1 {-# LANGUAGE GADTs, NoMonoLocalBinds, FlexibleContexts, ViewPatterns #-}
2 -- Norman likes local bindings
3 -- If this module lives on I'd like to get rid of this flag in due course
5 {-# OPTIONS_GHC -fno-warn-warnings-deprecations #-}
6 {-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}
7 #if __GLASGOW_HASKELL__ >= 701
8 -- GHC 7.0.1 improved incomplete pattern warnings with GADTs
9 {-# OPTIONS_GHC -fwarn-incomplete-patterns #-}
14 , dualLiveLattice, dualLiveTransfers, dualLiveness
15 --, insertSpillsAndReloads --- XXX todo check live-in at entry against formals
16 , dualLivenessWithInsertion
19 , removeDeadAssignmentsAndReloads
27 import OptimizationFuel
30 import Outputable hiding (empty)
31 import qualified Outputable as PP
36 import Compiler.Hoopl hiding (Unique)
38 import Prelude hiding (succ, zip)
40 {- Note [Overview of spill/reload]
41 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
42 The point of this module is to insert spills and reloads to
43 establish the invariant that at a call (or at any proc point with
44 an established protocol) all live variables not expected in
45 registers are sitting on the stack. We use a backward analysis to
46 insert spills and reloads. It should be followed by a
47 forward transformation to sink reloads as deeply as possible, so as
48 to reduce register pressure.
50 A variable can be expected to be live in a register, live on the
51 stack, or both. This analysis ensures that spills and reloads are
52 inserted as needed to make sure that every live variable needed
53 after a call is available on the stack. Spills are pushed back to
54 their reaching definitions, but reloads are dropped wherever needed
55 and will have to be sunk by a later forward transformation.
58 data DualLive = DualLive { on_stack :: RegSet, in_regs :: RegSet }
60 dualUnion :: DualLive -> DualLive -> DualLive
61 dualUnion (DualLive s r) (DualLive s' r') =
62 DualLive (s `unionUniqSets` s') (r `unionUniqSets` r')
64 dualUnionList :: [DualLive] -> DualLive
65 dualUnionList ls = DualLive ss rs
66 where ss = unionManyUniqSets $ map on_stack ls
67 rs = unionManyUniqSets $ map in_regs ls
69 changeStack, changeRegs :: (RegSet -> RegSet) -> DualLive -> DualLive
70 changeStack f live = live { on_stack = f (on_stack live) }
71 changeRegs f live = live { in_regs = f (in_regs live) }
74 dualLiveLattice :: DataflowLattice DualLive
75 dualLiveLattice = DataflowLattice "variables live in registers and on stack" empty add
76 where empty = DualLive emptyRegSet emptyRegSet
77 add _ (OldFact old) (NewFact new) = (changeIf $ change1 || change2, DualLive stack regs)
78 where (change1, stack) = add1 (on_stack old) (on_stack new)
79 (change2, regs) = add1 (in_regs old) (in_regs new)
80 add1 old new = if sizeUniqSet join > sizeUniqSet old then (True, join) else (False, old)
81 where join = unionUniqSets old new
83 dualLivenessWithInsertion :: BlockSet -> CmmGraph -> FuelUniqSM CmmGraph
84 dualLivenessWithInsertion procPoints g =
85 liftM fst $ dataflowPassBwd g [] $ analRewBwd dualLiveLattice
86 (dualLiveTransfers (g_entry g) procPoints)
87 (insertSpillAndReloadRewrites g procPoints)
89 dualLiveness :: BlockSet -> CmmGraph -> FuelUniqSM (BlockEnv DualLive)
90 dualLiveness procPoints g =
91 liftM snd $ dataflowPassBwd g [] $ analBwd dualLiveLattice $ dualLiveTransfers (g_entry g) procPoints
93 dualLiveTransfers :: BlockId -> BlockSet -> (BwdTransfer CmmNode DualLive)
94 dualLiveTransfers entry procPoints = mkBTransfer3 first middle last
95 where first :: CmmNode C O -> DualLive -> DualLive
96 first (CmmEntry id) live = check live id $ -- live at procPoint => spill
97 if id /= entry && setMember id procPoints
98 then DualLive { on_stack = on_stack live `plusRegSet` in_regs live
99 , in_regs = emptyRegSet }
101 where check live id x = if id == entry then noLiveOnEntry id (in_regs live) x else x
103 middle :: CmmNode O O -> DualLive -> DualLive
104 middle m = changeStack updSlots
106 where -- Reuse middle of liveness analysis from CmmLive
107 updRegs = case getBTransfer3 xferLive of (_, middle, _) -> middle m
109 updSlots live = foldSlotsUsed reload (foldSlotsDefd spill live m) m
110 spill live s@(RegSlot r, _, _) = check s $ deleteFromRegSet live r
112 reload live s@(RegSlot r, _, _) = check s $ extendRegSet live r
114 check (RegSlot (LocalReg _ ty), o, w) x
115 | o == w && w == widthInBytes (typeWidth ty) = x
116 check _ _ = panic "middleDualLiveness unsupported: slices"
117 last :: CmmNode O C -> FactBase DualLive -> DualLive
118 last l fb = case l of
119 CmmBranch id -> lkp id
120 l@(CmmCall {cml_cont=Nothing}) -> changeRegs (gen l . kill l) empty
121 l@(CmmCall {cml_cont=Just k}) -> call l k
122 l@(CmmForeignCall {succ=k}) -> call l k
123 l@(CmmCondBranch _ t f) -> changeRegs (gen l . kill l) $ dualUnion (lkp t) (lkp f)
124 l@(CmmSwitch _ tbl) -> changeRegs (gen l . kill l) $ dualUnionList $ map lkp (catMaybes tbl)
125 where empty = fact_bot dualLiveLattice
126 lkp id = empty `fromMaybe` lookupFact id fb
127 call l k = DualLive (on_stack (lkp k)) (gen l emptyRegSet)
129 gen :: UserOfLocalRegs a => a -> RegSet -> RegSet
130 gen a live = foldRegsUsed extendRegSet live a
131 kill :: DefinerOfLocalRegs a => a -> RegSet -> RegSet
132 kill a live = foldRegsDefd deleteFromRegSet live a
134 insertSpillAndReloadRewrites :: CmmGraph -> BlockSet -> CmmBwdRewrite DualLive
135 insertSpillAndReloadRewrites graph procPoints = deepBwdRw3 first middle nothing
136 -- Beware: deepBwdRw with one polymorphic function seems more reasonable here,
137 -- but GHC miscompiles it, see bug #4044.
138 where first :: CmmNode C O -> Fact O DualLive -> CmmReplGraph C O
139 first e@(CmmEntry id) live = return $
140 if id /= (g_entry graph) && setMember id procPoints then
141 case map reload (uniqSetToList spill_regs) of
143 is -> Just $ mkFirst e <*> mkMiddles is
146 -- If we are splitting procedures, we need the LastForeignCall
147 -- to spill its results to the stack because they will only
148 -- be used by a separate procedure (so they can't stay in LocalRegs).
150 spill_regs = if splitting then in_regs live
151 else in_regs live `minusRegSet` defs
152 defs = case mapLookup id firstDefs of
154 Nothing -> emptyRegSet
155 -- A LastForeignCall may contain some definitions, which take place
156 -- on return from the function call. Therefore, we build a map (firstDefs)
157 -- from BlockId to the set of variables defined on return to the BlockId.
158 firstDefs = mapFold addLive emptyBlockMap (toBlockMap graph)
159 addLive :: CmmBlock -> BlockEnv RegSet -> BlockEnv RegSet
160 addLive b env = case lastNode b of
161 CmmForeignCall {succ=k, res=defs} -> add k (mkRegSet defs) env
163 add bid defs env = mapInsert bid defs'' env
164 where defs'' = case mapLookup bid env of
165 Just defs' -> timesRegSet defs defs'
168 middle :: CmmNode O O -> Fact O DualLive -> CmmReplGraph O O
169 middle (CmmAssign (CmmLocal reg) (CmmLoad (CmmStackSlot (RegSlot reg') _) _)) _ | reg == reg' = return Nothing
170 middle m@(CmmAssign (CmmLocal reg) _) live = return $
171 if reg `elemRegSet` on_stack live then -- must spill
172 my_trace "Spilling" (f4sep [text "spill" <+> ppr reg,
173 text "after"{-, ppr m-}]) $
174 Just $ mkMiddles $ [m, spill reg]
176 middle _ _ = return Nothing
178 nothing _ _ = return Nothing
180 regSlot :: LocalReg -> CmmExpr
181 regSlot r = CmmStackSlot (RegSlot r) (widthInBytes $ typeWidth $ localRegType r)
183 spill, reload :: LocalReg -> CmmNode O O
184 spill r = CmmStore (regSlot r) (CmmReg $ CmmLocal r)
185 reload r = CmmAssign (CmmLocal r) (CmmLoad (regSlot r) $ localRegType r)
187 removeDeadAssignmentsAndReloads :: BlockSet -> CmmGraph -> FuelUniqSM CmmGraph
188 removeDeadAssignmentsAndReloads procPoints g =
189 liftM fst $ dataflowPassBwd g [] $ analRewBwd dualLiveLattice
190 (dualLiveTransfers (g_entry g) procPoints)
192 where rewrites = deepBwdRw3 nothing middle nothing
193 -- Beware: deepBwdRw with one polymorphic function seems more reasonable here,
194 -- but GHC panics while compiling, see bug #4045.
195 middle :: CmmNode O O -> Fact O DualLive -> CmmReplGraph O O
196 middle (CmmAssign (CmmLocal reg') _) live | not (reg' `elemRegSet` in_regs live) = return $ Just emptyGraph
197 -- XXX maybe this should be somewhere else...
198 middle (CmmAssign lhs (CmmReg rhs)) _ | lhs == rhs = return $ Just emptyGraph
199 middle (CmmStore lhs (CmmLoad rhs _)) _ | lhs == rhs = return $ Just emptyGraph
200 middle _ _ = return Nothing
202 nothing _ _ = return Nothing
204 ----------------------------------------------------------------
205 --- Usage information
207 -- We decorate all register assignments with usage information,
208 -- that is, the maximum number of times the register is referenced
209 -- while it is live along all outgoing control paths. There are a few
212 -- - If a register goes dead, and then becomes live again, the usages
213 -- of the disjoint live range don't count towards the original range.
215 -- a = 1; // used once
217 -- a = 2; // used once
220 -- - A register may be used multiple times, but these all reside in
221 -- different control paths, such that any given execution only uses
222 -- it once. In that case, the usage count may still be 1.
224 -- a = 1; // used once
231 -- This policy corresponds to an inlining strategy that does not
232 -- duplicate computation but may increase binary size.
234 -- - If we naively implement a usage count, we have a counting to
235 -- infinity problem across joins. Furthermore, knowing that
236 -- something is used 2 or more times in one runtime execution isn't
237 -- particularly useful for optimizations (inlining may be beneficial,
238 -- but there's no way of knowing that without register pressure
242 -- // first iteration, b used once
243 -- // second iteration, b used twice
244 -- // third iteration ...
247 -- // b used zero times
249 -- There is an orthogonal question, which is that for every runtime
250 -- execution, the register may be used only once, but if we inline it
251 -- in every conditional path, the binary size might increase a lot.
252 -- But tracking this information would be tricky, because it violates
253 -- the finite lattice restriction Hoopl requires for termination;
254 -- we'd thus need to supply an alternate proof, which is probably
255 -- something we should defer until we actually have an optimization
256 -- that would take advantage of this. (This might also interact
257 -- strangely with liveness information.)
260 -- // a is used one time, but in X different paths
267 -- This analysis is very similar to liveness analysis; we just keep a
268 -- little extra info. (Maybe we should move it to CmmLive, and subsume
269 -- the old liveness analysis.)
271 data RegUsage = SingleUse | ManyUse
272 deriving (Ord, Eq, Show)
273 -- Absence in map = ZeroUse
276 -- minBound is bottom, maxBound is top, least-upper-bound is max
277 -- ToDo: Put this in Hoopl. Note that this isn't as useful as I
278 -- originally hoped, because you usually want to leave out the bottom
279 -- element when you have things like this put in maps. Maybe f is
280 -- useful on its own as a combining function.
281 boundedOrdLattice :: (Bounded a, Ord a) => String -> DataflowLattice a
282 boundedOrdLattice n = DataflowLattice n minBound f
283 where f _ (OldFact x) (NewFact y)
284 | x >= y = (NoChange, x)
285 | otherwise = (SomeChange, y)
288 -- Custom node type we'll rewrite to. CmmAssign nodes to local
289 -- registers are replaced with AssignLocal nodes.
290 data WithRegUsage n e x where
291 Plain :: n e x -> WithRegUsage n e x
292 AssignLocal :: LocalReg -> CmmExpr -> RegUsage -> WithRegUsage n O O
294 instance UserOfLocalRegs (n e x) => UserOfLocalRegs (WithRegUsage n e x) where
295 foldRegsUsed f z (Plain n) = foldRegsUsed f z n
296 foldRegsUsed f z (AssignLocal _ e _) = foldRegsUsed f z e
298 instance DefinerOfLocalRegs (n e x) => DefinerOfLocalRegs (WithRegUsage n e x) where
299 foldRegsDefd f z (Plain n) = foldRegsDefd f z n
300 foldRegsDefd f z (AssignLocal r _ _) = foldRegsDefd f z r
302 instance NonLocal n => NonLocal (WithRegUsage n) where
303 entryLabel (Plain n) = entryLabel n
304 successors (Plain n) = successors n
306 liftRegUsage :: Graph n e x -> Graph (WithRegUsage n) e x
307 liftRegUsage = mapGraph Plain
309 eraseRegUsage :: Graph (WithRegUsage CmmNode) e x -> Graph CmmNode e x
310 eraseRegUsage = mapGraph f
311 where f :: WithRegUsage CmmNode e x -> CmmNode e x
312 f (AssignLocal l e _) = CmmAssign (CmmLocal l) e
315 type UsageMap = UniqFM RegUsage
317 usageLattice :: DataflowLattice UsageMap
318 usageLattice = DataflowLattice "usage counts for registers" emptyUFM (joinUFM f)
319 where f _ (OldFact x) (NewFact y)
320 | x >= y = (NoChange, x)
321 | otherwise = (SomeChange, y)
323 -- We reuse the names 'gen' and 'kill', although we're doing something
324 -- slightly different from the Dragon Book
325 usageTransfer :: BwdTransfer (WithRegUsage CmmNode) UsageMap
326 usageTransfer = mkBTransfer3 first middle last
328 middle :: WithRegUsage CmmNode O O -> UsageMap -> UsageMap
329 middle n f = gen_kill n f
330 last :: WithRegUsage CmmNode O C -> FactBase UsageMap -> UsageMap
331 -- Checking for CmmCall/CmmForeignCall is unnecessary, because
332 -- spills/reloads have already occurred by the time we do this
334 -- XXX Deprecated warning is puzzling: what label are we
336 -- ToDo: With a bit more cleverness here, we can avoid
337 -- disappointment and heartbreak associated with the inability
338 -- to inline into CmmCall and CmmForeignCall by
339 -- over-estimating the usage to be ManyUse.
340 last n f = gen_kill n (joinOutFacts usageLattice n f)
341 gen_kill a = gen a . kill a
342 gen a f = foldRegsUsed increaseUsage f a
343 kill a f = foldRegsDefd delFromUFM f a
344 increaseUsage f r = addToUFM_C combine f r SingleUse
345 where combine _ _ = ManyUse
347 usageRewrite :: BwdRewrite FuelUniqSM (WithRegUsage CmmNode) UsageMap
348 usageRewrite = mkBRewrite3 first middle last
349 where first _ _ = return Nothing
350 middle :: Monad m => WithRegUsage CmmNode O O -> UsageMap -> m (Maybe (Graph (WithRegUsage CmmNode) O O))
351 middle (Plain (CmmAssign (CmmLocal l) e)) f
353 $ case lookupUFM f l of
354 Nothing -> emptyGraph
355 Just usage -> mkMiddle (AssignLocal l e usage)
356 middle _ _ = return Nothing
357 last _ _ = return Nothing
359 type CmmGraphWithRegUsage = GenCmmGraph (WithRegUsage CmmNode)
360 annotateUsage :: CmmGraph -> FuelUniqSM (CmmGraphWithRegUsage)
361 annotateUsage vanilla_g =
362 let g = modifyGraph liftRegUsage vanilla_g
363 in liftM fst $ dataflowPassBwd g [(g_entry g, fact_bot usageLattice)] $
364 analRewBwd usageLattice usageTransfer usageRewrite
366 ----------------------------------------------------------------
367 --- Assignment tracking
369 -- The idea is to maintain a map of local registers do expressions,
370 -- such that the value of that register is the same as the value of that
371 -- expression at any given time. We can then do several things,
372 -- as described by Assignment.
374 -- Assignment describes the various optimizations that are valid
375 -- at a given point in the program.
377 -- This assignment can always be inlined. It is cheap or single-use.
379 -- This assignment should be sunk down to its first use. (This will
380 -- increase code size if the register is used in multiple control flow
381 -- paths, but won't increase execution time, and the reduction of
382 -- register pressure is worth it.)
384 -- We cannot safely optimize occurrences of this local register. (This
385 -- corresponds to top in the lattice structure.)
388 -- Extract the expression that is being assigned to
389 xassign :: Assignment -> Maybe CmmExpr
390 xassign (AlwaysInline e) = Just e
391 xassign (AlwaysSink e) = Just e
392 xassign NeverOptimize = Nothing
394 -- Extracts the expression, but only if they're the same constructor
395 xassign2 :: (Assignment, Assignment) -> Maybe (CmmExpr, CmmExpr)
396 xassign2 (AlwaysInline e, AlwaysInline e') = Just (e, e')
397 xassign2 (AlwaysSink e, AlwaysSink e') = Just (e, e')
400 -- Note: We'd like to make decisions about "not optimizing" as soon as
401 -- possible, because this will make running the transfer function more
403 type AssignmentMap = UniqFM Assignment
405 assignmentLattice :: DataflowLattice AssignmentMap
406 assignmentLattice = DataflowLattice "assignments for registers" emptyUFM (joinUFM add)
407 where add _ (OldFact old) (NewFact new)
409 (NeverOptimize, _) -> (NoChange, NeverOptimize)
410 (_, NeverOptimize) -> (SomeChange, NeverOptimize)
411 (xassign2 -> Just (e, e'))
412 | e == e' -> (NoChange, old)
413 | otherwise -> (SomeChange, NeverOptimize)
414 _ -> (SomeChange, NeverOptimize)
416 -- Deletes sinks from assignment map, because /this/ is the place
417 -- where it will be sunk to.
418 deleteSinks :: UserOfLocalRegs n => n -> AssignmentMap -> AssignmentMap
419 deleteSinks n m = foldRegsUsed (adjustUFM f) m n
420 where f (AlwaysSink _) = NeverOptimize
423 -- Invalidates any expressions that use a register.
424 invalidateUsersOf :: CmmReg -> AssignmentMap -> AssignmentMap
425 -- foldUFM_Directly :: (Unique -> elt -> a -> a) -> a -> UniqFM elt -> a
426 invalidateUsersOf reg m = foldUFM_Directly f m m -- [foldUFM performance]
427 where f u (xassign -> Just e) m | reg `regUsedIn` e = addToUFM_Directly m u NeverOptimize
429 {- This requires the entire spine of the map to be continually rebuilt,
430 - which causes crazy memory usage!
431 invalidateUsersOf reg = mapUFM (invalidateUsers' reg)
432 where invalidateUsers' reg (xassign -> Just e) | reg `regUsedIn` e = NeverOptimize
433 invalidateUsers' _ old = old
436 -- Note [foldUFM performance]
437 -- These calls to fold UFM no longer leak memory, but they do cause
438 -- pretty killer amounts of allocation. So they'll be something to
439 -- optimize; we need an algorithmic change to prevent us from having to
440 -- traverse the /entire/ map continually.
442 middleAssignment :: WithRegUsage CmmNode O O -> AssignmentMap -> AssignmentMap
444 -- Algorithm for annotated assignments:
445 -- 1. Delete any sinking assignments that were used by this instruction
446 -- 2. Add the assignment to our list of valid local assignments with
447 -- the correct optimization policy.
448 -- 3. Look for all assignments that reference that register and
450 middleAssignment n@(AssignLocal r e usage) assign
451 = invalidateUsersOf (CmmLocal r) . add . deleteSinks n $ assign
452 where add m = addToUFM m r
454 SingleUse -> AlwaysInline e
456 decide CmmLit{} = AlwaysInline e
457 decide CmmReg{} = AlwaysInline e
458 decide CmmLoad{} = AlwaysSink e
459 decide CmmStackSlot{} = AlwaysSink e
460 decide CmmMachOp{} = AlwaysSink e
461 -- We'll always inline simple operations on the global
462 -- registers, to reduce register pressure: Sp - 4 or Hp - 8
463 -- EZY: Justify this optimization more carefully.
464 decide CmmRegOff{} = AlwaysInline e
466 -- Algorithm for unannotated assignments of global registers:
467 -- 1. Delete any sinking assignments that were used by this instruction
468 -- 2. Look for all assignments that reference this register and
470 middleAssignment (Plain n@(CmmAssign reg@(CmmGlobal _) _)) assign
471 = invalidateUsersOf reg . deleteSinks n $ assign
473 -- Algorithm for unannotated assignments of *local* registers: do
474 -- nothing (it's a reload, so no state should have changed)
475 middleAssignment (Plain (CmmAssign (CmmLocal _) _)) assign = assign
477 -- Algorithm for stores:
478 -- 1. Delete any sinking assignments that were used by this instruction
479 -- 2. Look for all assignments that load from memory locations that
480 -- were clobbered by this store and invalidate them.
481 middleAssignment (Plain n@(CmmStore lhs rhs)) assign
482 = let m = deleteSinks n assign
483 in foldUFM_Directly f m m -- [foldUFM performance]
484 where f u (xassign -> Just x) m | (lhs, rhs) `clobbers` (u, x) = addToUFM_Directly m u NeverOptimize
487 = mapUFM_Directly p . deleteSinks n $ assign
488 -- ToDo: There's a missed opportunity here: even if a memory
489 -- access we're attempting to sink gets clobbered at some
490 -- location, it's still /better/ to sink it to right before the
491 -- point where it gets clobbered. How might we do this?
492 -- Unfortunately, it's too late to change the assignment...
493 where p r (xassign -> Just x) | (lhs, rhs) `clobbers` (r, x) = NeverOptimize
497 -- Assumption: Unsafe foreign calls don't clobber memory
498 middleAssignment (Plain n@(CmmUnsafeForeignCall{})) assign
499 = foldRegsDefd (\m r -> addToUFM m r NeverOptimize) (deleteSinks n assign) n
501 middleAssignment (Plain (CmmComment {})) assign
505 -- * Writes using Hp do not overlap with any other memory locations
506 -- (An important invariant being relied on here is that we only ever
507 -- use Hp to allocate values on the heap, which appears to be the
508 -- case given hpReg usage, and that our heap writing code doesn't
509 -- do anything stupid like overlapping writes.)
510 -- * Stack slots do not overlap with any other memory locations
511 -- * Stack slots for different areas do not overlap
512 -- * Stack slots within the same area and different offsets may
513 -- overlap; we need to do a size check (see 'overlaps').
514 -- * Register slots only overlap with themselves. (But this shouldn't
515 -- happen in practice, because we'll fail to inline a reload across
517 -- * Non stack-slot stores always conflict with each other. (This is
518 -- not always the case; we could probably do something special for Hp)
519 clobbers :: (CmmExpr, CmmExpr) -- (lhs, rhs) of clobbering CmmStore
520 -> (Unique, CmmExpr) -- (register, expression) that may be clobbered
522 clobbers (CmmRegOff (CmmGlobal Hp) _, _) (_, _) = False
523 clobbers (CmmReg (CmmGlobal Hp), _) (_, _) = False
524 -- ToDo: Also catch MachOp case
525 clobbers (ss@CmmStackSlot{}, CmmReg (CmmLocal r)) (u, CmmLoad (ss'@CmmStackSlot{}) _)
526 | getUnique r == u, ss == ss' = False -- No-op on the stack slot (XXX: Do we need this special case?)
527 clobbers (CmmStackSlot (CallArea a) o, rhs) (_, expr) = f expr
528 where f (CmmLoad (CmmStackSlot (CallArea a') o') t)
529 = (a, o, widthInBytes (cmmExprWidth rhs)) `overlaps` (a', o', widthInBytes (typeWidth t))
530 f (CmmLoad e _) = containsStackSlot e
531 f (CmmMachOp _ es) = or (map f es)
533 -- Maybe there's an invariant broken if this actually ever
535 containsStackSlot (CmmLoad{}) = True -- load of a load, all bets off
536 containsStackSlot (CmmMachOp _ es) = or (map containsStackSlot es)
537 containsStackSlot (CmmStackSlot{}) = True
538 containsStackSlot _ = False
539 clobbers (CmmStackSlot (RegSlot l) _, _) (_, expr) = f expr
540 where f (CmmLoad (CmmStackSlot (RegSlot l') _) _) = l == l'
542 clobbers _ (_, e) = f e
543 where f (CmmLoad (CmmStackSlot _ _) _) = False
544 f (CmmLoad{}) = True -- conservative
545 f (CmmMachOp _ es) = or (map f es)
548 -- Check for memory overlapping.
555 type CallSubArea = (AreaId, Int, Int) -- area, offset, width
556 overlaps :: CallSubArea -> CallSubArea -> Bool
557 overlaps (a, _, _) (a', _, _) | a /= a' = False
558 overlaps (_, o, w) (_, o', w') =
561 in (s' < o) && (s < o) -- Not LTE, because [ I32 ][ I32 ] is OK
563 lastAssignment :: WithRegUsage CmmNode O C -> AssignmentMap -> [(Label, AssignmentMap)]
564 -- Variables are dead across calls, so invalidating all mappings is justified
565 lastAssignment (Plain (CmmCall _ (Just k) _ _ _)) assign = [(k, mapUFM (const NeverOptimize) assign)]
566 lastAssignment (Plain (CmmForeignCall {succ=k})) assign = [(k, mapUFM (const NeverOptimize) assign)]
567 lastAssignment l assign = map (\id -> (id, deleteSinks l assign)) $ successors l
569 assignmentTransfer :: FwdTransfer (WithRegUsage CmmNode) AssignmentMap
570 assignmentTransfer = mkFTransfer3 (flip const) middleAssignment ((mkFactBase assignmentLattice .) . lastAssignment)
572 assignmentRewrite :: FwdRewrite FuelUniqSM (WithRegUsage CmmNode) AssignmentMap
573 assignmentRewrite = mkFRewrite3 first middle last
575 first _ _ = return Nothing
576 middle :: WithRegUsage CmmNode O O -> AssignmentMap -> GenCmmReplGraph (WithRegUsage CmmNode) O O
577 middle (Plain m) assign = return $ rewrite assign (precompute assign m) mkMiddle m
578 middle (AssignLocal l e u) assign = return $ rewriteLocal assign (precompute assign (CmmAssign (CmmLocal l) e)) mkMiddle l e u
579 last (Plain l) assign = return $ rewrite assign (precompute assign l) mkLast l
580 -- Tuple is (inline?, reloads)
581 precompute assign n = foldRegsUsed f (False, []) n -- duplicates are harmless
582 where f (i, l) r = case lookupUFM assign r of
583 Just (AlwaysSink e) -> (i, (Plain (CmmAssign (CmmLocal r) e)):l)
584 Just (AlwaysInline _) -> (True, l)
585 Just NeverOptimize -> (i, l)
586 -- This case can show up when we have
587 -- limited optimization fuel.
589 rewrite _ (False, []) _ _ = Nothing
590 -- Note [CmmCall Inline Hack]
591 -- ToDo: Conservative hack: don't do any inlining on CmmCalls, since
592 -- the code produced here tends to be unproblematic and I need
593 -- to write lint passes to ensure that we don't put anything in
594 -- the arguments that could be construed as a global register by
595 -- some later translation pass. (For example, slots will turn
596 -- into dereferences of Sp). This is the same hack in spirit as
597 -- was in cmm/CmmOpt.hs. Fix this up to only bug out if certain
598 -- CmmExprs are involved.
599 -- ToDo: We miss an opportunity here, where all possible
600 -- inlinings should instead be sunk.
601 rewrite _ (True, []) _ n | not (inlinable n) = Nothing -- see [CmmCall Inline Hack]
602 rewrite assign (i, xs) mk n = Just $ mkMiddles xs <*> mk (Plain (inline i assign n))
604 rewriteLocal _ (False, []) _ _ _ _ = Nothing
605 rewriteLocal assign (i, xs) mk l e u = Just $ mkMiddles xs <*> mk n'
606 where n' = AssignLocal l e' u
607 e' = if i then wrapRecExp (inlineExp assign) e else e
608 -- inlinable check omitted, since we can always inline into
611 inline :: Bool -> AssignmentMap -> CmmNode e x -> CmmNode e x
613 inline True _ n | not (inlinable n) = n -- see [CmmCall Inline Hack]
614 inline True assign n = mapExpDeep (inlineExp assign) n
616 inlineExp assign old@(CmmReg (CmmLocal r))
617 = case lookupUFM assign r of
618 Just (AlwaysInline x) -> x
620 inlineExp assign old@(CmmRegOff (CmmLocal r) i)
621 = case lookupUFM assign r of
622 Just (AlwaysInline x) ->
624 (CmmRegOff r' i') -> CmmRegOff r' (i + i')
625 _ -> CmmMachOp (MO_Add rep) [x, CmmLit (CmmInt (fromIntegral i) rep)]
626 where rep = typeWidth (localRegType r)
628 inlineExp _ old = old
630 inlinable :: CmmNode e x -> Bool
631 inlinable (CmmCall{}) = False
632 inlinable (CmmForeignCall{}) = False
635 rewriteAssignments :: CmmGraph -> FuelUniqSM CmmGraph
636 rewriteAssignments g = do
637 g' <- annotateUsage g
638 g'' <- liftM fst $ dataflowPassFwd g' [(g_entry g, fact_bot assignmentLattice)] $
639 analRewFwd assignmentLattice assignmentTransfer assignmentRewrite
640 return (modifyGraph eraseRegUsage g'')
642 ---------------------
645 ppr_regs :: String -> RegSet -> SDoc
646 ppr_regs s regs = text s <+> commafy (map ppr $ uniqSetToList regs)
647 where commafy xs = hsep $ punctuate comma xs
649 instance Outputable DualLive where
650 ppr (DualLive {in_regs = regs, on_stack = stack}) =
651 if isEmptyUniqSet regs && isEmptyUniqSet stack then
652 text "<nothing-live>"
654 nest 2 $ fsep [if isEmptyUniqSet regs then PP.empty
655 else (ppr_regs "live in regs =" regs),
656 if isEmptyUniqSet stack then PP.empty
657 else (ppr_regs "live on stack =" stack)]
659 -- ToDo: Outputable instance for UsageMap and AssignmentMap
661 my_trace :: String -> SDoc -> a -> a
662 my_trace = if False then pprTrace else \_ _ a -> a
664 f4sep :: [SDoc] -> SDoc
666 f4sep (d:ds) = fsep (d : map (nest 4) ds)