Make rehashing and resizing less generic #282
Conversation
| @@ -1312,6 +1224,19 @@ impl<A: Allocator + Clone> RawTableInner<A> { | |||
| Bucket::from_base_index(self.data_end(), index) | |||
| } | |||
|
|
|||
| #[cfg_attr(feature = "inline-more", inline)] | |||
| unsafe fn bucket_ptr(&self, index: usize, size_of: usize) -> *mut u8 { | |||
There was a problem hiding this comment.
This function absolutely needs to be #[inline] at minimum, it is not generic.
There was a problem hiding this comment.
It is generic over A, so I figured this should match the inlining of fn bucket. Though I would put #[inline(always)] on both of them (on release mode).
There was a problem hiding this comment.
There's a bunch of small functions which are beneficial to inline. I'll open a PR for those. It seems the compile time benefit of #119 mostly comes from removing inline on get, insert, etc.
| // that we haven't rehashed yet. We unfortunately can't preserve the | ||
| // element since we lost their hash and have no way of recovering it | ||
| // without risking another panic. | ||
| self.prepare_rehash_in_place(); |
There was a problem hiding this comment.
There's no reason for prepare_rehash_in_place to be a separate function any more now that rehash_in_place is moved to RawTableInner.
There was a problem hiding this comment.
I find rehash_in_place to be sufficiently large and complex already, so I don't mind this being in a separate function for code clarity. I'm less sure about prepare_resize staying a separate function though.
|
@bors r+ |
|
📌 Commit bf8635d has been approved by |
|
☀️ Test successful - checks-actions |
This makes the code in
rehash_in_place,resizeandreserve_rehashless generic onT. It also improves the performance of rustc. That performance increase in partially attributed to the use of#[inline(always)].This is the effect on rustc runtime:
rustc_driver's code size is increased by 0.02%.This is the effect on compile time this has on my HashMap compile time benchmark:
The
hashmap-instances:debugcompile time could be further improved if there was a way to apply#[inline(always)]only on release builds.