1 {-# LANGUAGE OverloadedStrings #-}
2 {-# LANGUAGE ForeignFunctionInterface #-}
3 module Mate.X86CodeGen where
8 import qualified Data.Map as M
9 import qualified Data.Set as S
10 import qualified Data.ByteString.Lazy as B
14 import Foreign.C.Types
18 import qualified JVM.Assembler as J
19 import JVM.Assembler hiding (Instruction)
23 import Harpy.X86Disassembler
25 import Mate.BasicBlocks
30 foreign import ccall "dynamic"
31 code_int :: FunPtr (CInt -> CInt -> IO CInt) -> (CInt -> CInt -> IO CInt)
33 foreign import ccall "getaddr"
36 foreign import ccall "getMallocAddr"
37 getMallocAddr :: CUInt
39 foreign import ccall "callertrap"
42 foreign import ccall "register_signal"
43 register_signal :: IO ()
45 test_01, test_02, test_03 :: IO ()
48 (entry, end) <- testCase "./tests/Fib" "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" "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" "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 :: B.ByteString -> B.ByteString -> IO (Ptr Word8, Int)
96 testCase cf method = do
97 cls <- getClassFile cf
98 hmap <- parseMethod cls method
101 Nothing -> error "sorry, no code generation"
103 let ebb = emitFromBB method 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, TMap)
122 emitFromBB :: B.ByteString -> Class Resolved -> MapBB -> CodeGen e s (CompileInfo, [Instruction])
123 emitFromBB method cls hmap = do
124 llmap <- sequence [newNamedLabel ("bb_" ++ show x) | (x,_) <- M.toList hmap]
125 let lmap = zip (Prelude.fst $ unzip $ M.toList hmap) llmap
129 -- TODO(bernhard): determine a reasonable value.
130 -- e.g. (locals used) * 4
131 sub esp (0x60 :: Word32)
133 (calls, bbstarts) <- efBB (0,(hmap M.! 0)) M.empty M.empty lmap
136 return ((ep, bbstarts, end, calls), d)
138 getLabel :: BlockID -> [(BlockID, Label)] -> Label
139 getLabel _ [] = error "label not found!"
140 getLabel i ((x,l):xs) = if i==x then l else getLabel i xs
142 efBB :: (BlockID, BasicBlock) -> TMap -> BBStarts -> [(BlockID, Label)] -> CodeGen e s (TMap, BBStarts)
143 efBB (bid, bb) calls bbstarts lmap =
144 if M.member bid bbstarts then
145 return (calls, bbstarts)
147 bb_offset <- getCodeOffset
148 let bbstarts' = M.insert bid bb_offset bbstarts
149 defineLabel $ getLabel bid lmap
150 cs <- mapM emit' $ code bb
151 let calls' = calls `M.union` (M.fromList $ catMaybes cs)
153 Return -> return (calls', bbstarts')
155 efBB (t, hmap M.! t) calls' bbstarts' lmap
157 efBB (t, hmap M.! t) calls' bbstarts' lmap
158 TwoTarget t1 t2 -> do
159 (calls'', bbstarts'') <- efBB (t1, hmap M.! t1) calls' bbstarts' lmap
160 efBB (t2, hmap M.! t2) calls'' bbstarts'' lmap
161 -- TODO(bernhard): also use metainformation
162 -- TODO(bernhard): implement `emit' as function which accepts a list of
163 -- instructions, so we can use patterns for optimizations
165 getCurrentOffset :: CodeGen e s (Word32)
166 getCurrentOffset = do
168 let w32_ep = (fromIntegral $ ptrToIntPtr ep) :: Word32
169 offset <- getCodeOffset
170 return $ w32_ep + (fromIntegral offset)
172 emit' :: J.Instruction -> CodeGen e s (Maybe (Word32, TrapInfo))
173 emit' (INVOKESPECIAL cpidx) = emit' (INVOKESTATIC cpidx)
174 emit' (INVOKESTATIC cpidx) = do
175 let l = buildMethodID cls cpidx
176 calladdr <- getCurrentOffset
177 newNamedLabel (show l) >>= defineLabel
178 -- causes SIGILL. in the signal handler we patch it to the acutal call.
179 -- place a nop at the end, therefore the disasm doesn't screw up
180 emit32 (0xffff9090 :: Word32) >> emit8 (0x90 :: Word8)
181 -- discard arguments on stack
182 let argcnt = (methodGetArgsCount cls cpidx) * 4
183 when (argcnt > 0) (add esp argcnt)
184 -- push result on stack if method has a return value
185 when (methodHaveReturnValue cls cpidx) (push eax)
186 return $ Just $ (calladdr, MI l)
187 emit' (PUTSTATIC cpidx) = do
189 trapaddr <- getCurrentOffset
190 mov (Addr 0x00000000) eax -- it's a trap
191 return $ Just $ (trapaddr, SFI $ buildFieldID cls cpidx)
192 emit' (GETSTATIC cpidx) = do
193 trapaddr <- getCurrentOffset
194 mov eax (Addr 0x00000000) -- it's a trap
196 return $ Just $ (trapaddr, SFI $ buildFieldID cls cpidx)
197 emit' insn = emit insn >> return Nothing
199 emit :: J.Instruction -> CodeGen e s ()
200 emit POP = do -- print dropped value
201 calladdr <- getCurrentOffset
202 -- '5' is the size of the `call' instruction ( + immediate)
203 let w32_calladdr = 5 + calladdr
204 let trapaddr = (fromIntegral getaddr :: Word32)
205 call (trapaddr - w32_calladdr)
206 add esp (4 :: Word32)
207 emit DUP = pop (Disp 0, esp)
208 emit (NEW objidx) = do
209 -- TODO(bernhard): determine right amount...
211 push (amount :: Word32)
212 calladdr <- getCurrentOffset
213 let w32_calladdr = 5 + calladdr
214 let malloaddr = (fromIntegral getMallocAddr :: Word32)
215 call (malloaddr - w32_calladdr)
216 add esp (4 :: Word32)
218 -- TODO(bernhard): save reference somewhere for GC
219 emit (BIPUSH val) = push ((fromIntegral val) :: Word32)
220 emit (SIPUSH val) = push ((fromIntegral $ ((fromIntegral val) :: Int16)) :: Word32)
221 emit (ICONST_0) = push (0 :: Word32)
222 emit (ICONST_1) = push (1 :: Word32)
223 emit (ICONST_2) = push (2 :: Word32)
224 emit (ICONST_4) = push (4 :: Word32)
225 emit (ICONST_5) = push (5 :: Word32)
226 emit (ALOAD_ x) = emit (ILOAD_ x)
228 push (Disp (cArgs_ x), ebp)
229 emit (ALOAD x) = emit (ILOAD x)
231 push (Disp (cArgs x), ebp)
232 emit (ASTORE_ x) = emit (ISTORE_ x)
233 emit (ISTORE_ x) = do
235 mov (Disp (cArgs_ x), ebp) eax
236 emit (ASTORE x) = emit (ISTORE x)
239 mov (Disp (cArgs x), ebp) eax
241 emit (GETFIELD x) = do
242 pop eax -- this pointer
243 push (Disp (fromIntegral x * 4), eax) -- get field
244 emit (PUTFIELD x) = do
245 pop ebx -- value to write
246 pop eax -- this pointer
247 mov (Disp (fromIntegral x * 4), eax) ebx -- set field
249 emit IADD = do pop ebx; pop eax; add eax ebx; push eax
250 emit ISUB = do pop ebx; pop eax; sub eax ebx; push eax
251 emit IMUL = do pop ebx; pop eax; mul ebx; push eax
252 emit (IINC x imm) = do
253 add (Disp (cArgs x), ebp) (s8_w32 imm)
255 emit (IF_ICMP cond _) = do
258 cmp ebx eax -- intel syntax is swapped (TODO(bernhard): test that plz)
259 let sid = case successor bb of TwoTarget _ t -> t; _ -> error "bad"
260 let l = getLabel sid lmap
262 C_EQ -> je l; C_NE -> jne l
263 C_LT -> jl l; C_GT -> jg l
264 C_GE -> jge l; C_LE -> jle l
266 emit (IF cond _) = do
268 cmp eax (0 :: Word32) -- TODO(bernhard): test that plz
269 let sid = case successor bb of TwoTarget _ t -> t; _ -> error "bad"
270 let l = getLabel sid lmap
272 C_EQ -> je l; C_NE -> jne l
273 C_LT -> jl l; C_GT -> jg l
274 C_GE -> jge l; C_LE -> jle l
277 let sid = case successor bb of OneTarget t -> t; _ -> error "bad"
278 jmp $ getLabel sid lmap
280 emit RETURN = do mov esp ebp; pop ebp; ret
286 emit invalid = error $ "insn not implemented yet: " ++ (show invalid)
288 -- for locals we use a different storage
289 cArgs :: Word8 -> Word32
290 cArgs x = if (x' >= thisMethodArgCnt)
291 -- TODO(bernhard): maybe s/(-4)/(-8)/
292 then fromIntegral $ (-4) * (x' - thisMethodArgCnt + 1)
294 where x' = fromIntegral x
296 cArgs_ :: IMM -> Word32
297 cArgs_ x = cArgs $ case x of I0 -> 0; I1 -> 1; I2 -> 2; I3 -> 3
299 thisMethodArgCnt :: Word32
300 thisMethodArgCnt = isNonStatic + (fromIntegral $ length args)
302 (Just m) = lookupMethod method cls
303 (MethodSignature args _) = methodSignature m
304 isNonStatic = if S.member ACC_STATIC (methodAccessFlags m)
306 else 1 -- one argument for the this pointer
309 -- sign extension from w8 to w32 (over s8)
310 -- unfortunately, hs-java is using Word8 everywhere (while
311 -- it should be Int8 actually)
312 s8_w32 :: Word8 -> Word32
313 s8_w32 w8 = fromIntegral s8
314 where s8 = (fromIntegral w8) :: Int8