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