[libc++] Optimize vector::emplace_back - #218658
Merged
Merged
Conversation
Contributor
Author
|
/libcxx-bot benchmark libcxx/test/benchmarks/containers/sequence/vector.bench.cpp
|
philnik777
force-pushed
the
optimize_vector_emplace_back
branch
from
August 25, 2026 11:26
260eca1 to
cedd02e
Compare
Contributor
Author
|
/libcxx-bot benchmark libcxx/test/benchmarks/containers/sequence/vector.bench.cpp
Benchmark results for linux-x86_64-20260812:Benchmark results for macos-26.5-arm64-20260812:Benchmark results for macos-26.5-arm64-hardenedfast-20260821: |
philnik777
marked this pull request as ready for review
August 26, 2026 13:18
philnik777
added
libc++
|
@llvm/pr-subscribers-libcxx Author: Nikolas Klauser (philnik777) ChangesThis avoids using a lot of stack memory for Full diff: https://github.com/llvm/llvm-project/pull/218658.diff 1 Files Affected:
diff --git a/libcxx/include/__vector/vector.h b/libcxx/include/__vector/vector.h
index ebcfcc17d4b6d..dad2f5f48596a 100644
--- a/libcxx/include/__vector/vector.h
+++ b/libcxx/include/__vector/vector.h
@@ -754,9 +754,6 @@ class _LIBCPP_WARN_UNUSED vector {
__annotate_shrink(__old_size);
}
- template <class... _Args>
- _LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_HIDE_FROM_ABI inline pointer __emplace_back_slow_path(_Args&&... __args);
-
// The following functions are no-ops outside of AddressSanitizer mode.
// We call annotations for every allocator, unless explicitly disabled.
//
@@ -1071,34 +1068,25 @@ _LIBCPP_CONSTEXPR_SINCE_CXX20 void vector<_Tp, _Allocator>::shrink_to_fit() _NOE
}
}
-template <class _Tp, class _Allocator>
-template <class... _Args>
-_LIBCPP_CONSTEXPR_SINCE_CXX20 typename vector<_Tp, _Allocator>::pointer
-vector<_Tp, _Allocator>::__emplace_back_slow_path(_Args&&... __args) {
- _SplitBuffer __v(__recommend(size() + 1), size(), this->__layout_.__alloc());
- // __v.emplace_back(std::forward<_Args>(__args)...);
- pointer __end = __v.end();
- __alloc_traits::construct(this->__layout_.__alloc(), std::__to_address(__end), std::forward<_Args>(__args)...);
- __v.__set_sentinel(++__end);
- __layout_.__relocate(__v);
- return __end;
-}
-
// This makes the compiler inline `__else()` if `__cond` is known to be false. Currently LLVM doesn't do that without
// the `__builtin_constant_p`, since it considers `__else` unlikely even through it's known to be run.
// See https://llvm.org/PR154292
-template <class _If, class _Else>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 void __if_likely_else(bool __cond, _If __if, _Else __else) {
+template <class _If, class _Else, class... _Args>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 decltype(std::declval<_If>()(std::declval<_Args&&>()...))
+__if_likely_else(bool __cond, _If __if, _Else __else, _Args&&... __args) {
+ static_assert(
+ is_same<decltype(__if(std::forward<_Args>(__args)...)), decltype(__else(std::forward<_Args>(__args)...))>::value,
+ "Return type of if and else branch have to be the same type");
if (__builtin_constant_p(__cond)) {
if (__cond)
- __if();
+ return __if(std::forward<_Args>(__args)...);
else
- __else();
+ return __else(std::forward<_Args>(__args)...);
} else {
if (__cond) [[__likely__]]
- __if();
+ return __if(std::forward<_Args>(__args)...);
else
- __else();
+ return __else(std::forward<_Args>(__args)...);
}
}
@@ -1106,16 +1094,23 @@ template <class _Tp, class _Alloc>
template <class... _Args>
_LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_HIDE_FROM_ABI typename vector<_Tp, _Alloc>::__emplace_back_result_t
vector<_Tp, _Alloc>::emplace_back(_Args&&... __args) {
- pointer __end = __layout_.__end_ptr();
std::__if_likely_else(
size() != capacity(),
- [&] {
- __emplace_back_assume_capacity(std::forward<_Args>(__args)...);
- ++__end;
+ [](vector& __self, _Args&&... __largs) static {
+ __self.__emplace_back_assume_capacity(std::forward<_Args>(__largs)...);
+ },
+ [](vector& __self, _Args&&... __largs) static {
+ _SplitBuffer __v(__self.__recommend(__self.size() + 1), __self.size(), __self.__layout_.__alloc());
+ // __v.emplace_back(std::forward<_Args>(__args)...);
+ pointer __end = __v.end();
+ __alloc_traits::construct(
+ __self.__layout_.__alloc(), std::__to_address(__end), std::forward<_Args>(__largs)...);
+ __v.__set_sentinel(++__end);
+ __self.__layout_.__relocate(__v);
},
- [&] { __end = __emplace_back_slow_path(std::forward<_Args>(__args)...); });
+ *this,
+ std::forward<_Args>(__args)...);
- __layout_.__set_bound_using_pointer(__end);
#if _LIBCPP_STD_VER >= 17
return back();
#endif
|
Contributor
Author
|
@liza371 This might also fix your issue with |
Contributor
Oh wow this does help a lot this is awesome. Just for reference here's the binary size diff with this specific optimization: https://chrome-supersize.firebaseapp.com/viewer.html?load_url=https%3A%2F%2Fstorage.googleapis.com%2Fchrome-supersize%2Foneoffs%2Fb16984ce99c702355a5b2b4c52574e82cec41fb9_4b112225b8291de813e1d486a4265bea054f4115.sizediff&group_by=template You also rightly pointed out I didn't include an analysis with the ABI turned off. Here's an analysis with the sized based vector ABI turned off for reference I haven't run this analysis with the unstable ABI fully turned off as chromium depends on other features pretty heavily so it would be non-trivial to do so. |
Member
|
@liza371 That tool looks amazing. I'm trying to understand the output though: does it mean that applying this patch increases the overall size of Chromium by 140.35 KiB (for unstable ABI) and 62.68 KiB (for stable ABI) no matter what? Isn't that a net regression? I think this is just how I'm reading the tool's output. |
ldionne
reviewed
Sep 1, 2026
ldionne
left a comment
ldionne
left a comment
Member
There was a problem hiding this comment.
I like the code changes, but the benchmarks are not very conclusive. Do we think this is simply because our benchmarks are not sensitive enough to capture this improvement?
Contributor
|
@ldionne not quite, I should have linked my other PR sorry about that. I opened #216822 to address the binary size bloat we were seeing in Chromium, and I have links there to show the initial bloat. Applying this patch still leads to an increase in Chromium's binary size, but it's a much smaller increase than we saw when we pulled in size-based vector. I initially was also trying to land some changes to emplace_back that @cjdb wrote but @philnik777 noted that they would remove a lot of other performance optimizations, so I shelved that. #216822 still has some small patches, but I got sidetracked and now I'm admittedly still not entirely sure why the inlining does help with the bloat, so I wanted to do some more investigating to get a more definitive answer than "idk the tool says so" before coming back to that PR. |
Contributor
Author
Yes, I think so. |
ldionne
approved these changes
Sep 3, 2026
ldionne
left a comment
ldionne
left a comment
Member
There was a problem hiding this comment.
@liza371 Just to clarify, even if we land this patch, you're still blocked from rolling libc++ because of the size increase, it's just not as bad, right? In other words, it seems that we still need to figure out what made __relocate (from this) go larger, no matter what.
@liza371 What optimization level does Chrome build with?
Either way, I think this change is a net positive so let's go with it, but let's figure out what went wrong with __relocate as well.
Contributor
|
@ldionne yes that's correct. We're building with I did just attempt a build with These are both without the |
Iasonaskrpr
pushed a commit
to Iasonaskrpr/llvm-project
that referenced
this pull request
Sep 4, 2026
This avoids using a lot of stack memory for `emplace_back` by making the outlined lambda `static`. This also reduces the number of instructions to execute a simple `emplace_back`.
addmisol
pushed a commit
to addmisol/llvm-project2
that referenced
this pull request
Sep 6, 2026
This avoids using a lot of stack memory for `emplace_back` by making the outlined lambda `static`. This also reduces the number of instructions to execute a simple `emplace_back`.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This avoids using a lot of stack memory for
emplace_backby making the outlined lambdastatic. This also reduces the number of instructions to execute a simpleemplace_back.