But since I use secp256k1 curve only for testing and research I do no care much for any of possible vulnerabilities and attacks.
The safest (not necessary the fastest) secp256k1 is the one used in Bitcoin Core. But I don't use it because I keep getting wrong answers when I do arithmetic. Maybe the privkey bytes are not being filled correctly or something.
Perhaps you should open an issue or discussion topic (
https://github.com/bitcoin-core/secp256k1/issues or
https://github.com/bitcoin-core/secp256k1/discussions/categories/q-a) on the library, because that's definitely not supposed to happen.
To update this thread, it appears to be that the multiplication has the secp256k1 characteristic modulus hardcoded into the algorithm, making it suitable only for public key multiplication.
I was thinking about making an adapted version of the multiplication algorithm that uses the curve order (subtracted from 2^256) in its place, so that private key multiplication is covered as well.
I haven't explored the 10x26 legs imply too deep, but I'm assuming it's a similar case.
Since you're here though, let me ask: Was the multiplication assembly (and possibly the C version of it in another file using int128) only intended for public keys?