I think I now know why I had difficulty getting funding

[compilerbitch]

I tried explaining my PhD work to someone over dinner. He was an ex-EE type, degree in electronics, PhD in information theory, so he could at least get the gist. Anyway, it became apparent quite quickly, after mentioning what I was trying to do, that he just flat-out didn't believe it was possible.

I did a mental rewind of what I'd been talking about, and actually I would be inclined to agree, but for one thing: I already know the technology works, because I've actually made the 'big result' work once already. Clearly, I'm going to have to work pretty hard to show that what I'm doing is both feasible and actually not a complete fraud -- PE sounds impossible, at best something that should be NP-complete and not doable in a general sense. But, no -- it does totally work, and even can be shown to be near enough time and space linear with the size of the generated hardware.

Someone said that one should never worry about good ideas being stolen -- the best ideas take a lifetime's effort to become recognised, ramming them down as many throats as possible. I suspect that hardware PE is going to be one of those. I was talking to someone 'in the know' a couple of weeks ago, who said that there would be no way that a chip company would fund this kind of research, despite its potential impact on the way that chips are designed. Seemingly, EEs are ultra-conservative, running a mile at any mention of the 'F-word' (functional programming). PE is probably an order of magnitude scarier.

This is scary. It definitely doesn't bode well for publication, and I'm going to have to show 'extraordinary evidence' to back up my 'extraordinary' claim. The annoying part of this is that PE has been known about since 1920ish (Kleene's s-m-n theorem), and was shown to work for software decades ago. Compiler hackers don't flinch at this stuff, it's just the electronics types.

So, since my funding applications all went out to chip companies and the tier of EDA companies that support them, it follows that none of them would even vaguely consider what I was suggesting. If I failed, they would waste their money. If I succeeded, I'd be upsetting the food chain horribly.

Luckily, my funding has nothing to do with the chip industry. My IPR is my own, so I don't have those concerns directly. However, the bigger problem will probably be in getting published -- the compilery people may be put off by the EE, and the EE people may well just think I'm bluffing.

Oh well, in for a penny, in for a pound...

Comments

[trurl.livejournal.com]

Compiler hackers don't flinch at this stuff, it's just the electronics types.



Heh. Given my general ignorance of EE, I'm very surprised that partial evaluation would be so foreign to them.

[compilerbitch.livejournal.com]

Indeed. PE for the simpler classes of circuit is actually pretty straigtforward, and can be easily shown to work. It does get tricky when dealing with asynchronous circuits, however, which I suspect will probably end up taking up more of my time. The problem is that if you don't care about glitches (hazards in some of the EE literature), the familiar axiom of replacement for Boolean expressions is fine. However, if the glitches matter, as they do in many cases inside asynchronous logic, the standard axiom of replacement falls apart and needs to be redefined. I'm fiddling around with a 5-value logic that captures the notion of transitions between states (and their relative cleanliness) -- it looks likely that it will provide a way of defining a hierarchy of alternative axioms of replacement that can be chosen to match the expected semantics of the underlying hardware. It's a bit of a sledgehammer approach in that it is much more general than is really necessary just to support PE, but if it gets the job done I'll be happy!

[compilerbitch.livejournal.com]

I should probably mention that EE people are not stupid -- their field is hard enough to be far from trivial. It's just that it has a very distinct culture that's a long way removed from the FP/compiler research world.

It's also fair to point out that FP people seem more open minded about EE than the other way around. I have no idea why this is.