- move uniform buffer cleanup and dead stripping into ShaderPreprocessor module's PreprocessShader function
- add "required symbols" to compiler input struct to specify additional symbols to keep during minification aside from those specified by the entrypoint; modify API such that both an entry point string and additional symbols can be specified (to avoid each backend needing to manually parse the compound RT entry point string)
- make use of ModifyShaderCompilerInput in all backends to set additional defines and required symbols on input struct up front; only use the AdditionalDefines map in cases where it's actually necessary
- remove the various per-platform defines for enabling minifier, no longer required now that this has been rolled out for all backends
- fix SCW directcompile mode; this had rotted due to pieces of the FShaderCompilerEnvironment having been added that weren't explicitly serialized to either cmdline or in the shader source. this now serializes as a base64 string written inside the USF containing all portions of the environment required for compilation (using the same serialization function as is used to write/read the SCW input file)
- use a debug flag for indicating we're in "direct compile" mode and should load the debug USF off disk, rather than the poorly named "bSkipPreprocessedCache" (this name is both inaccurate and also confusing due to the addition of the preprocessed job cache)
- modify platform "force wave32" mechanism to use a pragma directive to set a compiler define, instead of doing string replacement in the preprocessed source
- add a view version of the RT entrypoint parsing to use in preprocessing, note that other paths still need to construct fstrings due to further manipulation so keeping the FString path around too
- clean up backends manually checking the "directcompile" cmdline arg
#rb christopher.waters, Yuriy.ODonnell
#rb Chris.Waters
#rb Laura.Hermanns
[CL 30023082 by dan elksnitis in ue5-main branch]
- Disable COMPILER_SUPPORTS_WAVE_VOTE for Mac
- Fixed bug where Texture2DArray was flattened due to atomics, but bound as Texture2D
- Fixed bug where defines were read incorrectly in shader compiler
#rb Zack.Neyland, Jamie.Hayes
[CL 26291315 by carl lloyd in ue5-main branch]
* Early out when adding uniform buffer resource entries, by checking if uniform buffer had been processed, before processing individual entries.
* Changed resource entry list from a map to an array, saving the cost of map operations adding items to it.
* Resource entries for global uniform buffers cached once at startup.
* Resource entry member names stored in an external buffer, to avoid per-entry string memory allocation overhead.
* Miscellaneous smaller optimizations -- for example, using ByHash to avoid redundant hashing and FString construction.
#jira none
#rnx
#rb jason.nadro dan.elksnitis
#preflight 646cfdbc1134ffac7034af60
[CL 25596308 by jason hoerner in ue5-main branch]
This changelist is a merge of work done in //UE5/Dev-ParallelRendering to remove tech debt in the SRV/UAV create functions. See individual changelist descriptions in that stream for further details, but a summary of the change is as follows:
- We had multiple overloads of each of the RHICreateShaderResourceView and RHICreateUnorderedAccessView functions. Each function had a different set of supported view types and using them was complicated / ambiguous.
- There's now only 2 create functions, and a common FRHIViewDesc descriptor struct, which is stored on the base view class.
- FRHIView takes a reference on the underlying resource, ensuring that RHI views keep their viewed resource alive. This was not the case on some platforms, since it was previously the platform RHI implementation's responsibility.
- Platform RHI implementations resolve an FRHIViewDesc into a full FViewInfo struct using GetViewInfo. This centralizes the logic for computing num mips / num slices / format etc so it is the same across all platforms.
- Views must never be created with nullptr resources. This used to happen in the mesh streamer, but now all views require a real underlying resource. That resource can be a "BUF_NullResource" buffer, which is what the mesh streamer uses as a placeholder for buffers that have not streamed yet.
- We will eventually replace FRHITextureReference with a "null texture", similar to how BUF_NullResource works for buffers. This is not yet implemented, so there is no "null view" of a texture currently.
#rb kenzo.terelst,jeannoe.morissette,dmitriy.dyomin
#preflight 643534642855180717af410e
#jira none
[CL 24989901 by Luke Thatcher in ue5-main branch]
The shader code output contains a little bit of plaintext such as I/O identifiers and the shader name, which was generated for the DXC backend in a way that prevents caching.
This unifies the shader name in the output code and improves the cache hit ratio for a Lyra cook from 0% to roughly 50% cache hits (for certain platforms).
#rb Jason.Nadro, Carl.Lloyd
#fyi Brian.White, Arciel.Rekman, Brandon.Schaefer
#jira FORT-548603
#preflight 63e26c6c07207b2570ba2d5b
#rnx
[CL 24050212 by Laura Hermanns in ue5-main branch]
* Enabling requires re-building a metalpp with appropriate flags/defines set along with an OS version that supports the RT headers/libs.
#preflight 63c72f1a0225f00e14c05c46
#jira none
[CL 23750025 by zack neyland in ue5-main branch]
Metal 2.3 removed, Metal 3.0 added for Mac
#jira UE-163352
#rb adam.kinge bertrand.carre
#preflight 631ea7eb94758d0bf2008668
[CL 21960671 by axel riffard in ue5-main branch]
- Devices crashing when using Base_Vertex metal shader semantics
- Runtime error when using Virtual Texturing on A8 devices which is unsupported.
#jira UE-131863, UE-132217
#rb Will.Damon
[FYI] Jack.Porter
#preflight 62a75ad63d0a713cf8caf486
#ROBOMERGE-AUTHOR: carl.lloyd
#ROBOMERGE-SOURCE: CL 20631592 in //UE5/Release-5.0/... via CL 20649245
#ROBOMERGE-BOT: UE5 (Release-Engine-Staging -> Main) (v955-20579017)
[CL 20656300 by carl lloyd in ue5-main branch]
