Test on emulator passes, but fails on hardware

[Kuratius]

SqrtBench.zip

Expectation would be that the test either passes or fails on both, but the correct percentage is different.

[Jaklyy]

From a quick glance it sounds likely that the square root hardware on the ds takes shortcuts in the calculation That's very interesting

[Kuratius]

I did a bit more testing, I think the framecounter is off by 1. If I put a while(1); at the end of the first while loop, it freezes on frame 1 on the emulator and frame 2 on the nds. This affects the correct percentage because I use the framecounter to keep track of the number of iterations when calculating how many were correct.

[Jaklyy]

what exactly are you running the rom via on hardware? since the frame counter appears to be incremented via interrupt it could be possible its running the interrupt handler an extra time before/during the first frame due to a bug with the bootloader. because i can't actually reproduce that behavior when using a recent-ish build of twilightmenu++ on my 3ds

[Jaklyy]

...however i can reproduce it when installed directly to nand that's very interesting

[Kuratius]

My build of twilightmenu is probably 2-3 years old, but I'm starting the rom from unlaunch so I dont think it matters.