X-Git-Url: http://wien.tomnetworks.com/gitweb/?a=blobdiff_plain;f=Mate%2FBasicBlocks.hs;h=2f96a40fe66ee6ce5b86b644141b3e3cf269b964;hb=HEAD;hp=f92059f8a90d787f298fc6f3ce09a963910152a4;hpb=f339dd7125784b27a4bbf3eff1b38780a59c04fe;p=mate.git diff --git a/Mate/BasicBlocks.hs b/Mate/BasicBlocks.hs index f92059f..2f96a40 100644 --- a/Mate/BasicBlocks.hs +++ b/Mate/BasicBlocks.hs @@ -1,159 +1,221 @@ {-# LANGUAGE OverloadedStrings #-} -module Mate.BasicBlocks where - -import Data.Binary +module Mate.BasicBlocks( + BlockID, + BasicBlock, + BBEnd, + MapBB, + Method, + printMapBB, + parseMethod, + testCFG -- added by hs to perform benches from outside + )where + +import Data.Binary hiding (get) import Data.Int -import qualified Data.Map as H -import System.Environment 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.Common import JVM.ClassFile import JVM.Converter -import JVM.Dump import JVM.Assembler -import Debug.Trace - +import Mate.Types +import Mate.Debug import Mate.Utilities -type Name = String -- use "virtual register id" instead? -data Type = JInt | JFloat -- add more -type Variable = (Type,Name) - -type BlockID = Int --- Represents a CFG node -data BasicBlock = BasicBlock { - -- inputs :: [Variable], - -- outputs :: [Variable], - code :: [Instruction], - successor :: BBEnd } - --- describes (leaving) edges of a CFG node -data BBEnd = Return | OneTarget BlockID | TwoTarget BlockID BlockID deriving Show - -type MapBB = H.Map BlockID BasicBlock - --- for immediate representation for determine BBs -type Offset = (Int, Maybe BBEnd) -- (offset in bytecode, offset to jump target) -type OffIns = (Offset, Instruction) - - -main = do - args <- getArgs - case args of - [clspath] -> parseFile clspath "fib" -- TODO - _ -> error "Synopsis: dump-class File.class" - - -parseFile :: String -> B.ByteString -> IO () -parseFile clspath method = do - --clsFile <- decodeFile clspath - --putStrLn $ showListIx $ M.assocs $ constsPool (clsFile :: Class Pointers) - cls <- parseClassFile clspath - --dumpClass cls - let mainmethod = lookupMethod method cls -- "main|([Ljava/lang/String;)V" cf - let hmap = testCFG mainmethod - putStr "BlockIDs: " - let keys = fst $ unzip $ H.toList hmap - mapM_ (putStr . (flip (++)) ", " . show) keys - putStrLn "\n\nBasicBlocks:" - printMapBB keys hmap - -printMapBB :: [BlockID] -> MapBB -> IO () -printMapBB [] _ = return () -printMapBB (i:is) hmap = case H.lookup i hmap of - Just bb -> do - putStrLn $ "Block " ++ (show i) - mapM_ putStrLn (map ((++) "\t" . show) $ code bb) - case successor bb of - Return -> putStrLn "" - 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." - -test_01 = parseFile "./tests/Fib.class" "fib" -test_02 = parseFile "./tests/While.class" "f" -test_03 = parseFile "./tests/While.class" "g" - - -testCFG :: Maybe (Method Resolved) -> MapBB -testCFG (Just m) = case attrByName m "Code" of - Nothing -> error "no code" - Just bytecode -> let code = decodeMethod bytecode - instructions = codeInstructions code - in buildCFG instructions -testCFG _ = error "no method to build cfg" - - -buildCFG :: [Instruction] -> MapBB -buildCFG xs = buildCFG' H.empty xs' xs' - where - xs' :: [OffIns] - xs' = calculateInstructionOffset xs +-- (offset in bytecode, offset to jump target, ins) +type OffIns = (Int, Maybe BBEnd, Instruction) -buildCFG' :: MapBB -> [OffIns] -> [OffIns] -> MapBB -buildCFG' hmap [] _ = hmap -buildCFG' hmap (((off, Just entry), _):xs) insns = buildCFG' (insertlist entryi hmap) xs insns - where - insertlist :: [BlockID] -> MapBB -> MapBB - insertlist [] hmap = hmap - insertlist (x:xs) hmap = insertlist xs newhmap - where - newhmap = if H.member x hmap then hmap else H.insert x value hmap - value = parseBasicBlock x insns +type Target = BlockID +type BBState = S.Set Target +type AnalyseState = State BBState [OffIns] - entryi :: [BlockID] - entryi = (if off == 0 then [0] else []) ++ -- also consider the entrypoint - case entry of - TwoTarget t1 t2 -> [t1, t2] - OneTarget t -> [t] - Return -> trace "should not happen" [] -buildCFG' hmap (((_, Nothing), _):xs) insns = buildCFG' hmap xs insns +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_01 + test_02 + test_03 + test_04 + +test_01, test_02, test_03, test_04 :: IO () +test_01 = testInstance "./tests/Fib.class" "fib" +test_02 = testInstance "./tests/While.class" "f" +test_03 = testInstance "./tests/While.class" "g" +test_04 = testInstance "./tests/Fac.class" "fac" +-} + + +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 = (&&&) g eHandlerPC ce + where + g ce' = case eCatchType ce' of + 0 -> B.empty + x -> buildClassID cls x + + 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 + 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 = zip offsets 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 + (offsets, _, 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 - GOTO w16 -> onetarget w16 - IRETURN -> 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 `addW16Signed` w16) (off + 3)), 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 . (:[])