path: root/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 Q Public License version 1.0.               *)
(*                                                                     *)
(***********************************************************************)

(* Description of the AMD64 processor *)

open Misc
open Arch
open Cmm
open Reg
open Mach

(* Which ABI to use *)

let win64 =
  match Config.system with
  | "win64" | "mingw64" -> true
  | _                   -> false

(* Which asm conventions to use *)

let masm =
  match Config.ccomp_type with
  | "msvc" -> true
  | _      -> false

(* 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         trap 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): r11 should not be used for parameter passing, as it
     can be destroyed by the dynamic loader according to SVR4 ABI.
     Linux's dynamic loader also destroys r10.
*)

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
    Int -> 0
  | 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.create 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.create 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 rcx = phys_reg 5
let rdx = phys_reg 4
let rbp = phys_reg 12
let rxmm15 = phys_reg 115

let stack_slot slot ty =
  Reg.at_location ty (Stack slot)

(* Instruction selection *)

let word_addressed = false

(* Calling conventions *)

let calling_conventions first_int last_int first_float last_float make_stack
                        arg =
  let loc = Array.create (Array.length arg) Reg.dummy in
  let int = ref first_int in
  let float = ref first_float in
  let ofs = ref 0 in
  for i = 0 to Array.length arg - 1 do
    match arg.(i).typ with
      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
    | 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 !ofs 16)  (* keep stack 16-aligned *)

let incoming ofs = Incoming ofs
let outgoing ofs = Outgoing ofs
let not_supported ofs = fatal_error "Proc.loc_results: cannot call"

let loc_arguments arg =
  calling_conventions 0 9 100 109 outgoing arg
let loc_parameters arg =
  let (loc, ofs) = calling_conventions 0 9 100 109 incoming arg in loc
let loc_results res =
  let (loc, ofs) = calling_conventions 0 0 100 100 not_supported res in loc

(* 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 res in loc

let unix_loc_external_arguments arg =
  calling_conventions 2 7 100 107 outgoing 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.create (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).typ with
      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 =
  if win64 then win64_loc_external_arguments else unix_loc_external_arguments

let loc_exn_bucket = rax

(* 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_oper = function
    Iop(Icall_ind | Icall_imm _ | Iextcall(_, true)) -> all_phys_regs
  | Iop(Iextcall(_, false)) -> destroyed_at_c_call
  | Iop(Iintop(Idiv | Imod)) -> [| rax; rdx |]
  | Iop(Istore(Single, _)) -> [| rxmm15 |]
  | Iop(Ialloc _ | Iintop(Icomp _) | Iintop_imm((Idiv|Imod|Icomp _), _))
        -> [| rax |]
  | Iswitch(_, _) -> [| rax; rdx |]
  | _ ->
    if Config.with_frame_pointers then
(* prevent any use of the frame pointer ! *)
      [| rbp |]
    else
      [||]


let destroyed_at_raise = all_phys_regs

(* Maximal register pressure *)

let fp() = Config.with_frame_pointers

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) ->
    if fp() then [| 10; 16 |] else [| 11; 16 |]
  | Ialloc _ | Iintop(Icomp _) | Iintop_imm((Idiv|Imod|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 num_stack_slots = [| 0; 0 |]
let contains_calls = ref false

(* Calling the assembler *)

let assemble_file infile outfile =
  if masm then
    Ccomp.command (Config.asm ^
                   Filename.quote outfile ^ " " ^ Filename.quote infile ^
                   (if !Clflags.verbose then "" else ">NUL"))
  else
    Ccomp.command (Config.asm ^ " -o " ^
                   Filename.quote outfile ^ " " ^ Filename.quote infile)

let init () =
  if fp () then begin
    num_available_registers.(0) <- 12
  end else
    num_available_registers.(0) <- 13