The previous code was hardcoded to support a maximum of 128 cores and would assert if more cores were detected.
The problem was reported on UDN.
#jira UE-227879
#rb elizabeth.bunner, ben.woodhouse
#lockdown mark.lintott
[CL 37217479 by Martin Sevigny in 5.5 branch]
Fixed Sim Cache writes for CPU and engine tests reliant on it. (GPU sim cache write will have to wait given current timing) (UE-225994)
Fixed shader error due to redundant defaulting shader code. (UE-227814)
Fixed editor interactions with NDC assets (UE-227810)
- Inserting into variables array breaks existing entries
- Undo/Redo doesn't work in many cases
- Edits to NDC assets would not take effect for running systems until GC or re-init.
Made GPU/CPU Reads and Writes behave consistently in response to missing NDC data or OOB access. (UE-227886)
- Missing data returns defaulted values. (previously could be uninitialized)
- Valid params are correctly read/written even if there are missing params. Previously one missing parameter could prevent other access functioning.
- Overall success returns false if there are any missing params or access is OOB.
#jira UE-227619, UE-225994, UE-227814, UE-227810, UE-227886
#rb Stu.McKenna
#lockdown marc.audy
[CL 37203014 by Simon Tovey in 5.5 branch]
#rb [at]Lukasz.furman
[REVIEW] [at]Lukasz.Furman
#tests Verified that disable compiler optimizations flag functioned as expected.
[CL 37095919 by ben temple in 5.5 branch]
This also disables linking the static swappy lib, which was redundant (linker didn't use its symbols)
#rb Jack.Porter
[CL 37088336 by florian penzkofer in 5.5 branch]
While investigating Android app suspend workflow found that game thread missed APP_EVENT_STATE_APP_SUSPENDED and continued GEngineLoop.Tick(); and locked on FrameEndSync.Sync call because render thread is already suspended, but should be blocked EventHandlerEvent->Wait();
As a result after resuming render tread generates a frame from the previous cycle.
#jira UE-219578
#rb Chris.Babcock, ahmed.siddique
#rnx
[CL 37023037 by Yevgen Abramov in 5.5 branch]
This rolls back CL 35966216.
Due to way some platforms destruct TLS before terminating threads, `FMallocLeakDetection::SetDisabledForThisThread(false)` can incorrectly fail `check(Count >= 0)`, despite a paired `SetDisabledForThisThread(true)` call. This can happen when freeing memory as part of thread destruction.
#rb john.stiles, neil.henning
[CL 36905886 by ben clayton in 5.5 branch]
One change on the engine side I had to do is undefine our versions of FLT/DBL_MIN/MAX in Scalar.isph because these definition are not available as built-ins and generate warnings if present. That's something we already did for PI before.
#rb daniele.pieroni, tomasz.obrebski
#jira UE-225798
[CL 36824059 by Wojciech Krywult in 5.5 branch]
Selecting "link" or "copy" is done automatically by checking if UE is running from an installed place or if it is a dev/p4 stream place, you can override this with the option -ZenAutoLaunchInstallMode.
The ZenAutoLaunchInstallMode defaults to "auto" where it tries to detect if it is an installed engine or not. "link" forces the use of a link file and "copy" forces the copy-install option which is the legacy behavior.
Additional fixes:
FMacSystemWideCriticalSection which could fail to aquire the critical section causing launch to fail now uses flock in the same manner as FUnixSystemWideCriticalSection which resolves the issues
When checking if processes are running on Unix/Mac we now detect zombie processes properly
#jira UE-224517
#rb zack.neyland, Zousar.Shaker
[CL 36764363 by dan engelbrecht in 5.5 branch]
The game was deadlocking due to UI elements being accessed in non-main threads.
The main issue here is that running the game on an iPhone 13 with iOS 18.0 the game will almost certainly hang at some point.
With this fix, the game runs smoothly.
During gameplay, it seems to have improved performance.
But I was not able to measure yet due to the inability to run Insights Captures, it will OOM.
Context here: https://epicgames.slack.com/archives/C06S7N8LAD7/p1727302915935169
#rb adam.kinge, Peter.Sauerbrei
#rnx
[CL 36762847 by anderson ramos in 5.5 branch]
- -ZenPak used to indicate streaming using pak files instead of Zen loose file streaming
- added pak file hardware decompression on supported platforms
#jira UE-221577
#rb David.Harvey, daniele.pieroni
[CL 36761256 by tomasz obrebski in 5.5 branch]
