Skip to content

Commit 3dd012c

Browse files
HEnquistejmahler
HEnquist
and
ejmahler
authored
Neon (#78)
* WIP neon support * All except prime butterflies * Add prime butterflies * Enable prime butterflies in planner * Add more benches * Tweaks, cleanup * Format * Minor tweaks and some cleanup * Clean up docs * Format * Comparison bench and plotting tool for neon * Add instructions to asmtest example * Clean up asmtest example * Clean up, fix typo * Use easier numbers in test, to avoid failing on rounding errors * Format * Typo in avx fallback planner, ci for neon * Typo in ci name * Changes after review * Format * Renamed the "neon" feature to "neon-nightly" Co-authored-by: Elliott Mahler <[email protected]>
1 parent f80d7dd commit 3dd012c

20 files changed

+12650
-107
lines changed

‎.github/workflows/run_test.yml

Lines changed: 29 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -89,6 +89,35 @@ jobs:
8989
use-cross: true
9090
args: --release --target aarch64-unknown-linux-gnu
9191

92+
check_arm64_neon:
93+
name: Check and test Linux arm 64bit with neon
94+
runs-on: ubuntu-latest
95+
steps:
96+
- name: Checkout sources
97+
uses: actions/checkout@v2
98+
99+
- name: Install nightly toolchain
100+
uses: actions-rs/toolchain@v1
101+
with:
102+
profile: minimal
103+
toolchain: nightly
104+
target: aarch64-unknown-linux-gnu
105+
override: true
106+
107+
- name: Run cargo check
108+
uses: actions-rs/cargo@v1
109+
with:
110+
command: check
111+
use-cross: true
112+
args: --features neon-nightly --target aarch64-unknown-linux-gnu
113+
114+
- name: Run cargo test for arm
115+
uses: actions-rs/cargo@v1
116+
with:
117+
command: test
118+
use-cross: true
119+
args: --release --features neon-nightly --target aarch64-unknown-linux-gnu
120+
92121
check_x86:
93122
name: Check and test Linux x86 32bit
94123
runs-on: ubuntu-latest

‎Cargo.toml

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -21,8 +21,11 @@ default = ["avx", "sse"]
2121
# If both are enabled, RustFFT will use AVX if the CPU supports it. If not, it will check for SSE4.1.
2222
# If neither instruction set is available, it will fall back to the scalar code.
2323
# On every other platform, these features do nothing, and RustFFT will behave like they are not set.
24+
#
25+
# On AArch64, the "neon-nightly" feature enables compilation of Neon-accelerated code. It requires a nightly compiler, and is disabled by default.
2426
avx = []
2527
sse = []
28+
neon-nightly = []
2629

2730

2831
[dependencies]

‎README.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -41,6 +41,8 @@ Disabling them reduces compile time and binary size.
4141

4242
On other platform than x86_64, these features do nothing and RustFFT will behave like they are not set.
4343

44+
On AArch64, the `neon-nightly` feature enables compilation of Neon-accelerated code. It requires a nightly compiler, and is disabled by default. Be warned that new nightly versions may break RustFFT's Neon support.
45+
4446
## Stability/Future Breaking Changes
4547

4648
Version 5.0 contains several breaking API changes. In the interest of stability, we're committing to making no more breaking changes for 3 years, aka until 2024.

‎benches/bench_check_neon_2to1024.rs

Lines changed: 168 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,168 @@
1+
#![feature(test)]
2+
extern crate rustfft;
3+
extern crate test;
4+
5+
use paste::paste;
6+
use rustfft::num_complex::Complex;
7+
use rustfft::num_traits::Zero;
8+
use rustfft::Fft;
9+
use std::sync::Arc;
10+
use test::Bencher;
11+
12+
// Make fft using planner
13+
fn bench_planned_32(b: &mut Bencher, len: usize) {
14+
let mut planner = rustfft::FftPlannerNeon::new().unwrap();
15+
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len);
16+
17+
let mut buffer: Vec<Complex<f32>> = vec![Complex::zero(); len];
18+
let mut scratch: Vec<Complex<f32>> = vec![Complex::zero(); fft.get_inplace_scratch_len()];
19+
b.iter(|| {
20+
fft.process_with_scratch(&mut buffer, &mut scratch);
21+
});
22+
}
23+
24+
// Make fft using planner
25+
fn bench_planned_64(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+
29+
let mut buffer: Vec<Complex<f64>> = vec![Complex::zero(); len];
30+
let mut scratch: Vec<Complex<f64>> = vec![Complex::zero(); fft.get_inplace_scratch_len()];
31+
b.iter(|| {
32+
fft.process_with_scratch(&mut buffer, &mut scratch);
33+
});
34+
}
35+
36+
// Create benches using functions taking one argument
37+
macro_rules! make_benches {
38+
($name:ident, $fname:ident, { $($len:literal),* }) => {
39+
paste! {
40+
$(
41+
#[bench]
42+
fn [<bench_ $name _f32_ $len>](b: &mut Bencher) {
43+
[<bench_ $fname _32>](b, $len);
44+
}
45+
46+
#[bench]
47+
fn [<bench_ $name _f64_ $len>](b: &mut Bencher) {
48+
[<bench_ $fname _64>](b, $len);
49+
}
50+
)*
51+
}
52+
}
53+
}
54+
55+
make_benches!(from2to1024, planned, {2, 3, 4, 5, 6, 7, 8, 9 });
56+
make_benches!(from2to1024, planned, {10, 11, 12, 13, 14, 15, 16, 17, 18, 19 });
57+
make_benches!(from2to1024, planned, {20, 21, 22, 23, 24, 25, 26, 27, 28, 29 });
58+
make_benches!(from2to1024, planned, {30, 31, 32, 33, 34, 35, 36, 37, 38, 39 });
59+
make_benches!(from2to1024, planned, {40, 41, 42, 43, 44, 45, 46, 47, 48, 49 });
60+
make_benches!(from2to1024, planned, {50, 51, 52, 53, 54, 55, 56, 57, 58, 59 });
61+
make_benches!(from2to1024, planned, {60, 61, 62, 63, 64, 65, 66, 67, 68, 69 });
62+
make_benches!(from2to1024, planned, {70, 71, 72, 73, 74, 75, 76, 77, 78, 79 });
63+
make_benches!(from2to1024, planned, {80, 81, 82, 83, 84, 85, 86, 87, 88, 89 });
64+
make_benches!(from2to1024, planned, {90, 91, 92, 93, 94, 95, 96, 97, 98, 99 });
65+
66+
make_benches!(from2to1024, planned, {100, 101, 102, 103, 104, 105, 106, 107, 108, 109 });
67+
make_benches!(from2to1024, planned, {110, 111, 112, 113, 114, 115, 116, 117, 118, 119 });
68+
make_benches!(from2to1024, planned, {120, 121, 122, 123, 124, 125, 126, 127, 128, 129 });
69+
make_benches!(from2to1024, planned, {130, 131, 132, 133, 134, 135, 136, 137, 138, 139 });
70+
make_benches!(from2to1024, planned, {140, 141, 142, 143, 144, 145, 146, 147, 148, 149 });
71+
make_benches!(from2to1024, planned, {150, 151, 152, 153, 154, 155, 156, 157, 158, 159 });
72+
make_benches!(from2to1024, planned, {160, 161, 162, 163, 164, 165, 166, 167, 168, 169 });
73+
make_benches!(from2to1024, planned, {170, 171, 172, 173, 174, 175, 176, 177, 178, 179 });
74+
make_benches!(from2to1024, planned, {180, 181, 182, 183, 184, 185, 186, 187, 188, 189 });
75+
make_benches!(from2to1024, planned, {190, 191, 192, 193, 194, 195, 196, 197, 198, 199 });
76+
/*
77+
make_benches!(from2to1024, planned, {200, 201, 202, 203, 204, 205, 206, 207, 208, 209 });
78+
make_benches!(from2to1024, planned, {210, 211, 212, 213, 214, 215, 216, 217, 218, 219 });
79+
make_benches!(from2to1024, planned, {220, 221, 222, 223, 224, 225, 226, 227, 228, 229 });
80+
make_benches!(from2to1024, planned, {230, 231, 232, 233, 234, 235, 236, 237, 238, 239 });
81+
make_benches!(from2to1024, planned, {240, 241, 242, 243, 244, 245, 246, 247, 248, 249 });
82+
make_benches!(from2to1024, planned, {250, 251, 252, 253, 254, 255, 256, 257, 258, 259 });
83+
make_benches!(from2to1024, planned, {260, 261, 262, 263, 264, 265, 266, 267, 268, 269 });
84+
make_benches!(from2to1024, planned, {270, 271, 272, 273, 274, 275, 276, 277, 278, 279 });
85+
make_benches!(from2to1024, planned, {280, 281, 282, 283, 284, 285, 286, 287, 288, 289 });
86+
make_benches!(from2to1024, planned, {290, 291, 292, 293, 294, 295, 296, 297, 298, 299 });
87+
88+
make_benches!(from2to1024, planned, {300, 301, 302, 303, 304, 305, 306, 307, 308, 309 });
89+
make_benches!(from2to1024, planned, {310, 311, 312, 313, 314, 315, 316, 317, 318, 319 });
90+
make_benches!(from2to1024, planned, {320, 321, 322, 323, 324, 325, 326, 327, 328, 329 });
91+
make_benches!(from2to1024, planned, {330, 331, 332, 333, 334, 335, 336, 337, 338, 339 });
92+
make_benches!(from2to1024, planned, {340, 341, 342, 343, 344, 345, 346, 347, 348, 349 });
93+
make_benches!(from2to1024, planned, {350, 351, 352, 353, 354, 355, 356, 357, 358, 359 });
94+
make_benches!(from2to1024, planned, {360, 361, 362, 363, 364, 365, 366, 367, 368, 369 });
95+
make_benches!(from2to1024, planned, {370, 371, 372, 373, 374, 375, 376, 377, 378, 379 });
96+
make_benches!(from2to1024, planned, {380, 381, 382, 383, 384, 385, 386, 387, 388, 389 });
97+
make_benches!(from2to1024, planned, {390, 391, 392, 393, 394, 395, 396, 397, 398, 399 });
98+
99+
make_benches!(from2to1024, planned, {400, 401, 402, 403, 404, 405, 406, 407, 408, 409 });
100+
make_benches!(from2to1024, planned, {410, 411, 412, 413, 414, 415, 416, 417, 418, 419 });
101+
make_benches!(from2to1024, planned, {420, 421, 422, 423, 424, 425, 426, 427, 428, 429 });
102+
make_benches!(from2to1024, planned, {430, 431, 432, 433, 434, 435, 436, 437, 438, 439 });
103+
make_benches!(from2to1024, planned, {440, 441, 442, 443, 444, 445, 446, 447, 448, 449 });
104+
make_benches!(from2to1024, planned, {450, 451, 452, 453, 454, 455, 456, 457, 458, 459 });
105+
make_benches!(from2to1024, planned, {460, 461, 462, 463, 464, 465, 466, 467, 468, 469 });
106+
make_benches!(from2to1024, planned, {470, 471, 472, 473, 474, 475, 476, 477, 478, 479 });
107+
make_benches!(from2to1024, planned, {480, 481, 482, 483, 484, 485, 486, 487, 488, 489 });
108+
make_benches!(from2to1024, planned, {490, 491, 492, 493, 494, 495, 496, 497, 498, 499 });
109+
110+
make_benches!(from2to1024, planned, {500, 501, 502, 503, 504, 505, 506, 507, 508, 509 });
111+
make_benches!(from2to1024, planned, {510, 511, 512, 513, 514, 515, 516, 517, 518, 519 });
112+
make_benches!(from2to1024, planned, {520, 521, 522, 523, 524, 525, 526, 527, 528, 529 });
113+
make_benches!(from2to1024, planned, {530, 531, 532, 533, 534, 535, 536, 537, 538, 539 });
114+
make_benches!(from2to1024, planned, {540, 541, 542, 543, 544, 545, 546, 547, 548, 549 });
115+
make_benches!(from2to1024, planned, {550, 551, 552, 553, 554, 555, 556, 557, 558, 559 });
116+
make_benches!(from2to1024, planned, {560, 561, 562, 563, 564, 565, 566, 567, 568, 569 });
117+
make_benches!(from2to1024, planned, {570, 571, 572, 573, 574, 575, 576, 577, 578, 579 });
118+
make_benches!(from2to1024, planned, {580, 581, 582, 583, 584, 585, 586, 587, 588, 589 });
119+
make_benches!(from2to1024, planned, {590, 591, 592, 593, 594, 595, 596, 597, 598, 599 });
120+
121+
make_benches!(from2to1024, planned, {600, 601, 602, 603, 604, 605, 606, 607, 608, 609 });
122+
make_benches!(from2to1024, planned, {610, 611, 612, 613, 614, 615, 616, 617, 618, 619 });
123+
make_benches!(from2to1024, planned, {620, 621, 622, 623, 624, 625, 626, 627, 628, 629 });
124+
make_benches!(from2to1024, planned, {630, 631, 632, 633, 634, 635, 636, 637, 638, 639 });
125+
make_benches!(from2to1024, planned, {640, 641, 642, 643, 644, 645, 646, 647, 648, 649 });
126+
make_benches!(from2to1024, planned, {650, 651, 652, 653, 654, 655, 656, 657, 658, 659 });
127+
make_benches!(from2to1024, planned, {660, 661, 662, 663, 664, 665, 666, 667, 668, 669 });
128+
make_benches!(from2to1024, planned, {670, 671, 672, 673, 674, 675, 676, 677, 678, 679 });
129+
make_benches!(from2to1024, planned, {680, 681, 682, 683, 684, 685, 686, 687, 688, 689 });
130+
make_benches!(from2to1024, planned, {690, 691, 692, 693, 694, 695, 696, 697, 698, 699 });
131+
132+
make_benches!(from2to1024, planned, {700, 701, 702, 703, 704, 705, 706, 707, 708, 709 });
133+
make_benches!(from2to1024, planned, {710, 711, 712, 713, 714, 715, 716, 717, 718, 719 });
134+
make_benches!(from2to1024, planned, {720, 721, 722, 723, 724, 725, 726, 727, 728, 729 });
135+
make_benches!(from2to1024, planned, {730, 731, 732, 733, 734, 735, 736, 737, 738, 739 });
136+
make_benches!(from2to1024, planned, {740, 741, 742, 743, 744, 745, 746, 747, 748, 749 });
137+
make_benches!(from2to1024, planned, {750, 751, 752, 753, 754, 755, 756, 757, 758, 759 });
138+
make_benches!(from2to1024, planned, {760, 761, 762, 763, 764, 765, 766, 767, 768, 769 });
139+
make_benches!(from2to1024, planned, {770, 771, 772, 773, 774, 775, 776, 777, 778, 779 });
140+
make_benches!(from2to1024, planned, {780, 781, 782, 783, 784, 785, 786, 787, 788, 789 });
141+
make_benches!(from2to1024, planned, {790, 791, 792, 793, 794, 795, 796, 797, 798, 799 });
142+
143+
make_benches!(from2to1024, planned, {800, 801, 802, 803, 804, 805, 806, 807, 808, 809 });
144+
make_benches!(from2to1024, planned, {810, 811, 812, 813, 814, 815, 816, 817, 818, 819 });
145+
make_benches!(from2to1024, planned, {820, 821, 822, 823, 824, 825, 826, 827, 828, 829 });
146+
make_benches!(from2to1024, planned, {830, 831, 832, 833, 834, 835, 836, 837, 838, 839 });
147+
make_benches!(from2to1024, planned, {840, 841, 842, 843, 844, 845, 846, 847, 848, 849 });
148+
make_benches!(from2to1024, planned, {850, 851, 852, 853, 854, 855, 856, 857, 858, 859 });
149+
make_benches!(from2to1024, planned, {860, 861, 862, 863, 864, 865, 866, 867, 868, 869 });
150+
make_benches!(from2to1024, planned, {870, 871, 872, 873, 874, 875, 876, 877, 878, 879 });
151+
make_benches!(from2to1024, planned, {880, 881, 882, 883, 884, 885, 886, 887, 888, 889 });
152+
make_benches!(from2to1024, planned, {890, 891, 892, 893, 894, 895, 896, 897, 898, 899 });
153+
154+
make_benches!(from2to1024, planned, {900, 901, 902, 903, 904, 905, 906, 907, 908, 909 });
155+
make_benches!(from2to1024, planned, {910, 911, 912, 913, 914, 915, 916, 917, 918, 919 });
156+
make_benches!(from2to1024, planned, {920, 921, 922, 923, 924, 925, 926, 927, 928, 929 });
157+
make_benches!(from2to1024, planned, {930, 931, 932, 933, 934, 935, 936, 937, 938, 939 });
158+
make_benches!(from2to1024, planned, {940, 941, 942, 943, 944, 945, 946, 947, 948, 949 });
159+
make_benches!(from2to1024, planned, {950, 951, 952, 953, 954, 955, 956, 957, 958, 959 });
160+
make_benches!(from2to1024, planned, {960, 961, 962, 963, 964, 965, 966, 967, 968, 969 });
161+
make_benches!(from2to1024, planned, {970, 971, 972, 973, 974, 975, 976, 977, 978, 979 });
162+
make_benches!(from2to1024, planned, {980, 981, 982, 983, 984, 985, 986, 987, 988, 989 });
163+
make_benches!(from2to1024, planned, {990, 991, 992, 993, 994, 995, 996, 997, 998, 999 });
164+
165+
make_benches!(from2to1024, planned, {1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007, 1008, 1009 });
166+
make_benches!(from2to1024, planned, {1010, 1011, 1012, 1013, 1014, 1015, 1016, 1017, 1018, 1019 });
167+
make_benches!(from2to1024, planned, {1020, 1021, 1022, 1023, 1024 });
168+
*/

‎benches/bench_compare_scalar_neon.rs

Lines changed: 91 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,91 @@
1+
#![feature(test)]
2+
extern crate rustfft;
3+
extern crate test;
4+
5+
use paste::paste;
6+
use rustfft::num_complex::Complex;
7+
use rustfft::num_traits::Zero;
8+
use rustfft::Fft;
9+
use std::sync::Arc;
10+
use test::Bencher;
11+
12+
// Make fft using scalar planner
13+
fn bench_scalar_32(b: &mut Bencher, len: usize) {
14+
let mut planner = rustfft::FftPlannerScalar::new();
15+
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len);
16+
17+
let mut buffer: Vec<Complex<f32>> = vec![Complex::zero(); len];
18+
let mut scratch: Vec<Complex<f32>> = vec![Complex::zero(); fft.get_inplace_scratch_len()];
19+
b.iter(|| {
20+
fft.process_with_scratch(&mut buffer, &mut scratch);
21+
});
22+
}
23+
24+
// Make fft using scalar planner
25+
fn bench_scalar_64(b: &mut Bencher, len: usize) {
26+
let mut planner = rustfft::FftPlannerScalar::new();
27+
let fft: Arc<dyn Fft<f64>> = planner.plan_fft_forward(len);
28+
29+
let mut buffer: Vec<Complex<f64>> = vec![Complex::zero(); len];
30+
let mut scratch: Vec<Complex<f64>> = vec![Complex::zero(); fft.get_inplace_scratch_len()];
31+
b.iter(|| {
32+
fft.process_with_scratch(&mut buffer, &mut scratch);
33+
});
34+
}
35+
36+
// Make fft using sse planner
37+
fn bench_neon_32(b: &mut Bencher, len: usize) {
38+
let mut planner = rustfft::FftPlannerNeon::new().unwrap();
39+
let fft: Arc<dyn Fft<f32>> = planner.plan_fft_forward(len);
40+
41+
let mut buffer: Vec<Complex<f32>> = vec![Complex::zero(); len];
42+
let mut scratch: Vec<Complex<f32>> = vec![Complex::zero(); fft.get_inplace_scratch_len()];
43+
b.iter(|| {
44+
fft.process_with_scratch(&mut buffer, &mut scratch);
45+
});
46+
}
47+
48+
// Make fft using sse planner
49+
fn bench_neon_64(b: &mut Bencher, len: usize) {
50+
let mut planner = rustfft::FftPlannerNeon::new().unwrap();
51+
let fft: Arc<dyn Fft<f64>> = planner.plan_fft_forward(len);
52+
53+
let mut buffer: Vec<Complex<f64>> = vec![Complex::zero(); len];
54+
let mut scratch: Vec<Complex<f64>> = vec![Complex::zero(); fft.get_inplace_scratch_len()];
55+
b.iter(|| {
56+
fft.process_with_scratch(&mut buffer, &mut scratch);
57+
});
58+
}
59+
60+
61+
// Create benches using functions taking one argument
62+
macro_rules! make_benches {
63+
($name:ident, { $($len:literal),* }) => {
64+
paste! {
65+
$(
66+
#[bench]
67+
fn [<$name _ $len _f32_scalar>](b: &mut Bencher) {
68+
[<bench_scalar_32>](b, $len);
69+
}
70+
71+
#[bench]
72+
fn [<$name _ $len _f64_scalar>](b: &mut Bencher) {
73+
[<bench_scalar_64>](b, $len);
74+
}
75+
76+
#[bench]
77+
fn [<$name _ $len _f32_neon>](b: &mut Bencher) {
78+
[<bench_neon_32>](b, $len);
79+
}
80+
81+
#[bench]
82+
fn [<$name _ $len _f64_neon>](b: &mut Bencher) {
83+
[<bench_neon_64>](b, $len);
84+
}
85+
)*
86+
}
87+
}
88+
}
89+
90+
make_benches!(neoncomparison, {4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072});
91+
make_benches!(neoncomparison, { 262144, 524288, 1048576, 2097152, 4194304 });

0 commit comments

Comments
 (0)