* Issue is that the adapters were referenced by index
* Adapters enumeration was sometime performed using EnumAdapters() vs EnumAdapterByGpuPreference()
* Using the same index for both calls would result in different adapters being returned...
On AWS g4ad.16xlarge instances, this issue would cause the engine to think it had selected a discrete adapter, while in fact it created a WARP (software) adapter...
#rb christopher.waters
#preflight 611a817e8ff5540001d5e805
#ROBOMERGE-SOURCE: CL 17176596 via CL 17176598
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v855-17104924)
[CL 17176600 by sebastien lussier in ue5-release-engine-test branch]
useful for RVT on mobile platforms, to save the cost of runtime compression
#rb Dmitriy.Dyomin mihnea.balta
#ROBOMERGE-SOURCE: CL 16872220 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v836-16769935)
[CL 16872238 by yangke li in ue5-release-engine-test branch]
#rb kenzo.terelst
#preflight 60c9ebb93e1b3c0001400246
- The original code is using ::LoadLibrary with a relative path to find the dll and when running from staged data via -basedir= we will be relying on that path being relative to the current directory set by the command line. However many modules call FPlatformProcess::PushDllDirectory/PopDllDirectory to find dlls and when all of the paths are fully popped we will call ::SetDllDirectory with an empty string which will remove the current directory from the list of valid paths for ::LoadLibrary to search.
- To fix this we can call FPlatformProcess::PushDllDirectory/PopDllDirectory ourselves, with the location of the dll and then request the handle of the dll instead (examples of this can be found in other modules)
- When running with DX12 we no longer attempt to enable NvAftermath if we previously failed to find the dll in order to prevent a crash. We do this by recording if the dll has been loaded or not via a new global variable GDX12NVAfterMathModuleLoaded, similar to how the DX11 code path tracks this.
#ROBOMERGE-SOURCE: CL 16690777 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v835-16672529)
[CL 16690799 by paul chipchase in ue5-release-engine-test branch]
Integration touches several places in the engine:
1) SoundWave -- A check box enables Bink Audio as the codec of choice for that sound wave.
2) Decoder - Each supported platform's AudioMixer now returns BINK if the soundwave requests it. Additionally, the TargetPlatform returns BINK as an available codec, and returns it to the cooking code if the sound wave requests it.
3) Encode - TargetPlatform.Build.cs adds the encoder to the editor dependencies, and it gets picked up in the TPMM formats search.
#ROBOMERGE-SOURCE: CL 16682710 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v835-16672529)
[CL 16682731 by dan thompson in ue5-release-engine-test branch]
PipelineStateCache specifically can now enqueue commands but this should also catch any other uses before here.
Adding explicit Command List argument to FRHIResource::FlushPendingDeletes to a) emphasize that it enqueues commands and b) allow the caller the option to flush the command list afterwards.
#jira none
#rb Arne.Schober
#preflight 60a6c7f60569f300012d6d01
[CL 16411660 by christopher waters in ue5-main branch]
Adding usage text to rhi.DumpResourceMemory.
Tweaking rhi.DumpResourceMemory output to be more explicit with the number of items shown.
#jira none
#rb kenzo.terelst
#preflight 60917d13a370e100017d3813
[CL 16197762 by christopher waters in ue5-main branch]
