; lvar.
487 ;;;
488 ;;; We change the call to be to the last optional entry point and
489 ;;; change the call to be local. Due to our preconditions, the call
490 ;;; should eventually be converted to a let, but we can't do that now,
491 ;;; since there may be stray references to the e-p lambda due to
492 ;;; optional defaulting code.
493 ;;;
494 ;;; We also use variable types for the called function to construct an
495 ;;; assertion for the values lvar.
496 ;;;
497 ;;; See CONVERT-CALL for additional notes on MERGE-TAIL-SETS, etc.
522 while vars
523 do
525 arg
534 ;;; Convenience function to mark local calls as known bad.
540 default-name))))
548 node
554 ;;; Attempt to convert a call to a lambda. If the number of args is
555 ;;; wrong, we give a warning and mark the call as :ERROR to remove it
556 ;;; from future consideration. If the argcount is O.K. then we just
557 ;;; convert it.
566 'local-argument-mismatch
567 :format-control
568 "function called with ~R argument~:P, but wants exactly ~R"
570 \f
571 ;;;; &OPTIONAL, &MORE and &KEYWORD calls
573 ;;; This is similar to CONVERT-LAMBDA-CALL, but deals with
574 ;;; OPTIONAL-DISPATCHes. If only fixed args are supplied, then convert
575 ;;; a call to the correct entry point. If &KEY args are supplied, then
576 ;;; dispatch to a subfunction. We don't convert calls to functions
577 ;;; that have a &MORE (or &REST) arg.
586 'local-argument-mismatch
587 :format-control
588 "function called with ~R argument~:P, but wants at least ~R"
599 'local-argument-mismatch
600 :format-control
601 "function called with ~R argument~:P, but wants at most ~R"
602 :format-arguments
606 ;;; This function is used to convert a call to an entry point when
607 ;;; complex transformations need to be done on the original arguments.
608 ;;; ENTRY is the entry point function that we are calling. VARS is a
609 ;;; list of variable names which are bound to the original call
610 ;;; arguments. IGNORES is the subset of VARS which are ignored. ARGS
611 ;;; is the list of arguments to the entry point function.
612 ;;;
613 ;;; In order to avoid gruesome graph grovelling, we introduce a new
614 ;;; function that rearranges the arguments and calls the entry point.
615 ;;; We analyze the new function and the entry point immediately so
616 ;;; that everything gets converted during the single pass.
635 ;;; Use CONVERT-HAIRY-FUN-ENTRY to convert a &MORE-arg call to a known
636 ;;; function into a local call to the MAIN-ENTRY.
637 ;;;
638 ;;; First we verify that all keywords are constant and legal. If there
639 ;;; aren't, then we warn the user and don't attempt to convert the call.
640 ;;;
641 ;;; We massage the supplied &KEY arguments into the order expected
642 ;;; by the main entry. This is done by binding all the arguments to
643 ;;; the keyword call to variables in the introduced lambda, then
644 ;;; passing these values variables in the correct order when calling
645 ;;; the main entry. Unused arguments (such as the keywords themselves)
646 ;;; are discarded simply by not passing them along.
647 ;;;
648 ;;; If there is a &REST arg, then we bundle up the args and pass them
649 ;;; to LIST.
671 (:keyword
684 call
730 call
744 (:optional
748 t
750 (:rest
752 ;; &REST arguments may be accompanied by extra
753 ;; context and count arguments. We know this by
754 ;; the ARG-INFO-DEFAULT. Supply 0 and 0 or
755 ;; don't convert at all depending.
762 ;; We've already converted to use the more context
763 ;; instead of the rest list.
769 (:keyword
791 more-temps)))))
794 \f
795 ;;;; LET conversion
796 ;;;;
797 ;;;; Converting to a LET has differing significance to various parts
798 ;;;; of the compiler:
799 ;;;; -- The body of a LET is spliced in immediately after the
800 ;;;; corresponding combination node, making the control transfer
801 ;;;; explicit and allowing LETs to be mashed together into a single
802 ;;;; block. The value of the LET is delivered directly to the
803 ;;;; original lvar for the call, eliminating the need to
804 ;;;; propagate information from the dummy result lvar.
805 ;;;; -- As far as IR1 optimization is concerned, it is interesting in
806 ;;;; that there is only one expression that the variable can be bound
807 ;;;; to, and this is easily substituted for.
808 ;;;; -- LETs are interesting to environment analysis and to the back
809 ;;;; end because in most ways a LET can be considered to be "the
810 ;;;; same function" as its home function.
811 ;;;; -- LET conversion has dynamic scope implications, since control
812 ;;;; transfers within the same environment are local. In a local
813 ;;;; control transfer, cleanup code must be emitted to remove
814 ;;;; dynamic bindings that are no longer in effect.
816 ;;; Set up the control transfer to the called CLAMBDA. We split the
817 ;;; call block immediately after the call, and link the head of
818 ;;; CLAMBDA to the call block. The successor block after splitting
819 ;;; (where we return to) is returned.
820 ;;;
821 ;;; If the lambda is is a different component than the call, then we
822 ;;; call JOIN-COMPONENTS. This only happens in block compilation
823 ;;; before FIND-INITIAL-DFO.
840 next-block)))
842 ;;; Remove CLAMBDA from the tail set of anything it used to be in the
843 ;;; same set as; but leave CLAMBDA with a valid tail set value of
844 ;;; its own, for the benefit of code which might try to pull
845 ;;; something out of it (e.g. return type).
847 ;; Until sbcl-0.pre7.37.flaky5.2, we did
848 ;; (LET ((TAILS (LAMBDA-TAIL-SET CLAMBDA)))
849 ;; (SETF (TAIL-SET-FUNS TAILS)
850 ;; (DELETE CLAMBDA (TAIL-SET-FUNS TAILS))))
851 ;; (SETF (LAMBDA-TAIL-SET CLAMBDA) NIL)
852 ;; here. Apparently the idea behind the (SETF .. NIL) was that since
853 ;; TAIL-SET-FUNS no longer thinks we're in the tail set, it's
854 ;; inconsistent, and perhaps unsafe, for us to think we're in the
855 ;; tail set. Unfortunately..
856 ;;
857 ;; The (SETF .. NIL) caused problems in sbcl-0.pre7.37.flaky5.2 when
858 ;; I was trying to get Python to emit :EXTERNAL LAMBDAs directly
859 ;; (instead of only being able to emit funny little :TOPLEVEL stubs
860 ;; which you called in order to get the address of an external LAMBDA):
861 ;; the external function was defined in terms of internal function,
862 ;; which was LET-converted, and then things blew up downstream when
863 ;; FINALIZE-XEP-DEFINITION tried to find out its DEFINED-TYPE from
864 ;; the now-NILed-out TAIL-SET. So..
865 ;;
866 ;; To deal with this problem, we no longer NIL out
867 ;; (LAMBDA-TAIL-SET CLAMBDA) here. Instead:
868 ;; * If we're the only function in TAIL-SET-FUNS, it should
869 ;; be safe to leave ourself linked to it, and it to you.
870 ;; * If there are other functions in TAIL-SET-FUNS, then we're
871 ;; afraid of future optimizations on those functions causing
872 ;; the TAIL-SET object no longer to be valid to describe our
873 ;; return value. Thus, we delete ourselves from that object;
874 ;; but we save a newly-allocated tail-set, derived from the old
875 ;; one, for ourselves, for the use of later code (e.g.
876 ;; FINALIZE-XEP-DEFINITION) which might want to
877 ;; know about our return type.
886 ;; The documentation on TAIL-SET-INFO doesn't tell whether it could
887 ;; remain valid in this case, so we nuke it on the theory that
888 ;; missing information tends to be less dangerous than incorrect
889 ;; information.
892 ;;; Handle the PHYSENV semantics of LET conversion. We add CLAMBDA and
893 ;;; its LETs to LETs for the CALL's home function. We merge the calls
894 ;;; for CLAMBDA with the calls for the home function, removing CLAMBDA
895 ;;; in the process. We also merge the ENTRIES.
896 ;;;
897 ;;; We also unlink the function head from the component head and set
898 ;;; COMPONENT-REANALYZE to true to indicate that the DFO should be
899 ;;; recomputed.
919 ;; CLAMBDA belongs to HOME now.
931 ;; All of CLAMBDA's LETs belong to HOME now.
937 ;; CLAMBDA no longer has an independent existence as an entity
938 ;; which has LETs.
941 ;; HOME no longer calls CLAMBDA, and owns all of CLAMBDA's old
942 ;; DFO dependencies.
946 ;; CLAMBDA no longer has an independent existence as an entity
947 ;; which calls things or has DFO dependencies.
950 ;; All of CLAMBDA's ENTRIES belong to HOME now.
954 ;; CLAMBDA no longer has an independent existence as an entity
955 ;; with ENTRIES.
960 ;;; Handle the value semantics of LET conversion. Delete FUN's return
961 ;;; node, and change the control flow to transfer to NEXT-BLOCK
962 ;;; instead. Move all the uses of the result lvar to CALL's lvar.
981 ;; FIXME: Replace the call with unsafe CAST. -- APD, 2003-01-26
988 ;;; We are converting FUN to be a LET when the call is in a non-tail
989 ;;; position. Any previously tail calls in FUN are no longer tail
990 ;;; calls, and must be restored to normal calls which transfer to
991 ;;; NEXT-BLOCK (FUN's return point.) We can't do this by DO-USES on
992 ;;; the RETURN-RESULT, because the return might have been deleted (if
993 ;;; all calls were TR.)
1012 ;; The called function might be an assignment in the
1013 ;; case where we are currently converting that function.
1014 ;; In steady-state, assignments never appear as a called
1015 ;; function.
1016 (:assignment
1020 ;;; Deal with returning from a LET or assignment that we are
1021 ;;; converting. FUN is the function we are calling, CALL is a call to
1022 ;;; FUN, and NEXT-BLOCK is the return point for a non-tail call, or
1023 ;;; NULL if call is a tail call.
1024 ;;;
1025 ;;; If the call is not a tail call, then we must do
1026 ;;; UNCONVERT-TAIL-CALLS, since a tail call is a call which returns
1027 ;;; its value out of the enclosing non-let function. When call is
1028 ;;; non-TR, we must convert it back to an ordinary local call, since
1029 ;;; the value must be delivered to the receiver of CALL's value.
1030 ;;;
1031 ;;; We do different things depending on whether the caller and callee
1032 ;;; have returns left:
1034 ;;; -- If the callee has no return we just do MOVE-LET-CALL-CONT.
1035 ;;; Either the function doesn't return, or all returns are via
1036 ;;; tail-recursive local calls.
1037 ;;; -- If CALL is a non-tail call, or if both have returns, then
1038 ;;; we delete the callee's return, move its uses to the call's
1039 ;;; result lvar, and transfer control to the appropriate
1040 ;;; return point.
1041 ;;; -- If the callee has a return, but the caller doesn't, then we
1042 ;;; move the return to the caller.
1071 ;;; Actually do LET conversion. We call subfunctions to do most of the
1072 ;;; work. We do REOPTIMIZE-LVAR on the args and CALL's lvar so that
1073 ;;; LET-specific IR1 optimizations get a chance. We blow away any
1074 ;;; entry for the function in *FREE-FUNS* so that nobody will create
1075 ;;; new references to it.
1080 nil
1081 next-block)))
1085 ;; If CALL has a derive type NIL, it means that "its return" is
1086 ;; unreachable, but the next BIND is still reachable; in order to
1087 ;; not confuse MAYBE-TERMINATE-BLOCK...
1090 ;;; Reoptimize all of CALL's args and its result.
1099 ;;; Are there any declarations in force to say CLAMBDA shouldn't be
1100 ;;; LET converted?
1102 ;; From the user's point of view, LET-converting something that
1103 ;; has a name is inlining it. (The user can't see what we're doing
1104 ;; with anonymous things, and suppressing inlining
1105 ;; for such things can easily give Python acute indigestion, so
1106 ;; we don't.)
1107 ;;
1108 ;; A functional that is already inline-expanded in this componsne definitely
1109 ;; deserves let-conversion -- and in case of main entry points for inline
1110 ;; expanded optional dispatch, the main-etry isn't explicitly marked :INLINE
1111 ;; even if the function really is.
1114 ;; ANSI requires that explicit NOTINLINE be respected.
1116 ;; If (= LET-CONVERSION 0) we can guess that inlining
1117 ;; generally won't be appreciated, but if the user
1118 ;; specifically requests inlining, that takes precedence over
1119 ;; our general guess.
1123 ;;; We also don't convert calls to named functions which appear in the
1124 ;;; initial component, delaying this until optimization. This
1125 ;;; minimizes the likelihood that we will LET-convert a function which
1126 ;;; may have references added due to later local inline expansion.
1133 ;;; ir1opt usually takes care of forwarding let-bound values directly
1134 ;;; to their destination when possible. However, locall analysis
1135 ;;; greatly benefits from that transformation, and is executed in a
1136 ;;; distinct phase from ir1opt. After let-conversion, variables
1137 ;;; bound to functional values are immediately substituted away.
1138 ;;;
1139 ;;; When called from locall, component is non-nil, and the functionals
1140 ;;; are marked for reanalysis when appropriate.
1146 ;; only do that in the absence of assignment
1148 do
1153 ;; unlike propagate-let-args, we're only concerned with
1154 ;; functionals.
1156 ;; if the types match, we can mutate refs to point to
1157 ;; the functional instead of var
1167 nil)))
1168 leaf var)))
1169 ;; otherwise, we can still play LVAR-level tricks for single
1170 ;; destination variables.
1172 ;; Don't substitute single-ref variables on high-debug /
1173 ;; low speed, to improve the debugging experience.
1177 ;; if we've done something, the functional may now be used in
1178 ;; more analysis-friendly manners. Enqueue it if we're in
1179 ;; locall.
1181 component
1186 ;;; This function is called when there is some reason to believe that
1187 ;;; CLAMBDA might be converted into a LET. This is done after local
1188 ;;; call analysis, and also when a reference is deleted. We return
1189 ;;; true if we converted.
1190 ;;;
1191 ;;; COMPONENT is non-nil during local call analysis. It is used to
1192 ;;; re-enqueue functionals for reanalysis when they have been forwarded
1193 ;;; directly to destination nodes.
1199 ;; We only convert to a LET when the function is a normal local
1200 ;; function, has no XEP, and is referenced in exactly one local
1201 ;; call. Conversion is also inhibited if the only reference is in
1202 ;; a block about to be deleted.
1203 ;;
1204 ;; These rules limiting LET conversion may seem unnecessarily
1205 ;; restrictive, since there are some cases where we could do the
1206 ;; return with a jump that don't satisfy these requirements. The
1207 ;; reason for doing things this way is that it makes the concept
1208 ;; of a LET much more useful at the level of IR1 semantics. The
1209 ;; :ASSIGNMENT function kind provides another way to optimize
1210 ;; calls to single-return/multiple call functions.
1211 ;;
1212 ;; We don't attempt to convert calls to functions that have an
1213 ;; XEP, since we might be embarrassed later when we want to
1214 ;; convert a newly discovered local call. Also, see
1215 ;; OK-INITIAL-CONVERT-P.
1232 nil)))
1242 ; for me to handle - PK 2012-05-30
1244 t))))
1245 \f
1246 ;;;; tail local calls and assignments
1248 ;;; Return T if there are no cleanups between BLOCK1 and BLOCK2, or if
1249 ;;; they definitely won't generate any cleanup code. Currently we
1250 ;;; recognize lexical entry points that are only used locally (if at
1251 ;;; all).
1265 ;;; If a potentially TR local call really is TR, then convert it to
1266 ;;; jump directly to the called function. We also call
1267 ;;; MAYBE-CONVERT-TO-ASSIGNMENT. The first value is true if we
1268 ;;; tail-convert. The second is the value of M-C-T-A.
1275 ;; this is a tail call
1279 ;; If the call is in an XEP, we might decide to make it
1280 ;; non-tail so that we can use known return inside the
1281 ;; component.
1293 ;;; This is called when we believe it might make sense to convert
1294 ;;; CLAMBDA to an assignment. All this function really does is
1295 ;;; determine when a function with more than one call can still be
1296 ;;; combined with the calling function's environment. We can convert
1297 ;;; when:
1298 ;;; -- The function is a normal, non-entry function, and
1299 ;;; -- Except for one call, all calls must be tail recursive calls
1300 ;;; in the called function (i.e. are self-recursive tail calls)
1301 ;;; -- OK-INITIAL-CONVERT-P is true.
1302 ;;;
1303 ;;; There may be one outside call, and it need not be tail-recursive.
1304 ;;; Since all tail local calls have already been converted to direct
1305 ;;; transfers, the only control semantics needed are to splice in the
1306 ;;; body at the non-tail call. If there is no non-tail call, then we
1307 ;;; need only merge the environments. Both cases are handled by
1308 ;;; LET-CONVERT.
1309 ;;;
1310 ;;; ### It would actually be possible to allow any number of outside
1311 ;;; calls as long as they all return to the same place (i.e. have the
1312 ;;; same conceptual continuation.) A special case of this would be
1313 ;;; when all of the outside calls are tail recursive.
1340 t)
1342 nil))))))