runtime: add ERMS-based memmove support for modern CPU platforms
The current memmove implementation uses REP MOVSB to copy data larger than
2KB when the useAVXmemmove global variable is false and the CPU supports
the ERMS feature.
This feature is currently only enabled on CPUs in the Sandy Bridge (Client)
, Sandy Bridge (Server), Ivy Bridge (Client), and Ivy Bridge (Server)
microarchitectures.
For modern Intel CPU microarchitectures that support the ERMS feature, such
as Ice Lake (Server), Sapphire Rapids , REP MOVSB achieves better
performance than the AVX-based copy currently implemented in memmove.
Benchstat result:
goos: linux
goarch: amd64
pkg: runtime
cpu: Intel(R) Xeon(R) Gold 6348 CPU @ 2.60GHz
│ ./old.txt │ ./new.txt │
│ sec/op │ sec/op vs base │
Memmove/2048-2 25.24n ± 0% 24.27n ± 0% -3.84% (p=0.000 n=10)
Memmove/4096-2 44.87n ± 0% 33.16n ± 1% -26.11% (p=0.000 n=10)
geomean 33.65n 28.37n -15.71%
│ ./old.txt │ ./new.txt │
│ B/s │ B/s vs base │
Memmove/2048-2 75.56Gi ± 0% 78.59Gi ± 0% +4.02% (p=0.000 n=10)
Memmove/4096-2 85.01Gi ± 0% 115.05Gi ± 1% +35.34% (p=0.000 n=10)
geomean 80.14Gi 95.09Gi +18.65%
Fixes #66958
Change-Id: I1fafd1b51a16752f83ac15047cf3b29422a79d5d
GitHub-Last-Rev:
89cf5af32b1b41e1499282058656a8a5c7aed359
GitHub-Pull-Request: golang/go#66959
Reviewed-on: https://go-review.googlesource.com/c/go/+/580735
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>