1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
|
(* x86-svid-fn.sml
*
* C calling-sequence generator for x86.
*
* Mike Rainey (mrainey@cs.uchicago.edu)
*)
functor X86SVIDFn (
structure T : MLTREE
val abi : string
val ix : (T.stm,T.rexp,T.fexp,T.ccexp) X86InstrExt.sext -> T.sext
(* Note that the fast_loating_point flag must match the one passed
* to the code generator module.
*)
val fast_floating_point : bool ref
) : C_CALL =
struct
structure T = T
structure C = X86Cells
structure CB = CellsBasis
structure CTy = CType
structure IX = X86InstrExt
val wordTy = 32
fun gpr r = T.GPR(T.REG(32, r))
fun fpr f = T.FPR(T.FREG(80, f))
val spReg = T.REG (32, C.esp)
(* the C calling convention requires that the FP stack be empty on function
* entry. We add the fpStk list to the defs when the fast_floating_point flag
* is set.
*)
val st0 = C.ST 0
datatype loc_kind = datatype CLocKind.loc_kind
structure SA = StagedAllocationFn (
type reg_id = T.reg
datatype loc_kind = datatype loc_kind
val memSize = 4)
structure Gen = CCallGenFn(
structure T = T
structure C = C
val wordTy = wordTy
fun offSp 0 = spReg
| offSp offset = T.ADD (32, spReg, T.LI offset)
fun lobits {e, nBits, width} = e
fun sx {fromWidth, toWidth, e} = T.SX(toWidth, fromWidth, e)
fun f2f {fromWidth, toWidth, e} = e
structure SA = SA)
datatype c_arg = datatype Gen.c_arg
structure CCs = X86CConventionFn (
type reg_id = T.reg
val eax = C.eax
val edx = C.edx
val st0 = st0
structure SA = SA
)
val calleeSaveRegs = [C.ebx, C.esi, C.edi]
val callerSaveRegs = [C.eax, C.ecx, C.edx]
val calleeSaveFRegs = []
val callerSaveFRegs = []
(* assign a C type to a kind of machine location *)
fun kindOfCTy (CTy.C_float | CTy.C_double | CTy.C_long_double) = FPR
| kindOfCTy (CTy.C_unsigned _ | CTy.C_signed _ | CTy.C_PTR | CTy.C_ARRAY _) = GPR
(* convert a C type to reqs for staged allocation *)
fun cTyToReqs cTy = let
val {sz, align} = X86CSizes.sizeOfTy cTy
(* compute argument reqs for the flattened C type *)
val reqs = List.map (fn cTy => (sz * 8, kindOfCTy cTy, align))
(CTy.flattenCTy cTy)
in
case (cTy, abi)
of (CTy.C_STRUCT _, "Darwin") =>
(* for Darwin, structs <= 8 bytes are returned in GPRs *)
if (sz <= 4)
then [(8, GPR, align)]
else if (sz <= 8)
then [(8, GPR, align), (8, GPR, align)]
else reqs
| ( (CTy.C_unsigned CTy.I_long_long |
CTy.C_signed CTy.I_long_long ),
_ ) =>
(* 64-bit integers are returned in GPRs *)
[(8, GPR, align), (8, GPR, align)]
| _ => reqs
end
(* compute the parameter passing and return for a given C call *)
fun layout {conv, retTy, paramTys} = let
(* lay out the return parameters *)
val (resLocs, structRetLoc, store) = (case retTy
of CTy.C_void => ([], NONE, CCs.store0)
| retTy as CTy.C_STRUCT _ => raise Fail ""
| retTy => let
val {sz, align} = X86CSizes.sizeOfTy retTy
val (locs, store) = SA.allocateSeq CCs.returns (cTyToReqs retTy, CCs.store0)
in
(locs, NONE, store)
end
(* end case *))
(* lay out the parameters *)
val paramReqss = List.map cTyToReqs paramTys
val (paramLocss, store) = SA.allocateSeqs CCs.params (paramReqss, store)
(* number of bytes allocated for the call *)
val cStkSzB = let
val n = SA.find(store, CCs.cStack)
in
if (abi = "Mac OS X")
then X86CSizes.alignAddr(n, 16)
else n
end
in
{argLocs=paramLocss, argMem={szb=cStkSzB, align=4}, structRetLoc=structRetLoc, resLocs=resLocs}
end (* layout *)
val callerSaveRegs' = List.map gpr calleeSaveRegs
val calleeSaveRegs' = List.map gpr calleeSaveRegs
val calleeSaveFRegs' = []
val callerSaveFRegs' = []
(* the C calling convention requires that the FP stack be empty on function
* entry. We add the fpStk list to the defs when the fast_floating_point flag
* is set.
*)
val fpStk = List.tabulate(8, fn i => fpr (C.ST i))
(* List of registers defined by a C Call with the given return type; this list
* is the result registers plus the caller-save registers.
*)
fun definedRegs resTy = if !fast_floating_point
then let
val defs = callerSaveRegs' @ fpStk
in
case resTy
of (CTy.C_unsigned(CTy.I_long_long)) => gpr C.edx :: defs
| (CTy.C_signed(CTy.I_long_long)) => gpr C.edx :: defs
| _ => defs
(* end case *)
end
else (case resTy
of (CTy.C_float) => fpr st0 :: callerSaveRegs'
| (CTy.C_double) => fpr st0 :: callerSaveRegs'
| (CTy.C_long_double) => fpr st0 :: callerSaveRegs'
| (CTy.C_unsigned(CTy.I_long_long)) => gpr C.edx :: callerSaveRegs'
| (CTy.C_signed(CTy.I_long_long)) => gpr C.edx :: callerSaveRegs'
| _ => callerSaveRegs'
(* end case *))
fun fstp (32, f) = T.EXT(ix(IX.FSTPS(f)))
| fstp (64, f) = T.EXT(ix(IX.FSTPL(f)))
| fstp (80, f) = T.EXT(ix(IX.FSTPT(f)))
| fstp (sz, f) = raise Fail ("fstp(" ^ Int.toString sz ^ ",_)")
(* This annotation is used to indicate that a call returns a fp value
* in %st(0)
*)
val fpReturnValueInST0 = #create MLRiscAnnotations.RETURN_ARG C.ST0
fun genCall {
name, proto, paramAlloc, structRet, saveRestoreDedicated, callComment, args
} = let
val {argLocs, argMem, structRetLoc, resLocs} = layout proto
(* for functions that return a struct/union, pass the location as an
* implicit first argument. Because the callee removes this implicit
* argument from the stack, we must also keep track of the size of the
* explicit arguments.
*)
val (args, argLocs, explicitArgSzB) = (case structRetLoc
of SOME pos =>
(ARG(structRet pos)::args, [SA.BLOCK_OFFSET(wordTy, GPR, 0)]::argLocs, #szb argMem)
| NONE => (args, argLocs, #szb argMem)
(* end case *))
(* instruction to allocate space for arguments *)
val argAlloc = if (#szb argMem = 0 orelse paramAlloc argMem)
then []
else if abi = "Darwin"
then let
(* align the frame on a 16-byte boundary *)
val szb = X86CSizes.alignAddr(#szb argMem + 2*4, 16)-2*4
in
[T.MV(wordTy, C.esp, T.SUB(wordTy, spReg, T.LI(IntInf.fromInt szb)))]
end
else [T.MV(wordTy, C.esp, T.SUB(wordTy, spReg, T.LI(IntInf.fromInt(#szb argMem))))]
val (copyArgs, gprUses, fprUses) = Gen.writeLocs(args, argLocs)
(* the SVID specifies that the caller pops arguments, but the callee
* pops the arguments in a stdcall on Windows. I'm not sure what other
* differences there might be between the SVID and Windows ABIs. (JHR)
*)
val calleePops = (case #conv proto
of (""|"ccall") => false
| "stdcall" => true
| conv => raise Fail (concat [
"unknown calling convention \"", String.toString conv, "\""
])
(* end case *))
(* code to pop the arguments from the stack *)
val popArgs = if calleePops orelse (explicitArgSzB = 0)
then []
else [T.MV(wordTy, C.esp, T.ADD(wordTy, spReg, T.LI(IntInf.fromInt explicitArgSzB)))]
(* FIXME: support fast floating point *)
(* read return values *)
val (resultRegs, copyResult) = Gen.readLocs resLocs
val defs = definedRegs(#retTy proto)
val { save, restore } = saveRestoreDedicated defs
val callStm = T.CALL{
funct=name, targets=[], defs=defs, uses=[],
region = T.Region.memory,
pops = if calleePops
then Int32.fromInt(#szb argMem)
else Int32.fromInt(#szb argMem - explicitArgSzB)
}
val callStm = (case callComment
of NONE => callStm
| SOME c => T.ANNOTATION (callStm, #create MLRiscAnnotations.COMMENT c)
(* end case *))
val callStm = if !fast_floating_point
andalso ((#retTy proto = CTy.C_float)
orelse (#retTy proto = CTy.C_double)
orelse (#retTy proto = CTy.C_long_double))
then T.ANNOTATION(callStm, fpReturnValueInST0)
else callStm
(* assemble the call sequence *)
val callSeq = argAlloc @ copyArgs @ save @ [callStm] @ restore @ popArgs @ copyResult
in
{callseq=callSeq, result=resultRegs}
end
end (* X86SVIDFn *)
|
