Datapar search fixes#7288
Open
ArivoliR wants to merge 4 commits into
Open
Conversation
Signed-off-by: ArivoliR <arivoli2005@gmail.com>
Signed-off-by: ArivoliR <arivoli2005@gmail.com>
Signed-off-by: ArivoliR <arivoli2005@gmail.com>
Signed-off-by: ArivoliR <arivoli2005@gmail.com>
Up to standards ✅🟢 Issues
|
Contributor
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
This PR improves SIMD/datapar behavior for search and search_n, and adds performance benchmarks to compare HPX execution policies against the standard library.
Changes:
- Updated SIMD vector pack unaligned load/store to use explicit (element-aligned) pointer-based loads/stores for non-scalar packs.
- Reworked datapar implementations in
search.hppto use SIMD prefiltering (forsearch) and a sliding-window SIMD scan (forsearch_n). - Added new performance benchmarks for
searchandsearch_nand registered them in the performance CMake target list.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 8 comments.
Show a summary per file
| File | Description |
|---|---|
| libs/core/execution/include/hpx/execution/traits/detail/simd/vector_pack_load_store.hpp | Adjusts unaligned vector pack load/store to use pointer-based element-aligned operations for non-scalar packs. |
| libs/core/algorithms/include/hpx/parallel/datapar/search.hpp | Introduces new SIMD-accelerated scanning strategies for search and search_n. |
| libs/core/algorithms/tests/performance/benchmark_search.cpp | Adds a benchmark comparing hpx::search policies vs std::search. |
| libs/core/algorithms/tests/performance/benchmark_search_n.cpp | Adds a benchmark comparing hpx::search_n policies vs std::search_n. |
| libs/core/algorithms/tests/performance/CMakeLists.txt | Registers the new benchmark executables. |
|
|
||
| std::cout << "\n-------------- Benchmark Result --------------" | ||
| << std::endl; | ||
| auto fmt = "search ({1}) : {2}(sec)"; |
|
|
||
| std::cout << "\n-------------- Benchmark Result --------------" | ||
| << std::endl; | ||
| auto fmt = "search_n ({1}) : {2}(sec)"; |
Comment on lines
+32
to
+35
| for (int i = 0; i < test_count; ++i) | ||
| { | ||
| std::size_t pos = (n * 3 / 4) + | ||
| (static_cast<std::size_t>(i) * 1000003) % (n / 4 - needle_count); |
Comment on lines
+34
to
+35
| std::size_t pos = (n * 3 / 4) + | ||
| (static_cast<std::size_t>(i) * 1000003) % (n / 4 - needle_count); |
| #include <hpx/init.hpp> | ||
| #include <hpx/modules/testing.hpp> | ||
| #include <hpx/program_options.hpp> | ||
|
|
| #include <hpx/init.hpp> | ||
| #include <hpx/modules/testing.hpp> | ||
| #include <hpx/program_options.hpp> | ||
|
|
| // SIMD bulk pass | ||
| for (; i + pack_size <= scan_count; i += pack_size) | ||
| { | ||
| if (tok.was_cancelled(base_idx)) |
Comment on lines
+45
to
+54
| using value_type = | ||
| typename std::iterator_traits<Iter1>::value_type; | ||
| using pack_type = | ||
| hpx::parallel::traits::vector_pack_type_t<value_type>; | ||
| constexpr std::size_t pack_size = | ||
| hpx::parallel::traits::vector_pack_size_v<pack_type>; | ||
|
|
||
| // First-element broadcast: SIMD-scan haystack for matches of | ||
| // needle[0]; only candidates pay the scalar full-needle verify. | ||
| auto const needle0 = HPX_INVOKE(proj2, *s_first); |
Collaborator
|
Can one of the admins verify this patch? |
hkaiser
reviewed
May 22, 2026
Contributor
hkaiser
left a comment
hkaiser
left a comment
There was a problem hiding this comment.
Could you please apply the same fix for the other datapar backends (if necessary)?
Contributor
|
@ArivoliR Also, could you please add a test that verifies your unaligned load fix? |
Contributor
Author
On it! |
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
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.
Partially fixes #7287
Follow-up to PR #6995
Proposed Changes
vector_pack_load::unaligned(std-experimental-simd backend) to perform a real unaligned load instead of broadcasting*iteracross all lanes; symmetric fix to the store.sequential_search_nto use a sliding-window SIMD scan (broadcast value,none_ofskip,all_ofcarry-extend, lane-by-lane in mixed packs) instead of the per-position scan-with-const_loop_napproach.sequential_searchalong the same lines: SIMD-broadcastneedle[0]across the haystack, scalar-verify the full needle only on lane hits. Drops the per-positionzip_iterator+ nested SIMD dispatch.Any background context you want to provide?
Checklist
Not all points below apply to all pull requests.
Benchmarks
hpx::searchhpx::search_n