IOS7 on A5 devices load game crash generally

[unpeng]

No description provided.

[brujo5]

more info is nesesary..........

my iphone 5S is fine using the latest build.

angelxwind repo.

[unknownbrackets]

Well, the iPhone 5S doesn't use an A5, only the iPhone 4S does (and certain iPads and etc.)

It probably crashes on devices that aren't rooted and jailbroken, not sure, have not tested.

-[Unknown]

[unpeng]

A7 is fine, but 4s and ipad2 is broken.
Of course, the devices are jailbroken.

[unknownbrackets]

Did it used to work? It would help immensely to know which version was the first to not work, and which before that last did work.

-[Unknown]

[unpeng]

I'm testing.

[unpeng]

Ios5 on an A5 device works fine, but ios7 is broken.

0.9.1-2286 is crash when loading game.

[brujo5]

0.9.1-2286

is hold build

try to install the latest build from angelxwind repo.

is v0.9.6-372 after that reply here.

[unpeng]

The latest version of the a5 devices all crashes.
Wheather it is my own build or install from axe repo.

[brujo5]

@unpeng

um very strange issue,affected only A5 devices.

in IOS 5 all builds working on A5,but not in IOS 7,right?

also this problem no afected A6/A7 chips on Ios 7.

maybe you have very low Ram in ios 7,try to run ppsspp in safe mode?

[unpeng]

That's true, A5 device's RAM was too small. If cleans up the good memory to enter again, the crash probability is small

[bsauvage1]

So I have given it some more tests - games simply won't start on my ipad 2 anymore, whatever settings I use (and I've spent a lot of time trying them all).
Even when closing all background apps, it won't work.
What I can confirm is the last version that I tried and that still worked was v0.9.6-338-g367a5b7, and the last one I tried that crashed was v.0.9.6-566-gb333e5b - a total of 228 commits in between.
(but there may have been some intermediate versions that were still working - can't tell more than that as I wasn't able to test in the meantime).
Since moving on to Mavericks I haven't been able to make builds with Xcode, but I'm willing to try and narrow this down, if somebody can make some intermediate builds and give me a dropbox link, I can install them and let you know.

[unknownbrackets]

What happens when you try to build with Xcode on Mavericks?

-[Unknown]

[solarmystic]

I finally decided to take the plunge and JBed my Ipad 3 (which is on 6.1.x firmware) with posixpwn a few days ago. Tried out PPSSPP (angelxwind repo, v.0.9.6-566-gb333e5b) on it, and am facing the same issue in this report.

In my case though it seems like the HW Transform setting might be the key. PPSSPP crashes when loading quite a number of games with it on, but with it off, those games don't crash when loading. Games tested include Crisis Core (outside the Shinra building it crashes with a 100% certainty if HW Transform is enabled.).

[thedax]

This is why an archive of iOS builds would be useful, and not just the latest and greatest..

[hrydgard]

Is there an easy way to get logs from the iOS builds?

[unknownbrackets]

See #5191. iPhone Configuration Utility should do it, apparently.

-[Unknown]

[hrydgard]

Ah, good. Mostly interested if this is shader compilation errors or something else. @solarmystic could you install the iPhone Configuration Utility and see if you can grab a log?

Will have an iOS setup of my own soon, too.

[bsauvage1]

I have reached out to angelXwind see if by chance she keeps an archive of
recent builds. If positive, I'll scrub and revert.

If not I'll come back and ask advice to get Xcode to compile again (I'll
refresh the old forum thread).

On Tuesday, January 28, 2014, Henrik Rydgård notifications@github.com
wrote:

Ah, good. Mostly interested if this is shader compilation errors or
something else. @solarmystic https://github.com/solarmystic could you
install the iPhone Configuration Utility and see if you can grab a log?

Will have an iOS setup of my own soon, too.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5047#issuecomment-33437318
.

[bsauvage1]

She's explaining me how to fix Xcode in Mavericks:

Install this: http://iosopendev.com/download/
Make sure Code-signing is OFF in the xcodeproj.
Make sure you've pulled all submodules.
Also, try running "iod-setup sdk" after installing iOSOpenDev.
Xcode 5 kinda shuffled some things around, so it's not the same anymore.

I won't be able to try until next week but I'll give it a shot.
Meanwhile, anybody managed to get meaningful logs from the iPhone Configuration Utility ?

[unknownbrackets]

It seems this does not affect all games: #3179. That fact is probably very important.

Understanding what games do and don't work (and more importantly what those games do and don't do) would probably help a lot.

-[Unknown]

[bsauvage1]

Yes, I have tried systematically those games I have and here's the outcome:

Starting every time:
Hakuoki
God of war chains of Olympus

Starting sometimes (random, may depend on what programs are in the
background, but haven't been able to draw a conclusion yet):
Star ocean 2nd evolution
Ff7 crisis core

Not starting:
Persona 3 portable
Metal gear solid peace walker
Danganronpa
Kingdom hearts birth by sleep
Disgaea infinite
Monster hunter freedom unite
Patapon 3
The 3rd birthday
Corpse party

On Thursday, January 30, 2014, Unknown W. Brackets wrote:

It seems this does not affect all games: #3179#3179.
That fact is probably very important.

Understanding what games do and don't work (and more importantly what
those games do and don't do) would probably help a lot.

-[Unknown]

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5047#issuecomment-33658621
.

[bsauvage1]

@unknownbrackets
trying to build anew in Xcode 5 on Mavericks, and I now get the below error:
No architectures to compile for (ARCHS=armv7, VALID_ARCHS=i386 x86_64).
[already posted this in the iOS forum, as it may be a better place to do so than here]

[unknownbrackets]

Did you use the ios toolchain file with cmake as per the instructions? If yes, I don't know... maybe the toolchain file isn't working well with the new Xcode version.

-[Unknown]

[bsauvage1]

if by this you mean http://iosopendev.com/download/ then yes, unfortunately.

Sorry I'm very new at this, and would really need some help to get going... I hope one of the iOS devs specialists can help compile the old versions (between v0.9.6-338-g367a5b7 and v.0.9.6-566-gb333e5b) to narrow down the bug.

[bsauvage1]

I have reinstalled macports, then cmake, then started anew from scratch, and now stuck when compiling ffmpeg - gcc unable to create an executable file ...

[sky-zer]

May be relevant:
Monster Hunter Freedom Unite crash when loading first mission. iPad3 (A5X), iOS 7.0.4.
Interestingly, only happens with "Release" or "RelWithDebInfo" builds, "Debug" build doesn't crash.
Callstack:

[unknownbrackets]

Oh, it's in xxhash?

2e324c9

Try disabling the XXH_USE_UNALIGNED_ACCESS define.

-[Unknown]

[sky-zer]

Yep disabling the XXH_USE_UNALIGNED_ACCESS fixed it.

[unknownbrackets]

Sounds like we need an iOS check or something. Maybe clang provides us with information gcc doesn't here.

Interesting if this is affecting iOS 7 specially, though...

-[Unknown]

[sky-zer]

Tried to compile with older XCode (4.6.3 + iOS SDK 6.1) - Monster Hunter works without disabling XXH_USE_UNALIGNED_ACCESS. Seems like issue with newer Clang, not iOS 7.
But i think it wouldn't hurt to disable unaligned access for iOS only, moreover aligned access should be faster for it.

[unknownbrackets]

Ah, newer clang. It may even be that it was not generating unaligned accesses before even when requested, and now they changed it to be able to... or something.

Well, XXH_USE_UNALIGNED_ACCESS has confused both me and @xsacha. The comment above it and its name make it hard to be sure if defining it as 1 makes it use aligned addresses or not.

But yes, whatever is faster and less crashy seems better for sure. Doing it specifically for iOs makes sense.

Worth noticing: xxhash has been updated since we pulled it in:
http://code.google.com/p/xxhash/source/list

The latest change is related to alignment so this may be related. We should probably update it.

-[Unknown]

[solarmystic]

As a followup onto my previous comment (#5047 (comment)), I've discovered a build that does not crash with the games I've tested, even when Hardware Transform (which as I mentioned earlier on seems to be the decider) is enabled on my JB'ed Ipad 3 with 6.1.x firmware.

It's 0.9.6-187 038b0f5 which I manage to find thanks to this helpful poster (Nishe76, http://forums.ppsspp.org/showthread.php?tid=9055&pid=71689#pid71689) in the iOS section of the ppsspp forums.

Games tested:-
Monster Hunter Freedom Unite
Dead or Alive Paradise
Crisis Core

@hrydgard @unknownbrackets
I tried grabbing the iPhone Configuration Utility as suggested but it doesn't seem to log anything useful as to why PPSSPP crashes when HW Transform is enabled in those games. I was thinking it'd be as helpful as a MSVC stacktrace for Windows builds, but it provides the same level of information similar to a Windows Error Reporting crash dialogue. In short, not very helpful at all.

@brujo5
I think I understand, just a little bit, how frustrating it must be to be an iOS user of this application and not able to easily post helpful information about issues from your device.

Also, is there a way to take in app screenshots for the iOS ppsspp app to attach it to a github post when reporting issues?

EDIT:-
Thanks for that [Unknown]

[unknownbrackets]

Press the power and home buttons at the same time (iirc.) It will take a screenshot in your Photos.

So, I guess sometime after that build is when the problem started?

-[Unknown]

[bsauvage1]

Just to recap, there seems to be 2 problems at play:
-the HW Transform problem mentionned above, that started after -187
-the crash to desktop problem (that was hopefully fixed by the latest commit - will check once a new build is available) that appeared sometime between -338 and -566.

[bsauvage1]

With the latest build (v0.9.6-802) the crash to desktop is still present on A5 devices.

That said, turning off HW Transform seems to randomly make some games start (for example Danganronp, FF7CC , MGSPW, started - but still not P3P, Patapon 3, 3rd Bday - and these just don't start anymore: Star Ocean, GoW, Hakuoki).

[unknownbrackets]

It's probably not hardware transform, but rather vertex cache. The problem is likely xxhash.

Hardware transform off means vertex cache is also off. You should get better performance with vertex cache off and hardware transform on.

-[Unknown]

[bsauvage1]

HW Xfm ON + Vertex Cache off: still crash to springboard

HW xfm OFF : Vertex cache OFF:
launching: Hakuoki, Book of Shadows, Star Ocean
still crashing: Danganronpa, FF7CC, KHBBS, Disgaea, P3P, MHFU, Patapon3, 3rd Bday

[unknownbrackets]

Hmm. Then it's probably a shader compile error I guess.

I was under the impression that some games started to crash after the xxhash change, that weren't crashing before. Maybe this was only on non-A5 or non-iOS 7 devices.

-[Unknown]

[bsauvage1]

The commit you did to fix xxhash seemed to do the trick at least partially! So here the outcome:
Works with HW transform on: star ocean 2nd coming, GoW chains of Olympus,
Needs HW transform off and buffered rendering off: Hakuoki, danganronpa
Needs HW transform on and buffered rendering off: book of shadows
Works with HW transform on but need software skinning off: MGS peace walker, khbbs, ff7cc, p3p, 3rd bday
Doesn't work anymore whatever settings combo is used : monster hunter freedom unite

[bsauvage1]

So it seems the bulk issue is fixed and maybe this thread can be closed, @unpeng ?

[unknownbrackets]

How is this now?

-[Unknown]

[bsauvage1]

As far as I can tell, it's all good now since the commits from kaienfr

On Sunday, March 16, 2014, Unknown W. Brackets notifications@github.com
wrote:

How is this now?

-[Unknown]

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5047#issuecomment-37764171
.

[bsauvage1]

(I think this was fixed at same time as #5513 but as I didn't extensively
test loading saves, it would be best if @unpeng could confirm)

On Sunday, March 16, 2014, Benoit Sauvage
<benoit.sauvage@gmail.comjavascript:_e(%7B%7D,'cvml','benoit.sauvage@gmail.com');>
wrote:

As far as I can tell, it's all good now since the commits from kaienfr

On Sunday, March 16, 2014, Unknown W. Brackets notifications@github.com
wrote:

How is this now?

-[Unknown]

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5047#issuecomment-37764171
.