Add versions of wcscpy, wcschr, wcsrchr for power6/power7.
commit15d0da8cb36f5c96a3c649d29a2e2623e995cd49
authorWill Schmidt <will_schmidt@vnet.ibm.com>
Wed, 22 Aug 2012 16:04:42 +0000 (22 11:04 -0500)
committerRyan S. Arnold <rsa@linux.vnet.ibm.com>
Wed, 22 Aug 2012 16:04:42 +0000 (22 11:04 -0500)
tree3c1ccdcbc8fb94fb795fc13e519e02957d02e626
parentbcca089526c6859e775243731037a469aec3065c
Add versions of wcscpy, wcschr, wcsrchr for power6/power7.

Initially based on the versions found in wcsmbs/* ; these files have
been changed by hand unrolling, and adding some additional variables
to allow some read-ahead to occur, which then relieves some of the
wait-for-increment/wait-for-load/wait-for-compare-results pressure
that was slowing down every iteration through the while-loop.

For 64-bit Power7, These changes give an approx 20% throughput boost
for the wcschr and wcsrchr functions; and approx 40% boost for the
wcscpy function.  32-bit improvements appear to be slightly better
with ~ %30 and ~ %45 respectively.  Results for Power6 closely match
those for power7.
ChangeLog
NEWS
sysdeps/powerpc/powerpc32/power6/wcschr.c [new file with mode: 0644]
sysdeps/powerpc/powerpc32/power6/wcscpy.c [new file with mode: 0644]
sysdeps/powerpc/powerpc32/power6/wcsrchr.c [new file with mode: 0644]
sysdeps/powerpc/powerpc64/power6/wcschr.c [new file with mode: 0644]
sysdeps/powerpc/powerpc64/power6/wcscpy.c [new file with mode: 0644]
sysdeps/powerpc/powerpc64/power6/wcsrchr.c [new file with mode: 0644]