| blob | 7238730930a0a9bddc28a62fa2fc27ab7e2dea10 |
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 an LVAR whose sole use is a reference to a
22 ;;; constant leaf.
29 ;;; Return the constant value for an LVAR whose only use is a constant
30 ;;; node.
35 \f
36 ;;;; interface for obtaining results of type inference
38 ;;; Our best guess for the type of this lvar's value. Note that this
39 ;;; may be VALUES or FUNCTION type, which cannot be passed as an
40 ;;; argument to the normal type operations. See LVAR-TYPE.
41 ;;;
42 ;;; The result value is cached in the LVAR-%DERIVED-TYPE slot. If the
43 ;;; slot is true, just return that value, otherwise recompute and
44 ;;; stash the value there.
58 res))
64 ;;; Return the derived type for LVAR's first value. This is guaranteed
65 ;;; not to be a VALUES or FUNCTION type.
70 ;;; If LVAR is an argument of a function, return a type which the
71 ;;; function checks LVAR for.
80 ;; TODO: MV-COMBINATION
88 ;; FUN-TYPE might be (AND FUNCTION (SATISFIES ...)).
101 dest)))))
107 \f
108 ;;;; interface routines used by optimizers
110 ;;; This function is called by optimizers to indicate that something
111 ;;; interesting has happened to the value of LVAR. Optimizers must
112 ;;; make sure that they don't call for reoptimization when nothing has
113 ;;; happened, since optimization will fail to terminate.
114 ;;;
115 ;;; We clear any cached type for the lvar and set the reoptimize flags
116 ;;; on everything in sight.
126 ;; PREV may be missing.
145 ;;; Annotate NODE to indicate that its result has been proven to be
146 ;;; TYPEP to RTYPE. After IR1 conversion has happened, this is the
147 ;;; only correct way to supply information discovered about a node's
148 ;;; type. If you screw with the NODE-DERIVED-TYPE directly, then
149 ;;; information may be lost and reoptimization may not happen.
150 ;;;
151 ;;; What we do is intersect RTYPE with NODE's DERIVED-TYPE. If the
152 ;;; intersection is different from the old type, then we do a
153 ;;; REOPTIMIZE-LVAR on the NODE-LVAR.
166 "New inferred type ~S conflicts with old type:~
167 ~% ~S~%*** possible internal error? Please report this."
170 ;; If the new type consists of only one object, replace the
171 ;; node with a constant reference.
182 ;;; This is similar to DERIVE-NODE-TYPE, but asserts that it is an
183 ;;; error for LVAR's value not to be TYPEP to TYPE. We implement it
184 ;;; splitting off DEST a new CAST node; old LVAR will deliver values
185 ;;; to CAST. If we improve the assertion, we set TYPE-CHECK and
186 ;;; TYPE-ASSERTED to guarantee that the new assertion will be checked.
206 type-check type-asserted)
207 t))))))
209 \f
210 ;;;; IR1-OPTIMIZE
212 ;;; Do one forward pass over COMPONENT, deleting unreachable blocks
213 ;;; and doing IR1 optimizations. We can ignore all blocks that don't
214 ;;; have the REOPTIMIZE flag set. If COMPONENT-REOPTIMIZE is true when
215 ;;; we are done, then another iteration would be beneficial.
221 for last-block = block
224 ;; We delete blocks when there is either no predecessor or the
225 ;; block is in a lambda that has been deleted. These blocks
226 ;; would eventually be deleted by DFO recomputation, but doing
227 ;; it here immediately makes the effect available to IR1
228 ;; optimization.
234 ;; Preserve the BLOCK-SUCC invariant that almost every block has
235 ;; one successor (and a block with DELETE-P set is an acceptable
236 ;; exception).
271 ;;; Loop over the nodes in BLOCK, acting on (and clearing) REOPTIMIZE
272 ;;; flags.
273 ;;;
274 ;;; Note that although they are cleared here, REOPTIMIZE flags might
275 ;;; still be set upon return from this function, meaning that further
276 ;;; optimization is wanted (as a consequence of optimizations we did).
279 ;; We clear the node and block REOPTIMIZE flags before doing the
280 ;; optimization, not after. This ensures that the node or block will
281 ;; be reoptimized if necessary.
285 ;; As above, we clear the node REOPTIMIZE flag before optimizing.
290 ;; With a COMBINATION, we call PROPAGATE-FUN-CHANGE whenever
291 ;; the function changes, and call IR1-OPTIMIZE-COMBINATION if
292 ;; any argument changes.
297 ;; KLUDGE: We leave the NODE-OPTIMIZE flag set going into
298 ;; IR1-OPTIMIZE-RETURN, since IR1-OPTIMIZE-RETURN wants to
299 ;; clear the flag itself. -- WHN 2002-02-02, quoting original
300 ;; CMU CL comments
306 ;; With an EXIT, we derive the node's type from the VALUE's
307 ;; type.
318 ;;; Try to join with a successor block. If we succeed, we return true,
319 ;;; otherwise false.
326 ;; The successor has more than one predecessor.
328 ;; The successor is the current block (infinite loop).
330 ;; The next block has a different cleanup, and thus
331 ;; we may want to insert cleanup code between the
332 ;; two blocks at some point.
335 ;; The next block has a different home lambda, and
336 ;; thus the control transfer is a non-local exit.
339 nil)
342 t)))))
344 ;;; Join together two blocks. The code in BLOCK2 is moved into BLOCK1
345 ;;; and BLOCK2 is deleted from the DFO. We combine the optimize flags
346 ;;; for the two blocks so that any indicated optimization gets done.
379 ;;; Delete any nodes in BLOCK whose value is unused and which have no
380 ;;; side effects. We can delete sets of lexical variables when the set
381 ;;; variable has no references.
396 ;; ### For now, don't delete potentially
397 ;; flushable calls when they have the CALL
398 ;; attribute. Someday we should look at the
399 ;; functional args to determine if they have
400 ;; any side effects.
433 \f
434 ;;;; local call return type propagation
436 ;;; This function is called on RETURN nodes that have their REOPTIMIZE
437 ;;; flag set. It iterates over the uses of the RESULT, looking for
438 ;;; interesting stuff to update the TAIL-SET. If a use isn't a local
439 ;;; call, then we union its type together with the types of other such
440 ;;; uses. We assign to the RETURN-RESULT-TYPE the intersection of this
441 ;;; type with the RESULT's asserted type. We can make this
442 ;;; intersection now (potentially before type checking) because this
443 ;;; assertion on the result will eventually be checked (if
444 ;;; appropriate.)
445 ;;;
446 ;;; We call MAYBE-CONVERT-TAIL-LOCAL-CALL on each local non-MV
447 ;;; combination, which may change the succesor of the call to be the
448 ;;; called function, and if so, checks if the call can become an
449 ;;; assignment. If we convert to an assignment, we abort, since the
450 ;;; RETURN has been deleted.
469 ;; (values-type-intersection
470 ;; (continuation-asserted-type result) ; FIXME -- APD, 2002-01-26
472 ;; )
473 ))
475 nil)
477 ;;; Do stuff to realize that something has changed about the value
478 ;;; delivered to a return node. Since we consider the return values of
479 ;;; all functions in the tail set to be equivalent, this amounts to
480 ;;; bringing the entire tail set up to date. We iterate over the
481 ;;; returns for all the functions in the tail set, reanalyzing them
482 ;;; all (not treating NODE specially.)
483 ;;;
484 ;;; When we are done, we check whether the new type is different from
485 ;;; the old TAIL-SET-TYPE. If so, we set the type and also reoptimize
486 ;;; all the lvars for references to functions in the tail set. This
487 ;;; will cause IR1-OPTIMIZE-COMBINATION to derive the new type as the
488 ;;; results of the calls.
492 :restart
512 \f
513 ;;;; IF optimization
515 ;;; If the test has multiple uses, replicate the node when possible.
516 ;;; Also check whether the predicate is known to be true or false,
517 ;;; deleting the IF node in favor of the appropriate branch when this
518 ;;; is the case.
549 ;;; Create a new copy of an IF node that tests the value of the node
550 ;;; USE. The test must have >1 use, and must be immediately used by
551 ;;; USE. NODE must be the only node in its block (implying that
552 ;;; block-start = if-test).
553 ;;;
554 ;;; This optimization has an effect semantically similar to the
555 ;;; source-to-source transformation:
556 ;;; (IF (IF A B C) D E) ==>
557 ;;; (IF A (IF B D E) (IF C D E))
558 ;;;
559 ;;; We clobber the NODE-SOURCE-PATH of both the original and the new
560 ;;; node so that dead code deletion notes will definitely not consider
561 ;;; either node to be part of the original source. One node might
562 ;;; become unreachable, resulting in a spurious note.
574 :consequent cblock
596 \f
597 ;;;; exit IR1 optimization
599 ;;; This function attempts to delete an exit node, returning true if
600 ;;; it deletes the block as a consequence:
601 ;;; -- If the exit is degenerate (has no ENTRY), then we don't do
602 ;;; anything, since there is nothing to be done.
603 ;;; -- If the exit node and its ENTRY have the same home lambda then
604 ;;; we know the exit is local, and can delete the exit. We change
605 ;;; uses of the Exit-Value to be uses of the original lvar,
606 ;;; then unlink the node. If the exit is to a TR context, then we
607 ;;; must do MERGE-TAIL-SETS on any local calls which delivered
608 ;;; their value to this exit.
609 ;;; -- If there is no value (as in a GO), then we skip the value
610 ;;; semantics.
611 ;;;
612 ;;; This function is also called by environment analysis, since it
613 ;;; wants all exits to be optimized even if normal optimization was
614 ;;; omitted.
626 \f
627 ;;;; combination IR1 optimization
629 ;;; Report as we try each transform?
630 #!+sb-show
633 ;;; Do IR1 optimizations on a COMBINATION node.
642 (:local
658 ;; KLUDGE: The next test could be made more sensitive,
659 ;; only suppressing constant-folding of functions with
660 ;; CALL attributes when they're actually passed
661 ;; function arguments. -- WHN 19990918
665 ;; Even if the function is foldable in principle,
666 ;; it might be one of our low-level
667 ;; implementation-specific functions. Such
668 ;; functions don't necessarily exist at runtime on
669 ;; a plain vanilla ANSI Common Lisp
670 ;; cross-compilation host, in which case the
671 ;; cross-compiler can't fold it because the
672 ;; cross-compiler doesn't know how to evaluate it.
673 #+sb-xc-host
678 nil)))
692 #!+sb-show
698 #!+sb-show
705 ;;; If NODE doesn't return (i.e. return type is NIL), then terminate
706 ;;; the block there, and link it to the component tail.
707 ;;;
708 ;;; Except when called during IR1 convertion, we delete the
709 ;;; continuation if it has no other uses. (If it does have other uses,
710 ;;; we reoptimize.)
711 ;;;
712 ;;; Termination on the basis of a continuation type is
713 ;;; inhibited when:
714 ;;; -- The continuation is deleted (hence the assertion is spurious), or
715 ;;; -- We are in IR1 conversion (where THE assertions are subject to
716 ;;; weakening.) FIXME: Now THE assertions are not weakened, but new
717 ;;; uses can(?) be added later. -- APD, 2003-07-17
718 ;;;
719 ;;; Why do we need to consider LVAR type? -- APD, 2003-07-30
751 t))))
753 ;;; This is called both by IR1 conversion and IR1 optimization when
754 ;;; they have verified the type signature for the call, and are
755 ;;; wondering if something should be done to special-case the call. If
756 ;;; CALL is a call to a global function, then see whether it defined
757 ;;; or known:
758 ;;; -- If a DEFINED-FUN should be inline expanded, then convert
759 ;;; the expansion and change the call to call it. Expansion is
760 ;;; enabled if :INLINE or if SPACE=0. If the FUNCTIONAL slot is
761 ;;; true, we never expand, since this function has already been
762 ;;; converted. Local call analysis will duplicate the definition
763 ;;; if necessary. We claim that the parent form is LABELS for
764 ;;; context declarations, since we don't want it to be considered
765 ;;; a real global function.
766 ;;; -- If it is a known function, mark it as such by setting the KIND.
767 ;;;
768 ;;; We return the leaf referenced (NIL if not a leaf) and the
769 ;;; FUN-INFO assigned.
793 ;; called semi-inlining? A more descriptive name would
794 ;; be nice. -- WHN 2002-01-07
798 leaf t
809 nil))
816 ;;; Check whether CALL satisfies TYPE. If so, apply the type to the
817 ;;; call, and do MAYBE-TERMINATE-BLOCK and return the values of
818 ;;; RECOGNIZE-KNOWN-CALL. If an error, set the combination kind and
819 ;;; return NIL, NIL. If the type is just FUNCTION, then skip the
820 ;;; syntax check, arg/result type processing, but still call
821 ;;; RECOGNIZE-KNOWN-CALL, since the call might be to a known lambda,
822 ;;; and that checking is done by local call analysis.
833 ;; KLUDGE: Common Lisp is such a dynamic
834 ;; language that all we can do here in
835 ;; general is issue a STYLE-WARNING. It
836 ;; would be nice to issue a full WARNING
837 ;; in the special case of of type
838 ;; mismatches within a compilation unit
839 ;; (as in section 3.2.2.3 of the spec)
840 ;; but at least as of sbcl-0.6.11, we
841 ;; don't keep track of whether the
842 ;; mismatched data came from the same
843 ;; compilation unit, so we can't do that.
844 ;; -- WHN 2001-02-11
845 ;;
846 ;; FIXME: Actually, I think we could
847 ;; issue a full WARNING if the call
848 ;; violates a DECLAIM FTYPE.
858 ;;; This is called by IR1-OPTIMIZE when the function for a call has
859 ;;; changed. If the call is local, we try to LET-convert it, and
860 ;;; derive the result type. If it is a :FULL call, we validate it
861 ;;; against the type, which recognizes known calls, does inline
862 ;;; expansion, etc. If a call to a predicate in a non-conditional
863 ;;; position or to a function with a source transform, then we
864 ;;; reconvert the form to give IR1 another chance.
871 (:local
876 (:full
882 call))
888 predicate)
902 \f
903 ;;;; known function optimization
905 ;;; Add a failed optimization note to FAILED-OPTIMZATIONS for NODE,
906 ;;; FUN and ARGS. If there is already a note for NODE and TRANSFORM,
907 ;;; replace it, otherwise add a new one.
918 ;;; Attempt to transform NODE using TRANSFORM-FUNCTION, subject to the
919 ;;; call type constraint TRANSFORM-TYPE. If we are inhibited from
920 ;;; doing the transform for some reason and FLAME is true, then we
921 ;;; make a note of the message in FAILED-OPTIMIZATIONS for IR1
922 ;;; finalize to pick up. We return true if the transform failed, and
923 ;;; thus further transformation should be attempted. We return false
924 ;;; if either the transform succeeded or was aborted.
944 (:none
946 nil)
947 (:aborted
952 nil)
953 (:failure
959 t)
960 (:delayed
962 nil))))
965 type
969 t)
971 t))))
973 ;;; When we don't like an IR1 transform, we throw the severity/reason
974 ;;; and args.
975 ;;;
976 ;;; GIVE-UP-IR1-TRANSFORM is used to throw out of an IR1 transform,
977 ;;; aborting this attempt to transform the call, but admitting the
978 ;;; possibility that this or some other transform will later succeed.
979 ;;; If arguments are supplied, they are format arguments for an
980 ;;; efficiency note.
981 ;;;
982 ;;; ABORT-IR1-TRANSFORM is used to throw out of an IR1 transform and
983 ;;; force a normal call to the function at run time. No further
984 ;;; optimizations will be attempted.
985 ;;;
986 ;;; DELAY-IR1-TRANSFORM is used to throw out of an IR1 transform, and
987 ;;; delay the transform on the node until later. REASONS specifies
988 ;;; when the transform will be later retried. The :OPTIMIZE reason
989 ;;; causes the transform to be delayed until after the current IR1
990 ;;; optimization pass. The :CONSTRAINT reason causes the transform to
991 ;;; be delayed until after constraint propagation.
992 ;;;
993 ;;; FIXME: Now (0.6.11.44) that there are 4 variants of this (GIVE-UP,
994 ;;; ABORT, DELAY/:OPTIMIZE, DELAY/:CONSTRAINT) and we're starting to
995 ;;; do CASE operations on the various REASON values, it might be a
996 ;;; good idea to go OO, representing the reasons by objects, using
997 ;;; CLOS methods on the objects instead of CASE, and (possibly) using
998 ;;; SIGNAL instead of THROW.
1015 ;;; Clear any delayed transform with no reasons - these should have
1016 ;;; been tried in the last pass. Then remove the reason from the
1017 ;;; delayed transform reasons, and if any become empty then set
1018 ;;; reoptimize flags for the node. Return true if any transforms are
1019 ;;; to be retried.
1035 reoptimize))
1037 ;;; Take the lambda-expression RES, IR1 convert it in the proper
1038 ;;; environment, and then install it as the function for the call
1039 ;;; NODE. We do local call analysis so that the new function is
1040 ;;; integrated into the control flow.
1041 ;;;
1042 ;;; We require the original function source name in order to generate
1043 ;;; a meaningful debug name for the lambda we set up. (It'd be
1044 ;;; possible to do this starting from debug names as well as source
1045 ;;; names, but as of sbcl-0.7.1.5, there was no need for this
1046 ;;; generality, since source names are always known to our callers.)
1056 res
1059 source-name
1067 ;;; Replace a call to a foldable function of constant arguments with
1068 ;;; the result of evaluating the form. If there is an error during the
1069 ;;; evaluation, we give a warning and leave the call alone, making the
1070 ;;; call a :ERROR call.
1071 ;;;
1072 ;;; If there is more than one value, then we transform the call into a
1073 ;;; VALUES form.
1079 args
1080 call
1081 ;; Note: CMU CL had COMPILER-WARN here, and that
1082 ;; seems more natural, but it's probably not.
1083 ;;
1084 ;; It's especially not while bug 173 exists:
1085 ;; Expressions like
1086 ;; (COND (END
1087 ;; (UNLESS (OR UNSAFE? (<= END SIZE)))
1088 ;; ...))
1089 ;; can cause constant-folding TYPE-ERRORs (in
1090 ;; #'<=) when END can be proved to be NIL, even
1091 ;; though the code is perfectly legal and safe
1092 ;; because a NIL value of END means that the
1093 ;; #'<= will never be executed.
1094 ;;
1095 ;; Moreover, even without bug 173,
1096 ;; quite-possibly-valid code like
1097 ;; (COND ((NONINLINED-PREDICATE END)
1098 ;; (UNLESS (<= END SIZE))
1099 ;; ...))
1100 ;; (where NONINLINED-PREDICATE is something the
1101 ;; compiler can't do at compile time, but which
1102 ;; turns out to make the #'<= expression
1103 ;; unreachable when END=NIL) could cause errors
1104 ;; when the compiler tries to constant-fold (<=
1105 ;; END SIZE).
1106 ;;
1107 ;; So, with or without bug 173, it'd be
1108 ;; unnecessarily evil to do a full
1109 ;; COMPILER-WARNING (and thus return FAILURE-P=T
1110 ;; from COMPILE-FILE) for legal code, so we we
1111 ;; use a wimpier COMPILE-STYLE-WARNING instead.
1112 #'compiler-style-warn
1131 call
1135 fun-name))))))
1137 \f
1138 ;;;; local call optimization
1140 ;;; Propagate TYPE to LEAF and its REFS, marking things changed. If
1141 ;;; the leaf type is a function type, then just leave it alone, since
1142 ;;; TYPE is never going to be more specific than that (and
1143 ;;; TYPE-INTERSECTION would choke.)
1153 ;; KLUDGE: LET var substitution
1159 ;;; Iteration variable: exactly one SETQ of the form:
1160 ;;;
1161 ;;; (let ((var initial))
1162 ;;; ...
1163 ;;; (setq var (+ var step))
1164 ;;; ...)
1201 :low low
1202 :high high
1205 "check for iteration variable reoptimization"
1215 ;;; Figure out the type of a LET variable that has sets. We compute
1216 ;;; the union of the INITIAL-TYPE and the types of all the set
1217 ;;; values and to a PROPAGATE-TO-REFS with this type.
1232 ;;; If a LET variable, find the initial value's type and do
1233 ;;; PROPAGATE-FROM-SETS. We also derive the VALUE's type as the node's
1234 ;;; type.
1250 ;;; Return true if the value of REF will always be the same (and is
1251 ;;; thus legal to substitute.)
1263 (:global-function
1270 ;;; If we have a non-set LET var with a single use, then (if possible)
1271 ;;; replace the variable reference's LVAR with the arg lvar.
1272 ;;;
1273 ;;; We change the REF to be a reference to NIL with unused value, and
1274 ;;; let it be flushed as dead code. A side effect of this substitution
1275 ;;; is to delete the variable.
1282 ;; Think about (LET ((A ...)) (IF ... A ...)): two
1283 ;; LVAR-USEs should not be met on one path.
1286 ;; we should not change lifetime of unknown values lvars
1298 ;; While CRETURN and EXIT nodes may be known-values,
1299 ;; they have their own complications, such as
1300 ;; substitution into CRETURN may create new tail calls.
1301 nil)
1304 t))
1314 t)))
1316 ;;; Delete a LET, removing the call and bind nodes, and warning about
1317 ;;; any unreferenced variables. Note that FLUSH-DEAD-CODE will come
1318 ;;; along right away and delete the REF and then the lambda, since we
1319 ;;; flush the FUN lvar.
1331 ;;; This function is called when one of the arguments to a LET
1332 ;;; changes. We look at each changed argument. If the corresponding
1333 ;;; variable is set, then we call PROPAGATE-FROM-SETS. Otherwise, we
1334 ;;; consider substituting for the variable, and also propagate
1335 ;;; derived-type information for the arg to all the VAR's refs.
1336 ;;;
1337 ;;; Substitution is inhibited when the arg leaf's derived type isn't a
1338 ;;; subtype of the argument's leaf type. This prevents type checking
1339 ;;; from being defeated, and also ensures that the best representation
1340 ;;; for the variable can be used.
1341 ;;;
1342 ;;; Substitution of individual references is inhibited if the
1343 ;;; reference is in a different component from the home. This can only
1344 ;;; happen with closures over top level lambda vars. In such cases,
1345 ;;; the references may have already been compiled, and thus can't be
1346 ;;; retroactively modified.
1347 ;;;
1348 ;;; If all of the variables are deleted (have no references) when we
1349 ;;; are done, then we delete the LET.
1350 ;;;
1351 ;;; Note that we are responsible for clearing the LVAR-REOPTIMIZE
1352 ;;; flags.
1367 ;; (NODE-DERIVED-TYPE USE) would
1368 ;; be better -- APD, 2003-05-15
1375 t)
1378 nil)))
1379 leaf var)))
1380 t)))))
1391 ;;; This function is called when one of the args to a non-LET local
1392 ;;; call changes. For each changed argument corresponding to an unset
1393 ;;; variable, we compute the union of the types across all calls and
1394 ;;; propagate this type information to the var's refs.
1395 ;;;
1396 ;;; If the function has an XEP, then we don't do anything, since we
1397 ;;; won't discover anything.
1398 ;;;
1399 ;;; We can clear the LVAR-REOPTIMIZE flags for arguments in all calls
1400 ;;; corresponding to changed arguments in CALL, since the only use in
1401 ;;; IR1 optimization of the REOPTIMIZE flag for local call args is
1402 ;;; right here.
1415 vars))
1431 union)))))
1434 and type in union
1438 \f
1439 ;;;; multiple values optimization
1441 ;;; Do stuff to notice a change to a MV combination node. There are
1442 ;;; two main branches here:
1443 ;;; -- If the call is local, then it is already a MV let, or should
1444 ;;; become one. Note that although all :LOCAL MV calls must eventually
1445 ;;; be converted to :MV-LETs, there can be a window when the call
1446 ;;; is local, but has not been LET converted yet. This is because
1447 ;;; the entry-point lambdas may have stray references (in other
1448 ;;; entry points) that have not been deleted yet.
1449 ;;; -- The call is full. This case is somewhat similar to the non-MV
1450 ;;; combination optimization: we propagate return type information and
1451 ;;; notice non-returning calls. We also have an optimization
1452 ;;; which tries to convert MV-CALLs into MV-binds.
1455 (:local
1464 (:full
1487 ;;; Propagate derived type info from the values lvar to the vars.
1494 n-vars)))
1496 and type in types
1503 ;;; If possible, convert a general MV call to an MV-BIND. We can do
1504 ;;; this if:
1505 ;;; -- The call has only one argument, and
1506 ;;; -- The function has a known fixed number of arguments, or
1507 ;;; -- The argument yields a known fixed number of values.
1508 ;;;
1509 ;;; What we do is change the function in the MV-CALL to be a lambda
1510 ;;; that "looks like an MV bind", which allows
1511 ;;; IR1-OPTIMIZE-MV-COMBINATION to notice that this call can be
1512 ;;; converted (the next time around.) This new lambda just calls the
1513 ;;; actual function with the MV-BIND variables as arguments. Note that
1514 ;;; this new MV bind is not let-converted immediately, as there are
1515 ;;; going to be stray references from the entry-point functions until
1516 ;;; they get deleted.
1517 ;;;
1518 ;;; In order to avoid loss of argument count checking, we only do the
1519 ;;; transformation according to a known number of expected argument if
1520 ;;; safety is unimportant. We can always convert if we know the number
1521 ;;; of actual values, since the normal call that we build will still
1522 ;;; do any appropriate argument count checking.
1523 ;;;
1524 ;;; We only attempt the transformation if the called function is a
1525 ;;; constant reference. This allows us to just splice the leaf into
1526 ;;; the new function, instead of trying to somehow bind the function
1527 ;;; expression. The leaf must be constant because we are evaluating it
1528 ;;; again in a different place. This also has the effect of squelching
1529 ;;; multiple warnings when there is an argument count error.
1551 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
1552 at least ~R."
1553 total-nvals min)
1558 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
1559 at most ~R."
1560 total-nvals max)
1567 min)
1583 ;;; If we see:
1584 ;;; (multiple-value-bind
1585 ;;; (x y)
1586 ;;; (values xx yy)
1587 ;;; ...)
1588 ;;; Convert to:
1589 ;;; (let ((x xx)
1590 ;;; (y yy))
1591 ;;; ...)
1592 ;;;
1593 ;;; What we actually do is convert the VALUES combination into a
1594 ;;; normal LET combination calling the original :MV-LET lambda. If
1595 ;;; there are extra args to VALUES, discard the corresponding
1596 ;;; lvars. If there are insufficient args, insert references to NIL.
1619 for prev = node-prev then ctran
1625 use ctran))
1641 t)))
1643 ;;; If we see:
1644 ;;; (values-list (list x y z))
1645 ;;;
1646 ;;; Convert to:
1647 ;;; (values x y z)
1648 ;;;
1649 ;;; In implementation, this is somewhat similar to
1650 ;;; CONVERT-MV-BIND-TO-LET. We grab the args of LIST and make them
1651 ;;; args of the VALUES-LIST call, flushing the old argument lvar
1652 ;;; (allowing the LIST to be flushed.)
1653 ;;;
1654 ;;; FIXME: Thus we lose possible type assertions on (LIST ...).
1661 ;; FIXME: VALUES might not satisfy an assertion on NODE-LVAR.
1672 t)))
1674 ;;; If VALUES appears in a non-MV context, then effectively convert it
1675 ;;; to a PROG1. This allows the computation of the additional values
1676 ;;; to become dead code.
1686 val))
1687 nil))
1689 ;;; TODO:
1690 ;;; - CAST chains;
1707 lvar)
1708 ;; Pathwise removing of CAST
1711 next-block)
1736 ;; FIXME: Do it in one step.
1738 value
1742 ;; FIXME: Derived type.
1746 ;; KLUDGE: FILTER-LVAR does not work for non-returning
1747 ;; functions, so we declare the return type of
1748 ;; %COMPILE-TIME-TYPE-ERROR to be * and derive the real type
1749 ;; here.
1753 ;; FIXME: Is it necessary?