improve one_size_heap_list: use rwlock to speedup the allocation/free

[JiakunYan]

The original implementation uses an HPX mutex to ensure thread safety. This PR replaces it with a pthread rwlock, which greatly improves its performance. Multiple allocation/free calls can proceed simultaneously under the read lock, and the pthread read lock only costs an atomic fetch-and-add in most cases.

Need to discuss: directly using pthread rwlock in HPX codebase may not be a good idea, but implementing the whole pthread rwlock algorithm in HPX can also be cumbersome.

[StellarBot]

Can one of the admins verify this patch?

[codacy-production]

Coverage summary from Codacy

See diff coverage on Codacy

Coverage variation	Diff coverage
-0.05%	91.27%

Coverage variation details

	Coverable lines	Covered lines	Coverage
Common ancestor commit (5ffdfbe)	193075	164320	85.11%
Head commit (49e0553)	189172 (-3903)	160902 (-3418)	85.06% (-0.05%)

Coverage variation is the difference between the coverage for the head and common ancestor commits of the pull request branch: <coverage of head commit> - <coverage of common ancestor commit>

Diff coverage details

	Coverable lines	Covered lines	Diff coverage
Pull request (#6370)	355	324	91.27%

Diff coverage is the percentage of lines that are covered by tests out of the coverable lines that the pull request added or modified: <covered lines added or modified>/<coverable lines added or modified> * 100%

See your quality gate settings    Change summary preferences

[hkaiser]

@JiakunYan I think we can improve the implementation of our shared_mutex by placing its state data into a single atomic and start locking things internally only if really needed; i.e. replace

hpx/libs/core/synchronization/include/hpx/synchronization/shared_mutex.hpp

Lines 29 to 35 in 103a7b8

	struct state_data
	{
	unsigned shared_count;
	bool exclusive;
	bool upgrade;
	bool exclusive_waiting_blocked;
	};

[JiakunYan]

@hkaiser I did some brief benchmarking using Octotiger on Rostam. The performance of the new shared lock is pretty good! (around 5% improvement for a specific octotiger configuration)

[hkaiser]

@JiakunYan this looks good now and the test errors are unrelated. Can this be merged now?

[JiakunYan]

@JiakunYan this looks good now and the test errors are unrelated. Can this be merged now?

Sure!

[hkaiser]

@JiakunYan this looks good now and the test errors are unrelated. Can this be merged now?

Sure!

Could you fix the conflicts, please?

[JiakunYan]

@JiakunYan this looks good now and the test errors are unrelated. Can this be merged now?

Sure!

Could you fix the conflicts, please?

I tried rebasing this PR to the current master branch. However, git is behaving somehow weirdly. It kept missing to apply some changes in the Trying to streamline shared_mutex. Could you try rebasing it on your end and see what will happen?

[hkaiser]

@JiakunYan this looks good now and the test errors are unrelated. Can this be merged now?

Sure!

Could you fix the conflicts, please?

I tried rebasing this PR to the current master branch. However, git is behaving somehow weirdly. It kept missing to apply some changes in the Trying to streamline shared_mutex. Could you try rebasing it on your end and see what will happen?

If I rebase my branch, then you will not be able to merge it to your branch because of conflicts. I'll resolve the conflicts manually and merge your work manually as well. Thanks!