64-bit indexing for DeviceSegmentedReduce by jecs · Pull Request #414 · NVIDIA/cccl

jecs · 2023-09-07T18:03:02Z

Description

Added NumSegmentsT parameter to DeviceSegmentedReduce methods, so now num_segments can be of any integral type
Added OffsetT type resolution in a manner similar to DeviceReduce: set to std::common_type_t of BeginOffsetIterT::value_type and EndOffsetIterT::value_type. (see detail/choose_offset.cuh)
In a manner similar to DeviceReduce, we use static_cast to convert num_segments to OffsetT before dispatching
OffsetT type resolution uses std::common_type_t, std::void_t, and one inline constexpr variable (is_iterator_v).

Checklist

New or existing tests cover these changes.
The documentation is up to date with these changes.

* Added NumSegmentsT parameter to methods, so now num_segments can be of any integral type * Added OffsetT type resolution in a manner similar to DeviceReduce: set to std::common_type_t of BeginOffsetIterT::value_type and EndOffsetIterT::value_type. (see detail/choose_offset.cuh) * In a manner similar to DeviceReduce, we use static_cast to convert num_segments to OffsetT before dispatching * OffsetT type resolution uses std::common_type_t, std::void_t, and one inline constexpr variable (is_iterator_v).

copy-pr-bot · 2023-09-07T18:03:05Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

jecs · 2023-09-07T18:05:28Z

Hi, should I create an Issue and add it to the description? I originally submitted this PR to the old Cub repository.

gevtushenko

Thank you for the contribution! I've left a few comments, but before proceeding with the review, we'll need a test case that covers 64-bit problem sizes. Could you please add one? In order to keep moderate memory usage, you could try to rely on counting or constant iterators instead of actually allocating huge inputs.

jecs · 2023-09-08T02:08:49Z

Thank you for the contribution! I've left a few comments, but before proceeding with the review, we'll need a test case that covers 64-bit problem sizes. Could you please add one? In order to keep moderate memory usage, you could try to rely on counting or constant iterators instead of actually allocating huge inputs.

Thank you for the feedback! Yes, sure thing. I'll write a quick test using some kind of counter iterator- or constant iterator-based approach.

miscco

Thanks a lot for the contribution, I am so sorry about our support situation

* Removed NumSegmentsT portion * Added catch2 test * Modified previous test slightly to use correct type for num_segments * Made type deduction SFINAE-friendly

jecs · 2023-09-11T20:44:23Z

@miscco @senior-zero Ready. (I hope.) I walked back 64-bit num_segments. I hope to submit that in a new Pull Request in the near future. For the test, I simply copied the previous test and modified it slightly.

…e-reduce-64bit

miscco · 2023-09-13T06:24:22Z

/ok to test

jecs · 2023-09-13T13:36:40Z

Whoops! I deleted the offending lines and ran C++11 tests on my end. Everything seems to work now. Getting your code to be compliant with several standards is tricky business.

gevtushenko · 2023-09-19T20:13:57Z

@jecs your repo seems to be protected. I've filed a PR with a few fixes: jecs#1 for your branch

Device reduce 64bit

gevtushenko · 2023-09-19T20:42:31Z

/ok to test

jecs · 2023-09-21T19:19:26Z

@senior-zero I added you as a collaborator on my fork in case you'd like to make more changes.

gevtushenko · 2023-09-22T15:07:14Z

/ok to test

jecs · 2023-09-27T14:47:45Z

@senior-zero @miscco Thank you both so much for your excellent feedback and for guiding me through the process! I learned a lot from this experience. It's my first time submitting a pull request to a public project like this. And thank you for teaching me about the "forbidden fruit" hidden within ::detail and also for taking the wheel towards the end.

gevtushenko · 2023-09-27T15:04:59Z

@jecs thank you for the contribution!

- Retain offset_t/SegmentSizeT comments across all fixed-size env overloads - Retain PR NVIDIA#414 breaking-change comments in ArgMin/ArgMax env overloads - Refactor fixed-size ArgMin/ArgMax into shared private helpers (fixed_size_arg_min_dispatch, fixed_size_arg_max_dispatch) to reduce duplication between env and non-env overloads

* Add fixed-size segment env overloads for DeviceSegmentedReduce * Fix critical issues plus<> specilization * feedback address - Retain offset_t/SegmentSizeT comments across all fixed-size env overloads - Retain PR #414 breaking-change comments in ArgMin/ArgMax env overloads - Refactor fixed-size ArgMin/ArgMax into shared private helpers (fixed_size_arg_min_dispatch, fixed_size_arg_max_dispatch) to reduce duplication between env and non-env overloads * Refactor * More refactoring * Refactor: unify fixed-size ArgMin/ArgMax dispatch into a single function * Make using offset_t = int explicit * Substitute cudaStream_t with cuda::stream and use always stream.sync() but outside of the literalinclude guards * Guard Sum with sfinae * Sfinae operator<<(os,...) for cuda::std::pairs * Do not specialize std::plus<> --------- Co-authored-by: Bernhard Manfred Gruber <bernhardmgruber@gmail.com>

…A#8097) * Add fixed-size segment env overloads for DeviceSegmentedReduce * Fix critical issues plus<> specilization * feedback address - Retain offset_t/SegmentSizeT comments across all fixed-size env overloads - Retain PR NVIDIA#414 breaking-change comments in ArgMin/ArgMax env overloads - Refactor fixed-size ArgMin/ArgMax into shared private helpers (fixed_size_arg_min_dispatch, fixed_size_arg_max_dispatch) to reduce duplication between env and non-env overloads * Refactor * More refactoring * Refactor: unify fixed-size ArgMin/ArgMax dispatch into a single function * Make using offset_t = int explicit * Substitute cudaStream_t with cuda::stream and use always stream.sync() but outside of the literalinclude guards * Guard Sum with sfinae * Sfinae operator<<(os,...) for cuda::std::pairs * Do not specialize std::plus<> --------- Co-authored-by: Bernhard Manfred Gruber <bernhardmgruber@gmail.com>

jecs requested review from a team as code owners September 7, 2023 18:03

jecs requested review from elstehle and griwes and removed request for a team September 7, 2023 18:03

jecs added 5 commits September 7, 2023 15:18

Fixed minor mistake in OffsetT static_assert

525aef9

Fixed type checking in choose_offset.cuh

6905d7f

Undid erroneous changes due to using old Cub repository

2b9e144

Undid the last few erroneous changes

e3d3ec6

Converted tabs to spaces

34c733d

gevtushenko reviewed Sep 8, 2023

View reviewed changes

Comment thread cub/cub/device/dispatch/dispatch_reduce.cuh Resolved

Comment thread cub/cub/detail/choose_offset.cuh Resolved

Comment thread cub/cub/detail/choose_offset.cuh Resolved

miscco requested changes Sep 8, 2023

View reviewed changes

64-bit indexing for DeviceSegmentedReduce

5b82798

* Removed NumSegmentsT portion * Added catch2 test * Modified previous test slightly to use correct type for num_segments * Made type deduction SFINAE-friendly

jecs and others added 2 commits September 11, 2023 16:54

fixed minor formatting issue

d7a7ed4

Merge branch 'NVIDIA:main' into device-reduce-64bit

5f694d7

gevtushenko reviewed Sep 11, 2023

View reviewed changes

Comment thread cub/test/catch2_test_device_segmented_reduce_iterators_64bit.cu Resolved

jecs added 3 commits September 11, 2023 20:28

Made test input sequences with number of elements >= 2^32

a4c032b

Merge branch 'device-reduce-64bit' of github.com:jecs/cccl into devic…

39bad5d

…e-reduce-64bit

Removed some cruft

cfef155

jecs requested a review from miscco September 12, 2023 20:56

Removed defaulted assignment operators for C++11 compatibility

129b95c

jecs requested a review from gevtushenko September 13, 2023 13:43

Test signed offset types in segmented reduce

31554f8

gevtushenko requested changes Sep 19, 2023

View reviewed changes

Comment thread cub/cub/device/device_segmented_reduce.cuh Resolved

Return old offset type to segmented arg reductions

46dc4c8

Merge pull request #1 from senior-zero/device-reduce-64bit

b598341

Device reduce 64bit

jecs requested a review from gevtushenko September 20, 2023 18:49

gevtushenko approved these changes Sep 21, 2023

View reviewed changes

Merge branch 'main' into device-reduce-64bit

a270c0d

gevtushenko merged commit 94b358b into NVIDIA:main Sep 27, 2023

jecs deleted the device-reduce-64bit branch September 28, 2023 15:57

gevtushenko mentioned this pull request Dec 15, 2023

[BUG]: Improve error reporting for segmented reduce #1221

Closed

1 task done

elstehle mentioned this pull request Feb 21, 2024

[EPIC]: CUB large input support #50

Open

23 of 25 tasks

gonidelis mentioned this pull request Mar 6, 2024

[BUG]: cuda::std::iterator_traits does not expose proper member types in old C++ dialects #1509

Closed

1 task done

elstehle mentioned this pull request Aug 15, 2024

Describe potential solutions for algorithm implementations and their pros/cons #1771

Closed

elstehle mentioned this pull request Oct 9, 2024

Add support for large num_items to DeviceReduce::{ArgMin,ArgMax} #2515

Closed

6 tasks done

elstehle mentioned this pull request Oct 29, 2024

Adds support for large num_items to DeviceReduce::{ArgMin,ArgMax} #2647

Merged

2 tasks done

elstehle mentioned this pull request Dec 12, 2024

[FEA]: Introduce new DeviceReduce::Arg{Min,Max} interface with two output iterators #3146

Closed

1 task done

elstehle mentioned this pull request Nov 5, 2025

Add cccl_add_xfail_compile_target_test CMake function #6434

Merged

Conversation

jecs commented Sep 7, 2023 • edited by gevtushenko Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Checklist

Uh oh!

copy-pr-bot Bot commented Sep 7, 2023

Uh oh!

jecs commented Sep 7, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gevtushenko left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jecs commented Sep 8, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

miscco left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jecs commented Sep 11, 2023

Uh oh!

Uh oh!

miscco commented Sep 13, 2023

Uh oh!

jecs commented Sep 13, 2023

Uh oh!

Uh oh!

gevtushenko commented Sep 19, 2023

Uh oh!

gevtushenko commented Sep 19, 2023

Uh oh!

jecs commented Sep 21, 2023

Uh oh!

gevtushenko commented Sep 22, 2023

Uh oh!

jecs commented Sep 27, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gevtushenko commented Sep 27, 2023

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jecs commented Sep 7, 2023 •

edited by gevtushenko

Loading

jecs commented Sep 7, 2023 •

edited

Loading

jecs commented Sep 8, 2023 •

edited

Loading

jecs commented Sep 27, 2023 •

edited

Loading