From e1ebdb879053f1267245110cad9b33849b3d74f3 Mon Sep 17 00:00:00 2001 From: Huon Wilson Date: Wed, 15 Jan 2014 11:58:02 +1100 Subject: [PATCH] std::trie: optimise insert slightly. This reduces the number of moves/memcpy's we do, which makes insert faster, especially in cases of keys with long equal prefixes (the _low_bits tests): Before: bench_insert_large ... bench: 553966 ns/iter (+/- 64050) bench_insert_large_low_bits ... bench: 1048151 ns/iter (+/- 92484) bench_insert_small ... bench: 168840 ns/iter (+/- 22410) bench_insert_small_low_bits ... bench: 185069 ns/iter (+/- 38332) After: bench_insert_large ... bench: 422132 ns/iter (+/- 35112) bench_insert_large_low_bits ... bench: 339083 ns/iter (+/- 34421) bench_insert_small ... bench: 134539 ns/iter (+/- 15254) bench_insert_small_low_bits ... bench: 88775 ns/iter (+/- 5746) --- src/libstd/trie.rs | 68 ++++++++++++++++++++++++---------------------- 1 file changed, 35 insertions(+), 33 deletions(-) diff --git a/src/libstd/trie.rs b/src/libstd/trie.rs index 7987e25a2f3..d8df84bbba8 100644 --- a/src/libstd/trie.rs +++ b/src/libstd/trie.rs @@ -12,7 +12,7 @@ use prelude::*; use uint; -use util::{swap, replace}; +use util::replace; use vec; // FIXME: #5244: need to manually update the TrieNode constructor @@ -415,39 +415,41 @@ fn find_mut<'r, T>(child: &'r mut Child, key: uint, idx: uint) -> Option<&'r fn insert(count: &mut uint, child: &mut Child, key: uint, value: T, idx: uint) -> Option { - let mut tmp = Nothing; - let ret; - swap(&mut tmp, child); + // we branch twice to avoid having to do the `replace` when we + // don't need to; this is much faster, especially for keys that + // have long shared prefixes. + match *child { + Nothing => { + *count += 1; + *child = External(key, value); + return None; + } + Internal(ref mut x) => { + return insert(&mut x.count, &mut x.children[chunk(key, idx)], key, value, idx + 1); + } + External(stored_key, ref mut stored_value) if stored_key == key => { + // swap in the new value and return the old. + return Some(replace(stored_value, value)); + } + _ => {} + } - *child = match tmp { - External(stored_key, stored_value) => { - if stored_key == key { - ret = Some(stored_value); - External(stored_key, value) - } else { - // conflict - split the node - let mut new = ~TrieNode::new(); - insert(&mut new.count, - &mut new.children[chunk(stored_key, idx)], - stored_key, stored_value, idx + 1); - ret = insert(&mut new.count, &mut new.children[chunk(key, idx)], - key, value, idx + 1); - Internal(new) - } - } - Internal(x) => { - let mut x = x; - ret = insert(&mut x.count, &mut x.children[chunk(key, idx)], key, - value, idx + 1); - Internal(x) - } - Nothing => { - *count += 1; - ret = None; - External(key, value) - } - }; - return ret; + // conflict, an external node with differing keys: we have to + // split the node, so we need the old value by value; hence we + // have to move out of `child`. + match replace(child, Nothing) { + External(stored_key, stored_value) => { + let mut new = ~TrieNode::new(); + insert(&mut new.count, + &mut new.children[chunk(stored_key, idx)], + stored_key, stored_value, idx + 1); + let ret = insert(&mut new.count, &mut new.children[chunk(key, idx)], + key, value, idx + 1); + *child = Internal(new); + return ret; + } + _ => unreachable!() + } } fn remove(count: &mut uint, child: &mut Child, key: uint,