PATCH: Improve 64 bit memchr, memrchr, rawmemchr with SSE2

[Dmitrieva Liubov <liubov dot dmitrieva at gmail dot com>]

Hello,

This patch improves 64 bit memchr, rawmemchr, memrchr functions for all
64 bit architectures.

Main feature is using 64bytes loop with pmaxub instructions.

For example, it improves performance:

-- for memchr
---- by up to 2X on atom, up to 2X on core7.

-- for memrchr
---- by up to 4.5X on atom, up to 6X on core7.

-- for rawmemchr
---- by up to 2X on atom, up to 6.4X on core7.

"Make check" was run successfully on my atom and core2 machines, I
built rpms and checked them as well.

2011-08-31  Liubov Dmitrieva  <liubov.dmitrieva@gmail.com>

	* sysdeps/x86_64/multiarch/rawmemchr.S: Update.
	Use new sse2 version for core i3 - i7 as it's faster
	than sse42 version.
	(bit_Prefer_PMINUB_for_stringop): New.

	* sysdeps/x86_64/rawmemchr.S: Update.
	Replace with faster SSE2 version.
	* sysdeps/x86_64/memrchr.S: New file.
	* sysdeps/x86_64/memchr.S: Update.
	Replace with faster SSE2 version.

--
Liubov Dmitrieva
Software engineer
Intel Corporation

Attachment: memchr_family_64.patch
Description: Binary data

Attachment: ChangeLog.memchr_family_64
Description: Binary data