Adjust cycle counts so they are accurate to the JIT block level (Fixes OoT virtual console and other games) #3601
Conversation
| @@ -424,6 +452,8 @@ void Advance() | |||
| slicelength = maxSliceLength; | |||
| PowerPC::ppcState.downcount = CyclesToDowncount(slicelength); | |||
| } | |||
|
|
|||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
phire
force-pushed
the
AccurateEventScheduling
branch
from
e23dcc3
to
6735cfc
Compare
|
Should this be a candidate for 5.0 inclusion? My gut feeling says "no" since it touches JIT stuff, but if it fixes a regression I'll let it in. |
|
As far as I know it doesn't fix any regressions, more or less none of the Virtual Console games worked in 4.0. It would be nice to have this in 5.0 as it would rounds out our virtual console support to "more or less all of them except Majora's Mask" I'm not really concerned about the JIT change, it's minor and takes a few min to verify. I'm much more concerned about the massive timing changes it makes. |
|
PAL Virtual Console games seem to crash with this change. Super Mario 64 and Super Smash Bros. PAL, and The Legend of Zelda: Majora's Mask. Honestly I'm not sure if anything outside of the NTSC Virtual Console games is working? |
|
Maybe something's broken on my end, I can't even boot Pokemon Puzzle League. |
| if (!GlobalTimerIsSane) | ||
| { | ||
| // We are doing a mid-block update, so we need to adjust for the inaccuracy in globalTimer | ||
| int downcount = DowncountToCycles(PowerPC::ppcState.downcount); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
Fire Emblem: Radiant Dawn and Yu-gi-oh False Bound Kingdom both test positive based on the debug code for the same thing afflicting the N64 titles like Ocarina of Time. It's very possible the long term random hangs that can happen are fixed as I've not been able to reproduce them. Then again, I don't really like Fire Emblem Radiant Dawn and haven't played it much seriously. It's a hard game. N64 games are now more or less perfect. The only exception is Majora's Mask, which has a saving issue. Rogue Squadron 2 SEEMS more stable in a test I did on Dualcore, but, it still crashes when mashing sometimes. Seems to be less than master, at least; can play multiple missions and not have it happen. phire and degasus seem to concur that the actual bug is unrelated to this. |
|
Does not Fix Fire Emblem: Radiant Dawn (PAL) Fixes REGRESSION Ed, Edd, Eddy (not booting in LLE/HLE in various configurations,) https://bugs.dolphin-emu.org/issues/7639 |
|
Review status: 0 of 6 files reviewed at latest revision, 14 unresolved discussions. Source/Core/Common/SysConf.h, line 48 [r3] (raw file): Source/Core/Common/SysConf.h, line 50 [r3] (raw file): Source/Core/Common/SysConf.h, line 51 [r3] (raw file): Source/Core/Core/CoreTiming.cpp, line 216 [r3] (raw file): Source/Core/Core/HW/SystemTimers.cpp, line 162 [r3] (raw file): Source/Core/Core/PowerPC/Jit64/Jit_SystemRegisters.cpp, line 277 [r3] (raw file): Comments from the review on Reviewable.io |
|
Review status: 0 of 6 files reviewed at latest revision, 14 unresolved discussions. Source/Core/Common/SysConf.h, line 48 [r3] (raw file): Source/Core/Core/CoreTiming.cpp, line 216 [r3] (raw file): Source/Core/Core/PowerPC/Jit64/Jit_SystemRegisters.cpp, line 277 [r3] (raw file): Comments from the review on Reviewable.io |
|
Reviewed 1 of 4 files at r1, 2 of 2 files at r2, 4 of 4 files at r3. Source/Core/Common/SysConf.cpp, line 170 [r3] (raw file): Source/Core/Core/CoreTiming.cpp, line 327 [r3] (raw file): Source/Core/Core/HW/SystemTimers.cpp, line 263 [r3] (raw file): Comments from the review on Reviewable.io |
|
Fixes FMV timings in Metroid Prime 1, Metroid Prime 2, and Metroid Prime 3. https://bugs.dolphin-emu.org/issues/9321 |
|
Fixes issue https://bugs.dolphin-emu.org/issues/6938 as far as I can tell. |
|
Found a regression: multi-game demo discs heavily lag when attempting when doing anything. The issue is very strange; as long as you're holding a button it lags down to 0 - 1 FPS (60 VPS) O.o. Tested 11/12 D86E01, D85E01 in about 10 seconds of play, timebase registers were checked 251363187 times, 83773753 were rapid. |
|
Affects but does not fix issue 7192 -> https://bugs.dolphin-emu.org/issues/7192 F-Zero GX video gets further. Still likely a dual-issue thing. |
|
Do some of the VC games require specific settings to work with this PR? I figured I'd test out a few, but a lot of VC games that are working in 4.0-8880 are now crashing Dolphin instantly with identical settings. Here's a few that I can't get working now for reference. 2020 Super Baseball (NeoGeo) |
|
All of those games are working for me; in fact, I tested them myself. I tried this Pull Request originally and apparently my NAND was corrupt, and this PR triggered that. Try running the games with Portable.txt and see if they still crash. |
|
Yeah, might be another corrupt NAND. Could you upload the file: |
|
Actually, I might have been wrong about it being |
|
Ahh that makes sense. I'm still using my personal NAND that I dumped from my Wii like 3 years ago. I still have a backup of it so I'll give that a shot. Although, it's not from a virgin Wii since I already had Homebrew Channel installed, a few emulators, and Wiiware games when I dumped it. Not sure if that will affect anything. |
|
I was using a dumped NAND as well. |
|
Fortunately I also had a second NAND backup from a friend's Wii on my hard drive that I forgot existed, so I got to test this twice. After a fresh extraction, the same games crash the emulator using either of the two dumps, so it looks like it's consistent with "real" NANDs. |
|
@BigheadSMZ Please remove the link to your NAND. Distributing a full NAND is illegal because it contains copyrighted software. |
phire
force-pushed
the
AccurateEventScheduling
branch
from
a5d94ae
to
2385ccb
Compare
|
I tracked down the cause of the Nintendo demo disk slowdown. That have a feature that plays a random video if you don't press any buttons in 15 seconds. So every time you pressed (or held) a button, it would reset the deincrementer interrupt to 15 seconds. It's worth noting that 15 * 486 million has overflowed the a 32 bit variable and has the 31st (sign) bit set. Which causes dolphin to schedule an event in the past (which it fires immediately). Previously, dolphin would delay this past event by upto 10,000 cycles, so the impact wasn't too huge but with my more accurate timing the event is now firing immediately and continues to be re-scheduled while the button is held down. I'm assuming this also fixes the play_rec.dat bug, though I haven't tested yet. |
This showed up as a warning while I was debugging dolphin-emu#3601 Turns out it was unrelated, but the cleaup is nice.
|
Now working without needing various hacks to get past the sysconf issues. As an added bonus, fixes an issue with the draw distance I never reported in Starfox Assault. Objects no longer flicker at the edge of the draw distance. |
Previously GlobalTimer was only updated at the end of each slice when CoreTiming::Advance() was called, so it could be upto 20,000 cycles off. This was causing huge problems with games which made heavy use of the time base register, such as OoT (virtual console) and Pokemon puzzle. I've also made it so event scheduling will be accurate to the jit block level, instead of accurate to the slice.
Events scheduled more than 4.12 seconds in the future (2.96 seconds for Wii games) would overflow the sign bit and get scheduled in the past instead, causing them to fire instantly.
phire
force-pushed
the
AccurateEventScheduling
branch
from
2385ccb
to
035aa57
Compare
|
I've fixed the various style issues which were bought up, and created a inverted copy of lastOCfactor to minimize on divisions. |
|
Reviewed 1 of 4 files at r6. Source/Core/Core/CoreTiming.cpp, line 216 [r3] (raw file): Source/Core/Core/CoreTiming.cpp, line 53 [r6] (raw file): Comments from the review on Reviewable.io |
|
Reviewed 2 of 4 files at r6. Comments from the review on Reviewable.io |
|
Review status: 3 of 4 files reviewed at latest revision, 4 unresolved discussions, some commit checks failed. Source/Core/Core/CoreTiming.cpp, line 53 [r6] (raw file): Source/Core/Core/CoreTiming.cpp, line 210 [r6] (raw file): Comments from the review on Reviewable.io |
The inverse operation is more common, especially when games check the timer rapidly. So we do the division once and store the inverted copy.
The interpeted version is correct.
phire
force-pushed
the
AccurateEventScheduling
branch
from
430f0c2
to
d61baef
Compare
|
Review status: 2 of 5 files reviewed at latest revision, 3 unresolved discussions. Source/Core/Core/CoreTiming.cpp, line 53 [r6] (raw file): Comments from the review on Reviewable.io |
|
Reviewed 1 of 1 files at r7, 1 of 1 files at r8. Comments from the review on Reviewable.io |
|
Please finish reviewing/merge this. Would be nice for the Progress Report, which is very light right now. |
|
The review of this PR is done. Ready to merge. Review status: 4 of 5 files reviewed at latest revision, 2 unresolved discussions. Comments from the review on Reviewable.io |
|
Reviewed 1 of 4 files at r6. Comments from the review on Reviewable.io |
|
This pr breaks boot for Metroid other M, some user reported that they where having problems with this game and reverting this change fixed the issue. |
|
Thanks Tinob, I'm debugging this. |
|
LLE audio or changing the emulated CPU clock rate affects this; meaning it's probably yet another timing bug in core. |
|
I discovered why LLE and doesn't reset and HLE does: LLE's hang happens on DSP initialization, HLE's hang happens on LYT (which, the next step is loading the savefiles.) Instead of loading the save files, after about 7 seconds of emulated time, it aborts and loads the sysconf and resets the game. |
|
@phire Thanks for letting me know, Excelent debuguing :) |
Previously GlobalTimer was only updated at the end of each slice when CoreTiming::Advance() was called, so it could be upto 20,000 cycles off. This was causing huge problems with games which made heavy use of the time base register, such as OoT (virtual console) and Pokemon puzzle.
I've also made it so event scheduling will be accurate to the JIT block level, instead of accurate to the slice.