annotate src/fftw-3.3.8/genfft/gen_hc2hc.ml @ 82:d0c2a83c1364

Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
rev   line source
Chris@82 1 (*
Chris@82 2 * Copyright (c) 1997-1999 Massachusetts Institute of Technology
Chris@82 3 * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82 4 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82 5 *
Chris@82 6 * This program is free software; you can redistribute it and/or modify
Chris@82 7 * it under the terms of the GNU General Public License as published by
Chris@82 8 * the Free Software Foundation; either version 2 of the License, or
Chris@82 9 * (at your option) any later version.
Chris@82 10 *
Chris@82 11 * This program is distributed in the hope that it will be useful,
Chris@82 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Chris@82 14 * GNU General Public License for more details.
Chris@82 15 *
Chris@82 16 * You should have received a copy of the GNU General Public License
Chris@82 17 * along with this program; if not, write to the Free Software
Chris@82 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Chris@82 19 *
Chris@82 20 *)
Chris@82 21
Chris@82 22 open Util
Chris@82 23 open Genutil
Chris@82 24 open C
Chris@82 25
Chris@82 26
Chris@82 27 type ditdif = DIT | DIF
Chris@82 28 let ditdif = ref DIT
Chris@82 29 let usage = "Usage: " ^ Sys.argv.(0) ^ " -n <number> [ -dit | -dif ]"
Chris@82 30
Chris@82 31 let urs = ref Stride_variable
Chris@82 32
Chris@82 33 let speclist = [
Chris@82 34 "-dit",
Chris@82 35 Arg.Unit(fun () -> ditdif := DIT),
Chris@82 36 " generate a DIT codelet";
Chris@82 37
Chris@82 38 "-dif",
Chris@82 39 Arg.Unit(fun () -> ditdif := DIF),
Chris@82 40 " generate a DIF codelet";
Chris@82 41
Chris@82 42 "-with-rs",
Chris@82 43 Arg.String(fun x -> urs := arg_to_stride x),
Chris@82 44 " specialize for given R-stride";
Chris@82 45 ]
Chris@82 46
Chris@82 47 let rioarray = "cr"
Chris@82 48 and iioarray = "ci"
Chris@82 49
Chris@82 50 let genone sign n transform load store vrs =
Chris@82 51 let locations = unique_array_c n in
Chris@82 52 let input =
Chris@82 53 locative_array_c n
Chris@82 54 (C.array_subscript rioarray vrs)
Chris@82 55 (C.array_subscript iioarray vrs)
Chris@82 56 locations "BUG" in
Chris@82 57 let output = transform sign n (load n input) in
Chris@82 58 let ioloc =
Chris@82 59 locative_array_c n
Chris@82 60 (C.array_subscript rioarray vrs)
Chris@82 61 (C.array_subscript iioarray vrs)
Chris@82 62 locations "BUG" in
Chris@82 63 let odag = store n ioloc output in
Chris@82 64 let annot = standard_optimizer odag
Chris@82 65 in annot
Chris@82 66
Chris@82 67 let byi = Complex.times Complex.i
Chris@82 68 let byui = Complex.times (Complex.uminus Complex.i)
Chris@82 69
Chris@82 70 let sym1 n f i =
Chris@82 71 Complex.plus [Complex.real (f i); byi (Complex.imag (f (n - 1 - i)))]
Chris@82 72
Chris@82 73 let sym2 n f i = if (i < n - i) then f i else byi (f i)
Chris@82 74 let sym2i n f i = if (i < n - i) then f i else byui (f i)
Chris@82 75
Chris@82 76 let generate n =
Chris@82 77 let rs = "rs"
Chris@82 78 and twarray = "W"
Chris@82 79 and m = "m" and mb = "mb" and me = "me" and ms = "ms" in
Chris@82 80
Chris@82 81 let sign = !Genutil.sign
Chris@82 82 and name = !Magic.codelet_name
Chris@82 83 and byvl x = choose_simd x (ctimes (CVar "VL", x)) in
Chris@82 84
Chris@82 85 let (bytwiddle, num_twiddles, twdesc) = Twiddle.twiddle_policy 1 false in
Chris@82 86 let nt = num_twiddles n in
Chris@82 87
Chris@82 88 let byw = bytwiddle n sign (twiddle_array nt twarray) in
Chris@82 89
Chris@82 90 let vrs = either_stride (!urs) (C.SVar rs) in
Chris@82 91
Chris@82 92 let asch =
Chris@82 93 match !ditdif with
Chris@82 94 | DIT ->
Chris@82 95 genone sign n
Chris@82 96 (fun sign n input ->
Chris@82 97 ((sym1 n) @@ (sym2 n)) (Fft.dft sign n (byw input)))
Chris@82 98 load_array_c store_array_c vrs
Chris@82 99 | DIF ->
Chris@82 100 genone sign n
Chris@82 101 (fun sign n input ->
Chris@82 102 byw (Fft.dft sign n (((sym2i n) @@ (sym1 n)) input)))
Chris@82 103 load_array_c store_array_c vrs
Chris@82 104 in
Chris@82 105
Chris@82 106 let vms = CVar "ms"
Chris@82 107 and vrioarray = CVar rioarray
Chris@82 108 and viioarray = CVar iioarray
Chris@82 109 and vm = CVar m and vmb = CVar mb and vme = CVar me
Chris@82 110 in
Chris@82 111 let body = Block (
Chris@82 112 [Decl ("INT", m)],
Chris@82 113 [For (list_to_comma
Chris@82 114 [Expr_assign (vm, vmb);
Chris@82 115 Expr_assign (CVar twarray,
Chris@82 116 CPlus [CVar twarray;
Chris@82 117 ctimes (CPlus [vmb; CUminus (Integer 1)],
Chris@82 118 Integer nt)])],
Chris@82 119 Binop (" < ", vm, vme),
Chris@82 120 list_to_comma
Chris@82 121 [Expr_assign (vm, CPlus [vm; byvl (Integer 1)]);
Chris@82 122 Expr_assign (vrioarray, CPlus [vrioarray; byvl vms]);
Chris@82 123 Expr_assign (viioarray,
Chris@82 124 CPlus [viioarray; CUminus (byvl vms)]);
Chris@82 125 Expr_assign (CVar twarray, CPlus [CVar twarray;
Chris@82 126 byvl (Integer nt)]);
Chris@82 127 make_volatile_stride (2*n) (CVar rs)
Chris@82 128 ],
Chris@82 129 Asch asch)])
Chris@82 130 in
Chris@82 131
Chris@82 132 let tree =
Chris@82 133 Fcn ("static void", name,
Chris@82 134 [Decl (C.realtypep, rioarray);
Chris@82 135 Decl (C.realtypep, iioarray);
Chris@82 136 Decl (C.constrealtypep, twarray);
Chris@82 137 Decl (C.stridetype, rs);
Chris@82 138 Decl ("INT", mb);
Chris@82 139 Decl ("INT", me);
Chris@82 140 Decl ("INT", ms)],
Chris@82 141 finalize_fcn body)
Chris@82 142 in
Chris@82 143 let twinstr =
Chris@82 144 Printf.sprintf "static const tw_instr twinstr[] = %s;\n\n"
Chris@82 145 (twinstr_to_string "VL" (twdesc n))
Chris@82 146 and desc =
Chris@82 147 Printf.sprintf
Chris@82 148 "static const hc2hc_desc desc = {%d, \"%s\", twinstr, &GENUS, %s};\n\n"
Chris@82 149 n name (flops_of tree)
Chris@82 150 and register = "X(khc2hc_register)"
Chris@82 151
Chris@82 152 in
Chris@82 153 let init =
Chris@82 154 "\n" ^
Chris@82 155 twinstr ^
Chris@82 156 desc ^
Chris@82 157 (declare_register_fcn name) ^
Chris@82 158 (Printf.sprintf "{\n%s(p, %s, &desc);\n}" register name)
Chris@82 159 in
Chris@82 160
Chris@82 161 (unparse tree) ^ "\n" ^ init
Chris@82 162
Chris@82 163
Chris@82 164 let main () =
Chris@82 165 begin
Chris@82 166 parse (speclist @ Twiddle.speclist) usage;
Chris@82 167 print_string (generate (check_size ()));
Chris@82 168 end
Chris@82 169
Chris@82 170 let _ = main()