Add PSOPrecache file which wraps engine level PSO structs and functions
- FPSOPrecacheParams used to wrap certain parameters which drive the PSO collection (mostly set from components)
- IPSOCollector interface used as engine entry to the mesh pass processors in renderer project
- FPSOCollectorCreateManager to wrap all statically registered IPSOCollector CreateFuncions
- PSOCollectorStats structs and helper functions
Add PSOPrecache support to PipelineStateCache:
- functions to precache compute and graphics PSOs
- functions to check current precache state
- FPrecacheGraphicsPipelineCache which wraps all currently precached PSOs
IMaterial entry function to PrecachePSOs with given vertex factories and params (implemented by UMaterial & UMaterialInstance) which forwards call to FMaterial and finally the FMaterialShaderMap
- FMaterialShaderMap will iterate all registered IPSOCollector to collect all possible PSOs used and forward them to the PipelineStateCache for actual (async) precaching
RHI Changes:
- Add RHI specific functions to retrieve & compare the FRHIVertexDeclaration and FGraphicsPipelineStateInitializer hash for each RHI
- Add helper functions to match RHI*State data by retrieving initializer
Minimal vertex factory changes for PSO precaching as prep - actual changes in upcoming CLs
MeshPassProcessor base functions to collect PSOs which can be used by all mesh pass processor implemention (actual changes in upcoming CLs)
- Cache complete precache data hash of stored minimal PSO so it can be used for fast actual PSO initializer hash computation (full PSO initializer is build at runtime)
- Cache if minimal PSO was successfully precache and option to skip draw when PSO is still precaching (disabled by default and doesn't work yet via GPU scene based rendering)
- Add helper functions to setup the per mesh pass render target data used during PSO collection
Refactor SceneTextureConfig so it can be initialized from an init struct and move all function to setup the internal members from SceneTexture to SceneTextureConfig using the data from the init struct (so it's available in Engine project)
Fix GBufferInfo format for velocity target (was always floating point while it should be unorm when not android
Make certain helper functions definitions to h file so they can be used everywhere (SupportsNaniteRendering, FTextureRenderTarget2DResource::GetFormat & CreateFlags)
#preflight 63189a5a967ffc68fb9044a5
#jira UE-139584
#rb Mihnea.Balta
[CL 21907508 by kenzo terelst in ue5-main branch]
- GetRayTracingCallableLibrary() similar to how we handle hitgroups.
FindRayTracingCallableIndex(...) to query callable shader index in RTPSO.
#rb yuriy.odonnell
#preflight 628620c99016c6dd8974b510
[CL 20278150 by tiago costa in ue5-main branch]
Changed macro used to resolve overloads for these 2 functions to the one requiring one actual float/double type in the arguments. Removed old macro that allowed mixing signed/unsigned int types and replaced it with one that requires all the types to be signed to match any overloads (tested but currently unused).
Resolved some compile errors that resulted from newly ambiguous calls that mixed signed/unsigned integer types.
#jira UE-143473
#rb Andrew.Davidson
#preflight 6215382b30639b44d2f52355
#ROBOMERGE-AUTHOR: zak.middleton
#ROBOMERGE-SOURCE: CL 19098314 via CL 19099699 via CL 19100188 via CL 19100709 via CL 19109894
#ROBOMERGE-BOT: UE5 (Release-Engine-Staging -> Main) (v921-19075845)
[CL 19147214 by zak middleton in ue5-main branch]
- Provides about 8% runtime memory savings (in local tests).
- Also, adds more compression types for shaders.
- Impact on the shader compilation (in SCWs) seems to be negligible in local tests.
#rb Devin.Doucette, Charles.Bloom
[REVIEW] [at]Devin.Doucette, [at]Charles.Bloom, [at]Jason.Nadro
#jira UE-136845
#ROBOMERGE-AUTHOR: arciel.rekman
#ROBOMERGE-SOURCE: CL 18502862 via CL 18503105 via CL 18503112 via CL 18505939 via CL 18505950
#ROBOMERGE-BOT: STARSHIP (Release-Engine-Staging -> Release-Engine-Test) (v899-18417669)
[CL 18505961 by arciel rekman in ue5-release-engine-test branch]
- Splitting symbol generation and shader "debug info" generation into two different cvars. This should now allow us to generate symbols post-packaging without generating different shader variants. This should also enable us to always generate symbols and still have the final deduplicated shader sizes.
- r.Shaders.Symbols is now the primary way to enable shader symbols for debugging.
- r.Shaders.GenerateSymbols can be enabled to always generate symbols without writing them to disk.
- r.Shaders.WriteSymbols can be used to force write symbols if they were previously generated.
- r.Shaders.ExtraData can be used to generate shader names for engine runtime systems to use.
- NotifyShaderCooked replaced with NotifyShaderCompiled, prep for upcoming non-cook symbol writing.
- PC DXC shader compiles should only generate debug info when requested, not all the time.
- PC FXC shader compiles should generate debug info when requested, not just when r.Shaders.Optimize is 0.
#jira none
#rb arciel.rekman, lukas.hermanns
#preflight 611ad9035e737200015992db
#ROBOMERGE-SOURCE: CL 17186155 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v855-17104924)
[CL 17186162 by christopher waters in ue5-release-engine-test branch]
- number of supplied parameters less or more than the number of specifiers used in format string
- invalid format string specifiers used
#rb Matt.Peters
#ROBOMERGE-SOURCE: CL 16315056 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v804-16311228)
[CL 16315068 by ionut matasaru in ue5-release-engine-test branch]
- FShaderMapResourceCode derives from FThreadSafeRefCountedObject which has a member that keeps track of the number of references to the object. We were doing deep copies using the Copy Constructor of FShaderMapResourceCode which resulted in a member wise copy of the num refs. This is not the behavior you'd expect.
- The offending line of code that caused the memory leak was in FShaderMapBase::AssignCopy:
- Code = new FShaderMapResourceCode(*Source.Code);
- Like FRefCountBase I've deleted the copy constructor and assignment operator for FRefCountedObject and FThreadSafeRefCountedObject to catch issues at compile time where we are copying across NumRefs incorrectly.
- To fix the issue I've manually created copy constructors for FGPUProfilerEventNodeStats and FShaderMapResourceCode.
#rb Devin.Doucette, Dave.Jones, Arciel.Rekman, Ben.Ingram
#review-16115628 @Devin.Doucette, @Dave.Jones, @Arciel.Rekman, @Ben.Ingram
#jira UE-108598, UE-113736
[CL 16116577 by Jason Nadro in ue5-main branch]
Fix issues causing material DDC values to be non-determinisic
- Shader code within resources sorted by hash
- Debug source code values sorted by type
Also add mechanism to serialize any MemoryLayout described type to string
- Makes it easier to inspect material data to find sources of non-determinism
#rb none
#ROBOMERGE-OWNER: ben.ingram
#ROBOMERGE-AUTHOR: ben.ingram
#ROBOMERGE-SOURCE: CL 12944988 in //UE4/Release-4.25/... via CL 12944998 via CL 12957822
#ROBOMERGE-BOT: RELEASE (Release-Engine-Staging -> Main) (v682-12900288)
[CL 12957827 by ben ingram in Main branch]
Fix generation of shader debug data
#rb none
#ROBOMERGE-SOURCE: CL 11850208 in //UE4/Release-4.25/... via CL 11850214
#ROBOMERGE-BOT: RELEASE (Release-4.25Plus -> Main) (v656-11643781)
[CL 11850701 by ben ingram in Main branch]
