Jackass The Game runs super slow

[Panderner]

Jackass The Game runs super slow similar to Test Drive Unlimited and Killzone Liberation.

It runs at 4-5 fps (5-6%) on Realme C2

[Panderner]

However it rans full speed on pc running windows on all backends:

Specs: AMD A8-7650K Radeon R7 GTX 770 8GB RAM Windows 10 Pro

[hrydgard]

GE frame dump?

[Panderner]

Yes Here's a GE Dump: recording.ppdmp.zip

[hrydgard]

Yeah this does the same thing as the Sonic games for color correction, but slightly differently - it actually textures directly from the framebuffer being rendered to, which (currently) forces PPSSPP to do "traditional" "depal" through a framebuffer copy. Not too bad on desktop, but gets horribly expensive on mobile tiler hardware :(

There are a few ways, more and less hacky, that we could fix this... For now your best bet is probably playing at 1x rendering resolution to reduce the speed hit but it'll probably still be too slow.

[Panderner]

Even 1x resolution can't run this game on my phone.

[hrydgard]

well, you'll have to run it on PC until I get around to figuring out a way to fix it. Good to have a report and a nice dump to repro it though, very helpful!

[Panderner]

Also POCO M3 rans full speed on all backends but Realme C2 does not:

[Panderner]

Performance comparison for this game:

POCO M3:

Realme C2:

[71knight]

Come to find out the helio P22 processor is only half as powerful as a Snapdragon 662. Especially in single Core test. So no wonder it runs the game slower than the other phone.

[Panderner]

This is a device issue not a ppsspp realme c2 is a weak phone unlike Poco M3 🤷

@Gamemulatorer can you test this game on your phone? We need testing more devices if the game runs super slow.

[Panderner]

It's running smooth on my Redmi Note 9 both VK and OGL

Oh ok

[unknownbrackets]

I think we could maybe more generically detect a situation where we have a:

• Linear CLUT ramp
• CLUT16/CLUT32 source with masking for a single color channel
• RGBA write mask only writing the same channel
• Through mode so we have coordinates (non-overlapping)
• No other tests (depth, stencil, etc.) active

And queue the draws separately to combine as a single batch. If they all had the same source texture and the same destination, it almost seems like we could ignore overlap checks in every practical case.

-[Unknown]

[hrydgard]

Not super easy to do that detection though, preventing flushing between the draws.

To do that sort of thing easily, ideally we'd generate a higher level representation of the frame that we could postprocess (higher level than current VK and GL renderqueues)....

[Panderner]

I tested Samsung Galaxy A12 (Not Mine) using vulkan. It's a same gpu as my Realme C2

[ghost]

This game is also super lag 🤣

[Panderner]

This game is also super lag

@Gamemulatorer can you send me the screenshot?

[ghost]

Maybe this game is only lag and graphics issue on Mali GPU?

Vulkan

jackass.ppdmp.zip

OpenGL

OpenGL got higher fps compare to Vulkan 🤔

[Panderner]

Thank you @Gamemulatorer for testing this game for your Redmi Note 9. I also tested my Realme C2 which is using a PowerVR GPU is also running slow, My POCO M3 and my PC runs smoothly.

[hrydgard]

Yeah, it's well known that this game is running slowly. It's gonna need some special tricks like the Sonic games.

[ghost]

Yeah, it's well known that this game is running slowly. It's gonna need some special tricks like the Sonic games.

So this issues is related to this https://github.com/hrydgard/ppsspp/issues/10908

[hrydgard]

Yes

[Panderner]

Maybe this game is only lag and graphics issue on Mali GPU?

Vulkan

jackass.ppdmp.zip

OpenGL

OpenGL got higher fps compare to Vulkan 🤔

Umm @Gamemulatorer can you Lower the resolution to 1x?

[hrydgard]

Gonna see if the Killzone workaround #15934 can be adapted for this.

[hrydgard]

Nope, this one is more like Sonic Rivals, although it's not using a temporary buffer, it's doing "depal" directly from the framebuffer to the same pixels on itself in little tiles, somehow that doesn't cause conflicts on the hardware. Should have read my own post from before. We end up depal-ing the entire texture, for every tile, and there's a LOT of tiles.

Fortunately it's through-mode, we should be able to limit the region of the texture we depal for each draw fairly easily, and then maybe do a little post-rearrangement... Actually, it turns out we already do.

So all that's left now I think is to implement something like ApplySonicHack to the queue runner...

[Panderner]

My Realme C2 is broken unable to test this again hopely my Poco M3 runs ok for my phone this must be PowerVR GPU specific.

[ghost]

My Realme C2 is broken unable to test this again hopely my Poco M3 runs ok for my phone this must be PowerVR GPU specific.

This should improve now on your powervr gpu phone since there's a lot of improvement recently in powervr gpu https://github.com/hrydgard/ppsspp/pulls?q=is%3Apr+label%3A%22PowerVR+GPU%22+is%3Aclosed https://github.com/hrydgard/ppsspp/pull/16802#issuecomment-1411847872

[hrydgard]

Possibly, but the powervr improvements are mostly about speed of shader compilation and stuff like that. This game still runs very inefficiently and needs fixing.

[ghost]

Except for the performance issue of this there's also a graphics issue below the screen. HW Vulkan/OpenGL SOFTWARE

[hrydgard]

New report from Discord, by "oh well":

I've tried to run it numerous times in the past, it never really works. Like it can load until the minigame menu, but when I try to play any of the first minigames (e.g. Pachinko Precipice), it gets stuck on the loading screen

[ISASLIMBOY28]

Run really slow on Tecno Spark 10 Pro (Helio G88/Mali-G52 MC2/8GB Ram). On Vulkan, got 10ish fps. On OpenGL, Got 20ish fps.

[hrydgard]

I'm aware that this is still not fixed. It's a tricky one, haven't gotten around to it - the potential fix idea I have is quite fiddly.

[Panderner]

It's seems this game is likely runs very slow on non-adreno device.