Faster SIMD alternative

Author: maneatingape

README.md

@@ -53,7 +53,7 @@ Place input files in `input/yearYYYY/dayDD.txt` including leading zeroes. For ex
 ## Performance
 
 Benchmarks are measured using the built-in `cargo bench` tool run on an [Apple M2 Max][apple-link].
-All 225 solutions from 2023 to to 2015 complete sequentially in **625 milliseconds**.
+All 225 solutions from 2023 to 2015 complete sequentially in **622 milliseconds**.
 Interestingly 84% of the total time is spent on just 9 solutions.
 Performance is reasonable even on older hardware, for example a 2011 MacBook Pro with an
 [Intel i7-2720QM][intel-link] processor takes 3.5 seconds to run the same 225 solutions.
@@ -66,7 +66,7 @@ Performance is reasonable even on older hardware, for example a 2011 MacBook Pro
 | [2022](#2022) | 10 |
 | [2021](#2021) | 10 |
 | [2020](#2020) | 286 |
-| [2019](#2019) | 19 |
+| [2019](#2019) | 16 |
 | [2018](#2018) | 36 |
 | [2017](#2017) | 102 |
 | [2016](#2016) | 133 |
@@ -221,7 +221,7 @@ Performance is reasonable even on older hardware, for example a 2011 MacBook Pro
 | 13 | [Care Package](https://adventofcode.com/2019/day/13) | [Source](src/year2019/day13.rs) | 2513 |
 | 14 | [Space Stoichiometry](https://adventofcode.com/2019/day/14) | [Source](src/year2019/day14.rs) | 17 |
 | 15 | [Oxygen System](https://adventofcode.com/2019/day/15) | [Source](src/year2019/day15.rs) | 361 |
-| 16 | [Flawed Frequency Transmission](https://adventofcode.com/2019/day/16) | [Source](src/year2019/day16.rs) | 4124 |
+| 16 | [Flawed Frequency Transmission](https://adventofcode.com/2019/day/16) | [Source](src/year2019/day16.rs) | 1960 |
 | 17 | [Set and Forget](https://adventofcode.com/2019/day/17) | [Source](src/year2019/day17.rs) | 341 |
 | 18 | [Many-Worlds Interpretation](https://adventofcode.com/2019/day/18) | [Source](src/year2019/day18.rs) | 1077 |
 | 19 | [Tractor Beam](https://adventofcode.com/2019/day/19) | [Source](src/year2019/day19.rs) | 674 |

docs/pie-2019.svg

@@ -148,6 +148,11 @@ pub fn part2(input: &[u8]) -> usize {
     let upper = size * 10_000;
     assert!(lower <= start && start < upper);
 
+    compute(&digits, size, start, upper)
+}
+
+#[cfg(not(feature = "simd"))]
+fn compute(digits: &[usize], size: usize, start: usize, upper: usize) -> usize {
     let mut coefficients = [0; 8];
     let mut result = [0; 8];
 
@@ -163,6 +168,30 @@ pub fn part2(input: &[u8]) -> usize {
     result.fold_decimal()
 }
 
+#[cfg(feature = "simd")]
+fn compute(digits: &[usize], size: usize, start: usize, upper: usize) -> usize {
+    use std::simd::Mask;
+    use std::simd::Simd;
+
+    let mask: Mask<i32, 8> = Mask::from_bitmask(1);
+    let tens: Simd<u32, 8> = Simd::splat(10);
+
+    let mut coefficients: Simd<u32, 8> = Simd::splat(0);
+    let mut result: Simd<u32, 8> = Simd::splat(0);
+
+    for (k, index) in (start..upper).enumerate() {
+        coefficients = mask.select(
+            Simd::splat(binomial_mod_10(k + 99, k) as u32),
+            coefficients.rotate_elements_right::<1>(),
+        );
+
+        let next = Simd::splat(digits[index % size] as u32);
+        result += next * coefficients;
+    }
+
+    (result % tens).to_array().fold_decimal() as usize
+}
+
 /// Computes C(n, k) % 2
 ///
 /// This collapses to a special case of a product of only 4 possible values: