(I just realized: you may be in the dubious position of being less pragmatic about something than Richard Stallman. It's not often someone can say that!)
It's silly to assume no change in behavior, any legal counsel from that point on will advise in favor of patenting the hardware implementation, and instead of patented software running on commodity hardware we'll end up with patented "appliances" and "reference hardware implementations" with broad claims extending to other types of hardware implementations.
If you start an arguement at two extremes, and give ground where your opponents do not, you weaken your position and put the compromise between them more in their favor. And when you keep giving ground, you keep showing a willingness to concede defeat, and your opponents will just ignore you. If you stand your ground, you are a stalwart wall of opposition that needs to be confronted, rather than ignored, eventually.
In many cases (gay rights, civil liberties, patent law, equality / freedom) that stalwart position may require multiple generations. You don't get instant results when dealing with the entrenched status quo that is horribly wrong. But when you give ground you render your entire argument and stance illegitimate because your opponents can dismiss you as you keep conceding ground and giving up on your positions.
Patents, as a whole, are archaic and bad. The turnaround time for any new discovery is now a matter of days, or months, rather than the years it would take in centuries past to distribute and market goods and services. In the digital world, everything is already information, it is all number patterns in electrical terms, and we had a track record of not letting people monopolize knowledge and information in the realm of mathematics before, and the digital era should never have been any different in my opinion. Of course, it is my opinion and all. But I remain dilligent in my stance for it.
I have lost a lot of respect for Stallman because it seems in recent years his dialogue has propagated a lot of "buts" in his arguments. He concedes ground on what he believes in and that illegitimizes him more than anyone calling him cookey would in my book.
In that state of the world, if they patent the hardware solution, I don't care. I can still implement in software and I have a legal workaround to avoid infringing. I'm driving the truck through the hole.
Apple might be able to patent multi touch if they hypothetically invented it, since that's a hardware thing.
But once that touch data makes it to the CPU, what is done with it afterwards would be unpatentable.
Even if Apple themselves somehow did slide to unlock in hardware, competitors could still implement a software version.
Actually it's quite helpful. Something implemented in hardware has one fixed royalty per device, no matter what operating system or other customization has been performed. That implementation becomes a capability of the device, a manufacturers selling point. A software implementation is "shipped" over and over again as the software is updated, the user changes software packages, etc.
I'm pretty ambivalent on hardware patents, having grown up reading Don Lancaster, etc. But they could be a workable way to fund codec development while not encroaching on software freedom.
Sure, why wouldn't it be? I'm not suggesting his idea is feasible or not. But as he stated it, in this case if you were able to reverse engineer the software bits required to make the magic happen on the chip and reimplement your own version you wouldn't be on the hook for actionable patent infringement whether you were the developer or a user.
OTOH you wouldn't be able to implement the hardware side of that system under these rules.
FPGA is pretty much general purpose metahardware. It's programmable so you could program it to be any chip, thus not covered by the effects of a patent.
Sufficiently popular software algorithms like mpeg4 (including avc) are often implemented in ASICs for speed. Stallman's suggestion does nothing to help in those cases.
If you're manufacturing or buying dedicated hardware to do one specific task well I don't see why would patents need not apply? The codec chip is clearly a "hardware machine", not a general purpose processor. It's something I imagine companies are prepared to pay license fees for. Further, nobody will accidentally use a MPEG4 hardware decoder in some device they make which is a different situation from how the accidental problems manifest in software.
A clear and strong reform of patent would be preferable, and has been for long the free software foundations strategy. In the case of Bilski v. Kappos, one can read a brief made by the fsf, which rather clearly presents their view. I also strongly promote the idea to watch/read a talk by eben moglen at https://www.softwarefreedom.org/events/2009/Moglen_Cardozo/C...
As for the complaints that this might create more legal absurdities, there are plenty of those already. An imperfect solution is better than none.
Or maybe he's had a change in opinion recently. I have to say, this is the only cogent, practical argument I've heard out of him over the last few years. I really dislike his dogma on the free software issue.
He has principles that he tries to live by and advocates strongly but I have never viewed him as an idealist nor do I see much evidence for it.
FPGAs? Is an algorithm implemented on a FPGA a software or hardware implementation?
Sufficiently popular software algorithms like mpeg4 (including avc) are often implemented in ASICs for speed. Stallman's suggestion does nothing to help in those cases.
What if Intel introduces new instructions that assist with some patented algorithm but do not carry out the complete algorithm? Software completes the algorithm using the chip instruction. Is that covered under Stallman's software patent immunity proposal?
The dichotomy between software implementations and hardware implementations is unhelpful. If you oppose software patents on principle, whether because you think they're harmful or because you think they're math and are not supposed to be patentable in the first place, why let the camel's nose into the tent by campaigning to allow hardware (ASIC?) implementation patents, but not allowing pure-software implementations?
Perhaps this is better framed as an economic argument. If cost were no object, more algorithms would be implemented in ASICs. There's a limit to the total chip area you'd want to fit into a computer, but a lot of algorithms could be implemented in a few custom ASICs. It's likely that some of those algorithms would be covered by patents. Stallman's proposal seems like discrimination against algorithms that are important enough to make faster execution worth a lot of money.
If software patents are not valid, and I don't believe they are valid, then allowing them to apply to hardware implementations is just applying a band-aid and punting on the real issue. Stallman's proposal may be pragmatic, in that it reduces risk to most start-ups and other entities worried about violating patents in software, but it sustains the confusion about what software patents are and whether they can be valid. I think there's more at stake than the money at issue in patent lawsuits every year. I think broad appreciation of the value of open culture, and recognition that algorithms should be part of that, is worth more than any extra value companies might be able to extract from hardware implementations of patented algorithms due to artificial monopolies created by patent protection.