cannam@95: (*
cannam@95:  * Copyright (c) 1997-1999 Massachusetts Institute of Technology
cannam@95:  * Copyright (c) 2003, 2007-11 Matteo Frigo
cannam@95:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
cannam@95:  *
cannam@95:  * This program is free software; you can redistribute it and/or modify
cannam@95:  * it under the terms of the GNU General Public License as published by
cannam@95:  * the Free Software Foundation; either version 2 of the License, or
cannam@95:  * (at your option) any later version.
cannam@95:  *
cannam@95:  * This program is distributed in the hope that it will be useful,
cannam@95:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@95:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@95:  * GNU General Public License for more details.
cannam@95:  *
cannam@95:  * You should have received a copy of the GNU General Public License
cannam@95:  * along with this program; if not, write to the Free Software
cannam@95:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@95:  *
cannam@95:  *)
cannam@95: 
cannam@95: open Util
cannam@95: open Genutil
cannam@95: open C
cannam@95: 
cannam@95: 
cannam@95: type ditdif = DIT | DIF
cannam@95: let ditdif = ref DIT
cannam@95: let usage = "Usage: " ^ Sys.argv.(0) ^ " -n <number> [ -dit | -dif ]"
cannam@95: 
cannam@95: let urs = ref Stride_variable
cannam@95: let ums = ref Stride_variable
cannam@95: 
cannam@95: let speclist = [
cannam@95:   "-dit",
cannam@95:   Arg.Unit(fun () -> ditdif := DIT),
cannam@95:   " generate a DIT codelet";
cannam@95: 
cannam@95:   "-dif",
cannam@95:   Arg.Unit(fun () -> ditdif := DIF),
cannam@95:   " generate a DIF codelet";
cannam@95: 
cannam@95:   "-with-rs",
cannam@95:   Arg.String(fun x -> urs := arg_to_stride x),
cannam@95:   " specialize for given R-stride";
cannam@95: 
cannam@95:   "-with-ms",
cannam@95:   Arg.String(fun x -> ums := arg_to_stride x),
cannam@95:   " specialize for given ms"
cannam@95: ]
cannam@95: 
cannam@95: let byi = Complex.times Complex.i
cannam@95: let byui = Complex.times (Complex.uminus Complex.i)
cannam@95: 
cannam@95: let shuffle_eo fe fo i = if i mod 2 == 0 then fe (i/2) else fo ((i-1)/2)
cannam@95: 
cannam@95: let generate n =
cannam@95:   let rs = "rs"
cannam@95:   and twarray = "W"
cannam@95:   and m = "m" and mb = "mb" and me = "me" and ms = "ms"
cannam@95: 
cannam@95:   (* the array names are from the point of view of the complex array
cannam@95:      (output in R2C, input in C2R) *)
cannam@95:   and arp = "Rp" (* real, positive *)
cannam@95:   and aip = "Ip" (* imag, positive *)
cannam@95:   and arm = "Rm" (* real, negative *)
cannam@95:   and aim = "Im" (* imag, negative *)
cannam@95: 
cannam@95:   in
cannam@95: 
cannam@95:   let sign = !Genutil.sign 
cannam@95:   and name = !Magic.codelet_name 
cannam@95:   and byvl x = choose_simd x (ctimes (CVar "VL", x)) in
cannam@95: 
cannam@95:   let (bytwiddle, num_twiddles, twdesc) = Twiddle.twiddle_policy 1 false in
cannam@95:   let nt = num_twiddles n in
cannam@95: 
cannam@95:   let byw = bytwiddle n sign (twiddle_array nt twarray) in
cannam@95: 
cannam@95:   let vrs = either_stride (!urs) (C.SVar rs) in
cannam@95: 
cannam@95:   (* assume a single location.  No point in doing alias analysis *)
cannam@95:   let the_location = (Unique.make (), Unique.make ()) in
cannam@95:   let locations _ = the_location in
cannam@95: 
cannam@95:   let rlocp = (locative_array_c n 
cannam@95: 		 (C.array_subscript arp vrs)
cannam@95: 		 (C.array_subscript aip vrs)
cannam@95: 		 locations "BUG")
cannam@95:   and rlocm = (locative_array_c n 
cannam@95: 		 (C.array_subscript arm vrs)
cannam@95: 		 (C.array_subscript aim vrs)
cannam@95: 		 locations "BUG")
cannam@95:   and clocp = (locative_array_c n 
cannam@95: 		 (C.array_subscript arp vrs)
cannam@95: 		 (C.array_subscript aip vrs)
cannam@95: 		 locations "BUG")
cannam@95:   and clocm = (locative_array_c n 
cannam@95: 		 (C.array_subscript arm vrs)
cannam@95: 		 (C.array_subscript aim vrs)
cannam@95: 		 locations "BUG")
cannam@95:   in
cannam@95:   let rloc i = if i mod 2 == 0 then rlocp (i/2) else rlocm ((i-1)/2)
cannam@95:   and cloc i = if i < n - i then clocp i else clocm (n-1-i)
cannam@95:   and sym n f i = if (i < n - i) then f i else Complex.conj (f i)
cannam@95:   and sym1 f i = 
cannam@95:     if i mod 2 == 0 then
cannam@95:       Complex.plus [f i; Complex.conj (f (i+1))]
cannam@95:     else
cannam@95:       Complex.times (Complex.uminus Complex.i)
cannam@95: 	(Complex.plus [f (i-1); Complex.uminus (Complex.conj (f i))])
cannam@95:   and sym1i f i = 
cannam@95:     if i mod 2 == 0 then
cannam@95:       Complex.plus [f i; Complex.times Complex.i (f (i+1))]
cannam@95:     else
cannam@95:       Complex.conj
cannam@95: 	(Complex.plus [f (i-1); 
cannam@95: 		       Complex.times (Complex.uminus Complex.i) (f i)])
cannam@95:   in
cannam@95: 
cannam@95:   let asch = 
cannam@95:     match !ditdif with
cannam@95:     | DIT -> 
cannam@95: 	let output = 
cannam@95: 	  (Complex.times Complex.half) @@
cannam@95: 	    (Fft.dft sign n (byw (sym1 (load_array_c n rloc)))) in
cannam@95: 	let odag = store_array_c n cloc (sym n output) in
cannam@95: 	  standard_optimizer odag 
cannam@95: 
cannam@95:     | DIF -> 
cannam@95: 	let output = 
cannam@95: 	  byw (Fft.dft sign n (sym n (load_array_c n cloc)))
cannam@95: 	in
cannam@95: 	let odag = store_array_c n rloc (sym1i output) in
cannam@95: 	  standard_optimizer odag 
cannam@95:   in
cannam@95: 
cannam@95:   let vms = CVar "ms" 
cannam@95:   and varp = CVar arp
cannam@95:   and vaip = CVar aip
cannam@95:   and varm = CVar arm
cannam@95:   and vaim = CVar aim
cannam@95:   and vm = CVar m and vmb = CVar mb and vme = CVar me 
cannam@95:   in
cannam@95:   let body = Block (
cannam@95:     [Decl ("INT", m)],
cannam@95:     [For (list_to_comma
cannam@95: 	    [Expr_assign (vm, vmb);
cannam@95: 	     Expr_assign (CVar twarray, 
cannam@95: 			  CPlus [CVar twarray; 
cannam@95: 				 ctimes (CPlus [vmb; CUminus (Integer 1)],
cannam@95: 					 Integer nt)])],
cannam@95: 	  Binop (" < ", vm, vme),
cannam@95: 	  list_to_comma 
cannam@95: 	    [Expr_assign (vm, CPlus [vm; byvl (Integer 1)]);
cannam@95: 	     Expr_assign (varp, CPlus [varp; byvl vms]);
cannam@95: 	     Expr_assign (vaip, CPlus [vaip; byvl vms]);
cannam@95: 	     Expr_assign (varm, CPlus [varm; CUminus (byvl vms)]);
cannam@95: 	     Expr_assign (vaim, CPlus [vaim; CUminus (byvl vms)]);
cannam@95: 	     Expr_assign (CVar twarray, CPlus [CVar twarray; 
cannam@95: 					       byvl (Integer nt)]);
cannam@95: 	     make_volatile_stride (4*n) (CVar rs)
cannam@95: 	   ],
cannam@95: 	  Asch asch)]
cannam@95:     )
cannam@95:   in
cannam@95: 
cannam@95:   let tree = 
cannam@95:     Fcn ("static void", name,
cannam@95: 	 [Decl (C.realtypep, arp);
cannam@95: 	  Decl (C.realtypep, aip);
cannam@95: 	  Decl (C.realtypep, arm);
cannam@95: 	  Decl (C.realtypep, aim);
cannam@95: 	  Decl (C.constrealtypep, twarray);
cannam@95: 	  Decl (C.stridetype, rs);
cannam@95: 	  Decl ("INT", mb);
cannam@95: 	  Decl ("INT", me);
cannam@95: 	  Decl ("INT", ms)],
cannam@95:          finalize_fcn body)
cannam@95:   in
cannam@95:   let twinstr = 
cannam@95:     Printf.sprintf "static const tw_instr twinstr[] = %s;\n\n" 
cannam@95:       (twinstr_to_string "VL" (twdesc n))
cannam@95:   and desc = 
cannam@95:     Printf.sprintf
cannam@95:       "static const hc2c_desc desc = {%d, \"%s\", twinstr, &GENUS, %s};\n\n"
cannam@95:       n name (flops_of tree)
cannam@95:   and register = "X(khc2c_register)"
cannam@95: 
cannam@95:   in
cannam@95:   let init =
cannam@95:     "\n" ^
cannam@95:     twinstr ^ 
cannam@95:     desc ^
cannam@95:     (declare_register_fcn name) ^
cannam@95:     (Printf.sprintf "{\n%s(p, %s, &desc, HC2C_VIA_DFT);\n}" register name)
cannam@95:   in
cannam@95: 
cannam@95:   (unparse tree) ^ "\n" ^ init
cannam@95: 
cannam@95: 
cannam@95: let main () =
cannam@95:   begin 
cannam@95:     parse (speclist @ Twiddle.speclist) usage;
cannam@95:     print_string (generate (check_size ()));
cannam@95:   end
cannam@95: 
cannam@95: let _ = main()