2 {-# LANGUAGE OverloadedStrings #-}
3 {-# LANGUAGE ForeignFunctionInterface #-}
5 module Mate.X86CodeGen where
8 import Data.BinaryState
11 import qualified Data.Map as M
12 import qualified Data.Set as S
13 import qualified Data.ByteString.Lazy as B
16 import Foreign hiding (xor)
17 import Foreign.C.Types
19 import qualified JVM.Assembler as J
20 import JVM.Assembler hiding (Instruction)
24 import Harpy.X86Disassembler
26 import Mate.BasicBlocks
33 foreign import ccall "dynamic"
34 code_int :: FunPtr (CInt -> CInt -> IO CInt) -> (CInt -> CInt -> IO CInt)
36 foreign import ccall "getMallocObjectAddr"
37 getMallocObjectAddr :: CUInt
39 foreign import ccall "register_signal"
40 register_signal :: IO ()
43 type EntryPoint = Ptr Word8
44 type EntryPointOffset = Int
45 type PatchInfo = (BlockID, EntryPointOffset)
47 type BBStarts = M.Map BlockID Int
49 type CompileInfo = (EntryPoint, BBStarts, Int, TrapMap)
52 emitFromBB :: B.ByteString -> Class Resolved -> MapBB -> CodeGen e s (CompileInfo, [Instruction])
53 emitFromBB method cls hmap = do
54 llmap <- sequence [newNamedLabel ("bb_" ++ show x) | (x,_) <- M.toList hmap]
55 let lmap = zip (Prelude.fst $ unzip $ M.toList hmap) llmap
59 -- TODO(bernhard): determine a reasonable value.
60 -- e.g. (locals used) * 4
61 sub esp (0x60 :: Word32)
63 (calls, bbstarts) <- efBB (0,(hmap M.! 0)) M.empty M.empty lmap
66 return ((ep, bbstarts, end, calls), d)
68 getLabel :: BlockID -> [(BlockID, Label)] -> Label
69 getLabel _ [] = error "label not found!"
70 getLabel i ((x,l):xs) = if i==x then l else getLabel i xs
72 efBB :: (BlockID, BasicBlock) -> TrapMap -> BBStarts -> [(BlockID, Label)] -> CodeGen e s (TrapMap, BBStarts)
73 efBB (bid, bb) calls bbstarts lmap =
74 if M.member bid bbstarts then
75 return (calls, bbstarts)
77 bb_offset <- getCodeOffset
78 let bbstarts' = M.insert bid bb_offset bbstarts
79 defineLabel $ getLabel bid lmap
80 cs <- mapM emit' $ code bb
81 let calls' = calls `M.union` (M.fromList $ catMaybes cs)
83 Return -> return (calls', bbstarts')
85 efBB (t, hmap M.! t) calls' bbstarts' lmap
87 efBB (t, hmap M.! t) calls' bbstarts' lmap
89 (calls'', bbstarts'') <- efBB (t1, hmap M.! t1) calls' bbstarts' lmap
90 efBB (t2, hmap M.! t2) calls'' bbstarts'' lmap
91 -- TODO(bernhard): also use metainformation
92 -- TODO(bernhard): implement `emit' as function which accepts a list of
93 -- instructions, so we can use patterns for optimizations
95 getCurrentOffset :: CodeGen e s (Word32)
98 let w32_ep = (fromIntegral $ ptrToIntPtr ep) :: Word32
99 offset <- getCodeOffset
100 return $ w32_ep + (fromIntegral offset)
102 emitInvoke :: Word16 -> Bool -> CodeGen e s (Maybe (Word32, TrapInfo))
103 emitInvoke cpidx hasThis = do
104 let l = buildMethodID cls cpidx
105 calladdr <- getCurrentOffset
106 newNamedLabel (show l) >>= defineLabel
107 -- causes SIGILL. in the signal handler we patch it to the acutal call.
108 -- place a nop at the end, therefore the disasm doesn't screw up
109 emit32 (0xffff9090 :: Word32) >> emit8 (0x90 :: Word8)
110 -- discard arguments on stack
111 let argcnt = ((if hasThis then 1 else 0) + (methodGetArgsCount cls cpidx)) * 4
112 when (argcnt > 0) (add esp argcnt)
113 -- push result on stack if method has a return value
114 when (methodHaveReturnValue cls cpidx) (push eax)
115 return $ Just $ (calladdr, MI l)
117 emit' :: J.Instruction -> CodeGen e s (Maybe (Word32, TrapInfo))
118 emit' (INVOKESPECIAL cpidx) = emitInvoke cpidx True
119 emit' (INVOKESTATIC cpidx) = emitInvoke cpidx False
120 emit' (INVOKEINTERFACE cpidx _) = do
121 -- get methodInfo entry
122 let mi@(MethodInfo methodname ifacename msig@(MethodSignature args _)) = buildMethodID cls cpidx
123 newNamedLabel (show mi) >>= defineLabel
124 -- objref lives somewhere on the argument stack
125 mov eax (Disp ((*4) $ fromIntegral $ length args), esp)
126 -- get method-table-ptr, keep it in eax (for trap handling)
127 mov eax (Disp 0, eax)
128 -- get interface-table-ptr
129 mov ebx (Disp 0, eax)
131 offset <- liftIO $ getInterfaceMethodOffset ifacename methodname (encode msig)
132 -- make actual (indirect) call
133 calladdr <- getCurrentOffset
134 call (Disp offset, ebx)
135 -- discard arguments on stack (+4 for "this")
136 let argcnt = 4 + ((methodGetArgsCount cls cpidx) * 4)
137 when (argcnt > 0) (add esp argcnt)
138 -- push result on stack if method has a return value
139 when (methodHaveReturnValue cls cpidx) (push eax)
140 -- note, the "mi" has the wrong class reference here.
141 -- we figure that out at run-time, in the methodpool,
142 -- depending on the method-table-ptr
143 return $ Just $ (calladdr, II mi)
144 emit' (INVOKEVIRTUAL cpidx) = do
145 -- get methodInfo entry
146 let mi@(MethodInfo methodname objname msig@(MethodSignature args _)) = buildMethodID cls cpidx
147 newNamedLabel (show mi) >>= defineLabel
148 -- objref lives somewhere on the argument stack
149 mov eax (Disp ((*4) $ fromIntegral $ length args), esp)
150 -- get method-table-ptr
151 mov eax (Disp 0, eax)
153 let nameAndSig = methodname `B.append` (encode msig)
154 offset <- liftIO $ getMethodOffset objname nameAndSig
155 -- make actual (indirect) call
156 calladdr <- getCurrentOffset
157 call (Disp offset, eax)
158 -- discard arguments on stack (+4 for "this")
159 let argcnt = 4 + ((methodGetArgsCount cls cpidx) * 4)
160 when (argcnt > 0) (add esp argcnt)
161 -- push result on stack if method has a return value
162 when (methodHaveReturnValue cls cpidx) (push eax)
163 -- note, the "mi" has the wrong class reference here.
164 -- we figure that out at run-time, in the methodpool,
165 -- depending on the method-table-ptr
166 return $ Just $ (calladdr, VI mi)
167 emit' (PUTSTATIC cpidx) = do
169 trapaddr <- getCurrentOffset
170 mov (Addr 0x00000000) eax -- it's a trap
171 return $ Just $ (trapaddr, SFI $ buildStaticFieldID cls cpidx)
172 emit' (GETSTATIC cpidx) = do
173 trapaddr <- getCurrentOffset
174 mov eax (Addr 0x00000000) -- it's a trap
176 return $ Just $ (trapaddr, SFI $ buildStaticFieldID cls cpidx)
177 emit' insn = emit insn >> return Nothing
179 emit :: J.Instruction -> CodeGen e s ()
180 emit POP = do -- dropp value
181 add esp (4 :: Word32)
182 emit DUP = push (Disp 0, esp)
183 emit AASTORE = emit IASTORE
187 add ebx (1 :: Word32)
189 mov (ecx, ebx, S4) eax
190 emit AALOAD = emit IALOAD
193 add ebx (1 :: Word32)
196 emit ARRAYLENGTH = do
199 emit (ANEWARRAY _) = emit (NEWARRAY 10) -- 10 == T_INT
200 emit (NEWARRAY typ) = do
201 let tsize = case decodeS (0 :: Integer) (B.pack [typ]) of
203 _ -> error $ "newarray: type not implemented yet"
204 -- get length from stack, but leave it there
205 mov eax (Disp 0, esp)
206 mov ebx (tsize :: Word32)
207 -- multiple amount with native size of one element
208 mul ebx -- result is in eax
209 add eax (4 :: Word32) -- for "length" entry
210 -- push amount of bytes to allocate
213 pop eax -- ref to arraymemory
215 mov (Disp 0, eax) ebx -- store length at offset 0
216 push eax -- push ref again
217 emit (NEW objidx) = do
218 let objname = buildClassID cls objidx
219 amount <- liftIO $ getObjectSize objname
220 push (amount :: Word32)
222 -- TODO(bernhard): save reference somewhere for GC
223 -- set method table pointer
224 mtable <- liftIO $ getMethodTable objname
225 mov (Disp 0, eax) mtable
226 emit (CHECKCAST _) = nop -- TODO(bernhard): ...
227 emit (BIPUSH val) = push ((fromIntegral val) :: Word32)
228 emit (SIPUSH val) = push ((fromIntegral $ ((fromIntegral val) :: Int16)) :: Word32)
229 emit (ICONST_0) = push (0 :: Word32)
230 emit (ICONST_1) = push (1 :: Word32)
231 emit (ICONST_2) = push (2 :: Word32)
232 emit (ICONST_3) = push (3 :: Word32)
233 emit (ICONST_4) = push (4 :: Word32)
234 emit (ICONST_5) = push (5 :: Word32)
235 emit (ALOAD_ x) = emit (ILOAD_ x)
237 push (Disp (cArgs_ x), ebp)
238 emit (ALOAD x) = emit (ILOAD x)
240 push (Disp (cArgs x), ebp)
241 emit (ASTORE_ x) = emit (ISTORE_ x)
242 emit (ISTORE_ x) = do
244 mov (Disp (cArgs_ x), ebp) eax
245 emit (ASTORE x) = emit (ISTORE x)
248 mov (Disp (cArgs x), ebp) eax
250 emit (LDC1 x) = emit (LDC2 $ fromIntegral x)
252 value <- case (constsPool cls) M.! x of
253 (CString s) -> liftIO $ getUniqueStringAddr s
254 _ -> error $ "LDCI... missing impl."
256 emit (GETFIELD x) = do
257 pop eax -- this pointer
258 let (cname, fname) = buildFieldOffset cls x
259 offset <- liftIO $ getFieldOffset cname fname
260 push (Disp (fromIntegral $ offset), eax) -- get field
261 emit (PUTFIELD x) = do
262 pop ebx -- value to write
263 pop eax -- this pointer
264 let (cname, fname) = buildFieldOffset cls x
265 offset <- liftIO $ getFieldOffset cname fname
266 mov (Disp (fromIntegral $ offset), eax) ebx -- set field
268 emit IADD = do pop ebx; pop eax; add eax ebx; push eax
269 emit ISUB = do pop ebx; pop eax; sub eax ebx; push eax
270 emit IMUL = do pop ebx; pop eax; mul ebx; push eax
271 emit IXOR = do pop ebx; pop eax; xor eax ebx; push eax
272 emit (IINC x imm) = do
273 add (Disp (cArgs x), ebp) (s8_w32 imm)
275 emit (IF_ACMP cond x) = emit (IF_ICMP cond x)
276 emit (IF_ICMP cond _) = do
279 cmp ebx eax -- intel syntax is swapped (TODO(bernhard): test that plz)
280 let sid = case successor bb of TwoTarget _ t -> t; _ -> error "bad"
281 let l = getLabel sid lmap
283 C_EQ -> je l; C_NE -> jne l
284 C_LT -> jl l; C_GT -> jg l
285 C_GE -> jge l; C_LE -> jle l
287 emit (IF cond _) = do
289 cmp eax (0 :: Word32) -- TODO(bernhard): test that plz
290 let sid = case successor bb of TwoTarget _ t -> t; _ -> error "bad"
291 let l = getLabel sid lmap
293 C_EQ -> je l; C_NE -> jne l
294 C_LT -> jl l; C_GT -> jg l
295 C_GE -> jge l; C_LE -> jle l
298 let sid = case successor bb of OneTarget t -> t; _ -> error "bad"
299 jmp $ getLabel sid lmap
301 emit RETURN = do mov esp ebp; pop ebp; ret
302 emit ARETURN = emit IRETURN
308 emit invalid = error $ "insn not implemented yet: " ++ (show invalid)
310 callMalloc :: CodeGen e s ()
312 calladdr <- getCurrentOffset
313 let w32_calladdr = 5 + calladdr
314 let malloaddr = (fromIntegral getMallocObjectAddr :: Word32)
315 call (malloaddr - w32_calladdr)
316 add esp (4 :: Word32)
319 -- for locals we use a different storage
320 cArgs :: Word8 -> Word32
321 cArgs x = if (x' >= thisMethodArgCnt)
322 -- TODO(bernhard): maybe s/(-4)/(-8)/
323 then fromIntegral $ (-4) * (x' - thisMethodArgCnt + 1)
324 else 4 + (thisMethodArgCnt * 4) - (4 * x')
325 where x' = fromIntegral x
327 cArgs_ :: IMM -> Word32
328 cArgs_ x = cArgs $ case x of I0 -> 0; I1 -> 1; I2 -> 2; I3 -> 3
330 thisMethodArgCnt :: Word32
331 thisMethodArgCnt = isNonStatic + (fromIntegral $ length args)
333 (Just m) = lookupMethod method cls
334 (MethodSignature args _) = methodSignature m
335 isNonStatic = if S.member ACC_STATIC (methodAccessFlags m)
337 else 1 -- one argument for the this pointer
340 -- sign extension from w8 to w32 (over s8)
341 -- unfortunately, hs-java is using Word8 everywhere (while
342 -- it should be Int8 actually)
343 s8_w32 :: Word8 -> Word32
344 s8_w32 w8 = fromIntegral s8
345 where s8 = (fromIntegral w8) :: Int8