reflect: cache IsVariadic calls in Call
These calls are cacheable, so do that to avoid doing extra work.
This opportunity was discovered while taking a look at a CPU profile
while investigating #7818.
I added a BenchmarkCallMethod, which is similar to BechmarkCall but
for a method receiver.
Benchmark results, including the new BenchmarkCallMethod:
name old time/op new time/op delta
Call-16 22.0ns ±19% 20.2ns ±17% -8.08% (p=0.000 n=40+40)
CallMethod-16 100ns ± 3% 91ns ± 2% -9.13% (p=0.000 n=40+39)
CallArgCopy/size=128-16 15.7ns ± 1% 14.3ns ± 4% -8.98% (p=0.000 n=38+37)
CallArgCopy/size=256-16 15.9ns ± 3% 15.0ns ± 5% -6.12% (p=0.000 n=39+39)
CallArgCopy/size=1024-16 18.8ns ± 6% 17.1ns ± 6% -9.03% (p=0.000 n=38+38)
CallArgCopy/size=4096-16 26.6ns ± 3% 25.2ns ± 4% -5.19% (p=0.000 n=39+40)
CallArgCopy/size=65536-16 379ns ± 3% 371ns ± 5% -2.11% (p=0.000 n=39+40)
name old alloc/op new alloc/op delta
Call-16 0.00B 0.00B ~ (all equal)
CallMethod-16 0.00B 0.00B ~ (all equal)
name old allocs/op new allocs/op delta
Call-16 0.00 0.00 ~ (all equal)
CallMethod-16 0.00 0.00 ~ (all equal)
name old speed new speed delta
CallArgCopy/size=128-16 8.13GB/s ± 1% 8.92GB/s ± 4% +9.77% (p=0.000 n=38+38)
CallArgCopy/size=256-16 16.1GB/s ± 3% 17.1GB/s ± 5% +6.56% (p=0.000 n=39+39)
CallArgCopy/size=1024-16 54.6GB/s ± 6% 60.1GB/s ± 5% +9.93% (p=0.000 n=38+38)
CallArgCopy/size=4096-16 154GB/s ± 5% 163GB/s ± 4% +5.63% (p=0.000 n=40+40)
CallArgCopy/size=65536-16 173GB/s ± 3% 177GB/s ± 5% +2.18% (p=0.000 n=39+40)
Updates #7818.
Change-Id: I94f88811ea9faf3dc2543984a13b360b5db66a4b
GitHub-Last-Rev:
9bbaa1854aa32cade905eddb05737e6224c372a1
GitHub-Pull-Request: golang/go#43475
Reviewed-on: https://go-review.googlesource.com/c/go/+/281252
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Daniel Martí <mvdan@mvdan.cc>
Trust: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Go Bot <gobot@golang.org>