DSPLLE: Split SRS into SRS and SRSH and implement conditional variants of RTI

[Pokechu22]

• The SRS instruction works differently from the LRS instruction. LRS supports, in order, $ax0.l, $ax1.l, $ax0.h, $ax1.h, $ac0.l, $ac1.l, $ac0.m, and $ac1.m. SRS, however, goes $ac0.h, $ac1.h, invalid, invalid, $ac0.l, $ac1.l, $ac0.m, $ac1.m. As such I've split SRS into SRSH and SRS. SRS with ax registers was used in the free rom, so I've replaced that. It does seem that SRSH appears in the official rom, so this change may improve accuracy.

  The behavior of these instructions can be verified with the srs_test.ds program. What was previously SRS @0x00, $AX0.H or SRS @0x00, $AX1.H does not actually make any stores, and SRS @0x00, $AX0.L or SRS @0x00, $AX1.L actually stores $ac0.h/$ac1.h (including the sign extension).

• The RTI instruction supports conditions. I highly doubt anything actually uses this, because conditionally returning from an interrupt is rarely useful. Testing this required rejiggering some of the DSPSpy base code to allow custom interrupt handlers to be used.

  The behavior of these instructions can be verified with the rti_test.ds program. Dolphin previously ignored the RTINZ and RTIZ instructions and always had 0x3333 in $ax1.l. Now, Dolphin correctly shows 0x1111 or 0x2222 depending on whether $ax1.h is 0 (i.e. it shows, in order, 0x2222, 0x1111, 0x2222), matching hardware. I haven't actually tested the other conditional variants, and am assuming that they all match the same conditional variants used by other instructions.

Since the free DSP rom needed to be updated, I've also included Tilka/dolphin@84a7cda which makes some other rom-related changes that were not considered important enough for a separate pull request.

Since the old free DSP roms no longer work correctly, I've changed the on-screen display warnings to a AskYesNoFmtT. The message says "If you select "No", audio might be garbled.", which is not strictly accurate (the invalid SRS instructions were actually in ROM code used by the GBA uCode, and using the old ROM with this PR seems to only break GBA transfers and not audio in general). However, it would only affect users who manually copied the free ROM into their user GC folder; users with the official DSP ROM or with no DSP ROM should not see this message.

I will update the gamecube DSP manual for these new instructions in a separate pull request. I already have a large number of manual updates prepared, and would prefer to include this with them.

The test_main change causes this PR to conflict with #10032; although Git will not report a merge conflict, they should not be merged at the same time. It's an easy fix (reg_mask_test.ds will just need test_main: added to it), but it can't be performed until one of the PRs is merged.

[BhaaLseN]

Code seems good, but I haven't tested it.

[Tilka]

What if:

• try to load ROM from file, warn if its hash doesn't match the original ROM
• otherwise assemble free ROM from embedded text

This way we don't need to update hashes all the time.

[Pokechu22]

That could work, but I didn't actually find the hashes too hard to update. I modified the hashes mismatch message to include the hashes and got them that way (and then I changed the message back).

I feel like it would be useful to have the hash comparison even if we were assembling the free ROM from text, as a sanity check to make sure the assembly worked well. So maybe it'd be better to just always include the hashes in the mismatch message?

[Tilka]

I just feel silly maintaining a changelog and a list of file hashes. That's what Git is for.

[Rumi-Larry]

I just feel silly maintaining a changelog and a list of file hashes. That's what Git is for.

IMO, given the small size of it. keeping this documentation here is fine. Otherwise there would need to be another repository just for the replacement ROMS, which would have been a good idea when they were being created, but now creating a new repository for them seems silly too.