| blob | 79b70a163f479cdbb114e3a1d096a8bf19ebdcb5 |
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.
28 ;; check for EQL types and singleton numeric types
31 ;;; Same as above except it doesn't consider EQL types
38 ;;; Return the constant value for an LVAR whose only use is a constant
39 ;;; node.
44 leaf)
51 value))))
52 \f
53 ;;;; interface for obtaining results of type inference
55 ;;; Our best guess for the type of this lvar's value. Note that this
56 ;;; may be VALUES or FUNCTION type, which cannot be passed as an
57 ;;; argument to the normal type operations. See LVAR-TYPE.
58 ;;;
59 ;;; The result value is cached in the LVAR-%DERIVED-TYPE slot. If the
60 ;;; slot is true, just return that value, otherwise recompute and
61 ;;; stash the value there.
71 res))
74 res)))
81 ;;; Return the derived type for LVAR's first value. This is guaranteed
82 ;;; not to be a VALUES or FUNCTION type.
87 ;;; LVAR-CONSERVATIVE-TYPE
88 ;;;
89 ;;; Certain types refer to the contents of an object, which can
90 ;;; change without type derivation noticing: CONS types and ARRAY
91 ;;; types suffer from this:
92 ;;;
93 ;;; (let ((x (the (cons fixnum fixnum) (cons a b))))
94 ;;; (setf (car x) c)
95 ;;; (+ (car x) (cdr x)))
96 ;;;
97 ;;; Python doesn't realize that the SETF CAR can change the type of X -- so we
98 ;;; cannot use LVAR-TYPE which gets the derived results. Worse, still, instead
99 ;;; of (SETF CAR) we might have a call to a user-defined function FOO which
100 ;;; does the same -- so there is no way to use the derived information in
101 ;;; general.
102 ;;;
103 ;;; So, the conservative option is to use the derived type if the leaf has
104 ;;; only a single ref -- in which case there cannot be a prior call that
105 ;;; mutates it. Otherwise we use the declared type or punt to the most general
106 ;;; type we know to be correct for sure.
110 ;; Recompute using NODE-CONSERVATIVE-TYPE instead of derived type if
111 ;; necessary -- picking off some easy cases up front.
113 ;; Can't use CSUBTYPEP!
116 derived-type)
120 derived-type)
126 derived-type)
130 derived-type))))
143 derived-values-type))
144 derived-values-type)))
150 type)
156 ;; ADJUST-ARRAY may change dimensions, but rank stays same.
159 old
161 ;; Complexity cannot change.
163 ;; Element type cannot change.
166 ;; Simple arrays cannot change at all.
167 type))
169 ;; Conservative union type is an union of conservative types.
174 ;; Catch-all.
175 ;;
176 ;; If the type contains some CONS types, the conservative type contains all
177 ;; of them.
180 ;; Similarly for non-simple arrays -- it should be possible to preserve
181 ;; more information here, but really...
185 type)))
189 ;; Excluding T is necessary, because we do want type derivation to
190 ;; be able to narrow it down in case someone (most like a macro-expansion...)
191 ;; actually declares something as having type T.
192 nil)
194 ;; Covered by the next case as well, but this is a quick test.
195 t)
197 t)))
199 ;;; If LVAR is an argument of a function, return a type which the
200 ;;; function checks LVAR for.
209 ;; TODO: MV-COMBINATION
217 ;; FUN-TYPE might be (AND FUNCTION (SATISFIES ...)).
230 dest)))))
232 \f
233 ;;;; interface routines used by optimizers
235 ;;; This function is called by optimizers to indicate that something
236 ;;; interesting has happened to the value of LVAR. Optimizers must
237 ;;; make sure that they don't call for reoptimization when nothing has
238 ;;; happened, since optimization will fail to terminate.
239 ;;;
240 ;;; We clear any cached type for the lvar and set the reoptimize flags
241 ;;; on everything in sight.
251 ;; PREV may be missing.
270 ;;; Annotate NODE to indicate that its result has been proven to be
271 ;;; TYPEP to RTYPE. After IR1 conversion has happened, this is the
272 ;;; only correct way to supply information discovered about a node's
273 ;;; type. If you screw with the NODE-DERIVED-TYPE directly, then
274 ;;; information may be lost and reoptimization may not happen.
275 ;;;
276 ;;; What we do is intersect RTYPE with NODE's DERIVED-TYPE. If the
277 ;;; intersection is different from the old type, then we do a
278 ;;; REOPTIMIZE-LVAR on the NODE-LVAR.
284 initial-type)))
295 "New inferred type ~S conflicts with old type:~
296 ~% ~S~%*** possible internal error? Please report this."
299 ;; If the new type consists of only one object, replace the
300 ;; node with a constant reference.
311 ;;; This is similar to DERIVE-NODE-TYPE, but asserts that it is an
312 ;;; error for LVAR's value not to be TYPEP to TYPE. We implement it
313 ;;; splitting off DEST a new CAST node; old LVAR will deliver values
314 ;;; to CAST. If we improve the assertion, we set TYPE-CHECK and
315 ;;; TYPE-ASSERTED to guarantee that the new assertion will be checked.
324 t))))
326 \f
327 ;;;; IR1-OPTIMIZE
329 ;;; Do one forward pass over COMPONENT, deleting unreachable blocks
330 ;;; and doing IR1 optimizations. We can ignore all blocks that don't
331 ;;; have the REOPTIMIZE flag set. If COMPONENT-REOPTIMIZE is true when
332 ;;; we are done, then another iteration would be beneficial.
338 for last-block = block
341 ;; We delete blocks when there is either no predecessor or the
342 ;; block is in a lambda that has been deleted. These blocks
343 ;; would eventually be deleted by DFO recomputation, but doing
344 ;; it here immediately makes the effect available to IR1
345 ;; optimization.
351 ;; Preserve the BLOCK-SUCC invariant that almost every block has
352 ;; one successor (and a block with DELETE-P set is an acceptable
353 ;; exception).
388 ;;; Loop over the nodes in BLOCK, acting on (and clearing) REOPTIMIZE
389 ;;; flags.
390 ;;;
391 ;;; Note that although they are cleared here, REOPTIMIZE flags might
392 ;;; still be set upon return from this function, meaning that further
393 ;;; optimization is wanted (as a consequence of optimizations we did).
396 ;; We clear the node and block REOPTIMIZE flags before doing the
397 ;; optimization, not after. This ensures that the node or block will
398 ;; be reoptimized if necessary.
402 ;; As above, we clear the node REOPTIMIZE flag before optimizing.
407 ;; With a COMBINATION, we call PROPAGATE-FUN-CHANGE whenever
408 ;; the function changes, and call IR1-OPTIMIZE-COMBINATION if
409 ;; any argument changes.
414 ;; KLUDGE: We leave the NODE-OPTIMIZE flag set going into
415 ;; IR1-OPTIMIZE-RETURN, since IR1-OPTIMIZE-RETURN wants to
416 ;; clear the flag itself. -- WHN 2002-02-02, quoting original
417 ;; CMU CL comments
423 ;; With an EXIT, we derive the node's type from the VALUE's
424 ;; type.
429 ;; PROPAGATE-FROM-SETS can do a better job if NODE-REOPTIMIZE
430 ;; is accurate till the node actually has been reoptimized.
438 ;;; Try to join with a successor block. If we succeed, we return true,
439 ;;; otherwise false.
446 ;; the successor has more than one predecessor;
448 ;; the successor is the current block (infinite loop);
450 ;; the next block has a different cleanup, and thus
451 ;; we may want to insert cleanup code between the
452 ;; two blocks at some point;
455 ;; the next block has a different home lambda, and
456 ;; thus the control transfer is a non-local exit.
459 ;; Stack analysis phase wants ENTRY to start a block...
465 ;; ... and a DX-allocator to end a block.
467 ;; FIXME: This is a partial workaround for bug 303.
469 nil)
472 t)))))
474 ;;; Join together two blocks. The code in BLOCK2 is moved into BLOCK1
475 ;;; and BLOCK2 is deleted from the DFO. We combine the optimize flags
476 ;;; for the two blocks so that any indicated optimization gets done.
509 ;;; Delete any nodes in BLOCK whose value is unused and which have no
510 ;;; side effects. We can delete sets of lexical variables when the set
511 ;;; variable has no references.
557 victim)
558 \f
559 ;;;; local call return type propagation
561 ;;; This function is called on RETURN nodes that have their REOPTIMIZE
562 ;;; flag set. It iterates over the uses of the RESULT, looking for
563 ;;; interesting stuff to update the TAIL-SET. If a use isn't a local
564 ;;; call, then we union its type together with the types of other such
565 ;;; uses. We assign to the RETURN-RESULT-TYPE the intersection of this
566 ;;; type with the RESULT's asserted type. We can make this
567 ;;; intersection now (potentially before type checking) because this
568 ;;; assertion on the result will eventually be checked (if
569 ;;; appropriate.)
570 ;;;
571 ;;; We call MAYBE-CONVERT-TAIL-LOCAL-CALL on each local non-MV
572 ;;; combination, which may change the successor of the call to be the
573 ;;; called function, and if so, checks if the call can become an
574 ;;; assignment. If we convert to an assignment, we abort, since the
575 ;;; RETURN has been deleted.
594 ;; (values-type-intersection
595 ;; (continuation-asserted-type result) ; FIXME -- APD, 2002-01-26
597 ;; )
598 ))
600 nil)
602 ;;; Do stuff to realize that something has changed about the value
603 ;;; delivered to a return node. Since we consider the return values of
604 ;;; all functions in the tail set to be equivalent, this amounts to
605 ;;; bringing the entire tail set up to date. We iterate over the
606 ;;; returns for all the functions in the tail set, reanalyzing them
607 ;;; all (not treating NODE specially.)
608 ;;;
609 ;;; When we are done, we check whether the new type is different from
610 ;;; the old TAIL-SET-TYPE. If so, we set the type and also reoptimize
611 ;;; all the lvars for references to functions in the tail set. This
612 ;;; will cause IR1-OPTIMIZE-COMBINATION to derive the new type as the
613 ;;; results of the calls.
617 :restart
637 \f
638 ;;;; IF optimization
640 ;;; Utility: return T if both argument cblocks are equivalent. For now,
641 ;;; detect only blocks that read the same leaf into the same lvar, and
642 ;;; continue to the same block.
655 ;;; Check whether the predicate is known to be true or false,
656 ;;; deleting the IF node in favor of the appropriate branch when this
657 ;;; is the case.
658 ;;; Similarly, when both branches are equivalent, branch directly to either
659 ;;; of them.
660 ;;; Also, if the test has multiple uses, replicate the node when possible...
661 ;;; in fact, splice in direct jumps to the right branch if possible.
673 alternative)
675 consequent)
678 alternative))))
686 ;; When we know that we only have a single successor, kill the victim
687 ;; ... unless the victim and the remaining successor are the same.
696 ;; When if/if conversion would leave (if ... (if nil ...)) or
697 ;; (if ... (if not-nil ...)), splice the correct successor right
698 ;; in.
717 ;; only one use left. Just kill the now-useless
718 ;; branch to avoid spurious code deletion notes.
721 node test block
727 ;; Finally, duplicate EQ-nil tests
734 ;; leave the last use as is, instead of replacing
735 ;; the (singly-referenced) CIF node with a duplicate.
738 ;;; Create a new copy of an IF node that tests the value of the node
739 ;;; USE. The test must have >1 use, and must be immediately used by
740 ;;; USE. NODE must be the only node in its block (implying that
741 ;;; block-start = if-test).
742 ;;;
743 ;;; This optimization has an effect semantically similar to the
744 ;;; source-to-source transformation:
745 ;;; (IF (IF A B C) D E) ==>
746 ;;; (IF A (IF B D E) (IF C D E))
747 ;;;
748 ;;; We clobber the NODE-SOURCE-PATH of both the original and the new
749 ;;; node so that dead code deletion notes will definitely not consider
750 ;;; either node to be part of the original source. One node might
751 ;;; become unreachable, resulting in a spurious note.
763 :consequent cblock
785 \f
786 ;;;; exit IR1 optimization
788 ;;; This function attempts to delete an exit node, returning true if
789 ;;; it deletes the block as a consequence:
790 ;;; -- If the exit is degenerate (has no ENTRY), then we don't do
791 ;;; anything, since there is nothing to be done.
792 ;;; -- If the exit node and its ENTRY have the same home lambda then
793 ;;; we know the exit is local, and can delete the exit. We change
794 ;;; uses of the Exit-Value to be uses of the original lvar,
795 ;;; then unlink the node. If the exit is to a TR context, then we
796 ;;; must do MERGE-TAIL-SETS on any local calls which delivered
797 ;;; their value to this exit.
798 ;;; -- If there is no value (as in a GO), then we skip the value
799 ;;; semantics.
800 ;;;
801 ;;; This function is also called by environment analysis, since it
802 ;;; wants all exits to be optimized even if normal optimization was
803 ;;; omitted.
813 ;; We can't simply use DELETE-FILTER to unlink the node
814 ;; and substitute some LVAR magic, as this can confuse the
815 ;; stack analysis if there's another EXIT to the same
816 ;; continuation. Instead, we fabricate a new block (in
817 ;; the same lexenv as the ENTRY, so it can't be merged
818 ;; backwards), insert a gimmicked CAST node to link up the
819 ;; LVAR holding the value being returned to the LVAR which
820 ;; is expecting to accept the value, thus placing the
821 ;; return value where it needs to be while still providing
822 ;; the hook required for stack analysis.
823 ;; -- AJB, 2014-Mar-03
828 (%make-cast
831 :value value
835 ;; We only expect a single successor to EXIT-BLOCK,
836 ;; because it contains an EXIT node (which must end its
837 ;; block) and the only blocks that have more than once
838 ;; successor are those with IF nodes (which also must
839 ;; end their blocks). Still, just to be sure, we use a
840 ;; construct that guarantees an error if this
841 ;; expectation is violated.
846 ;; Finish creating the new block.
850 ;; Link the new block into the control sequence.
855 ;; Finish re-pointing the value-holding LVAR to the
856 ;; CAST node.
861 ;; Register the CAST node as providing a value to the
862 ;; LVAR for the continuation.
866 ;; Remove the EXIT node.
869 ;; And, because we created a new block, we need to
870 ;; force component reanalysis (to assign a DFO number
871 ;; to the block if nothing else).
875 \f
876 ;;;; combination IR1 optimization
878 ;;; Report as we try each transform?
879 #!+sb-show
887 "The return value of ~A should not be discarded."
890 ;;; Do IR1 optimizations on a COMBINATION node.
900 (:local
905 (:error
909 (:full
930 ;; Check against the DEFINED-TYPE unless TYPE is already good.
940 (:known
948 ;; If somebody is really sure that they want to modify
949 ;; constants, let them.
978 ;; First give the VM a peek at the call
982 (:direct
983 ;; The VM knows how to handle this.
984 )
985 (:transform
986 ;; The VM mostly knows how to handle this. We need
987 ;; to massage the call slightly, though.
990 ;; Let transforms have a crack at it.
992 #!+sb-show
998 #!+sb-show
1013 ;;; If NODE doesn't return (i.e. return type is NIL), then terminate
1014 ;;; the block there, and link it to the component tail.
1015 ;;;
1016 ;;; Except when called during IR1 convertion, we delete the
1017 ;;; continuation if it has no other uses. (If it does have other uses,
1018 ;;; we reoptimize.)
1019 ;;;
1020 ;;; Termination on the basis of a continuation type is
1021 ;;; inhibited when:
1022 ;;; -- The continuation is deleted (hence the assertion is spurious), or
1023 ;;; -- We are in IR1 conversion (where THE assertions are subject to
1024 ;;; weakening.) FIXME: Now THE assertions are not weakened, but new
1025 ;;; uses can(?) be added later. -- APD, 2003-07-17
1026 ;;;
1027 ;;; Why do we need to consider LVAR type? -- APD, 2003-07-30
1038 ;; Even if the combination will never return, don't terminate if this
1039 ;; is the tail call of a XEP: doing that would inhibit TCO.
1066 t))))
1068 ;;; This is called both by IR1 conversion and IR1 optimization when
1069 ;;; they have verified the type signature for the call, and are
1070 ;;; wondering if something should be done to special-case the call. If
1071 ;;; CALL is a call to a global function, then see whether it defined
1072 ;;; or known:
1073 ;;; -- If a DEFINED-FUN should be inline expanded, then convert
1074 ;;; the expansion and change the call to call it. Expansion is
1075 ;;; enabled if :INLINE or if SPACE=0. If the FUNCTIONAL slot is
1076 ;;; true, we never expand, since this function has already been
1077 ;;; converted. Local call analysis will duplicate the definition
1078 ;;; if necessary. We claim that the parent form is LABELS for
1079 ;;; context declarations, since we don't want it to be considered
1080 ;;; a real global function.
1081 ;;; -- If it is a known function, mark it as such by setting the KIND.
1082 ;;;
1083 ;;; We return the leaf referenced (NIL if not a leaf) and the
1084 ;;; FUN-INFO assigned.
1108 ;; Inline: if the function has already been converted at another call
1109 ;; site in this component, we point this REF to the functional. If not,
1110 ;; we convert the expansion.
1111 ;;
1112 ;; For :INLINE case local call analysis will copy the expansion later,
1113 ;; but for :MAYBE-INLINE and NIL cases we only get one copy of the
1114 ;; expansion per component.
1115 ;;
1116 ;; FIXME: We also convert in :INLINE & FUNCTIONAL-KIND case below. What
1117 ;; is it for?
1121 name
1123 leaf
1124 inlinep
1126 ;; Allow backward references to this function from following
1127 ;; forms. (Reused only if policy matches.)
1133 ;; Convert.
1139 ;; If we've already converted, change ref to the converted
1140 ;; functional.
1152 ;;; Check whether CALL satisfies TYPE. If so, apply the type to the
1153 ;;; call, and do MAYBE-TERMINATE-BLOCK and return the values of
1154 ;;; RECOGNIZE-KNOWN-CALL. If an error, set the combination kind and
1155 ;;; return NIL, NIL. If the type is just FUNCTION, then skip the
1156 ;;; syntax check, arg/result type processing, but still call
1157 ;;; RECOGNIZE-KNOWN-CALL, since the call might be to a known lambda,
1158 ;;; and that checking is done by local call analysis.
1167 ;; Using the defined-type too early is a bit of a waste: during
1168 ;; conversion we cannot use the untrusted ASSERT-CALL-TYPE, etc.
1173 ;; Don't validate multiple times against the same leaf --
1174 ;; it doesn't add any information, but may generate the same warning
1175 ;; multiple times.
1179 :result-test nil
1181 #'compiler-warn
1184 same-file-p)
1190 :result-test nil
1200 ;;; This is called by IR1-OPTIMIZE when the function for a call has
1201 ;;; changed. If the call is local, we try to LET-convert it, and
1202 ;;; derive the result type. If it is a :FULL call, we validate it
1203 ;;; against the type, which recognizes known calls, does inline
1204 ;;; expansion, etc. If a call to a predicate in a non-conditional
1205 ;;; position or to a function with a source transform, then we
1206 ;;; reconvert the form to give IR1 another chance.
1213 (:local
1218 (:full
1226 call))
1234 predicate)
1248 \f
1249 ;;;; known function optimization
1251 ;;; Add a failed optimization note to FAILED-OPTIMZATIONS for NODE,
1252 ;;; FUN and ARGS. If there is already a note for NODE and TRANSFORM,
1253 ;;; replace it, otherwise add a new one.
1264 ;;; Attempt to transform NODE using TRANSFORM-FUNCTION, subject to the
1265 ;;; call type constraint TRANSFORM-TYPE. If we are inhibited from
1266 ;;; doing the transform for some reason and FLAME is true, then we
1267 ;;; make a note of the message in FAILED-OPTIMIZATIONS for IR1
1268 ;;; finalize to pick up. We return true if the transform failed, and
1269 ;;; thus further transformation should be attempted. We return false
1270 ;;; if either the transform succeeded or was aborted.
1290 (:none
1292 nil)
1293 (:aborted
1298 nil)
1299 (:failure
1305 t)
1306 (:delayed
1308 nil))))
1311 type
1315 t)
1317 t))))
1319 ;;; When we don't like an IR1 transform, we throw the severity/reason
1320 ;;; and args.
1321 ;;;
1322 ;;; GIVE-UP-IR1-TRANSFORM is used to throw out of an IR1 transform,
1323 ;;; aborting this attempt to transform the call, but admitting the
1324 ;;; possibility that this or some other transform will later succeed.
1325 ;;; If arguments are supplied, they are format arguments for an
1326 ;;; efficiency note.
1327 ;;;
1328 ;;; ABORT-IR1-TRANSFORM is used to throw out of an IR1 transform and
1329 ;;; force a normal call to the function at run time. No further
1330 ;;; optimizations will be attempted.
1331 ;;;
1332 ;;; DELAY-IR1-TRANSFORM is used to throw out of an IR1 transform, and
1333 ;;; delay the transform on the node until later. REASONS specifies
1334 ;;; when the transform will be later retried. The :OPTIMIZE reason
1335 ;;; causes the transform to be delayed until after the current IR1
1336 ;;; optimization pass. The :CONSTRAINT reason causes the transform to
1337 ;;; be delayed until after constraint propagation.
1338 ;;;
1339 ;;; FIXME: Now (0.6.11.44) that there are 4 variants of this (GIVE-UP,
1340 ;;; ABORT, DELAY/:OPTIMIZE, DELAY/:CONSTRAINT) and we're starting to
1341 ;;; do CASE operations on the various REASON values, it might be a
1342 ;;; good idea to go OO, representing the reasons by objects, using
1343 ;;; CLOS methods on the objects instead of CASE, and (possibly) using
1344 ;;; SIGNAL instead of THROW.
1362 ;;; Poor man's catching and resignalling
1363 ;;; Implicit %GIVE-UP macrolet will resignal the give-up "condition"
1373 (:delayed
1381 ;;; Clear any delayed transform with no reasons - these should have
1382 ;;; been tried in the last pass. Then remove the reason from the
1383 ;;; delayed transform reasons, and if any become empty then set
1384 ;;; reoptimize flags for the node. Return true if any transforms are
1385 ;;; to be retried.
1401 reoptimize))
1403 ;;; Take the lambda-expression RES, IR1 convert it in the proper
1404 ;;; environment, and then install it as the function for the call
1405 ;;; NODE. We do local call analysis so that the new function is
1406 ;;; integrated into the control flow.
1407 ;;;
1408 ;;; We require the original function source name in order to generate
1409 ;;; a meaningful debug name for the lambda we set up. (It'd be
1410 ;;; possible to do this starting from debug names as well as source
1411 ;;; names, but as of sbcl-0.7.1.5, there was no need for this
1412 ;;; generality, since source names are always known to our callers.)
1418 ;; The internal variables of a transform are not going to be
1419 ;; interesting to the debugger, so there's no sense in
1420 ;; suppressing the substitution of variables with only one use
1421 ;; (the extra variables can slow down constraint propagation).
1422 ;;
1423 ;; This needs to be done before the WITH-IR1-ENVIRONMENT-FROM-NODE,
1424 ;; so that it will bind *LEXENV* to the right environment.
1437 res
1449 t)
1458 ;;; Return T if the function is foldable and if it's marked as CALL
1459 ;;; all function arguments are FOLDABLE too.
1465 nil)
1473 args)))
1477 ;;; Replace a call to a foldable function of constant arguments with
1478 ;;; the result of evaluating the form. If there is an error during the
1479 ;;; evaluation, we give a warning and leave the call alone, making the
1480 ;;; call a :ERROR call.
1481 ;;;
1482 ;;; If there is more than one value, then we transform the call into a
1483 ;;; VALUES form.
1494 args
1495 call
1496 ;; Note: CMU CL had COMPILER-WARN here, and that
1497 ;; seems more natural, but it's probably not.
1498 ;;
1499 ;; It's especially not while bug 173 exists:
1500 ;; Expressions like
1501 ;; (COND (END
1502 ;; (UNLESS (OR UNSAFE? (<= END SIZE)))
1503 ;; ...))
1504 ;; can cause constant-folding TYPE-ERRORs (in
1505 ;; #'<=) when END can be proved to be NIL, even
1506 ;; though the code is perfectly legal and safe
1507 ;; because a NIL value of END means that the
1508 ;; #'<= will never be executed.
1509 ;;
1510 ;; Moreover, even without bug 173,
1511 ;; quite-possibly-valid code like
1512 ;; (COND ((NONINLINED-PREDICATE END)
1513 ;; (UNLESS (<= END SIZE))
1514 ;; ...))
1515 ;; (where NONINLINED-PREDICATE is something the
1516 ;; compiler can't do at compile time, but which
1517 ;; turns out to make the #'<= expression
1518 ;; unreachable when END=NIL) could cause errors
1519 ;; when the compiler tries to constant-fold (<=
1520 ;; END SIZE).
1521 ;;
1522 ;; So, with or without bug 173, it'd be
1523 ;; unnecessarily evil to do a full
1524 ;; COMPILER-WARNING (and thus return FAILURE-P=T
1525 ;; from COMPILE-FILE) for legal code, so we we
1526 ;; use a wimpier COMPILE-STYLE-WARNING instead.
1528 ;; On the other hand, for code we control, we
1529 ;; should be able to work around any bug
1530 ;; 173-related problems, and in particular we
1531 ;; want to be alerted to calls to our own
1532 ;; functions which aren't being folded away; a
1533 ;; COMPILER-WARNING is butch enough to stop the
1534 ;; SBCL build itself in its tracks.
1554 call
1558 fun-name))))))
1560 \f
1561 ;;;; local call optimization
1563 ;;; Propagate TYPE to LEAF and its REFS, marking things changed.
1564 ;;;
1565 ;;; If the leaf type is a function type, then just leave it alone, since TYPE
1566 ;;; is never going to be more specific than that (and TYPE-INTERSECTION would
1567 ;;; choke.)
1568 ;;;
1569 ;;; Also, if the type is one requiring special care don't touch it if the leaf
1570 ;;; has multiple references -- otherwise LVAR-CONSERVATIVE-TYPE is screwed.
1585 ;; KLUDGE: LET var substitution
1591 ;;; Iteration variable: exactly one SETQ of the form:
1592 ;;;
1593 ;;; (let ((var initial))
1594 ;;; ...
1595 ;;; (setq var (+ var step))
1596 ;;; ...)
1622 ;; Detect cases like (LOOP FOR 1.0 to 5.0 ...), where
1623 ;; the initial and the step are of different types,
1624 ;; and the step is less contagious.
1627 step-type))))
1666 :low low
1667 :high high
1670 "check for iteration variable reoptimization"
1680 ;;; Figure out the type of a LET variable that has sets. We compute
1681 ;;; the union of the INITIAL-TYPE and the types of all the set
1682 ;;; values and to a PROPAGATE-TO-REFS with this type.
1702 ;;; If a LET variable, find the initial value's type and do
1703 ;;; PROPAGATE-FROM-SETS. We also derive the VALUE's type as the node's
1704 ;;; type.
1720 ;;; Return true if the value of REF will always be the same (and is
1721 ;;; thus legal to substitute.)
1733 (:global-function
1740 ;;; If we have a non-set LET var with a single use, then (if possible)
1741 ;;; replace the variable reference's LVAR with the arg lvar.
1742 ;;;
1743 ;;; We change the REF to be a reference to NIL with unused value, and
1744 ;;; let it be flushed as dead code. A side effect of this substitution
1745 ;;; is to delete the variable.
1753 ;; Think about (LET ((A ...)) (IF ... A ...)): two
1754 ;; LVAR-USEs should not be met on one path. Another problem
1755 ;; is with dynamic-extent.
1758 ;; If the destinatation is dynamic extent, don't substitute unless
1759 ;; the source is as well.
1764 ;; we should not change lifetime of unknown values lvars
1771 ;; CASTs can disappear, don't substitute if
1772 ;; DEST-LVAR has other uses (this will be
1773 ;; insufficient if we have a CAST-CAST chain, but
1774 ;; works well for a single CAST)
1782 ;; While CRETURN and EXIT nodes may be known-values,
1783 ;; they have their own complications, such as
1784 ;; substitution into CRETURN may create new tail calls.
1785 nil)
1788 t))
1794 ;; Really it is (EQ (LVAR-USES LVAR) REF):
1795 t)
1800 ;; KLUDGE: it should be (TYPE-CHECK 0)
1804 ;; FIXME
1805 t)
1813 t)))
1815 ;;; Delete a LET, removing the call and bind nodes, and warning about
1816 ;;; any unreferenced variables. Note that FLUSH-DEAD-CODE will come
1817 ;;; along right away and delete the REF and then the lambda, since we
1818 ;;; flush the FUN lvar.
1833 ;;; This function is called when one of the arguments to a LET
1834 ;;; changes. We look at each changed argument. If the corresponding
1835 ;;; variable is set, then we call PROPAGATE-FROM-SETS. Otherwise, we
1836 ;;; consider substituting for the variable, and also propagate
1837 ;;; derived-type information for the arg to all the VAR's refs.
1838 ;;;
1839 ;;; Substitution is inhibited when the arg leaf's derived type isn't a
1840 ;;; subtype of the argument's leaf type. This prevents type checking
1841 ;;; from being defeated, and also ensures that the best representation
1842 ;;; for the variable can be used.
1843 ;;;
1844 ;;; Substitution of individual references is inhibited if the
1845 ;;; reference is in a different component from the home. This can only
1846 ;;; happen with closures over top level lambda vars. In such cases,
1847 ;;; the references may have already been compiled, and thus can't be
1848 ;;; retroactively modified.
1849 ;;;
1850 ;;; If all of the variables are deleted (have no references) when we
1851 ;;; are done, then we delete the LET.
1852 ;;;
1853 ;;; Note that we are responsible for clearing the LVAR-REOPTIMIZE
1854 ;;; flags.
1869 ;; (NODE-DERIVED-TYPE USE) would
1870 ;; be better -- APD, 2003-05-15
1877 t)
1880 nil)))
1881 leaf var)))
1882 t)))))
1894 ;;; This function is called when one of the args to a non-LET local
1895 ;;; call changes. For each changed argument corresponding to an unset
1896 ;;; variable, we compute the union of the types across all calls and
1897 ;;; propagate this type information to the var's refs.
1898 ;;;
1899 ;;; If the function has an entry-fun, then we don't do anything: since
1900 ;;; it has a XEP we would not discover anything.
1901 ;;;
1902 ;;; If the function is an optional-entry-point, we will just make sure
1903 ;;; &REST lists are known to be lists. Doing the regular rigamarole
1904 ;;; can erronously propagate too strict types into refs: see
1905 ;;; BUG-655203-REGRESSION in tests/compiler.pure.lisp.
1906 ;;;
1907 ;;; We can clear the LVAR-REOPTIMIZE flags for arguments in all calls
1908 ;;; corresponding to changed arguments in CALL, since the only use in
1909 ;;; IR1 optimization of the REOPTIMIZE flag for local call args is
1910 ;;; right here.
1915 ;; We can still make sure &REST is known to be a list.
1920 ;; The normal case.
1928 vars))
1944 union)))))
1947 and type in union
1951 \f
1952 ;;;; multiple values optimization
1954 ;;; Do stuff to notice a change to a MV combination node. There are
1955 ;;; two main branches here:
1956 ;;; -- If the call is local, then it is already a MV let, or should
1957 ;;; become one. Note that although all :LOCAL MV calls must eventually
1958 ;;; be converted to :MV-LETs, there can be a window when the call
1959 ;;; is local, but has not been LET converted yet. This is because
1960 ;;; the entry-point lambdas may have stray references (in other
1961 ;;; entry points) that have not been deleted yet.
1962 ;;; -- The call is full. This case is somewhat similar to the non-MV
1963 ;;; combination optimization: we propagate return type information and
1964 ;;; notice non-returning calls. We also have an optimization
1965 ;;; which tries to convert MV-CALLs into MV-binds.
1971 (:local
1979 (:full
2001 ;;; Propagate derived type info from the values lvar to the vars.
2008 n-vars)))
2010 and type in types
2017 ;;; If possible, convert a general MV call to an MV-BIND. We can do
2018 ;;; this if:
2019 ;;; -- The call has only one argument, and
2020 ;;; -- The function has a known fixed number of arguments, or
2021 ;;; -- The argument yields a known fixed number of values.
2022 ;;;
2023 ;;; What we do is change the function in the MV-CALL to be a lambda
2024 ;;; that "looks like an MV bind", which allows
2025 ;;; IR1-OPTIMIZE-MV-COMBINATION to notice that this call can be
2026 ;;; converted (the next time around.) This new lambda just calls the
2027 ;;; actual function with the MV-BIND variables as arguments. Note that
2028 ;;; this new MV bind is not let-converted immediately, as there are
2029 ;;; going to be stray references from the entry-point functions until
2030 ;;; they get deleted.
2031 ;;;
2032 ;;; In order to avoid loss of argument count checking, we only do the
2033 ;;; transformation according to a known number of expected argument if
2034 ;;; safety is unimportant. We can always convert if we know the number
2035 ;;; of actual values, since the normal call that we build will still
2036 ;;; do any appropriate argument count checking.
2037 ;;;
2038 ;;; We only attempt the transformation if the called function is a
2039 ;;; constant reference. This allows us to just splice the leaf into
2040 ;;; the new function, instead of trying to somehow bind the function
2041 ;;; expression. The leaf must be constant because we are evaluating it
2042 ;;; again in a different place. This also has the effect of squelching
2043 ;;; multiple warnings when there is an argument count error.
2065 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2066 at least ~R."
2067 total-nvals min)
2072 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2073 at most ~R."
2074 total-nvals max)
2081 min)
2100 ;;; If we see:
2101 ;;; (multiple-value-bind
2102 ;;; (x y)
2103 ;;; (values xx yy)
2104 ;;; ...)
2105 ;;; Convert to:
2106 ;;; (let ((x xx)
2107 ;;; (y yy))
2108 ;;; ...)
2109 ;;;
2110 ;;; What we actually do is convert the VALUES combination into a
2111 ;;; normal LET combination calling the original :MV-LET lambda. If
2112 ;;; there are extra args to VALUES, discard the corresponding
2113 ;;; lvars. If there are insufficient args, insert references to NIL.
2136 for prev = node-prev then ctran
2142 use ctran))
2156 ;; Propagate derived types from the VALUES call to its args:
2157 ;; transforms can leave the VALUES call with a better type
2158 ;; than its args have, so make sure not to throw that away.
2164 ;; Propagate declared types of MV-BIND variables.
2167 t)))
2169 ;;; If we see:
2170 ;;; (values-list (list x y z))
2171 ;;;
2172 ;;; Convert to:
2173 ;;; (values x y z)
2174 ;;;
2175 ;;; In implementation, this is somewhat similar to
2176 ;;; CONVERT-MV-BIND-TO-LET. We grab the args of LIST and make them
2177 ;;; args of the VALUES-LIST call, flushing the old argument lvar
2178 ;;; (allowing the LIST to be flushed.)
2179 ;;;
2180 ;;; FIXME: Thus we lose possible type assertions on (LIST ...).
2187 ;; FIXME: VALUES might not satisfy an assertion on NODE-LVAR.
2199 t)))
2201 ;;; If VALUES appears in a non-MV context, then effectively convert it
2202 ;;; to a PROG1. This allows the computation of the additional values
2203 ;;; to become dead code.
2215 val))
2216 nil))
2218 ;;; TODO:
2219 ;;; - CAST chains;
2238 ;; Vestigial exits are only introduced when eliminating a local
2239 ;; RETURN-FROM. We may delete them only when we can show that
2240 ;; there are no other code paths that use the entry LVAR that
2241 ;; are live from within the block that contained the deleted
2242 ;; EXIT (our predecessor block). The conservative version of
2243 ;; this is that there are no EXITs for any ENTRY introduced
2244 ;; between the LEXENV of the deleted EXIT and the LEXENV of the
2245 ;; target ENTRY.
2260 #+sb-xc-host context
2285 lvar)
2286 ;; Pathwise removing of CAST
2289 next-block)
2323 ;; No need to transform into an analog of
2324 ;; %COMPILE-TIME-TYPE-ERROR, %CHECK-BOUND will signal at
2325 ;; run-time and %CHECK-BOUND ir2-converter will signal at
2326 ;; compile-time if it survives further stages of ir1
2327 ;; optimization.
2330 ;; FIXME: Do it in one step.
2334 value
2340 ;; FIXME: Derived type.
2344 ',detail
2346 ;; KLUDGE: FILTER-LVAR does not work for non-returning
2347 ;; functions, so we declare the return type of
2348 ;; %COMPILE-TIME-TYPE-ERROR to be * and derive the real type
2349 ;; here.
2353 ;; FIXME: Is it necessary?