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
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
|
(*
Copyright (c) 2016-17 David C.J. Matthews
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License version 2.1 as published by the Free Software Foundation.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*)
functor X86FOREIGNCALL(
structure X86CODE: X86CODESIG
structure X86OPTIMISE:
sig
type operation
type code
type operations = operation list
type address = Address.address
(* Optimise and code-generate. *)
val generateCode: {code: code, ops: operations, labelCount: int} -> address
structure Sharing:
sig
type operation = operation
type code = code
end
end
structure DEBUG: DEBUGSIG
sharing X86CODE.Sharing = X86OPTIMISE.Sharing
): FOREIGNCALLSIG
=
struct
open X86CODE
open Address
exception InternalError = Misc.InternalError
val pushR = PushToStack o RegisterArg
fun moveRR{source, output} = MoveToRegister{source=RegisterArg source, output=output}
fun loadMemory(reg, base, offset) =
MoveToRegister{source=MemoryArg{base=base, offset=offset, index=NoIndex}, output=reg}
and storeMemory(reg, base, offset) =
StoreRegToMemory{toStore=reg, address={base=base, offset=offset, index=NoIndex}}
fun createProfileObject _ (*functionName*) =
let
(* The profile object is a single mutable with the F_bytes bit set. *)
open Address
val profileObject = RunCall.allocateByteMemory(0w1, Word.fromLargeWord(Word8.toLargeWord(Word8.orb(F_mutable, F_bytes))))
fun clear 0w0 = ()
| clear i = (assignByte(profileObject, i-0w1, 0w0); clear (i-0w1))
val () = clear(Word.fromInt wordSize)
in
toMachineWord profileObject
end
val makeEntryPoint: string -> machineWord = RunCall.rtsCallFull1 "PolyCreateEntryPointObject"
datatype abi = X86_32 | X64Win | X64Unix
local
(* Get the ABI. On 64-bit Windows and Unix use different calling conventions. *)
val getABICall: unit -> int = RunCall.rtsCallFast0 "PolyGetABI"
in
fun getABI() =
case getABICall() of
0 => X86_32
| 1 => X64Unix
| 2 => X64Win
| n => raise InternalError ("Unknown ABI type " ^ Int.toString n)
end
val noException = 1
(* Full RTS call version. An extra argument is passed that contains the thread ID.
This allows the taskData object to be found which is needed if the code allocates
any ML memory or raises an exception. It also saves the stack and heap pointers
in case of a GC. *)
fun rtsCallFull (functionName, nArgs (* Not counting the thread ID *), debugSwitches) =
let
val entryPointAddr = makeEntryPoint functionName
(* Get the ABI. On 64-bit Windows and Unix use different calling conventions. *)
val abi = getABI()
(* Branch to check for exception. *)
val exLabel = Label{labelNo=0} (* There's just one label in this function. *)
(* Unix X64. The first six arguments are in rdi, rsi, rdx, rcx, r8, r9.
The rest are on the stack.
Windows X64. The first four arguments are in rcx, rdx, r8 and r9. The rest are
on the stack. The caller must ensure the stack is aligned on 16-byte boundary
and must allocate 32-byte save area for the register args.
rbx, rbp, rdi, rsi, rsp, r12-r15 are saved by the called function.
X86/32. Arguments are pushed to the stack.
ebx, edi, esi, ebp and esp are saved by the called function.
We use esi to hold the argument data pointer and edi to save the ML stack pointer
Our ML conventions use eax, ebx for the first two arguments in X86/32 and
rax, ebx, r8, r9, r10 for the first five arguments in X86/64.
*)
(* The ML stack pointer needs to be
loaded into a register that will be saved by the callee.
The entry point doesn't need to be but must be a register
that isn't used for an argument. *)
(* We have to save the stack pointer to the argument structure but
we still need a register if we have args on the stack. *)
val (entryPtrReg, saveMLStackPtrReg) =
if isX64 then (r11, r13) else (ecx, edi)
val stackSpace =
case abi of
X64Unix => memRegSize
| X64Win => memRegSize + 32 (* Requires 32-byte save area. *)
| X86_32 =>
let
(* GCC likes to keep the stack on a 16-byte alignment. *)
val argSpace = (nArgs+1)*4
val align = argSpace mod 16
in
(* Add sufficient space so that esp will be 16-byte aligned *)
if align = 0
then memRegSize
else memRegSize + 16 - align
end
val code =
[
MoveToRegister{source=AddressConstArg entryPointAddr, output=entryPtrReg}, (* Load the entry point ref. *)
loadMemory(entryPtrReg, entryPtrReg, 0)(* Load its value. *)
] @
(
(* Save heap ptr. This is in r15 in X86/64 *)
if isX64 then [storeMemory(r15, ebp, memRegLocalMPointer)] (* Save heap ptr *)
else []
) @
[
moveRR{source=esp, output=saveMLStackPtrReg}, (* Save this in case it is needed for arguments. *)
(* Have to save the stack pointer to the arg structure in case we need to scan the stack for a GC. *)
storeMemory(esp, ebp, memRegStackPtr), (* Save ML stack and switch to C stack. *)
loadMemory(esp, ebp, memRegCStackPtr), (*moveRR{source=ebp, output=esp},*) (* Load the saved C stack pointer. *)
(* Set the stack pointer past the data on the stack. For Windows/64 add in a 32 byte save area *)
ArithToGenReg{opc=SUB, output=esp, source=NonAddressConstArg(LargeInt.fromInt stackSpace)}
] @
(
case (abi, nArgs) of (* Set the argument registers. *)
(X64Unix, 0) => [ loadMemory(edi, ebp, memRegThreadSelf) ]
| (X64Unix, 1) => [ loadMemory(edi, ebp, memRegThreadSelf), moveRR{source=eax, output=esi} ]
| (X64Unix, 2) =>
[ loadMemory(edi, ebp, memRegThreadSelf), moveRR{source=eax, output=esi}, moveRR{source=ebx, output=edx} ]
| (X64Unix, 3) =>
[ loadMemory(edi, ebp, memRegThreadSelf), moveRR{source=eax, output=esi}, moveRR{source=ebx, output=edx}, moveRR{source=r8, output=ecx} ]
| (X64Win, 0) => [ loadMemory(ecx, ebp, memRegThreadSelf) ]
| (X64Win, 1) => [ loadMemory(ecx, ebp, memRegThreadSelf), moveRR{source=eax, output=edx} ]
| (X64Win, 2) =>
[ loadMemory(ecx, ebp, memRegThreadSelf), moveRR{source=eax, output=edx}, moveRR{source=ebx, output=r8} ]
| (X64Win, 3) =>
[ loadMemory(ecx, ebp, memRegThreadSelf), moveRR{source=eax, output=edx}, moveRR{source=r8, output=r9}, moveRR{source=ebx, output=r8} ]
| (X86_32, 0) => [ PushToStack(MemoryArg{base=ebp, offset=memRegThreadSelf, index=NoIndex}) ]
| (X86_32, 1) => [ pushR eax, PushToStack(MemoryArg{base=ebp, offset=memRegThreadSelf, index=NoIndex}) ]
| (X86_32, 2) => [ pushR ebx, pushR eax, PushToStack(MemoryArg{base=ebp, offset=memRegThreadSelf, index=NoIndex}) ]
| (X86_32, 3) =>
[
(* We need to move an argument from the ML stack. *)
PushToStack(MemoryArg{base=saveMLStackPtrReg, offset=4, index=NoIndex}), pushR ebx, pushR eax,
PushToStack(MemoryArg{base=ebp, offset=memRegThreadSelf, index=NoIndex})
]
| _ => raise InternalError "rtsCall: Abi/argument count not implemented"
) @
[
CallFunction(DirectReg entryPtrReg), (* Call the function *)
moveRR{source=saveMLStackPtrReg, output=esp} (* Restore the ML stack pointer *)
] @
(
if isX64 then [loadMemory(r15, ebp, memRegLocalMPointer) ] (* Copy back the heap ptr *)
else []
) @
[
ArithMemConst{opc=CMP, offset=memRegExceptionPacket, base=ebp, source=noException},
ConditionalBranch{test=JNE, predict=PredictNotTaken, label=exLabel},
(* Remove any arguments that have been passed on the stack. *)
ReturnFromFunction(Int.max(case abi of X86_32 => nArgs-2 | _ => nArgs-5, 0)),
JumpLabel exLabel, (* else raise the exception *)
loadMemory(eax, ebp, memRegExceptionPacket),
RaiseException
]
val profileObject = createProfileObject functionName
val newCode = codeCreate (functionName, profileObject, debugSwitches)
val createdCode = X86OPTIMISE.generateCode{code=newCode, labelCount=1(*One label.*), ops=code}
(* Have to create a closure for this *)
open Address
val closure = allocWordData(0w1, Word8.orb (F_mutable, F_words), toMachineWord 0w0)
in
assignWord(closure, 0w0, toMachineWord createdCode);
lock closure;
closure
end
(* This is a quicker version but can only be used if the RTS entry does
not allocated ML memory, raise an exception or need to suspend the thread. *)
fun rtsCallFast (functionName, nArgs, debugSwitches) =
let
val entryPointAddr = makeEntryPoint functionName
(* Get the ABI. On 64-bit Windows and Unix use different calling conventions. *)
val abi = getABI()
val (entryPtrReg, saveMLStackPtrReg) =
if isX64 then (r11, r13) else (ecx, edi)
val stackSpace =
case abi of
X64Unix => memRegSize
| X64Win => memRegSize + 32 (* Requires 32-byte save area. *)
| X86_32 =>
let
(* GCC likes to keep the stack on a 16-byte alignment. *)
val argSpace = nArgs*4
val align = argSpace mod 16
in
(* Add sufficient space so that esp will be 16-byte aligned *)
if align = 0
then memRegSize
else memRegSize + 16 - align
end
val code =
[
MoveToRegister{source=AddressConstArg entryPointAddr, output=entryPtrReg}, (* Load the entry point ref. *)
loadMemory(entryPtrReg, entryPtrReg, 0),(* Load its value. *)
moveRR{source=esp, output=saveMLStackPtrReg}, (* Save ML stack and switch to C stack. *)
loadMemory(esp, ebp, memRegCStackPtr),
(* Set the stack pointer past the data on the stack. For Windows/64 add in a 32 byte save area *)
ArithToGenReg{opc=SUB, output=esp, source=NonAddressConstArg(LargeInt.fromInt stackSpace)}
] @
(
case (abi, nArgs) of (* Set the argument registers. *)
(_, 0) => []
| (X64Unix, 1) => [ moveRR{source=eax, output=edi} ]
| (X64Unix, 2) =>
[ moveRR{source=eax, output=edi}, moveRR{source=ebx, output=esi} ]
| (X64Unix, 3) =>
[ moveRR{source=eax, output=edi}, moveRR{source=ebx, output=esi}, moveRR{source=r8, output=edx} ]
| (X64Unix, 4) =>
[ moveRR{source=eax, output=edi}, moveRR{source=ebx, output=esi}, moveRR{source=r8, output=edx}, moveRR{source=r9, output=ecx} ]
| (X64Win, 1) => [ moveRR{source=eax, output=ecx} ]
| (X64Win, 2) => [ moveRR{source=eax, output=ecx}, moveRR{source=ebx, output=edx} ]
| (X64Win, 3) =>
[ moveRR{source=eax, output=ecx}, moveRR{source=ebx, output=edx} (* Arg3 is already in r8. *) ]
| (X64Win, 4) =>
[ moveRR{source=eax, output=ecx}, moveRR{source=ebx, output=edx} (* Arg3 is already in r8 and arg4 in r9. *) ]
| (X86_32, 1) => [ pushR eax ]
| (X86_32, 2) => [ pushR ebx, pushR eax ]
| (X86_32, 3) =>
[
(* We need to move an argument from the ML stack. *)
loadMemory(edx, saveMLStackPtrReg, 4), pushR edx, pushR ebx, pushR eax
]
| (X86_32, 4) =>
[
(* We need to move an arguments from the ML stack. *)
loadMemory(edx, saveMLStackPtrReg, 4), pushR edx,
loadMemory(edx, saveMLStackPtrReg, 8), pushR edx,
pushR ebx, pushR eax
]
| _ => raise InternalError "rtsCall: Abi/argument count not implemented"
) @
[
CallFunction(DirectReg entryPtrReg), (* Call the function *)
moveRR{source=saveMLStackPtrReg, output=esp}, (* Restore the ML stack pointer *)
(* Remove any arguments that have been passed on the stack. *)
ReturnFromFunction(Int.max(case abi of X86_32 => nArgs-2 | _ => nArgs-5, 0))
]
val profileObject = createProfileObject functionName
val newCode = codeCreate (functionName, profileObject, debugSwitches)
val createdCode = X86OPTIMISE.generateCode{code=newCode, labelCount=0, ops=code}
(* Have to create a closure for this *)
open Address
val closure = allocWordData(0w1, Word8.orb (F_mutable, F_words), toMachineWord 0w0)
in
assignWord(closure, 0w0, toMachineWord createdCode);
lock closure;
closure
end
(* RTS call with one double-precision floating point argument and a floating point result.
First version. This will probably be merged into the above code in due
course.
Currently ML always uses boxed values for floats. *)
fun rtsCallFastFloattoFloat (functionName, debugSwitches) =
let
val entryPointAddr = makeEntryPoint functionName
(* Get the ABI. On 64-bit Windows and Unix use different calling conventions. *)
val abi = getABI()
val (entryPtrReg, saveMLStackPtrReg) =
if isX64 then (r11, r13) else (ecx, edi)
val stackSpace =
case abi of
X64Unix => memRegSize
| X64Win => memRegSize + 32 (* Requires 32-byte save area. *)
| X86_32 =>
let
(* GCC likes to keep the stack on a 16-byte alignment. *)
val argSpace = 8 (*nArgs*4*) (* One "double" value. *)
val align = argSpace mod 16
in
(* Add sufficient space so that esp will be 16-byte aligned *)
if align = 0
then memRegSize
else memRegSize + 16 - align
end
(* Constants for a box for a float *)
val fpBoxSize = 8 div wordSize
val fpBoxLengthWord32 = IntInf.orb(IntInf.fromInt fpBoxSize, IntInf.<<(Word8.toLargeInt F_bytes, 0w24))
val code =
[
MoveToRegister{source=AddressConstArg entryPointAddr, output=entryPtrReg}, (* Load the entry point ref. *)
loadMemory(entryPtrReg, entryPtrReg, 0),(* Load its value. *)
moveRR{source=esp, output=saveMLStackPtrReg}, (* Save ML stack and switch to C stack. *)
loadMemory(esp, ebp, memRegCStackPtr),
(* Set the stack pointer past the data on the stack. For Windows/64 add in a 32 byte save area *)
ArithToGenReg{opc=SUB, output=esp, source=NonAddressConstArg(LargeInt.fromInt stackSpace)}
] @
(
case abi of
(* X64 on both Windows and Unix take the first arg in xmm0. We need to
unbox the value pointed at by rax. *)
X64Unix => [ XMMArith { opc= SSE2Move, source=MemoryArg{base=eax, offset=0, index=NoIndex}, output=xmm0 } ]
| X64Win => [ XMMArith { opc= SSE2Move, source=MemoryArg{base=eax, offset=0, index=NoIndex}, output=xmm0 } ]
| X86_32 =>
(* eax contains the address of the value. This must be unboxed onto the stack. *)
[
FPLoadFromMemory{address={base=eax, offset=0, index=NoIndex}, precision=DoublePrecision},
ArithToGenReg{ opc=SUB, output=esp, source=NonAddressConstArg 8},
FPStoreToMemory{ address={base=esp, offset=0, index=NoIndex}, precision=DoublePrecision, andPop=true }
]
) @
[
CallFunction(DirectReg entryPtrReg), (* Call the function *)
moveRR{source=saveMLStackPtrReg, output=esp} (* Restore the ML stack pointer *)
] @
(
(* Put the floating point result into a box. *)
case abi of
X86_32 =>
[
AllocStore{size=fpBoxSize, output=eax, saveRegs=[]},
StoreConstToMemory{toStore=fpBoxLengthWord32,
address={offset= ~wordSize, base=eax, index=NoIndex}},
FPStoreToMemory{ address={base=eax, offset=0, index=NoIndex}, precision=DoublePrecision, andPop=true },
StoreInitialised
]
| _ => (* X64 The result is in xmm0 *)
[
AllocStore{size=fpBoxSize, output=eax, saveRegs=[]},
StoreConstToMemory{toStore=LargeInt.fromInt fpBoxSize,
address={offset= ~wordSize, base=eax, index=NoIndex}},
StoreNonWordConst{size=Size8Bit, toStore=Word8.toLargeInt F_bytes, address={offset= ~1, base=eax, index=NoIndex}},
XMMStoreToMemory { address={base=eax, offset=0, index=NoIndex}, precision=DoublePrecision, toStore=xmm0 },
StoreInitialised
]
) @
[
(* Remove any arguments that have been passed on the stack. *)
ReturnFromFunction 0
]
val profileObject = createProfileObject functionName
val newCode = codeCreate (functionName, profileObject, debugSwitches)
val createdCode = X86OPTIMISE.generateCode{code=newCode, labelCount=0, ops=code}
(* Have to create a closure for this *)
open Address
val closure = allocWordData(0w1, Word8.orb (F_mutable, F_words), toMachineWord 0w0)
in
assignWord(closure, 0w0, toMachineWord createdCode);
lock closure;
closure
end
(* RTS call with one general (i.e. ML word) argument and a floating point result.
This is used only to convert arbitrary precision values to floats.
In due course this will be merged into the other functions. *)
fun rtsCallFastGeneraltoFloat (functionName, debugSwitches) =
let
val entryPointAddr = makeEntryPoint functionName
(* Get the ABI. On 64-bit Windows and Unix use different calling conventions. *)
val abi = getABI()
val (entryPtrReg, saveMLStackPtrReg) =
if isX64 then (r11, r13) else (ecx, edi)
val stackSpace =
case abi of
X64Unix => memRegSize
| X64Win => memRegSize + 32 (* Requires 32-byte save area. *)
| X86_32 =>
let
(* GCC likes to keep the stack on a 16-byte alignment. *)
val argSpace = 8 (*nArgs*4*) (* One "double" value. *)
val align = argSpace mod 16
in
(* Add sufficient space so that esp will be 16-byte aligned *)
if align = 0
then memRegSize
else memRegSize + 16 - align
end
(* Constants for a box for a float *)
val fpBoxSize = 8 div wordSize
val fpBoxLengthWord32 = IntInf.orb(IntInf.fromInt fpBoxSize, IntInf.<<(Word8.toLargeInt F_bytes, 0w24))
val code =
[
MoveToRegister{source=AddressConstArg entryPointAddr, output=entryPtrReg}, (* Load the entry point ref. *)
loadMemory(entryPtrReg, entryPtrReg, 0),(* Load its value. *)
moveRR{source=esp, output=saveMLStackPtrReg}, (* Save ML stack and switch to C stack. *)
loadMemory(esp, ebp, memRegCStackPtr),
(* Set the stack pointer past the data on the stack. For Windows/64 add in a 32 byte save area *)
ArithToGenReg{opc=SUB, output=esp, source=NonAddressConstArg(LargeInt.fromInt stackSpace)}
] @
(
case abi of
X64Unix => [ moveRR{source=eax, output=edi} ]
| X64Win => [ moveRR{source=eax, output=ecx} ]
| X86_32 => [ pushR eax ]
) @
[
CallFunction(DirectReg entryPtrReg), (* Call the function *)
moveRR{source=saveMLStackPtrReg, output=esp} (* Restore the ML stack pointer *)
] @
(
(* Put the floating point result into a box. *)
case abi of
X86_32 =>
[
AllocStore{size=fpBoxSize, output=eax, saveRegs=[]},
StoreConstToMemory{toStore=fpBoxLengthWord32,
address={offset= ~wordSize, base=eax, index=NoIndex}},
FPStoreToMemory{ address={base=eax, offset=0, index=NoIndex}, precision=DoublePrecision, andPop=true },
StoreInitialised
]
| _ => (* X64 The result is in xmm0 *)
[
AllocStore{size=fpBoxSize, output=eax, saveRegs=[]},
StoreConstToMemory{toStore=LargeInt.fromInt fpBoxSize,
address={offset= ~wordSize, base=eax, index=NoIndex}},
StoreNonWordConst{size=Size8Bit, toStore=Word8.toLargeInt F_bytes, address={offset= ~1, base=eax, index=NoIndex}},
XMMStoreToMemory { address={base=eax, offset=0, index=NoIndex}, precision=DoublePrecision, toStore=xmm0 },
StoreInitialised
]
) @
[
(* Remove any arguments that have been passed on the stack. *)
ReturnFromFunction 0
]
val profileObject = createProfileObject functionName
val newCode = codeCreate (functionName, profileObject, debugSwitches)
val createdCode = X86OPTIMISE.generateCode{code=newCode, labelCount=0, ops=code}
(* Have to create a closure for this *)
open Address
val closure = allocWordData(0w1, Word8.orb (F_mutable, F_words), toMachineWord 0w0)
in
assignWord(closure, 0w0, toMachineWord createdCode);
lock closure;
closure
end
end;
|
