You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
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
The semi_join.cu takes about 6 minutes to compile on my Linux 18.04 desktop when doing a full build of libcudf. The join_kind template parameter used internally in cudf::detail::left_semi_anti_join for left_semi_join and left_anti_join APIs is not used in a constexpr or to pass to any other templated function. This PR moves the template parameter to a runtime parameter on the detail functions reducing the compile time for semi_join.cu by ~2x.
Another improvement includes un-inlining the is_trivial_join utility function to reduce the compile time for files that include join_common_utils.hpp.
Finally, the device vector used as a gather map in detail::left_semi_anti_join was wrapped with a column_view in order to call detail::gather without iterators. This allowed not including the heavy gather.cuh. This improved the compile time about 10% and reduced the object file semi_join.cu.o size by 2x.
Merging #8914 (be7eab5) into branch-21.10 (18f7c01) will decrease coverage by 0.08%.
The diff coverage is n/a.
❗ Current head be7eab5 differs from pull request most recent head 2270b90. Consider uploading reports for the commit 2270b90 to get more accurate results
In #8815 I also created a join_common_utils.cuh header for code that's common for normal and conditional joins in the CUDA layers. A significant chunk of that file is also inline functions. I doubt they'll have as significant of an impact, but perhaps it makes sense to break some of those out of the header as well so that they don't get recompiled for both sets of join TUs? I believe at least a few of them are not templated.
In #8815 I also created a join_common_utils.cuh header for code that's common for normal and conditional joins in the CUDA layers. A significant chunk of that file is also inline functions. I doubt they'll have as significant of an impact, but perhaps it makes sense to break some of those out of the header as well so that they don't get recompiled for both sets of join TUs? I believe at least a few of them are not templated.
Yes, thanks. I see there are 3 host-only utility functions that could be moved to a .cpp or .cu file. Perhaps I could add a join_utils.cpp for these and the is_trivial_join utility as well.
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.
davidwendt
added
2 - In Progress
The
semi_join.cutakes about 6 minutes to compile on my Linux 18.04 desktop when doing a full build of libcudf. Thejoin_kindtemplate parameter used internally incudf::detail::left_semi_anti_joinforleft_semi_joinandleft_anti_joinAPIs is not used in aconstexpror to pass to any other templated function. This PR moves the template parameter to a runtime parameter on the detail functions reducing the compile time forsemi_join.cuby ~2x.Another improvement includes un-inlining the
is_trivial_joinutility function to reduce the compile time for files that includejoin_common_utils.hpp.Finally, the device vector used as a gather map in
detail::left_semi_anti_joinwas wrapped with acolumn_viewin order to calldetail::gatherwithout iterators. This allowed not including the heavygather.cuh. This improved the compile time about 10% and reduced the object filesemi_join.cu.osize by 2x.