| blob | 6fcb1e7a1e393929799397e71da9bcec604c739c |
1 ;;;; This file implements the IR1 optimization phase of the compiler.
2 ;;;; IR1 optimization is a grab-bag of optimizations that don't make
3 ;;;; major changes to the block-level control flow and don't use flow
4 ;;;; analysis. These optimizations can mostly be classified as
5 ;;;; "meta-evaluation", but there is a sizable top-down component as
6 ;;;; well.
8 ;;;; This software is part of the SBCL system. See the README file for
9 ;;;; more information.
10 ;;;;
11 ;;;; This software is derived from the CMU CL system, which was
12 ;;;; written at Carnegie Mellon University and released into the
13 ;;;; public domain. The software is in the public domain and is
14 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
15 ;;;; files for more information.
18 \f
19 ;;;; interface for obtaining results of constant folding
21 ;;; Return true for a CONTINUATION whose sole use is a reference to a
22 ;;; constant leaf.
29 ;;; Return the constant value for a continuation whose only use is a
30 ;;; constant node.
35 \f
36 ;;;; interface for obtaining results of type inference
38 ;;; Return a (possibly values) type that describes what we have proven
39 ;;; about the type of Cont without taking any type assertions into
40 ;;; consideration. This is just the union of the NODE-DERIVED-TYPE of
41 ;;; all the uses. Most often people use CONTINUATION-DERIVED-TYPE or
42 ;;; CONTINUATION-TYPE instead of using this function directly.
51 res))
55 (:inside-block
58 ;;; Our best guess for the type of this continuation's value. Note
59 ;;; that this may be VALUES or FUNCTION type, which cannot be passed
60 ;;; as an argument to the normal type operations. See
61 ;;; CONTINUATION-TYPE. This may be called on deleted continuations,
62 ;;; always returning *.
63 ;;;
64 ;;; What we do is call CONTINUATION-PROVEN-TYPE and check whether the
65 ;;; result is a subtype of the assertion. If so, return the proven
66 ;;; type and set TYPE-CHECK to nil. Otherwise, return the intersection
67 ;;; of the asserted and proven types, and set TYPE-CHECK T. If
68 ;;; TYPE-CHECK already has a non-null value, then preserve it. Only in
69 ;;; the somewhat unusual circumstance of a newly discovered assertion
70 ;;; will we change TYPE-CHECK from NIL to T.
71 ;;;
72 ;;; The result value is cached in the CONTINUATION-%DERIVED-TYPE slot.
73 ;;; If the slot is true, just return that value, otherwise recompute
74 ;;; and stash the value there.
89 ;; It's physically impossible for a runtime type check to
90 ;; distinguish between the various subtypes of FUNCTION, so
91 ;; it'd be pointless to do more type checks here.
94 ;; FIXME: This should depend on optimization
95 ;; policy. This is for SPEED > SAFETY:
97 ;; and this is for SAFETY >= SPEED:
108 ;;; Call CONTINUATION-DERIVED-TYPE to make sure the slot is up to
109 ;;; date, then return it.
116 ;;; Return the derived type for CONT's first value. This is guaranteed
117 ;;; not to be a VALUES or FUNCTION type.
122 ;;; If CONT is an argument of a function, return a type which the
123 ;;; function checks CONT for.
132 ;; TODO: MV-COMBINATION
138 ;; FUN-TYPE might be (AND FUNCTION (SATISFIES ...)).
149 ;; TODO: &KEY
151 for arg of-type continuation in args
152 and type of-type ctype in arg-types
154 type))
156 \f
157 ;;;; interface routines used by optimizers
159 ;;; This function is called by optimizers to indicate that something
160 ;;; interesting has happened to the value of Cont. Optimizers must
161 ;;; make sure that they don't call for reoptimization when nothing has
162 ;;; happened, since optimization will fail to terminate.
163 ;;;
164 ;;; We clear any cached type for the continuation and set the
165 ;;; reoptimize flags on everything in sight, unless the continuation
166 ;;; is deleted (in which case we do nothing.)
167 ;;;
168 ;;; Since this can get called during IR1 conversion, we have to be
169 ;;; careful not to fly into space when the Dest's Prev is missing.
190 ;;; Annotate Node to indicate that its result has been proven to be
191 ;;; typep to RType. After IR1 conversion has happened, this is the
192 ;;; only correct way to supply information discovered about a node's
193 ;;; type. If you screw with the Node-Derived-Type directly, then
194 ;;; information may be lost and reoptimization may not happen.
195 ;;;
196 ;;; What we do is intersect Rtype with Node's Derived-Type. If the
197 ;;; intersection is different from the old type, then we do a
198 ;;; Reoptimize-Continuation on the Node-Cont.
210 "New inferred type ~S conflicts with old type:~
211 ~% ~S~%*** possible internal error? Please report this."
217 ;;; This is similar to DERIVE-NODE-TYPE, but asserts that it is an
218 ;;; error for CONT's value not to be TYPEP to TYPE. If we improve the
219 ;;; assertion, we set TYPE-CHECK and TYPE-ASSERTED to guarantee that
220 ;;; the new assertion will be checked.
230 type-check type-asserted)
231 t))
235 ;;; Assert that CALL is to a function of the specified TYPE. It is
236 ;;; assumed that the call is legal and has only constants in the
237 ;;; keyword positions.
264 \f
265 ;;;; IR1-OPTIMIZE
267 ;;; Do one forward pass over COMPONENT, deleting unreachable blocks
268 ;;; and doing IR1 optimizations. We can ignore all blocks that don't
269 ;;; have the REOPTIMIZE flag set. If COMPONENT-REOPTIMIZE is true when
270 ;;; we are done, then another iteration would be beneficial.
280 ;; Preserve the BLOCK-SUCC invariant that almost every block has
281 ;; one successor (and a block with DELETE-P set is an acceptable
282 ;; exception).
287 pred))
315 ;; We delete blocks when there is either no predecessor or the
316 ;; block is in a lambda that has been deleted. These blocks
317 ;; would eventually be deleted by DFO recomputation, but doing
318 ;; it here immediately makes the effect available to IR1
319 ;; optimization.
326 ;;; Loop over the nodes in BLOCK, acting on (and clearing) REOPTIMIZE
327 ;;; flags.
328 ;;;
329 ;;; Note that although they are cleared here, REOPTIMIZE flags might
330 ;;; still be set upon return from this function, meaning that further
331 ;;; optimization is wanted (as a consequence of optimizations we did).
334 ;; We clear the node and block REOPTIMIZE flags before doing the
335 ;; optimization, not after. This ensures that the node or block will
336 ;; be reoptimized if necessary.
340 ;; As above, we clear the node REOPTIMIZE flag before optimizing.
345 ;; With a COMBINATION, we call PROPAGATE-FUN-CHANGE whenever
346 ;; the function changes, and call IR1-OPTIMIZE-COMBINATION if
347 ;; any argument changes.
352 ;; KLUDGE: We leave the NODE-OPTIMIZE flag set going into
353 ;; IR1-OPTIMIZE-RETURN, since IR1-OPTIMIZE-RETURN wants to
354 ;; clear the flag itself. -- WHN 2002-02-02, quoting original
355 ;; CMU CL comments
361 ;; With an EXIT, we derive the node's type from the VALUE's
362 ;; type. We don't propagate CONT's assertion to the VALUE,
363 ;; since if we did, this would move the checking of CONT's
364 ;; assertion to the exit. This wouldn't work with CATCH and
365 ;; UWP, where the EXIT node is just a placeholder for the
366 ;; actual unknown exit.
374 ;;; Try to join with a successor block. If we succeed, we return true,
375 ;;; otherwise false.
385 ;; The successor has more than one predecessor.
387 ;; The last node's CONT is also used somewhere else.
389 ;; The successor is the current block (infinite loop).
391 ;; The next block has a different cleanup, and thus
392 ;; we may want to insert cleanup code between the
393 ;; two blocks at some point.
396 ;; The next block has a different home lambda, and
397 ;; thus the control transfer is a non-local exit.
400 nil)
401 ;; Joining is easy when the successor's START
402 ;; continuation is the same from our LAST's CONT.
405 t)
406 ;; If they differ, then we can still join when the last
407 ;; continuation has no next and the next continuation
408 ;; has no uses.
411 ;; In this case, we replace the next
412 ;; continuation with the last before joining the blocks.
414 ;; If NEXT-CONT does have a dest, it must be
415 ;; unreachable, since there are no USES.
416 ;; DELETE-CONTINUATION will mark the dest block as
417 ;; DELETE-P [and also this block, unless it is no
418 ;; longer backward reachable from the dest block.]
424 t)
426 nil))))))
428 ;;; Join together two blocks which have the same ending/starting
429 ;;; continuation. The code in BLOCK2 is moved into BLOCK1 and BLOCK2
430 ;;; is deleted from the DFO. We combine the optimize flags for the two
431 ;;; blocks so that any indicated optimization gets done.
464 ;;; Delete any nodes in BLOCK whose value is unused and which have no
465 ;;; side effects. We can delete sets of lexical variables when the set
466 ;;; variable has no references.
480 ;; ### For now, don't delete potentially
481 ;; flushable calls when they have the CALL
482 ;; attribute. Someday we should look at the
483 ;; functional args to determine if they have
484 ;; any side effects.
488 ;; FIXME: when bug 203
489 ;; will be fixed, remove
490 ;; this check
498 :lossage-fun nil
530 \f
531 ;;;; local call return type propagation
533 ;;; This function is called on RETURN nodes that have their REOPTIMIZE
534 ;;; flag set. It iterates over the uses of the RESULT, looking for
535 ;;; interesting stuff to update the TAIL-SET. If a use isn't a local
536 ;;; call, then we union its type together with the types of other such
537 ;;; uses. We assign to the RETURN-RESULT-TYPE the intersection of this
538 ;;; type with the RESULT's asserted type. We can make this
539 ;;; intersection now (potentially before type checking) because this
540 ;;; assertion on the result will eventually be checked (if
541 ;;; appropriate.)
542 ;;;
543 ;;; We call MAYBE-CONVERT-TAIL-LOCAL-CALL on each local non-MV
544 ;;; combination, which may change the succesor of the call to be the
545 ;;; called function, and if so, checks if the call can become an
546 ;;; assignment. If we convert to an assignment, we abort, since the
547 ;;; RETURN has been deleted.
568 ;;; Do stuff to realize that something has changed about the value
569 ;;; delivered to a return node. Since we consider the return values of
570 ;;; all functions in the tail set to be equivalent, this amounts to
571 ;;; bringing the entire tail set up to date. We iterate over the
572 ;;; returns for all the functions in the tail set, reanalyzing them
573 ;;; all (not treating Node specially.)
574 ;;;
575 ;;; When we are done, we check whether the new type is different from
576 ;;; the old TAIL-SET-TYPE. If so, we set the type and also reoptimize
577 ;;; all the continuations for references to functions in the tail set.
578 ;;; This will cause IR1-OPTIMIZE-COMBINATION to derive the new type as
579 ;;; the results of the calls.
600 \f
601 ;;;; IF optimization
603 ;;; If the test has multiple uses, replicate the node when possible.
604 ;;; Also check whether the predicate is known to be true or false,
605 ;;; deleting the IF node in favor of the appropriate branch when this
606 ;;; is the case.
638 ;;; Create a new copy of an IF node that tests the value of the node
639 ;;; USE. The test must have >1 use, and must be immediately used by
640 ;;; USE. NODE must be the only node in its block (implying that
641 ;;; block-start = if-test).
642 ;;;
643 ;;; This optimization has an effect semantically similar to the
644 ;;; source-to-source transformation:
645 ;;; (IF (IF A B C) D E) ==>
646 ;;; (IF A (IF B D E) (IF C D E))
647 ;;;
648 ;;; We clobber the NODE-SOURCE-PATH of both the original and the new
649 ;;; node so that dead code deletion notes will definitely not consider
650 ;;; either node to be part of the original source. One node might
651 ;;; become unreachable, resulting in a spurious note.
663 :consequent cblock
687 \f
688 ;;;; exit IR1 optimization
690 ;;; This function attempts to delete an exit node, returning true if
691 ;;; it deletes the block as a consequence:
692 ;;; -- If the exit is degenerate (has no ENTRY), then we don't do
693 ;;; anything, since there is nothing to be done.
694 ;;; -- If the exit node and its ENTRY have the same home lambda then
695 ;;; we know the exit is local, and can delete the exit. We change
696 ;;; uses of the Exit-Value to be uses of the original continuation,
697 ;;; then unlink the node. If the exit is to a TR context, then we
698 ;;; must do MERGE-TAIL-SETS on any local calls which delivered
699 ;;; their value to this exit.
700 ;;; -- If there is no value (as in a GO), then we skip the value
701 ;;; semantics.
702 ;;;
703 ;;; This function is also called by environment analysis, since it
704 ;;; wants all exits to be optimized even if normal optimization was
705 ;;; omitted.
726 \f
727 ;;;; combination IR1 optimization
729 ;;; Report as we try each transform?
730 #!+sb-show
733 ;;; Do IR1 optimizations on a COMBINATION node.
741 (:local
757 ;; KLUDGE: The next test could be made more sensitive,
758 ;; only suppressing constant-folding of functions with
759 ;; CALL attributes when they're actually passed
760 ;; function arguments. -- WHN 19990918
764 ;; Even if the function is foldable in principle,
765 ;; it might be one of our low-level
766 ;; implementation-specific functions. Such
767 ;; functions don't necessarily exist at runtime on
768 ;; a plain vanilla ANSI Common Lisp
769 ;; cross-compilation host, in which case the
770 ;; cross-compiler can't fold it because the
771 ;; cross-compiler doesn't know how to evaluate it.
772 #+sb-xc-host
787 #!+sb-show
793 #!+sb-show
800 ;;; If CALL is to a function that doesn't return (i.e. return type is
801 ;;; NIL), then terminate the block there, and link it to the component
802 ;;; tail. We also change the call's CONT to be a dummy continuation to
803 ;;; prevent the use from confusing things.
804 ;;;
805 ;;; Except when called during IR1 [FIXME: What does this mean? Except
806 ;;; during IR1 conversion? What about IR1 optimization?], we delete
807 ;;; the continuation if it has no other uses. (If it does have other
808 ;;; uses, we reoptimize.)
809 ;;;
810 ;;; Termination on the basis of a continuation type assertion is
811 ;;; inhibited when:
812 ;;; -- The continuation is deleted (hence the assertion is spurious), or
813 ;;; -- We are in IR1 conversion (where THE assertions are subject to
814 ;;; weakening.)
848 t))))
850 ;;; This is called both by IR1 conversion and IR1 optimization when
851 ;;; they have verified the type signature for the call, and are
852 ;;; wondering if something should be done to special-case the call. If
853 ;;; CALL is a call to a global function, then see whether it defined
854 ;;; or known:
855 ;;; -- If a DEFINED-FUN should be inline expanded, then convert
856 ;;; the expansion and change the call to call it. Expansion is
857 ;;; enabled if :INLINE or if SPACE=0. If the FUNCTIONAL slot is
858 ;;; true, we never expand, since this function has already been
859 ;;; converted. Local call analysis will duplicate the definition
860 ;;; if necessary. We claim that the parent form is LABELS for
861 ;;; context declarations, since we don't want it to be considered
862 ;;; a real global function.
863 ;;; -- If it is a known function, mark it as such by setting the KIND.
864 ;;;
865 ;;; We return the leaf referenced (NIL if not a leaf) and the
866 ;;; FUN-INFO assigned.
867 ;;;
868 ;;; FIXME: The IR1-CONVERTING-NOT-OPTIMIZING-P argument is what the
869 ;;; old CMU CL code called IR1-P, without explanation. My (WHN
870 ;;; 2002-01-09) tentative understanding of it is that we can call this
871 ;;; operation either in initial IR1 conversion or in later IR1
872 ;;; optimization, and it tells which is which. But it would be good
873 ;;; for someone who really understands it to check whether this is
874 ;;; really right.
898 ;; called semi-inlining? A more descriptive name would
899 ;; be nice. -- WHN 2002-01-07
903 leaf t
914 nil))
921 ;;; Check whether CALL satisfies TYPE. If so, apply the type to the
922 ;;; call, and do MAYBE-TERMINATE-BLOCK and return the values of
923 ;;; RECOGNIZE-KNOWN-CALL. If an error, set the combination kind and
924 ;;; return NIL, NIL. If the type is just FUNCTION, then skip the
925 ;;; syntax check, arg/result type processing, but still call
926 ;;; RECOGNIZE-KNOWN-CALL, since the call might be to a known lambda,
927 ;;; and that checking is done by local call analysis.
938 ;; KLUDGE: Common Lisp is such a dynamic
939 ;; language that all we can do here in
940 ;; general is issue a STYLE-WARNING. It
941 ;; would be nice to issue a full WARNING
942 ;; in the special case of of type
943 ;; mismatches within a compilation unit
944 ;; (as in section 3.2.2.3 of the spec)
945 ;; but at least as of sbcl-0.6.11, we
946 ;; don't keep track of whether the
947 ;; mismatched data came from the same
948 ;; compilation unit, so we can't do that.
949 ;; -- WHN 2001-02-11
950 ;;
951 ;; FIXME: Actually, I think we could
952 ;; issue a full WARNING if the call
953 ;; violates a DECLAIM FTYPE.
963 ;;; This is called by IR1-OPTIMIZE when the function for a call has
964 ;;; changed. If the call is local, we try to LET-convert it, and
965 ;;; derive the result type. If it is a :FULL call, we validate it
966 ;;; against the type, which recognizes known calls, does inline
967 ;;; expansion, etc. If a call to a predicate in a non-conditional
968 ;;; position or to a function with a source transform, then we
969 ;;; reconvert the form to give IR1 another chance.
976 (:local
981 (:full
987 call))
993 predicate)
996 ;; FIXME: This SYMBOLP is part of a literal
997 ;; translation of a test in the old CMU CL
998 ;; source, and it's not quite clear what
999 ;; the old source meant. Did it mean "has a
1000 ;; valid name"? Or did it mean "is an
1001 ;; ordinary function name, not a SETF
1002 ;; function"? Either way, the old CMU CL
1003 ;; code probably didn't deal with SETF
1004 ;; functions correctly, and neither does
1005 ;; this new SBCL code, and that should be fixed.
1015 \f
1016 ;;;; known function optimization
1018 ;;; Add a failed optimization note to FAILED-OPTIMZATIONS for NODE,
1019 ;;; FUN and ARGS. If there is already a note for NODE and TRANSFORM,
1020 ;;; replace it, otherwise add a new one.
1031 ;;; Attempt to transform NODE using TRANSFORM-FUNCTION, subject to the
1032 ;;; call type constraint TRANSFORM-TYPE. If we are inhibited from
1033 ;;; doing the transform for some reason and FLAME is true, then we
1034 ;;; make a note of the message in FAILED-OPTIMIZATIONS for IR1
1035 ;;; finalize to pick up. We return true if the transform failed, and
1036 ;;; thus further transformation should be attempted. We return false
1037 ;;; if either the transform succeeded or was aborted.
1057 (:none
1059 nil)
1060 (:aborted
1065 nil)
1066 (:failure
1072 t)
1073 (:delayed
1075 nil))))
1078 type
1082 t)
1084 t))))
1086 ;;; When we don't like an IR1 transform, we throw the severity/reason
1087 ;;; and args.
1088 ;;;
1089 ;;; GIVE-UP-IR1-TRANSFORM is used to throw out of an IR1 transform,
1090 ;;; aborting this attempt to transform the call, but admitting the
1091 ;;; possibility that this or some other transform will later succeed.
1092 ;;; If arguments are supplied, they are format arguments for an
1093 ;;; efficiency note.
1094 ;;;
1095 ;;; ABORT-IR1-TRANSFORM is used to throw out of an IR1 transform and
1096 ;;; force a normal call to the function at run time. No further
1097 ;;; optimizations will be attempted.
1098 ;;;
1099 ;;; DELAY-IR1-TRANSFORM is used to throw out of an IR1 transform, and
1100 ;;; delay the transform on the node until later. REASONS specifies
1101 ;;; when the transform will be later retried. The :OPTIMIZE reason
1102 ;;; causes the transform to be delayed until after the current IR1
1103 ;;; optimization pass. The :CONSTRAINT reason causes the transform to
1104 ;;; be delayed until after constraint propagation.
1105 ;;;
1106 ;;; FIXME: Now (0.6.11.44) that there are 4 variants of this (GIVE-UP,
1107 ;;; ABORT, DELAY/:OPTIMIZE, DELAY/:CONSTRAINT) and we're starting to
1108 ;;; do CASE operations on the various REASON values, it might be a
1109 ;;; good idea to go OO, representing the reasons by objects, using
1110 ;;; CLOS methods on the objects instead of CASE, and (possibly) using
1111 ;;; SIGNAL instead of THROW.
1128 ;;; Clear any delayed transform with no reasons - these should have
1129 ;;; been tried in the last pass. Then remove the reason from the
1130 ;;; delayed transform reasons, and if any become empty then set
1131 ;;; reoptimize flags for the node. Return true if any transforms are
1132 ;;; to be retried.
1148 reoptimize))
1150 ;;; Take the lambda-expression RES, IR1 convert it in the proper
1151 ;;; environment, and then install it as the function for the call
1152 ;;; NODE. We do local call analysis so that the new function is
1153 ;;; integrated into the control flow.
1154 ;;;
1155 ;;; We require the original function source name in order to generate
1156 ;;; a meaningful debug name for the lambda we set up. (It'd be
1157 ;;; possible to do this starting from debug names as well as source
1158 ;;; names, but as of sbcl-0.7.1.5, there was no need for this
1159 ;;; generality, since source names are always known to our callers.)
1166 res
1169 source-name
1177 ;;; Replace a call to a foldable function of constant arguments with
1178 ;;; the result of evaluating the form. We insert the resulting
1179 ;;; constant node after the call, stealing the call's continuation. We
1180 ;;; give the call a continuation with no DEST, which should cause it
1181 ;;; and its arguments to go away. If there is an error during the
1182 ;;; evaluation, we give a warning and leave the call alone, making the
1183 ;;; call a :ERROR call.
1184 ;;;
1185 ;;; If there is more than one value, then we transform the call into a
1186 ;;; VALUES form.
1192 args
1193 call
1194 ;; Note: CMU CL had COMPILER-WARN here, and that
1195 ;; seems more natural, but it's probably not.
1196 ;;
1197 ;; It's especially not while bug 173 exists:
1198 ;; Expressions like
1199 ;; (COND (END
1200 ;; (UNLESS (OR UNSAFE? (<= END SIZE)))
1201 ;; ...))
1202 ;; can cause constant-folding TYPE-ERRORs (in
1203 ;; #'<=) when END can be proved to be NIL, even
1204 ;; though the code is perfectly legal and safe
1205 ;; because a NIL value of END means that the
1206 ;; #'<= will never be executed.
1207 ;;
1208 ;; Moreover, even without bug 173,
1209 ;; quite-possibly-valid code like
1210 ;; (COND ((NONINLINED-PREDICATE END)
1211 ;; (UNLESS (<= END SIZE))
1212 ;; ...))
1213 ;; (where NONINLINED-PREDICATE is something the
1214 ;; compiler can't do at compile time, but which
1215 ;; turns out to make the #'<= expression
1216 ;; unreachable when END=NIL) could cause errors
1217 ;; when the compiler tries to constant-fold (<=
1218 ;; END SIZE).
1219 ;;
1220 ;; So, with or without bug 173, it'd be
1221 ;; unnecessarily evil to do a full
1222 ;; COMPILER-WARNING (and thus return FAILURE-P=T
1223 ;; from COMPILE-FILE) for legal code, so we we
1224 ;; use a wimpier COMPILE-STYLE-WARNING instead.
1225 #'compiler-style-warn
1231 call
1235 fun-name)))))
1237 \f
1238 ;;;; local call optimization
1240 ;;; Propagate TYPE to LEAF and its REFS, marking things changed. If
1241 ;;; the leaf type is a function type, then just leave it alone, since
1242 ;;; TYPE is never going to be more specific than that (and
1243 ;;; TYPE-INTERSECTION would choke.)
1255 ;;; Figure out the type of a LET variable that has sets. We compute
1256 ;;; the union of the initial value Type and the types of all the set
1257 ;;; values and to a PROPAGATE-TO-REFS with this type.
1266 ;;; If a LET variable, find the initial value's type and do
1267 ;;; PROPAGATE-FROM-SETS. We also derive the VALUE's type as the node's
1268 ;;; type.
1282 ;;; Return true if the value of REF will always be the same (and is
1283 ;;; thus legal to substitute.)
1297 ;;; If we have a non-set LET var with a single use, then (if possible)
1298 ;;; replace the variable reference's CONT with the arg continuation.
1299 ;;; This is inhibited when:
1300 ;;; -- CONT has other uses, or
1301 ;;; -- CONT receives multiple values, or
1302 ;;; -- the reference is in a different environment from the variable, or
1303 ;;; -- either continuation has a funky TYPE-CHECK annotation.
1304 ;;; -- the continuations have incompatible assertions, so the new asserted type
1305 ;;; would be NIL.
1306 ;;; -- the var's DEST has a different policy than the ARG's (think safety).
1307 ;;;
1308 ;;; We change the REF to be a reference to NIL with unused value, and
1309 ;;; let it be flushed as dead code. A side effect of this substitution
1310 ;;; is to delete the variable.
1318 dest
1325 cont-atype
1337 t)))
1339 ;;; Delete a LET, removing the call and bind nodes, and warning about
1340 ;;; any unreferenced variables. Note that FLUSH-DEAD-CODE will come
1341 ;;; along right away and delete the REF and then the lambda, since we
1342 ;;; flush the FUN continuation.
1354 ;;; This function is called when one of the arguments to a LET
1355 ;;; changes. We look at each changed argument. If the corresponding
1356 ;;; variable is set, then we call PROPAGATE-FROM-SETS. Otherwise, we
1357 ;;; consider substituting for the variable, and also propagate
1358 ;;; derived-type information for the arg to all the VAR's refs.
1359 ;;;
1360 ;;; Substitution is inhibited when the arg leaf's derived type isn't a
1361 ;;; subtype of the argument's asserted type. This prevents type
1362 ;;; checking from being defeated, and also ensures that the best
1363 ;;; representation for the variable can be used.
1364 ;;;
1365 ;;; Substitution of individual references is inhibited if the
1366 ;;; reference is in a different component from the home. This can only
1367 ;;; happen with closures over top level lambda vars. In such cases,
1368 ;;; the references may have already been compiled, and thus can't be
1369 ;;; retroactively modified.
1370 ;;;
1371 ;;; If all of the variables are deleted (have no references) when we
1372 ;;; are done, then we delete the LET.
1373 ;;;
1374 ;;; Note that we are responsible for clearing the
1375 ;;; CONTINUATION-REOPTIMIZE flags.
1396 t)
1399 nil)))
1400 leaf var))
1401 t)))))
1412 ;;; This function is called when one of the args to a non-LET local
1413 ;;; call changes. For each changed argument corresponding to an unset
1414 ;;; variable, we compute the union of the types across all calls and
1415 ;;; propagate this type information to the var's refs.
1416 ;;;
1417 ;;; If the function has an XEP, then we don't do anything, since we
1418 ;;; won't discover anything.
1419 ;;;
1420 ;;; We can clear the Continuation-Reoptimize flags for arguments in
1421 ;;; all calls corresponding to changed arguments in Call, since the
1422 ;;; only use in IR1 optimization of the Reoptimize flag for local call
1423 ;;; args is right here.
1436 vars))
1452 union)))))
1457 vars union)))
1460 \f
1461 ;;;; multiple values optimization
1463 ;;; Do stuff to notice a change to a MV combination node. There are
1464 ;;; two main branches here:
1465 ;;; -- If the call is local, then it is already a MV let, or should
1466 ;;; become one. Note that although all :LOCAL MV calls must eventually
1467 ;;; be converted to :MV-LETs, there can be a window when the call
1468 ;;; is local, but has not been LET converted yet. This is because
1469 ;;; the entry-point lambdas may have stray references (in other
1470 ;;; entry points) that have not been deleted yet.
1471 ;;; -- The call is full. This case is somewhat similar to the non-MV
1472 ;;; combination optimization: we propagate return type information and
1473 ;;; notice non-returning calls. We also have an optimization
1474 ;;; which tries to convert MV-CALLs into MV-binds.
1477 (:local
1486 (:full
1509 ;;; Propagate derived type info from the values continuation to the
1510 ;;; vars.
1522 vars
1529 ;;; If possible, convert a general MV call to an MV-BIND. We can do
1530 ;;; this if:
1531 ;;; -- The call has only one argument, and
1532 ;;; -- The function has a known fixed number of arguments, or
1533 ;;; -- The argument yields a known fixed number of values.
1534 ;;;
1535 ;;; What we do is change the function in the MV-CALL to be a lambda
1536 ;;; that "looks like an MV bind", which allows
1537 ;;; IR1-OPTIMIZE-MV-COMBINATION to notice that this call can be
1538 ;;; converted (the next time around.) This new lambda just calls the
1539 ;;; actual function with the MV-BIND variables as arguments. Note that
1540 ;;; this new MV bind is not let-converted immediately, as there are
1541 ;;; going to be stray references from the entry-point functions until
1542 ;;; they get deleted.
1543 ;;;
1544 ;;; In order to avoid loss of argument count checking, we only do the
1545 ;;; transformation according to a known number of expected argument if
1546 ;;; safety is unimportant. We can always convert if we know the number
1547 ;;; of actual values, since the normal call that we build will still
1548 ;;; do any appropriate argument count checking.
1549 ;;;
1550 ;;; We only attempt the transformation if the called function is a
1551 ;;; constant reference. This allows us to just splice the leaf into
1552 ;;; the new function, instead of trying to somehow bind the function
1553 ;;; expression. The leaf must be constant because we are evaluating it
1554 ;;; again in a different place. This also has the effect of squelching
1555 ;;; multiple warnings when there is an argument count error.
1577 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
1578 at least ~R."
1579 total-nvals min)
1584 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
1585 at most ~R."
1586 total-nvals max)
1593 min)
1609 ;;; If we see:
1610 ;;; (multiple-value-bind
1611 ;;; (x y)
1612 ;;; (values xx yy)
1613 ;;; ...)
1614 ;;; Convert to:
1615 ;;; (let ((x xx)
1616 ;;; (y yy))
1617 ;;; ...)
1618 ;;;
1619 ;;; What we actually do is convert the VALUES combination into a
1620 ;;; normal LET combination calling the original :MV-LET lambda. If
1621 ;;; there are extra args to VALUES, discard the corresponding
1622 ;;; continuations. If there are insufficient args, insert references
1623 ;;; to NIL.
1646 and prev = node-prev then cont
1666 t)))
1668 ;;; If we see:
1669 ;;; (values-list (list x y z))
1670 ;;;
1671 ;;; Convert to:
1672 ;;; (values x y z)
1673 ;;;
1674 ;;; In implementation, this is somewhat similar to
1675 ;;; CONVERT-MV-BIND-TO-LET. We grab the args of LIST and make them
1676 ;;; args of the VALUES-LIST call, flushing the old argument
1677 ;;; continuation (allowing the LIST to be flushed.)
1693 t)))
1695 ;;; If VALUES appears in a non-MV context, then effectively convert it
1696 ;;; to a PROG1. This allows the computation of the additional values
1697 ;;; to become dead code.
1707 val))
1708 nil))