| blob | 5bba1fcfd50db49999c223c468e7138a07f56e51 |
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 ;;; Return the constant value for an LVAR whose only use is a constant
32 ;;; node.
37 leaf)
44 value))))
45 \f
46 ;;;; interface for obtaining results of type inference
48 ;;; Our best guess for the type of this lvar's value. Note that this
49 ;;; may be VALUES or FUNCTION type, which cannot be passed as an
50 ;;; argument to the normal type operations. See LVAR-TYPE.
51 ;;;
52 ;;; The result value is cached in the LVAR-%DERIVED-TYPE slot. If the
53 ;;; slot is true, just return that value, otherwise recompute and
54 ;;; stash the value there.
64 res))
67 res)))
80 ;;; Return the derived type for LVAR's first value. This is guaranteed
81 ;;; not to be a VALUES or FUNCTION type.
86 ;;; LVAR-CONSERVATIVE-TYPE
87 ;;;
88 ;;; Certain types refer to the contents of an object, which can
89 ;;; change without type derivation noticing: CONS types and ARRAY
90 ;;; types suffer from this:
91 ;;;
92 ;;; (let ((x (the (cons fixnum fixnum) (cons a b))))
93 ;;; (setf (car x) c)
94 ;;; (+ (car x) (cdr x)))
95 ;;;
96 ;;; Python doesn't realize that the SETF CAR can change the type of X -- so we
97 ;;; cannot use LVAR-TYPE which gets the derived results. Worse, still, instead
98 ;;; of (SETF CAR) we might have a call to a user-defined function FOO which
99 ;;; does the same -- so there is no way to use the derived information in
100 ;;; general.
101 ;;;
102 ;;; So, the conservative option is to use the derived type if the leaf has
103 ;;; only a single ref -- in which case there cannot be a prior call that
104 ;;; mutates it. Otherwise we use the declared type or punt to the most general
105 ;;; type we know to be correct for sure.
109 ;; Recompute using NODE-CONSERVATIVE-TYPE instead of derived type if
110 ;; necessary -- picking off some easy cases up front.
112 ;; Can't use CSUBTYPEP!
115 derived-type)
119 derived-type)
125 derived-type)
129 derived-type))))
142 derived-values-type))
143 derived-values-type)))
149 type)
155 ;; ADJUST-ARRAY may change dimensions, but rank stays same.
158 old
160 ;; Complexity cannot change.
162 ;; Element type cannot change.
165 ;; Simple arrays cannot change at all.
166 type))
168 ;; Conservative union type is an union of conservative types.
173 ;; Catch-all.
174 ;;
175 ;; If the type contains some CONS types, the conservative type contains all
176 ;; of them.
179 ;; Similarly for non-simple arrays -- it should be possible to preserve
180 ;; more information here, but really...
184 type)))
188 ;; Excluding T is necessary, because we do want type derivation to
189 ;; be able to narrow it down in case someone (most like a macro-expansion...)
190 ;; actually declares something as having type T.
191 nil)
193 ;; Covered by the next case as well, but this is a quick test.
194 t)
196 t)))
198 ;;; If LVAR is an argument of a function, return a type which the
199 ;;; function checks LVAR for.
208 ;; TODO: MV-COMBINATION
216 ;; FUN-TYPE might be (AND FUNCTION (SATISFIES ...)).
229 dest)))))
235 \f
236 ;;;; interface routines used by optimizers
246 ;;; This function is called by optimizers to indicate that something
247 ;;; interesting has happened to the value of LVAR. Optimizers must
248 ;;; make sure that they don't call for reoptimization when nothing has
249 ;;; happened, since optimization will fail to terminate.
250 ;;;
251 ;;; We clear any cached type for the lvar and set the reoptimize flags
252 ;;; on everything in sight.
262 ;; PREV may be missing.
281 ;;; Annotate NODE to indicate that its result has been proven to be
282 ;;; TYPEP to RTYPE. After IR1 conversion has happened, this is the
283 ;;; only correct way to supply information discovered about a node's
284 ;;; type. If you screw with the NODE-DERIVED-TYPE directly, then
285 ;;; information may be lost and reoptimization may not happen.
286 ;;;
287 ;;; What we do is intersect RTYPE with NODE's DERIVED-TYPE. If the
288 ;;; intersection is different from the old type, then we do a
289 ;;; REOPTIMIZE-LVAR on the NODE-LVAR.
295 initial-type)))
306 "New inferred type ~S conflicts with old type:~
307 ~% ~S~%*** possible internal error? Please report this."
310 ;; If the new type consists of only one object, replace the
311 ;; node with a constant reference.
322 ;;; This is similar to DERIVE-NODE-TYPE, but asserts that it is an
323 ;;; error for LVAR's value not to be TYPEP to TYPE. We implement it
324 ;;; splitting off DEST a new CAST node; old LVAR will deliver values
325 ;;; to CAST. If we improve the assertion, we set TYPE-CHECK and
326 ;;; TYPE-ASSERTED to guarantee that the new assertion will be checked.
335 t))))
337 \f
338 ;;;; IR1-OPTIMIZE
340 ;;; Do one forward pass over COMPONENT, deleting unreachable blocks
341 ;;; and doing IR1 optimizations. We can ignore all blocks that don't
342 ;;; have the REOPTIMIZE flag set. If COMPONENT-REOPTIMIZE is true when
343 ;;; we are done, then another iteration would be beneficial.
349 for last-block = block
352 ;; We delete blocks when there is either no predecessor or the
353 ;; block is in a lambda that has been deleted. These blocks
354 ;; would eventually be deleted by DFO recomputation, but doing
355 ;; it here immediately makes the effect available to IR1
356 ;; optimization.
362 ;; Preserve the BLOCK-SUCC invariant that almost every block has
363 ;; one successor (and a block with DELETE-P set is an acceptable
364 ;; exception).
399 ;;; Loop over the nodes in BLOCK, acting on (and clearing) REOPTIMIZE
400 ;;; flags.
401 ;;;
402 ;;; Note that although they are cleared here, REOPTIMIZE flags might
403 ;;; still be set upon return from this function, meaning that further
404 ;;; optimization is wanted (as a consequence of optimizations we did).
407 ;; We clear the node and block REOPTIMIZE flags before doing the
408 ;; optimization, not after. This ensures that the node or block will
409 ;; be reoptimized if necessary.
413 ;; As above, we clear the node REOPTIMIZE flag before optimizing.
418 ;; With a COMBINATION, we call PROPAGATE-FUN-CHANGE whenever
419 ;; the function changes, and call IR1-OPTIMIZE-COMBINATION if
420 ;; any argument changes.
425 ;; KLUDGE: We leave the NODE-OPTIMIZE flag set going into
426 ;; IR1-OPTIMIZE-RETURN, since IR1-OPTIMIZE-RETURN wants to
427 ;; clear the flag itself. -- WHN 2002-02-02, quoting original
428 ;; CMU CL comments
434 ;; With an EXIT, we derive the node's type from the VALUE's
435 ;; type.
440 ;; PROPAGATE-FROM-SETS can do a better job if NODE-REOPTIMIZE
441 ;; is accurate till the node actually has been reoptimized.
449 ;;; Try to join with a successor block. If we succeed, we return true,
450 ;;; otherwise false.
457 ;; the successor has more than one predecessor;
459 ;; the successor is the current block (infinite loop);
461 ;; the next block has a different cleanup, and thus
462 ;; we may want to insert cleanup code between the
463 ;; two blocks at some point;
466 ;; the next block has a different home lambda, and
467 ;; thus the control transfer is a non-local exit.
470 ;; Stack analysis phase wants ENTRY to start a block...
476 ;; ... and a DX-allocator to end a block.
478 ;; FIXME: This is a partial workaround for bug 303.
480 nil)
483 t)))))
485 ;;; Join together two blocks. The code in BLOCK2 is moved into BLOCK1
486 ;;; and BLOCK2 is deleted from the DFO. We combine the optimize flags
487 ;;; for the two blocks so that any indicated optimization gets done.
520 ;;; Delete any nodes in BLOCK whose value is unused and which have no
521 ;;; side effects. We can delete sets of lexical variables when the set
522 ;;; variable has no references.
568 victim)
569 \f
570 ;;;; local call return type propagation
572 ;;; This function is called on RETURN nodes that have their REOPTIMIZE
573 ;;; flag set. It iterates over the uses of the RESULT, looking for
574 ;;; interesting stuff to update the TAIL-SET. If a use isn't a local
575 ;;; call, then we union its type together with the types of other such
576 ;;; uses. We assign to the RETURN-RESULT-TYPE the intersection of this
577 ;;; type with the RESULT's asserted type. We can make this
578 ;;; intersection now (potentially before type checking) because this
579 ;;; assertion on the result will eventually be checked (if
580 ;;; appropriate.)
581 ;;;
582 ;;; We call MAYBE-CONVERT-TAIL-LOCAL-CALL on each local non-MV
583 ;;; combination, which may change the successor of the call to be the
584 ;;; called function, and if so, checks if the call can become an
585 ;;; assignment. If we convert to an assignment, we abort, since the
586 ;;; RETURN has been deleted.
605 ;; (values-type-intersection
606 ;; (continuation-asserted-type result) ; FIXME -- APD, 2002-01-26
608 ;; )
609 ))
611 nil)
613 ;;; Do stuff to realize that something has changed about the value
614 ;;; delivered to a return node. Since we consider the return values of
615 ;;; all functions in the tail set to be equivalent, this amounts to
616 ;;; bringing the entire tail set up to date. We iterate over the
617 ;;; returns for all the functions in the tail set, reanalyzing them
618 ;;; all (not treating NODE specially.)
619 ;;;
620 ;;; When we are done, we check whether the new type is different from
621 ;;; the old TAIL-SET-TYPE. If so, we set the type and also reoptimize
622 ;;; all the lvars for references to functions in the tail set. This
623 ;;; will cause IR1-OPTIMIZE-COMBINATION to derive the new type as the
624 ;;; results of the calls.
628 :restart
648 \f
649 ;;;; IF optimization
651 ;;; Utility: return T if both argument cblocks are equivalent. For now,
652 ;;; detect only blocks that read the same leaf into the same lvar, and
653 ;;; continue to the same block.
666 ;;; Check whether the predicate is known to be true or false,
667 ;;; deleting the IF node in favor of the appropriate branch when this
668 ;;; is the case.
669 ;;; Similarly, when both branches are equivalent, branch directly to either
670 ;;; of them.
671 ;;; Also, if the test has multiple uses, replicate the node when possible...
672 ;;; in fact, splice in direct jumps to the right branch if possible.
684 alternative)
686 consequent)
689 alternative))))
697 ;; When we know that we only have a single successor, kill the victim
698 ;; ... unless the victim and the remaining successor are the same.
707 ;; When if/if conversion would leave (if ... (if nil ...)) or
708 ;; (if ... (if not-nil ...)), splice the correct successor right
709 ;; in.
728 ;; only one use left. Just kill the now-useless
729 ;; branch to avoid spurious code deletion notes.
732 node test block
738 ;; Finally, duplicate EQ-nil tests
745 ;; leave the last use as is, instead of replacing
746 ;; the (singly-referenced) CIF node with a duplicate.
749 ;;; Create a new copy of an IF node that tests the value of the node
750 ;;; USE. The test must have >1 use, and must be immediately used by
751 ;;; USE. NODE must be the only node in its block (implying that
752 ;;; block-start = if-test).
753 ;;;
754 ;;; This optimization has an effect semantically similar to the
755 ;;; source-to-source transformation:
756 ;;; (IF (IF A B C) D E) ==>
757 ;;; (IF A (IF B D E) (IF C D E))
758 ;;;
759 ;;; We clobber the NODE-SOURCE-PATH of both the original and the new
760 ;;; node so that dead code deletion notes will definitely not consider
761 ;;; either node to be part of the original source. One node might
762 ;;; become unreachable, resulting in a spurious note.
774 :consequent cblock
796 \f
797 ;;;; exit IR1 optimization
799 ;;; This function attempts to delete an exit node, returning true if
800 ;;; it deletes the block as a consequence:
801 ;;; -- If the exit is degenerate (has no ENTRY), then we don't do
802 ;;; anything, since there is nothing to be done.
803 ;;; -- If the exit node and its ENTRY have the same home lambda then
804 ;;; we know the exit is local, and can delete the exit. We change
805 ;;; uses of the Exit-Value to be uses of the original lvar,
806 ;;; then unlink the node. If the exit is to a TR context, then we
807 ;;; must do MERGE-TAIL-SETS on any local calls which delivered
808 ;;; their value to this exit.
809 ;;; -- If there is no value (as in a GO), then we skip the value
810 ;;; semantics.
811 ;;;
812 ;;; This function is also called by environment analysis, since it
813 ;;; wants all exits to be optimized even if normal optimization was
814 ;;; omitted.
824 ;; We can't simply use DELETE-FILTER to unlink the node
825 ;; and substitute some LVAR magic, as this can confuse the
826 ;; stack analysis if there's another EXIT to the same
827 ;; continuation. Instead, we fabricate a new block (in
828 ;; the same lexenv as the ENTRY, so it can't be merged
829 ;; backwards), insert a gimmicked CAST node to link up the
830 ;; LVAR holding the value being returned to the LVAR which
831 ;; is expecting to accept the value, thus placing the
832 ;; return value where it needs to be while still providing
833 ;; the hook required for stack analysis.
834 ;; -- AJB, 2014-Mar-03
839 (%make-cast
842 :value value
846 ;; We only expect a single successor to EXIT-BLOCK,
847 ;; because it contains an EXIT node (which must end its
848 ;; block) and the only blocks that have more than once
849 ;; successor are those with IF nodes (which also must
850 ;; end their blocks). Still, just to be sure, we use a
851 ;; construct that guarantees an error if this
852 ;; expectation is violated.
857 ;; Finish creating the new block.
861 ;; Link the new block into the control sequence.
866 ;; Finish re-pointing the value-holding LVAR to the
867 ;; CAST node.
872 ;; Register the CAST node as providing a value to the
873 ;; LVAR for the continuation.
877 ;; Remove the EXIT node.
880 ;; And, because we created a new block, we need to
881 ;; force component reanalysis (to assign a DFO number
882 ;; to the block if nothing else).
886 \f
887 ;;;; combination IR1 optimization
889 ;;; Report as we try each transform?
890 #!+sb-show
898 "The return value of ~A should not be discarded."
901 ;;; Do IR1 optimizations on a COMBINATION node.
911 (:local
916 (:error
920 (:full
943 ;; Check against the DEFINED-TYPE unless TYPE is already good.
953 (:known
961 ;; If somebody is really sure that they want to modify
962 ;; constants, let them.
975 ;; KLUDGE: The next test could be made more sensitive,
976 ;; only suppressing constant-folding of functions with
977 ;; CALL attributes when they're actually passed
978 ;; function arguments. -- WHN 19990918
998 ;; First give the VM a peek at the call
1002 (:direct
1003 ;; The VM knows how to handle this.
1004 )
1005 (:transform
1006 ;; The VM mostly knows how to handle this. We need
1007 ;; to massage the call slightly, though.
1010 ;; Let transforms have a crack at it.
1012 #!+sb-show
1018 #!+sb-show
1033 ;;; If NODE doesn't return (i.e. return type is NIL), then terminate
1034 ;;; the block there, and link it to the component tail.
1035 ;;;
1036 ;;; Except when called during IR1 convertion, we delete the
1037 ;;; continuation if it has no other uses. (If it does have other uses,
1038 ;;; we reoptimize.)
1039 ;;;
1040 ;;; Termination on the basis of a continuation type is
1041 ;;; inhibited when:
1042 ;;; -- The continuation is deleted (hence the assertion is spurious), or
1043 ;;; -- We are in IR1 conversion (where THE assertions are subject to
1044 ;;; weakening.) FIXME: Now THE assertions are not weakened, but new
1045 ;;; uses can(?) be added later. -- APD, 2003-07-17
1046 ;;;
1047 ;;; Why do we need to consider LVAR type? -- APD, 2003-07-30
1058 ;; Even if the combination will never return, don't terminate if this
1059 ;; is the tail call of a XEP: doing that would inhibit TCO.
1086 t))))
1088 ;;; This is called both by IR1 conversion and IR1 optimization when
1089 ;;; they have verified the type signature for the call, and are
1090 ;;; wondering if something should be done to special-case the call. If
1091 ;;; CALL is a call to a global function, then see whether it defined
1092 ;;; or known:
1093 ;;; -- If a DEFINED-FUN should be inline expanded, then convert
1094 ;;; the expansion and change the call to call it. Expansion is
1095 ;;; enabled if :INLINE or if SPACE=0. If the FUNCTIONAL slot is
1096 ;;; true, we never expand, since this function has already been
1097 ;;; converted. Local call analysis will duplicate the definition
1098 ;;; if necessary. We claim that the parent form is LABELS for
1099 ;;; context declarations, since we don't want it to be considered
1100 ;;; a real global function.
1101 ;;; -- If it is a known function, mark it as such by setting the KIND.
1102 ;;;
1103 ;;; We return the leaf referenced (NIL if not a leaf) and the
1104 ;;; FUN-INFO assigned.
1128 ;; Inline: if the function has already been converted at another call
1129 ;; site in this component, we point this REF to the functional. If not,
1130 ;; we convert the expansion.
1131 ;;
1132 ;; For :INLINE case local call analysis will copy the expansion later,
1133 ;; but for :MAYBE-INLINE and NIL cases we only get one copy of the
1134 ;; expansion per component.
1135 ;;
1136 ;; FIXME: We also convert in :INLINE & FUNCTIONAL-KIND case below. What
1137 ;; is it for?
1141 name
1143 leaf
1144 inlinep
1146 ;; Allow backward references to this function from following
1147 ;; forms. (Reused only if policy matches.)
1153 ;; Convert.
1159 ;; If we've already converted, change ref to the converted
1160 ;; functional.
1172 ;;; Check whether CALL satisfies TYPE. If so, apply the type to the
1173 ;;; call, and do MAYBE-TERMINATE-BLOCK and return the values of
1174 ;;; RECOGNIZE-KNOWN-CALL. If an error, set the combination kind and
1175 ;;; return NIL, NIL. If the type is just FUNCTION, then skip the
1176 ;;; syntax check, arg/result type processing, but still call
1177 ;;; RECOGNIZE-KNOWN-CALL, since the call might be to a known lambda,
1178 ;;; and that checking is done by local call analysis.
1187 ;; Using the defined-type too early is a bit of a waste: during
1188 ;; conversion we cannot use the untrusted ASSERT-CALL-TYPE, etc.
1193 ;; Don't validate multiple times against the same leaf --
1194 ;; it doesn't add any information, but may generate the same warning
1195 ;; multiple times.
1199 :result-test nil
1201 #'compiler-warn
1204 same-file-p)
1210 :result-test nil
1220 ;;; This is called by IR1-OPTIMIZE when the function for a call has
1221 ;;; changed. If the call is local, we try to LET-convert it, and
1222 ;;; derive the result type. If it is a :FULL call, we validate it
1223 ;;; against the type, which recognizes known calls, does inline
1224 ;;; expansion, etc. If a call to a predicate in a non-conditional
1225 ;;; position or to a function with a source transform, then we
1226 ;;; reconvert the form to give IR1 another chance.
1233 (:local
1238 (:full
1246 call))
1254 predicate)
1268 \f
1269 ;;;; known function optimization
1271 ;;; Add a failed optimization note to FAILED-OPTIMZATIONS for NODE,
1272 ;;; FUN and ARGS. If there is already a note for NODE and TRANSFORM,
1273 ;;; replace it, otherwise add a new one.
1284 ;;; Attempt to transform NODE using TRANSFORM-FUNCTION, subject to the
1285 ;;; call type constraint TRANSFORM-TYPE. If we are inhibited from
1286 ;;; doing the transform for some reason and FLAME is true, then we
1287 ;;; make a note of the message in FAILED-OPTIMIZATIONS for IR1
1288 ;;; finalize to pick up. We return true if the transform failed, and
1289 ;;; thus further transformation should be attempted. We return false
1290 ;;; if either the transform succeeded or was aborted.
1310 (:none
1312 nil)
1313 (:aborted
1318 nil)
1319 (:failure
1325 t)
1326 (:delayed
1328 nil))))
1331 type
1335 t)
1337 t))))
1339 ;;; When we don't like an IR1 transform, we throw the severity/reason
1340 ;;; and args.
1341 ;;;
1342 ;;; GIVE-UP-IR1-TRANSFORM is used to throw out of an IR1 transform,
1343 ;;; aborting this attempt to transform the call, but admitting the
1344 ;;; possibility that this or some other transform will later succeed.
1345 ;;; If arguments are supplied, they are format arguments for an
1346 ;;; efficiency note.
1347 ;;;
1348 ;;; ABORT-IR1-TRANSFORM is used to throw out of an IR1 transform and
1349 ;;; force a normal call to the function at run time. No further
1350 ;;; optimizations will be attempted.
1351 ;;;
1352 ;;; DELAY-IR1-TRANSFORM is used to throw out of an IR1 transform, and
1353 ;;; delay the transform on the node until later. REASONS specifies
1354 ;;; when the transform will be later retried. The :OPTIMIZE reason
1355 ;;; causes the transform to be delayed until after the current IR1
1356 ;;; optimization pass. The :CONSTRAINT reason causes the transform to
1357 ;;; be delayed until after constraint propagation.
1358 ;;;
1359 ;;; FIXME: Now (0.6.11.44) that there are 4 variants of this (GIVE-UP,
1360 ;;; ABORT, DELAY/:OPTIMIZE, DELAY/:CONSTRAINT) and we're starting to
1361 ;;; do CASE operations on the various REASON values, it might be a
1362 ;;; good idea to go OO, representing the reasons by objects, using
1363 ;;; CLOS methods on the objects instead of CASE, and (possibly) using
1364 ;;; SIGNAL instead of THROW.
1381 ;;; Poor man's catching and resignalling
1382 ;;; Implicit %GIVE-UP macrolet will resignal the give-up "condition"
1392 (:delayed
1400 ;;; Clear any delayed transform with no reasons - these should have
1401 ;;; been tried in the last pass. Then remove the reason from the
1402 ;;; delayed transform reasons, and if any become empty then set
1403 ;;; reoptimize flags for the node. Return true if any transforms are
1404 ;;; to be retried.
1420 reoptimize))
1422 ;;; Take the lambda-expression RES, IR1 convert it in the proper
1423 ;;; environment, and then install it as the function for the call
1424 ;;; NODE. We do local call analysis so that the new function is
1425 ;;; integrated into the control flow.
1426 ;;;
1427 ;;; We require the original function source name in order to generate
1428 ;;; a meaningful debug name for the lambda we set up. (It'd be
1429 ;;; possible to do this starting from debug names as well as source
1430 ;;; names, but as of sbcl-0.7.1.5, there was no need for this
1431 ;;; generality, since source names are always known to our callers.)
1437 ;; The internal variables of a transform are not going to be
1438 ;; interesting to the debugger, so there's no sense in
1439 ;; suppressing the substitution of variables with only one use
1440 ;; (the extra variables can slow down constraint propagation).
1441 ;;
1442 ;; This needs to be done before the WITH-IR1-ENVIRONMENT-FROM-NODE,
1443 ;; so that it will bind *LEXENV* to the right environment.
1456 res
1465 ;;; Replace a call to a foldable function of constant arguments with
1466 ;;; the result of evaluating the form. If there is an error during the
1467 ;;; evaluation, we give a warning and leave the call alone, making the
1468 ;;; call a :ERROR call.
1469 ;;;
1470 ;;; If there is more than one value, then we transform the call into a
1471 ;;; VALUES form.
1477 args
1478 call
1479 ;; Note: CMU CL had COMPILER-WARN here, and that
1480 ;; seems more natural, but it's probably not.
1481 ;;
1482 ;; It's especially not while bug 173 exists:
1483 ;; Expressions like
1484 ;; (COND (END
1485 ;; (UNLESS (OR UNSAFE? (<= END SIZE)))
1486 ;; ...))
1487 ;; can cause constant-folding TYPE-ERRORs (in
1488 ;; #'<=) when END can be proved to be NIL, even
1489 ;; though the code is perfectly legal and safe
1490 ;; because a NIL value of END means that the
1491 ;; #'<= will never be executed.
1492 ;;
1493 ;; Moreover, even without bug 173,
1494 ;; quite-possibly-valid code like
1495 ;; (COND ((NONINLINED-PREDICATE END)
1496 ;; (UNLESS (<= END SIZE))
1497 ;; ...))
1498 ;; (where NONINLINED-PREDICATE is something the
1499 ;; compiler can't do at compile time, but which
1500 ;; turns out to make the #'<= expression
1501 ;; unreachable when END=NIL) could cause errors
1502 ;; when the compiler tries to constant-fold (<=
1503 ;; END SIZE).
1504 ;;
1505 ;; So, with or without bug 173, it'd be
1506 ;; unnecessarily evil to do a full
1507 ;; COMPILER-WARNING (and thus return FAILURE-P=T
1508 ;; from COMPILE-FILE) for legal code, so we we
1509 ;; use a wimpier COMPILE-STYLE-WARNING instead.
1511 ;; On the other hand, for code we control, we
1512 ;; should be able to work around any bug
1513 ;; 173-related problems, and in particular we
1514 ;; want to be alerted to calls to our own
1515 ;; functions which aren't being folded away; a
1516 ;; COMPILER-WARNING is butch enough to stop the
1517 ;; SBCL build itself in its tracks.
1537 call
1541 fun-name))))))
1543 \f
1544 ;;;; local call optimization
1546 ;;; Propagate TYPE to LEAF and its REFS, marking things changed.
1547 ;;;
1548 ;;; If the leaf type is a function type, then just leave it alone, since TYPE
1549 ;;; is never going to be more specific than that (and TYPE-INTERSECTION would
1550 ;;; choke.)
1551 ;;;
1552 ;;; Also, if the type is one requiring special care don't touch it if the leaf
1553 ;;; has multiple references -- otherwise LVAR-CONSERVATIVE-TYPE is screwed.
1568 ;; KLUDGE: LET var substitution
1574 ;;; Iteration variable: exactly one SETQ of the form:
1575 ;;;
1576 ;;; (let ((var initial))
1577 ;;; ...
1578 ;;; (setq var (+ var step))
1579 ;;; ...)
1605 ;; Detect cases like (LOOP FOR 1.0 to 5.0 ...), where
1606 ;; the initial and the step are of different types,
1607 ;; and the step is less contagious.
1610 step-type))))
1649 :low low
1650 :high high
1653 "check for iteration variable reoptimization"
1663 ;;; Figure out the type of a LET variable that has sets. We compute
1664 ;;; the union of the INITIAL-TYPE and the types of all the set
1665 ;;; values and to a PROPAGATE-TO-REFS with this type.
1685 ;;; If a LET variable, find the initial value's type and do
1686 ;;; PROPAGATE-FROM-SETS. We also derive the VALUE's type as the node's
1687 ;;; type.
1703 ;;; Return true if the value of REF will always be the same (and is
1704 ;;; thus legal to substitute.)
1716 (:global-function
1723 ;;; If we have a non-set LET var with a single use, then (if possible)
1724 ;;; replace the variable reference's LVAR with the arg lvar.
1725 ;;;
1726 ;;; We change the REF to be a reference to NIL with unused value, and
1727 ;;; let it be flushed as dead code. A side effect of this substitution
1728 ;;; is to delete the variable.
1736 ;; Think about (LET ((A ...)) (IF ... A ...)): two
1737 ;; LVAR-USEs should not be met on one path. Another problem
1738 ;; is with dynamic-extent.
1741 ;; If the destinatation is dynamic extent, don't substitute unless
1742 ;; the source is as well.
1747 ;; we should not change lifetime of unknown values lvars
1759 ;; While CRETURN and EXIT nodes may be known-values,
1760 ;; they have their own complications, such as
1761 ;; substitution into CRETURN may create new tail calls.
1762 nil)
1765 t))
1771 ;; Really it is (EQ (LVAR-USES LVAR) REF):
1772 t)
1777 ;; KLUDGE: it should be (TYPE-CHECK 0)
1781 ;; FIXME
1782 t)
1790 t)))
1792 ;;; Delete a LET, removing the call and bind nodes, and warning about
1793 ;;; any unreferenced variables. Note that FLUSH-DEAD-CODE will come
1794 ;;; along right away and delete the REF and then the lambda, since we
1795 ;;; flush the FUN lvar.
1810 ;;; This function is called when one of the arguments to a LET
1811 ;;; changes. We look at each changed argument. If the corresponding
1812 ;;; variable is set, then we call PROPAGATE-FROM-SETS. Otherwise, we
1813 ;;; consider substituting for the variable, and also propagate
1814 ;;; derived-type information for the arg to all the VAR's refs.
1815 ;;;
1816 ;;; Substitution is inhibited when the arg leaf's derived type isn't a
1817 ;;; subtype of the argument's leaf type. This prevents type checking
1818 ;;; from being defeated, and also ensures that the best representation
1819 ;;; for the variable can be used.
1820 ;;;
1821 ;;; Substitution of individual references is inhibited if the
1822 ;;; reference is in a different component from the home. This can only
1823 ;;; happen with closures over top level lambda vars. In such cases,
1824 ;;; the references may have already been compiled, and thus can't be
1825 ;;; retroactively modified.
1826 ;;;
1827 ;;; If all of the variables are deleted (have no references) when we
1828 ;;; are done, then we delete the LET.
1829 ;;;
1830 ;;; Note that we are responsible for clearing the LVAR-REOPTIMIZE
1831 ;;; flags.
1846 ;; (NODE-DERIVED-TYPE USE) would
1847 ;; be better -- APD, 2003-05-15
1854 t)
1857 nil)))
1858 leaf var)))
1859 t)))))
1871 ;;; This function is called when one of the args to a non-LET local
1872 ;;; call changes. For each changed argument corresponding to an unset
1873 ;;; variable, we compute the union of the types across all calls and
1874 ;;; propagate this type information to the var's refs.
1875 ;;;
1876 ;;; If the function has an entry-fun, then we don't do anything: since
1877 ;;; it has a XEP we would not discover anything.
1878 ;;;
1879 ;;; If the function is an optional-entry-point, we will just make sure
1880 ;;; &REST lists are known to be lists. Doing the regular rigamarole
1881 ;;; can erronously propagate too strict types into refs: see
1882 ;;; BUG-655203-REGRESSION in tests/compiler.pure.lisp.
1883 ;;;
1884 ;;; We can clear the LVAR-REOPTIMIZE flags for arguments in all calls
1885 ;;; corresponding to changed arguments in CALL, since the only use in
1886 ;;; IR1 optimization of the REOPTIMIZE flag for local call args is
1887 ;;; right here.
1892 ;; We can still make sure &REST is known to be a list.
1897 ;; The normal case.
1905 vars))
1921 union)))))
1924 and type in union
1928 \f
1929 ;;;; multiple values optimization
1931 ;;; Do stuff to notice a change to a MV combination node. There are
1932 ;;; two main branches here:
1933 ;;; -- If the call is local, then it is already a MV let, or should
1934 ;;; become one. Note that although all :LOCAL MV calls must eventually
1935 ;;; be converted to :MV-LETs, there can be a window when the call
1936 ;;; is local, but has not been LET converted yet. This is because
1937 ;;; the entry-point lambdas may have stray references (in other
1938 ;;; entry points) that have not been deleted yet.
1939 ;;; -- The call is full. This case is somewhat similar to the non-MV
1940 ;;; combination optimization: we propagate return type information and
1941 ;;; notice non-returning calls. We also have an optimization
1942 ;;; which tries to convert MV-CALLs into MV-binds.
1948 (:local
1956 (:full
1978 ;;; Propagate derived type info from the values lvar to the vars.
1985 n-vars)))
1987 and type in types
1994 ;;; If possible, convert a general MV call to an MV-BIND. We can do
1995 ;;; this if:
1996 ;;; -- The call has only one argument, and
1997 ;;; -- The function has a known fixed number of arguments, or
1998 ;;; -- The argument yields a known fixed number of values.
1999 ;;;
2000 ;;; What we do is change the function in the MV-CALL to be a lambda
2001 ;;; that "looks like an MV bind", which allows
2002 ;;; IR1-OPTIMIZE-MV-COMBINATION to notice that this call can be
2003 ;;; converted (the next time around.) This new lambda just calls the
2004 ;;; actual function with the MV-BIND variables as arguments. Note that
2005 ;;; this new MV bind is not let-converted immediately, as there are
2006 ;;; going to be stray references from the entry-point functions until
2007 ;;; they get deleted.
2008 ;;;
2009 ;;; In order to avoid loss of argument count checking, we only do the
2010 ;;; transformation according to a known number of expected argument if
2011 ;;; safety is unimportant. We can always convert if we know the number
2012 ;;; of actual values, since the normal call that we build will still
2013 ;;; do any appropriate argument count checking.
2014 ;;;
2015 ;;; We only attempt the transformation if the called function is a
2016 ;;; constant reference. This allows us to just splice the leaf into
2017 ;;; the new function, instead of trying to somehow bind the function
2018 ;;; expression. The leaf must be constant because we are evaluating it
2019 ;;; again in a different place. This also has the effect of squelching
2020 ;;; multiple warnings when there is an argument count error.
2042 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2043 at least ~R."
2044 total-nvals min)
2049 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2050 at most ~R."
2051 total-nvals max)
2058 min)
2077 ;;; If we see:
2078 ;;; (multiple-value-bind
2079 ;;; (x y)
2080 ;;; (values xx yy)
2081 ;;; ...)
2082 ;;; Convert to:
2083 ;;; (let ((x xx)
2084 ;;; (y yy))
2085 ;;; ...)
2086 ;;;
2087 ;;; What we actually do is convert the VALUES combination into a
2088 ;;; normal LET combination calling the original :MV-LET lambda. If
2089 ;;; there are extra args to VALUES, discard the corresponding
2090 ;;; lvars. If there are insufficient args, insert references to NIL.
2113 for prev = node-prev then ctran
2119 use ctran))
2133 ;; Propagate derived types from the VALUES call to its args:
2134 ;; transforms can leave the VALUES call with a better type
2135 ;; than its args have, so make sure not to throw that away.
2141 ;; Propagate declared types of MV-BIND variables.
2144 t)))
2146 ;;; If we see:
2147 ;;; (values-list (list x y z))
2148 ;;;
2149 ;;; Convert to:
2150 ;;; (values x y z)
2151 ;;;
2152 ;;; In implementation, this is somewhat similar to
2153 ;;; CONVERT-MV-BIND-TO-LET. We grab the args of LIST and make them
2154 ;;; args of the VALUES-LIST call, flushing the old argument lvar
2155 ;;; (allowing the LIST to be flushed.)
2156 ;;;
2157 ;;; FIXME: Thus we lose possible type assertions on (LIST ...).
2164 ;; FIXME: VALUES might not satisfy an assertion on NODE-LVAR.
2176 t)))
2178 ;;; If VALUES appears in a non-MV context, then effectively convert it
2179 ;;; to a PROG1. This allows the computation of the additional values
2180 ;;; to become dead code.
2192 val))
2193 nil))
2195 ;;; TODO:
2196 ;;; - CAST chains;
2211 ;; Vestigial exits are only introduced when eliminating a local
2212 ;; RETURN-FROM. We may delete them only when we can show that
2213 ;; there are no other code paths that use the entry LVAR that
2214 ;; are live from within the block that contained the deleted
2215 ;; EXIT (our predecessor block). The conservative version of
2216 ;; this is that there are no EXITs for any ENTRY introduced
2217 ;; between the LEXENV of the deleted EXIT and the LEXENV of the
2218 ;; target ENTRY.
2233 #+sb-xc-host context
2249 lvar)
2250 ;; Pathwise removing of CAST
2253 next-block)
2281 ;; FIXME: Do it in one step.
2285 value
2291 ;; FIXME: Derived type.
2295 ',detail
2297 ;; KLUDGE: FILTER-LVAR does not work for non-returning
2298 ;; functions, so we declare the return type of
2299 ;; %COMPILE-TIME-TYPE-ERROR to be * and derive the real type
2300 ;; here.
2304 ;; FIXME: Is it necessary?