tree daf3fee03b35231a1902eace1b9a0537639903eb
parent d6320f1a58f1f7820daee06a086c83a0274a777f
author Lynn Boger <laboger@linux.vnet.ibm.com> 1649443800 -0500
committer Lynn Boger <laboger@linux.vnet.ibm.com> 1649771646 +0000

runtime: improve memmove for ppc64x

This improves performance of memmove for larger moves by
unrolling the main loop from 32 byte to 64 byte moves.

The improvement of the relevant sizes on a power9:

Memmove/64      5.11ns ± 0%    5.00ns ± 0%   -2.21%
Memmove/128     8.26ns ± 0%    5.88ns ± 0%  -28.83%
Memmove/256     12.7ns ± 0%     8.6ns ± 0%  -31.94%
Memmove/512     17.9ns ± 0%    14.3ns ± 0%  -19.87%
Memmove/1024    33.3ns ± 0%    27.0ns ± 0%  -18.92%
Memmove/2048    72.1ns ± 0%    51.8ns ± 0%  -28.25%
Memmove/4096     126ns ± 0%     110ns ± 0%  -12.63%


Change-Id: I74162a9f152d7752a8281da1b89a66da99a3fdc9
Reviewed-on: https://go-review.googlesource.com/c/go/+/399499
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>