{-# LANGUAGE OverloadedStrings #-}
module Mate.BasicBlocks(
BlockID,
- BasicBlock (..),
- BBEnd (..),
+ BasicBlock,
+ BBEnd,
MapBB,
+ Method,
printMapBB,
parseMethod,
- test_main,
testCFG -- added by hs to perform benches from outside
)where
-import Data.Binary
+import Data.Binary hiding (get)
import Data.Int
-import Data.List
import qualified Data.Map as M
+import qualified Data.Set as S
import qualified Data.ByteString.Lazy as B
+import Data.Maybe
+import Control.Monad.State
+import Control.Applicative
+import Control.Arrow
import JVM.ClassFile
import JVM.Converter
import JVM.Assembler
-import Mate.Utilities
import Mate.Types
+import Mate.Debug
+import Mate.Utilities
+-- (offset in bytecode, offset to jump target, ins)
+type OffIns = (Int, Maybe BBEnd, Instruction)
--- for immediate representation to determine BBs
-type Offset = (Int, Maybe BBEnd) -- (offset in bytecode, offset to jump target)
-type OffIns = (Offset, Instruction)
+type Target = BlockID
+type BBState = S.Set Target
+type AnalyseState = State BBState [OffIns]
-printMapBB :: Maybe MapBB -> IO ()
-printMapBB Nothing = putStrLn "No BasicBlock"
-printMapBB (Just hmap) = do
- putStr "BlockIDs: "
- let keys = fst $ unzip $ M.toList hmap
- mapM_ (putStr . (flip (++)) ", " . show) keys
- putStrLn "\n\nBasicBlocks:"
- printMapBB' keys hmap
- where
- printMapBB' :: [BlockID] -> MapBB -> IO ()
- printMapBB' [] _ = return ()
- printMapBB' (i:is) hmap' = case M.lookup i hmap' of
- Just bb -> do
- putStrLn $ "Block " ++ (show i)
- mapM_ putStrLn (map ((++) "\t" . show) $ code bb)
- case successor bb of
- Return -> putStrLn ""
- FallThrough t1 -> putStrLn $ "Sucessor: " ++ (show t1) ++ "\n"
- OneTarget t1 -> putStrLn $ "Sucessor: " ++ (show t1) ++ "\n"
- TwoTarget t1 t2 -> putStrLn $ "Sucessor: " ++ (show t1) ++ ", " ++ (show t2) ++ "\n"
- printMapBB' is hmap
- Nothing -> error $ "BlockID " ++ show i ++ " not found."
-
-testInstance :: String -> B.ByteString -> IO ()
-testInstance cf method = do
- cls <- parseClassFile cf
- hmap <- parseMethod cls method
- printMapBB hmap
+emptyBasicBlock :: BasicBlock
+emptyBasicBlock = BasicBlock
+ { code = []
+ , bblength = 0
+ , successor = Return }
+
+printMapBB :: MapBB -> IO ()
+printMapBB hmap = do
+ printfBb "BlockIDs: "
+ let keys = M.keys hmap
+ mapM_ (printfBb . flip (++) ", " . show) keys
+ printfBb "\n\nBasicBlocks:\n"
+ printMapBB' keys hmap
+ where
+ printMapBB' :: [BlockID] -> MapBB -> IO ()
+ printMapBB' [] _ = return ()
+ printMapBB' (i:is) hmap' = case M.lookup i hmap' of
+ Just bb -> do
+ printfBb $ "Block " ++ show i ++ ". len: " ++ (show $ bblength bb) ++ "\n"
+ mapM_ (printfBb . flip (++) "\n" . (++) "\t" . show) $ code bb
+ printfBb $ case successor bb of
+ Return -> ""
+ FallThrough t1 -> "Sucessor: " ++ show t1 ++ "\n"
+ OneTarget t1 -> "Sucessor: " ++ show t1 ++ "\n"
+ TwoTarget t1 t2 -> "Sucessor: " ++ show t1 ++ ", " ++ show t2 ++ "\n"
+ printMapBB' is hmap
+ Nothing -> error $ "BlockID " ++ show i ++ " not found."
+
+{-
+testInstance :: String -> B.ByteString -> MethodSignature -> IO ()
+testInstance cf method sig = do
+ cls <- parseClassFile cf
+ hmap <- parseMethod cls method sig
+ printMapBB hmap
test_main :: IO ()
test_main = do
test_02 = testInstance "./tests/While.class" "f"
test_03 = testInstance "./tests/While.class" "g"
test_04 = testInstance "./tests/Fac.class" "fac"
-
-
-parseMethod :: Class Resolved -> B.ByteString -> IO (Maybe MapBB)
-parseMethod cls method = do
- putStr "BB: analysing: "
- let msig = methodSignature $ (classMethods cls) !! 1
- B.putStrLn (method `B.append` ": " `B.append` (encode msig))
- return $ testCFG $ lookupMethod method cls
-
-
-testCFG :: Maybe (Method Resolved) -> Maybe MapBB
-testCFG (Just m) = case attrByName m "Code" of
- Nothing -> Nothing
- Just bytecode -> Just $ buildCFG $ codeInstructions $ decodeMethod bytecode
-testCFG _ = Nothing
-
-
-buildCFG :: [Instruction] -> MapBB
-buildCFG xs = buildCFG' M.empty xs' xs'
+-}
+
+
+parseMethod :: Class Direct -> B.ByteString -> MethodSignature -> IO RawMethod
+parseMethod cls methodname sig = do
+ let method = fromMaybe
+ (error $ "method " ++ (show . toString) methodname ++ " not found")
+ (lookupMethodSig methodname sig cls)
+ let codeseg = fromMaybe
+ (error $ "codeseg " ++ (show . toString) methodname ++ " not found")
+ (attrByName method "Code")
+ let decoded = decodeMethod codeseg
+ let mapbb = testCFG decoded
+ let locals = fromIntegral (codeMaxLocals decoded)
+ let stacks = fromIntegral (codeStackSize decoded)
+ let codelen = fromIntegral (codeLength decoded)
+ let methoddirect = methodInfoToMethod (MethodInfo methodname "" sig) cls
+ let isStatic = methodIsStatic methoddirect
+ let nametype = methodNameType methoddirect
+ let argscount = methodGetArgsCount nametype + (if isStatic then 0 else 1)
+
+ let exceptionMap :: ExceptionMap
+ exceptionMap = foldl f M.empty $ codeExceptions decoded
+ where
+ f emap ce =
+ if M.member key emap
+ then M.adjust (value:) key emap
+ else M.insert key [value] emap
+ where
+ key = (&&&) eStartPC eEndPC ce
+ value = (&&&) (buildClassID cls . eCatchType) eHandlerPC ce
+
+ let msig = methodSignature method
+ printfBb $ printf "BB: analysing \"%s\"\n" $ toString (methodname `B.append` ": " `B.append` encode msig)
+ printMapBB mapbb
+ return $ RawMethod mapbb exceptionMap locals stacks argscount codelen
+
+
+testCFG :: Code -> MapBB
+testCFG c = buildCFG (codeInstructions c) (codeExceptions c)
where
- xs' :: [OffIns]
- xs' = markBackwardTargets $ calculateInstructionOffset xs
-
--- get already calculated jmp-targets and mark the predecessor of the
--- target-instruction as "FallThrough". we just care about backwards
--- jumps here (forward jumps are handled in buildCFG')
-markBackwardTargets :: [OffIns] -> [OffIns]
-markBackwardTargets [] = []
-markBackwardTargets (x:[]) = [x]
-markBackwardTargets insns@(x@((x_off,x_bbend),x_ins):y@((y_off,_),_):xs) =
- (x_new):(markBackwardTargets (y:xs))
- where
- x_new = if isTarget then checkX y_off else x
- checkX w16 = case x_bbend of
- Just _ -> x -- already marked, don't change
- Nothing -> ((x_off, Just $ FallThrough w16), x_ins) -- mark previous insn
-
- -- look through all remaining insns in the stream if there is a jmp to `y'
- isTarget = case find cmpOffset insns of Just _ -> True; Nothing -> False
- cmpOffset ((_,(Just (OneTarget w16))),_) = w16 == y_off
- cmpOffset ((_,(Just (TwoTarget _ w16))),_) = w16 == y_off
- cmpOffset _ = False
-
-
-buildCFG' :: MapBB -> [OffIns] -> [OffIns] -> MapBB
-buildCFG' hmap [] _ = hmap
-buildCFG' hmap (((off, entry), _):xs) insns = buildCFG' (insertlist entryi hmap) xs insns
- where
- insertlist :: [BlockID] -> MapBB -> MapBB
- insertlist [] hmap' = hmap'
- insertlist (y:ys) hmap' = insertlist ys newhmap
- where
- newhmap = if M.member y hmap' then hmap' else M.insert y value hmap'
- value = parseBasicBlock y insns
-
- entryi :: [BlockID]
- entryi = (if off == 0 then [0] else []) ++ -- also consider the entrypoint
- case entry of
- Just (TwoTarget t1 t2) -> [t1, t2]
- Just (OneTarget t) -> [t]
- Just (FallThrough t) -> [t]
- Just (Return) -> []
- Nothing -> []
-
+ buildCFG :: [Instruction] -> [CodeException] -> MapBB
+ buildCFG xs excps = execState (mapM buildCFG' $ alltargets ++ handlerEntries) M.empty
+ where
+ (offins, targets) = runState (calculateInstructionOffset tryBlocks xs) S.empty
+ alltargets = S.toList $ S.insert 0 targets
+ tryBlocks = map (fromIntegral . eStartPC) excps
+ handlerEntries = map (fromIntegral . eHandlerPC) excps
+
+ buildCFG' :: Int -> State MapBB ()
+ buildCFG' off = do
+ let value = parseBasicBlock off offins
+ modify (M.insert off value)
parseBasicBlock :: Int -> [OffIns] -> BasicBlock
-parseBasicBlock i insns = BasicBlock insonly endblock
+parseBasicBlock i insns = emptyBasicBlock
+ { code = insonly
+ , bblength = lastoff - i + (insnLength lastins)
+ , successor = endblock }
where
- startlist = dropWhile (\((x,_),_) -> x < i) insns
- (Just ((_,(Just endblock)),_), is) = takeWhilePlusOne validins startlist
- insonly = snd $ unzip is
+ (lastblock, is) = takeWhilePlusOne validins omitins insns
+ (_, _, insonly) = unzip3 is
+ (lastoff, Just endblock, lastins) = fromJust lastblock
- -- also take last (non-matched) element and return it
- takeWhilePlusOne :: (a -> Bool) -> [a] -> (Maybe a,[a])
- takeWhilePlusOne _ [] = (Nothing,[])
- takeWhilePlusOne p (x:xs)
- | p x = let (lastins, list) = takeWhilePlusOne p xs in (lastins, (x:list))
- | otherwise = (Just x,[x])
+ -- also take last (non-matched) element and return it
+ takeWhilePlusOne :: (a -> Bool) -> (a -> Bool) -> [a] -> (Maybe a, [a])
+ takeWhilePlusOne _ _ [] = (Nothing, [])
+ takeWhilePlusOne p omit (x:xs)
+ | omit x = next
+ | p x = second (x:) next
+ | otherwise = (Just x, [x])
+ where
+ next = takeWhilePlusOne p omit xs
- validins :: ((Int, Maybe BBEnd), Instruction) -> Bool
- validins ((_,x),_) = case x of Just _ -> False; Nothing -> True
+ validins :: OffIns -> Bool
+ validins (_, x, _) = isNothing x
+ omitins :: OffIns -> Bool
+ omitins (off, _, _) = off < i
-calculateInstructionOffset :: [Instruction] -> [OffIns]
-calculateInstructionOffset = cio' (0, Nothing)
+
+calculateInstructionOffset :: [BlockID] -> [Instruction] -> AnalyseState
+calculateInstructionOffset exstarts = cio' 0
where
- newoffset :: Instruction -> Int -> Offset
- newoffset x off = (off + (fromIntegral $ B.length $ encodeInstructions [x]), Nothing)
-
- addW16Signed :: Int -> Word16 -> Int
- addW16Signed i w16 = i + (fromIntegral s16)
- where s16 = (fromIntegral w16) :: Int16
-
- cio' :: Offset -> [Instruction] -> [OffIns]
- cio' _ [] = []
- -- TODO(bernhard): add more instruction with offset (IF_ACMP, JSR, ...)
- cio' (off,_) (x:xs) = case x of
- IF _ w16 -> twotargets w16
- IF_ICMP _ w16 -> twotargets w16
- IF_ACMP _ w16 -> twotargets w16
- GOTO w16 -> onetarget w16
- IRETURN -> notarget
- ARETURN -> notarget
- RETURN -> notarget
- _ -> ((off, Nothing), x):next
- where
- notarget = ((off, Just Return), x):next
- onetarget w16 = ((off, Just $ OneTarget $ (off `addW16Signed` w16)), x):next
- twotargets w16 = ((off, Just $ TwoTarget (off + 3) (off `addW16Signed` w16)), x):next
- next = cio' (newoffset x off) xs
+ addW16Signed i w16 = i + fromIntegral s16
+ where s16 = fromIntegral w16 :: Int16
+
+ cio' :: Int -> [Instruction] -> AnalyseState
+ cio' _ [] = return $ []
+ cio' off (x:xs) = case x of
+ IF _ w16 -> twotargets w16
+ IF_ICMP _ w16 -> twotargets w16
+ IF_ACMP _ w16 -> twotargets w16
+ IFNONNULL w16 -> twotargets w16
+ IFNULL w16 -> twotargets w16
+ GOTO w16 -> onetarget w16
+ ATHROW -> notarget
+ IRETURN -> notarget
+ ARETURN -> notarget
+ RETURN -> notarget
+ _ -> if newoffset `elem` exstarts
+ then do
+ modify (S.insert newoffset)
+ ((off, Just $ OneTarget newoffset, x):) <$> next
+ else normalins
+ where
+ normalins = do
+ tailinsns <- next -- eval remaining instructions
+ isNextInsATarget <- (S.member newoffset) <$> get
+ let bbtyp = if isNextInsATarget
+ then Just $ FallThrough newoffset
+ else Nothing
+ return $ (off, bbtyp, x):tailinsns
+ notarget = ((off, Just Return, x):) <$> next
+ onetarget w16 = do
+ let jump = off `addW16Signed` w16
+ modify (S.insert jump)
+ ((off, Just $ OneTarget jump, x):) <$> next
+ twotargets w16 = do
+ let nojump = off + 3
+ modify (S.insert nojump)
+ let jump = off `addW16Signed` w16
+ modify (S.insert jump)
+ ((off, Just $ TwoTarget nojump jump, x):) <$> next
+ next = cio' newoffset xs
+ newoffset = off + insLen
+ insLen = insnLength x
+
+-- TODO(bernhard): does GHC memomize results? i.e. does it calculate the size
+-- of `NOP' only once?
+insnLength :: Num a => Instruction -> a
+insnLength = fromIntegral . B.length . encodeInstructions . (:[])