| blob | 132b3da0f94ef0a373ae141d8c7d320be5efd2a7 |
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.
323 context)))
325 t))))
333 :value lvar
335 :arg-count arg-count
347 type-check type-asserted)
348 t)
349 cast)))
357 caller policy)))
360 \f
361 ;;;; IR1-OPTIMIZE
363 ;;; Do one forward pass over COMPONENT, deleting unreachable blocks
364 ;;; and doing IR1 optimizations. We can ignore all blocks that don't
365 ;;; have the REOPTIMIZE flag set. If COMPONENT-REOPTIMIZE is true when
366 ;;; we are done, then another iteration would be beneficial.
372 for last-block = block
375 ;; We delete blocks when there is either no predecessor or the
376 ;; block is in a lambda that has been deleted. These blocks
377 ;; would eventually be deleted by DFO recomputation, but doing
378 ;; it here immediately makes the effect available to IR1
379 ;; optimization.
385 ;; Preserve the BLOCK-SUCC invariant that almost every block has
386 ;; one successor (and a block with DELETE-P set is an acceptable
387 ;; exception).
422 ;;; Loop over the nodes in BLOCK, acting on (and clearing) REOPTIMIZE
423 ;;; flags.
424 ;;;
425 ;;; Note that although they are cleared here, REOPTIMIZE flags might
426 ;;; still be set upon return from this function, meaning that further
427 ;;; optimization is wanted (as a consequence of optimizations we did).
430 ;; We clear the node and block REOPTIMIZE flags before doing the
431 ;; optimization, not after. This ensures that the node or block will
432 ;; be reoptimized if necessary.
436 ;; As above, we clear the node REOPTIMIZE flag before optimizing.
441 ;; With a COMBINATION, we call PROPAGATE-FUN-CHANGE whenever
442 ;; the function changes, and call IR1-OPTIMIZE-COMBINATION if
443 ;; any argument changes.
448 ;; KLUDGE: We leave the NODE-OPTIMIZE flag set going into
449 ;; IR1-OPTIMIZE-RETURN, since IR1-OPTIMIZE-RETURN wants to
450 ;; clear the flag itself. -- WHN 2002-02-02, quoting original
451 ;; CMU CL comments
457 ;; With an EXIT, we derive the node's type from the VALUE's
458 ;; type.
463 ;; PROPAGATE-FROM-SETS can do a better job if NODE-REOPTIMIZE
464 ;; is accurate till the node actually has been reoptimized.
472 ;;; Try to join with a successor block. If we succeed, we return true,
473 ;;; otherwise false.
480 ;; the successor has more than one predecessor;
482 ;; the successor is the current block (infinite loop);
484 ;; the next block has a different cleanup, and thus
485 ;; we may want to insert cleanup code between the
486 ;; two blocks at some point;
489 ;; the next block has a different home lambda, and
490 ;; thus the control transfer is a non-local exit.
493 ;; Stack analysis phase wants ENTRY to start a block...
499 ;; ... and a DX-allocator to end a block.
501 ;; FIXME: This is a partial workaround for bug 303.
503 nil)
506 t)))))
508 ;;; Join together two blocks. The code in BLOCK2 is moved into BLOCK1
509 ;;; and BLOCK2 is deleted from the DFO. We combine the optimize flags
510 ;;; for the two blocks so that any indicated optimization gets done.
543 ;;; Delete any nodes in BLOCK whose value is unused and which have no
544 ;;; side effects. We can delete sets of lexical variables when the set
545 ;;; variable has no references.
591 victim)
592 \f
593 ;;;; local call return type propagation
595 ;;; This function is called on RETURN nodes that have their REOPTIMIZE
596 ;;; flag set. It iterates over the uses of the RESULT, looking for
597 ;;; interesting stuff to update the TAIL-SET. If a use isn't a local
598 ;;; call, then we union its type together with the types of other such
599 ;;; uses. We assign to the RETURN-RESULT-TYPE the intersection of this
600 ;;; type with the RESULT's asserted type. We can make this
601 ;;; intersection now (potentially before type checking) because this
602 ;;; assertion on the result will eventually be checked (if
603 ;;; appropriate.)
604 ;;;
605 ;;; We call MAYBE-CONVERT-TAIL-LOCAL-CALL on each local non-MV
606 ;;; combination, which may change the successor of the call to be the
607 ;;; called function, and if so, checks if the call can become an
608 ;;; assignment. If we convert to an assignment, we abort, since the
609 ;;; RETURN has been deleted.
628 ;; (values-type-intersection
629 ;; (continuation-asserted-type result) ; FIXME -- APD, 2002-01-26
631 ;; )
632 ))
634 nil)
636 ;;; Do stuff to realize that something has changed about the value
637 ;;; delivered to a return node. Since we consider the return values of
638 ;;; all functions in the tail set to be equivalent, this amounts to
639 ;;; bringing the entire tail set up to date. We iterate over the
640 ;;; returns for all the functions in the tail set, reanalyzing them
641 ;;; all (not treating NODE specially.)
642 ;;;
643 ;;; When we are done, we check whether the new type is different from
644 ;;; the old TAIL-SET-TYPE. If so, we set the type and also reoptimize
645 ;;; all the lvars for references to functions in the tail set. This
646 ;;; will cause IR1-OPTIMIZE-COMBINATION to derive the new type as the
647 ;;; results of the calls.
651 :restart
671 \f
672 ;;;; IF optimization
674 ;;; Utility: return T if both argument cblocks are equivalent. For now,
675 ;;; detect only blocks that read the same leaf into the same lvar, and
676 ;;; continue to the same block.
689 ;;; Check whether the predicate is known to be true or false,
690 ;;; deleting the IF node in favor of the appropriate branch when this
691 ;;; is the case.
692 ;;; Similarly, when both branches are equivalent, branch directly to either
693 ;;; of them.
694 ;;; Also, if the test has multiple uses, replicate the node when possible...
695 ;;; in fact, splice in direct jumps to the right branch if possible.
707 alternative)
709 consequent)
712 alternative))))
720 ;; When we know that we only have a single successor, kill the victim
721 ;; ... unless the victim and the remaining successor are the same.
730 ;; When if/if conversion would leave (if ... (if nil ...)) or
731 ;; (if ... (if not-nil ...)), splice the correct successor right
732 ;; in.
751 ;; only one use left. Just kill the now-useless
752 ;; branch to avoid spurious code deletion notes.
755 node test block
761 ;; Finally, duplicate EQ-nil tests
768 ;; leave the last use as is, instead of replacing
769 ;; the (singly-referenced) CIF node with a duplicate.
772 ;;; Create a new copy of an IF node that tests the value of the node
773 ;;; USE. The test must have >1 use, and must be immediately used by
774 ;;; USE. NODE must be the only node in its block (implying that
775 ;;; block-start = if-test).
776 ;;;
777 ;;; This optimization has an effect semantically similar to the
778 ;;; source-to-source transformation:
779 ;;; (IF (IF A B C) D E) ==>
780 ;;; (IF A (IF B D E) (IF C D E))
781 ;;;
782 ;;; We clobber the NODE-SOURCE-PATH of both the original and the new
783 ;;; node so that dead code deletion notes will definitely not consider
784 ;;; either node to be part of the original source. One node might
785 ;;; become unreachable, resulting in a spurious note.
797 :consequent cblock
819 \f
820 ;;;; exit IR1 optimization
822 ;;; This function attempts to delete an exit node, returning true if
823 ;;; it deletes the block as a consequence:
824 ;;; -- If the exit is degenerate (has no ENTRY), then we don't do
825 ;;; anything, since there is nothing to be done.
826 ;;; -- If the exit node and its ENTRY have the same home lambda then
827 ;;; we know the exit is local, and can delete the exit. We change
828 ;;; uses of the Exit-Value to be uses of the original lvar,
829 ;;; then unlink the node. If the exit is to a TR context, then we
830 ;;; must do MERGE-TAIL-SETS on any local calls which delivered
831 ;;; their value to this exit.
832 ;;; -- If there is no value (as in a GO), then we skip the value
833 ;;; semantics.
834 ;;;
835 ;;; This function is also called by environment analysis, since it
836 ;;; wants all exits to be optimized even if normal optimization was
837 ;;; omitted.
847 ;; We can't simply use DELETE-FILTER to unlink the node
848 ;; and substitute some LVAR magic, as this can confuse the
849 ;; stack analysis if there's another EXIT to the same
850 ;; continuation. Instead, we fabricate a new block (in
851 ;; the same lexenv as the ENTRY, so it can't be merged
852 ;; backwards), insert a gimmicked CAST node to link up the
853 ;; LVAR holding the value being returned to the LVAR which
854 ;; is expecting to accept the value, thus placing the
855 ;; return value where it needs to be while still providing
856 ;; the hook required for stack analysis.
857 ;; -- AJB, 2014-Mar-03
862 (%make-cast
865 :value value
869 ;; We only expect a single successor to EXIT-BLOCK,
870 ;; because it contains an EXIT node (which must end its
871 ;; block) and the only blocks that have more than once
872 ;; successor are those with IF nodes (which also must
873 ;; end their blocks). Still, just to be sure, we use a
874 ;; construct that guarantees an error if this
875 ;; expectation is violated.
880 ;; Finish creating the new block.
884 ;; Link the new block into the control sequence.
889 ;; Finish re-pointing the value-holding LVAR to the
890 ;; CAST node.
895 ;; Register the CAST node as providing a value to the
896 ;; LVAR for the continuation.
900 ;; Remove the EXIT node.
903 ;; And, because we created a new block, we need to
904 ;; force component reanalysis (to assign a DFO number
905 ;; to the block if nothing else).
909 \f
910 ;;;; combination IR1 optimization
912 ;;; Report as we try each transform?
913 #!+sb-show
921 "The return value of ~A should not be discarded."
924 ;;; Do IR1 optimizations on a COMBINATION node.
934 (:local
939 (:error
943 (:full
964 ;; Check against the DEFINED-TYPE unless TYPE is already good.
974 (:known
982 ;; If somebody is really sure that they want to modify
983 ;; constants, let them.
1012 ;; First give the VM a peek at the call
1016 (:direct
1017 ;; The VM knows how to handle this.
1018 )
1019 (:transform
1020 ;; The VM mostly knows how to handle this. We need
1021 ;; to massage the call slightly, though.
1024 ;; Let transforms have a crack at it.
1026 #!+sb-show
1032 #!+sb-show
1047 ;;; If NODE doesn't return (i.e. return type is NIL), then terminate
1048 ;;; the block there, and link it to the component tail.
1049 ;;;
1050 ;;; Except when called during IR1 convertion, we delete the
1051 ;;; continuation if it has no other uses. (If it does have other uses,
1052 ;;; we reoptimize.)
1053 ;;;
1054 ;;; Termination on the basis of a continuation type is
1055 ;;; inhibited when:
1056 ;;; -- The continuation is deleted (hence the assertion is spurious), or
1057 ;;; -- We are in IR1 conversion (where THE assertions are subject to
1058 ;;; weakening.) FIXME: Now THE assertions are not weakened, but new
1059 ;;; uses can(?) be added later. -- APD, 2003-07-17
1060 ;;;
1061 ;;; Why do we need to consider LVAR type? -- APD, 2003-07-30
1072 ;; Even if the combination will never return, don't terminate if this
1073 ;; is the tail call of a XEP: doing that would inhibit TCO.
1100 t))))
1102 ;;; This is called both by IR1 conversion and IR1 optimization when
1103 ;;; they have verified the type signature for the call, and are
1104 ;;; wondering if something should be done to special-case the call. If
1105 ;;; CALL is a call to a global function, then see whether it defined
1106 ;;; or known:
1107 ;;; -- If a DEFINED-FUN should be inline expanded, then convert
1108 ;;; the expansion and change the call to call it. Expansion is
1109 ;;; enabled if :INLINE or if SPACE=0. If the FUNCTIONAL slot is
1110 ;;; true, we never expand, since this function has already been
1111 ;;; converted. Local call analysis will duplicate the definition
1112 ;;; if necessary. We claim that the parent form is LABELS for
1113 ;;; context declarations, since we don't want it to be considered
1114 ;;; a real global function.
1115 ;;; -- If it is a known function, mark it as such by setting the KIND.
1116 ;;;
1117 ;;; We return the leaf referenced (NIL if not a leaf) and the
1118 ;;; FUN-INFO assigned.
1142 ;; Inline: if the function has already been converted at another call
1143 ;; site in this component, we point this REF to the functional. If not,
1144 ;; we convert the expansion.
1145 ;;
1146 ;; For :INLINE case local call analysis will copy the expansion later,
1147 ;; but for :MAYBE-INLINE and NIL cases we only get one copy of the
1148 ;; expansion per component.
1149 ;;
1150 ;; FIXME: We also convert in :INLINE & FUNCTIONAL-KIND case below. What
1151 ;; is it for?
1155 name
1157 leaf
1158 inlinep
1160 ;; Allow backward references to this function from following
1161 ;; forms. (Reused only if policy matches.)
1167 ;; Convert.
1173 ;; If we've already converted, change ref to the converted
1174 ;; functional.
1186 ;;; Check whether CALL satisfies TYPE. If so, apply the type to the
1187 ;;; call, and do MAYBE-TERMINATE-BLOCK and return the values of
1188 ;;; RECOGNIZE-KNOWN-CALL. If an error, set the combination kind and
1189 ;;; return NIL, NIL. If the type is just FUNCTION, then skip the
1190 ;;; syntax check, arg/result type processing, but still call
1191 ;;; RECOGNIZE-KNOWN-CALL, since the call might be to a known lambda,
1192 ;;; and that checking is done by local call analysis.
1201 ;; Using the defined-type too early is a bit of a waste: during
1202 ;; conversion we cannot use the untrusted ASSERT-CALL-TYPE, etc.
1207 ;; Don't validate multiple times against the same leaf --
1208 ;; it doesn't add any information, but may generate the same warning
1209 ;; multiple times.
1213 :result-test nil
1215 #'compiler-warn
1218 same-file-p)
1225 :result-test nil
1235 ;;; This is called by IR1-OPTIMIZE when the function for a call has
1236 ;;; changed. If the call is local, we try to LET-convert it, and
1237 ;;; derive the result type. If it is a :FULL call, we validate it
1238 ;;; against the type, which recognizes known calls, does inline
1239 ;;; expansion, etc. If a call to a predicate in a non-conditional
1240 ;;; position or to a function with a source transform, then we
1241 ;;; reconvert the form to give IR1 another chance.
1248 (:local
1253 (:full
1261 call))
1269 predicate)
1283 \f
1284 ;;;; known function optimization
1286 ;;; Add a failed optimization note to FAILED-OPTIMZATIONS for NODE,
1287 ;;; FUN and ARGS. If there is already a note for NODE and TRANSFORM,
1288 ;;; replace it, otherwise add a new one.
1299 ;;; Attempt to transform NODE using TRANSFORM-FUNCTION, subject to the
1300 ;;; call type constraint TRANSFORM-TYPE. If we are inhibited from
1301 ;;; doing the transform for some reason and FLAME is true, then we
1302 ;;; make a note of the message in FAILED-OPTIMIZATIONS for IR1
1303 ;;; finalize to pick up. We return true if the transform failed, and
1304 ;;; thus further transformation should be attempted. We return false
1305 ;;; if either the transform succeeded or was aborted.
1328 (:none
1330 nil)
1331 (:aborted
1336 nil)
1337 (:failure
1343 t)
1344 (:delayed
1346 nil))))
1349 type
1353 t)
1355 t))))
1357 ;;; When we don't like an IR1 transform, we throw the severity/reason
1358 ;;; and args.
1359 ;;;
1360 ;;; GIVE-UP-IR1-TRANSFORM is used to throw out of an IR1 transform,
1361 ;;; aborting this attempt to transform the call, but admitting the
1362 ;;; possibility that this or some other transform will later succeed.
1363 ;;; If arguments are supplied, they are format arguments for an
1364 ;;; efficiency note.
1365 ;;;
1366 ;;; ABORT-IR1-TRANSFORM is used to throw out of an IR1 transform and
1367 ;;; force a normal call to the function at run time. No further
1368 ;;; optimizations will be attempted.
1369 ;;;
1370 ;;; DELAY-IR1-TRANSFORM is used to throw out of an IR1 transform, and
1371 ;;; delay the transform on the node until later. REASONS specifies
1372 ;;; when the transform will be later retried. The :OPTIMIZE reason
1373 ;;; causes the transform to be delayed until after the current IR1
1374 ;;; optimization pass. The :CONSTRAINT reason causes the transform to
1375 ;;; be delayed until after constraint propagation.
1376 ;;;
1377 ;;; FIXME: Now (0.6.11.44) that there are 4 variants of this (GIVE-UP,
1378 ;;; ABORT, DELAY/:OPTIMIZE, DELAY/:CONSTRAINT) and we're starting to
1379 ;;; do CASE operations on the various REASON values, it might be a
1380 ;;; good idea to go OO, representing the reasons by objects, using
1381 ;;; CLOS methods on the objects instead of CASE, and (possibly) using
1382 ;;; SIGNAL instead of THROW.
1400 ;;; Poor man's catching and resignalling
1401 ;;; Implicit %GIVE-UP macrolet will resignal the give-up "condition"
1411 (:delayed
1419 ;;; Clear any delayed transform with no reasons - these should have
1420 ;;; been tried in the last pass. Then remove the reason from the
1421 ;;; delayed transform reasons, and if any become empty then set
1422 ;;; reoptimize flags for the node. Return true if any transforms are
1423 ;;; to be retried.
1439 reoptimize))
1441 ;;; Take the lambda-expression RES, IR1 convert it in the proper
1442 ;;; environment, and then install it as the function for the call
1443 ;;; NODE. We do local call analysis so that the new function is
1444 ;;; integrated into the control flow.
1445 ;;;
1446 ;;; We require the original function source name in order to generate
1447 ;;; a meaningful debug name for the lambda we set up. (It'd be
1448 ;;; possible to do this starting from debug names as well as source
1449 ;;; names, but as of sbcl-0.7.1.5, there was no need for this
1450 ;;; generality, since source names are always known to our callers.)
1456 ;; The internal variables of a transform are not going to be
1457 ;; interesting to the debugger, so there's no sense in
1458 ;; suppressing the substitution of variables with only one use
1459 ;; (the extra variables can slow down constraint propagation).
1460 ;;
1461 ;; This needs to be done before the WITH-IR1-ENVIRONMENT-FROM-NODE,
1462 ;; so that it will bind *LEXENV* to the right environment.
1475 res
1487 t)
1496 ;;; Return T if the function is foldable and if it's marked as CALL
1497 ;;; all function arguments are FOLDABLE too.
1503 nil)
1511 args)))
1515 ;;; Replace a call to a foldable function of constant arguments with
1516 ;;; the result of evaluating the form. If there is an error during the
1517 ;;; evaluation, we give a warning and leave the call alone, making the
1518 ;;; call a :ERROR call.
1519 ;;;
1520 ;;; If there is more than one value, then we transform the call into a
1521 ;;; VALUES form.
1532 args
1533 call
1534 ;; Note: CMU CL had COMPILER-WARN here, and that
1535 ;; seems more natural, but it's probably not.
1536 ;;
1537 ;; It's especially not while bug 173 exists:
1538 ;; Expressions like
1539 ;; (COND (END
1540 ;; (UNLESS (OR UNSAFE? (<= END SIZE)))
1541 ;; ...))
1542 ;; can cause constant-folding TYPE-ERRORs (in
1543 ;; #'<=) when END can be proved to be NIL, even
1544 ;; though the code is perfectly legal and safe
1545 ;; because a NIL value of END means that the
1546 ;; #'<= will never be executed.
1547 ;;
1548 ;; Moreover, even without bug 173,
1549 ;; quite-possibly-valid code like
1550 ;; (COND ((NONINLINED-PREDICATE END)
1551 ;; (UNLESS (<= END SIZE))
1552 ;; ...))
1553 ;; (where NONINLINED-PREDICATE is something the
1554 ;; compiler can't do at compile time, but which
1555 ;; turns out to make the #'<= expression
1556 ;; unreachable when END=NIL) could cause errors
1557 ;; when the compiler tries to constant-fold (<=
1558 ;; END SIZE).
1559 ;;
1560 ;; So, with or without bug 173, it'd be
1561 ;; unnecessarily evil to do a full
1562 ;; COMPILER-WARNING (and thus return FAILURE-P=T
1563 ;; from COMPILE-FILE) for legal code, so we we
1564 ;; use a wimpier COMPILE-STYLE-WARNING instead.
1566 ;; On the other hand, for code we control, we
1567 ;; should be able to work around any bug
1568 ;; 173-related problems, and in particular we
1569 ;; want to be alerted to calls to our own
1570 ;; functions which aren't being folded away; a
1571 ;; COMPILER-WARNING is butch enough to stop the
1572 ;; SBCL build itself in its tracks.
1592 call
1596 fun-name))))))
1598 \f
1599 ;;;; local call optimization
1601 ;;; Propagate TYPE to LEAF and its REFS, marking things changed.
1602 ;;;
1603 ;;; If the leaf type is a function type, then just leave it alone, since TYPE
1604 ;;; is never going to be more specific than that (and TYPE-INTERSECTION would
1605 ;;; choke.)
1606 ;;;
1607 ;;; Also, if the type is one requiring special care don't touch it if the leaf
1608 ;;; has multiple references -- otherwise LVAR-CONSERVATIVE-TYPE is screwed.
1623 ;; KLUDGE: LET var substitution
1629 ;;; Iteration variable: exactly one SETQ of the form:
1630 ;;;
1631 ;;; (let ((var initial))
1632 ;;; ...
1633 ;;; (setq var (+ var step))
1634 ;;; ...)
1660 ;; Detect cases like (LOOP FOR 1.0 to 5.0 ...), where
1661 ;; the initial and the step are of different types,
1662 ;; and the step is less contagious.
1665 step-type))))
1704 :low low
1705 :high high
1708 "check for iteration variable reoptimization"
1718 ;;; Figure out the type of a LET variable that has sets. We compute
1719 ;;; the union of the INITIAL-TYPE and the types of all the set
1720 ;;; values and to a PROPAGATE-TO-REFS with this type.
1740 ;;; If a LET variable, find the initial value's type and do
1741 ;;; PROPAGATE-FROM-SETS. We also derive the VALUE's type as the node's
1742 ;;; type.
1758 ;;; Return true if the value of REF will always be the same (and is
1759 ;;; thus legal to substitute.)
1771 (:global-function
1778 ;;; If we have a non-set LET var with a single use, then (if possible)
1779 ;;; replace the variable reference's LVAR with the arg lvar.
1780 ;;;
1781 ;;; We change the REF to be a reference to NIL with unused value, and
1782 ;;; let it be flushed as dead code. A side effect of this substitution
1783 ;;; is to delete the variable.
1791 ;; Think about (LET ((A ...)) (IF ... A ...)): two
1792 ;; LVAR-USEs should not be met on one path. Another problem
1793 ;; is with dynamic-extent.
1796 ;; If the destinatation is dynamic extent, don't substitute unless
1797 ;; the source is as well.
1802 ;; we should not change lifetime of unknown values lvars
1809 ;; CASTs can disappear, don't substitute if
1810 ;; DEST-LVAR has other uses (this will be
1811 ;; insufficient if we have a CAST-CAST chain, but
1812 ;; works well for a single CAST)
1820 ;; While CRETURN and EXIT nodes may be known-values,
1821 ;; they have their own complications, such as
1822 ;; substitution into CRETURN may create new tail calls.
1823 nil)
1826 t))
1832 ;; Really it is (EQ (LVAR-USES LVAR) REF):
1833 t)
1838 ;; KLUDGE: it should be (TYPE-CHECK 0)
1842 ;; FIXME
1843 t)
1851 t)))
1853 ;;; Delete a LET, removing the call and bind nodes, and warning about
1854 ;;; any unreferenced variables. Note that FLUSH-DEAD-CODE will come
1855 ;;; along right away and delete the REF and then the lambda, since we
1856 ;;; flush the FUN lvar.
1871 ;;; This function is called when one of the arguments to a LET
1872 ;;; changes. We look at each changed argument. If the corresponding
1873 ;;; variable is set, then we call PROPAGATE-FROM-SETS. Otherwise, we
1874 ;;; consider substituting for the variable, and also propagate
1875 ;;; derived-type information for the arg to all the VAR's refs.
1876 ;;;
1877 ;;; Substitution is inhibited when the arg leaf's derived type isn't a
1878 ;;; subtype of the argument's leaf type. This prevents type checking
1879 ;;; from being defeated, and also ensures that the best representation
1880 ;;; for the variable can be used.
1881 ;;;
1882 ;;; Substitution of individual references is inhibited if the
1883 ;;; reference is in a different component from the home. This can only
1884 ;;; happen with closures over top level lambda vars. In such cases,
1885 ;;; the references may have already been compiled, and thus can't be
1886 ;;; retroactively modified.
1887 ;;;
1888 ;;; If all of the variables are deleted (have no references) when we
1889 ;;; are done, then we delete the LET.
1890 ;;;
1891 ;;; Note that we are responsible for clearing the LVAR-REOPTIMIZE
1892 ;;; flags.
1907 ;; (NODE-DERIVED-TYPE USE) would
1908 ;; be better -- APD, 2003-05-15
1915 t)
1918 nil)))
1919 leaf var)))
1920 t)))))
1932 ;;; This function is called when one of the args to a non-LET local
1933 ;;; call changes. For each changed argument corresponding to an unset
1934 ;;; variable, we compute the union of the types across all calls and
1935 ;;; propagate this type information to the var's refs.
1936 ;;;
1937 ;;; If the function has an entry-fun, then we don't do anything: since
1938 ;;; it has a XEP we would not discover anything.
1939 ;;;
1940 ;;; If the function is an optional-entry-point, we will just make sure
1941 ;;; &REST lists are known to be lists. Doing the regular rigamarole
1942 ;;; can erronously propagate too strict types into refs: see
1943 ;;; BUG-655203-REGRESSION in tests/compiler.pure.lisp.
1944 ;;;
1945 ;;; We can clear the LVAR-REOPTIMIZE flags for arguments in all calls
1946 ;;; corresponding to changed arguments in CALL, since the only use in
1947 ;;; IR1 optimization of the REOPTIMIZE flag for local call args is
1948 ;;; right here.
1953 ;; We can still make sure &REST is known to be a list.
1958 ;; The normal case.
1966 vars))
1982 union)))))
1985 and type in union
1989 \f
1990 ;;;; multiple values optimization
1992 ;;; Do stuff to notice a change to a MV combination node. There are
1993 ;;; two main branches here:
1994 ;;; -- If the call is local, then it is already a MV let, or should
1995 ;;; become one. Note that although all :LOCAL MV calls must eventually
1996 ;;; be converted to :MV-LETs, there can be a window when the call
1997 ;;; is local, but has not been LET converted yet. This is because
1998 ;;; the entry-point lambdas may have stray references (in other
1999 ;;; entry points) that have not been deleted yet.
2000 ;;; -- The call is full. This case is somewhat similar to the non-MV
2001 ;;; combination optimization: we propagate return type information and
2002 ;;; notice non-returning calls. We also have an optimization
2003 ;;; which tries to convert MV-CALLs into MV-binds.
2009 (:local
2014 do
2019 (:full
2041 ;;; Propagate derived type info from the values lvar to the vars.
2052 and type in types
2057 do
2072 sum nvals)))
2076 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2077 at least ~R."
2078 total-nvals min)
2083 "MULTIPLE-VALUE-CALL with ~R values when the function expects ~
2084 at most ~R."
2085 total-nvals max)
2091 min)
2109 ;;; If we see:
2110 ;;; (multiple-value-bind
2111 ;;; (x y)
2112 ;;; (values xx yy)
2113 ;;; ...)
2114 ;;; Convert to:
2115 ;;; (let ((x xx)
2116 ;;; (y yy))
2117 ;;; ...)
2118 ;;;
2119 ;;; What we actually do is convert the VALUES combination into a
2120 ;;; normal LET combination calling the original :MV-LET lambda. If
2121 ;;; there are extra args to VALUES, discard the corresponding
2122 ;;; lvars. If there are insufficient args, insert references to NIL.
2146 for prev = node-prev then ctran
2152 use ctran))
2166 ;; Propagate derived types from the VALUES call to its args:
2167 ;; transforms can leave the VALUES call with a better type
2168 ;; than its args have, so make sure not to throw that away.
2174 ;; Propagate declared types of MV-BIND variables.
2177 t)))
2179 ;;; If we see:
2180 ;;; (values-list (list x y z))
2181 ;;;
2182 ;;; Convert to:
2183 ;;; (values x y z)
2184 ;;;
2185 ;;; In implementation, this is somewhat similar to
2186 ;;; CONVERT-MV-BIND-TO-LET. We grab the args of LIST and make them
2187 ;;; args of the VALUES-LIST call, flushing the old argument lvar
2188 ;;; (allowing the LIST to be flushed.)
2189 ;;;
2190 ;;; FIXME: Thus we lose possible type assertions on (LIST ...).
2197 ;; FIXME: VALUES might not satisfy an assertion on NODE-LVAR.
2209 t)))
2211 ;;; If VALUES appears in a non-MV context, then effectively convert it
2212 ;;; to a PROG1. This allows the computation of the additional values
2213 ;;; to become dead code.
2225 val))
2226 nil))
2228 ;;; TODO:
2229 ;;; - CAST chains;
2248 ;; Vestigial exits are only introduced when eliminating a local
2249 ;; RETURN-FROM. We may delete them only when we can show that
2250 ;; there are no other code paths that use the entry LVAR that
2251 ;; are live from within the block that contained the deleted
2252 ;; EXIT (our predecessor block) and that all uses of the entry
2253 ;; LVAR have the same dynamic-extent environment. The
2254 ;; conservative version of this is that there are no EXITs for
2255 ;; any ENTRY introduced between the LEXENV of the deleted EXIT
2256 ;; and the LEXENV of the target ENTRY and that both the LEXENV
2257 ;; of the deleted EXIT and the LEXENV of the ENTRY have the same
2258 ;; set of :DYNAMIC-EXTENT variables and functions.
2272 ;; (CDAR VARS) is the actual variable we're looking at,
2273 ;; except that if it's a symbol-macro then it's (MACRO
2274 ;; . <form>), and that's not a leaf, so ignore the
2275 ;; non-LEAF-P variables.
2281 ;; Like with VARS, we can have FUNS of the form (MACRO
2282 ;; . <something>), though in this case it's an expander
2283 ;; function. Again, ignore the non-LEAF-P functions.
2290 #+sb-xc-host context
2322 lvar)
2323 ;; Pathwise removing of CAST
2326 next-block)
2360 ;; No need to transform into an analog of
2361 ;; %COMPILE-TIME-TYPE-ERROR, %CHECK-BOUND will signal at
2362 ;; run-time and %CHECK-BOUND ir2-converter will signal at
2363 ;; compile-time if it survives further stages of ir1
2364 ;; optimization.
2367 ;; FIXME: Do it in one step.
2371 value
2377 ;; FIXME: Derived type.
2381 ',detail
2383 ;; KLUDGE: FILTER-LVAR does not work for non-returning
2384 ;; functions, so we declare the return type of
2385 ;; %COMPILE-TIME-TYPE-ERROR to be * and derive the real type
2386 ;; here.
2390 ;; FIXME: Is it necessary?