Chris@42: (*
Chris@42:  * Copyright (c) 1997-1999 Massachusetts Institute of Technology
Chris@42:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42:  *
Chris@42:  * This program is free software; you can redistribute it and/or modify
Chris@42:  * it under the terms of the GNU General Public License as published by
Chris@42:  * the Free Software Foundation; either version 2 of the License, or
Chris@42:  * (at your option) any later version.
Chris@42:  *
Chris@42:  * This program is distributed in the hope that it will be useful,
Chris@42:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@42:  * GNU General Public License for more details.
Chris@42:  *
Chris@42:  * You should have received a copy of the GNU General Public License
Chris@42:  * along with this program; if not, write to the Free Software
Chris@42:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@42:  *
Chris@42:  *)
Chris@42: 
Chris@42: open Util
Chris@42: open Genutil
Chris@42: open C
Chris@42: 
Chris@42: 
Chris@42: type ditdif = DIT | DIF
Chris@42: let ditdif = ref DIT
Chris@42: let usage = "Usage: " ^ Sys.argv.(0) ^ " -n <number> [ -dit | -dif ]"
Chris@42: 
Chris@42: let urs = ref Stride_variable
Chris@42: let ums = ref Stride_variable
Chris@42: 
Chris@42: let speclist = [
Chris@42:   "-dit",
Chris@42:   Arg.Unit(fun () -> ditdif := DIT),
Chris@42:   " generate a DIT codelet";
Chris@42: 
Chris@42:   "-dif",
Chris@42:   Arg.Unit(fun () -> ditdif := DIF),
Chris@42:   " generate a DIF codelet";
Chris@42: 
Chris@42:   "-with-rs",
Chris@42:   Arg.String(fun x -> urs := arg_to_stride x),
Chris@42:   " specialize for given R-stride";
Chris@42: 
Chris@42:   "-with-ms",
Chris@42:   Arg.String(fun x -> ums := arg_to_stride x),
Chris@42:   " specialize for given ms"
Chris@42: ]
Chris@42: 
Chris@42: let byi = Complex.times Complex.i
Chris@42: let byui = Complex.times (Complex.uminus Complex.i)
Chris@42: 
Chris@42: let shuffle_eo fe fo i = if i mod 2 == 0 then fe (i/2) else fo ((i-1)/2)
Chris@42: 
Chris@42: let generate n =
Chris@42:   let rs = "rs"
Chris@42:   and twarray = "W"
Chris@42:   and m = "m" and mb = "mb" and me = "me" and ms = "ms"
Chris@42: 
Chris@42:   (* the array names are from the point of view of the complex array
Chris@42:      (output in R2C, input in C2R) *)
Chris@42:   and arp = "Rp" (* real, positive *)
Chris@42:   and aip = "Ip" (* imag, positive *)
Chris@42:   and arm = "Rm" (* real, negative *)
Chris@42:   and aim = "Im" (* imag, negative *)
Chris@42: 
Chris@42:   in
Chris@42: 
Chris@42:   let sign = !Genutil.sign 
Chris@42:   and name = !Magic.codelet_name 
Chris@42:   and byvl x = choose_simd x (ctimes (CVar "VL", x)) 
Chris@42:   and bytwvl x = choose_simd x (ctimes (CVar "TWVL", x)) 
Chris@42:   and bytwvl_vl x = choose_simd x (ctimes (CVar "(TWVL/VL)", x)) in
Chris@42: 
Chris@42:   let (bytwiddle, num_twiddles, twdesc) = Twiddle.twiddle_policy 1 true in
Chris@42:   let nt = num_twiddles n in
Chris@42: 
Chris@42:   let byw = bytwiddle n sign (twiddle_array nt twarray) in
Chris@42: 
Chris@42:   let vrs = either_stride (!urs) (C.SVar rs) in
Chris@42:   let sms = stride_to_string "ms" !ums in
Chris@42:   let msms = "-" ^ sms in
Chris@42: 
Chris@42:   (* assume a single location.  No point in doing alias analysis *)
Chris@42:   let the_location = (Unique.make (), Unique.make ()) in
Chris@42:   let locations _ = the_location in
Chris@42: 
Chris@42:   let rlocp = (locative_array_c n 
Chris@42: 		 (C.array_subscript arp vrs)
Chris@42: 		 (C.array_subscript aip vrs)
Chris@42: 		 locations sms)
Chris@42:   and rlocm = (locative_array_c n 
Chris@42: 		 (C.array_subscript arm vrs)
Chris@42: 		 (C.array_subscript aim vrs)
Chris@42: 		 locations msms)
Chris@42:   and clocp = (locative_array_c n 
Chris@42: 		 (C.array_subscript arp vrs)
Chris@42: 		 (C.array_subscript aip vrs)
Chris@42: 		 locations sms)
Chris@42:   and clocm = (locative_array_c n 
Chris@42: 		 (C.array_subscript arm vrs)
Chris@42: 		 (C.array_subscript aim vrs)
Chris@42: 		 locations msms)
Chris@42:   in
Chris@42:   let rloc i = if i mod 2 == 0 then rlocp (i/2) else rlocm ((i-1)/2)
Chris@42:   and cloc i = if i < n - i then clocp i else clocm (n-1-i)
Chris@42:   and sym n f i =
Chris@42:     if (i < n - i) then 
Chris@42:       f i
Chris@42:     else 
Chris@42:       Complex.times (Complex.nan Expr.CONJ) (f i)
Chris@42:   and sym1 f i = 
Chris@42:     if i mod 2 == 0 then
Chris@42:       Complex.plus [f i; 
Chris@42: 		    Complex.times (Complex.nan Expr.CONJ) (f (i+1))]
Chris@42:     else
Chris@42:       Complex.times (Complex.nan Expr.I)
Chris@42: 	(Complex.plus [Complex.uminus (f (i-1));
Chris@42: 		       Complex.times (Complex.nan Expr.CONJ) (f i)])
Chris@42:   and sym1i f i = 
Chris@42:     if i mod 2 == 0 then
Chris@42:       Complex.plus [f i; 
Chris@42: 		    Complex.times (Complex.nan Expr.I) (f (i+1))]
Chris@42:     else
Chris@42:       Complex.times (Complex.nan Expr.CONJ)
Chris@42: 	(Complex.plus [f (i-1); 
Chris@42: 		       Complex.uminus
Chris@42: 			 (Complex.times (Complex.nan Expr.I) (f i))])
Chris@42:   in
Chris@42: 
Chris@42:   let asch = 
Chris@42:     match !ditdif with
Chris@42:     | DIT -> 
Chris@42: 	let output = 
Chris@42: 	  (Complex.times Complex.half) @@
Chris@42: 	    (Trig.dft_via_rdft sign n (byw (sym1 (load_array_r n rloc)))) in
Chris@42: 	let odag = store_array_r n cloc (sym n output) in
Chris@42: 	  standard_optimizer odag 
Chris@42: 
Chris@42:     | DIF -> 
Chris@42: 	let output = 
Chris@42: 	  byw (Trig.dft_via_rdft sign n (sym n (load_array_r n cloc)))
Chris@42: 	in
Chris@42: 	let odag = store_array_r n rloc (sym1i output) in
Chris@42: 	  standard_optimizer odag 
Chris@42:   in
Chris@42: 
Chris@42:   let vms = CVar sms 
Chris@42:   and varp = CVar arp
Chris@42:   and vaip = CVar aip
Chris@42:   and varm = CVar arm
Chris@42:   and vaim = CVar aim
Chris@42:   and vm = CVar m and vmb = CVar mb and vme = CVar me 
Chris@42:   in
Chris@42:   let body = Block (
Chris@42:     [Decl ("INT", m)],
Chris@42:     [For (list_to_comma
Chris@42: 	    [Expr_assign (vm, vmb);
Chris@42: 	     Expr_assign (CVar twarray, 
Chris@42: 			  CPlus [CVar twarray; 
Chris@42: 				 ctimes (CPlus [vmb; CUminus (Integer 1)],
Chris@42: 					 bytwvl_vl (Integer nt))])],
Chris@42: 	  Binop (" < ", vm, vme),
Chris@42: 	  list_to_comma 
Chris@42: 	    [Expr_assign (vm, CPlus [vm; byvl (Integer 1)]);
Chris@42: 	     Expr_assign (varp, CPlus [varp; byvl vms]);
Chris@42: 	     Expr_assign (vaip, CPlus [vaip; byvl vms]);
Chris@42: 	     Expr_assign (varm, CPlus [varm; CUminus (byvl vms)]);
Chris@42: 	     Expr_assign (vaim, CPlus [vaim; CUminus (byvl vms)]);
Chris@42: 	     Expr_assign (CVar twarray, CPlus [CVar twarray; 
Chris@42: 					       bytwvl (Integer nt)]);
Chris@42: 	     make_volatile_stride (4*n) (CVar rs)
Chris@42: 	   ],
Chris@42: 	  Asch asch)]
Chris@42:     )
Chris@42:   in
Chris@42: 
Chris@42:   let tree = 
Chris@42:     Fcn ("static void", name,
Chris@42: 	 [Decl (C.realtypep, arp);
Chris@42: 	  Decl (C.realtypep, aip);
Chris@42: 	  Decl (C.realtypep, arm);
Chris@42: 	  Decl (C.realtypep, aim);
Chris@42: 	  Decl (C.constrealtypep, twarray);
Chris@42: 	  Decl (C.stridetype, rs);
Chris@42: 	  Decl ("INT", mb);
Chris@42: 	  Decl ("INT", me);
Chris@42: 	  Decl ("INT", ms)],
Chris@42:          finalize_fcn body)
Chris@42:   in
Chris@42:   let twinstr = 
Chris@42:     Printf.sprintf "static const tw_instr twinstr[] = %s;\n\n" 
Chris@42:       (twinstr_to_string "VL" (twdesc n))
Chris@42:   and desc = 
Chris@42:     Printf.sprintf
Chris@42:       "static const hc2c_desc desc = {%d, %s, twinstr, &GENUS, %s};\n\n"
Chris@42:       n (stringify name) (flops_of tree)
Chris@42:   and register = "X(khc2c_register)"
Chris@42: 
Chris@42:   in
Chris@42:   let init =
Chris@42:     "\n" ^
Chris@42:     twinstr ^ 
Chris@42:     desc ^
Chris@42:     (declare_register_fcn name) ^
Chris@42:     (Printf.sprintf "{\n%s(p, %s, &desc, HC2C_VIA_DFT);\n}" register name)
Chris@42:   in
Chris@42: 
Chris@42:   (unparse tree) ^ "\n" ^ init
Chris@42: 
Chris@42: 
Chris@42: let main () =
Chris@42:   begin 
Chris@42:     Simdmagic.simd_mode := true;
Chris@42:     parse (speclist @ Twiddle.speclist) usage;
Chris@42:     print_string (generate (check_size ()));
Chris@42:   end
Chris@42: 
Chris@42: let _ = main()