Chris@10
|
1 (*
|
Chris@10
|
2 * Copyright (c) 1997-1999 Massachusetts Institute of Technology
|
Chris@10
|
3 * Copyright (c) 2003, 2007-11 Matteo Frigo
|
Chris@10
|
4 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
|
Chris@10
|
5 *
|
Chris@10
|
6 * This program is free software; you can redistribute it and/or modify
|
Chris@10
|
7 * it under the terms of the GNU General Public License as published by
|
Chris@10
|
8 * the Free Software Foundation; either version 2 of the License, or
|
Chris@10
|
9 * (at your option) any later version.
|
Chris@10
|
10 *
|
Chris@10
|
11 * This program is distributed in the hope that it will be useful,
|
Chris@10
|
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@10
|
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@10
|
14 * GNU General Public License for more details.
|
Chris@10
|
15 *
|
Chris@10
|
16 * You should have received a copy of the GNU General Public License
|
Chris@10
|
17 * along with this program; if not, write to the Free Software
|
Chris@10
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@10
|
19 *
|
Chris@10
|
20 *)
|
Chris@10
|
21
|
Chris@10
|
22 open Util
|
Chris@10
|
23 open Genutil
|
Chris@10
|
24 open C
|
Chris@10
|
25
|
Chris@10
|
26 type ditdif = DIT | DIF
|
Chris@10
|
27 let ditdif = ref DIT
|
Chris@10
|
28
|
Chris@10
|
29 let usage = "Usage: " ^ Sys.argv.(0) ^ " -n <number> [ -dit | -dif ]"
|
Chris@10
|
30
|
Chris@10
|
31 let reload_twiddle = ref false
|
Chris@10
|
32
|
Chris@10
|
33 let urs = ref Stride_variable
|
Chris@10
|
34 let uvs = ref Stride_variable
|
Chris@10
|
35 let ums = ref Stride_variable
|
Chris@10
|
36
|
Chris@10
|
37 let speclist = [
|
Chris@10
|
38 "-dit",
|
Chris@10
|
39 Arg.Unit(fun () -> ditdif := DIT),
|
Chris@10
|
40 " generate a DIT codelet";
|
Chris@10
|
41
|
Chris@10
|
42 "-dif",
|
Chris@10
|
43 Arg.Unit(fun () -> ditdif := DIF),
|
Chris@10
|
44 " generate a DIF codelet";
|
Chris@10
|
45
|
Chris@10
|
46 "-reload-twiddle",
|
Chris@10
|
47 Arg.Unit(fun () -> reload_twiddle := true),
|
Chris@10
|
48 " do not collect common twiddle factors";
|
Chris@10
|
49
|
Chris@10
|
50 "-with-rs",
|
Chris@10
|
51 Arg.String(fun x -> urs := arg_to_stride x),
|
Chris@10
|
52 " specialize for given input stride";
|
Chris@10
|
53
|
Chris@10
|
54 "-with-vs",
|
Chris@10
|
55 Arg.String(fun x -> uvs := arg_to_stride x),
|
Chris@10
|
56 " specialize for given vector stride";
|
Chris@10
|
57
|
Chris@10
|
58 "-with-ms",
|
Chris@10
|
59 Arg.String(fun x -> ums := arg_to_stride x),
|
Chris@10
|
60 " specialize for given ms"
|
Chris@10
|
61 ]
|
Chris@10
|
62
|
Chris@10
|
63 let generate n =
|
Chris@10
|
64 let rioarray = "rio"
|
Chris@10
|
65 and iioarray = "iio"
|
Chris@10
|
66 and rs = "rs" and vs = "vs"
|
Chris@10
|
67 and twarray = "W"
|
Chris@10
|
68 and m = "m" and mb = "mb" and me = "me" and ms = "ms" in
|
Chris@10
|
69
|
Chris@10
|
70 let sign = !Genutil.sign
|
Chris@10
|
71 and name = !Magic.codelet_name in
|
Chris@10
|
72
|
Chris@10
|
73 let (bytwiddle, num_twiddles, twdesc) = Twiddle.twiddle_policy 0 false in
|
Chris@10
|
74 let nt = num_twiddles n in
|
Chris@10
|
75
|
Chris@10
|
76 let svs = either_stride (!uvs) (C.SVar vs)
|
Chris@10
|
77 and srs = either_stride (!urs) (C.SVar rs) in
|
Chris@10
|
78
|
Chris@10
|
79 let byw =
|
Chris@10
|
80 if !reload_twiddle then
|
Chris@10
|
81 array n (fun v -> bytwiddle n sign (twiddle_array nt twarray))
|
Chris@10
|
82 else
|
Chris@10
|
83 let a = bytwiddle n sign (twiddle_array nt twarray)
|
Chris@10
|
84 in fun v -> a
|
Chris@10
|
85 in
|
Chris@10
|
86
|
Chris@10
|
87 let locations = unique_v_array_c n n in
|
Chris@10
|
88
|
Chris@10
|
89 let ioi =
|
Chris@10
|
90 locative_v_array_c n n
|
Chris@10
|
91 (C.varray_subscript rioarray svs srs)
|
Chris@10
|
92 (C.varray_subscript iioarray svs srs)
|
Chris@10
|
93 locations "BUG"
|
Chris@10
|
94 and ioo =
|
Chris@10
|
95 locative_v_array_c n n
|
Chris@10
|
96 (C.varray_subscript rioarray svs srs)
|
Chris@10
|
97 (C.varray_subscript iioarray svs srs)
|
Chris@10
|
98 locations "BUG"
|
Chris@10
|
99 in
|
Chris@10
|
100
|
Chris@10
|
101 let lioi = load_v_array_c n n ioi in
|
Chris@10
|
102 let output =
|
Chris@10
|
103 match !ditdif with
|
Chris@10
|
104 | DIT -> array n (fun v -> Fft.dft sign n (byw v (lioi v)))
|
Chris@10
|
105 | DIF -> array n (fun v -> byw v (Fft.dft sign n (lioi v)))
|
Chris@10
|
106 in
|
Chris@10
|
107
|
Chris@10
|
108 let odag = store_v_array_c n n ioo (transpose output) in
|
Chris@10
|
109 let annot = standard_optimizer odag in
|
Chris@10
|
110
|
Chris@10
|
111 let vm = CVar m and vmb = CVar mb and vme = CVar me in
|
Chris@10
|
112
|
Chris@10
|
113 let body = Block (
|
Chris@10
|
114 [Decl ("INT", m)],
|
Chris@10
|
115 [For (list_to_comma
|
Chris@10
|
116 [Expr_assign (vm, vmb);
|
Chris@10
|
117 Expr_assign (CVar twarray,
|
Chris@10
|
118 CPlus [CVar twarray;
|
Chris@10
|
119 ctimes (vmb, Integer nt)])],
|
Chris@10
|
120 Binop (" < ", vm, vme),
|
Chris@10
|
121 list_to_comma
|
Chris@10
|
122 [Expr_assign (vm, CPlus [vm; Integer 1]);
|
Chris@10
|
123 Expr_assign (CVar rioarray, CPlus [CVar rioarray; CVar ms]);
|
Chris@10
|
124 Expr_assign (CVar iioarray, CPlus [CVar iioarray; CVar ms]);
|
Chris@10
|
125 Expr_assign (CVar twarray, CPlus [CVar twarray; Integer nt]);
|
Chris@10
|
126 make_volatile_stride (2*n) (CVar rs);
|
Chris@10
|
127 make_volatile_stride (2*0) (CVar vs)
|
Chris@10
|
128 ],
|
Chris@10
|
129 Asch annot)]) in
|
Chris@10
|
130
|
Chris@10
|
131 let tree =
|
Chris@10
|
132 Fcn (("static void"), name,
|
Chris@10
|
133 [Decl (C.realtypep, rioarray);
|
Chris@10
|
134 Decl (C.realtypep, iioarray);
|
Chris@10
|
135 Decl (C.constrealtypep, twarray);
|
Chris@10
|
136 Decl (C.stridetype, rs);
|
Chris@10
|
137 Decl (C.stridetype, vs);
|
Chris@10
|
138 Decl ("INT", mb);
|
Chris@10
|
139 Decl ("INT", me);
|
Chris@10
|
140 Decl ("INT", ms)],
|
Chris@10
|
141 finalize_fcn body)
|
Chris@10
|
142 in
|
Chris@10
|
143 let twinstr =
|
Chris@10
|
144 Printf.sprintf "static const tw_instr twinstr[] = %s;\n\n"
|
Chris@10
|
145 (Twiddle.twinstr_to_c_string (twdesc n))
|
Chris@10
|
146
|
Chris@10
|
147 and desc =
|
Chris@10
|
148 Printf.sprintf
|
Chris@10
|
149 "static const ct_desc desc = {%d, \"%s\", twinstr, &GENUS, %s, %s, %s, %s};\n\n"
|
Chris@10
|
150 n name (flops_of tree)
|
Chris@10
|
151 (stride_to_solverparm !urs) (stride_to_solverparm !uvs)
|
Chris@10
|
152 (stride_to_solverparm !ums)
|
Chris@10
|
153
|
Chris@10
|
154 and register =
|
Chris@10
|
155 match !ditdif with
|
Chris@10
|
156 | DIT -> "X(kdft_ditsq_register)"
|
Chris@10
|
157 | DIF -> "X(kdft_difsq_register)"
|
Chris@10
|
158 in
|
Chris@10
|
159 let init =
|
Chris@10
|
160 "\n" ^
|
Chris@10
|
161 twinstr ^
|
Chris@10
|
162 desc ^
|
Chris@10
|
163 (declare_register_fcn name) ^
|
Chris@10
|
164 (Printf.sprintf "{\n%s(p, %s, &desc);\n}" register name)
|
Chris@10
|
165 in
|
Chris@10
|
166
|
Chris@10
|
167 (unparse tree) ^ "\n" ^ init
|
Chris@10
|
168
|
Chris@10
|
169
|
Chris@10
|
170 let main () =
|
Chris@10
|
171 begin
|
Chris@10
|
172 parse (speclist @ Twiddle.speclist) usage;
|
Chris@10
|
173 print_string (generate (check_size ()));
|
Chris@10
|
174 end
|
Chris@10
|
175
|
Chris@10
|
176 let _ = main()
|