Trie #350

amukkara · 2023-08-11T17:49:22Z

Trie and associated bit-vector code with tests

GPUtester · 2023-08-11T17:49:24Z

Can one of the admins verify this patch?

Admins can comment ok to test to allow this one PR to run or add to allowlist to allow all future PRs from the same author to run.

rapids-bot · 2023-08-11T17:49:24Z

Pull requests from external contributors require approval from a NVIDIA organization member with write permissions or greater before CI can begin.

PointKernel · 2023-08-16T23:36:45Z

ok to test

sleeepyjack · 2023-08-22T22:35:32Z

/ok to test

copy-pr-bot · 2023-08-30T02:39:52Z

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.

This PR adds `dynamic_bitset` code that will be used in Trie data structure. Trie will be integrated in a separate PR #350. Since `dynamic_bitset` is not intended to be part of public-facing API, all files (.cuh and .inl) are located in include/cuco/detail/trie/dynamic_bitset Tests are added in tests/dynamic_bitset --------- Co-authored-by: Yunsong Wang <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

sleeepyjack · 2023-09-12T14:24:49Z

/ok to test

sleeepyjack · 2023-09-12T14:29:24Z

@amukkara can you sign all of your commits in this PR (starting at c1b448e)? Easiest way to do this is do an interactive rebase, pick->edit every commit that is not signed, and then do a git commit --amend --no-edit -S && git rebase --continue until you land at HEAD.

amukkara · 2023-09-12T19:51:21Z

@amukkara can you sign all of your commits in this PR (starting at c1b448e)? Easiest way to do this is do an interactive rebase, pick->edit every commit that is not signed, and then do a git commit --amend --no-edit -S && git rebase --continue until you land at HEAD.

Done

This improves insertion perfomance massively

A key is a list of labels. Iterators use LabelIt, rather than KeyIt.

sleeepyjack

Reviewed just the benchmarks

sleeepyjack · 2023-10-17T22:47:04Z

benchmarks/trie/lookup_bench.cu

+
+  distribution::unique lengths_dist;
+  distribution::gaussian labels_dist{0.5};
+  generate_labels(labels, offsets, num_keys, max_key_length, lengths_dist, labels_dist);


This should be placed into some namespace, e.g., ´cuco::test::`

Added namespace coco::test::trie 025c59a

sleeepyjack · 2023-10-17T22:47:13Z

benchmarks/trie/lookup_bench.cu

+#include <defaults.hpp>
+#include <utils.hpp>
+
+#include "../../tests/trie/trie_utils.hpp"


You can use the system path here, i.e., <cuco/tests/…>.

<cuco/tests/..> does not compile. Using <../tests/> in recent commit.

Should we change top-level CMakeLists or benchmarks/CMakeLists to access <cuco/tests/..> from benchmarks?

benchmarks/trie/dynamic_bitset/find_next_bench.cu

benchmarks/trie/insert_bench.cu

sleeepyjack · 2023-10-17T22:50:42Z

benchmarks/trie/insert_bench.cu

+  auto const num_keys       = state.get_int64_or_default("NumKeys", 100 * 1000);
+  auto const max_key_length = state.get_int64_or_default("MaxKeyLength", 10);
+
+  using LabelType = int;


Would it make sense to make this a configurable parameter?

7f1d083 makes LabelType configurable. Testing char, int for now.

Both in tests and benchmarks

amukkara marked this pull request as ready for review August 11, 2023 17:50

amukkara force-pushed the trie branch from 90be9f7 to 8a4ea8a Compare August 16, 2023 17:38

amukkara mentioned this pull request Aug 17, 2023

Add dynamic_bitset #352

Merged

amukkara marked this pull request as draft August 18, 2023 05:21

amukkara force-pushed the trie branch 4 times, most recently from 7921760 to ef30e24 Compare August 30, 2023 02:39

amukkara force-pushed the trie branch from ef30e24 to f8837a1 Compare September 11, 2023 17:03

amukkara added 13 commits September 12, 2023 19:47

Trie checkin

85de9c2

Coding style

c724f61

Change template parameter name

ac5af70

Includes

ffdde23

Misc coding style changes

18e9501

Remove bitvector template parameter

c5c1ec5

Remove some host-side structures

152f716

bit_vector -> dynamic_bitset

643ba31

dynamic_bitset API change

53b2ba4

Template variable naming style

fdca1c2

Minor

10f1b05

Add allocator template parameter

748c1ea

Misc coding style

d2e339b

amukkara force-pushed the trie branch from f8837a1 to 74effe6 Compare September 12, 2023 19:50

amukkara added 19 commits September 12, 2023 22:23

Run a dummy test() on bitsets

c6d300f

Use allocator for member classes

c57e491

Buffer bitset updates on host

4e4a08f

Minor changes in lookup test

6a8af91

Use iterators as parameters to insert()

248d71f

Move key generation utilities into different file

25eaa9d

Limit grid size to 128

2753a38

Add performance test

01c35df

Store labels in host vector during insertions

7151733

This improves insertion perfomance massively

Parallelize input preprocessing

9f7d6d3

Move dynamic_bitset tests into tests/trie

342bcee

Consolidate trie utils in a single file

81f9d46

find_next host-bulk API in dynamic_bitset

03821ef

Dynamic bitset benchmarks

f70d528

Consistent use of LabelType

bd4549e

A key is a list of labels. Iterators use LabelIt, rather than KeyIt.

Trie benchmarks

5f67117

Reorganize trie utils

660bdc0

Use cuco key generators

6f2b0b8

Add key length and key count nvbench axes

799d968

sleeepyjack reviewed Oct 17, 2023

View reviewed changes

amukkara added 8 commits October 18, 2023 01:19

Add namespaces

025c59a

Remove custom key comparator

6184e50

Include multiple label types

7f1d083

Both in tests and benchmarks

Change key distribution to uniform

5ed0498

Move file

6842144

Delete file read utilities

4dcb7ce

Change nvbench axes order

9892378

Trie examples

fc3fa7a

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Trie #350

Trie #350

amukkara commented Aug 11, 2023 •

edited

Loading

GPUtester commented Aug 11, 2023

rapids-bot bot commented Aug 11, 2023

PointKernel commented Aug 16, 2023

sleeepyjack commented Aug 22, 2023

copy-pr-bot bot commented Aug 30, 2023

sleeepyjack commented Sep 12, 2023

sleeepyjack commented Sep 12, 2023 •

edited

Loading

amukkara commented Sep 12, 2023

sleeepyjack left a comment

sleeepyjack Oct 17, 2023

amukkara Oct 18, 2023

sleeepyjack Oct 17, 2023

amukkara Oct 18, 2023

sleeepyjack Oct 17, 2023

amukkara Oct 18, 2023

Trie #350

Are you sure you want to change the base?

Trie #350

Conversation

amukkara commented Aug 11, 2023 • edited Loading

GPUtester commented Aug 11, 2023

rapids-bot bot commented Aug 11, 2023

PointKernel commented Aug 16, 2023

sleeepyjack commented Aug 22, 2023

copy-pr-bot bot commented Aug 30, 2023

sleeepyjack commented Sep 12, 2023

sleeepyjack commented Sep 12, 2023 • edited Loading

amukkara commented Sep 12, 2023

sleeepyjack left a comment

Choose a reason for hiding this comment

sleeepyjack Oct 17, 2023

Choose a reason for hiding this comment

amukkara Oct 18, 2023

Choose a reason for hiding this comment

sleeepyjack Oct 17, 2023

Choose a reason for hiding this comment

amukkara Oct 18, 2023

Choose a reason for hiding this comment

sleeepyjack Oct 17, 2023

Choose a reason for hiding this comment

amukkara Oct 18, 2023

Choose a reason for hiding this comment

amukkara commented Aug 11, 2023 •

edited

Loading

sleeepyjack commented Sep 12, 2023 •

edited

Loading