annotate src/fftw-3.3.8/genfft/gen_hc2c.ml @ 167:bd3cc4d1df30

Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam <cannam@all-day-breakfast.com>
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
rev   line source
cannam@167 1 (*
cannam@167 2 * Copyright (c) 1997-1999 Massachusetts Institute of Technology
cannam@167 3 * Copyright (c) 2003, 2007-14 Matteo Frigo
cannam@167 4 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
cannam@167 5 *
cannam@167 6 * This program is free software; you can redistribute it and/or modify
cannam@167 7 * it under the terms of the GNU General Public License as published by
cannam@167 8 * the Free Software Foundation; either version 2 of the License, or
cannam@167 9 * (at your option) any later version.
cannam@167 10 *
cannam@167 11 * This program is distributed in the hope that it will be useful,
cannam@167 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@167 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
cannam@167 14 * GNU General Public License for more details.
cannam@167 15 *
cannam@167 16 * You should have received a copy of the GNU General Public License
cannam@167 17 * along with this program; if not, write to the Free Software
cannam@167 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
cannam@167 19 *
cannam@167 20 *)
cannam@167 21
cannam@167 22 open Util
cannam@167 23 open Genutil
cannam@167 24 open C
cannam@167 25
cannam@167 26
cannam@167 27 type ditdif = DIT | DIF
cannam@167 28 let ditdif = ref DIT
cannam@167 29 let usage = "Usage: " ^ Sys.argv.(0) ^ " -n <number> [ -dit | -dif ]"
cannam@167 30
cannam@167 31 let urs = ref Stride_variable
cannam@167 32
cannam@167 33 let speclist = [
cannam@167 34 "-dit",
cannam@167 35 Arg.Unit(fun () -> ditdif := DIT),
cannam@167 36 " generate a DIT codelet";
cannam@167 37
cannam@167 38 "-dif",
cannam@167 39 Arg.Unit(fun () -> ditdif := DIF),
cannam@167 40 " generate a DIF codelet";
cannam@167 41
cannam@167 42 "-with-rs",
cannam@167 43 Arg.String(fun x -> urs := arg_to_stride x),
cannam@167 44 " specialize for given R-stride";
cannam@167 45 ]
cannam@167 46
cannam@167 47 let byi = Complex.times Complex.i
cannam@167 48 let byui = Complex.times (Complex.uminus Complex.i)
cannam@167 49
cannam@167 50 let sym n f i = if (i < n - i) then f i else Complex.conj (f i)
cannam@167 51
cannam@167 52 let shuffle_eo fe fo i = if i mod 2 == 0 then fe (i/2) else fo ((i-1)/2)
cannam@167 53
cannam@167 54 let generate n =
cannam@167 55 let rs = "rs"
cannam@167 56 and twarray = "W"
cannam@167 57 and m = "m" and mb = "mb" and me = "me" and ms = "ms"
cannam@167 58
cannam@167 59 (* the array names are from the point of view of the complex array
cannam@167 60 (output in R2C, input in C2R) *)
cannam@167 61 and arp = "Rp" (* real, positive *)
cannam@167 62 and aip = "Ip" (* imag, positive *)
cannam@167 63 and arm = "Rm" (* real, negative *)
cannam@167 64 and aim = "Im" (* imag, negative *)
cannam@167 65
cannam@167 66 in
cannam@167 67
cannam@167 68 let sign = !Genutil.sign
cannam@167 69 and name = !Magic.codelet_name
cannam@167 70 and byvl x = choose_simd x (ctimes (CVar "VL", x)) in
cannam@167 71
cannam@167 72 let (bytwiddle, num_twiddles, twdesc) = Twiddle.twiddle_policy 1 false in
cannam@167 73 let nt = num_twiddles n in
cannam@167 74
cannam@167 75 let byw = bytwiddle n sign (twiddle_array nt twarray) in
cannam@167 76
cannam@167 77 let vrs = either_stride (!urs) (C.SVar rs) in
cannam@167 78
cannam@167 79 (* assume a single location. No point in doing alias analysis *)
cannam@167 80 let the_location = (Unique.make (), Unique.make ()) in
cannam@167 81 let locations _ = the_location in
cannam@167 82
cannam@167 83 let locr = (locative_array_c n
cannam@167 84 (C.array_subscript arp vrs)
cannam@167 85 (C.array_subscript arm vrs)
cannam@167 86 locations "BUG")
cannam@167 87 and loci = (locative_array_c n
cannam@167 88 (C.array_subscript aip vrs)
cannam@167 89 (C.array_subscript aim vrs)
cannam@167 90 locations "BUG")
cannam@167 91 and locp = (locative_array_c n
cannam@167 92 (C.array_subscript arp vrs)
cannam@167 93 (C.array_subscript aip vrs)
cannam@167 94 locations "BUG")
cannam@167 95 and locm = (locative_array_c n
cannam@167 96 (C.array_subscript arm vrs)
cannam@167 97 (C.array_subscript aim vrs)
cannam@167 98 locations "BUG")
cannam@167 99 in
cannam@167 100 let locri i = if i mod 2 == 0 then locr (i/2) else loci ((i-1)/2)
cannam@167 101 and locpm i = if i < n - i then locp i else locm (n-1-i)
cannam@167 102 in
cannam@167 103
cannam@167 104 let asch =
cannam@167 105 match !ditdif with
cannam@167 106 | DIT ->
cannam@167 107 let output = Fft.dft sign n (byw (load_array_c n locri)) in
cannam@167 108 let odag = store_array_c n locpm (sym n output) in
cannam@167 109 standard_optimizer odag
cannam@167 110
cannam@167 111 | DIF ->
cannam@167 112 let output = byw (Fft.dft sign n (sym n (load_array_c n locpm))) in
cannam@167 113 let odag = store_array_c n locri output in
cannam@167 114 standard_optimizer odag
cannam@167 115 in
cannam@167 116
cannam@167 117 let vms = CVar "ms"
cannam@167 118 and varp = CVar arp
cannam@167 119 and vaip = CVar aip
cannam@167 120 and varm = CVar arm
cannam@167 121 and vaim = CVar aim
cannam@167 122 and vm = CVar m and vmb = CVar mb and vme = CVar me
cannam@167 123 in
cannam@167 124 let body = Block (
cannam@167 125 [Decl ("INT", m)],
cannam@167 126 [For (list_to_comma
cannam@167 127 [Expr_assign (vm, vmb);
cannam@167 128 Expr_assign (CVar twarray,
cannam@167 129 CPlus [CVar twarray;
cannam@167 130 ctimes (CPlus [vmb; CUminus (Integer 1)],
cannam@167 131 Integer nt)])],
cannam@167 132 Binop (" < ", vm, vme),
cannam@167 133 list_to_comma
cannam@167 134 [Expr_assign (vm, CPlus [vm; byvl (Integer 1)]);
cannam@167 135 Expr_assign (varp, CPlus [varp; byvl vms]);
cannam@167 136 Expr_assign (vaip, CPlus [vaip; byvl vms]);
cannam@167 137 Expr_assign (varm, CPlus [varm; CUminus (byvl vms)]);
cannam@167 138 Expr_assign (vaim, CPlus [vaim; CUminus (byvl vms)]);
cannam@167 139 Expr_assign (CVar twarray, CPlus [CVar twarray;
cannam@167 140 byvl (Integer nt)]);
cannam@167 141 make_volatile_stride (4*n) (CVar rs)
cannam@167 142 ],
cannam@167 143 Asch asch)])
cannam@167 144 in
cannam@167 145
cannam@167 146 let tree =
cannam@167 147 Fcn ("static void", name,
cannam@167 148 [Decl (C.realtypep, arp);
cannam@167 149 Decl (C.realtypep, aip);
cannam@167 150 Decl (C.realtypep, arm);
cannam@167 151 Decl (C.realtypep, aim);
cannam@167 152 Decl (C.constrealtypep, twarray);
cannam@167 153 Decl (C.stridetype, rs);
cannam@167 154 Decl ("INT", mb);
cannam@167 155 Decl ("INT", me);
cannam@167 156 Decl ("INT", ms)],
cannam@167 157 finalize_fcn body)
cannam@167 158 in
cannam@167 159 let twinstr =
cannam@167 160 Printf.sprintf "static const tw_instr twinstr[] = %s;\n\n"
cannam@167 161 (twinstr_to_string "VL" (twdesc n))
cannam@167 162 and desc =
cannam@167 163 Printf.sprintf
cannam@167 164 "static const hc2c_desc desc = {%d, \"%s\", twinstr, &GENUS, %s};\n\n"
cannam@167 165 n name (flops_of tree)
cannam@167 166 and register = "X(khc2c_register)"
cannam@167 167
cannam@167 168 in
cannam@167 169 let init =
cannam@167 170 "\n" ^
cannam@167 171 twinstr ^
cannam@167 172 desc ^
cannam@167 173 (declare_register_fcn name) ^
cannam@167 174 (Printf.sprintf "{\n%s(p, %s, &desc, HC2C_VIA_RDFT);\n}" register name)
cannam@167 175 in
cannam@167 176
cannam@167 177 (unparse tree) ^ "\n" ^ init
cannam@167 178
cannam@167 179
cannam@167 180 let main () =
cannam@167 181 begin
cannam@167 182 parse (speclist @ Twiddle.speclist) usage;
cannam@167 183 print_string (generate (check_size ()));
cannam@167 184 end
cannam@167 185
cannam@167 186 let _ = main()