If I have a method that works better than your method (let's say, simply traverse the range backwards),
smoke
The Scooby Doo method, like all transformational advancements, is tough to swallow for people indoctrinated with current or old generationally transmitted habits. They will usually attack it though all facts are under their eyes, crystal clear.
A change in perception, or the simply a social revolution of changes in how we think,, like the ones new generations bring, usually solves such problems in the long term.
Join the Scooby Doo method movement today!
Whenever I see a post going like "splitting the range is a good idea and not an issue at all" after all the uncountable and obvious proofs to the contrary, I do 50 pushups, to compensate for the 41% extra steps required to solve the same problem in 2 ranges.
smoke
That's basic math and is correct, as proven countless times (in actual math, not yours, since I remember how you recently suggested to solve Puzzle 135 by first converting it magically to a 100 bits problem or something).
50 was too easy from a point, now it's 100. How many do you do?
After adapting my own kernel to load/store stuff using L2 (instead of only once, before and after all the jumps) I reached 9.7 GK/s on RTX 4090 (64 jump points, DP 32), which was an increase of 75% in speed, and I haven't even tried to do micro-optimizations on it, like before.
smoke...
[/smoke]
Old news. Since that time, I tweaked things a lot more. I now get around
10900 MK/s on a RTX 4090 to jump the kangaroos. And around
14 GK/s on a RTX 5090. It seems that the performance is greater if combining both the L1 cache with the L2 persistent memory, to squeeze in a couple more kangaroos to fill the execution pipeline.
Why do you need the code for? Maybe learn to actual code, instead of trolling around total non-sense and sharing
script kiddie scripts while asking for free stuff. Here's a hint: I'm not gonna share my code to you even if you pay 5 million dollars for it.