rustfft 6.2.0

High-performance FFT library written in pure Rust.
Documentation 
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

use core::arch::aarch64::*;

//  __  __       _   _               _________  _     _ _
// |  \/  | __ _| |_| |__           |___ /___ \| |__ (_) |_
// | |\/| |/ _` | __| '_ \   _____    |_ \ __) | '_ \| | __|
// | |  | | (_| | |_| | | | |_____|  ___) / __/| |_) | | |_
// |_|  |_|\__,_|\__|_| |_|         |____/_____|_.__/|_|\__|
//

pub struct Rotate90F32 {
    //sign_lo: float32x4_t,
    sign_hi: float32x2_t,
    sign_both: float32x4_t,
}

impl Rotate90F32 {
    pub fn new(positive: bool) -> Self {
        // There doesn't seem to be any need for rotating just the first element, but let's keep the code just in case
        //let sign_lo = unsafe {
        //    if positive {
        //        _mm_set_ps(0.0, 0.0, 0.0, -0.0)
        //    }
        //    else {
        //        _mm_set_ps(0.0, 0.0, -0.0, 0.0)
        //    }
        //};
        let sign_hi = unsafe {
            if positive {
                vld1_f32([-0.0, 0.0].as_ptr())
            } else {
                vld1_f32([0.0, -0.0].as_ptr())
            }
        };
        let sign_both = unsafe {
            if positive {
                vld1q_f32([-0.0, 0.0, -0.0, 0.0].as_ptr())
            } else {
                vld1q_f32([0.0, -0.0, 0.0, -0.0].as_ptr())
            }
        };
        Self {
            //sign_lo,
            sign_hi,
            sign_both,
        }
    }

    #[inline(always)]
    pub unsafe fn rotate_hi(&self, values: float32x4_t) -> float32x4_t {
        vcombine_f32(
            vget_low_f32(values),
            vreinterpret_f32_u32(veor_u32(
                vrev64_u32(vreinterpret_u32_f32(vget_high_f32(values))),
                vreinterpret_u32_f32(self.sign_hi),
            )),
        )
    }

    // There doesn't seem to be any need for rotating just the first element, but let's keep the code just in case
    //#[inline(always)]
    //pub unsafe fn rotate_lo(&self, values: __m128) -> __m128 {
    //    let temp = _mm_shuffle_ps(values, values, 0xE1);
    //    _mm_xor_ps(temp, self.sign_lo)
    //}

    #[inline(always)]
    pub unsafe fn rotate_both(&self, values: float32x4_t) -> float32x4_t {
        let temp = vrev64q_f32(values);
        vreinterpretq_f32_u32(veorq_u32(
            vreinterpretq_u32_f32(temp),
            vreinterpretq_u32_f32(self.sign_both),
        ))
    }
}

// Pack low (1st) complex
// left: r1.re, r1.im, r2.re, r2.im
// right: l1.re, l1.im, l2.re, l2.im
// --> r1.re, r1.im, l1.re, l1.im
#[inline(always)]
pub unsafe fn extract_lo_lo_f32(left: float32x4_t, right: float32x4_t) -> float32x4_t {
    //_mm_shuffle_ps(left, right, 0x44)
    vreinterpretq_f32_f64(vtrn1q_f64(
        vreinterpretq_f64_f32(left),
        vreinterpretq_f64_f32(right),
    ))
}

// Pack high (2nd) complex
// left: r1.re, r1.im, r2.re, r2.im
// right: l1.re, l1.im, l2.re, l2.im
// --> r2.re, r2.im, l2.re, l2.im
#[inline(always)]
pub unsafe fn extract_hi_hi_f32(left: float32x4_t, right: float32x4_t) -> float32x4_t {
    vreinterpretq_f32_f64(vtrn2q_f64(
        vreinterpretq_f64_f32(left),
        vreinterpretq_f64_f32(right),
    ))
}

// Pack low (1st) and high (2nd) complex
// left: r1.re, r1.im, r2.re, r2.im
// right: l1.re, l1.im, l2.re, l2.im
// --> r1.re, r1.im, l2.re, l2.im
#[inline(always)]
pub unsafe fn extract_lo_hi_f32(left: float32x4_t, right: float32x4_t) -> float32x4_t {
    vcombine_f32(vget_low_f32(left), vget_high_f32(right))
}

// Pack  high (2nd) and low (1st) complex
// left: r1.re, r1.im, r2.re, r2.im
// right: l1.re, l1.im, l2.re, l2.im
// --> r2.re, r2.im, l1.re, l1.im
#[inline(always)]
pub unsafe fn extract_hi_lo_f32(left: float32x4_t, right: float32x4_t) -> float32x4_t {
    vcombine_f32(vget_high_f32(left), vget_low_f32(right))
}

// Reverse complex
// values: a.re, a.im, b.re, b.im
// --> b.re, b.im, a.re, a.im
#[inline(always)]
pub unsafe fn reverse_complex_elements_f32(values: float32x4_t) -> float32x4_t {
    vcombine_f32(vget_high_f32(values), vget_low_f32(values))
}

// Reverse complex and then negate hi complex
// values: a.re, a.im, b.re, b.im
// --> b.re, b.im, -a.re, -a.im
#[inline(always)]
pub unsafe fn reverse_complex_and_negate_hi_f32(values: float32x4_t) -> float32x4_t {
    vcombine_f32(vget_high_f32(values), vneg_f32(vget_low_f32(values)))
}

// Invert sign of high (2nd) complex
// values: a.re, a.im, b.re, b.im
// -->  a.re, a.im, -b.re, -b.im
//#[inline(always)]
//pub unsafe fn negate_hi_f32(values: float32x4_t) -> float32x4_t {
//    vcombine_f32(vget_low_f32(values), vneg_f32(vget_high_f32(values)))
//}

// Duplicate low (1st) complex
// values: a.re, a.im, b.re, b.im
// --> a.re, a.im, a.re, a.im
#[inline(always)]
pub unsafe fn duplicate_lo_f32(values: float32x4_t) -> float32x4_t {
    vreinterpretq_f32_f64(vtrn1q_f64(
        vreinterpretq_f64_f32(values),
        vreinterpretq_f64_f32(values),
    ))
}

// Duplicate high (2nd) complex
// values: a.re, a.im, b.re, b.im
// --> b.re, b.im, b.re, b.im
#[inline(always)]
pub unsafe fn duplicate_hi_f32(values: float32x4_t) -> float32x4_t {
    vreinterpretq_f32_f64(vtrn2q_f64(
        vreinterpretq_f64_f32(values),
        vreinterpretq_f64_f32(values),
    ))
}

// transpose a 2x2 complex matrix given as [x0, x1], [x2, x3]
// result is [x0, x2], [x1, x3]
#[inline(always)]
pub unsafe fn transpose_complex_2x2_f32(left: float32x4_t, right: float32x4_t) -> [float32x4_t; 2] {
    let temp02 = extract_lo_lo_f32(left, right);
    let temp13 = extract_hi_hi_f32(left, right);
    [temp02, temp13]
}

// Complex multiplication.
// Each input contains two complex values, which are multiplied in parallel.
#[inline(always)]
pub unsafe fn mul_complex_f32(left: float32x4_t, right: float32x4_t) -> float32x4_t {
    // ARMv8.2-A introduced vcmulq_f32 and vcmlaq_f32 for complex multiplication, these intrinsics are not yet available.
    let temp1 = vtrn1q_f32(right, right);
    let temp2 = vtrn2q_f32(right, vnegq_f32(right));
    let temp3 = vmulq_f32(temp2, left);
    let temp4 = vrev64q_f32(temp3);
    vfmaq_f32(temp4, temp1, left)
}

//  __  __       _   _                __   _  _   _     _ _
// |  \/  | __ _| |_| |__            / /_ | || | | |__ (_) |_
// | |\/| |/ _` | __| '_ \   _____  | '_ \| || |_| '_ \| | __|
// | |  | | (_| | |_| | | | |_____| | (_) |__   _| |_) | | |_
// |_|  |_|\__,_|\__|_| |_|          \___/   |_| |_.__/|_|\__|
//

pub(crate) struct Rotate90F64 {
    sign: float64x2_t,
}

impl Rotate90F64 {
    pub fn new(positive: bool) -> Self {
        let sign = unsafe {
            if positive {
                vld1q_f64([-0.0, 0.0].as_ptr())
            } else {
                vld1q_f64([0.0, -0.0].as_ptr())
            }
        };
        Self { sign }
    }

    #[inline(always)]
    pub unsafe fn rotate(&self, values: float64x2_t) -> float64x2_t {
        let temp = vcombine_f64(vget_high_f64(values), vget_low_f64(values));
        vreinterpretq_f64_u64(veorq_u64(
            vreinterpretq_u64_f64(temp),
            vreinterpretq_u64_f64(self.sign),
        ))
    }
}

#[inline(always)]
pub unsafe fn mul_complex_f64(left: float64x2_t, right: float64x2_t) -> float64x2_t {
    // ARMv8.2-A introduced vcmulq_f64 and vcmlaq_f64 for complex multiplication, these intrinsics are not yet available.
    let temp = vcombine_f64(vneg_f64(vget_high_f64(left)), vget_low_f64(left));
    let sum = vmulq_laneq_f64::<0>(left, right);
    vfmaq_laneq_f64::<1>(sum, temp, right)
}

#[cfg(test)]
mod unit_tests {
    use super::*;
    use num_complex::Complex;

    #[test]
    fn test_mul_complex_f64() {
        unsafe {
            let right = vld1q_f64([1.0, 2.0].as_ptr());
            let left = vld1q_f64([5.0, 7.0].as_ptr());
            let res = mul_complex_f64(left, right);
            let expected = vld1q_f64([1.0 * 5.0 - 2.0 * 7.0, 1.0 * 7.0 + 2.0 * 5.0].as_ptr());
            assert_eq!(
                std::mem::transmute::<float64x2_t, Complex<f64>>(res),
                std::mem::transmute::<float64x2_t, Complex<f64>>(expected)
            );
        }
    }

    #[test]
    fn test_mul_complex_f32() {
        unsafe {
            let val1 = Complex::<f32>::new(1.0, 2.5);
            let val2 = Complex::<f32>::new(3.2, 4.75);
            let val3 = Complex::<f32>::new(5.75, 6.25);
            let val4 = Complex::<f32>::new(7.4, 8.5);

            let nbr2 = vld1q_f32([val3, val4].as_ptr() as *const f32);
            let nbr1 = vld1q_f32([val1, val2].as_ptr() as *const f32);
            let res = mul_complex_f32(nbr1, nbr2);
            let res = std::mem::transmute::<float32x4_t, [Complex<f32>; 2]>(res);
            let expected = [val1 * val3, val2 * val4];
            assert_eq!(res, expected);
        }
    }

    #[test]
    fn test_pack() {
        unsafe {
            let nbr2 = vld1q_f32([5.0, 6.0, 7.0, 8.0].as_ptr());
            let nbr1 = vld1q_f32([1.0, 2.0, 3.0, 4.0].as_ptr());
            let first = extract_lo_lo_f32(nbr1, nbr2);
            let second = extract_hi_hi_f32(nbr1, nbr2);
            let first = std::mem::transmute::<float32x4_t, [Complex<f32>; 2]>(first);
            let second = std::mem::transmute::<float32x4_t, [Complex<f32>; 2]>(second);
            let first_expected = [Complex::new(1.0, 2.0), Complex::new(5.0, 6.0)];
            let second_expected = [Complex::new(3.0, 4.0), Complex::new(7.0, 8.0)];
            assert_eq!(first, first_expected);
            assert_eq!(second, second_expected);
        }
    }
}