[PATCH RFC] Imporve 64bit memcpy performance for Haswell CPU with AVX instruction

[Ling Ma]

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 the whole
performance from 2% to 10%, and 12~ 60% when copy size is over 256bytes.

This version is based on latest ChangeLog,  and simplified memmove
according to Ondra's comment.

Thanks
Ling
-------------- next part --------------
A non-text attachment was scrubbed...
Name: memcpy-avx-unaligned.patch.tar.gz
Type: application/x-gzip
Size: 4617 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20140714/6168d1c9/attachment.bin>