This seems unlikely. (Or maybe that's the reason given, but it seems implausible to be true.) Competitors have more than enough know how to determine if you're infringing a patent, source or no source.

The US legal system has discovery for patent cases. You can sue and then subpoena their code base to confirm whether they are in violation before going to court (and really racking up the fees). Since these are US companies I think that it's more likely fear that others would see the horrible hacks or clever trade secrets.

In other countries (most of Asia), where there is no discivery, it's almost impossible to prove hardware or software patent violations so your case is kicked out of court immediately, even if your patent claims are valid and their product reads into your claims. That's why most patent suits end up in the US or Europe (or in the even faster ITC import injunction).

> Competitors have more than enough know how to determine if you're infringing a patent

How do you figure? The detectability of infringement is a key factor in deciding whether to file a patent.

Regardless, copyright infringement might also be an issue.

Well, it's hard to be specific since nobody ever mentions which patents they're talking about. But I would assume that somebody at AMD has the skills to determine whether Nvidia uses the "good matrix" technique, or whatever it is people are assuming AMD patented that Nvidia is trying to hide.

In court, there's a big difference between your assumption vs. published code.

Drivers are published code, freely available. If the driver does something, and you can't find an expert witness to testify it does what it does, maybe it doesn't do what you think?

Reading assembly with no comments is a different skill than reading C or C++, especially since it's not always clear what's an instruction and what's data, but it's still reading code, and there are tools to help you trace through it. And most of the drivers aren't writing code like it's a 64k demo (where the code is the data, and the data is the code, and they both modify each other)

But they need evidence.

Reverse engineering silicon to figure out if you used a specific type of patented algorithm is super hard.

Looking at open source code is waaaay easier.