summaryrefslogtreecommitdiff
path: root/asmcomp/amd64/selection.ml
blob: 3c0993f23486badbba2679c8fd285406a9383c8b (plain)
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
(**************************************************************************)
(*                                                                        *)
(*                                 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.          *)
(*                                                                        *)
(**************************************************************************)

(* Instruction selection for the AMD64 *)

open Arch
open Proc
open Cmm
open Mach

(* Auxiliary for recognizing addressing modes *)

type addressing_expr =
    Asymbol of string
  | Alinear of expression
  | Aadd of expression * expression
  | Ascale of expression * int
  | Ascaledadd of expression * expression * int

let rec select_addr exp =
  match exp with
    Cconst_symbol (s, _) when not !Clflags.dlcode ->
      (Asymbol s, 0)
  | Cop((Caddi | Caddv | Cadda), [arg; Cconst_int (m, _)], _) ->
      let (a, n) = select_addr arg in (a, n + m)
  | Cop(Csubi, [arg; Cconst_int (m, _)], _) ->
      let (a, n) = select_addr arg in (a, n - m)
  | Cop((Caddi | Caddv | Cadda), [Cconst_int (m, _); arg], _) ->
      let (a, n) = select_addr arg in (a, n + m)
  | Cop(Clsl, [arg; Cconst_int((1|2|3 as shift), _)], _) ->
      begin match select_addr arg with
        (Alinear e, n) -> (Ascale(e, 1 lsl shift), n lsl shift)
      | _ -> (Alinear exp, 0)
      end
  | Cop(Cmuli, [arg; Cconst_int((2|4|8 as mult), _)], _) ->
      begin match select_addr arg with
        (Alinear e, n) -> (Ascale(e, mult), n * mult)
      | _ -> (Alinear exp, 0)
      end
  | Cop(Cmuli, [Cconst_int((2|4|8 as mult), _); arg], _) ->
      begin match select_addr arg with
        (Alinear e, n) -> (Ascale(e, mult), n * mult)
      | _ -> (Alinear exp, 0)
      end
  | Cop((Caddi | Caddv | Cadda), [arg1; arg2], _) ->
      begin match (select_addr arg1, select_addr arg2) with
          ((Alinear e1, n1), (Alinear e2, n2)) ->
              (Aadd(e1, e2), n1 + n2)
        | ((Alinear e1, n1), (Ascale(e2, scale), n2)) ->
              (Ascaledadd(e1, e2, scale), n1 + n2)
        | ((Ascale(e1, scale), n1), (Alinear e2, n2)) ->
              (Ascaledadd(e2, e1, scale), n1 + n2)
        | (_, (Ascale(e2, scale), n2)) ->
              (Ascaledadd(arg1, e2, scale), n2)
        | ((Ascale(e1, scale), n1), _) ->
              (Ascaledadd(arg2, e1, scale), n1)
        | _ ->
              (Aadd(arg1, arg2), 0)
      end
  | arg ->
      (Alinear arg, 0)

(* Special constraints on operand and result registers *)

exception Use_default

let rax = phys_reg 0
let rcx = phys_reg 5
let rdx = phys_reg 4

let pseudoregs_for_operation op arg res =
  match op with
  (* Two-address binary operations: arg.(0) and res.(0) must be the same *)
    Iintop(Iadd|Isub|Imul|Iand|Ior|Ixor) | Iaddf|Isubf|Imulf|Idivf ->
      ([|res.(0); arg.(1)|], res)
  (* One-address unary operations: arg.(0) and res.(0) must be the same *)
  | Iintop_imm((Iadd|Isub|Imul|Iand|Ior|Ixor|Ilsl|Ilsr|Iasr), _)
  | Iabsf | Inegf
  | Ispecific(Ibswap (32|64)) ->
      (res, res)
  (* For xchg, args must be a register allowing access to high 8 bit register
     (rax, rbx, rcx or rdx). Keep it simple, just force the argument in rax. *)
  | Ispecific(Ibswap 16) ->
      ([| rax |], [| rax |])
  (* For imulq, first arg must be in rax, rax is clobbered, and result is in
     rdx. *)
  | Iintop(Imulh) ->
      ([| rax; arg.(1) |], [| rdx |])
  | Ispecific(Ifloatarithmem(_,_)) ->
      let arg' = Array.copy arg in
      arg'.(0) <- res.(0);
      (arg', res)
  (* For shifts with variable shift count, second arg must be in rcx *)
  | Iintop(Ilsl|Ilsr|Iasr) ->
      ([|res.(0); rcx|], res)
  (* For div and mod, first arg must be in rax, rdx is clobbered,
     and result is in rax or rdx respectively.
     Keep it simple, just force second argument in rcx. *)
  | Iintop(Idiv) ->
      ([| rax; rcx |], [| rax |])
  | Iintop(Imod) ->
      ([| rax; rcx |], [| rdx |])
  (* Other instructions are regular *)
  | _ -> raise Use_default

(* If you update [inline_ops], you may need to update [is_simple_expr] and/or
   [effects_of], below. *)
let inline_ops =
  [ "sqrt"; "caml_bswap16_direct"; "caml_int32_direct_bswap";
    "caml_int64_direct_bswap"; "caml_nativeint_direct_bswap" ]

(* The selector class *)

class selector = object (self)

inherit Spacetime_profiling.instruction_selection as super

method is_immediate n = n <= 0x7FFF_FFFF && n >= (-1-0x7FFF_FFFF)
  (* -1-.... : hack so that this can be compiled on 32-bit
     (cf 'make check_all_arches') *)

method is_immediate_natint n = n <= 0x7FFFFFFFn && n >= -0x80000000n

method! is_simple_expr e =
  match e with
  | Cop(Cextcall (fn, _, _, _, _), args, _)
    when List.mem fn inline_ops ->
      (* inlined ops are simple if their arguments are *)
      List.for_all self#is_simple_expr args
  | _ ->
      super#is_simple_expr e

method! effects_of e =
  match e with
  | Cop(Cextcall(fn, _, _, _, _), args, _)
    when List.mem fn inline_ops ->
      Selectgen.Effect_and_coeffect.join_list_map args self#effects_of
  | _ ->
      super#effects_of e

method select_addressing _chunk exp =
  let (a, d) = select_addr exp in
  (* PR#4625: displacement must be a signed 32-bit immediate *)
  if not (self # is_immediate d)
  then (Iindexed 0, exp)
  else match a with
    | Asymbol s ->
        (Ibased(s, d), Ctuple [])
    | Alinear e ->
        (Iindexed d, e)
    | Aadd(e1, e2) ->
        (Iindexed2 d, Ctuple[e1; e2])
    | Ascale(e, scale) ->
        (Iscaled(scale, d), e)
    | Ascaledadd(e1, e2, scale) ->
        (Iindexed2scaled(scale, d), Ctuple[e1; e2])

method! select_store is_assign addr exp =
  match exp with
    Cconst_int (n, _dbg) when self#is_immediate n ->
      (Ispecific(Istore_int(Nativeint.of_int n, addr, is_assign)), Ctuple [])
  | (Cconst_natint (n, _dbg)) when self#is_immediate_natint n ->
      (Ispecific(Istore_int(n, addr, is_assign)), Ctuple [])
  | (Cblockheader(n, _dbg))
      when self#is_immediate_natint n && not Config.spacetime ->
      (Ispecific(Istore_int(n, addr, is_assign)), Ctuple [])
  | _ ->
      super#select_store is_assign addr exp

method! select_operation op args dbg =
  match op with
  (* Recognize the LEA instruction *)
    Caddi | Caddv | Cadda | Csubi ->
      begin match self#select_addressing Word_int (Cop(op, args, dbg)) with
        (Iindexed _, _)
      | (Iindexed2 0, _) -> super#select_operation op args dbg
      | (addr, arg) -> (Ispecific(Ilea addr), [arg])
      end
  (* Recognize float arithmetic with memory. *)
  | Caddf ->
      self#select_floatarith true Iaddf Ifloatadd args
  | Csubf ->
      self#select_floatarith false Isubf Ifloatsub args
  | Cmulf ->
      self#select_floatarith true Imulf Ifloatmul args
  | Cdivf ->
      self#select_floatarith false Idivf Ifloatdiv args
  | Cextcall("sqrt", _, _, false, _) ->
     begin match args with
       [Cop(Cload ((Double|Double_u as chunk), _), [loc], _dbg)] ->
         let (addr, arg) = self#select_addressing chunk loc in
         (Ispecific(Ifloatsqrtf addr), [arg])
     | [arg] ->
         (Ispecific Isqrtf, [arg])
     | _ ->
         assert false
     end
  (* Recognize store instructions *)
  | Cstore ((Word_int|Word_val as chunk), _init) ->
      begin match args with
        [loc; Cop(Caddi, [Cop(Cload _, [loc'], _); Cconst_int (n, _dbg)], _)]
        when loc = loc' && self#is_immediate n ->
          let (addr, arg) = self#select_addressing chunk loc in
          (Ispecific(Ioffset_loc(n, addr)), [arg])
      | _ ->
          super#select_operation op args dbg
      end
  | Cextcall("caml_bswap16_direct", _, _, _, _) ->
      (Ispecific (Ibswap 16), args)
  | Cextcall("caml_int32_direct_bswap", _, _, _, _) ->
      (Ispecific (Ibswap 32), args)
  | Cextcall("caml_int64_direct_bswap", _, _, _, _)
  | Cextcall("caml_nativeint_direct_bswap", _, _, _, _) ->
      (Ispecific (Ibswap 64), args)
  (* AMD64 does not support immediate operands for multiply high signed *)
  | Cmulhi ->
      (Iintop Imulh, args)
  | Casr ->
      begin match args with
        (* Recognize sign extension *)
        [Cop(Clsl, [k; Cconst_int (32, _)], _); Cconst_int (32, _)] ->
          (Ispecific Isextend32, [k])
        | _ -> super#select_operation op args dbg
      end
  (* Recognize zero extension *)
  | Cand ->
    begin match args with
    | [arg; Cconst_int (0xffff_ffff, _)]
    | [arg; Cconst_natint (0xffff_ffffn, _)]
    | [Cconst_int (0xffff_ffff, _); arg]
    | [Cconst_natint (0xffff_ffffn, _); arg] ->
      Ispecific Izextend32, [arg]
    | _ -> super#select_operation op args dbg
    end
  | _ -> super#select_operation op args dbg

(* Recognize float arithmetic with mem *)

method select_floatarith commutative regular_op mem_op args =
  match args with
    [arg1; Cop(Cload ((Double|Double_u as chunk), _), [loc2], _)] ->
      let (addr, arg2) = self#select_addressing chunk loc2 in
      (Ispecific(Ifloatarithmem(mem_op, addr)),
                 [arg1; arg2])
  | [Cop(Cload ((Double|Double_u as chunk), _), [loc1], _); arg2]
        when commutative ->
      let (addr, arg1) = self#select_addressing chunk loc1 in
      (Ispecific(Ifloatarithmem(mem_op, addr)),
                 [arg2; arg1])
  | [arg1; arg2] ->
      (regular_op, [arg1; arg2])
  | _ ->
      assert false

method! mark_c_tailcall =
  contains_calls := true

(* Deal with register constraints *)

method! insert_op_debug env op dbg rs rd =
  try
    let (rsrc, rdst) = pseudoregs_for_operation op rs rd in
    self#insert_moves env rs rsrc;
    self#insert_debug env (Iop op) dbg rsrc rdst;
    self#insert_moves env rdst rd;
    rd
  with Use_default ->
    super#insert_op_debug env op dbg rs rd

end

let fundecl f = (new selector)#emit_fundecl f