| 1 |
#![allow(bare_trait_objects)] |
| 2 |
#![allow(non_snake_case)] |
| 3 |
#![feature(test)] |
| 4 |
extern crate test; |
| 5 |
extern crate rustfft; |
| 6 |
|
| 7 |
use std::sync::Arc; |
| 8 |
use test::Bencher; |
| 9 |
use rustfft::{Direction, FftNum, Fft, FftDirection, Length}; |
| 10 |
use rustfft::num_complex::Complex; |
| 11 |
use rustfft::num_traits::Zero; |
| 12 |
use rustfft::algorithm::*; |
| 13 |
use rustfft::algorithm::butterflies::*; |
| 14 |
|
| 15 |
struct Noop { |
| 16 |
len: usize, |
| 17 |
direction: FftDirection, |
| 18 |
} |
| 19 |
impl<T: FftNum> Fft<T> for Noop { |
| 20 |
fn process_with_scratch(&self, _buffer: &mut [Complex<T>], _scratch: &mut [Complex<T>]) {} |
| 21 |
fn process_outofplace_with_scratch(&self, _input: &mut [Complex<T>], _output: &mut [Complex<T>], _scratch: &mut [Complex<T>]) {} |
| 22 |
fn get_inplace_scratch_len(&self) -> usize { self.len } |
| 23 |
fn get_outofplace_scratch_len(&self) -> usize { 0 } |
| 24 |
} |
| 25 |
impl Length for Noop { |
| 26 |
fn len(&self) -> usize { self.len } |
| 27 |
} |
| 28 |
impl Direction for Noop { |
| 29 |
fn fft_direction(&self) -> FftDirection { self.direction } |
| 30 |
} |
| 31 |
|
| 32 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 33 |
/// for a given length |
| 34 |
fn bench_planned_f32(b: &mut Bencher, len: usize) { |
| 35 |
|
| 36 |
let mut planner = rustfft::FftPlanner::new(); |
| 37 |
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len); |
| 38 |
assert_eq!(fft.len(), len); |
| 39 |
|
| 40 |
let mut buffer = vec![Complex::zero(); len]; |
| 41 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 42 |
b.iter(|| { |
| 43 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 44 |
}); |
| 45 |
} |
| 46 |
|
| 47 |
|
| 48 |
// Powers of 4 |
| 49 |
#[bench] fn planned32_p2_00000064(b: &mut Bencher) { bench_planned_f32(b, 64); } |
| 50 |
#[bench] fn planned32_p2_00000128(b: &mut Bencher) { bench_planned_f32(b, 128); } |
| 51 |
#[bench] fn planned32_p2_00000256(b: &mut Bencher) { bench_planned_f32(b, 256); } |
| 52 |
#[bench] fn planned32_p2_00000512(b: &mut Bencher) { bench_planned_f32(b, 512); } |
| 53 |
#[bench] fn planned32_p2_00001024(b: &mut Bencher) { bench_planned_f32(b, 1024); } |
| 54 |
#[bench] fn planned32_p2_00002048(b: &mut Bencher) { bench_planned_f32(b, 2048); } |
| 55 |
#[bench] fn planned32_p2_00004096(b: &mut Bencher) { bench_planned_f32(b, 4096); } |
| 56 |
#[bench] fn planned32_p2_00016384(b: &mut Bencher) { bench_planned_f32(b, 16384); } |
| 57 |
#[bench] fn planned32_p2_00065536(b: &mut Bencher) { bench_planned_f32(b, 65536); } |
| 58 |
#[bench] fn planned32_p2_01048576(b: &mut Bencher) { bench_planned_f32(b, 1048576); } |
| 59 |
#[bench] fn planned32_p2_16777216(b: &mut Bencher) { bench_planned_f32(b, 16777216); } |
| 60 |
|
| 61 |
|
| 62 |
// Powers of 5 |
| 63 |
#[bench] fn planned32_p5_00125(b: &mut Bencher) { bench_planned_f32(b, 125); } |
| 64 |
#[bench] fn planned32_p5_00625(b: &mut Bencher) { bench_planned_f32(b, 625); } |
| 65 |
#[bench] fn planned32_p5_03125(b: &mut Bencher) { bench_planned_f32(b, 3125); } |
| 66 |
#[bench] fn planned32_p5_15625(b: &mut Bencher) { bench_planned_f32(b, 15625); } |
| 67 |
|
| 68 |
// Powers of 7 |
| 69 |
#[bench] fn planned32_p7_00343(b: &mut Bencher) { bench_planned_f32(b, 343); } |
| 70 |
#[bench] fn planned32_p7_02401(b: &mut Bencher) { bench_planned_f32(b, 2401); } |
| 71 |
#[bench] fn planned32_p7_16807(b: &mut Bencher) { bench_planned_f32(b, 16807); } |
| 72 |
|
| 73 |
// Prime lengths |
| 74 |
// Prime lengths |
| 75 |
#[bench] fn planned32_prime_0005(b: &mut Bencher) { bench_planned_f32(b, 5); } |
| 76 |
#[bench] fn planned32_prime_0017(b: &mut Bencher) { bench_planned_f32(b, 17); } |
| 77 |
#[bench] fn planned32_prime_0149(b: &mut Bencher) { bench_planned_f32(b, 149); } |
| 78 |
#[bench] fn planned32_prime_0151(b: &mut Bencher) { bench_planned_f32(b, 151); } |
| 79 |
#[bench] fn planned32_prime_0251(b: &mut Bencher) { bench_planned_f32(b, 251); } |
| 80 |
#[bench] fn planned32_prime_0257(b: &mut Bencher) { bench_planned_f32(b, 257); } |
| 81 |
#[bench] fn planned32_prime_1009(b: &mut Bencher) { bench_planned_f32(b, 1009); } |
| 82 |
#[bench] fn planned32_prime_1201(b: &mut Bencher) { bench_planned_f32(b, 1201); } |
| 83 |
#[bench] fn planned32_prime_2017(b: &mut Bencher) { bench_planned_f32(b, 2017); } |
| 84 |
#[bench] fn planned32_prime_2879(b: &mut Bencher) { bench_planned_f32(b, 2879); } |
| 85 |
#[bench] fn planned32_prime_32767(b: &mut Bencher) { bench_planned_f32(b, 32767); } |
| 86 |
#[bench] fn planned32_prime_65521(b: &mut Bencher) { bench_planned_f32(b, 65521); } |
| 87 |
#[bench] fn planned32_prime_65537(b: &mut Bencher) { bench_planned_f32(b, 65537); } |
| 88 |
#[bench] fn planned32_prime_746483(b: &mut Bencher) { bench_planned_f32(b,746483); } |
| 89 |
#[bench] fn planned32_prime_746497(b: &mut Bencher) { bench_planned_f32(b,746497); } |
| 90 |
|
| 91 |
//primes raised to a power |
| 92 |
#[bench] fn planned32_primepower_044521(b: &mut Bencher) { bench_planned_f32(b, 44521); } // 211^2 |
| 93 |
#[bench] fn planned32_primepower_160801(b: &mut Bencher) { bench_planned_f32(b, 160801); } // 401^2 |
| 94 |
|
| 95 |
// numbers times powers of two |
| 96 |
#[bench] fn planned32_composite_024576(b: &mut Bencher) { bench_planned_f32(b, 24576); } |
| 97 |
#[bench] fn planned32_composite_020736(b: &mut Bencher) { bench_planned_f32(b, 20736); } |
| 98 |
|
| 99 |
// power of 2 times large prime |
| 100 |
#[bench] fn planned32_composite_032192(b: &mut Bencher) { bench_planned_f32(b, 32192); } |
| 101 |
#[bench] fn planned32_composite_024028(b: &mut Bencher) { bench_planned_f32(b, 24028); } |
| 102 |
|
| 103 |
// small mixed composites times a large prime |
| 104 |
#[bench] fn planned32_composite_005472(b: &mut Bencher) { bench_planned_f32(b, 5472); } |
| 105 |
#[bench] fn planned32_composite_030270(b: &mut Bencher) { bench_planned_f32(b, 30270); } |
| 106 |
|
| 107 |
// small mixed composites |
| 108 |
#[bench] fn planned32_composite_000018(b: &mut Bencher) { bench_planned_f32(b, 00018); } |
| 109 |
#[bench] fn planned32_composite_000360(b: &mut Bencher) { bench_planned_f32(b, 00360); } |
| 110 |
#[bench] fn planned32_composite_001200(b: &mut Bencher) { bench_planned_f32(b, 01200); } |
| 111 |
#[bench] fn planned32_composite_044100(b: &mut Bencher) { bench_planned_f32(b, 44100); } |
| 112 |
#[bench] fn planned32_composite_048000(b: &mut Bencher) { bench_planned_f32(b, 48000); } |
| 113 |
#[bench] fn planned32_composite_046656(b: &mut Bencher) { bench_planned_f32(b, 46656); } |
| 114 |
#[bench] fn planned32_composite_100000(b: &mut Bencher) { bench_planned_f32(b, 100000); } |
| 115 |
|
| 116 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 117 |
/// for a given length |
| 118 |
fn bench_planned_f64(b: &mut Bencher, len: usize) { |
| 119 |
|
| 120 |
let mut planner = rustfft::FftPlanner::new(); |
| 121 |
let fft: Arc<dyn Fft<f64>> = planner.plan_fft_forward(len); |
| 122 |
|
| 123 |
let mut buffer = vec![Complex::zero(); len]; |
| 124 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 125 |
b.iter(|| { fft.process_with_scratch(&mut buffer, &mut scratch); }); |
| 126 |
} |
| 127 |
|
| 128 |
#[bench] fn planned64_p2_00000064(b: &mut Bencher) { bench_planned_f64(b, 64); } |
| 129 |
#[bench] fn planned64_p2_00000128(b: &mut Bencher) { bench_planned_f64(b, 128); } |
| 130 |
#[bench] fn planned64_p2_00000256(b: &mut Bencher) { bench_planned_f64(b, 256); } |
| 131 |
#[bench] fn planned64_p2_00000512(b: &mut Bencher) { bench_planned_f64(b, 512); } |
| 132 |
#[bench] fn planned64_p2_00001024(b: &mut Bencher) { bench_planned_f64(b, 1024); } |
| 133 |
#[bench] fn planned64_p2_00002048(b: &mut Bencher) { bench_planned_f64(b, 2048); } |
| 134 |
#[bench] fn planned64_p2_00004096(b: &mut Bencher) { bench_planned_f64(b, 4096); } |
| 135 |
#[bench] fn planned64_p2_00016384(b: &mut Bencher) { bench_planned_f64(b, 16384); } |
| 136 |
#[bench] fn planned64_p2_00065536(b: &mut Bencher) { bench_planned_f64(b, 65536); } |
| 137 |
#[bench] fn planned64_p2_01048576(b: &mut Bencher) { bench_planned_f64(b, 1048576); } |
| 138 |
//#[bench] fn planned64_p2_16777216(b: &mut Bencher) { bench_planned_f64(b, 16777216); } |
| 139 |
|
| 140 |
// Powers of 5 |
| 141 |
#[bench] fn planned64_p5_00125(b: &mut Bencher) { bench_planned_f64(b, 125); } |
| 142 |
#[bench] fn planned64_p5_00625(b: &mut Bencher) { bench_planned_f64(b, 625); } |
| 143 |
#[bench] fn planned64_p5_03125(b: &mut Bencher) { bench_planned_f64(b, 3125); } |
| 144 |
#[bench] fn planned64_p5_15625(b: &mut Bencher) { bench_planned_f64(b, 15625); } |
| 145 |
|
| 146 |
#[bench] fn planned64_p7_00343(b: &mut Bencher) { bench_planned_f64(b, 343); } |
| 147 |
#[bench] fn planned64_p7_02401(b: &mut Bencher) { bench_planned_f64(b, 2401); } |
| 148 |
#[bench] fn planned64_p7_16807(b: &mut Bencher) { bench_planned_f64(b, 16807); } |
| 149 |
|
| 150 |
// Prime lengths |
| 151 |
#[bench] fn planned64_prime_0005(b: &mut Bencher) { bench_planned_f64(b, 5); } |
| 152 |
#[bench] fn planned64_prime_0017(b: &mut Bencher) { bench_planned_f64(b, 17); } |
| 153 |
#[bench] fn planned64_prime_0149(b: &mut Bencher) { bench_planned_f64(b, 149); } |
| 154 |
#[bench] fn planned64_prime_0151(b: &mut Bencher) { bench_planned_f64(b, 151); } |
| 155 |
#[bench] fn planned64_prime_0251(b: &mut Bencher) { bench_planned_f64(b, 251); } |
| 156 |
#[bench] fn planned64_prime_0257(b: &mut Bencher) { bench_planned_f64(b, 257); } |
| 157 |
#[bench] fn planned64_prime_1009(b: &mut Bencher) { bench_planned_f64(b, 1009); } |
| 158 |
#[bench] fn planned64_prime_2017(b: &mut Bencher) { bench_planned_f64(b, 2017); } |
| 159 |
#[bench] fn planned64_prime_2879(b: &mut Bencher) { bench_planned_f64(b, 2879); } |
| 160 |
#[bench] fn planned64_prime_32767(b: &mut Bencher) { bench_planned_f64(b, 32767); } |
| 161 |
#[bench] fn planned64_prime_65521(b: &mut Bencher) { bench_planned_f64(b, 65521); } |
| 162 |
#[bench] fn planned64_prime_65537(b: &mut Bencher) { bench_planned_f64(b, 65537); } |
| 163 |
#[bench] fn planned64_prime_746483(b: &mut Bencher) { bench_planned_f64(b,746483); } |
| 164 |
#[bench] fn planned64_prime_746497(b: &mut Bencher) { bench_planned_f64(b,746497); } |
| 165 |
|
| 166 |
//primes raised to a power |
| 167 |
#[bench] fn planned64_primepower_044521(b: &mut Bencher) { bench_planned_f64(b, 44521); } // 211^2 |
| 168 |
#[bench] fn planned64_primepower_160801(b: &mut Bencher) { bench_planned_f64(b, 160801); } // 401^2 |
| 169 |
|
| 170 |
// numbers times powers of two |
| 171 |
#[bench] fn planned64_composite_024576(b: &mut Bencher) { bench_planned_f64(b, 24576); } |
| 172 |
#[bench] fn planned64_composite_020736(b: &mut Bencher) { bench_planned_f64(b, 20736); } |
| 173 |
|
| 174 |
// power of 2 times large prime |
| 175 |
#[bench] fn planned64_composite_032192(b: &mut Bencher) { bench_planned_f64(b, 32192); } |
| 176 |
#[bench] fn planned64_composite_024028(b: &mut Bencher) { bench_planned_f64(b, 24028); } |
| 177 |
|
| 178 |
// small mixed composites times a large prime |
| 179 |
#[bench] fn planned64_composite_030270(b: &mut Bencher) { bench_planned_f64(b, 30270); } |
| 180 |
|
| 181 |
// small mixed composites |
| 182 |
#[bench] fn planned64_composite_000018(b: &mut Bencher) { bench_planned_f64(b, 00018); } |
| 183 |
#[bench] fn planned64_composite_000360(b: &mut Bencher) { bench_planned_f64(b, 00360); } |
| 184 |
#[bench] fn planned64_composite_044100(b: &mut Bencher) { bench_planned_f64(b, 44100); } |
| 185 |
#[bench] fn planned64_composite_048000(b: &mut Bencher) { bench_planned_f64(b, 48000); } |
| 186 |
#[bench] fn planned64_composite_046656(b: &mut Bencher) { bench_planned_f64(b, 46656); } |
| 187 |
#[bench] fn planned64_composite_100000(b: &mut Bencher) { bench_planned_f64(b, 100000); } |
| 188 |
|
| 189 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 190 |
/// for a given length, specific to the Good-Thomas algorithm |
| 191 |
fn bench_good_thomas(b: &mut Bencher, width: usize, height: usize) { |
| 192 |
|
| 193 |
let mut planner = rustfft::FftPlanner::new(); |
| 194 |
let width_fft = planner.plan_fft_forward(width); |
| 195 |
let height_fft = planner.plan_fft_forward(height); |
| 196 |
|
| 197 |
let fft : Arc<Fft<f32>> = Arc::new(GoodThomasAlgorithm::new(width_fft, height_fft)); |
| 198 |
|
| 199 |
let mut buffer = vec![Complex::zero(); width * height]; |
| 200 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 201 |
b.iter(|| {fft.process_with_scratch(&mut buffer, &mut scratch);} ); |
| 202 |
} |
| 203 |
|
| 204 |
#[bench] fn good_thomas_0002_3(b: &mut Bencher) { bench_good_thomas(b, 2, 3); } |
| 205 |
#[bench] fn good_thomas_0003_4(b: &mut Bencher) { bench_good_thomas(b, 3, 4); } |
| 206 |
#[bench] fn good_thomas_0004_5(b: &mut Bencher) { bench_good_thomas(b, 4, 5); } |
| 207 |
#[bench] fn good_thomas_0007_32(b: &mut Bencher) { bench_good_thomas(b, 7, 32); } |
| 208 |
#[bench] fn good_thomas_0032_27(b: &mut Bencher) { bench_good_thomas(b, 32, 27); } |
| 209 |
#[bench] fn good_thomas_0256_243(b: &mut Bencher) { bench_good_thomas(b, 256, 243); } |
| 210 |
#[bench] fn good_thomas_2048_3(b: &mut Bencher) { bench_good_thomas(b, 2048, 3); } |
| 211 |
#[bench] fn good_thomas_2048_2187(b: &mut Bencher) { bench_good_thomas(b, 2048, 2187); } |
| 212 |
|
| 213 |
/// Times just the FFT setup (not execution) |
| 214 |
/// for a given length, specific to the Good-Thomas algorithm |
| 215 |
fn bench_good_thomas_setup(b: &mut Bencher, width: usize, height: usize) { |
| 216 |
|
| 217 |
let mut planner = rustfft::FftPlanner::new(); |
| 218 |
let width_fft = planner.plan_fft_forward(width); |
| 219 |
let height_fft = planner.plan_fft_forward(height); |
| 220 |
|
| 221 |
b.iter(|| { |
| 222 |
let fft : Arc<Fft<f32>> = Arc::new(GoodThomasAlgorithm::new(Arc::clone(&width_fft), Arc::clone(&height_fft))); |
| 223 |
test::black_box(fft); |
| 224 |
}); |
| 225 |
} |
| 226 |
|
| 227 |
#[bench] fn good_thomas_setup_0002_3(b: &mut Bencher) { bench_good_thomas_setup(b, 2, 3); } |
| 228 |
#[bench] fn good_thomas_setup_0003_4(b: &mut Bencher) { bench_good_thomas_setup(b, 3, 4); } |
| 229 |
#[bench] fn good_thomas_setup_0004_5(b: &mut Bencher) { bench_good_thomas_setup(b, 4, 5); } |
| 230 |
#[bench] fn good_thomas_setup_0007_32(b: &mut Bencher) { bench_good_thomas_setup(b, 7, 32); } |
| 231 |
#[bench] fn good_thomas_setup_0032_27(b: &mut Bencher) { bench_good_thomas_setup(b, 32, 27); } |
| 232 |
#[bench] fn good_thomas_setup_0256_243(b: &mut Bencher) { bench_good_thomas_setup(b, 256, 243); } |
| 233 |
#[bench] fn good_thomas_setup_2048_3(b: &mut Bencher) { bench_good_thomas_setup(b, 2048, 3); } |
| 234 |
#[bench] fn good_thomas_setup_2048_2187(b: &mut Bencher) { bench_good_thomas_setup(b, 2048, 2187); } |
| 235 |
|
| 236 |
/// Times just the FFT setup (not execution) |
| 237 |
/// for a given length, specific to MixedRadix |
| 238 |
fn bench_mixed_radix_setup(b: &mut Bencher, width: usize, height: usize) { |
| 239 |
|
| 240 |
let mut planner = rustfft::FftPlanner::new(); |
| 241 |
let width_fft = planner.plan_fft_forward(width); |
| 242 |
let height_fft = planner.plan_fft_forward(height); |
| 243 |
|
| 244 |
b.iter(|| { |
| 245 |
let fft : Arc<Fft<f32>> = Arc::new(MixedRadix::new(Arc::clone(&width_fft), Arc::clone(&height_fft))); |
| 246 |
test::black_box(fft); |
| 247 |
}); |
| 248 |
} |
| 249 |
|
| 250 |
#[bench] fn setup_mixed_radix_0002_3(b: &mut Bencher) { bench_mixed_radix_setup(b, 2, 3); } |
| 251 |
#[bench] fn setup_mixed_radix_0003_4(b: &mut Bencher) { bench_mixed_radix_setup(b, 3, 4); } |
| 252 |
#[bench] fn setup_mixed_radix_0004_5(b: &mut Bencher) { bench_mixed_radix_setup(b, 4, 5); } |
| 253 |
#[bench] fn setup_mixed_radix_0007_32(b: &mut Bencher) { bench_mixed_radix_setup(b, 7, 32); } |
| 254 |
#[bench] fn setup_mixed_radix_0032_27(b: &mut Bencher) { bench_mixed_radix_setup(b, 32, 27); } |
| 255 |
#[bench] fn setup_mixed_radix_0256_243(b: &mut Bencher) { bench_mixed_radix_setup(b, 256, 243); } |
| 256 |
#[bench] fn setup_mixed_radix_2048_3(b: &mut Bencher) { bench_mixed_radix_setup(b, 2048, 3); } |
| 257 |
#[bench] fn setup_mixed_radix_2048_2187(b: &mut Bencher) { bench_mixed_radix_setup(b, 2048, 2187); } |
| 258 |
|
| 259 |
/// Times just the FFT setup (not execution) |
| 260 |
/// for a given length, specific to MixedRadix |
| 261 |
fn bench_small_mixed_radix_setup(b: &mut Bencher, width: usize, height: usize) { |
| 262 |
|
| 263 |
let mut planner = rustfft::FftPlanner::new(); |
| 264 |
let width_fft = planner.plan_fft_forward(width); |
| 265 |
let height_fft = planner.plan_fft_forward(height); |
| 266 |
|
| 267 |
b.iter(|| { |
| 268 |
let fft : Arc<Fft<f32>> = Arc::new(MixedRadixSmall::new(Arc::clone(&width_fft), Arc::clone(&height_fft))); |
| 269 |
test::black_box(fft); |
| 270 |
}); |
| 271 |
} |
| 272 |
|
| 273 |
#[bench] fn setup_small_mixed_radix_0002_3(b: &mut Bencher) { bench_small_mixed_radix_setup(b, 2, 3); } |
| 274 |
#[bench] fn setup_small_mixed_radix_0003_4(b: &mut Bencher) { bench_small_mixed_radix_setup(b, 3, 4); } |
| 275 |
#[bench] fn setup_small_mixed_radix_0004_5(b: &mut Bencher) { bench_small_mixed_radix_setup(b, 4, 5); } |
| 276 |
#[bench] fn setup_small_mixed_radix_0007_32(b: &mut Bencher) { bench_small_mixed_radix_setup(b, 7, 32); } |
| 277 |
#[bench] fn setup_small_mixed_radix_0032_27(b: &mut Bencher) { bench_small_mixed_radix_setup(b, 32, 27); } |
| 278 |
|
| 279 |
|
| 280 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 281 |
/// for a given length, specific to the Mixed-Radix algorithm |
| 282 |
fn bench_mixed_radix(b: &mut Bencher, width: usize, height: usize) { |
| 283 |
|
| 284 |
let mut planner = rustfft::FftPlanner::new(); |
| 285 |
let width_fft = planner.plan_fft_forward(width); |
| 286 |
let height_fft = planner.plan_fft_forward(height); |
| 287 |
|
| 288 |
let fft : Arc<Fft<_>> = Arc::new(MixedRadix::new(width_fft, height_fft)); |
| 289 |
|
| 290 |
let mut buffer = vec![Complex{re: 0_f32, im: 0_f32}; fft.len()]; |
| 291 |
let mut scratch = vec![Complex{re: 0_f32, im: 0_f32}; fft.get_inplace_scratch_len()]; |
| 292 |
b.iter(|| {fft.process_with_scratch(&mut buffer, &mut scratch);} ); |
| 293 |
} |
| 294 |
|
| 295 |
#[bench] fn mixed_radix_0002_3(b: &mut Bencher) { bench_mixed_radix(b, 2, 3); } |
| 296 |
#[bench] fn mixed_radix_0003_4(b: &mut Bencher) { bench_mixed_radix(b, 3, 4); } |
| 297 |
#[bench] fn mixed_radix_0004_5(b: &mut Bencher) { bench_mixed_radix(b, 4, 5); } |
| 298 |
#[bench] fn mixed_radix_0007_32(b: &mut Bencher) { bench_mixed_radix(b, 7, 32); } |
| 299 |
#[bench] fn mixed_radix_0032_27(b: &mut Bencher) { bench_mixed_radix(b, 32, 27); } |
| 300 |
#[bench] fn mixed_radix_0256_243(b: &mut Bencher) { bench_mixed_radix(b, 256, 243); } |
| 301 |
#[bench] fn mixed_radix_2048_3(b: &mut Bencher) { bench_mixed_radix(b, 2048, 3); } |
| 302 |
#[bench] fn mixed_radix_2048_2187(b: &mut Bencher) { bench_mixed_radix(b, 2048, 2187); } |
| 303 |
|
| 304 |
fn plan_butterfly_fft(len: usize) -> Arc<Fft<f32>> { |
| 305 |
match len { |
| 306 |
2 => Arc::new(Butterfly2::new(FftDirection::Forward)), |
| 307 |
3 => Arc::new(Butterfly3::new(FftDirection::Forward)), |
| 308 |
4 => Arc::new(Butterfly4::new(FftDirection::Forward)), |
| 309 |
5 => Arc::new(Butterfly5::new(FftDirection::Forward)), |
| 310 |
6 => Arc::new(Butterfly6::new(FftDirection::Forward)), |
| 311 |
7 => Arc::new(Butterfly7::new(FftDirection::Forward)), |
| 312 |
8 => Arc::new(Butterfly8::new(FftDirection::Forward)), |
| 313 |
16 => Arc::new(Butterfly16::new(FftDirection::Forward)), |
| 314 |
32 => Arc::new(Butterfly32::new(FftDirection::Forward)), |
| 315 |
_ => panic!("Invalid butterfly size: {}", len), |
| 316 |
} |
| 317 |
} |
| 318 |
|
| 319 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 320 |
/// for a given length, specific to the MixedRadixSmall algorithm |
| 321 |
fn bench_mixed_radix_small(b: &mut Bencher, width: usize, height: usize) { |
| 322 |
|
| 323 |
let width_fft = plan_butterfly_fft(width); |
| 324 |
let height_fft = plan_butterfly_fft(height); |
| 325 |
|
| 326 |
let fft : Arc<Fft<_>> = Arc::new(MixedRadixSmall::new(width_fft, height_fft)); |
| 327 |
|
| 328 |
let mut signal = vec![Complex{re: 0_f32, im: 0_f32}; width * height]; |
| 329 |
let mut spectrum = signal.clone(); |
| 330 |
b.iter(|| {fft.process_with_scratch(&mut signal, &mut spectrum);} ); |
| 331 |
} |
| 332 |
|
| 333 |
#[bench] fn mixed_radix_small_0002_3(b: &mut Bencher) { bench_mixed_radix_small(b, 2, 3); } |
| 334 |
#[bench] fn mixed_radix_small_0003_4(b: &mut Bencher) { bench_mixed_radix_small(b, 3, 4); } |
| 335 |
#[bench] fn mixed_radix_small_0004_5(b: &mut Bencher) { bench_mixed_radix_small(b, 4, 5); } |
| 336 |
#[bench] fn mixed_radix_small_0007_32(b: &mut Bencher) { bench_mixed_radix_small(b, 7, 32); } |
| 337 |
|
| 338 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 339 |
/// for a given length, specific to the Mixed-Radix Double Butterfly algorithm |
| 340 |
fn bench_good_thomas_small(b: &mut Bencher, width: usize, height: usize) { |
| 341 |
|
| 342 |
let width_fft = plan_butterfly_fft(width); |
| 343 |
let height_fft = plan_butterfly_fft(height); |
| 344 |
|
| 345 |
let fft : Arc<Fft<_>> = Arc::new(GoodThomasAlgorithmSmall::new(width_fft, height_fft)); |
| 346 |
|
| 347 |
let mut signal = vec![Complex{re: 0_f32, im: 0_f32}; width * height]; |
| 348 |
let mut spectrum = signal.clone(); |
| 349 |
b.iter(|| {fft.process_with_scratch(&mut signal, &mut spectrum);} ); |
| 350 |
} |
| 351 |
|
| 352 |
#[bench] fn good_thomas_small_0002_3(b: &mut Bencher) { bench_good_thomas_small(b, 2, 3); } |
| 353 |
#[bench] fn good_thomas_small_0003_4(b: &mut Bencher) { bench_good_thomas_small(b, 3, 4); } |
| 354 |
#[bench] fn good_thomas_small_0004_5(b: &mut Bencher) { bench_good_thomas_small(b, 4, 5); } |
| 355 |
#[bench] fn good_thomas_small_0007_32(b: &mut Bencher) { bench_good_thomas_small(b, 7, 32); } |
| 356 |
|
| 357 |
|
| 358 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 359 |
/// for a given length, specific to Rader's algorithm |
| 360 |
fn bench_raders_scalar(b: &mut Bencher, len: usize) { |
| 361 |
|
| 362 |
let mut planner = rustfft::FftPlanner::new(); |
| 363 |
let inner_fft = planner.plan_fft_forward(len - 1); |
| 364 |
|
| 365 |
let fft : Arc<Fft<_>> = Arc::new(RadersAlgorithm::new(inner_fft)); |
| 366 |
|
| 367 |
let mut buffer = vec![Complex{re: 0_f32, im: 0_f32}; len]; |
| 368 |
let mut scratch = vec![Complex{re: 0_f32, im: 0_f32}; fft.get_inplace_scratch_len()]; |
| 369 |
b.iter(|| {fft.process_with_scratch(&mut buffer, &mut scratch);} ); |
| 370 |
} |
| 371 |
|
| 372 |
#[bench] fn raders_fft_scalar_prime_0005(b: &mut Bencher) { bench_raders_scalar(b, 5); } |
| 373 |
#[bench] fn raders_fft_scalar_prime_0017(b: &mut Bencher) { bench_raders_scalar(b, 17); } |
| 374 |
#[bench] fn raders_fft_scalar_prime_0149(b: &mut Bencher) { bench_raders_scalar(b, 149); } |
| 375 |
#[bench] fn raders_fft_scalar_prime_0151(b: &mut Bencher) { bench_raders_scalar(b, 151); } |
| 376 |
#[bench] fn raders_fft_scalar_prime_0251(b: &mut Bencher) { bench_raders_scalar(b, 251); } |
| 377 |
#[bench] fn raders_fft_scalar_prime_0257(b: &mut Bencher) { bench_raders_scalar(b, 257); } |
| 378 |
#[bench] fn raders_fft_scalar_prime_1009(b: &mut Bencher) { bench_raders_scalar(b, 1009); } |
| 379 |
#[bench] fn raders_fft_scalar_prime_2017(b: &mut Bencher) { bench_raders_scalar(b, 2017); } |
| 380 |
#[bench] fn raders_fft_scalar_prime_12289(b: &mut Bencher) { bench_raders_scalar(b, 12289); } |
| 381 |
#[bench] fn raders_fft_scalar_prime_18433(b: &mut Bencher) { bench_raders_scalar(b, 18433); } |
| 382 |
#[bench] fn raders_fft_scalar_prime_65521(b: &mut Bencher) { bench_raders_scalar(b, 65521); } |
| 383 |
#[bench] fn raders_fft_scalar_prime_65537(b: &mut Bencher) { bench_raders_scalar(b, 65537); } |
| 384 |
#[bench] fn raders_fft_scalar_prime_746483(b: &mut Bencher) { bench_raders_scalar(b,746483); } |
| 385 |
#[bench] fn raders_fft_scalar_prime_746497(b: &mut Bencher) { bench_raders_scalar(b,746497); } |
| 386 |
|
| 387 |
|
| 388 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 389 |
/// for a given length, specific to Bluestein's Algorithm |
| 390 |
fn bench_bluesteins_scalar_prime(b: &mut Bencher, len: usize) { |
| 391 |
let mut planner = rustfft::FftPlanner::new(); |
| 392 |
let inner_fft = planner.plan_fft_forward((len * 2 - 1).checked_next_power_of_two().unwrap()); |
| 393 |
let fft : Arc<Fft<f32>> = Arc::new(BluesteinsAlgorithm::new(len, inner_fft)); |
| 394 |
|
| 395 |
let mut buffer = vec![Zero::zero(); len]; |
| 396 |
let mut scratch = vec![Zero::zero(); fft.get_inplace_scratch_len()]; |
| 397 |
b.iter(|| { fft.process_with_scratch(&mut buffer, &mut scratch);} ); |
| 398 |
} |
| 399 |
|
| 400 |
#[bench] fn bench_bluesteins_scalar_prime_0005(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 5); } |
| 401 |
#[bench] fn bench_bluesteins_scalar_prime_0017(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 17); } |
| 402 |
#[bench] fn bench_bluesteins_scalar_prime_0149(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 149); } |
| 403 |
#[bench] fn bench_bluesteins_scalar_prime_0151(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 151); } |
| 404 |
#[bench] fn bench_bluesteins_scalar_prime_0251(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 251); } |
| 405 |
#[bench] fn bench_bluesteins_scalar_prime_0257(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 257); } |
| 406 |
#[bench] fn bench_bluesteins_scalar_prime_1009(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 1009); } |
| 407 |
#[bench] fn bench_bluesteins_scalar_prime_2017(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 2017); } |
| 408 |
#[bench] fn bench_bluesteins_scalar_prime_32767(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 32767); } |
| 409 |
#[bench] fn bench_bluesteins_scalar_prime_65521(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 65521); } |
| 410 |
#[bench] fn bench_bluesteins_scalar_prime_65537(b: &mut Bencher) { bench_bluesteins_scalar_prime(b, 65537); } |
| 411 |
#[bench] fn bench_bluesteins_scalar_prime_746483(b: &mut Bencher) { bench_bluesteins_scalar_prime(b,746483); } |
| 412 |
#[bench] fn bench_bluesteins_scalar_prime_746497(b: &mut Bencher) { bench_bluesteins_scalar_prime(b,746497); } |
| 413 |
|
| 414 |
|
| 415 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 416 |
/// for a given length, specific to Rader's algorithm |
| 417 |
fn bench_radix4(b: &mut Bencher, len: usize) { |
| 418 |
assert!(len % 4 == 0); |
| 419 |
|
| 420 |
let fft = Radix4::new(len, FftDirection::Forward); |
| 421 |
|
| 422 |
let mut signal = vec![Complex{re: 0_f32, im: 0_f32}; len]; |
| 423 |
let mut spectrum = signal.clone(); |
| 424 |
b.iter(|| {fft.process_outofplace_with_scratch(&mut signal, &mut spectrum, &mut []);} ); |
| 425 |
} |
| 426 |
|
| 427 |
#[bench] fn radix4_______64(b: &mut Bencher) { bench_radix4(b, 64); } |
| 428 |
#[bench] fn radix4______256(b: &mut Bencher) { bench_radix4(b, 256); } |
| 429 |
#[bench] fn radix4_____1024(b: &mut Bencher) { bench_radix4(b, 1024); } |
| 430 |
#[bench] fn radix4____65536(b: &mut Bencher) { bench_radix4(b, 65536); } |
| 431 |
#[bench] fn radix4__1048576(b: &mut Bencher) { bench_radix4(b, 1048576); } |
| 432 |
//#[bench] fn radix4_16777216(b: &mut Bencher) { bench_radix4(b, 16777216); } |
| 433 |
|
| 434 |
fn get_mixed_radix_power2(len: usize) -> Arc<dyn Fft<f32>> { |
| 435 |
match len { |
| 436 |
8 => Arc::new(Butterfly8::new( FftDirection::Forward)), |
| 437 |
16 => Arc::new(Butterfly16::new(FftDirection::Forward)), |
| 438 |
32 => Arc::new(Butterfly32::new(FftDirection::Forward)), |
| 439 |
_ => { |
| 440 |
let zeroes = len.trailing_zeros(); |
| 441 |
assert!(zeroes % 2 == 0); |
| 442 |
let half_zeroes = zeroes / 2; |
| 443 |
let inner = get_mixed_radix_power2(1 << half_zeroes); |
| 444 |
Arc::new(MixedRadix::new(Arc::clone(&inner), inner)) |
| 445 |
} |
| 446 |
} |
| 447 |
} |
| 448 |
|
| 449 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 450 |
/// for a given length, specific to Rader's algorithm |
| 451 |
fn bench_mixed_radix_power2(b: &mut Bencher, len: usize) { |
| 452 |
let fft = get_mixed_radix_power2(len); |
| 453 |
|
| 454 |
let mut buffer = vec![Zero::zero(); len]; |
| 455 |
let mut scratch = vec![Zero::zero(); fft.get_inplace_scratch_len()]; |
| 456 |
b.iter(|| { |
| 457 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 458 |
}); |
| 459 |
} |
| 460 |
|
| 461 |
#[bench] fn mixed_radix_power2__00000256(b: &mut Bencher) { bench_mixed_radix_power2(b, 256); } |
| 462 |
#[bench] fn mixed_radix_power2__00001024(b: &mut Bencher) { bench_mixed_radix_power2(b, 1024); } |
| 463 |
#[bench] fn mixed_radix_power2__00004096(b: &mut Bencher) { bench_mixed_radix_power2(b, 4096); } |
| 464 |
#[bench] fn mixed_radix_power2__00065536(b: &mut Bencher) { bench_mixed_radix_power2(b, 65536); } |
| 465 |
#[bench] fn mixed_radix_power2__01048576(b: &mut Bencher) { bench_mixed_radix_power2(b, 1048576); } |
| 466 |
#[bench] fn mixed_radix_power2__16777216(b: &mut Bencher) { bench_mixed_radix_power2(b, 16777216); } |
| 467 |
|
| 468 |
|
| 469 |
fn get_mixed_radix_inline_power2(len: usize) -> Arc<dyn Fft<f32>> { |
| 470 |
match len { |
| 471 |
8 => Arc::new(Butterfly8::new( FftDirection::Forward)), |
| 472 |
16 => Arc::new(Butterfly16::new(FftDirection::Forward)), |
| 473 |
32 => Arc::new(Butterfly32::new(FftDirection::Forward)), |
| 474 |
_ => { |
| 475 |
let zeroes = len.trailing_zeros(); |
| 476 |
assert!(zeroes % 2 == 0); |
| 477 |
let half_zeroes = zeroes / 2; |
| 478 |
let inner = get_mixed_radix_inline_power2(1 << half_zeroes); |
| 479 |
Arc::new(MixedRadix::new(Arc::clone(&inner), inner)) |
| 480 |
} |
| 481 |
} |
| 482 |
} |
| 483 |
|
| 484 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 485 |
/// for a given length, specific to Rader's algorithm |
| 486 |
fn bench_mixed_radix_inline_power2(b: &mut Bencher, len: usize) { |
| 487 |
let fft = get_mixed_radix_inline_power2(len); |
| 488 |
|
| 489 |
let mut buffer = vec![Zero::zero(); len]; |
| 490 |
let mut scratch = vec![Zero::zero(); fft.get_inplace_scratch_len()]; |
| 491 |
b.iter(|| { |
| 492 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 493 |
}); |
| 494 |
} |
| 495 |
|
| 496 |
#[bench] fn mixed_radix_power2_inline__00000256(b: &mut Bencher) { bench_mixed_radix_inline_power2(b, 256); } |
| 497 |
#[bench] fn mixed_radix_power2_inline__00001024(b: &mut Bencher) { bench_mixed_radix_inline_power2(b, 1024); } |
| 498 |
#[bench] fn mixed_radix_power2_inline__00004096(b: &mut Bencher) { bench_mixed_radix_inline_power2(b, 4096); } |
| 499 |
#[bench] fn mixed_radix_power2_inline__00065536(b: &mut Bencher) { bench_mixed_radix_inline_power2(b, 65536); } |
| 500 |
#[bench] fn mixed_radix_power2_inline__01048576(b: &mut Bencher) { bench_mixed_radix_inline_power2(b, 1048576); } |
| 501 |
#[bench] fn mixed_radix_power2_inline__16777216(b: &mut Bencher) { bench_mixed_radix_inline_power2(b, 16777216); } |
| 502 |
|
| 503 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 504 |
/// for a given length |
| 505 |
fn bench_butterfly32(b: &mut Bencher, len: usize) { |
| 506 |
|
| 507 |
let mut planner = rustfft::FftPlanner::new(); |
| 508 |
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len); |
| 509 |
|
| 510 |
let mut buffer = vec![Complex::zero(); len * 10]; |
| 511 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 512 |
b.iter(|| { fft.process_with_scratch(&mut buffer, &mut scratch); }); |
| 513 |
} |
| 514 |
|
| 515 |
#[bench] fn butterfly32_02(b: &mut Bencher) { bench_butterfly32(b, 2); } |
| 516 |
#[bench] fn butterfly32_03(b: &mut Bencher) { bench_butterfly32(b, 3); } |
| 517 |
#[bench] fn butterfly32_04(b: &mut Bencher) { bench_butterfly32(b, 4); } |
| 518 |
#[bench] fn butterfly32_05(b: &mut Bencher) { bench_butterfly32(b, 5); } |
| 519 |
#[bench] fn butterfly32_06(b: &mut Bencher) { bench_butterfly32(b, 6); } |
| 520 |
#[bench] fn butterfly32_07(b: &mut Bencher) { bench_butterfly32(b, 7); } |
| 521 |
#[bench] fn butterfly32_08(b: &mut Bencher) { bench_butterfly32(b, 8); } |
| 522 |
#[bench] fn butterfly32_09(b: &mut Bencher) { bench_butterfly32(b, 9); } |
| 523 |
#[bench] fn butterfly32_11(b: &mut Bencher) { bench_butterfly32(b, 11); } |
| 524 |
#[bench] fn butterfly32_12(b: &mut Bencher) { bench_butterfly32(b, 12); } |
| 525 |
#[bench] fn butterfly32_16(b: &mut Bencher) { bench_butterfly32(b, 16); } |
| 526 |
#[bench] fn butterfly32_24(b: &mut Bencher) { bench_butterfly32(b, 24); } |
| 527 |
#[bench] fn butterfly32_27(b: &mut Bencher) { bench_butterfly32(b, 27); } |
| 528 |
#[bench] fn butterfly32_32(b: &mut Bencher) { bench_butterfly32(b, 32); } |
| 529 |
#[bench] fn butterfly32_36(b: &mut Bencher) { bench_butterfly32(b, 36); } |
| 530 |
#[bench] fn butterfly32_48(b: &mut Bencher) { bench_butterfly32(b, 48); } |
| 531 |
#[bench] fn butterfly32_54(b: &mut Bencher) { bench_butterfly32(b, 54); } |
| 532 |
#[bench] fn butterfly32_64(b: &mut Bencher) { bench_butterfly32(b, 64); } |
| 533 |
#[bench] fn butterfly32_72(b: &mut Bencher) { bench_butterfly32(b, 72); } |
| 534 |
#[bench] fn butterfly32_128(b: &mut Bencher) { bench_butterfly32(b, 128); } |
| 535 |
#[bench] fn butterfly32_256(b: &mut Bencher) { bench_butterfly32(b, 256); } |
| 536 |
#[bench] fn butterfly32_512(b: &mut Bencher) { bench_butterfly32(b, 512); } |
| 537 |
|
| 538 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 539 |
/// for a given length |
| 540 |
fn bench_butterfly64(b: &mut Bencher, len: usize) { |
| 541 |
|
| 542 |
let mut planner = rustfft::FftPlanner::new(); |
| 543 |
let fft: Arc<dyn Fft<f64>> = planner.plan_fft_forward(len); |
| 544 |
|
| 545 |
let mut buffer = vec![Complex::zero(); len * 10]; |
| 546 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 547 |
b.iter(|| { fft.process_with_scratch(&mut buffer, &mut scratch); }); |
| 548 |
} |
| 549 |
|
| 550 |
#[bench] fn butterfly64_02(b: &mut Bencher) { bench_butterfly64(b, 2); } |
| 551 |
#[bench] fn butterfly64_03(b: &mut Bencher) { bench_butterfly64(b, 3); } |
| 552 |
#[bench] fn butterfly64_04(b: &mut Bencher) { bench_butterfly64(b, 4); } |
| 553 |
#[bench] fn butterfly64_05(b: &mut Bencher) { bench_butterfly64(b, 5); } |
| 554 |
#[bench] fn butterfly64_06(b: &mut Bencher) { bench_butterfly64(b, 6); } |
| 555 |
#[bench] fn butterfly64_07(b: &mut Bencher) { bench_butterfly64(b, 7); } |
| 556 |
#[bench] fn butterfly64_08(b: &mut Bencher) { bench_butterfly64(b, 8); } |
| 557 |
#[bench] fn butterfly64_09(b: &mut Bencher) { bench_butterfly64(b, 9); } |
| 558 |
#[bench] fn butterfly64_11(b: &mut Bencher) { bench_butterfly64(b, 11); } |
| 559 |
#[bench] fn butterfly64_12(b: &mut Bencher) { bench_butterfly64(b, 12); } |
| 560 |
#[bench] fn butterfly64_16(b: &mut Bencher) { bench_butterfly64(b, 16); } |
| 561 |
#[bench] fn butterfly64_18(b: &mut Bencher) { bench_butterfly64(b, 18); } |
| 562 |
#[bench] fn butterfly64_24(b: &mut Bencher) { bench_butterfly64(b, 24); } |
| 563 |
#[bench] fn butterfly64_27(b: &mut Bencher) { bench_butterfly64(b, 27); } |
| 564 |
#[bench] fn butterfly64_32(b: &mut Bencher) { bench_butterfly64(b, 32); } |
| 565 |
#[bench] fn butterfly64_36(b: &mut Bencher) { bench_butterfly64(b, 36); } |
| 566 |
#[bench] fn butterfly64_64(b: &mut Bencher) { bench_butterfly64(b, 64); } |
| 567 |
#[bench] fn butterfly64_128(b: &mut Bencher) { bench_butterfly64(b, 128); } |
| 568 |
#[bench] fn butterfly64_256(b: &mut Bencher) { bench_butterfly64(b, 256); } |
| 569 |
#[bench] fn butterfly64_512(b: &mut Bencher) { bench_butterfly64(b, 512); } |