What is the right way to determine, in CGo, if the environment supports AVX-512? - x86-64

Question

CGo appears to be removing preprocessor definitions. Has anyone else run into this?
In my other C code on the same machine, this works fine. In CGo, the "__AVX2__" definition gets removed.
I've tried this with and without the '-march=native'
How can I get it back ?
The code:
//go:build amd64 && !purego && gc
// +build amd64,!purego,gc
// #cgo CFLAGS: -march=native
package keccakAVX512
/*
// check if the amd64 machine supports AVX512 instructions at build time and call
// an assembly function using AVX512 if so.
#ifdef __AVX2__
void keccakF1600_AVX512(unsigned long* state);
this should fail with a syntax error but doesn't
#endif
// void keccakF1600_AVX512(unsigned long* state) {};
*/
import "C"
func KeccakF1600AMDAVX512(a *[25]uint64) {
C.keccakF1600_AVX512((*C.ulong)(&a[0]))
}
With the 'go tool cgo' command, I see that the default '-m64' is taking precedence, even with the CFLAGS set.

Answer 1

I found that the problem was in the #cgo syntax. It has to be inside the '/**/' comment block instead of the '//' comment block.
package keccakAVX512
/*
#cgo CFLAGS: -march=native
// check if the amd64 machine supports AVX512 instructions at build time and call
// an assembly function using AVX512 if so.
#ifdef __AVX2__
void keccakF1600_AVX512(unsigned long* state);
#else
void keccakF1600_AVX512(unsigned long* state) {};
#endif
*/
import "C"
func KeccakF1600AMDAVX512(a *[25]uint64) {
C.keccakF1600_AVX512((*C.ulong)(&a[0]))
}