This is an archive of the discontinued LLVM Phabricator instance.

Use one task per iteration in parallel_for_loop
ClosedPublic

Authored by • rafael on Nov 15 2016, 11:59 AM.

Reviewers

ruiu
grimar

Summary

This seems far more natural. A user can create larger chunks if the overhead is too large.

With this linking xul with "--threads --build-id=sha1 goes from 13.938177535 to 11.035953538 seconds on linux.

• rafael retitled this revision from to Use one task per iteration in parallel_for_loop.

• rafael updated this object.

• rafael added reviewers: ruiu, grimar.

Ah, okay, you already created a patch. LGTM.

This revision is now accepted and ready to land.Nov 15 2016, 1:29 PM

LGTM.

Was committed as r287042

Path

Size

include/

lld/

Core/

8 lines

	Show First 20 Lines • Show All 278 Lines • ▼ Show 20 Lines
	template <class Iterator, class Func>			template <class Iterator, class Func>
	void parallel_for_each(Iterator begin, Iterator end, Func func) {			void parallel_for_each(Iterator begin, Iterator end, Func func) {
	concurrency::parallel_for_each(begin, end, func);			concurrency::parallel_for_each(begin, end, func);
	}			}
	#else			#else
	template <class Iterator, class Func>			template <class Iterator, class Func>
	void parallel_for_each(Iterator begin, Iterator end, Func func) {			void parallel_for_each(Iterator begin, Iterator end, Func func) {
	TaskGroup tg;			TaskGroup tg;
	ptrdiff_t taskSize = 1024;			for (; begin != end; ++begin)
	while (taskSize <= std::distance(begin, end)) {			tg.spawn([=, &func] { func(*begin); });
	tg.spawn([=, &func] { std::for_each(begin, begin + taskSize, func); });
	begin += taskSize;
	}
	std::for_each(begin, end, func);
	}			}
	#endif			#endif
	} // end namespace lld			} // end namespace lld

	#endif // LLD_CORE_PARALLEL_H			#endif // LLD_CORE_PARALLEL_H