runtime: optimize memmove for 1-16 MB overlapping case on AMD64

[weishi-deng]

What version of Go are you using (go version)?

$ go1.16.4

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

GO111MODULE="auto"
GOARCH="amd64"
GOBIN=""
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOINSECURE=""
GONOPROXY=""
GONOSUMDB=""
GOOS="linux"
GOPRIVATE=""
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/usr/lib/golang"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/usr/lib/golang/pkg/tool/linux_amd64"
GCCGO="gccgo"
AR="ar"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD=""
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build287939000=/tmp/go-build -gno-record-gcc-switches"

What did you do?

I run some test cases using function 'runtime.memmove' when data size is over 1MB and below 16MB with address overlap.

What did you expect to see?

'runtime.memmove' choose the most efficient way (from the non-temporal store and temporal store) to copy data.

What did you see instead?

When the test case is with address overlap and the size is over 1MB and below 16MB, the non-temporal store copying is slower than temporal store copying, but 'runtime.memmove' chooses to copy data using non-temporal store copying.

[gopherbot]

Change https://golang.org/cl/356749 mentions this issue: runtime: use non-temporal store copying for size over 16MB and use temporal store copying for size is 1MB ~ 16MB in'memmove'.