| blob | 15606bb056881624d8c77ef5d92c657d2fe5e395 |
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 (but not singleton numeric types)
32 ;;; Return the constant value for an LVAR whose only use is a constant
33 ;;; node.
38 leaf)
45 value))))
46 \f
47 ;;;; interface for obtaining results of type inference
49 ;;; Our best guess for the type of this lvar's value. Note that this
50 ;;; may be VALUES or FUNCTION type, which cannot be passed as an
51 ;;; argument to the normal type operations. See LVAR-TYPE.
52 ;;;
53 ;;; The result value is cached in the LVAR-%DERIVED-TYPE slot. If the
54 ;;; slot is true, just return that value, otherwise recompute and
55 ;;; stash the value there.
65 res))
68 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)))))
236 \f
237 ;;;; interface routines used by optimizers
247 ;;; This function is called by optimizers to indicate that something
248 ;;; interesting has happened to the value of LVAR. Optimizers must
249 ;;; make sure that they don't call for reoptimization when nothing has
250 ;;; happened, since optimization will fail to terminate.
251 ;;;
252 ;;; We clear any cached type for the lvar and set the reoptimize flags
253 ;;; on everything in sight.
263 ;; PREV may be missing.
282 ;;; Annotate NODE to indicate that its result has been proven to be
283 ;;; TYPEP to RTYPE. After IR1 conversion has happened, this is the
284 ;;; only correct way to supply information discovered about a node's
285 ;;; type. If you screw with the NODE-DERIVED-TYPE directly, then
286 ;;; information may be lost and reoptimization may not happen.
287 ;;;
288 ;;; What we do is intersect RTYPE with NODE's DERIVED-TYPE. If the
289 ;;; intersection is different from the old type, then we do a
290 ;;; REOPTIMIZE-LVAR on the NODE-LVAR.
296 initial-type)))
307 "New inferred type ~S conflicts with old type:~
308 ~% ~S~%*** possible internal error? Please report this."
311 ;; If the new type consists of only one object, replace the
312 ;; node with a constant reference.
323 ;;; This is similar to DERIVE-NODE-TYPE, but asserts that it is an
324 ;;; error for LVAR's value not to be TYPEP to TYPE. We implement it
325 ;;; splitting off DEST a new CAST node; old LVAR will deliver values
326 ;;; to CAST. If we improve the assertion, we set TYPE-CHECK and
327 ;;; TYPE-ASSERTED to guarantee that the new assertion will be checked.
336 t))))
338 \f
339 ;;;; IR1-OPTIMIZE
341 ;;; Do one forward pass over COMPONENT, deleting unreachable blocks
342 ;;; and doing IR1 optimizations. We can ignore all blocks that don't
343 ;;; have the REOPTIMIZE flag set. If COMPONENT-REOPTIMIZE is true when
344 ;;; we are done, then another iteration would be beneficial.
350 for last-block = block
353 ;; We delete blocks when there is either no predecessor or the
354 ;; block is in a lambda that has been deleted. These blocks
355 ;; would eventually be deleted by DFO recomputation, but doing
356 ;; it here immediately makes the effect available to IR1
357 ;; optimization.
363 ;; Preserve the BLOCK-SUCC invariant that almost every block has
364 ;; one successor (and a block with DELETE-P set is an acceptable
365 ;; exception).
400 ;;; Loop over the nodes in BLOCK, acting on (and clearing) REOPTIMIZE
401 ;;; flags.
402 ;;;
403 ;;; Note that although they are cleared here, REOPTIMIZE flags might
404 ;;; still be set upon return from this function, meaning that further
405 ;;; optimization is wanted (as a consequence of optimizations we did).
408 ;; We clear the node and block REOPTIMIZE flags before doing the
409 ;; optimization, not after. This ensures that the node or block will
410 ;; be reoptimized if necessary.
414 ;; As above, we clear the node REOPTIMIZE flag before optimizing.
419 ;; With a COMBINATION, we call PROPAGATE-FUN-CHANGE whenever
420 ;; the function changes, and call IR1-OPTIMIZE-COMBINATION if
421 ;; any argument changes.
426 ;; KLUDGE: We leave the NODE-OPTIMIZE flag set going into
427 ;; IR1-OPTIMIZE-RETURN, since IR1-OPTIMIZE-RETURN wants to
428 ;; clear the flag itself. -- WHN 2002-02-02, quoting original
429 ;; CMU CL comments
435 ;; With an EXIT, we derive the node's type from the VALUE's
436 ;; type.
441 ;; PROPAGATE-FROM-SETS can do a better job if NODE-REOPTIMIZE
442 ;; is accurate till the node actually has been reoptimized.
450 ;;; Try to join with a successor block. If we succeed, we return true,
451 ;;; otherwise false.
458 ;; the successor has more than one predecessor;
460 ;; the successor is the current block (infinite loop);
462 ;; the next block has a different cleanup, and thus
463 ;; we may want to insert cleanup code between the
464 ;; two blocks at some point;
467 ;; the next block has a different home lambda, and
468 ;; thus the control transfer is a non-local exit.
471 ;; Stack analysis phase wants ENTRY to start a block...
477 ;; ... and a DX-allocator to end a block.
479 ;; FIXME: This is a partial workaround for bug 303.
481 nil)
484 t)))))
486 ;;; Join together two blocks. The code in BLOCK2 is moved into BLOCK1
487 ;;; and BLOCK2 is deleted from the DFO. We combine the optimize flags
488 ;;; for the two blocks so that any indicated optimization gets done.
521 ;;; Delete any nodes in BLOCK whose value is unused and which have no
522 ;;; side effects. We can delete sets of lexical variables when the set
523 ;;; variable has no references.
569 victim)
570 \f
571 ;;;; local call return type propagation
573 ;;; This function is called on RETURN nodes that have their REOPTIMIZE
574 ;;; flag set. It iterates over the uses of the RESULT, looking for
575 ;;; interesting stuff to update the TAIL-SET. If a use isn't a local
576 ;;; call, then we union its type together with the types of other such
577 ;;; uses. We assign to the RETURN-RESULT-TYPE the intersection of this
578 ;;; type with the RESULT's asserted type. We can make this
579 ;;; intersection now (potentially before type checking) because this
580 ;;; assertion on the result will eventually be checked (if
581 ;;; appropriate.)
582 ;;;
583 ;;; We call MAYBE-CONVERT-TAIL-LOCAL-CALL on each local non-MV
584 ;;; combination, which may change the successor of the call to be the
585 ;;; called function, and if so, checks if the call can become an
586 ;;; assignment. If we convert to an assignment, we abort, since the
587 ;;; RETURN has been deleted.
606 ;; (values-type-intersection
607 ;; (continuation-asserted-type result) ; FIXME -- APD, 2002-01-26
609 ;; )
610 ))
612 nil)
614 ;;; Do stuff to realize that something has changed about the value
615 ;;; delivered to a return node. Since we consider the return values of
616 ;;; all functions in the tail set to be equivalent, this amounts to
617 ;;; bringing the entire tail set up to date. We iterate over the
618 ;;; returns for all the functions in the tail set, reanalyzing them
619 ;;; all (not treating NODE specially.)
620 ;;;
621 ;;; When we are done, we check whether the new type is different from
622 ;;; the old TAIL-SET-TYPE. If so, we set the type and also reoptimize
623 ;;; all the lvars for references to functions in the tail set. This
624 ;;; will cause IR1-OPTIMIZE-COMBINATION to derive the new type as the
625 ;;; results of the calls.
629 :restart
649 \f
650 ;;;; IF optimization
652 ;;; Utility: return T if both argument cblocks are equivalent. For now,
653 ;;; detect only blocks that read the same leaf into the same lvar, and
654 ;;; continue to the same block.
667 ;;; Check whether the predicate is known to be true or false,
668 ;;; deleting the IF node in favor of the appropriate branch when this
669 ;;; is the case.
670 ;;; Similarly, when both branches are equivalent, branch directly to either
671 ;;; of them.
672 ;;; Also, if the test has multiple uses, replicate the node when possible...
673 ;;; in fact, splice in direct jumps to the right branch if possible.
685 alternative)
687 consequent)
690 alternative))))
698 ;; When we know that we only have a single successor, kill the victim
699 ;; ... unless the victim and the remaining successor are the same.
708 ;; When if/if conversion would leave (if ... (if nil ...)) or
709 ;; (if ... (if not-nil ...)), splice the correct successor right
710 ;; in.
729 ;; only one use left. Just kill the now-useless
730 ;; branch to avoid spurious code deletion notes.
733 node test block
739 ;; Finally, duplicate EQ-nil tests
746 ;; leave the last use as is, instead of replacing
747 ;; the (singly-referenced) CIF node with a duplicate.
750 ;;; Create a new copy of an IF node that tests the value of the node
751 ;;; USE. The test must have >1 use, and must be immediately used by
752 ;;; USE. NODE must be the only node in its block (implying that
753 ;;; block-start = if-test).
754 ;;;
755 ;;; This optimization has an effect semantically similar to the
756 ;;; source-to-source transformation:
757 ;;; (IF (IF A B C) D E) ==>
758 ;;; (IF A (IF B D E) (IF C D E))
759 ;;;
760 ;;; We clobber the NODE-SOURCE-PATH of both the original and the new
761 ;;; node so that dead code deletion notes will definitely not consider
762 ;;; either node to be part of the original source. One node might
763 ;;; become unreachable, resulting in a spurious note.
775 :consequent cblock
797 \f
798 ;;;; exit IR1 optimization
800 ;;; This function attempts to delete an exit node, returning true if
801 ;;; it deletes the block as a consequence:
802 ;;; -- If the exit is degenerate (has no ENTRY), then we don't do
803 ;;; anything, since there is nothing to be done.
804 ;;; -- If the exit node and its ENTRY have the same home lambda then
805 ;;; we know the exit is local, and can delete the exit. We change
806 ;;; uses of the Exit-Value to be uses of the original lvar,
807 ;;; then unlink the node. If the exit is to a TR context, then we
808 ;;; must do MERGE-TAIL-SETS on any local calls which delivered
809 ;;; their value to this exit.
810 ;;; -- If there is no value (as in a GO), then we skip the value
811 ;;; semantics.
812 ;;;
813 ;;; This function is also called by environment analysis, since it
814 ;;; wants all exits to be optimized even if normal optimization was
815 ;;; omitted.
825 ;; We can't simply use DELETE-FILTER to unlink the node
826 ;; and substitute some LVAR magic, as this can confuse the
827 ;; stack analysis if there's another EXIT to the same
828 ;; continuation. Instead, we fabricate a new block (in
829 ;; the same lexenv as the ENTRY, so it can't be merged
830 ;; backwards), insert a gimmicked CAST node to link up the
831 ;; LVAR holding the value being returned to the LVAR which
832 ;; is expecting to accept the value, thus placing the
833 ;; return value where it needs to be while still providing
834 ;; the hook required for stack analysis.
835 ;; -- AJB, 2014-Mar-03
840 (%make-cast
843 :value value
847 ;; We only expect a single successor to EXIT-BLOCK,
848 ;; because it contains an EXIT node (which must end its
849 ;; block) and the only blocks that have more than once
850 ;; successor are those with IF nodes (which also must
851 ;; end their blocks). Still, just to be sure, we use a
852 ;; construct that guarantees an error if this
853 ;; expectation is violated.
858 ;; Finish creating the new block.
862 ;; Link the new block into the control sequence.
867 ;; Finish re-pointing the value-holding LVAR to the
868 ;; CAST node.
873 ;; Register the CAST node as providing a value to the
874 ;; LVAR for the continuation.
878 ;; Remove the EXIT node.
881 ;; And, because we created a new block, we need to
882 ;; force component reanalysis (to assign a DFO number
883 ;; to the block if nothing else).
887 \f
888 ;;;; combination IR1 optimization
890 ;;; Report as we try each transform?
891 #!+sb-show
899 "The return value of ~A should not be discarded."
902 ;;; Do IR1 optimizations on a COMBINATION node.
912 (:local
917 (:error
921 (:full
944 ;; Check against the DEFINED-TYPE unless TYPE is already good.
954 (:known
962 ;; If somebody is really sure that they want to modify
963 ;; constants, let them.
976 ;; KLUDGE: The next test could be made more sensitive,
977 ;; only suppressing constant-folding of functions with
978 ;; CALL attributes when they're actually passed
979 ;; function arguments. -- WHN 19990918
999 ;; First give the VM a peek at the call
1003 (:direct
1004 ;; The VM knows how to handle this.
1005 )
1006 (:transform
1007 ;; The VM mostly knows how to handle this. We need
1008 ;; to massage the call slightly, though.
1011 ;; Let transforms have a crack at it.
1013 #!+sb-show
1019 #!+sb-show
1034 ;;; If NODE doesn't return (i.e. return type is NIL), then terminate
1035 ;;; the block there, and link it to the component tail.
1036 ;;;
1037 ;;; Except when called during IR1 convertion, we delete the
1038 ;;; continuation if it has no other uses. (If it does have other uses,
1039 ;;; we reoptimize.)
1040 ;;;
1041 ;;; Termination on the basis of a continuation type is
1042 ;;; inhibited when:
1043 ;;; -- The continuation is deleted (hence the assertion is spurious), or
1044 ;;; -- We are in IR1 conversion (where THE assertions are subject to
1045 ;;; weakening.) FIXME: Now THE assertions are not weakened, but new
1046 ;;; uses can(?) be added later. -- APD, 2003-07-17
1047 ;;;
1048 ;;; Why do we need to consider LVAR type? -- APD, 2003-07-30
1059 ;; Even if the combination will never return, don't terminate if this
1060 ;; is the tail call of a XEP: doing that would inhibit TCO.
1087 t))))
1089 ;;; This is called both by IR1 conversion and IR1 optimization when
1090 ;;; they have verified the type signature for the call, and are
1091 ;;; wondering if something should be done to special-case the call. If
1092 ;;; CALL is a call to a global function, then see whether it defined
1093 ;;; or known:
1094 ;;; -- If a DEFINED-FUN should be inline expanded, then convert
1095 ;;; the expansion and change the call to call it. Expansion is
1096 ;;; enabled if :INLINE or if SPACE=0. If the FUNCTIONAL slot is
1097 ;;; true, we never expand, since this function has already been
1098 ;;; converted. Local call analysis will duplicate the definition
1099 ;;; if necessary. We claim that the parent form is LABELS for
1100 ;;; context declarations, since we don't want it to be considered
1101 ;;; a real global function.
1102 ;;; -- If it is a known function, mark it as such by setting the KIND.
1103 ;;;
1104 ;;; We return the leaf referenced (NIL if not a leaf) and the
1105 ;;; FUN-INFO assigned.
1129 ;; Inline: if the function has already been converted at another call
1130 ;; site in this component, we point this REF to the functional. If not,
1131 ;; we convert the expansion.
1132 ;;
1133 ;; For :INLINE case local call analysis will copy the expansion later,
1134 ;; but for :MAYBE-INLINE and NIL cases we only get one copy of the
1135 ;; expansion per component.
1136 ;;
1137 ;; FIXME: We also convert in :INLINE & FUNCTIONAL-KIND case below. What
1138 ;; is it for?
1142 name
1144 leaf
1145 inlinep
1147 ;; Allow backward references to this function from following
1148 ;; forms. (Reused only if policy matches.)
1154 ;; Convert.
1160 ;; If we've already converted, change ref to the converted
1161 ;; functional.
1173 ;;; Check whether CALL satisfies TYPE. If so, apply the type to the
1174 ;;; call, and do MAYBE-TERMINATE-BLOCK and return the values of
1175 ;;; RECOGNIZE-KNOWN-CALL. If an error, set the combination kind and
1176 ;;; return NIL, NIL. If the type is just FUNCTION, then skip the
1177 ;;; syntax check, arg/result type processing, but still call
1178 ;;; RECOGNIZE-KNOWN-CALL, since the call might be to a known lambda,
1179 ;;; and that checking is done by local call analysis.
1188 ;; Using the defined-type too early is a bit of a waste: during
1189 ;; conversion we cannot use the untrusted ASSERT-CALL-TYPE, etc.
1194 ;; Don't validate multiple times against the same leaf --
1195 ;; it doesn't add any information, but may generate the same warning
1196 ;; multiple times.
1200 :result-test nil
1202 #'compiler-warn
1205 same-file-p)
1211 :result-test nil
1221 ;;; This is called by IR1-OPTIMIZE when the function for a call has
1222 ;;; changed. If the call is local, we try to LET-convert it, and
1223 ;;; derive the result type. If it is a :FULL call, we validate it
1224 ;;; against the type, which recognizes known calls, does inline
1225 ;;; expansion, etc. If a call to a predicate in a non-conditional
1226 ;;; position or to a function with a source transform, then we
1227 ;;; reconvert the form to give IR1 another chance.
1234 (:local
1239 (:full
1247 call))
1255 predicate)
1269 \f
1270 ;;;; known function optimization
1272 ;;; Add a failed optimization note to FAILED-OPTIMZATIONS for NODE,
1273 ;;; FUN and ARGS. If there is already a note for NODE and TRANSFORM,
1274 ;;; replace it, otherwise add a new one.
1285 ;;; Attempt to transform NODE using TRANSFORM-FUNCTION, subject to the
1286 ;;; call type constraint TRANSFORM-TYPE. If we are inhibited from
1287 ;;; doing the transform for some reason and FLAME is true, then we
1288 ;;; make a note of the message in FAILED-OPTIMIZATIONS for IR1
1289 ;;; finalize to pick up. We return true if the transform failed, and
1290 ;;; thus further transformation should be attempted. We return false
1291 ;;; if either the transform succeeded or was aborted.
1311 (:none
1313 nil)
1314 (:aborted
1319 nil)
1320 (:failure
1326 t)
1327 (:delayed
1329 nil))))
1332 type
1336 t)
1338 t))))
1340 ;;; When we don't like an IR1 transform, we throw the severity/reason
1341 ;;; and args.
1342 ;;;
1343 ;;; GIVE-UP-IR1-TRANSFORM is used to throw out of an IR1 transform,
1344 ;;; aborting this attempt to transform the call, but admitting the
1345 ;;; possibility that this or some other transform will later succeed.
1346 ;;; If arguments are supplied, they are format arguments for an
1347 ;;; efficiency note.
1348 ;;;
1349 ;;; ABORT-IR1-TRANSFORM is used to throw out of an IR1 transform and
1350 ;;; force a normal call to the function at run time. No further
1351 ;;; optimizations will be attempted.
1352 ;;;
1353 ;;; DELAY-IR1-TRANSFORM is used to throw out of an IR1 transform, and
1354 ;;; delay the transform on the node until later. REASONS specifies
1355 ;;; when the transform will be later retried. The :OPTIMIZE reason
1356 ;;; causes the transform to be delayed until after the current IR1
1357 ;;; optimization pass. The :CONSTRAINT reason causes the transform to
1358 ;;; be delayed until after constraint propagation.
1359 ;;;
1360 ;;; FIXME: Now (0.6.11.44) that there are 4 variants of this (GIVE-UP,
1361 ;;; ABORT, DELAY/:OPTIMIZE, DELAY/:CONSTRAINT) and we're starting to
1362 ;;; do CASE operations on the various REASON values, it might be a
1363 ;;; good idea to go OO, representing the reasons by objects, using
1364 ;;; CLOS methods on the objects instead of CASE, and (possibly) using
1365 ;;; SIGNAL instead of THROW.
1382 ;;; Poor man's catching and resignalling
1383 ;;; Implicit GIVE-UP macrolet will resignal the give-up "condition"
1393 (:delayed
1401 ;;; Clear any delayed transform with no reasons - these should have
1402 ;;; been tried in the last pass. Then remove the reason from the
1403 ;;; delayed transform reasons, and if any become empty then set
1404 ;;; reoptimize flags for the node. Return true if any transforms are
1405 ;;; to be retried.
1421 reoptimize))
1423 ;;; Take the lambda-expression RES, IR1 convert it in the proper
1424 ;;; environment, and then install it as the function for the call
1425 ;;; NODE. We do local call analysis so that the new function is
1426 ;;; integrated into the control flow.
1427 ;;;
1428 ;;; We require the original function source name in order to generate
1429 ;;; a meaningful debug name for the lambda we set up. (It'd be
1430 ;;; possible to do this starting from debug names as well as source
1431 ;;; names, but as of sbcl-0.7.1.5, there was no need for this
1432 ;;; generality, since source names are always known to our callers.)
1438 ;; The internal variables of a transform are not going to be
1439 ;; interesting to the debugger, so there's no sense in
1440 ;; suppressing the substitution of variables with only one use
1441 ;; (the extra variables can slow down constraint propagation).
1442 ;;
1443 ;; This needs to be done before the WITH-IR1-ENVIRONMENT-FROM-NODE,
1444 ;; so that it will bind *LEXENV* to the right environment.
1457 res
1466 ;;; Replace a call to a foldable function of constant arguments with
1467 ;;; the result of evaluating the form. If there is an error during the
1468 ;;; evaluation, we give a warning and leave the call alone, making the
1469 ;;; call a :ERROR call.
1470 ;;;
1471 ;;; If there is more than one value, then we transform the call into a
1472 ;;; VALUES form.
1478 args
1479 call
1480 ;; Note: CMU CL had COMPILER-WARN here, and that
1481 ;; seems more natural, but it's probably not.
1482 ;;
1483 ;; It's especially not while bug 173 exists:
1484 ;; Expressions like
1485 ;; (COND (END
1486 ;; (UNLESS (OR UNSAFE? (<= END SIZE)))
1487 ;; ...))
1488 ;; can cause constant-folding TYPE-ERRORs (in
1489 ;; #'<=) when END can be proved to be NIL, even
1490 ;; though the code is perfectly legal and safe
1491 ;; because a NIL value of END means that the
1492 ;; #'<= will never be executed.
1493 ;;
1494 ;; Moreover, even without bug 173,
1495 ;; quite-possibly-valid code like
1496 ;; (COND ((NONINLINED-PREDICATE END)
1497 ;; (UNLESS (<= END SIZE))
1498 ;; ...))
1499 ;; (where NONINLINED-PREDICATE is something the
1500 ;; compiler can't do at compile time, but which
1501 ;; turns out to make the #'<= expression
1502 ;; unreachable when END=NIL) could cause errors
1503 ;; when the compiler tries to constant-fold (<=
1504 ;; END SIZE).
1505 ;;
1506 ;; So, with or without bug 173, it'd be
1507 ;; unnecessarily evil to do a full
1508 ;; COMPILER-WARNING (and thus return FAILURE-P=T
1509 ;; from COMPILE-FILE) for legal code, so we we
1510 ;; use a wimpier COMPILE-STYLE-WARNING instead.
1512 ;; On the other hand, for code we control, we
1513 ;; should be able to work around any bug
1514 ;; 173-related problems, and in particular we
1515 ;; want to be alerted to calls to our own
1516 ;; functions which aren't being folded away; a
1517 ;; COMPILER-WARNING is butch enough to stop the
1518 ;; SBCL build itself in its tracks.
1538 call
1542 fun-name))))))
1544 \f
1545 ;;;; local call optimization
1547 ;;; Propagate TYPE to LEAF and its REFS, marking things changed.
1548 ;;;
1549 ;;; If the leaf type is a function type, then just leave it alone, since TYPE
1550 ;;; is never going to be more specific than that (and TYPE-INTERSECTION would
1551 ;;; choke.)
1552 ;;;
1553 ;;; Also, if the type is one requiring special care don't touch it if the leaf
1554 ;;; has multiple references -- otherwise LVAR-CONSERVATIVE-TYPE is screwed.
1569 ;; KLUDGE: LET var substitution
1575 ;;; Iteration variable: exactly one SETQ of the form:
1576 ;;;
1577 ;;; (let ((var initial))
1578 ;;; ...
1579 ;;; (setq var (+ var step))
1580 ;;; ...)
1606 ;; Detect cases like (LOOP FOR 1.0 to 5.0 ...), where
1607 ;; the initial and the step are of different types,
1608 ;; and the step is less contagious.
1611 step-type))))
1650 :low low
1651 :high high
1654 "check for iteration variable reoptimization"
1664 ;;; Figure out the type of a LET variable that has sets. We compute
1665 ;;; the union of the INITIAL-TYPE and the types of all the set
1666 ;;; values and to a PROPAGATE-TO-REFS with this type.
1686 ;;; If a LET variable, find the initial value's type and do
1687 ;;; PROPAGATE-FROM-SETS. We also derive the VALUE's type as the node's
1688 ;;; type.
1704 ;;; Return true if the value of REF will always be the same (and is
1705 ;;; thus legal to substitute.)
1717 (:global-function
1724 ;;; If we have a non-set LET var with a single use, then (if possible)
1725 ;;; replace the variable reference's LVAR with the arg lvar.
1726 ;;;
1727 ;;; We change the REF to be a reference to NIL with unused value, and
1728 ;;; let it be flushed as dead code. A side effect of this substitution
1729 ;;; is to delete the variable.
1737 ;; Think about (LET ((A ...)) (IF ... A ...)): two
1738 ;; LVAR-USEs should not be met on one path. Another problem
1739 ;; is with dynamic-extent.
1742 ;; If the destinatation is dynamic extent, don't substitute unless
1743 ;; the source is as well.
1748 ;; we should not change lifetime of unknown values lvars
1760 ;; While CRETURN and EXIT nodes may be known-values,
1761 ;; they have their own complications, such as
1762 ;; substitution into CRETURN may create new tail calls.
1763 nil)
1766 t))
1772 ;; Really it is (EQ (LVAR-USES LVAR) REF):
1773 t)
1778 ;; KLUDGE: it should be (TYPE-CHECK 0)
1782 ;; FIXME
1783 t)
1791 t)))
1793 ;;; Delete a LET, removing the call and bind nodes, and warning about
1794 ;;; any unreferenced variables. Note that FLUSH-DEAD-CODE will come
1795 ;;; along right away and delete the REF and then the lambda, since we
1796 ;;; flush the FUN lvar.
1811 ;;; This function is called when one of the arguments to a LET
1812 ;;; changes. We look at each changed argument. If the corresponding
1813 ;;; variable is set, then we call PROPAGATE-FROM-SETS. Otherwise, we
1814 ;;; consider substituting for the variable, and also propagate
1815 ;;; derived-type information for the arg to all the VAR's refs.
1816 ;;;
1817 ;;; Substitution is inhibited when the arg leaf's derived type isn't a
1818 ;;; subtype of the argument's leaf type. This prevents type checking
1819 ;;; from being defeated, and also ensures that the best representation
1820 ;;; for the variable can be used.
1821 ;;;
1822 ;;; Substitution of individual references is inhibited if the
1823 ;;; reference is in a different component from the home. This can only
1824 ;;; happen with closures over top level lambda vars. In such cases,
1825 ;;; the references may have already been compiled, and thus can't be
1826 ;;; retroactively modified.
1827 ;;;
1828 ;;; If all of the variables are deleted (have no references) when we
1829 ;;; are done, then we delete the LET.
1830 ;;;
1831 ;;; Note that we are responsible for clearing the LVAR-REOPTIMIZE
1832 ;;; flags.
1847 ;; (NODE-DERIVED-TYPE USE) would
1848 ;; be better -- APD, 2003-05-15
1855 t)
1858 nil)))
1859 leaf var)))
1860 t)))))
1872 ;;; This function is called when one of the args to a non-LET local
1873 ;;; call changes. For each changed argument corresponding to an unset
1874 ;;; variable, we compute the union of the types across all calls and
1875 ;;; propagate this type information to the var's refs.
1876 ;;;
1877 ;;; If the function has an entry-fun, then we don't do anything: since
1878 ;;; it has a XEP we would not discover anything.
1879 ;;;
1880 ;;; If the function is an optional-entry-point, we will just make sure
1881 ;;; &REST lists are known to be lists. Doing the regular rigamarole
1882 ;;; can erronously propagate too strict types into refs: see
1883 ;;; BUG-655203-REGRESSION in tests/compiler.pure.lisp.
1884 ;;;
1885 ;;; We can clear the LVAR-REOPTIMIZE flags for arguments in all calls
1886 ;;; corresponding to changed arguments in CALL, since the only use in
1887 ;;; IR1 optimization of the REOPTIMIZE flag for local call args is
1888 ;;; right here.
1893 ;; We can still make sure &REST is known to be a list.
1898 ;; The normal case.
1906 vars))
1922 union)))))
1925 and type in union
1929 \f
1930 ;;;; multiple values optimization
1932 ;;; Do stuff to notice a change to a MV combination node. There are
1933 ;;; two main branches here:
1934 ;;; -- If the call is local, then it is already a MV let, or should
1935 ;;; become one. Note that although all :LOCAL MV calls must eventually
1936 ;;; be converted to :MV-LETs, there can be a window when the call
1937 ;;; is local, but has not been LET converted yet. This is because
1938 ;;; the entry-point lambdas may have stray references (in other
1939 ;;; entry points) that have not been deleted yet.
1940 ;;; -- The call is full. This case is somewhat similar to the non-MV
1941 ;;; combination optimization: we propagate return type information and
1942 ;;; notice non-returning calls. We also have an optimization
1943 ;;; which tries to convert MV-CALLs into MV-binds.
1949 (:local
1957 (:full
1979 ;;; Propagate derived type info from the values lvar to the vars.
1986 n-vars)))
1988 and type in types
1995 ;;; If possible, convert a general MV call to an MV-BIND. We can do
1996 ;;; this if:
1997 ;;; -- The call has only one argument, and
1998 ;;; -- The function has a known fixed number of arguments, or
1999 ;;; -- The argument yields a known fixed number of values.
2000 ;;;
2001 ;;; What we do is change the function in the MV-CALL to be a lambda
2002 ;;; that "looks like an MV bind", which allows
2003 ;;; IR1-OPTIMIZE-MV-COMBINATION to notice that this call can be
2004 ;;; converted (the next time around.) This new lambda just calls the
2005 ;;; actual function with the MV-BIND variables as arguments. Note that
2006 ;;; this new MV bind is not let-converted immediately, as there are
2007 ;;; going to be stray references from the entry-point functions until
2008 ;;; they get deleted.
2009 ;;;
2010 ;;; In order to avoid loss of argument count checking, we only do the
2011 ;;; transformation according to a known number of expected argument if
2012 ;;; safety is unimportant. We can always convert if we know the number
2013 ;;; of actual values, since the normal call that we build will still
2014 ;;; do any appropriate argument count checking.
2015 ;;;
2016 ;;; We only attempt the transformation if the called function is a
2017 ;;; constant reference. This allows us to just splice the leaf into
2018 ;;; the new function, instead of trying to somehow bind the function
2019 ;;; expression. The leaf must be constant because we are evaluating it
2020 ;;; again in a different place. This also has the effect of squelching
2021 ;;; multiple warnings when there is an argument count error.
2043 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2044 at least ~R."
2045 total-nvals min)
2050 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2051 at most ~R."
2052 total-nvals max)
2059 min)
2078 ;;; If we see:
2079 ;;; (multiple-value-bind
2080 ;;; (x y)
2081 ;;; (values xx yy)
2082 ;;; ...)
2083 ;;; Convert to:
2084 ;;; (let ((x xx)
2085 ;;; (y yy))
2086 ;;; ...)
2087 ;;;
2088 ;;; What we actually do is convert the VALUES combination into a
2089 ;;; normal LET combination calling the original :MV-LET lambda. If
2090 ;;; there are extra args to VALUES, discard the corresponding
2091 ;;; lvars. If there are insufficient args, insert references to NIL.
2114 for prev = node-prev then ctran
2120 use ctran))
2134 ;; Propagate derived types from the VALUES call to its args:
2135 ;; transforms can leave the VALUES call with a better type
2136 ;; than its args have, so make sure not to throw that away.
2142 ;; Propagate declared types of MV-BIND variables.
2145 t)))
2147 ;;; If we see:
2148 ;;; (values-list (list x y z))
2149 ;;;
2150 ;;; Convert to:
2151 ;;; (values x y z)
2152 ;;;
2153 ;;; In implementation, this is somewhat similar to
2154 ;;; CONVERT-MV-BIND-TO-LET. We grab the args of LIST and make them
2155 ;;; args of the VALUES-LIST call, flushing the old argument lvar
2156 ;;; (allowing the LIST to be flushed.)
2157 ;;;
2158 ;;; FIXME: Thus we lose possible type assertions on (LIST ...).
2165 ;; FIXME: VALUES might not satisfy an assertion on NODE-LVAR.
2177 t)))
2179 ;;; If VALUES appears in a non-MV context, then effectively convert it
2180 ;;; to a PROG1. This allows the computation of the additional values
2181 ;;; to become dead code.
2193 val))
2194 nil))
2196 ;;; TODO:
2197 ;;; - CAST chains;
2212 ;; Vestigial exits are only introduced when eliminating a local
2213 ;; RETURN-FROM. We may delete them only when we can show that
2214 ;; there are no other code paths that use the entry LVAR that
2215 ;; are live from within the block that contained the deleted
2216 ;; EXIT (our predecessor block). The conservative version of
2217 ;; this is that there are no EXITs for any ENTRY introduced
2218 ;; between the LEXENV of the deleted EXIT and the LEXENV of the
2219 ;; target ENTRY.
2234 #+sb-xc-host context
2250 lvar)
2251 ;; Pathwise removing of CAST
2254 next-block)
2282 ;; FIXME: Do it in one step.
2286 value
2292 ;; FIXME: Derived type.
2296 ',detail
2298 ;; KLUDGE: FILTER-LVAR does not work for non-returning
2299 ;; functions, so we declare the return type of
2300 ;; %COMPILE-TIME-TYPE-ERROR to be * and derive the real type
2301 ;; here.
2305 ;; FIXME: Is it necessary?