In some other project I found out that cgo has a significant overhead in calling functions in C. This is a simple benchmark to demonstrate that.
An observant reader will note that I'm calling my functions weirdly. This is to defeat the compiler. At the first attempt I did this the overhead of Go function calls was almost zero and that's because the compiler helpfully inlined everything which defeated the whole purpose of the benchmark.
Even with go:noinline the compiler still happily inlines empty
functions. Not sure what can be done about it. Making the functions do
something requires a non-trivial amount of "something" to make them
not inline and the old tricks (like switch{}) don't seem to work
anymore. defer still works, but then the benchmark tests defer which
is also a bit in Go that isn't very fast.
This is the best I can do and it isn't great, but it shows the magnitude difference at least.
BenchmarkCFuncall-4 20000000 72.2 ns/op
BenchmarkGoFuncall-4 1000000000 2.03 ns/op
BenchmarkCFuncall-4 10000000 184 ns/op
BenchmarkGoFuncall-4 1000000000 2.00 ns/op