Optimize first() and last()

[cberner]

This change makes these methods almost 2x faster, by reducing the number of tree traversals from two to one.

[adamreichold]

I understand tree depth is somewhat limited, but this looks highly amenable to an iterative implementation by basically putting all of this in a loop and writing

page = self.mem.get_page_extended(child_page, self.hint)?;

here with a

return Ok(Some((key_guard, value_guard)));

above to avoid the any questions of call overhead and/or stack usage.

[cberner]

Thanks for the suggestion. I'm going to stick with this recursive implementation for consistency with most of the rest of the code base, but if you have any benchmarks on the overhead of the function calls, I'd be very interested and open to changing if there's meaningful overhead Sent from my phone

…

On Mon, Sep 30, 2024, 12:25 AM Adam Reichold ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In src/tree_store/btree.rs <#871 (comment)>: > + &self, + page: PageImpl, + ) -> Result<Option<(AccessGuard<'static, K>, AccessGuard<'static, V>)>> { + let node_mem = page.memory(); + match node_mem[0] { + LEAF => { + let accessor = LeafAccessor::new(page.memory(), K::fixed_width(), V::fixed_width()); + let (key_range, value_range) = accessor.entry_ranges(0).unwrap(); + let key_guard = AccessGuard::with_page(page.clone(), key_range); + let value_guard = AccessGuard::with_page(page, value_range); + Ok(Some((key_guard, value_guard))) + } + BRANCH => { + let accessor = BranchAccessor::new(&page, K::fixed_width()); + let child_page = accessor.child_page(0).unwrap(); + self.first_helper(self.mem.get_page_extended(child_page, self.hint)?) I understand tree depth is somewhat limited, but this looks highly amenable to an iterative implementation by basically putting all of this in a loop and writing page = self.mem.get_page_extended(child_page, self.hint)?; here with a return Ok(Some((key_guard, value_guard))); above to avoid the any questions of call overhead and/or stack usage. — Reply to this email directly, view it on GitHub <#871 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGNXQG7WZSACU6NKGQD2FDZZD4GVAVCNFSM6AAAAABPCKJOAGVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDGMZWGYYTANBVGU> . You are receiving this because you authored the thread.Message ID: ***@***.***>