annotate src/fftw-3.3.8/genfft/gen_mdct.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 (* generation of trigonometric transforms *)
Chris@82 23
Chris@82 24 open Util
Chris@82 25 open Genutil
Chris@82 26 open C
Chris@82 27
Chris@82 28
Chris@82 29 let usage = "Usage: " ^ Sys.argv.(0) ^ " -n <number>"
Chris@82 30
Chris@82 31 let uistride = ref Stride_variable
Chris@82 32 let uostride = ref Stride_variable
Chris@82 33 let uivstride = ref Stride_variable
Chris@82 34 let uovstride = ref Stride_variable
Chris@82 35 let normalization = ref 1
Chris@82 36
Chris@82 37 type mode =
Chris@82 38 | MDCT
Chris@82 39 | MDCT_MP3
Chris@82 40 | MDCT_VORBIS
Chris@82 41 | MDCT_WINDOW
Chris@82 42 | MDCT_WINDOW_SYM
Chris@82 43 | IMDCT
Chris@82 44 | IMDCT_MP3
Chris@82 45 | IMDCT_VORBIS
Chris@82 46 | IMDCT_WINDOW
Chris@82 47 | IMDCT_WINDOW_SYM
Chris@82 48 | NONE
Chris@82 49
Chris@82 50 let mode = ref NONE
Chris@82 51
Chris@82 52 let speclist = [
Chris@82 53 "-with-istride",
Chris@82 54 Arg.String(fun x -> uistride := arg_to_stride x),
Chris@82 55 " specialize for given input stride";
Chris@82 56
Chris@82 57 "-with-ostride",
Chris@82 58 Arg.String(fun x -> uostride := arg_to_stride x),
Chris@82 59 " specialize for given output stride";
Chris@82 60
Chris@82 61 "-with-ivstride",
Chris@82 62 Arg.String(fun x -> uivstride := arg_to_stride x),
Chris@82 63 " specialize for given input vector stride";
Chris@82 64
Chris@82 65 "-with-ovstride",
Chris@82 66 Arg.String(fun x -> uovstride := arg_to_stride x),
Chris@82 67 " specialize for given output vector stride";
Chris@82 68
Chris@82 69 "-normalization",
Chris@82 70 Arg.String(fun x -> normalization := int_of_string x),
Chris@82 71 " normalization integer to divide by";
Chris@82 72
Chris@82 73 "-mdct",
Chris@82 74 Arg.Unit(fun () -> mode := MDCT),
Chris@82 75 " generate an MDCT codelet";
Chris@82 76
Chris@82 77 "-mdct-mp3",
Chris@82 78 Arg.Unit(fun () -> mode := MDCT_MP3),
Chris@82 79 " generate an MDCT codelet with MP3 windowing";
Chris@82 80
Chris@82 81 "-mdct-window",
Chris@82 82 Arg.Unit(fun () -> mode := MDCT_WINDOW),
Chris@82 83 " generate an MDCT codelet with window array";
Chris@82 84
Chris@82 85 "-mdct-window-sym",
Chris@82 86 Arg.Unit(fun () -> mode := MDCT_WINDOW_SYM),
Chris@82 87 " generate an MDCT codelet with symmetric window array";
Chris@82 88
Chris@82 89 "-imdct",
Chris@82 90 Arg.Unit(fun () -> mode := IMDCT),
Chris@82 91 " generate an IMDCT codelet";
Chris@82 92
Chris@82 93 "-imdct-mp3",
Chris@82 94 Arg.Unit(fun () -> mode := IMDCT_MP3),
Chris@82 95 " generate an IMDCT codelet with MP3 windowing";
Chris@82 96
Chris@82 97 "-imdct-window",
Chris@82 98 Arg.Unit(fun () -> mode := IMDCT_WINDOW),
Chris@82 99 " generate an IMDCT codelet with window array";
Chris@82 100
Chris@82 101 "-imdct-window-sym",
Chris@82 102 Arg.Unit(fun () -> mode := IMDCT_WINDOW_SYM),
Chris@82 103 " generate an IMDCT codelet with symmetric window array";
Chris@82 104 ]
Chris@82 105
Chris@82 106 let unity_window n i = Complex.one
Chris@82 107
Chris@82 108 (* MP3 window(k) = sin(pi/(2n) * (k + 1/2)) *)
Chris@82 109 let mp3_window n k =
Chris@82 110 Complex.imag (Complex.exp (8 * n) (2*k + 1))
Chris@82 111
Chris@82 112 (* Vorbis window(k) = sin(pi/2 * (mp3_window(k))^2)
Chris@82 113 ... this is transcendental, though, so we can't do it with our
Chris@82 114 current Complex.exp function *)
Chris@82 115
Chris@82 116 let window_array n w =
Chris@82 117 array n (fun i ->
Chris@82 118 let stride = C.SInteger 1
Chris@82 119 and klass = Unique.make () in
Chris@82 120 let refr = C.array_subscript w stride i in
Chris@82 121 let kr = Variable.make_constant klass refr in
Chris@82 122 load_r (kr, kr))
Chris@82 123
Chris@82 124 let load_window w n i = w i
Chris@82 125 let load_window_sym w n i = w (if (i < n) then i else (2*n - 1 - i))
Chris@82 126
Chris@82 127 (* fixme: use same locations for input and output so that it works in-place? *)
Chris@82 128
Chris@82 129 (* Note: only correct for even n! *)
Chris@82 130 let load_array_mdct window n rarr iarr locations =
Chris@82 131 let twon = 2 * n in
Chris@82 132 let arr = load_array_c twon
Chris@82 133 (locative_array_c twon rarr iarr locations "BUG") in
Chris@82 134 let arrw = fun i -> Complex.times (window n i) (arr i) in
Chris@82 135 array n
Chris@82 136 ((Complex.times Complex.half) @@
Chris@82 137 (fun i ->
Chris@82 138 if (i < n/2) then
Chris@82 139 Complex.uminus (Complex.plus [arrw (i + n + n/2);
Chris@82 140 arrw (n + n/2 - 1 - i)])
Chris@82 141 else
Chris@82 142 Complex.plus [arrw (i - n/2);
Chris@82 143 Complex.uminus (arrw (n + n/2 - 1 - i))]))
Chris@82 144
Chris@82 145 let store_array_mdct window n rarr iarr locations arr =
Chris@82 146 store_array_r n (locative_array_c n rarr iarr locations "BUG") arr
Chris@82 147
Chris@82 148 let load_array_imdct window n rarr iarr locations =
Chris@82 149 load_array_c n (locative_array_c n rarr iarr locations "BUG")
Chris@82 150
Chris@82 151 let store_array_imdct window n rarr iarr locations arr =
Chris@82 152 let n2 = n/2 in
Chris@82 153 let threen2 = 3*n2 in
Chris@82 154 let arr2 = fun i ->
Chris@82 155 if (i < n2) then
Chris@82 156 arr (i + n2)
Chris@82 157 else if (i < threen2) then
Chris@82 158 Complex.uminus (arr (threen2 - 1 - i))
Chris@82 159 else
Chris@82 160 Complex.uminus (arr (i - threen2))
Chris@82 161 in
Chris@82 162 let arr2w = fun i -> Complex.times (window n i) (arr2 i) in
Chris@82 163 let twon = 2 * n in
Chris@82 164 store_array_r twon (locative_array_c twon rarr iarr locations "BUG") arr2w
Chris@82 165
Chris@82 166 let window_param = function
Chris@82 167 MDCT_WINDOW -> true
Chris@82 168 | MDCT_WINDOW_SYM -> true
Chris@82 169 | IMDCT_WINDOW -> true
Chris@82 170 | IMDCT_WINDOW_SYM -> true
Chris@82 171 | _ -> false
Chris@82 172
Chris@82 173 let generate n mode =
Chris@82 174 let iarray = "I"
Chris@82 175 and oarray = "O"
Chris@82 176 and istride = "istride"
Chris@82 177 and ostride = "ostride"
Chris@82 178 and window = "W"
Chris@82 179 and name = !Magic.codelet_name in
Chris@82 180
Chris@82 181 let vistride = either_stride (!uistride) (C.SVar istride)
Chris@82 182 and vostride = either_stride (!uostride) (C.SVar ostride)
Chris@82 183 in
Chris@82 184
Chris@82 185 let sivs = stride_to_string "ovs" !uovstride in
Chris@82 186 let sovs = stride_to_string "ivs" !uivstride in
Chris@82 187
Chris@82 188 let (transform, load_input, store_output) = match mode with
Chris@82 189 | MDCT -> Trig.dctIV, load_array_mdct unity_window,
Chris@82 190 store_array_mdct unity_window
Chris@82 191 | MDCT_MP3 -> Trig.dctIV, load_array_mdct mp3_window,
Chris@82 192 store_array_mdct unity_window
Chris@82 193 | MDCT_WINDOW -> Trig.dctIV, load_array_mdct
Chris@82 194 (load_window (window_array (2 * n) window)),
Chris@82 195 store_array_mdct unity_window
Chris@82 196 | MDCT_WINDOW_SYM -> Trig.dctIV, load_array_mdct
Chris@82 197 (load_window_sym (window_array n window)),
Chris@82 198 store_array_mdct unity_window
Chris@82 199 | IMDCT -> Trig.dctIV, load_array_imdct unity_window,
Chris@82 200 store_array_imdct unity_window
Chris@82 201 | IMDCT_MP3 -> Trig.dctIV, load_array_imdct unity_window,
Chris@82 202 store_array_imdct mp3_window
Chris@82 203 | IMDCT_WINDOW -> Trig.dctIV, load_array_imdct unity_window,
Chris@82 204 store_array_imdct (load_window (window_array (2 * n) window))
Chris@82 205 | IMDCT_WINDOW_SYM -> Trig.dctIV, load_array_imdct unity_window,
Chris@82 206 store_array_imdct (load_window_sym (window_array n window))
Chris@82 207 | _ -> failwith "must specify transform kind"
Chris@82 208 in
Chris@82 209
Chris@82 210 let locations = unique_array_c (2*n) in
Chris@82 211 let input =
Chris@82 212 load_input n
Chris@82 213 (C.array_subscript iarray vistride)
Chris@82 214 (C.array_subscript "BUG" vistride)
Chris@82 215 locations
Chris@82 216 in
Chris@82 217 let output = (Complex.times (Complex.inverse_int !normalization))
Chris@82 218 @@ (transform n input) in
Chris@82 219 let odag =
Chris@82 220 store_output n
Chris@82 221 (C.array_subscript oarray vostride)
Chris@82 222 (C.array_subscript "BUG" vostride)
Chris@82 223 locations
Chris@82 224 output
Chris@82 225 in
Chris@82 226 let annot = standard_optimizer odag in
Chris@82 227
Chris@82 228 let tree =
Chris@82 229 Fcn ("void", name,
Chris@82 230 ([Decl (C.constrealtypep, iarray);
Chris@82 231 Decl (C.realtypep, oarray)]
Chris@82 232 @ (if stride_fixed !uistride then []
Chris@82 233 else [Decl (C.stridetype, istride)])
Chris@82 234 @ (if stride_fixed !uostride then []
Chris@82 235 else [Decl (C.stridetype, ostride)])
Chris@82 236 @ (choose_simd []
Chris@82 237 (if stride_fixed !uivstride then [] else
Chris@82 238 [Decl ("int", sivs)]))
Chris@82 239 @ (choose_simd []
Chris@82 240 (if stride_fixed !uovstride then [] else
Chris@82 241 [Decl ("int", sovs)]))
Chris@82 242 @ (if (not (window_param mode)) then []
Chris@82 243 else [Decl (C.constrealtypep, window)])
Chris@82 244 ),
Chris@82 245 finalize_fcn (Asch annot))
Chris@82 246
Chris@82 247 in
Chris@82 248 (unparse tree) ^ "\n"
Chris@82 249
Chris@82 250
Chris@82 251 let main () =
Chris@82 252 begin
Chris@82 253 parse speclist usage;
Chris@82 254 print_string (generate (check_size ()) !mode);
Chris@82 255 end
Chris@82 256
Chris@82 257 let _ = main()