Forking Kangaroo 256: A Fresh Take

I'm working on forking Kangaroo 256 based on some concepts I have in mind that I think are better. Gonna update as I find time since I've got other commitments too. It's not that I'm against Kangaroo, just feel we need something fresh for future puzzles. Efficiency is key, and I have started tweaking some things already.

Can you break down your ideas a bit? Sounds interesting, but getting concepts from the code is a hassle. Not sure where to start.

Hey! Any chance you can add support for multi pubkey functionality? Also, thanks for the automatic DP selection; JLP isn't sharing his secret on that. Kinda frustrating.

The plan is to fine-tune the DP and jumps to get better success rates. Need to ensure the randomness is just right not too random, but enough to keep on track. Only realistic multi pub scenario I see is if they're all tied to one point with some additions/subtractions.

But how do you figure out the "best" DP? The original used x, y, z. What's your method for that, especially in terms of selecting based on range? And what’s this “too random” you mention? It's confusing.

Maybe check out this link for some more insights.

Choosing the best DP might come down to testing and some statistical analysis. Too high could mean slow performance, too low might mess with accuracy. Finding the right balance is key. And with jumps, gotta avoid making patterns too predictable.

I tried running this with CUDA 86, and it works but doesn't find matches on smaller challenges like 50 or 55. CPU seems to handle those better. Here are some logs for context.

Cool logs, man. I'm looking to bump up the success rate by about 50% and optimize it. Once I get a GPU, I'll get into converting it to a GPU setup.