forked from gonum/gonum
/
radix24.go
343 lines (306 loc) · 8.74 KB
/
radix24.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
// Copyright ©2020 The Gonum Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package fourier
import (
"math"
"math/bits"
"math/cmplx"
)
// CoefficientsRadix2 computes the Fourier coefficients of the input
// sequence, converting the time series in seq into the frequency spectrum,
// in place and returning it. This transform is unnormalized; a call to
// CoefficientsRadix2 followed by a call of SequenceRadix2 will multiply the
// input sequence by the length of the sequence.
//
// CoefficientsRadix2 does not allocate, requiring that FFT twiddle factors
// be calculated lazily. For performance reasons, this is done by successive
// multiplication, so numerical accuracies can accumulate for large inputs.
// If accuracy is needed, the FFT or CmplxFFT types should be used.
//
// If the length of seq is not an integer power of 2, CoefficientsRadix2 will
// panic.
func CoefficientsRadix2(seq []complex128) []complex128 {
x := seq
switch len(x) {
default:
if bits.OnesCount(uint(len(x))) != 1 {
panic("fourier: radix-2 fft called with non-power 2 length")
}
case 0, 1:
return x
case 2:
x[0], x[1] =
x[0]+x[1],
x[0]-x[1]
return x
case 4:
t := x[1] + x[3]
u := x[2]
v := negi(x[1] - x[3])
x[0], x[1], x[2], x[3] =
x[0]+u+t,
x[0]-u+v,
x[0]+u-t,
x[0]-u-v
return x
}
bitReversePermute(x)
for k := 0; k < len(x); k += 4 {
t := x[k+2] + x[k+3]
u := x[k+1]
v := negi(x[k+2] - x[k+3])
x[k], x[k+1], x[k+2], x[k+3] =
x[k]+u+t,
x[k]-u+v,
x[k]+u-t,
x[k]-u-v
}
for m := 4; m < len(x); m *= 2 {
f := swap(complex(math.Sincos(-math.Pi / float64(m))))
for k := 0; k < len(x); k += 2 * m {
w := 1 + 0i
for j := 0; j < m; j++ {
i := j + k
u := w * x[i+m]
x[i], x[i+m] =
x[i]+u,
x[i]-u
w *= f
}
}
}
return x
}
// bitReversePermute performs a bit-reversal permutation on x.
func bitReversePermute(x []complex128) {
if len(x) < 2 || bits.OnesCount(uint(len(x))) != 1 {
panic("fourier: invalid bitReversePermute call")
}
lz := bits.LeadingZeros(uint(len(x) - 1))
i := 0
for ; i < len(x)/2; i++ {
j := int(bits.Reverse(uint(i)) >> lz)
if i < j {
x[i], x[j] = x[j], x[i]
}
}
for i++; i < len(x); i += 2 {
j := int(bits.Reverse(uint(i)) >> lz)
if i < j {
x[i], x[j] = x[j], x[i]
}
}
}
// SequenceRadix2 computes the real periodic sequence from the Fourier
// coefficients, converting the frequency spectrum in coeff into a time
// series, in place and returning it. This transform is unnormalized; a
// call to CoefficientsRadix2 followed by a call of SequenceRadix2 will
// multiply the input sequence by the length of the sequence.
//
// SequenceRadix2 does not allocate, requiring that FFT twiddle factors
// be calculated lazily. For performance reasons, this is done by successive
// multiplication, so numerical accuracies can accumulate for large inputs.
// If accuracy is needed, the FFT or CmplxFFT types should be used.
//
// If the length of coeff is not an integer power of 2, SequenceRadix2
// will panic.
func SequenceRadix2(coeff []complex128) []complex128 {
x := coeff
for i, j := 1, len(x)-1; i < j; i, j = i+1, j-1 {
x[i], x[j] = x[j], x[i]
}
CoefficientsRadix2(x)
return x
}
// PadRadix2 returns the values in x in a slice that is an integer
// power of 2 long. If x already has an integer power of 2 length
// it is returned unaltered.
func PadRadix2(x []complex128) []complex128 {
if len(x) == 0 {
return x
}
b := bits.Len(uint(len(x)))
if len(x) == 1<<(b-1) {
return x
}
p := make([]complex128, 1<<b)
copy(p, x)
return p
}
// TrimRadix2 returns the largest slice of x that is has an integer
// power of 2 length, and a slice holding the remaining elements.
func TrimRadix2(x []complex128) (even, remains []complex128) {
if len(x) == 0 {
return x, nil
}
n := 1 << (bits.Len(uint(len(x))) - 1)
return x[:n], x[n:]
}
// CoefficientsRadix4 computes the Fourier coefficients of the input
// sequence, converting the time series in seq into the frequency spectrum,
// in place and returning it. This transform is unnormalized; a call to
// CoefficientsRadix4 followed by a call of SequenceRadix4 will multiply the
// input sequence by the length of the sequence.
//
// CoefficientsRadix4 does not allocate, requiring that FFT twiddle factors
// be calculated lazily. For performance reasons, this is done by successive
// multiplication, so numerical accuracies can accumulate for large inputs.
// If accuracy is needed, the FFT or CmplxFFT types should be used.
//
// If the length of seq is not an integer power of 4, CoefficientsRadix4 will
// panic.
func CoefficientsRadix4(seq []complex128) []complex128 {
x := seq
switch len(x) {
default:
if bits.OnesCount(uint(len(x))) != 1 || bits.TrailingZeros(uint(len(x)))&0x1 != 0 {
panic("fourier: radix-4 fft called with non-power 4 length")
}
case 0, 1:
return x
case 4:
t := x[1] + x[3]
u := x[2]
v := negi(x[1] - x[3])
x[0], x[1], x[2], x[3] =
x[0]+u+t,
x[0]-u+v,
x[0]+u-t,
x[0]-u-v
return x
}
bitPairReversePermute(x)
for k := 0; k < len(x); k += 4 {
t := x[k+1] + x[k+3]
u := x[k+2]
v := negi(x[k+1] - x[k+3])
x[k], x[k+1], x[k+2], x[k+3] =
x[k]+u+t,
x[k]-u+v,
x[k]+u-t,
x[k]-u-v
}
for m := 4; m < len(x); m *= 4 {
f := swap(complex(math.Sincos((-math.Pi / 2) / float64(m))))
for k := 0; k < len(x); k += m * 4 {
w := 1 + 0i
w2 := w
w3 := w2
for j := 0; j < m; j++ {
i := j + k
t := x[i+m]*w + x[i+3*m]*w3
u := x[i+2*m] * w2
v := negi(x[i+m]*w - x[i+3*m]*w3)
x[i], x[i+m], x[i+2*m], x[i+3*m] =
x[i]+u+t,
x[i]-u+v,
x[i]+u-t,
x[i]-u-v
wt := f
w *= wt
wt *= f
w2 *= wt
wt *= f
w3 *= wt
}
}
}
return x
}
// bitPairReversePermute performs a bit pair-reversal permutation on x.
func bitPairReversePermute(x []complex128) {
if len(x) < 4 || bits.OnesCount(uint(len(x))) != 1 || bits.TrailingZeros(uint(len(x)))&0x1 != 0 {
panic("fourier: invalid bitPairReversePermute call")
}
lz := bits.LeadingZeros(uint(len(x) - 1))
i := 0
for ; i < 3*len(x)/4; i++ {
j := int(reversePairs(uint(i)) >> lz)
if i < j {
x[i], x[j] = x[j], x[i]
}
}
for i++; i < len(x); i += 2 {
j := int(reversePairs(uint(i)) >> lz)
if i < j {
x[i], x[j] = x[j], x[i]
}
}
}
// SequenceRadix4 computes the real periodic sequence from the Fourier
// coefficients, converting the frequency spectrum in coeff into a time
// series, in place and returning it. This transform is unnormalized; a
// call to CoefficientsRadix4 followed by a call of SequenceRadix4 will
// multiply the input sequence by the length of the sequence.
//
// SequenceRadix4 does not allocate, requiring that FFT twiddle factors
// be calculated lazily. For performance reasons, this is done by successive
// multiplication, so numerical accuracies can accumulate for large inputs.
// If accuracy is needed, the FFT or CmplxFFT types should be used.
//
// If the length of coeff is not an integer power of 4, SequenceRadix4
// will panic.
func SequenceRadix4(coeff []complex128) []complex128 {
x := coeff
for i, j := 1, len(x)-1; i < j; i, j = i+1, j-1 {
x[i], x[j] = x[j], x[i]
}
CoefficientsRadix4(x)
return x
}
// PadRadix4 returns the values in x in a slice that is an integer
// power of 4 long. If x already has an integer power of 4 length
// it is returned unaltered.
func PadRadix4(x []complex128) []complex128 {
if len(x) == 0 {
return x
}
b := bits.Len(uint(len(x)))
if len(x) == 1<<(b-1) && b&0x1 == 1 {
return x
}
p := make([]complex128, 1<<((b+1)&^0x1))
copy(p, x)
return p
}
// TrimRadix4 returns the largest slice of x that is has an integer
// power of 4 length, and a slice holding the remaining elements.
func TrimRadix4(x []complex128) (even, remains []complex128) {
if len(x) == 0 {
return x, nil
}
n := 1 << ((bits.Len(uint(len(x))) - 1) &^ 0x1)
return x[:n], x[n:]
}
// reversePairs returns the value of x with its bit pairs in reversed order.
func reversePairs(x uint) uint {
if bits.UintSize == 32 {
return uint(reversePairs32(uint32(x)))
}
return uint(reversePairs64(uint64(x)))
}
const (
m1 = 0x3333333333333333
m2 = 0x0f0f0f0f0f0f0f0f
)
// reversePairs32 returns the value of x with its bit pairs in reversed order.
func reversePairs32(x uint32) uint32 {
const m = 1<<32 - 1
x = x>>2&(m1&m) | x&(m1&m)<<2
x = x>>4&(m2&m) | x&(m2&m)<<4
return bits.ReverseBytes32(x)
}
// reversePairs64 returns the value of x with its bit pairs in reversed order.
func reversePairs64(x uint64) uint64 {
const m = 1<<64 - 1
x = x>>2&(m1&m) | x&(m1&m)<<2
x = x>>4&(m2&m) | x&(m2&m)<<4
return bits.ReverseBytes64(x)
}
func negi(c complex128) complex128 {
return cmplx.Conj(swap(c))
}
func swap(c complex128) complex128 {
return complex(imag(c), real(c))
}