git.ipfire.org Git - thirdparty/glibc.git/commit

]> git.ipfire.org Git - thirdparty/glibc.git/commit - ChangeLog

projects / thirdparty / glibc.git / commit

author	Ling Ma <ling.ml@alibaba-inc.com>
	Mon, 14 Jul 2014 04:02:52 +0000 (00:02 -0400)
committer	H.J. Lu <hjl.tools@gmail.com>
	Wed, 30 Jul 2014 15:02:35 +0000 (08:02 -0700)
commit	05f3633da4f9df870d04dd77336e793746e57ed4
tree	bd980df4702ffd9ebfa9cb4dac8823a659f0c430	tree \| snapshot
parent	a53fbd8e6cd2f69bdfa3431d616a5f332aea6664	commit \| diff

Improve 64bit memcpy performance for Haswell CPU with AVX instruction

In this patch we take advantage of HSW memory bandwidth, manage to
reduce miss branch prediction by avoiding using branch instructions and
force destination to be aligned with avx instruction.

The CPU2006 403.gcc benchmark indicates this patch improves performance
from 2% to 10%.

12 files changed:

ChangeLog		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/Makefile		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/ifunc-impl-list.c		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/memcpy-avx-unaligned.S	[new file with mode: 0644]	blob
sysdeps/x86_64/multiarch/memcpy.S		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/memcpy_chk.S		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/memmove-avx-unaligned.S	[new file with mode: 0644]	blob
sysdeps/x86_64/multiarch/memmove.c		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/memmove_chk.c		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/mempcpy-avx-unaligned.S	[new file with mode: 0644]	blob
sysdeps/x86_64/multiarch/mempcpy.S		diff \| blob \| blame \| history
sysdeps/x86_64/multiarch/mempcpy_chk.S		diff \| blob \| blame \| history

A mirror of the official glibc repository

RSS Atom