Avoid default-init overhead for lookup map in assemble_pre #122
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
With e3b3545 std::vector started to take care of the memory for the lookup map in assemble_pre to avoid having to call allocate and deallocate manually. Unfortunately, std::vector also default initializes all values on construction, meaning that this change introduced an overhead of always zeroing the map values initially. Indeed it does not need to be zero'd, as all used values are set upon building it. As it needs to be of size n to be able to map any row index, zeroing it can actually be quite costly, especially since it is repeated for each assemble_pre. Very noticeable for example on DIMACS10/tx2010 from SuiteSparse matrix collection. This replaces the std::vector with std::unique_ptr to avoid the repeated unnecessary zeroing but still keep the implementation RAII.
|
Hm, I am unable to reproduce the pipeline failure for gcc-9, any way to get more information? Edit: Oh, it's the unrelated |
|
@mjacobse I’ve tried a re-run, sometimes the VM hardware is a bit flaky… (and SSMFE is not the most reliable). |
jfowkes
approved these changes
Jul 17, 2023
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Outdated suggestions cannot be applied.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.
With e3b3545,
std::vectorstarted to take care of the memory for the lookup map inassemble_preto avoid having to callallocateanddeallocatemanually. Butstd::vectoralso default initializes all values on construction, meaning that this change introduced an overhead of always zeroing the map values initially.Indeed it does not need to be zero'd, as all used values are set upon building it. As it needs to be of size
nto be able to map any row index, zeroing it can actually be somewhat costly, especially since it is repeated for eachassemble_pre. This change replaces thestd::vectorwithstd::unique_ptrto avoid the repeated unnecessary zeroing but still keep the implementation RAII.Here and in the plots below are some benchmarks with the same setup as in #119, except that repetitions for a matrix are run at different times and on random cores to try to highlight the baseline noise. For most test matrices I see no measureable change, but for a few larger matrices (especially the DIMACS10 group) there are small but significant performance improvements.
Note that e3b3545 only changed
assemple_pre, so this only deals with that function too.assemble_postdoes not use an RAII solution yet. If desired it should be possible to introducestd::unique_ptrfor it as well, supposedly without any influence on performance.