| 1 |
#![feature(test)] |
| 2 |
extern crate rustfft; |
| 3 |
extern crate test; |
| 4 |
|
| 5 |
use rustfft::num_complex::Complex; |
| 6 |
use rustfft::num_traits::Zero; |
| 7 |
use rustfft::Fft; |
| 8 |
use std::sync::Arc; |
| 9 |
use test::Bencher; |
| 10 |
|
| 11 |
|
| 12 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 13 |
/// for a given length |
| 14 |
fn bench_planned_f32(b: &mut Bencher, len: usize) { |
| 15 |
let mut planner = rustfft::FftPlannerNeon::new().unwrap(); |
| 16 |
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len); |
| 17 |
assert_eq!(fft.len(), len); |
| 18 |
|
| 19 |
let mut buffer = vec![Complex::zero(); len]; |
| 20 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 21 |
b.iter(|| { |
| 22 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 23 |
}); |
| 24 |
} |
| 25 |
fn bench_planned_f64(b: &mut Bencher, len: usize) { |
| 26 |
let mut planner = rustfft::FftPlannerNeon::new().unwrap(); |
| 27 |
let fft: Arc<dyn Fft<f64>> = planner.plan_fft_forward(len); |
| 28 |
assert_eq!(fft.len(), len); |
| 29 |
|
| 30 |
let mut buffer = vec![Complex::zero(); len]; |
| 31 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 32 |
b.iter(|| { |
| 33 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 34 |
}); |
| 35 |
} |
| 36 |
|
| 37 |
/// Times just the FFT execution (not allocation and pre-calculation) |
| 38 |
/// for a given length. |
| 39 |
/// Run the fft on a 10*len vector, similar to how the butterflies are often used. |
| 40 |
fn bench_planned_multi_f32(b: &mut Bencher, len: usize) { |
| 41 |
let mut planner = rustfft::FftPlannerNeon::new().unwrap(); |
| 42 |
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len); |
| 43 |
|
| 44 |
let mut buffer = vec![Complex::zero(); len * 10]; |
| 45 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 46 |
b.iter(|| { |
| 47 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 48 |
}); |
| 49 |
} |
| 50 |
fn bench_planned_multi_f64(b: &mut Bencher, len: usize) { |
| 51 |
let mut planner = rustfft::FftPlannerNeon::new().unwrap(); |
| 52 |
let fft: Arc<dyn Fft<f64>> = planner.plan_fft_forward(len); |
| 53 |
|
| 54 |
let mut buffer = vec![Complex::zero(); len * 10]; |
| 55 |
let mut scratch = vec![Complex::zero(); fft.get_inplace_scratch_len()]; |
| 56 |
b.iter(|| { |
| 57 |
fft.process_with_scratch(&mut buffer, &mut scratch); |
| 58 |
}); |
| 59 |
} |
| 60 |
|
| 61 |
// All butterflies |
| 62 |
#[bench] fn neon_butterfly32_02(b: &mut Bencher) { bench_planned_multi_f32(b, 2);} |
| 63 |
#[bench] fn neon_butterfly32_03(b: &mut Bencher) { bench_planned_multi_f32(b, 3);} |
| 64 |
#[bench] fn neon_butterfly32_04(b: &mut Bencher) { bench_planned_multi_f32(b, 4);} |
| 65 |
#[bench] fn neon_butterfly32_05(b: &mut Bencher) { bench_planned_multi_f32(b, 5);} |
| 66 |
#[bench] fn neon_butterfly32_06(b: &mut Bencher) { bench_planned_multi_f32(b, 6);} |
| 67 |
#[bench] fn neon_butterfly32_07(b: &mut Bencher) { bench_planned_multi_f32(b, 7);} |
| 68 |
#[bench] fn neon_butterfly32_08(b: &mut Bencher) { bench_planned_multi_f32(b, 8);} |
| 69 |
#[bench] fn neon_butterfly32_09(b: &mut Bencher) { bench_planned_multi_f32(b, 9);} |
| 70 |
#[bench] fn neon_butterfly32_10(b: &mut Bencher) { bench_planned_multi_f32(b, 10);} |
| 71 |
#[bench] fn neon_butterfly32_11(b: &mut Bencher) { bench_planned_multi_f32(b, 11);} |
| 72 |
#[bench] fn neon_butterfly32_12(b: &mut Bencher) { bench_planned_multi_f32(b, 12);} |
| 73 |
#[bench] fn neon_butterfly32_13(b: &mut Bencher) { bench_planned_multi_f32(b, 13);} |
| 74 |
#[bench] fn neon_butterfly32_15(b: &mut Bencher) { bench_planned_multi_f32(b, 15);} |
| 75 |
#[bench] fn neon_butterfly32_16(b: &mut Bencher) { bench_planned_multi_f32(b, 16);} |
| 76 |
#[bench] fn neon_butterfly32_17(b: &mut Bencher) { bench_planned_multi_f32(b, 17);} |
| 77 |
#[bench] fn neon_butterfly32_19(b: &mut Bencher) { bench_planned_multi_f32(b, 19);} |
| 78 |
#[bench] fn neon_butterfly32_23(b: &mut Bencher) { bench_planned_multi_f32(b, 23);} |
| 79 |
#[bench] fn neon_butterfly32_29(b: &mut Bencher) { bench_planned_multi_f32(b, 29);} |
| 80 |
#[bench] fn neon_butterfly32_31(b: &mut Bencher) { bench_planned_multi_f32(b, 31);} |
| 81 |
#[bench] fn neon_butterfly32_32(b: &mut Bencher) { bench_planned_multi_f32(b, 32);} |
| 82 |
|
| 83 |
#[bench] fn neon_butterfly64_02(b: &mut Bencher) { bench_planned_multi_f64(b, 2);} |
| 84 |
#[bench] fn neon_butterfly64_03(b: &mut Bencher) { bench_planned_multi_f64(b, 3);} |
| 85 |
#[bench] fn neon_butterfly64_04(b: &mut Bencher) { bench_planned_multi_f64(b, 4);} |
| 86 |
#[bench] fn neon_butterfly64_05(b: &mut Bencher) { bench_planned_multi_f64(b, 5);} |
| 87 |
#[bench] fn neon_butterfly64_06(b: &mut Bencher) { bench_planned_multi_f64(b, 6);} |
| 88 |
#[bench] fn neon_butterfly64_07(b: &mut Bencher) { bench_planned_multi_f64(b, 7);} |
| 89 |
#[bench] fn neon_butterfly64_08(b: &mut Bencher) { bench_planned_multi_f64(b, 8);} |
| 90 |
#[bench] fn neon_butterfly64_09(b: &mut Bencher) { bench_planned_multi_f64(b, 9);} |
| 91 |
#[bench] fn neon_butterfly64_10(b: &mut Bencher) { bench_planned_multi_f64(b, 10);} |
| 92 |
#[bench] fn neon_butterfly64_11(b: &mut Bencher) { bench_planned_multi_f64(b, 11);} |
| 93 |
#[bench] fn neon_butterfly64_12(b: &mut Bencher) { bench_planned_multi_f64(b, 12);} |
| 94 |
#[bench] fn neon_butterfly64_13(b: &mut Bencher) { bench_planned_multi_f64(b, 13);} |
| 95 |
#[bench] fn neon_butterfly64_15(b: &mut Bencher) { bench_planned_multi_f64(b, 15);} |
| 96 |
#[bench] fn neon_butterfly64_16(b: &mut Bencher) { bench_planned_multi_f64(b, 16);} |
| 97 |
#[bench] fn neon_butterfly64_17(b: &mut Bencher) { bench_planned_multi_f64(b, 17);} |
| 98 |
#[bench] fn neon_butterfly64_19(b: &mut Bencher) { bench_planned_multi_f64(b, 19);} |
| 99 |
#[bench] fn neon_butterfly64_23(b: &mut Bencher) { bench_planned_multi_f64(b, 23);} |
| 100 |
#[bench] fn neon_butterfly64_29(b: &mut Bencher) { bench_planned_multi_f64(b, 29);} |
| 101 |
#[bench] fn neon_butterfly64_31(b: &mut Bencher) { bench_planned_multi_f64(b, 31);} |
| 102 |
#[bench] fn neon_butterfly64_32(b: &mut Bencher) { bench_planned_multi_f64(b, 32);} |
| 103 |
|
| 104 |
// Powers of 2 |
| 105 |
#[bench] fn neon_planned32_p2_00000064(b: &mut Bencher) { bench_planned_f32(b, 64); } |
| 106 |
#[bench] fn neon_planned32_p2_00000128(b: &mut Bencher) { bench_planned_f32(b, 128); } |
| 107 |
#[bench] fn neon_planned32_p2_00000256(b: &mut Bencher) { bench_planned_f32(b, 256); } |
| 108 |
#[bench] fn neon_planned32_p2_00000512(b: &mut Bencher) { bench_planned_f32(b, 512); } |
| 109 |
#[bench] fn neon_planned32_p2_00001024(b: &mut Bencher) { bench_planned_f32(b, 1024); } |
| 110 |
#[bench] fn neon_planned32_p2_00002048(b: &mut Bencher) { bench_planned_f32(b, 2048); } |
| 111 |
#[bench] fn neon_planned32_p2_00004096(b: &mut Bencher) { bench_planned_f32(b, 4096); } |
| 112 |
#[bench] fn neon_planned32_p2_00016384(b: &mut Bencher) { bench_planned_f32(b, 16384); } |
| 113 |
#[bench] fn neon_planned32_p2_00065536(b: &mut Bencher) { bench_planned_f32(b, 65536); } |
| 114 |
#[bench] fn neon_planned32_p2_01048576(b: &mut Bencher) { bench_planned_f32(b, 1048576); } |
| 115 |
|
| 116 |
#[bench] fn neon_planned64_p2_00000064(b: &mut Bencher) { bench_planned_f64(b, 64); } |
| 117 |
#[bench] fn neon_planned64_p2_00000128(b: &mut Bencher) { bench_planned_f64(b, 128); } |
| 118 |
#[bench] fn neon_planned64_p2_00000256(b: &mut Bencher) { bench_planned_f64(b, 256); } |
| 119 |
#[bench] fn neon_planned64_p2_00000512(b: &mut Bencher) { bench_planned_f64(b, 512); } |
| 120 |
#[bench] fn neon_planned64_p2_00001024(b: &mut Bencher) { bench_planned_f64(b, 1024); } |
| 121 |
#[bench] fn neon_planned64_p2_00002048(b: &mut Bencher) { bench_planned_f64(b, 2048); } |
| 122 |
#[bench] fn neon_planned64_p2_00004096(b: &mut Bencher) { bench_planned_f64(b, 4096); } |
| 123 |
#[bench] fn neon_planned64_p2_00016384(b: &mut Bencher) { bench_planned_f64(b, 16384); } |
| 124 |
#[bench] fn neon_planned64_p2_00065536(b: &mut Bencher) { bench_planned_f64(b, 65536); } |
| 125 |
#[bench] fn neon_planned64_p2_01048576(b: &mut Bencher) { bench_planned_f64(b, 1048576); } |
| 126 |
|
| 127 |
|
| 128 |
// Powers of 7 |
| 129 |
#[bench] fn neon_planned32_p7_00343(b: &mut Bencher) { bench_planned_f32(b, 343); } |
| 130 |
#[bench] fn neon_planned32_p7_02401(b: &mut Bencher) { bench_planned_f32(b, 2401); } |
| 131 |
#[bench] fn neon_planned32_p7_16807(b: &mut Bencher) { bench_planned_f32(b, 16807); } |
| 132 |
|
| 133 |
#[bench] fn neon_planned64_p7_00343(b: &mut Bencher) { bench_planned_f64(b, 343); } |
| 134 |
#[bench] fn neon_planned64_p7_02401(b: &mut Bencher) { bench_planned_f64(b, 2401); } |
| 135 |
#[bench] fn neon_planned64_p7_16807(b: &mut Bencher) { bench_planned_f64(b, 16807); } |
| 136 |
|
| 137 |
// Prime lengths |
| 138 |
#[bench] fn neon_planned32_prime_0149(b: &mut Bencher) { bench_planned_f32(b, 149); } |
| 139 |
#[bench] fn neon_planned32_prime_0151(b: &mut Bencher) { bench_planned_f32(b, 151); } |
| 140 |
#[bench] fn neon_planned32_prime_0251(b: &mut Bencher) { bench_planned_f32(b, 251); } |
| 141 |
#[bench] fn neon_planned32_prime_0257(b: &mut Bencher) { bench_planned_f32(b, 257); } |
| 142 |
#[bench] fn neon_planned32_prime_2017(b: &mut Bencher) { bench_planned_f32(b, 2017); } |
| 143 |
#[bench] fn neon_planned32_prime_2879(b: &mut Bencher) { bench_planned_f32(b, 2879); } |
| 144 |
#[bench] fn neon_planned32_prime_65521(b: &mut Bencher) { bench_planned_f32(b, 65521); } |
| 145 |
#[bench] fn neon_planned32_prime_746497(b: &mut Bencher) { bench_planned_f32(b,746497); } |
| 146 |
|
| 147 |
#[bench] fn neon_planned64_prime_0149(b: &mut Bencher) { bench_planned_f64(b, 149); } |
| 148 |
#[bench] fn neon_planned64_prime_0151(b: &mut Bencher) { bench_planned_f64(b, 151); } |
| 149 |
#[bench] fn neon_planned64_prime_0251(b: &mut Bencher) { bench_planned_f64(b, 251); } |
| 150 |
#[bench] fn neon_planned64_prime_0257(b: &mut Bencher) { bench_planned_f64(b, 257); } |
| 151 |
#[bench] fn neon_planned64_prime_2017(b: &mut Bencher) { bench_planned_f64(b, 2017); } |
| 152 |
#[bench] fn neon_planned64_prime_2879(b: &mut Bencher) { bench_planned_f64(b, 2879); } |
| 153 |
#[bench] fn neon_planned64_prime_65521(b: &mut Bencher) { bench_planned_f64(b, 65521); } |
| 154 |
#[bench] fn neon_planned64_prime_746497(b: &mut Bencher) { bench_planned_f64(b,746497); } |
| 155 |
|
| 156 |
// small mixed composites |
| 157 |
#[bench] fn neon_planned32_composite_000018(b: &mut Bencher) { bench_planned_f32(b, 00018); } |
| 158 |
#[bench] fn neon_planned32_composite_000360(b: &mut Bencher) { bench_planned_f32(b, 00360); } |
| 159 |
#[bench] fn neon_planned32_composite_001200(b: &mut Bencher) { bench_planned_f32(b, 01200); } |
| 160 |
#[bench] fn neon_planned32_composite_044100(b: &mut Bencher) { bench_planned_f32(b, 44100); } |
| 161 |
#[bench] fn neon_planned32_composite_048000(b: &mut Bencher) { bench_planned_f32(b, 48000); } |
| 162 |
#[bench] fn neon_planned32_composite_046656(b: &mut Bencher) { bench_planned_f32(b, 46656); } |
| 163 |
|
| 164 |
#[bench] fn neon_planned64_composite_000018(b: &mut Bencher) { bench_planned_f64(b, 00018); } |
| 165 |
#[bench] fn neon_planned64_composite_000360(b: &mut Bencher) { bench_planned_f64(b, 00360); } |
| 166 |
#[bench] fn neon_planned64_composite_001200(b: &mut Bencher) { bench_planned_f64(b, 01200); } |
| 167 |
#[bench] fn neon_planned64_composite_044100(b: &mut Bencher) { bench_planned_f64(b, 44100); } |
| 168 |
#[bench] fn neon_planned64_composite_048000(b: &mut Bencher) { bench_planned_f64(b, 48000); } |
| 169 |
#[bench] fn neon_planned64_composite_046656(b: &mut Bencher) { bench_planned_f64(b, 46656); } |
| 170 |
|
| 171 |
|
| 172 |
|