Open simonlindholm opened 3 years ago
Apparently the suffix array is a bit slow and has resulted in TLEs.
One suggestion that came up in the AC discord server (https://discord.com/channels/555883512952258563/559415184826040340/809806400070483989) is splitting
for (int i = n; i--;) sa[--ws[x[y[i]]]] = y[i];
into multiple loops, e.g.
for (int i = n; i--;) tmp[i] = x[y[i]]; for (int i = n; i--;) tmp[i] = --ws[tmp[i]]; for (int i = n; i--;) sa[tmp[i]] = y[i];
which seems to help on random data. It's not obvious why, but I'm speculating that it's due to load mispredictions.
The suffix array seems to perform worse on repetitive data though (e.g. s[i] = i % 10), and the change makes performance worse in that case. :(
I'm also having trouble with this implementation; it is too slow to pass several CSES problems, such as https://cses.fi/problemset/task/1111/ or https://cses.fi/problemset/task/1110.
Apparently the suffix array is a bit slow and has resulted in TLEs.
One suggestion that came up in the AC discord server (https://discord.com/channels/555883512952258563/559415184826040340/809806400070483989) is splitting
into multiple loops, e.g.
which seems to help on random data. It's not obvious why, but I'm speculating that it's due to load mispredictions.
The suffix array seems to perform worse on repetitive data though (e.g. s[i] = i % 10), and the change makes performance worse in that case. :(