open Op
open PrepassSchedulingOracleDeps

module FU74 = struct
  (* Attempt at modeling the FU74 (HiFive Unmatched board) core *)

  let resource_bounds = [| 2; 1; 1; 1; 1 |]
  (* issue ; LSU ; BU ; FPU ; IMUL/IDIV *)

  let nr_non_pipelined_units = 1

  (* divider *)

  let latency_of_op (op : operation) (nargs : int) =
    match op with
    | OEmayundef _ -> 0
    | Omove | Ointconst _ | Olongconst _
    | Oaddrsymbol (_, _)
    | Oaddrstack _ | Ocast8signed | Ocast16signed | Oadd | Oaddimm _ | Oneg
    | Osub | Oand | Oandimm _ | Oor | Oorimm _ | Oxor | Oxorimm _ | Oshl
    | Oshlimm _ | Oshr | Oshrimm _ | Oshru | Oshruimm _ | Oshrximm _ | Olowlong
    | Ocast32signed | Ocast32unsigned | Oaddl | Oaddlimm _ | Onegl | Osubl
    | Oandl | Oandlimm _ | Oorl | Oorlimm _ | Oxorl | Oxorlimm _ | Oshll
    | Oshllimm _ | Oshrl | Oshrlimm _ | Oshrlu | Oshrluimm _ | Oshrxlimm _
    | Oselectl | Obits_of_single | Obits_of_float | OEseqw _ | OEsnew _
    | OEsequw _ | OEsneuw _ | OEsltw _ | OEsltuw _ | OEsltiw _ | OEsltiuw _
    | OEaddiw (_, _)
    | OEandiw _ | OEoriw _ | OExoriw _ | OEluiw _ | OEseql _ | OEsnel _
    | OEsequl _ | OEsneul _ | OEsltl _ | OEsltul _ | OEsltil _ | OEsltiul _
    | OEaddil (_, _)
    | OEandil _ | OEoril _ | OExoril _ | OEluil _ | OEloadli _ ->
        1
    | Osingleconst _ | Ofloatconst _ | Onegf | Oabsf | Onegfs | Oabsfs
    | Osingleoffloat | Ofloatofsingle | Ofloatofint | Ofloatofintu
    | Osingleofint | Osingleofintu | Osingle_of_bits ->
        2
    | Omul | Omulhs | Omulhu | Omull | Omullhs | Omullhu -> 3
    | Omulf -> 7
    | Omulfs -> 5
    | Ointoffloat | Ointuoffloat | Ointofsingle | Ointuofsingle | Olongoffloat
    | Olonguoffloat | Olongofsingle | Olonguofsingle | Osingleoflong
    | Osingleoflongu | OEfeqd | OEfltd | OEfled | OEfeqs | OEflts | OEfles ->
        4
    | Ofloatoflong | Ofloatoflongu | Ofloat_of_bits -> 6
    | Oaddf | Osubf | Oaddfs | Osubfs -> 7
    | Ocmp cond -> (
        match cond with
        | Ccomp _ | Ccompu _ | Ccompimm _ | Ccompuimm _ | Ccompl _ | Ccomplu _
        | Ccomplimm _ | Ccompluimm _ | CEbeqw _ | CEbnew _ | CEbequw _
        | CEbneuw _ | CEbltw _ | CEbltuw _ | CEbgew _ | CEbgeuw _ | CEbeql _
        | CEbnel _ | CEbequl _ | CEbneul _ | CEbltl _ | CEbltul _ | CEbgel _
        | CEbgeul _ ->
            1
        | Ccompf _ | Cnotcompf _ | Ccompfs _ | Cnotcompfs _ -> 4)
    | Odiv | Odivu | Omod | Omodu | Odivl | Odivlu | Omodl | Omodlu | Odivf
    | Odivfs ->
        68
    | _ -> 1

  let resources_of_op (op : operation) (nargs : int) =
    match op with
    | OEmayundef _ -> [| 0; 0; 0; 0; 0 |]
    | Omove | Ointconst _ | Olongconst _
    | Oaddrsymbol (_, _)
    | Oaddrstack _ | Ocast8signed | Ocast16signed | Oadd | Oaddimm _ | Oneg
    | Osub | Oand | Oandimm _ | Oor | Oorimm _ | Oxor | Oxorimm _ | Oshl
    | Oshlimm _ | Oshr | Oshrimm _ | Oshru | Oshruimm _ | Oshrximm _ | Olowlong
    | Ocast32signed | Ocast32unsigned | Oaddl | Oaddlimm _ | Onegl | Osubl
    | Oandl | Oandlimm _ | Oorl | Oorlimm _ | Oxorl | Oxorlimm _ | Oshll
    | Oshllimm _ | Oshrl | Oshrlimm _ | Oshrlu | Oshrluimm _ | Oshrxlimm _
    | Oselectl | Obits_of_single | Obits_of_float | OEseqw _ | OEsnew _
    | OEsequw _ | OEsneuw _ | OEsltw _ | OEsltuw _ | OEsltiw _ | OEsltiuw _
    | OEaddiw (_, _)
    | OEandiw _ | OEoriw _ | OExoriw _ | OEluiw _ | OEseql _ | OEsnel _
    | OEsequl _ | OEsneul _ | OEsltl _ | OEsltul _ | OEsltil _ | OEsltiul _
    | OEaddil (_, _)
    | OEandil _ | OEoril _ | OExoril _ | OEluil _ | OEloadli _ ->
        [| 1; 0; 0; 0; 0 |]
    | Omul | Omulhs | Omulhu | Omull | Omullhs | Omullhu | Odiv | Odivu | Omod
    | Omodu | Odivl | Odivlu | Omodl | Omodlu ->
        [| 1; 0; 0; 0; 1 |]
    | Omulf | Omulfs | Ointoffloat | Ointuoffloat | Ointofsingle | Ointuofsingle
    | Olongoffloat | Olonguoffloat | Olongofsingle | Olonguofsingle
    | Osingleoflong | Osingleoflongu | OEfeqd | OEfltd | OEfled | OEfeqs
    | OEflts | OEfles | Ofloatoflong | Ofloatoflongu | Ofloat_of_bits | Oaddf
    | Osubf | Oaddfs | Osubfs | Osingleconst _ | Ofloatconst _ | Onegf | Oabsf
    | Onegfs | Oabsfs | Osingleoffloat | Ofloatofsingle | Ofloatofint
    | Ofloatofintu | Osingleofint | Osingleofintu | Osingle_of_bits | Odivf
    | Odivfs ->
        [| 1; 0; 0; 1; 0 |]
    | Ocmp cond -> (
        match cond with
        | Ccomp _ | Ccompu _ | Ccompimm _ | Ccompuimm _ | Ccompl _ | Ccomplu _
        | Ccomplimm _ | Ccompluimm _ | CEbeqw _ | CEbnew _ | CEbequw _
        | CEbneuw _ | CEbltw _ | CEbltuw _ | CEbgew _ | CEbgeuw _ | CEbeql _
        | CEbnel _ | CEbequl _ | CEbneul _ | CEbltl _ | CEbltul _ | CEbgel _
        | CEbgeul _ ->
            [| 1; 0; 0; 0; 0 |]
        | Ccompf _ | Cnotcompf _ | Ccompfs _ | Cnotcompfs _ ->
            [| 1; 0; 0; 1; 0 |])
    | _ -> [| 1; 0; 0; 0; 0 |]

  let non_pipelined_resources_of_op (op : operation) (nargs : int) =
    match op with
    | Odiv | Odivu | Odivl | Odivlu | Odivf | Odivfs -> [| 68 |]
    | _ -> [| -1 |]

  let resources_of_cond (cond : condition) (nargs : int) =
    match cond with
    | Ccompf _ | Cnotcompf _ | Ccompfs _ | Cnotcompfs _ -> [| 1; 0; 1; 1; 0 |]
    | _ -> [| 1; 0; 1; 0; 0 |]

  let latency_of_load trap chunk (addr : addressing) (nargs : int) = 3

  let latency_of_call _ _ = 6

  let resources_of_load trap chunk addressing nargs = [| 1; 1; 0; 0; 0 |]

  let resources_of_store chunk addressing nargs = [| 1; 1; 0; 0; 0 |]

  let resources_of_call _ _ = resource_bounds

  let resources_of_builtin _ = resource_bounds
end

module Rocket = struct
  (* Attempt at modeling the Rocket core *)

  let resource_bounds = [| 1 |]

  let nr_non_pipelined_units = 1

  (* divider *)

  let latency_of_op (op : operation) (nargs : int) =
    match op with
    | Omul | Omulhs | Omulhu | Omull | Omullhs | Omullhu -> 4
    | Onegf -> 1 (*r [rd = - r1] *)
    | Oabsf (*r [rd = abs(r1)] *)
    | Oaddf (*r [rd = r1 + r2] *)
    | Osubf (*r [rd = r1 - r2] *)
    | Omulf ->
        6 (*r [rd = r1 * r2] *)
    | Onegfs -> 1 (*r [rd = - r1] *)
    | Oabsfs (*r [rd = abs(r1)] *)
    | Oaddfs (*r [rd = r1 + r2] *)
    | Osubfs (*r [rd = r1 - r2] *)
    | Omulfs ->
        4 (*r [rd = r1 * r2] *)
    | Osingleoffloat (*r [rd] is [r1] truncated to single-precision float *)
    | Ofloatofsingle (*r [rd] is [r1] extended to double-precision float *)
    (*c Conversions between int and float: *)
    | Ofloatconst _ | Osingleconst _
    | Ointoffloat (*r [rd = signed_int_of_float64(r1)] *)
    | Ointuoffloat (*r [rd = unsigned_int_of_float64(r1)] *)
    | Ofloatofint (*r [rd = float64_of_signed_int(r1)] *)
    | Ofloatofintu (*r [rd = float64_of_unsigned_int(r1)] *)
    | Ointofsingle (*r [rd = signed_int_of_float32(r1)] *)
    | Ointuofsingle (*r [rd = unsigned_int_of_float32(r1)] *)
    | Osingleofint (*r [rd = float32_of_signed_int(r1)] *)
    | Osingleofintu (*r [rd = float32_of_unsigned_int(r1)] *)
    | Olongoffloat (*r [rd = signed_long_of_float64(r1)] *)
    | Olonguoffloat (*r [rd = unsigned_long_of_float64(r1)] *)
    | Ofloatoflong (*r [rd = float64_of_signed_long(r1)] *)
    | Ofloatoflongu (*r [rd = float64_of_unsigned_long(r1)] *)
    | Olongofsingle (*r [rd = signed_long_of_float32(r1)] *)
    | Olonguofsingle (*r [rd = unsigned_long_of_float32(r1)] *)
    | Osingleoflong (*r [rd = float32_of_signed_long(r1)] *)
    | Osingleoflongu ->
        2 (*r [rd = float32_of_unsigned_int(r1)] *)
    | OEfeqd | OEfltd | OEfeqs | OEflts | OEfles | OEfled | Obits_of_single
    | Obits_of_float | Osingle_of_bits | Ofloat_of_bits ->
        2
    | OEloadli _ -> 2
    | Odiv | Odivu | Odivl | Odivlu -> 16
    | Odivfs -> 35
    | Odivf -> 50
    | Ocmp cond -> (
        match cond with
        | Ccomp _ | Ccompu _ | Ccompimm _ | Ccompuimm _ | Ccompl _ | Ccomplu _
        | Ccomplimm _ | Ccompluimm _ | CEbeqw _ | CEbnew _ | CEbequw _
        | CEbneuw _ | CEbltw _ | CEbltuw _ | CEbgew _ | CEbgeuw _ | CEbeql _
        | CEbnel _ | CEbequl _ | CEbneul _ | CEbltl _ | CEbltul _ | CEbgel _
        | CEbgeul _ ->
            1
        | Ccompf _ | Cnotcompf _ -> 2
        | Ccompfs _ | Cnotcompfs _ -> 2)
    | OEmayundef _ -> 0
    | _ -> 1

  let resources_of_op (op : operation) (nargs : int) =
    match op with OEmayundef _ -> [| 0 |] | _ -> resource_bounds

  let non_pipelined_resources_of_op (op : operation) (nargs : int) =
    match op with
    | Odiv | Odivu -> [| 29 |]
    | Odivfs -> [| 20 |]
    | Odivl | Odivlu | Odivf -> [| 50 |]
    | _ -> [| -1 |]

  let resources_of_cond (cond : condition) (nargs : int) = resource_bounds

  let latency_of_load trap chunk (addr : addressing) (nargs : int) = 3

  let latency_of_call _ _ = 6

  let resources_of_load trap chunk addressing nargs = resource_bounds

  let resources_of_store chunk addressing nargs = resource_bounds

  let resources_of_call _ _ = resource_bounds

  let resources_of_builtin _ = resource_bounds
end

module SweRV_EH1 = struct
  (* Attempt at modeling SweRV EH1
     [| issues ; LSU ; multiplier |] *)
  let resource_bounds = [| 2; 1; 1 |]

  let nr_non_pipelined_units = 1

  (* divider *)

  let latency_of_op (op : operation) (nargs : int) =
    match op with
    | Omul | Omulhs | Omulhu | Omull | Omullhs | Omullhu -> 3
    | Odiv | Odivu | Odivl | Odivlu -> 16
    | _ -> 1

  let resources_of_op (op : operation) (nargs : int) =
    match op with
    | Omul | Omulhs | Omulhu | Omull | Omullhs | Omullhu -> [| 1; 0; 1 |]
    | Odiv | Odivu | Odivl | Odivlu -> [| 0; 0; 0 |]
    | _ -> [| 1; 0; 0 |]

  let non_pipelined_resources_of_op (op : operation) (nargs : int) =
    match op with
    | Odiv | Odivu -> [| 29 |]
    | Odivfs -> [| 20 |]
    | Odivl | Odivlu | Odivf -> [| 50 |]
    | _ -> [| -1 |]

  let resources_of_cond (cond : condition) (nargs : int) = [| 1; 0; 0 |]

  let latency_of_load trap chunk (addr : addressing) (nargs : int) = 3

  let latency_of_call _ _ = 6

  let resources_of_load trap chunk addressing nargs = [| 1; 1; 0 |]

  let resources_of_store chunk addressing nargs = [| 1; 1; 0 |]

  let resources_of_call _ _ = resource_bounds

  let resources_of_builtin _ = resource_bounds
end

let get_opweights () : opweights =
  match !Clflags.option_mtune with
  | "rocket" | "" ->
      {
        pipelined_resource_bounds = Rocket.resource_bounds;
        nr_non_pipelined_units = Rocket.nr_non_pipelined_units;
        latency_of_op = Rocket.latency_of_op;
        resources_of_op = Rocket.resources_of_op;
        non_pipelined_resources_of_op = Rocket.non_pipelined_resources_of_op;
        latency_of_load = Rocket.latency_of_load;
        resources_of_load = Rocket.resources_of_load;
        resources_of_store = Rocket.resources_of_store;
        resources_of_cond = Rocket.resources_of_cond;
        latency_of_call = Rocket.latency_of_call;
        resources_of_call = Rocket.resources_of_call;
        resources_of_builtin = Rocket.resources_of_builtin;
      }
  | "SweRV_EH1" | "EH1" ->
      {
        pipelined_resource_bounds = SweRV_EH1.resource_bounds;
        nr_non_pipelined_units = SweRV_EH1.nr_non_pipelined_units;
        latency_of_op = SweRV_EH1.latency_of_op;
        resources_of_op = SweRV_EH1.resources_of_op;
        non_pipelined_resources_of_op = SweRV_EH1.non_pipelined_resources_of_op;
        latency_of_load = SweRV_EH1.latency_of_load;
        resources_of_load = SweRV_EH1.resources_of_load;
        resources_of_store = SweRV_EH1.resources_of_store;
        resources_of_cond = SweRV_EH1.resources_of_cond;
        latency_of_call = SweRV_EH1.latency_of_call;
        resources_of_call = SweRV_EH1.resources_of_call;
        resources_of_builtin = SweRV_EH1.resources_of_builtin;
      }
  | "FU74" | "sifive-u74" ->
      {
        pipelined_resource_bounds = FU74.resource_bounds;
        nr_non_pipelined_units = FU74.nr_non_pipelined_units;
        latency_of_op = FU74.latency_of_op;
        resources_of_op = FU74.resources_of_op;
        non_pipelined_resources_of_op = FU74.non_pipelined_resources_of_op;
        latency_of_load = FU74.latency_of_load;
        resources_of_load = FU74.resources_of_load;
        resources_of_store = FU74.resources_of_store;
        resources_of_cond = FU74.resources_of_cond;
        latency_of_call = FU74.latency_of_call;
        resources_of_call = FU74.resources_of_call;
        resources_of_builtin = FU74.resources_of_builtin;
      }
  | xxx -> failwith (Printf.sprintf "unknown -mtune: %s" xxx)