[v3] Further tweak sso_string::_M_swap

[Paolo Carlini]

Hi,

I'm happy to report that with this tweak, on testcases like
moveable_vector_insert.cc, thanks to move semantics on the v7-branch we
are typically 2 times faster than our "old" reference-counted
implementation, both for short and long strings.

Tested x86-linux, both mainline and v7-branch (which uses sso_string by
default)

Paolo.

///////////////////
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: CL_swap
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20051103/f37b13b9/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_swap
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20051103/f37b13b9/attachment-0001.ksh>