Improve performance of hash_character_ngrams using warp-per-string kernel #16212
Conversation
This file contains 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
davidwendt
added
2 - In Progress
|
Benchmark for |
bdice
reviewed
Jul 29, 2024
mhaseeb123
reviewed
Jul 31, 2024
cpp/src/text/generate_ngrams.cu
Outdated
| return; | ||
| } | ||
|
|
||
| namespace cg = cooperative_groups; |
There was a problem hiding this comment.
Thanks for accommodating cg changes @davidwendt. Just to confirm, do the new changes affect the performance results from before in any way?
There was a problem hiding this comment.
No. It looks like the performance is the same with cooperative groups as with the cub warp-reduce.
mhaseeb123
reviewed
Jul 31, 2024
cpp/src/text/generate_ngrams.cu
Outdated
| } | ||
|
|
||
| namespace cg = cooperative_groups; | ||
| auto warp = cg::tiled_partition<cudf::detail::warp_size>(cg::this_thread_block()); |
There was a problem hiding this comment.
Minor:
Suggested change
| auto warp = cg::tiled_partition<cudf::detail::warp_size>(cg::this_thread_block()); | |
| auto const warp = cg::tiled_partition<cudf::detail::warp_size>(cg::this_thread_block()); |
PointKernel
approved these changes
Jul 31, 2024
cpp/src/text/generate_ngrams.cu
Outdated
| } | ||
| } | ||
| auto const char_count = cg::reduce(warp, count, cg::plus<int>()); | ||
| if (lane_idx == 0) { d_counts[str_idx] = std::max(0, char_count - ngrams + 1); } |
There was a problem hiding this comment.
Suggested change
| if (lane_idx == 0) { d_counts[str_idx] = std::max(0, char_count - ngrams + 1); } | |
| if (lane_idx == 0) { d_counts[str_idx] = max(0, char_count - ngrams + 1); } |
non-blocking nit: use CUDA native functions instead of STL ones in device code since the former is hardware accelerated. cuda::std::max also works.
mhaseeb123
reviewed
Jul 31, 2024
| })); | ||
| auto [offsets, total_ngrams] = | ||
| cudf::detail::make_offsets_child_column(sizes_itr, sizes_itr + input.size(), stream, mr); | ||
| auto [offsets, total_ngrams] = [&] { |
mhaseeb123
approved these changes
Jul 31, 2024
|
/merge |
3 tasks
rapids-bot bot
pushed a commit
that referenced
this pull request
Oct 3, 2024
Fixes performance regression in `nvtext::generate_character_ngrams` introduced in #16212. Thread-per-row kernel is faster for smaller strings. Authors: - David Wendt (https://github.com/davidwendt) - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - Vukasin Milovanovic (https://github.com/vuule) - Bradley Dice (https://github.com/bdice) - Nghia Truong (https://github.com/ttnghia) URL: #16849
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.
Description
Improves the performance of
nvtext::hash_character_ngramsusing a warp-per-string kernel instead of a string per thread.Checklist