[Bug]: Slow downs

[alexanderzuckerman]

SunPad revision

1.0

Platform

iPhone

What happened?

SunPad-Diagnostic-20260811-012529.log
SunPad-Diagnostic-20260812-014603.log
SunPad-Diagnostic-20260812-024442.log
SunPad-Diagnostic-20260812-022816.log

Multiple slow downs when playing game on iPhone 15 Pro (1x)

Reviewed evidence

No response

[chrissotraidis]

Thanks for the detailed logs. I have reproduced the same class of slowdown on a physical iPhone at native 1x, including drops below the game's intended 30 FPS and corresponding loss of emulation speed. I am keeping this issue open.

The attached logs were useful for confirming the environment, but SunPad 1.0 did not yet record enough performance data to identify which subsystem was responsible. Our subsequent profiling indicates that the main CPU/game thread can become saturated. The cost varies by scene and includes generated game code, PowerPC floating-point/paired-single helpers, address and memory handling, and the iOS software vertex loader. Thermal pressure can make it worse, but it does not appear to be the sole cause.

We found two narrow CPU optimizations that recovered headroom in some tests, but later slowdowns still occurred, so I do not consider the problem solved and will not ship an experimental change based only on a favorable short run. Game Mode is also being evaluated, but it is only a scheduling aid and is not a complete fix.

The next step is improved bounded diagnostics and a repeatable scene capture so one test can separate CPU execution, vertex processing, shader compilation, audio starvation, scheduling, and thermal behavior. No additional logs from the current 1.0 build are needed right now; they would have the same measurement gap. Once that diagnostic build is ready, I may ask you to repeat a short, specific route on the iPhone 15 Pro.

Thank you for reporting this and for providing all four logs.

[chrissotraidis]

@alexanderzuckerman the latest preview .ipa has an experimental mode that is a test on CPU optimization for the game. Feel free to use it, and even better if you can use it and report back with logs from sessions running under the new build. When you put the new .ipa on your device, open the three dot menu at the top right and select the experimental mode.

[alexanderzuckerman]

I was able to update to the latest version but the screen doesn't look right

SunPad-Diagnostic-20260818-025200.log

[chrissotraidis]

@alexanderzuckerman Sorry we missed your follow-up. I reviewed the diagnostic: it shows Experimental Performance Mode enabled with 4× render scale. We later confirmed that this combination can produce severe visual corruption even when the FPS counter looks normal.

Preview 5 improves the diagnostics, but I cannot honestly say it fixes this configuration automatically. Please update to Preview 5, turn off Experimental Performance Mode in the ••• menu, restart SunPad, and set render scale to 1× with Original 4:3. That is the supported baseline.

If you would still like to test the experimental mode, please keep render scale at 1× or 2×, restart after changing it, and attach a screenshot plus ••• → Report a Problem… if the visual issue returns. We are keeping the slowdown issue open.

[IMTHECOOLEST]

Happens to me at base settings

[chrissotraidis]

@IMTHECOOLEST can you share logs, and which version you're running?

[chrissotraidis]

@alexanderzuckerman @IMTHECOOLEST Preview 6 is now available: https://github.com/chrissotraidis/sunpad/releases/tag/v0.1.0-preview.6

Please install it, choose Use Supported 30 FPS Mode in the ••• menu, set Original 4:3 and 1×, then fully close and reopen SunPad. Play until the slowdown or visual problem normally appears. If it happens again, use Report a Problem… and attach the new log plus a screenshot if the problem is visual. Preview 6 records the active runtime settings we need for the next investigation.

[alexanderzuckerman]

Latest-SunPad-Diagnostic.log

Significantly more stable! Appreciate the work you put into this

[chrissotraidis]

Great to hear, Alex. Thanks for retesting. The new log looks much healthier and confirms that the supported 30 FPS, Original 4:3, 1× profile is active. I’ll keep the issue open for now. If the sustained slowdown returns, please use Report a Problem after it happens and include the scene and approximate play time.

[alexanderzuckerman]

Latest-SunPad-Diagnostic.log

Latest update returns the slow downs particularly at Noki Bay

[chrissotraidis]

Thanks — I downloaded and analyzed the full new log. It is from Preview 8 (build 6), and it captured the Noki Bay slowdown clearly.

The log also exposed a concrete regression: earlier in the same process, 16:9 was selected and then changed back to Original 4:3. Dolphin's one-shot patch removal cleared its bookkeeping but did not restore Sunshine's original heatwave instruction. That left the containing 16 KiB StaticRecomp chunk in interpreter fallback; the log records that exact chunk failing verification six times. During the later Noki Bay interval, the game stayed mostly around 22–28 FPS at 1×/4:3 while the combined CPU-GPU thread was saturated and the phone was at Serious thermal pressure.

I've released SunPad 0.1.0 Preview 10. It restores the exact original GMSE01 instruction, re-invalidates the chunk so native StaticRecomp verification can recover, and avoids redundant patch applications.

Could you test one controlled pass?

1. Install Preview 10.
2. Select Use Supported 30 FPS Mode, Original 4:3, and 1×.
3. Fully close SunPad, let the phone cool if it is already hot, then reopen it.
4. Go directly to Noki Bay without changing display settings and play the same route for several minutes.
5. If it slows down, take a screenshot during the slowdown and immediately export a new ••• → Report a Problem… log. Please include roughly how many minutes it took.

This fixes the confirmed fallback leak, but I'm keeping the issue open until your scene-matched retest tells us whether Noki Bay also has a separate thermal/single-core limit.