bert/adventofcode
1
0
Fork 0
mirror of https://github.com/bertptrs/adventofcode.git synced 2025-12-25 21:00:31 +01:00
Code Issues Projects Releases Wiki Activity

Flatten DP buffers

Browse Source 
Going from 2d to 1d arrays improves cache locality and vectorization,
resulting in a roughly 40% speedup for part 2.
This commit is contained in:
Bert Peters
2023-12-15 17:39:59 +01:00
parent 673e8184ed
commit 682e6f06b8
1 changed files with 14 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
2023/src/day12.rs
View File

@@ -15,33 +15,32 @@ fn number_ways(line: &[u8], groups: &[u8]) -> u64 {
return 0;
};
let mut next = vec![vec![0; max_group as usize + 1]; groups.len() + 1];
let group_stride = max_group as usize + 1;
let mut next = vec![0; (groups.len() + 1) * group_stride];
let mut cur = next.clone();
cur[0][0] = 1;
cur[0] = 1;
for &c in line {
for entry in &mut next {
entry.fill(0);
}
next.fill(0);
for group_pos in 0..=groups.len() {
for (group_pos, positions) in cur.chunks_exact(group_stride).enumerate() {
let group = *groups.get(group_pos).unwrap_or(&0);
for cur_group in 0..=max_group {
let ways = cur[group_pos][cur_group as usize];
for (cur_group, &ways) in positions.iter().enumerate() {
if ways == 0 {
continue;
}
// Either defective or maybe defective
if c != b'.' && cur_group < group {
next[group_pos][cur_group as usize + 1] += ways;
if c != b'.' && cur_group < usize::from(group) {
next[group_pos * group_stride + cur_group as usize + 1] += ways;
}
if c != b'#' {
if cur_group == 0 {
next[group_pos][0] += ways;
} else if group == cur_group {
next[group_pos + 1][0] += ways;
next[group_pos * group_stride] += ways;
} else if usize::from(group) == cur_group {
next[(group_pos + 1) * group_stride] += ways;
}
}
}
@@ -50,7 +49,8 @@ fn number_ways(line: &[u8], groups: &[u8]) -> u64 {
mem::swap(&mut cur, &mut next);
}
cur[groups.len()][0] + cur[groups.len() - 1][groups[groups.len() - 1] as usize]
cur[groups.len() * group_stride]
+ cur[(groups.len() - 1) * group_stride + groups[groups.len() - 1] as usize]
}
fn parse_lines(i: &[u8]) -> IResult<&[u8], Vec<(&[u8], Vec<u8>)>> {