1 {-# LANGUAGE ScopedTypeVariables #-}
2 {-# OPTIONS -Wall -fno-warn-name-shadowing #-}
4 ( AGraph, (<*>), emptyAGraph, withFreshLabel, withUnique
5 , mkMiddle, mkMiddles, mkLast, mkZTail, mkBranch, mkLabel, mkIfThenElse, mkWhileDo
7 , emptyGraph, graphOfMiddles, graphOfZTail
8 , lgraphOfAGraph, graphOfAGraph, labelAGraph
19 import Prelude hiding (zip, unzip, last)
22 -------------------------------------------------------------------------
23 -- GENERIC ZIPPER-BASED CONTROL-FLOW GRAPH (CONSTRUCTOR VIEW) --
24 -------------------------------------------------------------------------
28 You can think of an AGraph like this: it is the program built by
29 composing in sequence three kinds of nodes:
30 * Label nodes (e.g. L2:)
31 * Middle nodes (e.g. x = y*3)
32 * Last nodes (e.g. if b then goto L1 else goto L2)
34 The constructors mkLabel, mkMiddle, and mkLast build single-node
35 AGraphs of the indicated type. The composition operator <*> glues
36 AGraphs together in sequence (in constant time).
42 if x<10 then goto L1 else goto L2
47 Notice that the AGraph may begin without a label, and may end without
48 a control transfer. Control *always* falls through a label and middle
49 node, and *never* falls through a Last node.
51 A 'AGraph m l' is simply an abstract version of a 'Graph m l' from
52 module 'ZipCfg'. The only difference is that the 'AGraph m l'
53 supports a constant-time splicing operation, written infix <*>.
54 That splicing operation, together with the constructor functions in
55 this module (and with 'labelAGraph'), is the recommended way to build
56 large graphs. Each construction or splice has constant cost, and to
57 turn an AGraph into a Graph requires time linear in the number of
58 nodes and N log N in the number of basic blocks.
60 The splicing operation warrants careful explanation. Like a Graph, an
61 AGraph is a control-flow graph which begins with a distinguished,
62 unlabelled sequence of middle nodes called the *entry*. An unlabelled
63 graph may also end with a sequence of middle nodes called the *exit*.
64 The entry may fall straight through to the exit, or it may fall into
65 the rest of the graph, which may include arbitrary control flow.
67 Using ASCII art, here are examples of the two kinds of graph. On the
68 left, the entry and exit sequences are labelled A and B, where the
69 control flow in the middle is labelled X. On the right, there is no
85 The AGraph has these properties:
87 * A AGraph is opaque; nothing about its structure can be observed.
89 * A AGraph may be turned into a LGraph in time linear in the number
90 of nodes and O(N log N) in the number of basic blocks.
92 * Two AGraphs may be spliced in constant time by writing g1 <*> g2
94 There are two rules for splicing, depending on whether the left-hand
95 graph falls through. If it does, the rule is as follows:
102 | X | <*> | Y | = | X |
120 And in the case where the left-hand graph does not fall through, the
129 | X | <*> | Y | = | X |
143 In this case C will become unreachable and is lost; when such a graph
144 is converted into a data structure, the system will bleat about
145 unreachable code. Also it must be assumed that there are branches
146 from somewhere in X to labelled blocks in Y; otherwise Y and D are
147 unreachable as well. (However, it may be the case that X branches
148 into some third AGraph, which in turn branches into D; the
149 representation is agnostic on this point.)
154 (<*>) :: AGraph m l -> AGraph m l -> AGraph m l
156 -- | A graph is built up by splicing together graphs each containing a
157 -- single node (where a label is considered a 'first' node. The empty
158 -- graph is a left and right unit for splicing. All of the AGraph
159 -- constructors (even complex ones like 'mkIfThenElse', as well as the
160 -- splicing operation <*>, are constant-time operations.
162 emptyAGraph :: AGraph m l
163 mkLabel :: LastNode l =>
164 BlockId -> AGraph m l -- graph contains the label
165 mkMiddle :: m -> AGraph m l -- graph contains the node
166 mkLast :: (Outputable m, Outputable l, LastNode l) =>
167 l -> AGraph m l -- graph contains the node
169 -- | This function provides access to fresh labels without requiring
170 -- clients to be programmed monadically.
171 withFreshLabel :: String -> (BlockId -> AGraph m l) -> AGraph m l
172 withUnique :: (Unique -> AGraph m l) -> AGraph m l
175 outOfLine :: (LastNode l, Outputable m, Outputable l)
176 => AGraph m l -> AGraph m l
177 -- ^ The argument is an AGraph that has an
178 -- empty entry sequence and no exit sequence.
179 -- The result is a new AGraph that has an empty entry sequence
180 -- connected to an empty exit sequence, with the original graph
181 -- sitting to the side out-of-line.
183 -- Example: mkMiddle (x = 3)
184 -- <*> outOfLine (mkLabel L <*> ...stuff...)
185 -- <*> mkMiddle (y = x)
186 -- Control will flow directly from x=3 to y=x;
187 -- the block starting with L is "on the side".
189 -- N.B. algebraically forall g g' : g <*> outOfLine g' == outOfLine g' <*> g
193 -- below for convenience
194 mkMiddles :: [m] -> AGraph m l
195 mkZTail :: (Outputable m, Outputable l, LastNode l) => ZTail m l -> AGraph m l
196 mkBranch :: (Outputable m, Outputable l, LastNode l) => BlockId -> AGraph m l
198 -- | For the structured control-flow constructs, a condition is
199 -- represented as a function that takes as arguments the labels to
200 -- goto on truth or falsehood.
202 mkIfThenElse :: (Outputable m, Outputable l, LastNode l)
203 => (BlockId -> BlockId -> AGraph m l) -- branch condition
204 -> AGraph m l -- code in the 'then' branch
205 -> AGraph m l -- code in the 'else' branch
206 -> AGraph m l -- resulting if-then-else construct
208 mkWhileDo :: (Outputable m, Outputable l, LastNode l)
209 => (BlockId -> BlockId -> AGraph m l) -- loop condition
210 -> AGraph m l -- body of the bloop
211 -> AGraph m l -- the final while loop
213 -- | Converting an abstract graph to a concrete form is expensive: the
214 -- cost is linear in the number of nodes in the answer, plus N log N
215 -- in the number of basic blocks. The conversion is also monadic
216 -- because it may require the allocation of fresh, unique labels.
218 graphOfAGraph :: AGraph m l -> UniqSM (Graph m l)
219 lgraphOfAGraph :: AGraph m l -> UniqSM (LGraph m l)
220 -- ^ allocate a fresh label for the entry point
221 labelAGraph :: BlockId -> AGraph m l -> UniqSM (LGraph m l)
222 -- ^ use the given BlockId as the label of the entry point
225 -- | The functions below build Graphs directly; for convenience, they
226 -- are included here with the rest of the constructor functions.
228 emptyGraph :: Graph m l
229 graphOfMiddles :: [m] -> Graph m l
230 graphOfZTail :: ZTail m l -> Graph m l
233 -- ================================================================
235 -- ================================================================
237 newtype AGraph m l = AGraph (Graph m l -> UniqSM (Graph m l))
238 -- an AGraph is a monadic function from a successor Graph to a new Graph
240 AGraph f1 <*> AGraph f2 = AGraph f
241 where f g = f2 g >>= f1 -- note right associativity
243 emptyAGraph = AGraph return
245 graphOfAGraph (AGraph f) = f emptyGraph
246 emptyGraph = Graph (ZLast LastExit) emptyBlockEnv
249 do Graph tail blocks <- graphOfAGraph g
250 return $ LGraph id $ insertBlock (Block id tail) blocks
252 lgraphOfAGraph g = do id <- freshBlockId "graph entry"
255 -------------------------------------
258 mkLabel id = AGraph f
259 where f (Graph tail blocks) =
260 return $ Graph (ZLast (mkBranchNode id))
261 (insertBlock (Block id tail) blocks)
263 mkBranch target = mkLast $ mkBranchNode target
265 mkMiddle m = AGraph f
266 where f (Graph tail blocks) = return $ Graph (ZTail m tail) blocks
268 mkMiddles ms = AGraph f
269 where f (Graph tail blocks) = return $ Graph (foldr ZTail tail ms) blocks
271 graphOfMiddles ms = Graph (foldr ZTail (ZLast LastExit) ms) emptyBlockEnv
272 graphOfZTail t = Graph t emptyBlockEnv
276 where f (Graph tail blocks) =
277 do note_this_code_becomes_unreachable tail
278 return $ Graph (ZLast (LastOther l)) blocks
280 mkZTail tail = AGraph f
281 where f (Graph utail blocks) =
282 do note_this_code_becomes_unreachable utail
283 return $ Graph tail blocks
285 withFreshLabel name ofId = AGraph f
286 where f g = do id <- freshBlockId name
287 let AGraph f' = ofId id
290 withUnique ofU = AGraph f
291 where f g = do u <- getUniqueUs
292 let AGraph f' = ofU u
295 outOfLine (AGraph f) = AGraph f'
296 where f' (Graph tail' blocks') =
297 do Graph emptyEntrance blocks <- f emptyGraph
298 note_this_code_becomes_unreachable emptyEntrance
299 return $ Graph tail' (blocks `plusUFM` blocks')
302 mkIfThenElse cbranch tbranch fbranch =
303 withFreshLabel "end of if" $ \endif ->
304 withFreshLabel "start of then" $ \tid ->
305 withFreshLabel "start of else" $ \fid ->
307 mkLabel tid <*> tbranch <*> mkBranch endif <*>
308 mkLabel fid <*> fbranch <*> mkLabel endif
311 mkWhileDo cbranch body =
312 withFreshLabel "loop test" $ \test ->
313 withFreshLabel "loop head" $ \head ->
314 withFreshLabel "end while" $ \endwhile ->
315 -- Forrest Baskett's while-loop layout
316 mkBranch test <*> mkLabel head <*> body <*> mkLabel test
317 <*> cbranch head endwhile <*> mkLabel endwhile
320 -- | Bleat if the insertion of a last node will create unreachable code
321 note_this_code_becomes_unreachable ::
322 (Monad m, LastNode l, Outputable middle, Outputable l) => ZTail middle l -> m ()
325 note_this_code_becomes_unreachable = u
326 where u (ZLast LastExit) = return ()
327 u (ZLast (LastOther l)) | isBranchNode l = return ()
328 -- Note [Branch follows branch]
329 u tail = fail ("unreachable code: " ++ showSDoc (ppr tail))
331 note_this_code_becomes_unreachable = return ()
335 Note [Branch follows branch]
336 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
337 Why do we say it's ok for a Branch to follow a Branch?
338 Because the standard constructor mkLabel-- has fall-through
339 semantics. So if you do a mkLabel, you finish the current block,
340 giving it a label, and start a new one that branches to that label.
341 Emitting a Branch at this point is fine:
342 goto L1; L2: ...stuff...