Implement parallel hpx::uninitialized_relocate_* algorithms for overlapping ranges by ArivoliR · Pull Request #6879 · TheHPXProject/hpx

ArivoliR · 2026-02-04T17:14:36Z

Fixes #6878

Proposed Changes

Implemented parallel hpx::uninitialized_relocate_* algorithms for overlapping ranges.
Chain-based algorithm: for a left-shift by d positions, the range is split into d independent chains where chain c processes indices c, c+d, c+2d, ....
Each chain can be executed in parallel since elements within a chain depend only on each other.

Any background context you want to provide?

P1144 Object relocation in terms of move plus destroy
GSoC Project Wiki

Checklist

Not all points below apply to all pull requests.

I have added a new feature and have added tests to go along with it.
I have fixed a bug and have added a regression test.
I have added a test using random numbers; I have made sure it uses a seed, and that random numbers generated are valid inputs for the tests.

StellarBot · 2026-02-04T17:15:03Z

Can one of the admins verify this patch?

hkaiser · 2026-02-04T23:48:38Z

@isidorostsa I believe this is another PR for you to look at. Would you have the time?

hkaiser · 2026-02-08T16:42:18Z

@ArivoliR Could you please rebase your branch onto master (and resolve the conflicts)?

hkaiser · 2026-02-09T13:08:15Z

+
+    //    test_uninitialized_relocate_overlap_forward(
+    //        hpx::execution::par(hpx::execution::task))
+    //        .get();


Do you plan to re-enable these tests?

No sir, those were only added temporarily for debugging. I’ll be removing them in the next commit. Thank you for pointing it out

Wouldn't it be useful to have those tests functioning?

Hi @hkaiser,
I’ve split the original overlap test into two clearer pieces as guided by Isidoros:

A sequenced-only semantics test that validates the expected behavior for overlapping ranges under execution::seq.

A policy comparison test that checks that execution::par (and par_unseq) produce the same observable results as the sequenced implementation.

For now I’ve left par(task) out and added a TODO. I haven’t added a test for the async behavior yet, since it requires explicitly handling the task/sender return path. I’d prefer to add a dedicated test for that once I’ve spent more time understanding the expected testing pattern.

You can ignore the sender&receiver part for now. I'd focus on the async/future implementation which should be easier to do. However I'd be fine for this ending up in a separate PR.

I don’t have a working test for the async/future-based path yet, and I’m still figuring out the right way to implement it. I’d prefer to follow up with a separate PR once I figure it out in sometime. If you’re okay with it, I’d be happy to close this PR here once the current commits are approved

hkaiser · 2026-02-10T19:52:44Z

There are still some minor things left:

inspect complains about a missing #include
the spell checker complains about the use of iff, please add that to the while list here: https://github.com/STEllAR-GROUP/hpx/blob/master/tools/.codespell_whitelist

The other reported problems are unrelated.

ArivoliR · 2026-02-16T17:55:42Z

There are still some minor things left:
inspect complains about a missing #include
the spell checker complains about the use of iff, please add that to the while list here: https://github.com/STEllAR-GROUP/hpx/blob/master/tools/.codespell_whitelist
The other reported problems are unrelated.

@hkaiser I believe this PR is ready for merge. I’m rebasing it now, please let me know if youd' like any further changes.

isidorostsa · 2026-02-26T06:49:30Z

In this PR, ranges that overlap in any direction will be handled by the sequential version of the code, instead of just the ones that overlap in the direction that would work with forward iteration. This means it is still a bug to call this with a range overlapping in the wrong direction, but at least its a bug in the sequential code.

In the case of relocation this will either end up destroying objects and moving from them afterwards (bug) or if the type is trivially relocatable it will keep memcpying the same object (not as bad, but still a bug). In the parallel code case it would do this but also with data races.

All in all, I think we should keep the logic of the check as it was before, since this change is not protecting us from any bugs, but looks like it is. But we should rewrite it in the style you wrote it, as it is clearer than what we had before.

For bug identification we should HPX_ASSERT that the ranges are not overlapping on either direction if we decide to go for the parallel version. When we implement the safe parallel version we can change the selection logic to fit this.

Sahilll10 · 2026-03-03T08:22:40Z

Hi, I've been going through this PR as part of researching the GSoC project on "Implement parallel hpx::uninitialized_relocate_ algorithms for overlapping ranges." Really helpful context here.

@isidorostsa - If I'm understanding correctly, the current check falls back to sequential for all overlaps including d_first > first cases where forward parallel chunking would actually be safe.

The fix should preserve full parallelism when the shift direction is safe, only serialize when there is a genuine hazard, and HPX_ASSERT for the truly undefined cases in the parallel path.

Would appreciate any correction if I’ve misunderstood something.

isidorostsa · 2026-03-16T18:00:57Z

@ArivoliR, thanks for working more on this. What is the state of this PR? What does it achieve compared to master?

hkaiser · 2026-03-16T22:22:37Z

@ArivoliR any comments to @isidorostsa's question?

ArivoliR · 2026-03-17T08:21:36Z

@ArivoliR, thanks for working more on this. What is the state of this PR? What does it achieve compared to master?

The PR essentially adds HPX_ASSERT(!overlap) before any parallel algorithm is executed in all three functions. This catches wrong-direction overlaps in debug builds before they reach parallel execution. There is no other major change. Pushing fix for clang-format now.

…ap detection Signed-off-by: ArivoliR <arivoli2005@gmail.com>

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

…structors For non-trivially relocatable types with a potentially-throwing move constructor, the parallel chain algorithm cannot guarantee the same move/destruction counts as sequential execution (P1144 requirement). Fall back to sequential in that case. Signed-off-by: ArivoliR <arivoli2005@gmail.com>

…ackward Signed-off-by: ArivoliR <arivoli2005@gmail.com>

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

hkaiser

LGTM, thanks!

Please resolve the merge conflicts, however.

ArivoliR · 2026-03-30T15:04:05Z

LGTM, thanks!

Please resolve the merge conflicts, however.

There are a few errors sir, still working on them. Will push the fixes soon when I'm done and resolve the conflicts as well

ArivoliR · 2026-03-30T15:43:30Z

To be specific, there was a clang-tidy error since there was a pointer to a non-trivially-copyable type.
And a few clang-format errors. Should be fixed now

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

…le-clean

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

codacy-production · 2026-03-31T12:38:04Z

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

_{TIP This summary will be updated as you push new changes. Give us feedback}

hkaiser

LGTM, thanks!

ArivoliR requested a review from hkaiser as a code owner February 4, 2026 17:14

ArivoliR changed the title ~~tests: add overlapping range tests for uninitialized_relocate~~ algorithms: handle overlapping ranges in uninitialized_relocate (WIP) Feb 4, 2026

hkaiser added the category: algorithms label Feb 4, 2026

hkaiser added this to the 2.0.0 milestone Feb 4, 2026

ArivoliR changed the title ~~algorithms: handle overlapping ranges in uninitialized_relocate (WIP)~~ algorithms: handle overlapping ranges in uninitialized_relocate Feb 6, 2026

isidorostsa reviewed Feb 7, 2026

View reviewed changes

ArivoliR force-pushed the trivial-relocatable branch from 51e050b to d4fd325 Compare February 9, 2026 08:53

hkaiser reviewed Feb 9, 2026

View reviewed changes

ArivoliR force-pushed the trivial-relocatable branch from d4fd325 to ac3fc68 Compare February 9, 2026 13:28

ArivoliR requested a review from isidorostsa February 9, 2026 13:49

ArivoliR force-pushed the trivial-relocatable branch from 2c7490c to fec562e Compare February 9, 2026 22:14

ArivoliR force-pushed the trivial-relocatable branch 2 times, most recently from 345ef7f to 4df7159 Compare February 11, 2026 17:37

ArivoliR requested a review from hkaiser February 12, 2026 05:02

ArivoliR force-pushed the trivial-relocatable branch from 4df7159 to 5fa5d27 Compare February 14, 2026 12:15

ArivoliR force-pushed the trivial-relocatable branch from 5fa5d27 to e851773 Compare February 16, 2026 17:55

ArivoliR force-pushed the trivial-relocatable branch 2 times, most recently from dcf51bc to 6d6d8ca Compare March 5, 2026 09:17

ArivoliR force-pushed the trivial-relocatable branch 2 times, most recently from 2fc4adb to 3dad9a8 Compare March 17, 2026 09:17

ArivoliR added 7 commits March 29, 2026 21:14

add parallel_uninitialized_relocate_n_backward and refactor CPO overl…

45e4d61

…ap detection Signed-off-by: ArivoliR <arivoli2005@gmail.com>

add parallel uninitialized_relocate for overlapping ranges

fd0621c

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

add assert for test that is not deterministic for parallel execution

c205154

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

add benchmark for uninitialized_relocate and uninitialized_relocate_b…

734cb44

…ackward Signed-off-by: ArivoliR <arivoli2005@gmail.com>

name lambdas and add comments

c21678d

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

Add no-leak assertion for overlapping exception

1f89822

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

ArivoliR force-pushed the trivial-relocatable branch 2 times, most recently from 0c2c61c to 9a6a0b8 Compare March 29, 2026 23:58

hkaiser previously approved these changes Mar 30, 2026

View reviewed changes

ArivoliR dismissed hkaiser’s stale review via 6d0817f March 30, 2026 15:29

ArivoliR force-pushed the trivial-relocatable branch from 9a6a0b8 to 6d0817f Compare March 30, 2026 15:29

Tests for right-overlapping trivial relocation

3e833c3

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

ArivoliR force-pushed the trivial-relocatable branch 2 times, most recently from 64c2d72 to aebb292 Compare March 30, 2026 16:13

Merge remote-tracking branch 'upstream/master' into trivial-relocatab…

bd5d836

…le-clean

ArivoliR force-pushed the trivial-relocatable branch from aebb292 to bd5d836 Compare March 30, 2026 16:14

ArivoliR requested a review from hkaiser March 30, 2026 19:19

hkaiser reviewed Mar 30, 2026

View reviewed changes

Comment thread libs/core/algorithms/include/hpx/parallel/algorithms/uninitialized_relocate.hpp

Remove explicit casts in memmove calls

c0b9cc5

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

ArivoliR force-pushed the trivial-relocatable branch 2 times, most recently from 7ab62e5 to 6f1cc34 Compare March 31, 2026 09:46

Add NOLINTNEXTLINE in memmove calls

d5080d9

Signed-off-by: ArivoliR <arivoli2005@gmail.com>

ArivoliR force-pushed the trivial-relocatable branch from 6f1cc34 to d5080d9 Compare March 31, 2026 12:36

ArivoliR requested a review from hkaiser March 31, 2026 15:48

hkaiser approved these changes Apr 3, 2026

View reviewed changes

hkaiser merged commit 4803e49 into TheHPXProject:master Apr 3, 2026
107 checks passed

Uh oh!

Conversation

ArivoliR commented Feb 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Proposed Changes

Any background context you want to provide?

Checklist

Uh oh!

StellarBot commented Feb 4, 2026

Uh oh!

hkaiser commented Feb 4, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hkaiser commented Feb 8, 2026

Uh oh!

hkaiser Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

ArivoliR Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

hkaiser Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

ArivoliR Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

hkaiser Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

ArivoliR Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

hkaiser commented Feb 10, 2026

Uh oh!

ArivoliR commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

isidorostsa commented Feb 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Sahilll10 commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

isidorostsa commented Mar 16, 2026

Uh oh!

hkaiser commented Mar 16, 2026

Uh oh!

ArivoliR commented Mar 17, 2026

Uh oh!

hkaiser left a comment

Choose a reason for hiding this comment

Uh oh!

ArivoliR commented Mar 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ArivoliR commented Mar 30, 2026

Uh oh!

Uh oh!

codacy-production Bot commented Mar 31, 2026

Up to standards ✅

Uh oh!

hkaiser left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

ArivoliR commented Feb 4, 2026 •

edited

Loading

ArivoliR commented Feb 16, 2026 •

edited

Loading

isidorostsa commented Feb 26, 2026 •

edited

Loading

Sahilll10 commented Mar 3, 2026 •

edited

Loading

ArivoliR commented Mar 30, 2026 •

edited

Loading