Kurohyou series clip loosen after certain circumstances (now: graphical errors after fights)

[lim2222]

Game or games this happens in

NPJH-50333,NPJH-50562

What area of the game

After mini game,cutscenes.

What happens

Screenrecorder-2022-09-04-10-20-39-733.mp4

What should happen

Screenrecorder-2022-09-04-10-22-24-622.mp4

GE frame capture

No response

Platform

Android

Mobile phone model or graphics card

Mi10t pro android 10

PPSSPP version affected

ppsspp-v1.13.1-475-g853a2df06-android.apk

Last working version

ppsspp-v1.13.1-471-g67c911d13-android.apk

Graphics backend (3D API)

Vulkan

Checklist

• Test in the latest git build in case it's already fixed.
• Search for other reports of the same issue.
• Try resetting settings or older versions and include if the issue is related.
• Try changing graphics settings to determine if one causes the glitch (especially speed hacks or enhancements/replacements.)
• Include logs or screenshots of issue.

[unknownbrackets]

Thanks for narrowing it down - that points at #15892.

I suspect this is some subtle case in format/stride matching...

-[Unknown]

[lim2222]

v1.13.2-1293-g7a4830ed4 fixed the after mini game and cutscenes flickering issue, however when screen recording flickering still there.

[hrydgard]

How are things now? And what method of screen recording?

[lim2222]

With MIUI 12 internal screen recorder,that not a big deal only flickering during screen record,ingame play will not happen:⁠-⁠D.
There is new issue of NPJH-50333 that with vulkan renderer winning a fight will crash.

[lim2222]

Crash as video shown
https://youtu.be/yYC47kPN1r4

[hrydgard]

Still crashes in the latest builds?

[lim2222]

Yes,crash during the last punch/kick.

[hrydgard]

Ok, thanks for testing..

[hrydgard]

Could you test some old versions to try to figure out about at which version number it started crashing?

Or get some ADB logs?

[lim2222]

v1.13.2-1214-ge6db0bef2 start crashing.
Sorry that my pc broken at the moment can't do adb log,this is the crash log generated by my phone may be useful

---

Build fingerprint: 'Xiaomi/apollopro_global/apollo:10/QKQ1.200419.002/V12.0.11.0.QJDMIXM:user/release-keys'
Revision: '0'
ABI: 'arm64'
Timestamp: 2022-11-24 10:36:49+0800
pid: 5185, tid: 5406, name: RenderMan >>> org.ppsspp.ppsspp <<<
uid: 10673
signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0xec
Cause: null pointer dereference
x0 00000071e8d49178 x1 0000000000000000 x2 0000000000000002 x3 00000000ffffffb5
x4 0000000000000001 x5 0000000000000000 x6 0000000000000000 x7 0000000000000001
x8 0000000000000001 x9 000000716668b318 x10 0000000000000002 x11 00000000fffffe24
x12 0000000000000001 x13 0000000000000002 x14 0000000000200000 x15 0000000000000000
x16 00000072782158f8 x17 0000007278206c60 x18 000000716af2a000 x19 000000716668b310
x20 00000071e922fd50 x21 0000000000000001 x22 0000000000000001 x23 000000718e0d0084
x24 0000000000000000 x25 00000071e922fd50 x26 0000000000000000 x27 0000000000000048
x28 00000071e9217698 x29 000000716b81e750
sp 000000716b81e530 lr 00000071e9293080 pc 00000071798d00ac

backtrace:
#00 pc 000000000017f0ac /vendor/lib64/hw/vulkan.kona.so (qglinternal::vkCreateFramebuffer(VkDevice_T*, VkFramebufferCreateInfo const*, VkAllocationCallbacks const*, VkFramebuffer_T**)+1676) (BuildId: d9ae83e3b2710252b7e131577d625059)
#1 pc 00000000007324f0 /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (VKRFramebuffer::Get(VKRRenderPass*, RenderPassType)+200) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#2 pc 000000000073dffc /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (VulkanQueueRunner::PerformBindFramebufferAsRenderTarget(VKRStep const&, VkCommandBuffer_T*)+176) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#3 pc 000000000073b5c4 /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (VulkanQueueRunner::PerformRenderPass(VKRStep const&, VkCommandBuffer_T*)+524) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#4 pc 000000000073b0e0 /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (VulkanQueueRunner::RunSteps(std::__ndk1::vector<VKRStep*, std::__ndk1::allocator<VKRStep*>>&, FrameData&, FrameDataShared&, bool)+432) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#5 pc 0000000000734240 /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (VulkanRenderManager::Run(VKRRenderThreadTask&)+412) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#6 pc 0000000000733318 /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (VulkanRenderManager::ThreadFunc()+320) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#7 pc 000000000073756c /data/app/org.ppsspp.ppsspp-0Qy6kFFIWmJ8xS16afEyMw==/lib/arm64/libppsspp_jni.so (ZNSt6__ndk114__thread_proxyINS_5tupleIJNS_10unique_ptrINS_15__thread_structENS_14default_deleteIS3_EEEEM19VulkanRenderManagerFvvEPS7_EEEEEPvSC+64) (BuildId: 535721a90eaff3e84768f7d414cf26620116d509)
#8 pc 00000000000e7480 /apex/com.android.runtime/lib64/bionic/libc.so (__pthread_start(void*)+36) (BuildId: 6dc18c350dc6bdab66bd0af0dcfd6e4c)
#9 pc 000000000008522c /apex/com.android.runtime/lib64/bionic/libc.so (__start_thread+64) (BuildId: 6dc18c350dc6bdab66bd0af0dcfd6e4c)

[lim2222]

By doing couple of tests i also noticed that some weird graphics behaviour happen compare with v1.13.1-347-g6cc8ca517 which is the best version i have test for these two games.

v1.13.1-347-g6cc8ca517 there is green colour background effect

v1.13.2-1214-ge6db0bef2 missing green colour background

v1.13.2-1214-ge6db0bef2 foot shadow present gap in NPJH-50333

v1.13.1-347-g6cc8ca517 foot shadow

v1.13.2-1214-ge6db0bef2 shoe in NPJH-50562

v1.13.1-347-g6cc8ca517 shoe

v1.13.2-1214-ge6db0bef2 flashing negative colour before start fighting

v1.13.2-1214-ge6db0bef2 flashing red colour before start fighting in NPJH-50562

[hrydgard]

Thank you, great stuff. I think I know what's going on with the crash now.

I wonder if the inversion/red color aren't actually intended? Hard to say with none of the youtube gameplay being from real hardware, sigh...

[unknownbrackets]

As I recall from #16101, it's not supposed to be flashing, but it's definitely not supposed to be grey. Pretty sure I compared it with the PSP via frame dump so I think the software renderer is doing what's intended.

-[Unknown]

[hrydgard]

This should likely be better now after #16426.

[lim2222]

v1.13.2-2030-gcb7329cd1 fixed vulkan fighting crash

[hrydgard]

Great, it should have fixed some of the visual problems too, do any remain?

[lim2222]

still flashing grey

[hrydgard]

OK, well, progress at least!

[lim2222]

v1.13.2-981-gca2962bea start flashing grey

[hrydgard]

cool, so that's #16014 , which first implemented loading CLUTs (palettes) from framebuffers (images rendered to). Not too surprising but good to confirm.

So I guess the question is in what way it goes wrong... There's one unhandled case which we might fall into I guess, where we should use a CLUT loaded from a framebuffer to texture from another framebuffer...

[lim2222]

Confirmed that there is green colour background effect from original hardware.

[hrydgard]

Is the green color effect still broken in the latest build?

[lim2222]

still missing in v1.13.2-2041-g21cc6b85e

[unknownbrackets]

#16454 should take care of the color before entering a fight. It should not be red or grayscale, but more of a cyan. Didn't look into the other issues, but decent chance this helps the green background as well.

-[Unknown]

[hrydgard]

@lim2222 Please test again - these effects should hopefully all work now.

Of course, do report if something is still not working.

[lim2222]

v1.13.2-2120-g612d2181a the flashing thing gone.
Now only minor glitch of the player's shadow has gap in convenient store,other place shadow is normal.

[lim2222]

green effect come back too,great job.

[hrydgard]

Could you get a GE dump of the convenience store problem? (I think I used to have one of the store, but can't find it)

[lim2222]

here is ge dump
NPJH50333 (1).zip

[hrydgard]

The split shadow is (not too surprisingly) depth related, there's a small discontinuity in the depth gradient of the floor, after converted from color to depth, at the same place as one of the color components wrapping around:

I don't see any unwanted filtering going on in Renderdoc, so there's likely a minor math error in the color->depth conversion shader, or in one of the passes leading up to it.

I'm postponing investigation of this minor glitch to after 1.14.0.