Faster (but unsafer) Cgo calls remove some extra assembly instruments.
Fast (but unsafe) Cgo calls via an assembly trampoline. Inspired by
rustgo. The UnsafeCall* function
takes a pointer to a Cgo function and a number of arguments and
executes the function on the "system stack" of the thread the current
Goroutine is running on. Usage looks like:
package hello
// #include <stdint.h>
// #include <stdio.h>
// void hello(uint64_t arg) {
// printf("Hello, C: %lld\n", arg)
// }
import "C"
import (
"github.com/petermattis/fastcgo"
)
func Hello(arg uint64) {
fastcgo.UnsafeCall4(C.hello, arg, 0, 0, 0)
}
The upshot is significantly faster transitions into C/C++:
go test -run=- -bench=. ./bench/
...
goos: darwin
goarch: amd64
pkg: github.com/choleraehyq/fastercgo/bench
BenchmarkGO-4 2000000000 1.28 ns/op
BenchmarkCGO-4 30000000 51.6 ns/op
BenchmarkFastCGO-4 300000000 4.34 ns/op
BenchmarkFasterCGO-4 500000000 3.81 ns/op
...
WARNING: No attempt is made to play nice with the Go scheduler or GC. The goroutine executing a fastcgo call will never be interrupted and will delay the starting or completion of GC until it finishes.