1 {-# LANGUAGE OverloadedStrings #-}
2 {-# LANGUAGE ForeignFunctionInterface #-}
3 module Mate.X86CodeGen where
8 import qualified Data.Map as M
9 import qualified Data.ByteString.Lazy as B
12 import Foreign.C.Types
16 import qualified JVM.Assembler as J
17 import JVM.Assembler hiding (Instruction)
22 import Harpy.X86Disassembler
24 import Mate.BasicBlocks
27 foreign import ccall "dynamic"
28 code_int :: FunPtr (CInt -> CInt -> IO CInt) -> (CInt -> CInt -> IO CInt)
30 foreign import ccall "getaddr"
33 foreign import ccall "callertrap"
36 foreign import ccall "register_signal"
37 register_signal :: IO ()
39 foreign import ccall "get_cmap"
40 get_cmap :: IO (Ptr ())
42 foreign import ccall "set_cmap"
43 set_cmap :: Ptr () -> IO ()
45 test_01, test_02, test_03 :: IO ()
48 (entry, end) <- testCase "./tests/Fib.class" "fib"
49 let entryFuncPtr = ((castPtrToFunPtr entry) :: FunPtr (CInt -> CInt -> IO CInt))
52 result <- code_int entryFuncPtr x 0
53 let iresult :: Int; iresult = fromIntegral result
54 let kk :: String; kk = if iresult == (fib x) then "OK" else "FAIL (" ++ (show (fib x)) ++ ")"
55 printf "result of fib(%2d): %3d\t\t%s\n" (fromIntegral x :: Int) iresult kk
56 ) $ ([0..10] :: [CInt])
57 printf "patched disasm:\n"
58 Right newdisasm <- disassembleBlock entry end
59 mapM_ (putStrLn . showAtt) newdisasm
64 | otherwise = (fib (n - 1)) + (fib (n - 2))
68 (entry,_) <- testCase "./tests/While.class" "f"
69 let entryFuncPtr = ((castPtrToFunPtr entry) :: FunPtr (CInt -> CInt -> IO CInt))
70 result <- code_int entryFuncPtr 5 4
71 let iresult :: Int; iresult = fromIntegral result
72 let kk :: String; kk = if iresult == 15 then "OK" else "FAIL"
73 printf "result of f(5,4): %3d\t\t%s\n" iresult kk
75 result2 <- code_int entryFuncPtr 4 3
76 let iresult2 :: Int; iresult2 = fromIntegral result2
77 let kk2 :: String; kk2 = if iresult2 == 10 then "OK" else "FAIL"
78 printf "result of f(4,3): %3d\t\t%s\n" iresult2 kk2
82 (entry,_) <- testCase "./tests/While.class" "g"
83 let entryFuncPtr = ((castPtrToFunPtr entry) :: FunPtr (CInt -> CInt -> IO CInt))
84 result <- code_int entryFuncPtr 5 4
85 let iresult :: Int; iresult = fromIntegral result
86 let kk :: String; kk = if iresult == 15 then "OK" else "FAIL"
87 printf "result of g(5,4): %3d\t\t%s\n" iresult kk
89 result2 <- code_int entryFuncPtr 4 3
90 let iresult2 :: Int; iresult2 = fromIntegral result2
91 let kk2 :: String; kk2 = if iresult2 == 10 then "OK" else "FAIL"
92 printf "result of g(4,3): %3d\t\t%s\n" iresult2 kk2
95 testCase :: String -> B.ByteString -> IO (Ptr Word8, Int)
96 testCase cf method = do
97 cls <- parseClassFile cf
98 hmap <- parseMethod cls method
101 Nothing -> error "sorry, no code generation"
103 let ebb = emitFromBB cls hmap'
104 (_, Right ((entry, bbstarts, end, _), disasm)) <- runCodeGen ebb () ()
105 let int_entry = ((fromIntegral $ ptrToIntPtr entry) :: Int)
107 mapM_ (putStrLn . showAtt) disasm
108 printf "basicblocks addresses:\n"
109 let b = map (\(x,y) -> (x,y + int_entry)) $ M.toList bbstarts
110 mapM_ (\(x,y) -> printf "\tBasicBlock %2d starts at 0x%08x\n" x y) b
113 type EntryPoint = Ptr Word8
114 type EntryPointOffset = Int
115 type PatchInfo = (BlockID, EntryPointOffset)
117 type BBStarts = M.Map BlockID Int
119 type CompileInfo = (EntryPoint, BBStarts, Int, CMap)
121 -- B.ByteString: encoded name: <Class>.<methodname><signature>
122 -- Class Resolved: classfile
123 -- Word16: index of invoke-instruction
124 type MethodInfo = (B.ByteString, Class Resolved, Word16)
126 -- Word32 = point of method call in generated code
127 -- MethodInfo = relevant information about callee
128 type CMap = M.Map Word32 MethodInfo
131 emitFromBB :: Class Resolved -> MapBB -> CodeGen e s (CompileInfo, [Instruction])
132 emitFromBB cls hmap = do
133 llmap <- sequence [newNamedLabel ("bb_" ++ show x) | (x,_) <- M.toList hmap]
134 let lmap = zip (Prelude.fst $ unzip $ M.toList hmap) llmap
139 (calls, bbstarts) <- efBB (0,(hmap M.! 0)) M.empty M.empty lmap
142 return ((ep, bbstarts, end, calls), d)
144 getLabel :: BlockID -> [(BlockID, Label)] -> Label
145 getLabel _ [] = error "label not found!"
146 getLabel i ((x,l):xs) = if i==x then l else getLabel i xs
148 efBB :: (BlockID, BasicBlock) -> CMap -> BBStarts -> [(BlockID, Label)] -> CodeGen e s (CMap, BBStarts)
149 efBB (bid, bb) calls bbstarts lmap =
150 if M.member bid bbstarts then
151 return (calls, bbstarts)
153 bb_offset <- getCodeOffset
154 let bbstarts' = M.insert bid bb_offset bbstarts
155 defineLabel $ getLabel bid lmap
156 cs <- mapM emit' $ code bb
157 let calls' = calls `M.union` (M.fromList $ catMaybes cs)
159 Return -> return (calls', bbstarts')
161 efBB (t, hmap M.! t) calls' bbstarts' lmap
162 TwoTarget t1 t2 -> do
163 (calls'', bbstarts'') <- efBB (t1, hmap M.! t1) calls' bbstarts' lmap
164 efBB (t2, hmap M.! t2) calls'' bbstarts'' lmap
165 -- TODO(bernhard): also use metainformation
166 -- TODO(bernhard): implement `emit' as function which accepts a list of
167 -- instructions, so we can use patterns for optimizations
169 emit' :: J.Instruction -> CodeGen e s (Maybe (Word32, MethodInfo))
170 emit' (INVOKESTATIC cpidx) = do
172 let w32_ep = (fromIntegral $ ptrToIntPtr ep) :: Word32
173 let l = buildMethodID cls cpidx
174 calladdr <- getCodeOffset
175 let w32_calladdr = w32_ep + (fromIntegral calladdr) :: Word32
176 newNamedLabel (toString l) >>= defineLabel
177 -- TODO(bernhard): better try SIGILL instead of SIGSEGV?
179 -- discard arguments (TODO(bernhard): don't hardcode it)
180 add esp (4 :: Word32)
181 -- push result on stack (TODO(bernhard): if any)
183 return $ Just $ (w32_calladdr, (l, cls, cpidx))
184 emit' insn = emit insn >> return Nothing
186 emit :: J.Instruction -> CodeGen e s ()
187 emit POP = do -- print dropped value
189 let w32_ep = (fromIntegral $ ptrToIntPtr ep) :: Word32
190 -- '5' is the size of the `call' instruction ( + immediate)
191 calladdr <- getCodeOffset
192 let w32_calladdr = 5 + w32_ep + (fromIntegral calladdr) :: Word32
193 let trapaddr = (fromIntegral getaddr :: Word32)
194 call (trapaddr - w32_calladdr)
195 emit (BIPUSH val) = push ((fromIntegral val) :: Word32)
196 emit (ICONST_1) = push (1 :: Word32)
197 emit (ICONST_2) = push (2 :: Word32)
198 emit (ICONST_5) = push (5 :: Word32)
200 push (Disp (cArgs_ x), ebp)
201 emit (ISTORE_ x) = do
203 mov (Disp (cArgs_ x), ebp) eax
204 emit IADD = do pop ebx; pop eax; add eax ebx; push eax
205 emit ISUB = do pop ebx; pop eax; sub eax ebx; push eax
206 emit (IINC x imm) = do
207 add (Disp (cArgs x), ebp) (s8_w32 imm)
209 emit (IF_ICMP cond _) = do
212 cmp ebx eax -- intel syntax is swapped (TODO(bernhard): test that plz)
213 let sid = case successor bb of TwoTarget _ t -> t; _ -> error "bad"
214 let l = getLabel sid lmap
216 C_EQ -> je l; C_NE -> jne l
217 C_LT -> jl l; C_GT -> jg l
218 C_GE -> jge l; C_LE -> jle l
220 emit (IF cond _) = do
222 cmp eax (0 :: Word32) -- TODO(bernhard): test that plz
223 let sid = case successor bb of TwoTarget _ t -> t; _ -> error "bad"
224 let l = getLabel sid lmap
226 C_EQ -> je l; C_NE -> jne l
227 C_LT -> jl l; C_GT -> jg l
228 C_GE -> jge l; C_LE -> jle l
231 let sid = case successor bb of OneTarget t -> t; _ -> error "bad"
232 jmp $ getLabel sid lmap
234 emit RETURN = do mov esp ebp; pop ebp; ret
240 emit _ = do cmovbe eax eax -- dummy
242 cArgs x = (8 + 4 * (fromIntegral x))
243 cArgs_ x = (8 + 4 * case x of I0 -> 0; I1 -> 1; I2 -> 2; I3 -> 3)
245 -- sign extension from w8 to w32 (over s8)
246 -- unfortunately, hs-java is using Word8 everywhere (while
247 -- it should be Int8 actually)
248 s8_w32 :: Word8 -> Word32
249 s8_w32 w8 = fromIntegral s8
250 where s8 = (fromIntegral w8) :: Int8