mirror of
https://github.com/tildearrow/furnace.git
synced 2024-11-24 05:25:12 +00:00
308 lines
11 KiB
OCaml
308 lines
11 KiB
OCaml
|
(*
|
||
|
* Copyright (c) 1997-1999 Massachusetts Institute of Technology
|
||
|
* Copyright (c) 2003, 2007-14 Matteo Frigo
|
||
|
* Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
||
|
*
|
||
|
* This program is free software; you can redistribute it and/or modify
|
||
|
* it under the terms of the GNU General Public License as published by
|
||
|
* the Free Software Foundation; either version 2 of the License, or
|
||
|
* (at your option) any later version.
|
||
|
*
|
||
|
* This program is distributed in the hope that it will be useful,
|
||
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
|
* GNU General Public License for more details.
|
||
|
*
|
||
|
* You should have received a copy of the GNU General Public License
|
||
|
* along with this program; if not, write to the Free Software
|
||
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||
|
*
|
||
|
*)
|
||
|
|
||
|
|
||
|
(* This is the part of the generator that actually computes the FFT
|
||
|
in symbolic form *)
|
||
|
|
||
|
open Complex
|
||
|
open Util
|
||
|
|
||
|
(* choose a suitable factor of n *)
|
||
|
let choose_factor n =
|
||
|
(* first choice: i such that gcd(i, n / i) = 1, i as big as possible *)
|
||
|
let choose1 n =
|
||
|
let rec loop i f =
|
||
|
if (i * i > n) then f
|
||
|
else if ((n mod i) == 0 && gcd i (n / i) == 1) then loop (i + 1) i
|
||
|
else loop (i + 1) f
|
||
|
in loop 1 1
|
||
|
|
||
|
(* second choice: the biggest factor i of n, where i < sqrt(n), if any *)
|
||
|
and choose2 n =
|
||
|
let rec loop i f =
|
||
|
if (i * i > n) then f
|
||
|
else if ((n mod i) == 0) then loop (i + 1) i
|
||
|
else loop (i + 1) f
|
||
|
in loop 1 1
|
||
|
|
||
|
in let i = choose1 n in
|
||
|
if (i > 1) then i
|
||
|
else choose2 n
|
||
|
|
||
|
let is_power_of_two n = (n > 0) && ((n - 1) land n == 0)
|
||
|
|
||
|
let rec dft_prime sign n input =
|
||
|
let sum filter i =
|
||
|
sigma 0 n (fun j ->
|
||
|
let coeff = filter (exp n (sign * i * j))
|
||
|
in coeff @* (input j)) in
|
||
|
let computation_even = array n (sum identity)
|
||
|
and computation_odd =
|
||
|
let sumr = array n (sum real)
|
||
|
and sumi = array n (sum ((times Complex.i) @@ imag)) in
|
||
|
array n (fun i ->
|
||
|
if (i = 0) then
|
||
|
(* expose some common subexpressions *)
|
||
|
input 0 @+
|
||
|
sigma 1 ((n + 1) / 2) (fun j -> input j @+ input (n - j))
|
||
|
else
|
||
|
let i' = min i (n - i) in
|
||
|
if (i < n - i) then
|
||
|
sumr i' @+ sumi i'
|
||
|
else
|
||
|
sumr i' @- sumi i') in
|
||
|
if (n >= !Magic.rader_min) then
|
||
|
dft_rader sign n input
|
||
|
else if (n == 2) then
|
||
|
computation_even
|
||
|
else
|
||
|
computation_odd
|
||
|
|
||
|
|
||
|
and dft_rader sign p input =
|
||
|
let half =
|
||
|
let one_half = inverse_int 2 in
|
||
|
times one_half
|
||
|
|
||
|
and make_product n a b =
|
||
|
let scale_factor = inverse_int n in
|
||
|
array n (fun i -> a i @* (scale_factor @* b i)) in
|
||
|
|
||
|
(* generates a convolution using ffts. (all arguments are the
|
||
|
same as to gen_convolution, below) *)
|
||
|
let gen_convolution_by_fft n a b addtoall =
|
||
|
let fft_a = dft 1 n a
|
||
|
and fft_b = dft 1 n b in
|
||
|
|
||
|
let fft_ab = make_product n fft_a fft_b
|
||
|
and dc_term i = if (i == 0) then addtoall else zero in
|
||
|
|
||
|
let fft_ab1 = array n (fun i -> fft_ab i @+ dc_term i)
|
||
|
and sum = fft_a 0 in
|
||
|
let conv = dft (-1) n fft_ab1 in
|
||
|
(sum, conv)
|
||
|
|
||
|
(* alternate routine for convolution. Seems to work better for
|
||
|
small sizes. I have no idea why. *)
|
||
|
and gen_convolution_by_fft_alt n a b addtoall =
|
||
|
let ap = array n (fun i -> half (a i @+ a ((n - i) mod n)))
|
||
|
and am = array n (fun i -> half (a i @- a ((n - i) mod n)))
|
||
|
and bp = array n (fun i -> half (b i @+ b ((n - i) mod n)))
|
||
|
and bm = array n (fun i -> half (b i @- b ((n - i) mod n)))
|
||
|
in
|
||
|
|
||
|
let fft_ap = dft 1 n ap
|
||
|
and fft_am = dft 1 n am
|
||
|
and fft_bp = dft 1 n bp
|
||
|
and fft_bm = dft 1 n bm in
|
||
|
|
||
|
let fft_abpp = make_product n fft_ap fft_bp
|
||
|
and fft_abpm = make_product n fft_ap fft_bm
|
||
|
and fft_abmp = make_product n fft_am fft_bp
|
||
|
and fft_abmm = make_product n fft_am fft_bm
|
||
|
and sum = fft_ap 0 @+ fft_am 0
|
||
|
and dc_term i = if (i == 0) then addtoall else zero in
|
||
|
|
||
|
let fft_ab1 = array n (fun i -> (fft_abpp i @+ fft_abmm i) @+ dc_term i)
|
||
|
and fft_ab2 = array n (fun i -> fft_abpm i @+ fft_abmp i) in
|
||
|
let conv1 = dft (-1) n fft_ab1
|
||
|
and conv2 = dft (-1) n fft_ab2 in
|
||
|
let conv = array n (fun i ->
|
||
|
conv1 i @+ conv2 i) in
|
||
|
(sum, conv)
|
||
|
|
||
|
(* generator of assignment list assigning conv to the convolution of
|
||
|
a and b, all of which are of length n. addtoall is added to
|
||
|
all of the elements of the result. Returns (sum, convolution) pair
|
||
|
where sum is the sum of the elements of a. *)
|
||
|
|
||
|
in let gen_convolution =
|
||
|
if (p <= !Magic.alternate_convolution) then
|
||
|
gen_convolution_by_fft_alt
|
||
|
else
|
||
|
gen_convolution_by_fft
|
||
|
|
||
|
(* fft generator for prime n = p using Rader's algorithm for
|
||
|
turning the fft into a convolution, which then can be
|
||
|
performed in a variety of ways *)
|
||
|
in
|
||
|
let g = find_generator p in
|
||
|
let ginv = pow_mod g (p - 2) p in
|
||
|
let input_perm = array p (fun i -> input (pow_mod g i p))
|
||
|
and omega_perm = array p (fun i -> exp p (sign * (pow_mod ginv i p)))
|
||
|
and output_perm = array p (fun i -> pow_mod ginv i p)
|
||
|
in let (sum, conv) =
|
||
|
(gen_convolution (p - 1) input_perm omega_perm (input 0))
|
||
|
in array p (fun i ->
|
||
|
if (i = 0) then
|
||
|
input 0 @+ sum
|
||
|
else
|
||
|
let i' = suchthat 0 (fun i' -> i = output_perm i')
|
||
|
in conv i')
|
||
|
|
||
|
(* our modified version of the conjugate-pair split-radix algorithm,
|
||
|
which reduces the number of multiplications by rescaling the
|
||
|
sub-transforms (power-of-two n's only) *)
|
||
|
and newsplit sign n input =
|
||
|
let rec s n k = (* recursive scale factor *)
|
||
|
if n <= 4 then
|
||
|
one
|
||
|
else
|
||
|
let k4 = (abs k) mod (n / 4) in
|
||
|
let k4' = if k4 <= (n / 8) then k4 else (n/4 - k4) in
|
||
|
(s (n / 4) k4') @* (real (exp n k4'))
|
||
|
|
||
|
and sinv n k = (* 1 / s(n,k) *)
|
||
|
if n <= 4 then
|
||
|
one
|
||
|
else
|
||
|
let k4 = (abs k) mod (n / 4) in
|
||
|
let k4' = if k4 <= (n / 8) then k4 else (n/4 - k4) in
|
||
|
(sinv (n / 4) k4') @* (sec n k4')
|
||
|
|
||
|
in let sdiv2 n k = (s n k) @* (sinv (2*n) k) (* s(n,k) / s(2*n,k) *)
|
||
|
and sdiv4 n k = (* s(n,k) / s(4*n,k) *)
|
||
|
let k4 = (abs k) mod n in
|
||
|
sec (4*n) (if k4 <= (n / 2) then k4 else (n - k4))
|
||
|
|
||
|
in let t n k = (exp n k) @* (sdiv4 (n/4) k)
|
||
|
|
||
|
and dft1 input = input
|
||
|
and dft2 input = array 2 (fun k -> (input 0) @+ ((input 1) @* exp 2 k))
|
||
|
|
||
|
in let rec newsplit0 sign n input =
|
||
|
if (n == 1) then dft1 input
|
||
|
else if (n == 2) then dft2 input
|
||
|
else let u = newsplit0 sign (n / 2) (fun i -> input (i*2))
|
||
|
and z = newsplitS sign (n / 4) (fun i -> input (i*4 + 1))
|
||
|
and z' = newsplitS sign (n / 4) (fun i -> input ((n + i*4 - 1) mod n))
|
||
|
and twid = array n (fun k -> s (n/4) k @* exp n (sign * k)) in
|
||
|
let w = array n (fun k -> twid k @* z (k mod (n / 4)))
|
||
|
and w' = array n (fun k -> conj (twid k) @* z' (k mod (n / 4))) in
|
||
|
let ww = array n (fun k -> w k @+ w' k) in
|
||
|
array n (fun k -> u (k mod (n / 2)) @+ ww k)
|
||
|
|
||
|
and newsplitS sign n input =
|
||
|
if (n == 1) then dft1 input
|
||
|
else if (n == 2) then dft2 input
|
||
|
else let u = newsplitS2 sign (n / 2) (fun i -> input (i*2))
|
||
|
and z = newsplitS sign (n / 4) (fun i -> input (i*4 + 1))
|
||
|
and z' = newsplitS sign (n / 4) (fun i -> input ((n + i*4 - 1) mod n)) in
|
||
|
let w = array n (fun k -> t n (sign * k) @* z (k mod (n / 4)))
|
||
|
and w' = array n (fun k -> conj (t n (sign * k)) @* z' (k mod (n / 4))) in
|
||
|
let ww = array n (fun k -> w k @+ w' k) in
|
||
|
array n (fun k -> u (k mod (n / 2)) @+ ww k)
|
||
|
|
||
|
and newsplitS2 sign n input =
|
||
|
if (n == 1) then dft1 input
|
||
|
else if (n == 2) then dft2 input
|
||
|
else let u = newsplitS4 sign (n / 2) (fun i -> input (i*2))
|
||
|
and z = newsplitS sign (n / 4) (fun i -> input (i*4 + 1))
|
||
|
and z' = newsplitS sign (n / 4) (fun i -> input ((n + i*4 - 1) mod n)) in
|
||
|
let w = array n (fun k -> t n (sign * k) @* z (k mod (n / 4)))
|
||
|
and w' = array n (fun k -> conj (t n (sign * k)) @* z' (k mod (n / 4))) in
|
||
|
let ww = array n (fun k -> (w k @+ w' k) @* (sdiv2 n k)) in
|
||
|
array n (fun k -> u (k mod (n / 2)) @+ ww k)
|
||
|
|
||
|
and newsplitS4 sign n input =
|
||
|
if (n == 1) then dft1 input
|
||
|
else if (n == 2) then
|
||
|
let f = dft2 input
|
||
|
in array 2 (fun k -> (f k) @* (sinv 8 k))
|
||
|
else let u = newsplitS2 sign (n / 2) (fun i -> input (i*2))
|
||
|
and z = newsplitS sign (n / 4) (fun i -> input (i*4 + 1))
|
||
|
and z' = newsplitS sign (n / 4) (fun i -> input ((n + i*4 - 1) mod n)) in
|
||
|
let w = array n (fun k -> t n (sign * k) @* z (k mod (n / 4)))
|
||
|
and w' = array n (fun k -> conj (t n (sign * k)) @* z' (k mod (n / 4))) in
|
||
|
let ww = array n (fun k -> w k @+ w' k) in
|
||
|
array n (fun k -> (u (k mod (n / 2)) @+ ww k) @* (sdiv4 n k))
|
||
|
|
||
|
in newsplit0 sign n input
|
||
|
|
||
|
and dft sign n input =
|
||
|
let rec cooley_tukey sign n1 n2 input =
|
||
|
let tmp1 =
|
||
|
array n2 (fun i2 ->
|
||
|
dft sign n1 (fun i1 -> input (i1 * n2 + i2))) in
|
||
|
let tmp2 =
|
||
|
array n1 (fun i1 ->
|
||
|
array n2 (fun i2 ->
|
||
|
exp n (sign * i1 * i2) @* tmp1 i2 i1)) in
|
||
|
let tmp3 = array n1 (fun i1 -> dft sign n2 (tmp2 i1)) in
|
||
|
(fun i -> tmp3 (i mod n1) (i / n1))
|
||
|
|
||
|
(*
|
||
|
* This is "exponent -1" split-radix by Dan Bernstein.
|
||
|
*)
|
||
|
and split_radix_dit sign n input =
|
||
|
let f0 = dft sign (n / 2) (fun i -> input (i * 2))
|
||
|
and f10 = dft sign (n / 4) (fun i -> input (i * 4 + 1))
|
||
|
and f11 = dft sign (n / 4) (fun i -> input ((n + i * 4 - 1) mod n)) in
|
||
|
let g10 = array n (fun k ->
|
||
|
exp n (sign * k) @* f10 (k mod (n / 4)))
|
||
|
and g11 = array n (fun k ->
|
||
|
exp n (- sign * k) @* f11 (k mod (n / 4))) in
|
||
|
let g1 = array n (fun k -> g10 k @+ g11 k) in
|
||
|
array n (fun k -> f0 (k mod (n / 2)) @+ g1 k)
|
||
|
|
||
|
and split_radix_dif sign n input =
|
||
|
let n2 = n / 2 and n4 = n / 4 in
|
||
|
let x0 = array n2 (fun i -> input i @+ input (i + n2))
|
||
|
and x10 = array n4 (fun i -> input i @- input (i + n2))
|
||
|
and x11 = array n4 (fun i ->
|
||
|
input (i + n4) @- input (i + n2 + n4)) in
|
||
|
let x1 k i =
|
||
|
exp n (k * i * sign) @* (x10 i @+ exp 4 (k * sign) @* x11 i) in
|
||
|
let f0 = dft sign n2 x0
|
||
|
and f1 = array 4 (fun k -> dft sign n4 (x1 k)) in
|
||
|
array n (fun k ->
|
||
|
if k mod 2 = 0 then f0 (k / 2)
|
||
|
else let k' = k mod 4 in f1 k' ((k - k') / 4))
|
||
|
|
||
|
and prime_factor sign n1 n2 input =
|
||
|
let tmp1 = array n2 (fun i2 ->
|
||
|
dft sign n1 (fun i1 -> input ((i1 * n2 + i2 * n1) mod n)))
|
||
|
in let tmp2 = array n1 (fun i1 ->
|
||
|
dft sign n2 (fun k2 -> tmp1 k2 i1))
|
||
|
in fun i -> tmp2 (i mod n1) (i mod n2)
|
||
|
|
||
|
in let algorithm sign n =
|
||
|
let r = choose_factor n in
|
||
|
if List.mem n !Magic.rader_list then
|
||
|
(* special cases *)
|
||
|
dft_rader sign n
|
||
|
else if (r == 1) then (* n is prime *)
|
||
|
dft_prime sign n
|
||
|
else if (gcd r (n / r)) == 1 then
|
||
|
prime_factor sign r (n / r)
|
||
|
else if (n mod 4 = 0 && n > 4) then
|
||
|
if !Magic.newsplit && is_power_of_two n then
|
||
|
newsplit sign n
|
||
|
else if !Magic.dif_split_radix then
|
||
|
split_radix_dif sign n
|
||
|
else
|
||
|
split_radix_dit sign n
|
||
|
else
|
||
|
cooley_tukey sign r (n / r)
|
||
|
in
|
||
|
array n (algorithm sign n input)
|