[GBA Test Suite] Hard crash on an FPE in the test suite, endless? run on SIO timing

[RivenSkaye]

Not sure how common it is in actual BIOS and games (and possibly homebrew content) but the BIOS math tests cause an instant hard crash and core dump on a floating point exception, and I have yet to see the SIO timing tests finish with either none connected and selecting any of the SIO addons from the settings menu

[shonumi]

Thanks for running that test suite! As you can see from the overall results, GBE+ is very inaccurate. It's a wonder it even runs anything at all. The focus over the years had strictly been on supporting accessories and peripherals, but with that out of the way I have been fixing a bunch of stuff (recently DMA related).

I'll take a closer look at that hard crash and endless SIO test this weekend.

[RivenSkaye]

It's a wonder it even runs anything at all.

Not at all, I've seen much worse code perform reasonably okay. That said, it's mostly slight inaccuracies and the actual major test failures are on unaligned and out-of-bounds ops.

If I see any obvious gains, I'll have a look at making a PR for fixes. As for my renewed interest, I adopted the orphaned AUR package to make it build for the first time since the Qt5 switch.

[RivenSkaye]

Any special advice on debugging or figuring this out? Or is it just a matter of matching the checks' source code and the emulator's (and possibly stepping through with a debugger) to figure out what goes wrong?

[shonumi]

I fixed the SIO Timing test freeze around Sunday night (CST anyway). I haven't committed the changes yet since I tend to wait like a week or more before pushing these days.

The tests work now without crashing or freezing. The SWI math issue was a floating-point exception triggered by a special case division (-MAX/-1) which is also a special case on the DS.

The SIO freeze was caused by GBE+ not handling SIO interrupts for Link Cable transfers when disconnected. The test waits indefinitely for a timer interrupt or a SIO interrupt, but GBE+ generated neither. I dunno why the test didn't time out, so I'll look into that later.

For now, however the SIO Timing tests at least run and fail. Timings are basically instantaneous at the moment. Probably won't address that until CPU timings are cleaned up.