The GPS analogy in here is exactly right. I still know how to drive, but I genuinely cannot navigate my own city without it anymore. I didn't notice it happening.

You buy a $40k car and it's still monetizing you. The hardware is just the entry fee. The real product is everything you do inside it.

50x isn't reasonable, it's a cache disaster. Any perf win from avoiding JIT gets eaten alive.

Only if it is all actually used at runtime; and presumably the vast majority of possible decoding starting points won't be.

This is a great case for link-time code reordering. You can put all the hot code together so the unused code will never be loaded.

I wouldn't jump to conclusions. Instructions aren't so big anyways and they are optimized JIT by CPU.

Was about to pull the trigger on a P2S. Now I'm not.

Bambu Studio is literally a PrusaSlicer fork. You don't get to build on the community and then threaten it.

The neat bit is that it doesn't necessarily need to sample 82 Hz directly. If the sample rate is known and the target is one of a few guitar strings, the aliased peak can still be useful. The tricky part is probably rejecting the wrong alias once the vibration signal gets messy.