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
|
# 2 "asmcomp/amd64/proc.ml"
(**************************************************************************)
(* *)
(* OCaml *)
(* *)
(* Xavier Leroy, projet Cristal, INRIA Rocquencourt *)
(* *)
(* Copyright 2000 Institut National de Recherche en Informatique et *)
(* en Automatique. *)
(* *)
(* All rights reserved. This file is distributed under the terms of *)
(* the GNU Lesser General Public License version 2.1, with the *)
(* special exception on linking described in the file LICENSE. *)
(* *)
(**************************************************************************)
(* Description of the AMD64 processor *)
open Misc
open Arch
open Cmm
open Reg
open Mach
let fp = Config.with_frame_pointers
(* Which ABI to use *)
let win64 = Arch.win64
(* Registers available for register allocation *)
(* Register map:
rax 0
rbx 1
rdi 2
rsi 3
rdx 4
rcx 5
r8 6
r9 7
r12 8
r13 9
r10 10
r11 11
rbp 12
r14 domain state pointer
r15 allocation pointer
xmm0 - xmm15 100 - 115 *)
(* Conventions:
rax - r13: OCaml function arguments
rax: OCaml and C function results
xmm0 - xmm9: OCaml function arguments
xmm0: OCaml and C function results
Under Unix:
rdi, rsi, rdx, rcx, r8, r9: C function arguments
xmm0 - xmm7: C function arguments
rbx, rbp, r12-r15 are preserved by C
xmm registers are not preserved by C
Under Win64:
rcx, rdx, r8, r9: C function arguments
xmm0 - xmm3: C function arguments
rbx, rbp, rsi, rdi r12-r15 are preserved by C
xmm6-xmm15 are preserved by C
Note (PR#5707, GPR#1304): PLT stubs (used for dynamic resolution of symbols
on Unix-like platforms) may clobber any register except those used for:
1. C parameter passing;
2. C return values;
3. C callee-saved registers.
This translates to the set { r10, r11 }. These registers hence cannot
be used for OCaml parameter passing and must also be marked as
destroyed across [Ialloc] and [Ipoll] (otherwise a call to
caml_call_gc@PLT might clobber these two registers before the assembly
stub saves them into the GC regs block).
*)
let int_reg_name =
match Config.ccomp_type with
| "msvc" ->
[| "rax"; "rbx"; "rdi"; "rsi"; "rdx"; "rcx"; "r8"; "r9";
"r12"; "r13"; "r10"; "r11"; "rbp" |]
| _ ->
[| "%rax"; "%rbx"; "%rdi"; "%rsi"; "%rdx"; "%rcx"; "%r8"; "%r9";
"%r12"; "%r13"; "%r10"; "%r11"; "%rbp" |]
let float_reg_name =
match Config.ccomp_type with
| "msvc" ->
[| "xmm0"; "xmm1"; "xmm2"; "xmm3"; "xmm4"; "xmm5"; "xmm6"; "xmm7";
"xmm8"; "xmm9"; "xmm10"; "xmm11";
"xmm12"; "xmm13"; "xmm14"; "xmm15" |]
| _ ->
[| "%xmm0"; "%xmm1"; "%xmm2"; "%xmm3"; "%xmm4"; "%xmm5"; "%xmm6"; "%xmm7";
"%xmm8"; "%xmm9"; "%xmm10"; "%xmm11";
"%xmm12"; "%xmm13"; "%xmm14"; "%xmm15" |]
let num_register_classes = 2
let register_class r =
match r.typ with
| Val | Int | Addr -> 0
| Float -> 1
let num_available_registers = [| 13; 16 |]
let first_available_register = [| 0; 100 |]
let register_name r =
if r < 100 then int_reg_name.(r) else float_reg_name.(r - 100)
(* Pack registers starting at %rax so as to reduce the number of REX
prefixes and thus improve code density *)
let rotate_registers = false
(* Representation of hard registers by pseudo-registers *)
let hard_int_reg =
let v = Array.make 13 Reg.dummy in
for i = 0 to 12 do v.(i) <- Reg.at_location Int (Reg i) done;
v
let hard_float_reg =
let v = Array.make 16 Reg.dummy in
for i = 0 to 15 do v.(i) <- Reg.at_location Float (Reg (100 + i)) done;
v
let all_phys_regs =
Array.append hard_int_reg hard_float_reg
let phys_reg n =
if n < 100 then hard_int_reg.(n) else hard_float_reg.(n - 100)
let rax = phys_reg 0
let rdx = phys_reg 4
let r10 = phys_reg 10
let r11 = phys_reg 11
let rbp = phys_reg 12
let rxmm15 = phys_reg 115
let destroyed_by_plt_stub =
if not X86_proc.use_plt then [| |] else [| r10; r11 |]
let num_destroyed_by_plt_stub = Array.length destroyed_by_plt_stub
let destroyed_by_plt_stub_set = Reg.set_of_array destroyed_by_plt_stub
let stack_slot slot ty =
Reg.at_location ty (Stack slot)
(* Instruction selection *)
let word_addressed = false
(* Calling conventions *)
let size_domainstate_args = 64 * size_int
let calling_conventions first_int last_int first_float last_float
make_stack first_stack
arg =
let loc = Array.make (Array.length arg) Reg.dummy in
let int = ref first_int in
let float = ref first_float in
let ofs = ref first_stack in
for i = 0 to Array.length arg - 1 do
match arg.(i) with
| Val | Int | Addr as ty ->
if !int <= last_int then begin
loc.(i) <- phys_reg !int;
incr int
end else begin
loc.(i) <- stack_slot (make_stack !ofs) ty;
ofs := !ofs + size_int
end;
assert (not (Reg.Set.mem loc.(i) destroyed_by_plt_stub_set))
| Float ->
if !float <= last_float then begin
loc.(i) <- phys_reg !float;
incr float
end else begin
loc.(i) <- stack_slot (make_stack !ofs) Float;
ofs := !ofs + size_float
end
done;
(loc, Misc.align (max 0 !ofs) 16) (* keep stack 16-aligned *)
let incoming ofs =
if ofs >= 0
then Incoming ofs
else Domainstate (ofs + size_domainstate_args)
let outgoing ofs =
if ofs >= 0
then Outgoing ofs
else Domainstate (ofs + size_domainstate_args)
let not_supported _ofs = fatal_error "Proc.loc_results: cannot call"
let loc_arguments arg =
calling_conventions 0 9 100 109 outgoing (- size_domainstate_args) arg
let loc_parameters arg =
let (loc, _ofs) =
calling_conventions 0 9 100 109 incoming (- size_domainstate_args) arg
in loc
let loc_results res =
let (loc, _ofs) = calling_conventions 0 0 100 100 not_supported 0 res
in loc
let max_arguments_for_tailcalls = 10 (* in regs *) + 64 (* in domain state *)
(* C calling conventions under Unix:
first integer args in rdi, rsi, rdx, rcx, r8, r9
first float args in xmm0 ... xmm7
remaining args on stack
return value in rax or xmm0.
C calling conventions under Win64:
first integer args in rcx, rdx, r8, r9
first float args in xmm0 ... xmm3
each integer arg consumes a float reg, and conversely
remaining args on stack
always 32 bytes reserved at bottom of stack.
Return value in rax or xmm0. *)
let loc_external_results res =
let (loc, _ofs) = calling_conventions 0 0 100 100 not_supported 0 res in loc
let unix_loc_external_arguments arg =
calling_conventions 2 7 100 107 outgoing 0 arg
let win64_int_external_arguments =
[| 5 (*rcx*); 4 (*rdx*); 6 (*r8*); 7 (*r9*) |]
let win64_float_external_arguments =
[| 100 (*xmm0*); 101 (*xmm1*); 102 (*xmm2*); 103 (*xmm3*) |]
let win64_loc_external_arguments arg =
let loc = Array.make (Array.length arg) Reg.dummy in
let reg = ref 0
and ofs = ref 32 in
for i = 0 to Array.length arg - 1 do
match arg.(i) with
| Val | Int | Addr as ty ->
if !reg < 4 then begin
loc.(i) <- phys_reg win64_int_external_arguments.(!reg);
incr reg
end else begin
loc.(i) <- stack_slot (Outgoing !ofs) ty;
ofs := !ofs + size_int
end
| Float ->
if !reg < 4 then begin
loc.(i) <- phys_reg win64_float_external_arguments.(!reg);
incr reg
end else begin
loc.(i) <- stack_slot (Outgoing !ofs) Float;
ofs := !ofs + size_float
end
done;
(loc, Misc.align !ofs 16) (* keep stack 16-aligned *)
let loc_external_arguments ty_args =
let arg = Cmm.machtype_of_exttype_list ty_args in
let loc, stack_ofs =
if win64
then win64_loc_external_arguments arg
else unix_loc_external_arguments arg
in
Array.map (fun reg -> [|reg|]) loc, stack_ofs
let loc_exn_bucket = rax
(** See "System V Application Binary Interface, AMD64 Architecture Processor
Supplement" (www.x86-64.org/documentation/abi.pdf) page 57, fig. 3.36. *)
let int_dwarf_reg_numbers =
[| 0; 3; 5; 4; 1; 2; 8; 9; 12; 13; 10; 11; 6 |]
let float_dwarf_reg_numbers =
[| 17; 18; 19; 20; 21; 22; 23; 24; 25; 26; 27; 28; 29; 30; 31; 32 |]
let dwarf_register_numbers ~reg_class =
match reg_class with
| 0 -> int_dwarf_reg_numbers
| 1 -> float_dwarf_reg_numbers
| _ -> Misc.fatal_errorf "Bad register class %d" reg_class
let stack_ptr_dwarf_register_number = 7
(* Volatile registers: none *)
let regs_are_volatile _rs = false
(* Registers destroyed by operations *)
let destroyed_at_c_call =
if win64 then
(* Win64: rbx, rbp, rsi, rdi, r12-r15, xmm6-xmm15 preserved *)
Array.of_list(List.map phys_reg
[0;4;5;6;7;10;11;
100;101;102;103;104;105])
else
(* Unix: rbp, rbx, r12-r15 preserved *)
Array.of_list(List.map phys_reg
[0;2;3;4;5;6;7;10;11;
100;101;102;103;104;105;106;107;
108;109;110;111;112;113;114;115])
let destroyed_at_alloc_or_poll =
if X86_proc.use_plt then
destroyed_by_plt_stub
else
[| r11 |]
let destroyed_at_oper = function
Iop(Icall_ind | Icall_imm _ | Iextcall { alloc = true; }) ->
all_phys_regs
| Iop(Iextcall { alloc = false; }) -> destroyed_at_c_call
| Iop(Iintop(Idiv | Imod)) | Iop(Iintop_imm((Idiv | Imod), _))
-> [| rax; rdx |]
| Iop(Istore(Single, _, _)) -> [| rxmm15 |]
| Iop(Ialloc _ | Ipoll _) -> destroyed_at_alloc_or_poll
| Iop(Iintop(Imulh | Icomp _) | Iintop_imm((Icomp _), _))
-> [| rax |]
| Iswitch(_, _) -> [| rax; rdx |]
| Itrywith _ -> [| r11 |]
| _ ->
if fp then
(* prevent any use of the frame pointer ! *)
[| rbp |]
else
[||]
let destroyed_at_raise = all_phys_regs
let destroyed_at_reloadretaddr = [| |]
(* Maximal register pressure *)
let safe_register_pressure = function
Iextcall _ -> if win64 then if fp then 7 else 8 else 0
| _ -> if fp then 10 else 11
let max_register_pressure = function
Iextcall _ ->
if win64 then
if fp then [| 7; 10 |] else [| 8; 10 |]
else
if fp then [| 3; 0 |] else [| 4; 0 |]
| Iintop(Idiv | Imod) | Iintop_imm((Idiv | Imod), _) ->
if fp then [| 10; 16 |] else [| 11; 16 |]
| Ialloc _ | Ipoll _ ->
if fp then [| 11 - num_destroyed_by_plt_stub; 16 |]
else [| 12 - num_destroyed_by_plt_stub; 16 |]
| Iintop(Icomp _) | Iintop_imm((Icomp _), _) ->
if fp then [| 11; 16 |] else [| 12; 16 |]
| Istore(Single, _, _) ->
if fp then [| 12; 15 |] else [| 13; 15 |]
| _ -> if fp then [| 12; 16 |] else [| 13; 16 |]
(* Layout of the stack frame *)
let frame_required fd =
fp || fd.fun_contains_calls ||
fd.fun_num_stack_slots.(0) > 0 || fd.fun_num_stack_slots.(1) > 0
let prologue_required fd =
frame_required fd
(* Calling the assembler *)
let assemble_file infile outfile =
X86_proc.assemble_file infile outfile
let init () =
if fp then begin
num_available_registers.(0) <- 12
end else
num_available_registers.(0) <- 13
|