Skip to content

Add Neon implementation of find for ARM64 targets #6003

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
hazzlim wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add Neon implementation of find for ARM64 targets #6003

hazzlim wants to merge 1 commit into from
+318 −15

Conversation

@hazzlim
Copy link
Contributor

@hazzlim hazzlim commented Jan 9, 2026 •
edited
Loading

This PR vectorize find using Neon 🚀

For char/uint8_t we just dispatch to memchr (as is done currently), as this is already well optimized using Neon.

Benchmark results (figures are relative speedup compared to existing code):

Benchmark MSVC SU Clang SU
bm<char, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/8021/3056 9.375 9.199
bm<char, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/63/62 4.39 4.3
bm<char, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/31/30 2.511 2.581
bm<char, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/15/14 1.491 1.398
bm<char, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/7/6 0.752 0.773
bm<uint16_t, not_highly_aligned_allocator, Op::FindSized>/8021/3056 2.338 2.237
bm<uint16_t, not_highly_aligned_allocator, Op::FindSized>/63/62 1.405 1.366
bm<uint16_t, not_highly_aligned_allocator, Op::FindSized>/31/30 1.187 1.145
bm<uint16_t, not_highly_aligned_allocator, Op::FindSized>/15/14 1.095 0.931
bm<uint16_t, not_highly_aligned_allocator, Op::FindSized>/7/6 1.023 0.769
bm<wchar_t, not_highly_aligned_allocator, Op::StringFind>/8021/3056 2.332 2.143
bm<wchar_t, not_highly_aligned_allocator, Op::StringFind>/63/62 1.368 1.319
bm<wchar_t, not_highly_aligned_allocator, Op::StringFind>/31/30 1.199 1.106
bm<wchar_t, not_highly_aligned_allocator, Op::StringFind>/15/14 1.147 1.022
bm<wchar_t, not_highly_aligned_allocator, Op::StringFind>/7/6 1.17 0.926
bm<wchar_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/8021/3056 5.455 5.331
bm<wchar_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/63/62 3.738 3.814
bm<wchar_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/31/30 2.554 2.653
bm<wchar_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/15/14 1.564 1.775
bm<wchar_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/7/6 0.999 1.023
bm<uint32_t, not_highly_aligned_allocator, Op::FindSized>/8021/3056 4 4.103
bm<uint32_t, not_highly_aligned_allocator, Op::FindSized>/63/62 3.178 3.409
bm<uint32_t, not_highly_aligned_allocator, Op::FindSized>/31/30 2.4 2.783
bm<uint32_t, not_highly_aligned_allocator, Op::FindSized>/15/14 1.886 1.894
bm<uint32_t, not_highly_aligned_allocator, Op::FindSized>/7/6 1.362 1.179
bm<char32_t, not_highly_aligned_allocator, Op::StringFind>/8021/3056 4.027 4
bm<char32_t, not_highly_aligned_allocator, Op::StringFind>/63/62 2.985 4
bm<char32_t, not_highly_aligned_allocator, Op::StringFind>/31/30 2.281 3.03
bm<char32_t, not_highly_aligned_allocator, Op::StringFind>/15/14 1.721 2.016
bm<char32_t, not_highly_aligned_allocator, Op::StringFind>/7/6 1.105 1.235
bm<char32_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/8021/3056 2.727 2.686
bm<char32_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/63/62 2.613 3.273
bm<char32_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/31/30 2.236 2.705
bm<char32_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/15/14 1.655 1.947
bm<char32_t, not_highly_aligned_allocator, Op::StringFindNotFirstOne>/7/6 1.099 1.207
bm<uint64_t, not_highly_aligned_allocator, Op::FindSized>/8021/3056 2.023 2.024
bm<uint64_t, not_highly_aligned_allocator, Op::FindSized>/63/62 1.948 2.119
bm<uint64_t, not_highly_aligned_allocator, Op::FindSized>/31/30 1.837 1.941
bm<uint64_t, not_highly_aligned_allocator, Op::FindSized>/15/14 1.581 1.51
bm<uint64_t, not_highly_aligned_allocator, Op::FindSized>/7/6 1.235 1.179

@hazzlim hazzlim requested a review from a team as a code owner January 9, 2026 15:00
@github-project-automation github-project-automation bot moved this to Initial Review in STL Code Reviews Jan 9, 2026
@StephanTLavavej StephanTLavavej added performance Must go faster ARM64 Related to the ARM64 architecture labels Jan 9, 2026
@StephanTLavavej StephanTLavavej self-assigned this Jan 9, 2026
AlexGuteniev
AlexGuteniev reviewed Jan 10, 2026
View reviewed changes
stl/src/vector_algorithms.cpp
return vgetq_lane_u64(vpaddq_u64(_Cmp, _Cmp), 0);
}

static uint8x16_t _Not_q(const uint8x16_t _Val) {
Copy link
Contributor

@AlexGuteniev AlexGuteniev Jan 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

noexcept

  • other occurrences

stl/src/vector_algorithms.cpp
#ifdef _M_ARM64EC
#ifdef _M_ARM64
struct _Find_traits_1 {
static uint8x16_t _Load_q(const void* _Ptr) noexcept {
Copy link
Contributor

@AlexGuteniev AlexGuteniev Jan 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

top level const (and in other occurrences

stl/src/vector_algorithms.cpp
}

static uint64x2_t _Not_q(const uint64x2_t _Val) {
return vreinterpretq_u64_u8(vmvnq_u8(vreinterpretq_u8_u64(_Val)));
Copy link
Contributor

@AlexGuteniev AlexGuteniev Jan 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Question: why is it preferrable to negate when in vector, rather than bits?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@AlexGuteniev AlexGuteniev AlexGuteniev left review comments

Reviewers whose approvals may not affect merge requirements

At least 1 approving review is required to merge this pull request.

Assignees

@StephanTLavavej StephanTLavavej

Labels

ARM64 Related to the ARM64 architecture performance Must go faster

Projects

STL Code Reviews
Status: Initial Review

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@hazzlim @AlexGuteniev @StephanTLavavej