No, it's unambiguously bad. You're transmitting a cleartext password to a system which doesn't have a business need to know it, and which wasn't designed to process secret data. There's a substantial risk that the database may leak that data in some unexpected way, e.g. by logging it when an error occurs or by showing the parameter in a process list. Worse, a stored procedure can potentially be covertly modified to store or exfiltrate the password while hashing it.
"As early as possible" is interesting. It could be done on the client, however you need to actually have the hash to check if a string hashes to the same as it because of the salts embedded in the strings. Using an algorithm without salting would allow you to hash the password on the client then allow arbitrary 32 (or however long your hash digests are) character strings on the server as a password equivalent. This would also protect from a misconfigured or malicious server logging unhashed passwords.
No, that's actually too early -- if you let the client hash the password, you're vulnerable to a "pass-the-hash" attack where the client submits a hash without knowing the password.
There are protocols like SRP and other PAKEs which allow this to be done securely, but it's uncommon for them to be used in web applications.