runtime: improve memmove on ppc64x/power10
Rewrite memmove asm function to use the new power10 instructions
lxvl and stxvl or the load and store vector with length which can
specify the number of bytes to be loaded/stored in a register,
thereby avoiding multiple instructions to process 8bytes, 4bytes,
2bytes and a single byte while storing the tail end bytes. On power9
and power8 the code remains unchanged.
The performance for all sizes<=16 improve on power10 with this change.
name old time/op new time/op delta
Memmove/1 2.87ns ± 0% 2.64ns ± 1% -8.11%
Memmove/2 2.85ns ± 0% 2.62ns ± 1% -8.12%
Memmove/3 2.78ns ± 0% 2.63ns ± 1% -5.33%
Memmove/4 2.83ns ± 0% 2.63ns ± 2% -7.33%
Memmove/5 2.78ns ± 0% 2.63ns ± 1% -5.40%
Memmove/6 2.61ns ± 3% 2.61ns ± 1% ~
Memmove/7 2.82ns ± 0% 2.61ns ± 1% -7.48%
Memmove/8 2.82ns ± 0% 2.65ns ± 1% -6.11%
Memmove/9 6.41ns ± 0% 2.62ns ± 1% -59.17%
Memmove/10 5.09ns ± 1% 2.60ns ± 1% -48.90%
Memmove/11 4.68ns ± 7% 2.59ns ± 1% -44.56%
Memmove/12 6.25ns ± 2% 2.60ns ± 1% -58.46%
Memmove/13 4.15ns ± 25% 2.59ns ± 1% -37.66%
Memmove/14 3.76ns ± 11% 2.59ns ± 1% -30.94%
Memmove/15 3.82ns ± 1% 2.60ns ± 1% -31.93%
Memmove/16 2.96ns ± 1% 2.59ns ± 1% -12.63%
MemmoveUnalignedDst/1 3.07ns ± 0% 2.77ns ± 0% -9.75%
MemmoveUnalignedDst/2 2.82ns ± 0% 2.77ns ± 0% -1.73%
MemmoveUnalignedDst/3 3.03ns ± 0% 2.77ns ± 0% -8.75%
MemmoveUnalignedDst/4 2.85ns ± 1% 2.77ns ± 0% -2.90%
MemmoveUnalignedDst/5 3.03ns ± 0% 2.77ns ± 0% -8.75%
MemmoveUnalignedDst/6 2.88ns ± 0% 2.77ns ± 0% -4.04%
MemmoveUnalignedDst/7 3.11ns ± 0% 2.77ns ± 0% -11.10%
MemmoveUnalignedDst/8 4.18ns ± 2% 2.77ns ± 0% -33.90%
MemmoveUnalignedDst/9 6.36ns ± 1% 2.77ns ± 0% -56.53%
MemmoveUnalignedDst/10 5.77ns ± 1% 2.77ns ± 0% -52.09%
MemmoveUnalignedDst/11 4.68ns ± 1% 2.77ns ± 0% -40.86%
MemmoveUnalignedDst/12 4.54ns ± 2% 2.77ns ± 0% -39.05%
MemmoveUnalignedDst/13 6.16ns ± 5% 2.77ns ± 0% -55.14%
MemmoveUnalignedDst/14 4.03ns ± 2% 2.77ns ± 0% -31.41%
MemmoveUnalignedDst/15 4.11ns ± 0% 2.77ns ± 0% -32.74%
MemmoveUnalignedDst/16 3.49ns ± 4% 2.79ns ± 1% -20.04%
MemmoveUnalignedSrc/1 3.06ns ± 0% 2.77ns ± 0% -9.68%
MemmoveUnalignedSrc/2 2.82ns ± 1% 2.77ns ± 0% -1.93%
MemmoveUnalignedSrc/3 3.04ns ± 0% 2.77ns ± 0% -8.95%
MemmoveUnalignedSrc/4 2.85ns ± 0% 2.77ns ± 0% -2.86%
MemmoveUnalignedSrc/5 3.04ns ± 0% 2.77ns ± 0% -8.97%
MemmoveUnalignedSrc/6 2.93ns ± 0% 2.77ns ± 0% -5.43%
MemmoveUnalignedSrc/7 3.13ns ± 0% 2.77ns ± 0% -11.56%
MemmoveUnalignedSrc/8 3.71ns ± 2% 2.77ns ± 0% -25.46%
MemmoveUnalignedSrc/9 6.04ns ± 0% 2.77ns ± 0% -54.16%
MemmoveUnalignedSrc/10 6.86ns ± 5% 2.77ns ± 0% -59.69%
MemmoveUnalignedSrc/11 4.18ns ± 3% 2.77ns ± 0% -33.81%
MemmoveUnalignedSrc/12 4.75ns ± 2% 2.77ns ± 0% -41.81%
MemmoveUnalignedSrc/13 4.78ns ± 3% 2.77ns ± 0% -42.15%
MemmoveUnalignedSrc/14 3.89ns ± 5% 2.77ns ± 0% -28.80%
MemmoveUnalignedSrc/15 4.09ns ± 0% 2.77ns ± 0% -32.30%
MemmoveUnalignedSrc/16 3.15ns ± 1% 2.77ns ± 0% -12.05%
Change-Id: Ia3c09d968dada71a794e5ccab3300ea9c46d8374
Reviewed-on: https://go-review.googlesource.com/c/go/+/470135
Run-TryBot: Archana Ravindar <aravind5@in.ibm.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>