You've already forked UnrealEngineUWP
mirror of
https://github.com/izzy2lost/UnrealEngineUWP.git
synced 2026-03-26 18:15:20 -07:00
c8b108b1ed
#rb none #lockdown Nick.Penwarden ============================ MAJOR FEATURES & CHANGES ============================ Change 3443778 by Ben.Woodhouse Aliasing for transient resources + new high level API Changelists integrated: 3368830 3368887 3377762 3377763 3379513 3381840 338204633821383385390 3385391 3385531 3396613 3388752 3396756 3397007 3397059 3397780 3397883 3401716 3415179 Change 3576521 by Bart.Hawthorne ShooterGame - update the game instance's online mode when the LAN option is changed Change 3587252 by Ben.Woodhouse Integrate from //UE4/Main/...@3584123 to //UE4/Dev-Console/... Change 3587943 by Luke.Thatcher [CONSOLE] [+] Improved XGE shader compiler - Added support for XGE interception interface, which allows asynchronous streaming of shader tasks into XGE, and eliminates the stalling/flushing found in the old system. - Old system has been renamed "Xml interface" - Both systems are available and selectable based on r.XGEShaderCompile.Mode CVar. Overview of the new system: - New system uses an additional XGEControlWorker process to manage communication between the engine and XGE. This is simply a copy of ShaderCompileWorker.exe (the implementation is part of SCW, but XGE requires the two files to be separate, hence the duplicate). - When shader jobs arrive from the engine, the shader compiler thread spawns an instance of XGEControlWorker within an XGE context, and connects to it via named pipes. - Job command lines are fed to the control worker via the pipe, which the worker enqueues within XGE by issuing a CreateProcess call, which is intercepted. Job completion is reported back to the engine via the named pipes. - New system is preferred over the old one, and is enabled with "-xgeshadercompile" command line switch or "r.XGEShaderCompile" cvar. - The new system depends on the "XGEController" engine module. Support for which is based on the WITH_XGE_CONTROLLER #define, or -xgecontroller/-noxgecontroller command line options. #jira UECON-263 Change 3590037 by Luke.Thatcher [CONSOLE] [!] Fix compile/link errors with XGE controller module - Various unity related fixes needed. Change 3590047 by Luke.Thatcher [CONSOLE] [!] Compile XGE controller module for Editor builds only - The module should only be used in uncooked, non-monolitic binaries (i.e. the editor). Change 3594074 by Luke.Thatcher [CONSOLE] [!] Fix Linux build - case sensitive header include was wrong in XGEControllerModule.cpp Change 3596292 by Luke.Thatcher [CONSOLE] [!] Actually fix unity/nopch build errors in XGE Controller! Change 3596776 by Jonathan.Fitzpatrick cherry pick CL 3595292 cooker change which fixes a condition where DLC would always cook with all maps regardless of which maps were specified. Change 3596795 by Jonathan.Fitzpatrick Back out changelist 3596776 Change 3596816 by Jonathan.Fitzpatrick Moved the DLC map check ahead of the "no maps found" condition so that include-all-maps can be skipped if DLC maps are found. Change 3604282 by Ben.Marsh UBT: Add support for pre-build and post-build steps when using MegaXGE. Change 3606862 by Ben.Marsh Fix missing header guard warning. Change 3611644 by Ben.Woodhouse Integrate from //UE4/Main/...@3610522 to //UE4/Dev-Console/... Change 3614089 by Joe.Barnes Call GRHICommandList.LatchBypass() after variables it checks are set. #jira ue-48130 Change 3617783 by Daniel.Eldar Headers for the HDR changes (they are pre-requisites for compilation). Integrated from: //UE4/Partner-Microsoft-DX12/Engine/Source/ThirdParty/Windows/DirectX/... to //UE4/Dev-Console/Engine/Source/ThirdParty/Windows/DirectX/... #jira UECON-226 Change 3619117 by Ben.Woodhouse Integrate from //UE4/Main/...@3617655 to //UE4/Dev-Console/... Change 3619123 by Ben.Woodhouse Integrate from //UE4/Main/...@3618361 to //UE4/Dev-Console/... Change 3619603 by Stewart.Lynch Merging //Fortnite/Dev-Athena/Engine/Source/Runtime/Engine/Private/Materials/MaterialInstanceConstant.cpp to //UE4/Dev-Console/Engine/Source/Runtime/Engine/Private/Materials/MaterialInstanceConstant.cpp Changed UMaterialInstance::PostLoad(); to Super::PostLoad(); Change 3619608 by Stewart.Lynch Merging //Fortnite/Dev-Athena/Engine/Source/Runtime/... to //UE4/Dev-Console/Engine/Source/Runtime/... LLM update * made sure that all csv stat tracking is disabled if running without -LLMCSV * added new UObject scopes * increased LLM_MAX_THREAD_SCOPE_STACK_DEPTH to 128 to cope with new UObject scopes * added material scopes * changed UMaterialInstance::PostLoad(); to Super::PostLoad(); Change 3619634 by Stewart.Lynch Merging //Fortnite/Main/... to //UE4/Dev-Console/... General LLM improvements * added tracking for misc task graph tasks (moves 20MB out of Untagged) * renamed EngineTick to EngineMisc * added tracking for FName * added tracking for GC_ProcessObjectArray potential leak * renamed index & vertex buffers stat to Meshes * added hooks for MemPro to track allocations from a single category. Currently defined out. I haven't added MemPro.cpp/h. * removed AVAILABLE_PHYSICAL stat from LLM csv * csv files now include the date in the filename * fixed potential threading bug when reading stat values to csv * added lots more scopes * started changing Stat scopes to enum scopes. Stat scopes will be phased out. * added tracking of FName memory * added llm tracking for CPU symbol allocations (20MB) * added tracking for GC * fixed tracking for TransientMemoryAllocator * added tracking for networking memory * added more audio memory tracking * added tracking for blueprints * added tracking for static meshes * show on screen warning if debug memory is enabled * added tracking for particles * renamed Phys to PhysX and added more scopes * renamed Slate to UI and added more scopes * much better coverage of networking memory * improved coverage of audio Change 3619642 by Stewart.Lynch Merging //Fortnite/Main/Engine/Source/Runtime/Core/... to //UE4/Dev-Console/Engine/Source/Runtime/Core/... Windows builds were incorrectly showing the debug memory enabled warning. Made IsDebugMemoryEnabled always return false on Windows. Change 3619799 by Luke.Thatcher [CONSOLE] [-] Disable peak memory tracking in LLM. Change 3619809 by Luke.Thatcher [CONSOLE] [^] Merging CL 3580894 from //Fortnite/Main/... to //UE4/Dev-Console/... - Add Render Target Pool Stats. Use "stat rendertargetpool" to view. Change 3619928 by Luke.Thatcher [CONSOLE] [^] Enable symbol compression on symstore task (Merging CL 3606945 from //Fortnite/Main/... to //UE4/Dev-Console/...) Change 3620571 by Keith.Judge Double the size of the occlusion query buffer in D3D12. This costs 256KB of memory. We need to refactor this in future to resize according to content on demand like D3D11.x. #jira FORT-53103 #jira FORT-53119 Change 3622419 by Daniel.Eldar [General] Fix for occasional ensure to do with retriving the new cvar value. #JIRA UE-46644 Change 3622491 by Daniel.Eldar [General] Added support for scaling on stat unit and unitgraph. #JIRA UE-46644 Change 3622610 by Daniel.Eldar [General] Fixed a compilation warning Change 3624659 by Ben.Woodhouse Integrate from //UE4/Main/...@3624175 to //UE4/Dev-Console/... Change 3624678 by Ben.Woodhouse Integrate from //UE4/Main/...@3624415 to //UE4/Dev-Console/... Change 3624878 by Ben.Woodhouse Integrate as edit CLs 3611835 3611844 3616942 3617607 Descriptor heap memory fix (saves up to ~100MB). * Fix a bug where rolling over a thread-local heap immediately after switching to context-local would crash due to an uninitialized syncpoint * Add a separate value for context local heaps (LOCAL_HEAP_SIZE) and set it to a sensible value (2MB) instead of the current 16MB. This means thread local heaps will be allocated in chunks of ~2MB instead of 16MB. There can be up to 6 thread-local heaps, since they're per-context. Measured peak descriptor heap memory with LLM: 45MB (compared to 145MB previously) This improves behaviour, but this code could use a bit of a rewrite to allow it to reclaim memory better. For example, the global heap memory is almost entirely wasted currently, since when a context switches to "context local strategy" it simply throws this memory away, never to be reclaimed. Having the pools thread-local is also not a good strategy. Ideally we want a thread-safe pool allocator for descriptor heaps which is shared between all the contexts . Change 3624882 by Ben.Woodhouse Fix virtual destructor warning in D3DX12.h Change 3626419 by Stewart.Lynch Merging //Fortnite/Main/... to //UE4/Dev-Console/... LLM Update - Memory reductions, Summary page, enum scopes, refactor and cleanup of tags * Remove all static arrays and hard limits from LLM. Everything is now dynamically allocated using the internal LLM allocators. The overhead when LLM is disabled is now only 48K (was 40MB) * re-wrote LLMMap. Now stores an int32 index rather then pointer in the HashMap array. Also, changed the Values to be arrays for structs instead of structs of arrays. Means that the tag can be stored in a single byte. Changed the size of the allocation size from int64 to int32. All this takes the memory down from around 600MB to 100MB. It was 120 bytes per allocation, now 29 bytes. * changed all LLM scopes over to enums. This has a number of benefits; LLM can be enable in Test, less CPU overhead, stored in a byte (LLM overhead /= 8) * summary page for content creators where all lower-level stats are grouped under one Engine stat * renamed ELLMScopeTag enum to ELLMTag * renamed LLM_SCOPED_TAG_WITH_ENUM macro to LLM_SCOPE * removed Tracker arg from LLM_SCOPE and added LLM_PLATFORM_SCOPE macro * fixed GenericPlatformMallocCrash stat. Although it seems not be be used anymore * fixed BackupOOMMemoryPool stat (now shows in both default and platform pages) * lots of changes adding/removing/renaming tags * added LLMArray and FLLMObjectAllocator classes * disabled asset tag tracking by default because it takes up so much memory even when not used * enable LLM in all non-shipping builds. In Test the on screendisplay won't show because it uses the stats system but it till still write out the csv. * all the stat macros have been left as they were and can be enabled on the LLM_STAT_TAGS_ENABLED define. These are needed for the asset tagging. * disabled LLM_TRACK_PEAK_MEMORY because there is a problem with the way it adds the peaks for multiple threads. This needs to be fixed. * added a CVar to control the csv write interval: LLM.LLMWriteInterval * added static arrays for the enum tags setup. Easier to manage and removes need for slow switch statements. * renamed FLLMThreadStateManager to FLLMTracker to make it consistent with the enum #jira UECON-264 Change 3626906 by Joe.Barnes Fix compile errors when nativizing when a property references a sub object of a dervied type with modified default properties Change 3628045 by Stewart.Lynch Merging //Fortnite/Main/Engine/Source/Runtime/Core/... to //UE4/Dev-Console/Engine/Source/Runtime/Core/... LLM - enabled LLM on Windows * fixed crash in FLLMObjectAllocator destructor. Only happens on Windows shutdown. * changed Align to AlignArbitrary in LLMArray. Small memory saving. * clear ThreadStateAllocator in FLLMTracker::Clear(). Small memory saving * set program size to zero on windows because it isn't relevant #jira UECON-282 Change 3632279 by Ben.Woodhouse Back out changelist3619862#jira UE-49472 [CL 3632432 by Joe Barnes in Main branch]