Jak & Daxter : "Unknown GetPointer" and make the game run slow FPS.

[daniel229]

DEMO,pressing triangle is to skip video. https://drive.google.com/file/d/0BzGZGDfFE68zc3Raa2t2ZzBLcWM

FuncReplacements = False normal FPS

[hrydgard]

Good find! This is a bit odd.

Commenting out

{ "memcpy", &Replace_memcpy, 0, 0 },

and

{ "memset", &Replace_memset, 0, 0 },

in ReplaceTables.cpp fixes it in the demo. Strangely, only commenting out memcpy but not memset causes the game to crash in fast memory mode. Feels like it's relying on some return value from memcpy to then do a memset?

Another strange thing - savestates seem broken in this demo?

[hrydgard]

Rather scary this one - the game relies on some non-return registers to be set correctly after the function. Fixed, but there might be more of this kind of bug out there..

[daniel229]

Great,how about the "Unknown GetPointer" before intro,also not happen when FuncReplacements = False .

[hrydgard]

Oh, I missed that one. I'll check soon

[hrydgard]

Hm, starting to feel like a timing issue? None of the "thread not dormant" or ReferEventFlagStatus errors happen when not using function replacement...

[hrydgard]

Timing issue indeed, bleh.

[daniel229]

Not fix this game,but seems help https://github.com/hrydgard/ppsspp/issues/6695 and https://github.com/hrydgard/ppsspp/issues/6665

[brujo5]

since build 1.0-83 1ecfd6cebcd8bb97722be4dc6fbc48f3ea71fe56 Have slowdowns in the game Dante inferno, drop fps from 40/40 to 29/40

In build 1.0-80/81 is fine.no slowdown

iPhone 5S

[hrydgard]

Thanks @brujo5

I'm starting to lean towards not replacing these functions that have highly variable timing, instead just hook and do the GPU copy, then let them run.

[unknownbrackets]

Well, we will have to hook memcpy on enter and exit, but yeah, that may be the simplest thing. I wonder what the perf impact is.

-[Unknown]

[brujo5]

@hrydgard

Here other game that work in slowmov also in non-buffered rendering since 1.0-83 . And cause overheat

before 1.0-81/82 60/60 fps In 1.0-83 29/60fps overheat

[unknownbrackets]

Even using hooks, this game crashes.

More importantly, in interpreter, it works, but in jit it doesn't work. But, the original memset/memcpy don't work with interp, so that's still an issue.

So, it's almost definitely a bug in the jit. I wonder if this happens on arm too?

-[Unknown]

[hrydgard]

@daniel229 how about build 1.0-84 in Tomb Raider?

Yeah, I don't get what's going on here...

[brujo5]

Will try later today. Hoop the update come out in cydia very soon.

[daniel229]

@unknownbrackets Is it the crash not randomly happening?Those errors may sometime happen,sometime not. @hrydgard I suppose you @brujo5

[unknownbrackets]

Yeah, I ended up getting random crashes in even interp with no replacements, but it didn't happen every time. It seems way more likely with replacements enabled.

-[Unknown]

[brujo5]

In build 1.0-86 retest dante inferno/tomb raider legends, slowdowns still going on but not too much lower fps. Example in Dante inferno build 1.0-83 Slowdowns 29/40 fps Now in 1.0-86 increase an bit slowdowns 34/40 fps

[hrydgard]

I think I'm gonna leave it like it is now for 1.0.1. We will have to look more into why replacements cause some misbehavior later..

[unknownbrackets]

I hope you're not using any adjusted cpu clock setting. We've already confirmed through testing that the current speed is faster than on real hardware, and matches replacements being off.

-[Unknown]

[brujo5]

Nope never use cpu speed hack on this game because make sound very choppy.

[sum2012]

@daniel229 Does v1.3-540-g29739c8 or newer version fixed ? I only tested beginning no problem

[benderscruffy01]

no problems so far i think this can be closed