Files
UnrealEngineUWP/Engine/Source/Developer/Apple/MetalShaderFormat/Private/MetalShaderFormat.cpp

1140 lines
41 KiB
C++
Raw Normal View History

// Copyright Epic Games, Inc. All Rights Reserved.
Copying //UE4/Dev-Build to //UE4/Dev-Main (Source: //UE4/Dev-Build @ 3092544) #lockdown Nick.Penwarden #rb none ========================== MAJOR FEATURES + CHANGES ========================== Change 3079316 on 2016/08/05 by Ben.Marsh Better PCH selection in ShaderFormatOpenGL and MetalShaderFormat - make sure Core is the first included header. Change 3080579 on 2016/08/08 by Ben.Marsh Slate: Move DEBUG_TAB_MANAGEMENT into SDockingTabStack.h to remove circular include dependency with DockingPrivate.h. Change 3080587 on 2016/08/08 by Ben.Marsh StandaloneRenderer: Move platform includes into a separate header so we can make individual headers self-contained, without having a circular dependency on StandaloneRendererPrivate.h. Change 3080789 on 2016/08/08 by Ben.Marsh Move BuildGraph tasks for chunking, posting builds, labeling builds, and merging manifests into the MCP project. While we do provide public interfaces for this functionality, we don't currently expect anyone outside Epic to be using them. Change 3080815 on 2016/08/08 by Ben.Marsh BuildGraph: Add a -GenerateDocs option, which writes out an HTML file constructed from C# XML documentation containing all the task information. Change 3081374 on 2016/08/08 by Ben.Marsh UBT: Invalidate the makefile if any UHT headers are deleted. Should fix issue where files are moved from one module to another, and the original module no longer contains any generated headers. Its include path needs to be removed from the compile environment. Change 3083152 on 2016/08/09 by Ben.Marsh PR #2667: Add Intel C++ Compiler support to Windows build (Contributed by JeffRous) Change 3084039 on 2016/08/10 by Ben.Marsh BuildGraph: Add additional markup for parameter attributes. Also improve some documation. Change 3084240 on 2016/08/10 by Ben.Marsh Plugins: Allow plugins in the project folder to replace plugins in the engine folder with the same name. Prohibit multiple plugins with the same name at other times. Change 3084337 on 2016/08/10 by Ben.Marsh UBT: Specify the -precompile option when generating project files for a target, so we include all valid modules for intellisense. Change 3085594 on 2016/08/11 by Ben.Marsh Change modules which reference a public header for their PCH to use a private PCH instead, even if it just includes the public header for now. Change 3085999 on 2016/08/11 by Ben.Marsh Add some missing #pragma once directives. Change 3086146 on 2016/08/11 by Ben.Marsh Core: Move prototype and linkage specifier for ConsoleCommandLibrary_* functions into header matching cpp file. Change 3086172 on 2016/08/11 by Ben.Marsh Fixup some C-style header guards to use #pragma once instead. Change 3087289 on 2016/08/12 by Ben.Marsh Split out UPackage and UMetaData into their own headers (they're already implemented in separate CPP files) Change 3087310 on 2016/08/12 by Ben.Marsh Move method stubs for FNullSlateSoundDevice into a CPP file, since they're exported from the SlateCore module. Change 3087341 on 2016/08/12 by Ben.Marsh UdpMessaging: Move PCH before #if PLATFORM_DESKTOP; it will only be defined if the definition is included. Change 3087457 on 2016/08/12 by Ben.Marsh Core: Reorganize the FTransform and FMatrix headers: Transform.h now includes TransformNonVectorized.h or TransformVectorized.h as appropriate, and UnrealMatrix.h is now Matrix.inl (and included from Matrix.h). Change 3088407 on 2016/08/13 by Ben.Marsh Replace use of Windows SIZE_T define with the regular C++ size_t. Change 3088416 on 2016/08/13 by Ben.Marsh Include a header from all .generated.cpp files (GeneratedCppIncludes.h) which includes all the basic types required to compile them, rather than assuming that the module PCH will include everything. Also include the real declarations of noexport classes in Object.h (now renamed to NoExportTypes.h for clarity) when the CPP macro is defined, so the .generated.deps.h file will automatically have the correct definitions for them at compile time rather than relying on them being in the private PCH. Finally, rename UObject.h to Object.h for consistency with the naming convention for all other UObject classes. UObject.h still exists for now, but outputs a deprecated message if included. Change 3088544 on 2016/08/14 by Ben.Marsh Core: Move the definition of the TEXT() macro into Platform.h, to avoid having to include OS headers to get it. Change 3088552 on 2016/08/14 by Ben.Marsh Fix compile errors for some modules that don't already include CoreUObject.h. Change 3088925 on 2016/08/15 by Ben.Marsh Remove circular include dependencies from VulkanRHI. Change 3088926 on 2016/08/15 by Ben.Marsh Remove duplicate definition for WITH_FIXED_AREA_ENTERING_COST from EngineDefines.h - always uses the definition from DetourNavMeshQuery.h instead. Change 3088930 on 2016/08/15 by Ben.Marsh Remove circular include dependency from PhysX. Change 3088935 on 2016/08/15 by Ben.Marsh OnlineSubsystemUtils: Move CPP files out of public header directory. Change 3088965 on 2016/08/15 by Ben.Marsh Add private PCH to Landscape, MoviePlayer, TaskGraph, XAudio2 and RealtimeProfiler modules. Change 3088966 on 2016/08/15 by Ben.Marsh Engine: Move CPP files out of public header directories. Change 3089520 on 2016/08/15 by Ben.Marsh BuildGraph: Change documentation command to output markdown. Change 3090299 on 2016/08/16 by Ben.Marsh D3D12RHI: Move around some implementations to fix circular header dependencies which are masked by delayed template instantiation. Change 3090303 on 2016/08/16 by Ben.Marsh Engine: Add a template specialization for TPointerIsConvertibleFromTo<AActor, const volatile UObject> to fix dependency on complete AActor definition for static assert in TWeakPointer<AActor>, which only appears if including Level.h without Actor.h. Delayed template instantiation usually masks this issue. Change 3091861 on 2016/08/17 by Ben.Marsh Remove circular header dependencies, and fix ambiguous include paths in OSVR. Change 3092068 on 2016/08/17 by Ben.Marsh Moving VulkanDynamicRHI into its own header. Change 3093133 on 2016/08/18 by Ben.Marsh EC: Include additional context lines for Clang errors. Change 3093147 on 2016/08/18 by Ben.Marsh UBT: Add an error message when attempting to do a single-file compile with the wrong target selected. Change 3093228 on 2016/08/18 by Ben.Marsh Remove redundant setting for remote server name from XML config, and set it to a valid machine in the engine config. [CL 3093264 by Ben Marsh in Main branch]
2016-08-18 10:28:43 -04:00
#include "MetalShaderFormat.h"
Copying //UE4/Dev-Core to //UE4/Dev-Main (Source: //UE4/Dev-Core @ 3847469) #lockdown Nick.Penwarden #rb none ============================ MAJOR FEATURES & CHANGES ============================ Change 3805828 by Gil.Gribb UE4 - Fixed a bug in the lock free stalling task queue and adjusted a comment. The code is not current used, so this is not actually change the way the code works. Change 3806784 by Ben.Marsh UAT: Remove code to compile UBT when using UE4Build. It should already be compiled as a dependency of UAT. Change 3807549 by Graeme.Thornton Add a cook timer around VerifyCanCookPackage. A licensee reports this taking a lot of time so it'll be good to account for it. Change 3807727 by Graeme.Thornton Unhide the text asset format experimental editor option Change 3807746 by Josh.Engebretson Remove WER from iOS platform Change 3807928 by Robert.Manuszewski When async loading, GC Clusters will be created after packages have been processed to avoid situations where some of the objects that are being added to a cluster haven't been fully loaded yet Change 3808221 by Steve.Robb GitHub #4307 - Made GetModulePtr() thread safe by not using GetModule() ^ I'm not convinced by how much thread-safer this is really, but it's tidier anyway. Change 3809233 by Graeme.Thornton TBA: Misc changes to text asset commandlet - Rename mode to "loadsave" - Add -outputFormat option which can be assigned "text" or "binary" - When saving binary, use a differentiated filename so that source assets aren't overwritten Change 3809518 by Ben.Marsh Remove the outdated UnrealSync automation script. Change 3809643 by Steve.Robb GitHub #4277 : fix bug; FMath::FormatIntToHumanReadable 3rd comma and negative value #jira UE-53037 Change 3809862 by Steve.Robb GitHub #3342 : [FRotator.h] Fix to DecompressAxisFromByte to be more efficient and reflect its intent accurately #jira UE-42593 Change 3811190 by Graeme.Thornton Add support for writing specific log channels to their own files Change 3811197 by Graeme.Thornton Minor updates to output formatting and timing for the text asset commandlet Change 3811257 by Robert.Manuszewski Cluster creation will now be time-sliced Change 3811565 by Steve.Robb Define out non-monolithic module functions. Change 3812561 by Steve.Robb GitHub #3886 : Enable Brace-Initialization for Declaring Variables Incorrect semi-colon search removed after discussion with author. Test added. #jira UE-48242 Change 3812864 by Steve.Robb Removal of some unproven code which was supposed to fix hot reloading BP class functions in plugins. See: https://udn.unrealengine.com/questions/376978/aitask-blueprint-nodes-disappear-when-their-module.html #jira UE-53089 Change 3820358 by Ben.Marsh PR #4358: Incredibuild use ShowAgent by default (Contributed by projectgheist) Change 3822594 by Ben.Marsh UAT: Improvements to log file handling. - Always create log files in the final location, rather than writing to a temp directory and copying in later. - Now supports -Verbose and -VeryVerbose for increasing log verbosity, rather than -Verbose=XXX. - Keep a backlog of log output before the log system is initialized, and flush it to the log file once it is. - Allow buildmachines to specify the uebp_FinalLogFolder environment variable, which is used to form paths for display. When build machines copy log files elsewhere after UAT finishes (eg. a network share), this allows error messages to display the right location. Change 3823695 by Ben.Marsh UGS: Fix issue where precompiled binaries would not be shown as available for a change until scrolling the last submitted code change into the buffer (other symptoms, like de-focussing the main window would cause it to go back to an unavailable state, since the changes buffer was shrunk). Now always queries changes up to the last change for which zipped binaries are available. Change 3823845 by Ben.Marsh UBT: Exclude C# projects for unsupported platforms when generating project files. Change 3824180 by Ben.Marsh UGS: Add an option to show changes by build machines, and move the "only show reviewed" option in there too (Options > Show Changes). #jira Change 3825777 by Steve.Robb Fix to return value of StringToBytes. Change 3825810 by Ben.Marsh UBT: Reduce length of include paths for MSVC toolchain. Change 3825822 by Robert.Manuszewski Optimized PIE lazy pointer fixup. Should be up to 8x faster now. Change 3826734 by Ben.Marsh Remove code to disable TextureFormatAndroid on Linux. It seems to be an editor dependency. Change 3827730 by Steve.Robb Try to avoid decltype(auto) if it's not supported. See: https://udn.unrealengine.com/questions/395644/build-417-with-c11-on-linux-ttuple-errors.html Change 3827745 by Steve.Robb Initializer list support for TMap. Change 3827770 by Steve.Robb GitHub #4399 : Added a CONSTEXPR qualifiers to FVariant::GetType() #jira UE-53813 Change 3829189 by Ben.Marsh UBT: Now always writes a minimal log file. By default, just contains the regular console output and any reasons why actions are outdated and needed to be executed. UAT directs child UBT instances to output logs into its own log folder, so that build machines can save them off. Change 3830444 by Steve.Robb BuildVersion and ModuleManifest moved to Core, and parsing of these files reimplemented to avoid a JSON library. This should be revisited when Core has its own JSON library. Change 3830718 by Ben.Marsh Fix incorrect group name being returned by FStatNameAndInfo::GetGroupName() for stat groups. The editor populates the viewport stats list by calling this for every registered stat and stat group (via FLevelViewportCommands::HandleNewStatGroup). The menu entry attempts to show the stat name with STAT_XXX stripped from the start as the menu item label, with the free-form text description as a tooltip. For stat groups, the it would previously just return the stat group name as "Groups" (due to the raw naming convention of "//Groups//STATGROUP_Foo//..."). Since this didn't match the expected naming convention in FLevelViewportCommands::HandleNewStat (ie. STAT_XXX or STATGROUP_XXX), it would fail to add it. When the first actual stat belonging to that group is added, it would add a menu entry for the group based on that, but the stat description no longer makes sense as a tooltip for the group. As a result, all the editor tooltips were junk. #jira UE-53845 Change 3831064 by Ben.Marsh Fix log file contention when spawning UBT recursively. Change 3832654 by Ben.Marsh UGS: Fix error panel not being selected when opened, and weird alignment/color issues on it. Change 3832680 by Ben.Marsh UGS: Fix failing to detect workspace if synced to a different stream. Seems to be a regression caused by recent P4D upgrade. Change 3832695 by Ben.Marsh UGS: Invert the options in the 'Show Changes' submenu for simplicity. Change 3833528 by Ben.Marsh UAT: Script to rewrite source files with public include paths relative to the 'Public' folder. Usage is: RebasePublicIncludePaths -UpdateDir=<Dir> [-Project=<Dir>] [-Write]. Change 3833543 by Ben.Marsh UBT: Allow targets to opt-out of having public include paths added for every dependent module. This reduces the command line length when building a target, which has recently become a problem with larger games (due to Microsoft's compiler embedding the command line into each object file, with a maximum length of 64kb). All engine modules are compiled with this enabled; games may opt into it by setting bLegacyPublicIncludePaths = false; from their .target.cs, as may individual modules. Change 3834354 by Robert.Manuszewski Archetype pointer will now be cached to avoid locking the object tables when acquiring its info. It should also be faster this way regardless of any locks. #jira UE-52035 Change 3834400 by Robert.Manuszewski Fixing crash on exit caused by cached archetypes not being cleaned up before static exit cleanup. #jira UE-52035 Change 3834947 by Steve.Robb USE_FORMAT_STRING_TYPE_CHECKING removed from FMsg::Logf and FMsg::Logf_Internal. Change 3835004 by Ben.Marsh Fix code that relies on dubious behavior of requiring referenced "include path only" modules having their _API macros set to be empty, even if the module is actually implemented in a separate DLL. Change 3835340 by Ben.Marsh Fix errors making installed build from directories with spaces in the name. Change 3835972 by Ben.Marsh UBT: Improved diagnostic message for targets which don't need a version file. Change 3836019 by Ben.Marsh UBT: Fix warnings caused by defining linkage macros for third party libraries. Change 3836269 by Ben.Marsh Fix message box larger than the screen height being created when a large number of modules are incompatible on startup. Change 3836543 by Ben.Marsh Enable SoundMod plugin on Linux, since it's already supported through the editor. Change 3836546 by Ben.Marsh PR #4412: fix type mismatch (Contributed by nakapon) Change 3836805 by Ben.Marsh Fix commandlet to compile marketplace plugins. Change 3836829 by Ben.Marsh UBT: Fix ability to precompile plugins from installed engine builds. Change 3837036 by Ben.Marsh UBT: Write the previous and new contents of intermediate files to the log if they change. Makes it easier to debug unexpected rebuilds. Change 3837037 by Ben.Marsh UBT: Fix engine modules having inconsistent definitions depending on whether modules are only referenced for their include paths vs being linked into a binary (due to different _API macro). Change 3837040 by Ben.Marsh UBT: Remove code that initializes members in ModuleRules and TargetRules objects before the constructor is run. This is no longer necessary, now that the backwards-compatible default constructors have been removed. Change 3837247 by Ben.Marsh UBT: Remove UELinkerFixups module, now that plugins and precompiled modules do not require hacks to force initialization (since they're linked in as object files). Encryption and signing keys are now set via macros expanded from the IMPLEMENT_PRIMARY_GAME_MODULE macro, via project-specific macros added in the TargetRules constructor. Change 3837262 by Ben.Marsh UBT: Set whether a module is an engine module or not via a default value for the rules assembly. All non-program engine and enterprise modules are created with this flag set to true; program targets and modules are now created from a different assembly that sets it to false. This removes hacks from UEBuildModule needed to adjust behavior for different module types based on the directory containing the module. Also add a bUseBackwardsCompatibleDefaults flag to the TargetRules class, also initialized to a default value from a setting passed to the RulesAssembly constructor. This controls whether modules created for the target should be configured to allow breaking changes to default settings, and is set to false for all engine targets, and true for all project targets. Change 3837343 by Ben.Marsh UBT: Remove the OverrideExecutableFileExtension target property. Change the only current use for this (the MayaLiveLinkPlugin target) to use a post build step to copy the file instead. Change 3837356 by Ben.Marsh Fix invalid character encodings. Change 3837727 by Graeme.Thornton UnrealPak: KeyGenerator: Only generate prime table when required, not all the time Change 3837823 by Ben.Marsh UBT: Output warnings and errors when compiling module rules assembly in a way that allows them to be double-clicked in the Visual Studio output window. Change 3837831 by Graeme.Thornton UBT: When parsing crypto settings, always load legacy data first, then allow the new system to override it. Provides the same key backwards compatibility that the editor settings class gives Change 3837857 by Robert.Manuszewski PR #4404: Make FGCArrayPool singleton global instead of per-CU (Contributed by mhutch) Change 3837943 by Robert.Manuszewski PR #4405: Fix FGarbageCollectionTracer (Contributed by mhutch) Change 3838451 by Ben.Marsh UBT: Fix exceptions thrown on a background thread while caching C++ includes not being caught and logged correctly. Now captures exceptions and re-throws on the main thread. #jira UE-53996 Change 3839519 by Ben.Marsh UBT: Simplify configuring bPrecompile and bUsePrecompile settings for modules. Each rules assembly can now be configured as installed, which defaults the module rules it creates to use precompiled data. Change 3843790 by Graeme.Thornton UnrealPak: Log the size of all encrypted data Change 3844258 by Ben.Marsh Fix plugin compile failure when created via new plugin wizard. Passing -plugin on the command line is unnecessary, and is now reserved for packaging external plugins for the marketplace. Also extend the length of time that the error toast stays visible, and don't delete the plugin on failure. #jira UE-54157 Change 3845796 by Ben.Marsh Workaround for slow performance of String.EndsWith() on Mono. Change 3845823 by Ben.Marsh Fix case sensitive matching of platform names in -TargetPlatform=X argument to BuildCookRun. #jira UE-54123 Change 3845901 by Arciel.Rekman Linux: fix crash due to lambda lifetime issues (UE-54040). - The lambda goes out of scope in FBufferVisualizationMenuCommands::CreateVisualizationCommands, crashing the editor if compiled with a recent clang (5.0+). (Edigrating 3819174 to Dev-Core) Change 3846439 by Ben.Marsh Revert CL 3822742 to always call Process.WaitForExit(). The Android target platform module in the editor spawns ADB.EXE, which inherits the editor's stdout/stderr handles and forks itself. Process.WaitForExit() waits for EOF on those pipes, which never occurs because the forked process never terminates. Proper fix is probably to have the engine explicitly duplicate stdout/stderr handles for new pipes to output process, but too risky before copying up to Main. Change 3816608 by Ben.Marsh UBT: Use DirectoryReference objects for all include paths. Change 3816954 by Ben.Marsh UBT: Remove bIncludeDependentLibrariesInLibrary option. This is not widely supported by platform toolchains, and is not used anywhere. Change 3816986 by Ben.Marsh UBT: Remove UEBuildBinaryConfig; UEBuildBinary objects are now just created directly. Change 3816991 by Ben.Marsh UBT: Deprecate PlatformSpecificDynamicallyLoadedModules. We no longer have any special behavior for these modules. Change 3823090 by Ben.Marsh UAT: Improve logging for child UAT instances. - Calling RunUAT now requires an identifier for prefixing into the parent log, which is also used to determine the name of the log folder. - Stdout is no longer written to its own output file, since it's written to the parent stdout, the parent log file, and the child log file anyway. - Log folders for child UAT instances are left intact, rather than being copied to the parent folder. The derived names for the copied names were confusing and hard to read. - Output from UAT is no longer returned as a string. It should not be parsed anyway (but may be huge!). ProcessResult now supports running without capturing output. Change 3826082 by Ben.Marsh UBT: Add a check to make sure that all modules that are precompiled are correctly marked to enable it, even if they are part of the build target. Change 3827025 by Ben.Marsh UBT: Move the compile output directory into a property on the module, and explicitly pass it to the toolchain when compiling. Change 3829927 by James.Hopkin Made HTTP interface const correct Change 3833533 by Ben.Marsh Rewrite engine source files to base include paths relative to the "Public" directory. This allows reducing the number of public include paths that have to be added for engine modules. Change 3835826 by Ben.Marsh UBT: Precompiled targets now generate a separate manifest for each precompiled module, rather than adding object files to a library. This fixes issues where object files from static libraries would not be linked into a target if a symbol in them was not referenced. Change 3835969 by Ben.Marsh UBT: Fix cases where text is being written directly to the console rather than via logging functions. Change 3837777 by Steve.Robb Format string type checking added to FOutputDevice::Logf. Fixes for those. Change 3838569 by Steve.Robb Algo moved up a folder. [CL 3847482 by Ben Marsh in Main branch]
2018-01-20 11:19:29 -05:00
#include "Modules/ModuleInterface.h"
#include "Modules/ModuleManager.h"
Copying //UE4/Dev-Build to //UE4/Dev-Main (Source: //UE4/Dev-Build @ 3209340) #lockdown Nick.Penwarden #rb none ========================== MAJOR FEATURES + CHANGES ========================== Change 3209340 on 2016/11/23 by Ben.Marsh Convert UE4 codebase to an "include what you use" model - where every header just includes the dependencies it needs, rather than every source file including large monolithic headers like Engine.h and UnrealEd.h. Measured full rebuild times around 2x faster using XGE on Windows, and improvements of 25% or more for incremental builds and full rebuilds on most other platforms. * Every header now includes everything it needs to compile. * There's a CoreMinimal.h header that gets you a set of ubiquitous types from Core (eg. FString, FName, TArray, FVector, etc...). Most headers now include this first. * There's a CoreTypes.h header that sets up primitive UE4 types and build macros (int32, PLATFORM_WIN64, etc...). All headers in Core include this first, as does CoreMinimal.h. * Every .cpp file includes its matching .h file first. * This helps validate that each header is including everything it needs to compile. * No engine code includes a monolithic header such as Engine.h or UnrealEd.h any more. * You will get a warning if you try to include one of these from the engine. They still exist for compatibility with game projects and do not produce warnings when included there. * There have only been minor changes to our internal games down to accommodate these changes. The intent is for this to be as seamless as possible. * No engine code explicitly includes a precompiled header any more. * We still use PCHs, but they're force-included on the compiler command line by UnrealBuildTool instead. This lets us tune what they contain without breaking any existing include dependencies. * PCHs are generated by a tool to get a statistical amount of coverage for the source files using it, and I've seeded the new shared PCHs to contain any header included by > 15% of source files. Tool used to generate this transform is at Engine\Source\Programs\IncludeTool. [CL 3209342 by Ben Marsh in Main branch]
2016-11-23 15:48:37 -05:00
#include "Interfaces/IShaderFormat.h"
#include "Interfaces/IShaderFormatModule.h"
#include "ShaderCore.h"
#include "ShaderCodeArchive.h"
#include "ShaderPreprocessTypes.h"
#include "hlslcc.h"
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
#include "MetalShaderResources.h"
#include "HAL/FileManager.h"
#include "HAL/PlatformFileManager.h"
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
#include "Serialization/Archive.h"
Copying //UE4/Dev-Core to //UE4/Dev-Main (Source: //UE4/Dev-Core @ 3847469) #lockdown Nick.Penwarden #rb none ============================ MAJOR FEATURES & CHANGES ============================ Change 3805828 by Gil.Gribb UE4 - Fixed a bug in the lock free stalling task queue and adjusted a comment. The code is not current used, so this is not actually change the way the code works. Change 3806784 by Ben.Marsh UAT: Remove code to compile UBT when using UE4Build. It should already be compiled as a dependency of UAT. Change 3807549 by Graeme.Thornton Add a cook timer around VerifyCanCookPackage. A licensee reports this taking a lot of time so it'll be good to account for it. Change 3807727 by Graeme.Thornton Unhide the text asset format experimental editor option Change 3807746 by Josh.Engebretson Remove WER from iOS platform Change 3807928 by Robert.Manuszewski When async loading, GC Clusters will be created after packages have been processed to avoid situations where some of the objects that are being added to a cluster haven't been fully loaded yet Change 3808221 by Steve.Robb GitHub #4307 - Made GetModulePtr() thread safe by not using GetModule() ^ I'm not convinced by how much thread-safer this is really, but it's tidier anyway. Change 3809233 by Graeme.Thornton TBA: Misc changes to text asset commandlet - Rename mode to "loadsave" - Add -outputFormat option which can be assigned "text" or "binary" - When saving binary, use a differentiated filename so that source assets aren't overwritten Change 3809518 by Ben.Marsh Remove the outdated UnrealSync automation script. Change 3809643 by Steve.Robb GitHub #4277 : fix bug; FMath::FormatIntToHumanReadable 3rd comma and negative value #jira UE-53037 Change 3809862 by Steve.Robb GitHub #3342 : [FRotator.h] Fix to DecompressAxisFromByte to be more efficient and reflect its intent accurately #jira UE-42593 Change 3811190 by Graeme.Thornton Add support for writing specific log channels to their own files Change 3811197 by Graeme.Thornton Minor updates to output formatting and timing for the text asset commandlet Change 3811257 by Robert.Manuszewski Cluster creation will now be time-sliced Change 3811565 by Steve.Robb Define out non-monolithic module functions. Change 3812561 by Steve.Robb GitHub #3886 : Enable Brace-Initialization for Declaring Variables Incorrect semi-colon search removed after discussion with author. Test added. #jira UE-48242 Change 3812864 by Steve.Robb Removal of some unproven code which was supposed to fix hot reloading BP class functions in plugins. See: https://udn.unrealengine.com/questions/376978/aitask-blueprint-nodes-disappear-when-their-module.html #jira UE-53089 Change 3820358 by Ben.Marsh PR #4358: Incredibuild use ShowAgent by default (Contributed by projectgheist) Change 3822594 by Ben.Marsh UAT: Improvements to log file handling. - Always create log files in the final location, rather than writing to a temp directory and copying in later. - Now supports -Verbose and -VeryVerbose for increasing log verbosity, rather than -Verbose=XXX. - Keep a backlog of log output before the log system is initialized, and flush it to the log file once it is. - Allow buildmachines to specify the uebp_FinalLogFolder environment variable, which is used to form paths for display. When build machines copy log files elsewhere after UAT finishes (eg. a network share), this allows error messages to display the right location. Change 3823695 by Ben.Marsh UGS: Fix issue where precompiled binaries would not be shown as available for a change until scrolling the last submitted code change into the buffer (other symptoms, like de-focussing the main window would cause it to go back to an unavailable state, since the changes buffer was shrunk). Now always queries changes up to the last change for which zipped binaries are available. Change 3823845 by Ben.Marsh UBT: Exclude C# projects for unsupported platforms when generating project files. Change 3824180 by Ben.Marsh UGS: Add an option to show changes by build machines, and move the "only show reviewed" option in there too (Options > Show Changes). #jira Change 3825777 by Steve.Robb Fix to return value of StringToBytes. Change 3825810 by Ben.Marsh UBT: Reduce length of include paths for MSVC toolchain. Change 3825822 by Robert.Manuszewski Optimized PIE lazy pointer fixup. Should be up to 8x faster now. Change 3826734 by Ben.Marsh Remove code to disable TextureFormatAndroid on Linux. It seems to be an editor dependency. Change 3827730 by Steve.Robb Try to avoid decltype(auto) if it's not supported. See: https://udn.unrealengine.com/questions/395644/build-417-with-c11-on-linux-ttuple-errors.html Change 3827745 by Steve.Robb Initializer list support for TMap. Change 3827770 by Steve.Robb GitHub #4399 : Added a CONSTEXPR qualifiers to FVariant::GetType() #jira UE-53813 Change 3829189 by Ben.Marsh UBT: Now always writes a minimal log file. By default, just contains the regular console output and any reasons why actions are outdated and needed to be executed. UAT directs child UBT instances to output logs into its own log folder, so that build machines can save them off. Change 3830444 by Steve.Robb BuildVersion and ModuleManifest moved to Core, and parsing of these files reimplemented to avoid a JSON library. This should be revisited when Core has its own JSON library. Change 3830718 by Ben.Marsh Fix incorrect group name being returned by FStatNameAndInfo::GetGroupName() for stat groups. The editor populates the viewport stats list by calling this for every registered stat and stat group (via FLevelViewportCommands::HandleNewStatGroup). The menu entry attempts to show the stat name with STAT_XXX stripped from the start as the menu item label, with the free-form text description as a tooltip. For stat groups, the it would previously just return the stat group name as "Groups" (due to the raw naming convention of "//Groups//STATGROUP_Foo//..."). Since this didn't match the expected naming convention in FLevelViewportCommands::HandleNewStat (ie. STAT_XXX or STATGROUP_XXX), it would fail to add it. When the first actual stat belonging to that group is added, it would add a menu entry for the group based on that, but the stat description no longer makes sense as a tooltip for the group. As a result, all the editor tooltips were junk. #jira UE-53845 Change 3831064 by Ben.Marsh Fix log file contention when spawning UBT recursively. Change 3832654 by Ben.Marsh UGS: Fix error panel not being selected when opened, and weird alignment/color issues on it. Change 3832680 by Ben.Marsh UGS: Fix failing to detect workspace if synced to a different stream. Seems to be a regression caused by recent P4D upgrade. Change 3832695 by Ben.Marsh UGS: Invert the options in the 'Show Changes' submenu for simplicity. Change 3833528 by Ben.Marsh UAT: Script to rewrite source files with public include paths relative to the 'Public' folder. Usage is: RebasePublicIncludePaths -UpdateDir=<Dir> [-Project=<Dir>] [-Write]. Change 3833543 by Ben.Marsh UBT: Allow targets to opt-out of having public include paths added for every dependent module. This reduces the command line length when building a target, which has recently become a problem with larger games (due to Microsoft's compiler embedding the command line into each object file, with a maximum length of 64kb). All engine modules are compiled with this enabled; games may opt into it by setting bLegacyPublicIncludePaths = false; from their .target.cs, as may individual modules. Change 3834354 by Robert.Manuszewski Archetype pointer will now be cached to avoid locking the object tables when acquiring its info. It should also be faster this way regardless of any locks. #jira UE-52035 Change 3834400 by Robert.Manuszewski Fixing crash on exit caused by cached archetypes not being cleaned up before static exit cleanup. #jira UE-52035 Change 3834947 by Steve.Robb USE_FORMAT_STRING_TYPE_CHECKING removed from FMsg::Logf and FMsg::Logf_Internal. Change 3835004 by Ben.Marsh Fix code that relies on dubious behavior of requiring referenced "include path only" modules having their _API macros set to be empty, even if the module is actually implemented in a separate DLL. Change 3835340 by Ben.Marsh Fix errors making installed build from directories with spaces in the name. Change 3835972 by Ben.Marsh UBT: Improved diagnostic message for targets which don't need a version file. Change 3836019 by Ben.Marsh UBT: Fix warnings caused by defining linkage macros for third party libraries. Change 3836269 by Ben.Marsh Fix message box larger than the screen height being created when a large number of modules are incompatible on startup. Change 3836543 by Ben.Marsh Enable SoundMod plugin on Linux, since it's already supported through the editor. Change 3836546 by Ben.Marsh PR #4412: fix type mismatch (Contributed by nakapon) Change 3836805 by Ben.Marsh Fix commandlet to compile marketplace plugins. Change 3836829 by Ben.Marsh UBT: Fix ability to precompile plugins from installed engine builds. Change 3837036 by Ben.Marsh UBT: Write the previous and new contents of intermediate files to the log if they change. Makes it easier to debug unexpected rebuilds. Change 3837037 by Ben.Marsh UBT: Fix engine modules having inconsistent definitions depending on whether modules are only referenced for their include paths vs being linked into a binary (due to different _API macro). Change 3837040 by Ben.Marsh UBT: Remove code that initializes members in ModuleRules and TargetRules objects before the constructor is run. This is no longer necessary, now that the backwards-compatible default constructors have been removed. Change 3837247 by Ben.Marsh UBT: Remove UELinkerFixups module, now that plugins and precompiled modules do not require hacks to force initialization (since they're linked in as object files). Encryption and signing keys are now set via macros expanded from the IMPLEMENT_PRIMARY_GAME_MODULE macro, via project-specific macros added in the TargetRules constructor. Change 3837262 by Ben.Marsh UBT: Set whether a module is an engine module or not via a default value for the rules assembly. All non-program engine and enterprise modules are created with this flag set to true; program targets and modules are now created from a different assembly that sets it to false. This removes hacks from UEBuildModule needed to adjust behavior for different module types based on the directory containing the module. Also add a bUseBackwardsCompatibleDefaults flag to the TargetRules class, also initialized to a default value from a setting passed to the RulesAssembly constructor. This controls whether modules created for the target should be configured to allow breaking changes to default settings, and is set to false for all engine targets, and true for all project targets. Change 3837343 by Ben.Marsh UBT: Remove the OverrideExecutableFileExtension target property. Change the only current use for this (the MayaLiveLinkPlugin target) to use a post build step to copy the file instead. Change 3837356 by Ben.Marsh Fix invalid character encodings. Change 3837727 by Graeme.Thornton UnrealPak: KeyGenerator: Only generate prime table when required, not all the time Change 3837823 by Ben.Marsh UBT: Output warnings and errors when compiling module rules assembly in a way that allows them to be double-clicked in the Visual Studio output window. Change 3837831 by Graeme.Thornton UBT: When parsing crypto settings, always load legacy data first, then allow the new system to override it. Provides the same key backwards compatibility that the editor settings class gives Change 3837857 by Robert.Manuszewski PR #4404: Make FGCArrayPool singleton global instead of per-CU (Contributed by mhutch) Change 3837943 by Robert.Manuszewski PR #4405: Fix FGarbageCollectionTracer (Contributed by mhutch) Change 3838451 by Ben.Marsh UBT: Fix exceptions thrown on a background thread while caching C++ includes not being caught and logged correctly. Now captures exceptions and re-throws on the main thread. #jira UE-53996 Change 3839519 by Ben.Marsh UBT: Simplify configuring bPrecompile and bUsePrecompile settings for modules. Each rules assembly can now be configured as installed, which defaults the module rules it creates to use precompiled data. Change 3843790 by Graeme.Thornton UnrealPak: Log the size of all encrypted data Change 3844258 by Ben.Marsh Fix plugin compile failure when created via new plugin wizard. Passing -plugin on the command line is unnecessary, and is now reserved for packaging external plugins for the marketplace. Also extend the length of time that the error toast stays visible, and don't delete the plugin on failure. #jira UE-54157 Change 3845796 by Ben.Marsh Workaround for slow performance of String.EndsWith() on Mono. Change 3845823 by Ben.Marsh Fix case sensitive matching of platform names in -TargetPlatform=X argument to BuildCookRun. #jira UE-54123 Change 3845901 by Arciel.Rekman Linux: fix crash due to lambda lifetime issues (UE-54040). - The lambda goes out of scope in FBufferVisualizationMenuCommands::CreateVisualizationCommands, crashing the editor if compiled with a recent clang (5.0+). (Edigrating 3819174 to Dev-Core) Change 3846439 by Ben.Marsh Revert CL 3822742 to always call Process.WaitForExit(). The Android target platform module in the editor spawns ADB.EXE, which inherits the editor's stdout/stderr handles and forks itself. Process.WaitForExit() waits for EOF on those pipes, which never occurs because the forked process never terminates. Proper fix is probably to have the engine explicitly duplicate stdout/stderr handles for new pipes to output process, but too risky before copying up to Main. Change 3816608 by Ben.Marsh UBT: Use DirectoryReference objects for all include paths. Change 3816954 by Ben.Marsh UBT: Remove bIncludeDependentLibrariesInLibrary option. This is not widely supported by platform toolchains, and is not used anywhere. Change 3816986 by Ben.Marsh UBT: Remove UEBuildBinaryConfig; UEBuildBinary objects are now just created directly. Change 3816991 by Ben.Marsh UBT: Deprecate PlatformSpecificDynamicallyLoadedModules. We no longer have any special behavior for these modules. Change 3823090 by Ben.Marsh UAT: Improve logging for child UAT instances. - Calling RunUAT now requires an identifier for prefixing into the parent log, which is also used to determine the name of the log folder. - Stdout is no longer written to its own output file, since it's written to the parent stdout, the parent log file, and the child log file anyway. - Log folders for child UAT instances are left intact, rather than being copied to the parent folder. The derived names for the copied names were confusing and hard to read. - Output from UAT is no longer returned as a string. It should not be parsed anyway (but may be huge!). ProcessResult now supports running without capturing output. Change 3826082 by Ben.Marsh UBT: Add a check to make sure that all modules that are precompiled are correctly marked to enable it, even if they are part of the build target. Change 3827025 by Ben.Marsh UBT: Move the compile output directory into a property on the module, and explicitly pass it to the toolchain when compiling. Change 3829927 by James.Hopkin Made HTTP interface const correct Change 3833533 by Ben.Marsh Rewrite engine source files to base include paths relative to the "Public" directory. This allows reducing the number of public include paths that have to be added for engine modules. Change 3835826 by Ben.Marsh UBT: Precompiled targets now generate a separate manifest for each precompiled module, rather than adding object files to a library. This fixes issues where object files from static libraries would not be linked into a target if a symbol in them was not referenced. Change 3835969 by Ben.Marsh UBT: Fix cases where text is being written directly to the console rather than via logging functions. Change 3837777 by Steve.Robb Format string type checking added to FOutputDevice::Logf. Fixes for those. Change 3838569 by Steve.Robb Algo moved up a folder. [CL 3847482 by Ben Marsh in Main branch]
2018-01-20 11:19:29 -05:00
#include "Misc/ConfigCacheIni.h"
#include "MetalBackend.h"
#include "Misc/FileHelper.h"
#include "Misc/Paths.h"
#include "FileUtilities/ZipArchiveWriter.h"
#include "MetalShaderCompiler.h"
#include "DataDrivenShaderPlatformInfo.h"
DEFINE_LOG_CATEGORY(LogMetalCompilerSetup)
DEFINE_LOG_CATEGORY(LogMetalShaderCompiler)
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
#define WRITE_METAL_SHADER_SOURCE_ARCHIVE 0
// Set this define to get additional logging information about Metal toolchain setup.
#define CHECK_METAL_COMPILER_TOOLCHAIN_SETUP 0
extern bool PreprocessMetalShader(const FShaderCompilerInput& Input, const FShaderCompilerEnvironment& Environment, FShaderPreprocessOutput& PreprocessOutput);
extern void CompileMetalShader(const FShaderCompilerInput& Input, const FShaderPreprocessOutput& InPreprocessOutput, FShaderCompilerOutput& Output);
extern void OutputMetalDebugData(const FShaderCompilerInput& Input, const FShaderPreprocessOutput& PreprocessOutput, const FShaderCompilerOutput& Output);
extern bool StripShader_Metal(TArray<uint8>& Code, class FString const& DebugPath, bool const bNative);
extern uint64 AppendShader_Metal(class FString const& ArchivePath, const FSHAHash& Hash, TArray<uint8>& Code);
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
extern bool FinalizeLibrary_Metal(class FName const& Format, class FString const& ArchivePath, class FString const& LibraryPath, TSet<uint64> const& Shaders, class FString const& DebugOutputDir);
/** Version for shader format, this becomes part of the DDC key. */
static const FGuid UE_SHADER_METAL_VER = FGuid("282D7BF5-C5F8-47B7-B065-E165D60FC17B");
class FMetalShaderFormat : public IShaderFormat
{
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3607928) #lockdown Nick.Penwarden ============================ MAJOR FEATURES & CHANGES ============================ Change 3441680 by Uriel.Doyon Added units to point light intensity, to allow the user to specify the value in candelas or lumens. New point light actors now configure the intensity in candelas by default. Replaced viewport exposure settings by an EV100 slider. Hidding the tone mapper in the show flag now still applies the exposure. Added a new AutoExposure method called EV100 which allows to specify : - MinEV100, MaxEV100 - Calibration Constnat - Exposure Compensation #jira UE-42783 Change 3454934 by Chris.Bunner Backing out changelists 3441680, 3454636 and 3454844 for the sake of integration stability. Change 3512118 by Marc.Olano Fix rare Sobol shader data problem. Mismatch with CPU code after a large number of points Resubmit of portion of //UE4/Dev-Rendering@3509854 that was rolled back to avoid massive shader recompiles during integration testing Change 3512129 by Benjamin.Hyder Fixing up content in TM-SobolNoise Change 3512151 by Rolando.Caloca DR - Fixed some layouts that were general - Added some extra dump information Change 3512160 by Benjamin.Hyder Still Fixing TM-Sobol Change 3512180 by Marc.Olano PCSS for spotlights. Like directional PCSS this is experimental, activated by r.Shadow.FilterMethod. Change 3512261 by Michael.Lentine Move Subsurface to shared properties. Previously the same code could be executed multiple times without being optimized out if multiple inputs used the same subsurface output. #jira UE-44405 Change 3512288 by Rolando.Caloca DR - Fix issue when recycling image handles Change 3512338 by Michael.Lentine Fix precision if user enters a multiple of 90 degree rotation for transforms. This will only work for exact values. Generally comparing float point numbers using == is unsafe but it should be ok in this case as they are exact values entered from the UI. We may want to later expand this to include thresholding using a value ~1e-7. #jira UE-46137 Change 3512424 by Michael.Lentine Regenerate BaseColor.uasset and Specular.uasset to not have the notforclient flags set. #jira UE-44315 Change 3512686 by Brian.Karis Fix for quadric assert in infiltrator. Due to bad tangents in source mesh. Change 3512696 by Brian.Karis Unrevert TAA. Fixed DOF NaN artifacts Change 3512717 by Marcus.Wassmer PR #3714: Fix typo in EOcclusionCombineMode (Contributed by Mumbles4) Change 3513112 by Richard.Wallis Crash when packaging for iOS with Shared Material Native Libraries and Share Material Shader Code from windows platform. Offline shader compile for archiving not done - shader header has missing offline compile flag for native Metal library archiving. Fix includes: - Handle offline compile failure when not running on Mac and no remote is configured (or remote fails). (I think it's this point at which the crash in the bug report is at). - Make sure remote can build for native Metal libraries and archive correctly - this should now support Linux platforms or Mac to Mac (if enabled in MetalShaderCompiler.cpp) for testing if required. - Updated to include remote calling into the xcode 9 Metal pch fix already submitted by Mark Satt. #jira UE-45657 Change 3513357 by Richard.Wallis Windows compile fix. Change 3513375 by Guillaume.Abadie Exposes the possibility to manually destroy the GPU ressource of UTextureRenderTarget2D. Change 3513685 by Richard.Hinckley #jira UEDOC-3822 Fixing a comment that refers to a non-existent function, for documentation purposes. Change 3513705 by Marc.Olano Updates to Sobol test levels in RenderTest project Change 3513730 by Rolando.Caloca DR - Fix mip size copying resolve targets - Fix compute fence - Fix descriptor set texture layout - More dump info Change 3513742 by Marc.Olano Texture-free numeric print for shader debugging Change 3513777 by Daniel.Wright Handled edge case where no furthest samples are found in precomputed visibility Change 3514852 by Rolando.Caloca DR - Fix -directcompile on SCW Change 3515049 by Rolando.Caloca DR - hlslcc dump crash fix Change 3515167 by Rolando.Caloca DR - hlslcc - Fix bogus string pointer - Allow reading from non-scalar UAVs Change 3515745 by Rolando.Caloca DR - Linux compile fix Change 3515862 by Rolando.Caloca DR - Remove old reference to CCT - Link with hlslcc debug libs on SCW debug config for easier debugging Change 3516292 by Rolando.Caloca DR - glslang exe fixes Change 3516568 by Rolando.Caloca DR - hlslcc - Copy fix for *Buffer as functionparameters Change 3516659 by Marcus.Wassmer Fix some d3derrors with distance fields Change 3516801 by Daniel.Wright Fixed crash when doing editor 'Force Delete' on a static mesh whose distance field is still being built. Any UObject reference that is to an asset can be NULL'ed by the editor. Change 3516825 by Rolando.Caloca DR - Some initial fixes for structured buffers Change 3516843 by Rolando.Caloca DR - Fix for Vulkan dist fields Change 3516869 by Marcus.Wassmer Add format to the createrendertarget blueprint node Change 3516957 by Daniel.Wright Fixed bUsesDistortion being editable Change 3516965 by Daniel.Wright Still mark the distance field task completed, even if the static mesh has been deleted Change 3517039 by Yujiang.Wang GitHub #2655: Optimization for shadow map resolution selection for spot lights * Use the radius of the inscribed sphere at the cone end as the spot light's screen radius Note: slight drop of shadow quality of spot lights may occur when they are far away from the camera. This is intended, since before this optimization they tend to be always rendered with the maximum shadow map resolution (2048), which is very costly #jira UE-33982 Change 3517069 by Yujiang.Wang Fix for ScissorRect settings in d3d11 being lost under certain scenarios * Scissor rectangle is always enabled in the low-level d3d11 pipeline, and it is expected that at least one ScissorRect is present no matter whether RHISetScissorRect is called with bEnable=false (when it is false we just use a big rect to make it effectively disabled) * However FD3D11StateCacheBase::ClearState() clears all the states, which removes scissor rectangles and causes problems for certain routines (FScene::UpdateSkyCaptureContents) * Now SetScissorRectIfRequiredWhenSettingViewport will always set a effectively disabled ScissorRect on each FD3D11DynamicRHI::RHISetViewport call, just like d3d12 does #jira UE-45465 UE-44760 Change 3517134 by Yujiang.Wang CIS fix Change 3517662 by Rolando.Caloca DR - Execute upload Vulkan cmds on the RHI thread - Fix crash with structured buffer Change 3517677 by Rolando.Caloca DR - Update/copy textures on RHI thread Change 3517680 by Rolando.Caloca DR - Copy texture bulk data on rhi thread Change 3517748 by Marcus.Wassmer temporary workaround for one class of GPU crashes Change 3518832 by Rolando.Caloca DR - Copy & extend 3518077 - Fix for movable skylight shader missing on simple forward (low lighting quality mode) Change 3519973 by Richard.Wallis Jittering in Engine Menu Dropdown Options. Jitter fix: Fix some areas that hadn't been changed from RoundToInt (from previous CL's) to CeilToInt. #jira UE-46505 Change 3520849 by Uriel.Doyon Fixed issue with investigate texture command and dynamic component entries. Change 3521064 by Guillaume.Abadie Returns absolute path of shader files on error to avoid work loss in visual studio that can't figure out that a sln relative and absolute path might leading to same file on disk. Change 3521834 by Rolando.Caloca DR - Fix decals on Vulkan Change 3521892 by Rolando.Caloca DR - Fix Vulkan texture streaming Change 3523181 by Rolando.Caloca DR - Copy from 3523176 UE4.17 - Fix Vulkan scissor causing text to not clip Change 3523534 by Yujiang.Wang UE-46631: Implement a scalable LongGPUTask to fix ProfileGPU * A new, scalable, platform-independent IssueLongGPUTask is now implemented in UtilityShaders * Removed IssueLongGPUTask and G*Vector4VertexDeclaration from RHI implementations * The measurement of the execution time of a basic LongGPUTask unit is kicked off on the very first frame #jira UE-46631 Change 3524552 by Yujiang.Wang Fix iteration number calculation of LongGPUTask Change 3524975 by Joe.Graf Moved the Hamming-weight function from StaticMeshDrawList.inl to FGenericPlatformMath Added SSE versions using _mm_popcnt_u64 for platforms that support it Added a SSE check to gracefully exit when missing the instruction and it was expected to be there #CodeReview: arciel.rekman, brian.karis Change 3525306 by Daniel.Wright Fixed ensure from LPV Change 3525346 by Rolando.Caloca DR - Fix linking issue Change 3525459 by Daniel.Wright Volumetric Lightmaps - higher quality precomputed GI on dynamic objects and GI on Volumetric Fog * Enabled by default on all maps, effective after a lighting build. This replaces the existing Precomputed Light Volume and Indirect Lighting Cache features. * New Lightmass World Settings: VolumeLightingMethod, VolumetricLightmapDetailCellSize and VolumetricLightmapMaximumBrickMemoryMb. * Lightmass computes lighting samples in an adaptive grid, with higher density around geometry inside the importance volume. Positions outside the importance volume get lit with the border texels. * Improved Lightmass volume solver to use importance photons and full adaptive final gather, so volume samples have similar quality to 2d lightmaps. * A static indirection texture is built covering the importance volume and flattening the brick tree by storing the offset to the highest density brick at each indirection cell. * Seamless and efficient GPU interpolation across density levels is achieved by adding a single row of padding to bricks, copied from neighbors, and stitching up bricks with lower density neighbors * The Volumetric lightmap stores Irradiance as a 3 band SH, which is 27 floats, quantized into 28 bytes, 7 texture lookups. * A full screen barebones material using Volumetric Lightmaps costs .42ms on 970 GTX, while Indirect Lighting Cache Point costs .32ms * Sky bent normal is also stored for stationary skylights and Directional Light Shadowing for Single Sample Shadow receiving. * Volumetric fog, Movable components, unbuilt Static Components, SingleSampleShadow receiving and Capsule Shadows use Volumetric Lightmaps if available * New Visualization show flag for Volumetric Lightmap sample points * Level streaming of volume light data is not currently supported with this method Change 3525461 by Daniel.Wright Lowered default r.Shadow.RadiusThreshold for Epic shadow settings as it was causing a lot of visible artifacts from small objects popping out. This will increase shadowmap cost slightly (13.5ms RT -> 14.3ms RT in Fortnite on PS4, no measurable GPU difference). Change 3526459 by Rolando.Caloca DR - Fix validation error Change 3526474 by Rolando.Caloca DR - Integrate from GV Change 3526487 by Daniel.Wright Disabled Volumetric Lightmap filtering with neighbors due to artifacts Fix linux compile errors Change 3526833 by Rolando.Caloca DR - Workaround for hlslcc Change 3526991 by Uriel.Doyon Integrated 3526859 : Texture mip bias is now reset whenever the streaming budget increases. This fixes an issue where textures persistently become low res after a memory spike. Change 3527574 by Rolando.Caloca DR - Added some missing resource entries for SCW direct mode Change 3527625 by Rolando.Caloca DR - Copy from 3527113 UE4.17 - Fix Vulkan not calling Present Change 3528461 by Brian.Karis Support larger hash sizes. Added uint list hashing function. Change 3528780 by Rolando.Caloca DR - Default Vulkan resources Change 3528818 by Rolando.Caloca DR - glslang - Added missing accessor Change 3528839 by Rolando.Caloca DR - Fix virtual path issue when using non-engine relative absolute paths Change 3528900 by Daniel.Wright Fixed variable shadowing Change 3529039 by Rolando.Caloca DR - Read Spirv reflection data (not used yet) Change 3529040 by Joe.Graf Fixed the 32bit compile failures for the popcnt optimization #CodeReview: arciel.rekman Change 3529060 by Rolando.Caloca DR - hlslcc - New flag for keeping resource names Change 3529344 by Rolando.Caloca DR - Delete unused file Change 3529723 by Brian.Karis Fixed static analysis cleaner. Change 3531357 by Michael.Trepka Updated Mac glslang libraries with latest changes. Also, updated the Xcode project (generated with CMake) and moved it to a different location so that it no longer uses hardcoded absolute paths. It should be easy to rebuild these libraries in the future. Change 3531517 by Joe.Graf Added support for ddx_fine, ddy_fine, ddx_coarse, ddy_coarse to hlslcc #CodeReview: arciel.rekman, mark.satterthwaite, rolando.caloca Change 3531626 by Joe.Graf Mac version of the popcount optimization Changed Linux version to use the same builtin #CodeReview: mark.satterthwaite, arciel.rekman Change 3531837 by Chris.Bunner SetScissorRectIfRequiredWhenSettingViewport sets the viewport size by default rather than disabling the scissor rect. #jra UE-46753 Change 3533415 by Joe.Graf Renamed the SSSE3 checks per feedback #CodeReview: arciel.rekman Change 3533480 by Michael.Lentine Use more accurate descriptions for shader recompile options Change 3533511 by Joe.Graf Updated the GenericPlatformMisc to match the SSSE3 name change #CodeReview: arciel.rekman Change 3533521 by Marcus.Wassmer Fix scenerenderer leak when updating out of view planar reflections Change 3533528 by Joe.Graf Updated comments #CodeReview: n/a Change 3533608 by Mark.Satterthwaite New manual Xcode project for glslang so that we include all the necessary code and can link again. Change 3534260 by Mark.Satterthwaite Fix the Xcode 9 Beta 3 compile errors in MetalRHI without breaking Xcode 8.3.3. Change 3535789 by Yujiang.Wang Fix for wrong hair shading in forward shading * IBL reflections should be turned off for hairs Change 3537059 by Ben.Marsh Fixing case of iOS directories, pt1 Change 3537060 by Ben.Marsh Fixing case of iOS directories, pt2 Change 3538297 by Michael.Lentine Add shader comparison test. Adding the basic test case. Adding logic to Common.ush to enable FP16 conditionally on a define (which is not set by default) Adding more exported functionality to automation for use in the shader test. Change 3538309 by Michael.Lentine Add missing file from Shader Test CL. Change 3538751 by Michael.Lentine Add missing pragma once. Change 3539236 by Michael.Lentine Do not ignore return values. Change 3539237 by Michael.Lentine Check in the correct file Change 3540343 by Rolando.Caloca DR - Added t.DumpHitches.AllThreads Change 3540661 by Yujiang.Wang Fix spot tube light direction * The tube direction for a spot light was pointing along the light direction, now it is along the local Z axis which is perpendicular to the light direction. Lightmass is also touched * A new LightTangent is added to FDeferredLightData * Packed all the values from LightSceneProxy->GetParameters into a single FLightParameters struct to avoid copy-pasting them everywhere Change 3541129 by Rolando.Caloca DR - vk - Copy all Vulkan fixes from 4.17 Change 3541347 by Yujiang.Wang Fix wrong ViewFlags being set between objects when rendering shadow depth maps * Bug caused by trying to share DrawRenderState between objects, but SetViewFlagsForShadowPass was designed to start from a fresh render state * Now SetViewFlagsForShadowPass recalculates and sets the flags on each call Change 3542603 by Rolando.Caloca DR - vk - Allow sharing samplers on Vulkan Change 3542639 by Jian.Ru Changed warning text to better indicate that global clip plane needs to be enabled for planar reflection #RB Marcus.Wassmer Change 3543167 by Michael.Lentine Fix naming for the shader comparison tests. Change 3543210 by Uriel.Doyon Fixed an issue when computing material scales where the default material ends up being used instead of the required material. In that case, we used the default settings for texture streaming (assuming a scale of 1). Change 3543221 by Brian.Karis Simplifier optimizations Change 3543239 by Arciel.Rekman hlslcc: remove FCustomStd* workarounds. - This was previous attempt to work around problems arising from different STL used for building libhlslcc (in the cross-toolchain) and possibly different STL used for building engine (on the system). - The same problem has been resolved by bundling libc++. Change 3543946 by Michael.Lentine Add comparison output. Change 3544277 by Brian.Karis Fixed uninitialized var error Change 3544404 by Rolando.Caloca DR - Fix broken textures Change 3544503 by Jian.Ru Ensure lighting failure delegates are always called #RB Marcus.Wassmer,Daniel.Wright #3689 Change 3545241 by Daniel.Wright Fixed spotlight whole scene shadows using a radius 2x too long Change 3545347 by Daniel.Wright Fixed shadow occlusion culling broken by shadowmap caching change. FProjectedShadowKey is now computed correctly for whole scene shadows and SDCM_StaticPrimitivesOnly shadowmaps will fall back to the query for a SDCM_MovablePrimitivesOnly, since the static primitives shadowmap's query is not issued every frame. Change 3546196 by Marcus.Wassmer Fix minor typo Change 3546459 by Daniel.Wright ULevel::PostEditChangeProperty recreates rendering resources if MapBuildData is modified - fixes a crash when Force Deleting the MapBuildData package. Change 3546469 by Jian.Ru Take into account CVarStaticMeshLODDistanceScale during static mesh LOD calculation Change 3546804 by Daniel.Wright [Copy] Added SendAllEndOfFrameUpdates draw event to wrap skin cache events Change 3546814 by Daniel.Wright [Copy] Only use skylight OcclusionMaxDistance for the global distance field if it casts shadows Change 3546815 by Daniel.Wright [Copy] Snap volumetric fog light function target resolution to a factor of 32 to avoid constant texture reallocation Change 3546817 by Daniel.Wright [Copy] Warmup time warning Change 3546828 by Daniel.Wright [Copy] Fixed UWorld::DestroyActor in PIE calling InvalidateLightingCacheDetailed which can do a FlushRenderingCommands and cause a large hitch Change 3546836 by Daniel.Wright [Copy] ULightComponent::InvalidateLightingCacheInner uses MarkRenderStateDirty instead of slow reregister + FlushRendingCommands, and only for lights which might have static lighting data Change 3546849 by Rolando.Caloca DR - vk - Fix missing samplerstates - Fixes for structured buffers - Add missing Draw and Dispatch Indirect Change 3547516 by Brian.Karis Linear time 5-coloring for planar graphs. Brought in the Planarity library written by John Boyer, heavily edited and trimmed down to only include code necesary for graph coloring. Put behind a simple wrapper. Change 3547542 by Brian.Karis Linear time 5-coloring for planar graphs. Brought in the Planarity library written by John Boyer, heavily edited and trimmed down to only include code necesary for graph coloring. Put behind a simple wrapper. Change 3547563 by Brian.Karis Fixed some compiler warnings and hopefully some errors. Change 3547610 by Brian.Karis Replaced macros with inlined functions Change 3547620 by Brian.Karis Clean up includes Change 3547770 by Marcus.Wassmer GPU Crash for MTBF analytics Change 3547773 by Marcus.Wassmer Updated doxygen comment for new analytic Change 3548244 by Rolando.Caloca DR - Fix for translucency Change 3548352 by Yujiang.Wang Added soft source radius for point and spot lights * Soft source radius controls how 'blurry' the shape of specular lighting looks * Implemented by LobeRoughness modification * Better approximation for spherical lights so that they don't look sharp when the radius is large using 'smoothed representative point' method * Suppoted LightTangent in forward shading Change 3548530 by Brian.Karis Fix for mac build Change 3548770 by Rolando.Caloca DR - vk - Prereq work for Vulkan parallel RHI contexts Change 3548772 by Jian.Ru Fixed an issue that caused an ensure when switching levels in D3D10. #rb Marcus.Wassmer Change 3548865 by Daniel.Wright With shadowmap caching of whole scene shadows, only one of the cache modes issues an occlusion query. Fixes a crash where the static primitive shadowmap is culled but the movable primitive shadowmap is visible, which is normally not possible. Change 3548952 by Rolando.Caloca DR - Allow separate samplers in the shaders on Vulkan Change 3549197 by Marcus.Wassmer Fix DX12 PIx not working in cooked builds Change 3549209 by Daniel.Wright Occlusion culling for CSM, from the main camera, controlled by 'r.Shadow.OcclusionCullCascadedShadowMaps'. Disabled by default as rapid view changes don't work well with latent occlusion queries. Change 3549943 by Ben.Marsh Include better diagnostic information when a modified build product is detected after running a build step. Change 3550546 by Rolando.Caloca DR - Fix merge issue Change 3550962 by Marcus.Wassmer EarlyZ Masking requires full depth prepass, so just force it to. Change 3551062 by Daniel.Wright Handle NULL skylight Change 3551104 by Rolando.Caloca DR - vk - Remove assert to match other platforms Change 3551221 by Rolando.Caloca DR - vk - Add mirror clamp to edge extension - Fix framebuffer deletion Change 3551224 by Daniel.Wright Volumetric lightmap increase density around static lights affecting a voxel brighter than LightBrightnessSubdivideThreshold. Change 3551495 by Rolando.Caloca DR - vk - Intiial support for async queue Change 3552101 by Rolando.Caloca DR - vk - Fix for async Change 3552102 by Rolando.Caloca DR - SkinCache - Fix potential leak on staging buffers for recompute tangents - Integrate changes from 4.17 for memory optimizations Change 3552104 by Rolando.Caloca DR - vk - Support for SRVs for index buffers Change 3552838 by Rolando.Caloca DR - vk - Enable debug markers if found Change 3553106 by Rolando.Caloca DR - vk - Fixes for index buffer SRVs Change 3553107 by Rolando.Caloca DR - vk - Enable recompute tangents on Vulkan Change 3553154 by Rolando.Caloca DR - vk - Fix crash with null uav Change 3553342 by Yujiang.Wang Fix redundant skylights in AdvancedPreviewScene * PreviewScene was changed to using a skylight instead of ambient cubemap to support forward shading * AdvancedPreviewScene originally had a skylight, now it is changed to using the one inherited from PreviewScene Change 3553481 by Rolando.Caloca DR - Integrate fix for D3D12 support of index buffers SRVs #jira UE-47674 Change 3553715 by Rolando.Caloca DR - Fix crash when launching PC with -featureleveles31 Change 3553725 by Rolando.Caloca DR - Redo fix Change 3553803 by Rolando.Caloca DR - Shader compile fixes for ES3.1 Change 3553963 by Rolando.Caloca DR - vk - Remove extra IRDump Change 3554741 by Ben.Marsh CIS fix. Change 3555222 by Rolando.Caloca DR - vk - static analysis fix Change 3555362 by Rolando.Caloca DR - vk - Prep work for separate present queue Change 3556800 by Daniel.Wright Fixed screenshot for simple volume material doc Change 3556942 by Brian.Karis Fixed Bokeh DOF regression. Change 3556959 by Rolando.Caloca DR - vk - Rework staging buffer peak usage Change 3557497 by Daniel.Wright Better display name for Unbound property on post process volume Change 3557499 by Daniel.Wright Disable r.GenerateLandscapeGIData by default, opt-in for kite demo. Projects that want to use heightfield GI need to opt-in to r.GenerateLandscapeGIData. Change 3557068 by Olaf.Piesche Configurable spawn rate scaling reference value; sets the zero-scale reference value (default: 2), so additional quality levels can be added and scaling customized further. IMPORTANT: This sets the reference to 3 in PS4Scalability.ini; effects on PS4 are again going to have reduced spawn rates versus PC and Neo, as intended by the FX artists starting with this change. #tests QAGame test maps Change 3558123 by Rolando.Caloca DR - vk - static analysis fix Change 3558685 by Yujiang.Wang Github #3323: Two sided foliage lightmap directionality fix * Subsurface is not intended to work with lightmaps that don't have directionality, however we still want it to look similar to a directional one * Now it uses a constant directionality value #jira UE-42523 Change 3559052 by Brian.Karis Hopefully fix static analysis Change 3559113 by Rolando.Caloca DR - Fix crash witrh planar reflections Change 3559275 by Yujiang.Wang Fix race condition on several scalability CVars between rendering thread and game thread Change 3559612 by Rolando.Caloca DR - vk - SM5 with uniform buffers backend support Change 3559716 by Rolando.Caloca DR - hlslcc - Fix linker warning on SCW debug Change 3559768 by Rolando.Caloca DR - vk - Keep ub names for bindings Change 3560195 by Rolando.Caloca DR - accessor Change 3560275 by Rolando.Caloca DR - vk - Support for uniform buffers Change 3560913 by Rolando.Caloca DR - vk - Fix static analysis Change 3561145 by Rolando.Caloca DR - Don't crash if out of resource table bits Change 3561194 by Rolando.Caloca DR - vk - Integrate timestamp fixes Change 3562009 by Rolando.Caloca DR - vk - Workaround for bad UTexture data Change 3563884 by Chris.Bunner VK_NULL_HANDLE fix. Change 3563885 by Jian.Ru Ignore a warning caused by enabling distance field generation so that test Cube_Blue and Cube_Section don't fail. #rb Chris.Bunner Change 3565943 by Jian.Ru Add extra warning log triggered when attempt to create FRWBuffer greater than 256MB in ComputeLightGrid() #rb Chris.Bunner Change 3569479 by Michael.Lentine Integrate rhino shader changes to dev-rendering Change 3569511 by Michael.Lentine Fix formating and string out on windows. Change 3569572 by Yujiang.Wang Fix MeasureLongGPUTaskExecutionTime crashing on AMD on Macs Change 3569614 by Yujiang.Wang Flush rendering commands before measuring the long GPU task's excution time to get accurate results Change 3570524 by Jian.Ru Add extra parentheses to avoid compilation warning #rb Chris.Bunner Change 3570722 by Chris.Bunner Static analysis workaround - same code, just validating compile-time assumptions a little further. Change 3570880 by Jian.Ru Add small depth offset to avoid depth test failing during velocity pass #jira UE-37556 Change 3572532 by Jian.Ru Disable a warning to let tests pass #jira UE-48021 Change 3573109 by Michael.Lentine Checkin Michael.Trepka's fix for external dynamic libraries on mac. This is needed to make the build go green on mac. Change 3573995 by Jian.Ru Move an include out of define to let nightly build pass Change 3574777 by Chris.Bunner Continued merge fixes. Change 3574792 by Rolando.Caloca DR - Rename todo Change 3574794 by Chris.Bunner Re-adding includes lost in a pre-merge merge. Change 3574879 by Michael.Trepka Disabled a couple of Mac deprecation warnings Change 3574932 by Chris.Bunner Merge fix. Change 3575048 by Michael.Trepka Fixed iOS compile warnings Change 3575530 by Chris.Bunner Duplicating static analysis fix CL 3539836. Change 3575582 by Chris.Bunner Fixed GetDimensions return type in depth resolve shaders. Compile error fix. Change 3576326 by Chris.Bunner Static analysis fixes. Change 3576513 by Michael.Trepka Updated Mac MCPP lib to be compatible with OS X 10.9 Change 3576555 by Richard.Wallis Metal Validation Errors. Dummy black volume texture is in the wrong format in the Metal shader for the VolumetricLightmapIndirectionTexture. Create a new dummy texture with pixel format PF_R8G8B8A8_UINT. #jira UE-47549 Change 3576562 by Chris.Bunner OpenGL SetStreamSource stride updates. Change 3576589 by Michael.Trepka Fixed Mac CIS warnings and errors in Dev-Rendering Change 3576708 by Jian.Ru Fix cascade preview viewport background color not changing #jira UE-39687 Change 3576827 by Rolando.Caloca DR - Minor fix for licensee Change 3576973 by Chris.Bunner Fixing up HLSLCC language spec mismatch (potential shader compile crashes in GL and Vulkan). Change 3577729 by Rolando.Caloca DR - Fix for info on SCW crashes Change 3578723 by Chris.Bunner Fixed issue where custom material attribute was using display name as hlsl function name. Change 3578797 by Chris.Bunner Fixed pixel inspector crashing on high-precision normals gbuffer format. #jira UE-48094 Change 3578815 by Yujiang.Wang Fix for UE-48207 Orion cooked windows server crash on startup * Crash caused by rendering features not available in a dedicated server build * Skip over MeasureLongGPUTaskExecutionTime when !FApp::CanEvenRender() #jira UE-48207 Change 3578828 by Daniel.Wright Disable volumetric lightmap 3d texture creation on mobile Change 3579473 by Daniel.Wright Added View.SharedBilinearClampSampler and View.SharedBilinearWrapSampler. Used these to reduce base pass sampler counts with volumetric lightmaps. Change 3580088 by Jian.Ru Fix QAGame TM-CharacterMovement crashing on PIE #jira UE-48031 Change 3580388 by Daniel.Wright Fixed shadowed light injection into volumetric fog fallout from Rhino merge Change 3580407 by Michael.Trepka Updated Mac UnrealPak binaries Change 3581094 by Michael.Trepka Fix for ScreenSpaceReflections not working properly on iOS 11 Change 3581242 by Michael.Trepka Fixed a crash on startup on Mac when launching TM-ShaderModels in QAGame #jira UE-48255 Change 3581489 by Olaf.Piesche Replicating CL 3578030 from Fortnite-Main to fix #jira UE-46475 #jira FORT-47068, FORT-49705 Don't inappropriaely touch game thread data on the render thread. Push SubUV cutout data into a RT side object owned by the sprite dynamic data. #tests FN LastPerfTest Change 3581544 by Simon.Tovey Fix for ensure accessing cvar from task thread. #tests no more ensure Change 3581934 by Chris.Bunner Fixed ConsoleVariables.ini break from merge. Change 3581968 by Jian.Ru Fix QAGame TM-ShaderModels PIE crash when resizing game viewport #jira UE-48251 Change 3581989 by Richard.Wallis Fix for NULL PrecomputedLightingBuffer. It is null for first frame request in forward rendering so should have the GEmptyPrecomputedLightingUniformBuffer set in these cases after it's been initially tried to be set not before. #jira UE-46955 Change 3582632 by Chris.Bunner Resolved merge error. Change 3582722 by Rolando.Caloca DR - Workaround for PF_R8G8B8A8_UINT on GL #jira UE-48208 Change 3584096 by Rolando.Caloca DR - Fix for renderdoc crashing in shipping #jira UE-46867 Change 3584245 by Jian.Ru Fix System.Promotion.Editor.Particle Editor test crash #jira UE-48235 Change 3584359 by Yujiang.Wang Fix for UE-48315 Wall behind base in Monolith is flickering white in -game Orion * Caused by dot(N, V) being negative * Clamp to (0, 1) #jira UE-48315 Change 3587864 by Mark.Satterthwaite Fix the GPU hang on iOS caused by changes to the Depth-Stencil MSAA handling: you can't store the MSAA stencil results on iOS < 10 unless you use the slower MTLStoreActionStoreAndMultisampleResolve which we don't need for the mobile renderer. #jira UE-48342 Change 3587866 by Mark.Satterthwaite Correctly fix iOS compilation errors against Xcode 9 Beta 5 and Xcode 8.3.3 - duplicating function definitions is guaranteed to be wrong. Change 3588168 by Mark.Satterthwaite Move the Xcode version into the Metal shader format header, not the DDC key, so that we can handle bad compiler/driver combinations in the runtime and don't force all users to recompile every time the Xcode version changes. Change 3588192 by Rolando.Caloca DR - Fix d3d12 linker error when EXECUTE_DEBUG_COMMAND_LISTS is enabled Change 3588291 by Rolando.Caloca DR - Fix for d3d12 command list crash: Commited resources can not have aliasing barriers #jira UE-48299 Change 3590134 by Michael.Trepka Copy of CL 3578963 Reset automation tests timer after shader compilation when preparing for screenshots taking to make sure tests don't time out. Change 3590405 by Rolando.Caloca DR - hlslcc - support for sqrt(uint) Change 3590436 by Mark.Satterthwaite Rebuild Mac hlslcc for CL #3590405 - without the various compiler workarounds left over from before the recent code changes. Change 3590674 by Rolando.Caloca DR - vk - Integration from working branch - Fixes distance field maps - Compute pipelines stored in saved file - Adds GRHIRequiresRenderTargetForPixelShaderUAVs for platforms that need dummy render targets Change 3590699 by Rolando.Caloca DR - Fix distance fields mem leak Change 3590815 by Rolando.Caloca DR - vk - Fixes for uniform buffers and empty resource tables Change 3590818 by Mark.Satterthwaite Temporarily switch back to OpenVR v1.0.6 for Mac only until I can clarify what to do about a required but missing API hook for Metal. Re-enabled and fixed compile errors with Mac SteamVR plugin code. Change 3590905 by Mark.Satterthwaite For Metal shader compilation where the bytecode compiler is unavailable force the debug compiler flag and disable the archiving flag because storing text requires this. #jira UE-48163 Change 3590961 by Mark.Satterthwaite Submitted on Richard Wallis's behalf as he's on holiday: Mac fixes for Compute Skin Cache rendering issues (resulting in incorrect positions and tangents) and for recomputing tangents. Problem sampling from buffers/textures as floats with packed data. Some of the data appears as denorms so get flushed to zero then reinterpreted as uints via asuint or in Metal as_type<uint>(). Fix here for Metal seems to be to use uint types for the skin cache SRV's and as_type<> to floats instead. There could be some other areas where we're unpacking via floats that could affect Metal and I'm not sure how this will impact on other platforms. #jira UE-46688, UE-39256, UE-47215 Change 3590965 by Mark.Satterthwaite Remove the Z-bias workaround from Metal MRT as it isn't required and actually causes more problems. Change 3590969 by Mark.Satterthwaite Make all Metal shader platforms compile such that half may be used, unless the material specifies full precision. Change 3591871 by Rolando.Caloca DR - Enable PCSS on Vulkan & Metal - Enable capsule shadows on Vulkan Change 3592014 by Mark.Satterthwaite Remove support for Mac OS X El Capitan (10.11) including the stencil view workaround. Bump the minimum Metal shader standard for Metal SM4, SM5 & Metal MRT to v1.2 (macOS 10.12 Sierra & iOS 10) so we can use FMAs and other newer shader language features globally. Enable the new GRHIRequiresRenderTargetForPixelShaderUAVs flag as Metal is like Vulkan and needs a target for fragment rendering. Also fix the filename for direct-compile & remove the old batch file generation in the Metal shader compiler. Change 3592171 by Rolando.Caloca DR - CIS fix Change 3592753 by Jian.Ru repeat Daniel's fix on xb1 profilegpu crash (draw events cannot live beyond present) Change 3594595 by Rolando.Caloca DR - Fix D3D shader compiling run time stack corruption failure on debug triggering falsely Change 3594794 by Michael.Trepka Call FPlatformMisc::PumpMessages() before attempting to toggle fullscreen on Mac to fix an issue on some Macs running 10.13 beta that would ignore the toggle fullscreen call freezing the app Change 3594999 by Mark.Satterthwaite Disable MallocBinned2 for iOS as on Rhino it worked but on iOS 10.0.2 there are bugs (munmap uses 64kb granularity, not the 4096 the code expects given the reported page-size). While we are here remove the spurious FORCE_MALLOC_ANSI from the iOS platform header. #jira UE-48342 Change 3595004 by Mark.Satterthwaite Disable Metal's Deferred Store Actions and combined Depth/Stencil formats on iOS < 10.3 as there are bugs on earlier versions of iOS 10. #jira UE-48342 Change 3595386 by Mark.Satterthwaite Silence the deprecation warning for kIOSurfaceIsGlobal until SteamVR switches to one of the newer IOSurface sharing mechanisms. Change 3595394 by Rolando.Caloca DR - Added function for tracking down errors in the hlsl parser - Added support for simple #if 0...#endif Change 3599352 by Rolando.Caloca DR - Fixes for HlslParser - Added missing attributes for functions - Fixed nested assignment Change 3602440 by Michael.Trepka Fixed Metal shader compilation from Windows with remote compilation disabled #jira UE-48163 Change 3602898 by Chris.Bunner Resaving assets. Change 3603731 by Jian.Ru fix a crash caused by a material destroyed before the decal component #jira UE-48587 Change 3604629 by Rolando.Caloca DR - Workaround for PF_R8G8B8A8_UINT on Android #jira UE-48208 Change 3604984 by Peter.Sauerbrei fix for orientation not being limited to that specified in the plist #jira UE-48360 Change 3605738 by Chris.Bunner Allow functional screenshot tests to request a camera cut (e.g. tests relying on temporal aa history). #jira UE-48748 Change 3606009 by Mark.Satterthwaite Correctly implement ClipDistance for Metal as an array of floats as required by the spec. and fix a few irritating issues from the merge that should not have. - When compiling a tessellation vertex shader in the SCW direct mode we can't evaluate non-existant defines and we don't actually need to. - The define names, values & shader file name are irrelevant to the Metal output key, but the shader format name & Metal standard really do matter - should speed up Metal shader compilation a bit. - Move the shader vertex layer clip-distance to index 2 to avoid conflicts. - Don't default initialise the debug code string for Metal shaders or it won't print out the actual code.... #jira UE-47663 Change 3606108 by Mark.Satterthwaite Temporary hack to avoid a crash in AVPlayer. #jira UE-48758 Change 3606121 by Mark.Satterthwaite Fix Windows compilation. Change 3606992 by Chris.Bunner Static analysis fix. [CL 3608256 by Marcus Wassmer in Main branch]
2017-08-24 15:38:57 -04:00
public:
FMetalShaderFormat()
{
FMetalCompilerToolchain::CreateAndInit();
}
virtual ~FMetalShaderFormat()
{
FMetalCompilerToolchain::Destroy();
}
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3607928) #lockdown Nick.Penwarden ============================ MAJOR FEATURES & CHANGES ============================ Change 3441680 by Uriel.Doyon Added units to point light intensity, to allow the user to specify the value in candelas or lumens. New point light actors now configure the intensity in candelas by default. Replaced viewport exposure settings by an EV100 slider. Hidding the tone mapper in the show flag now still applies the exposure. Added a new AutoExposure method called EV100 which allows to specify : - MinEV100, MaxEV100 - Calibration Constnat - Exposure Compensation #jira UE-42783 Change 3454934 by Chris.Bunner Backing out changelists 3441680, 3454636 and 3454844 for the sake of integration stability. Change 3512118 by Marc.Olano Fix rare Sobol shader data problem. Mismatch with CPU code after a large number of points Resubmit of portion of //UE4/Dev-Rendering@3509854 that was rolled back to avoid massive shader recompiles during integration testing Change 3512129 by Benjamin.Hyder Fixing up content in TM-SobolNoise Change 3512151 by Rolando.Caloca DR - Fixed some layouts that were general - Added some extra dump information Change 3512160 by Benjamin.Hyder Still Fixing TM-Sobol Change 3512180 by Marc.Olano PCSS for spotlights. Like directional PCSS this is experimental, activated by r.Shadow.FilterMethod. Change 3512261 by Michael.Lentine Move Subsurface to shared properties. Previously the same code could be executed multiple times without being optimized out if multiple inputs used the same subsurface output. #jira UE-44405 Change 3512288 by Rolando.Caloca DR - Fix issue when recycling image handles Change 3512338 by Michael.Lentine Fix precision if user enters a multiple of 90 degree rotation for transforms. This will only work for exact values. Generally comparing float point numbers using == is unsafe but it should be ok in this case as they are exact values entered from the UI. We may want to later expand this to include thresholding using a value ~1e-7. #jira UE-46137 Change 3512424 by Michael.Lentine Regenerate BaseColor.uasset and Specular.uasset to not have the notforclient flags set. #jira UE-44315 Change 3512686 by Brian.Karis Fix for quadric assert in infiltrator. Due to bad tangents in source mesh. Change 3512696 by Brian.Karis Unrevert TAA. Fixed DOF NaN artifacts Change 3512717 by Marcus.Wassmer PR #3714: Fix typo in EOcclusionCombineMode (Contributed by Mumbles4) Change 3513112 by Richard.Wallis Crash when packaging for iOS with Shared Material Native Libraries and Share Material Shader Code from windows platform. Offline shader compile for archiving not done - shader header has missing offline compile flag for native Metal library archiving. Fix includes: - Handle offline compile failure when not running on Mac and no remote is configured (or remote fails). (I think it's this point at which the crash in the bug report is at). - Make sure remote can build for native Metal libraries and archive correctly - this should now support Linux platforms or Mac to Mac (if enabled in MetalShaderCompiler.cpp) for testing if required. - Updated to include remote calling into the xcode 9 Metal pch fix already submitted by Mark Satt. #jira UE-45657 Change 3513357 by Richard.Wallis Windows compile fix. Change 3513375 by Guillaume.Abadie Exposes the possibility to manually destroy the GPU ressource of UTextureRenderTarget2D. Change 3513685 by Richard.Hinckley #jira UEDOC-3822 Fixing a comment that refers to a non-existent function, for documentation purposes. Change 3513705 by Marc.Olano Updates to Sobol test levels in RenderTest project Change 3513730 by Rolando.Caloca DR - Fix mip size copying resolve targets - Fix compute fence - Fix descriptor set texture layout - More dump info Change 3513742 by Marc.Olano Texture-free numeric print for shader debugging Change 3513777 by Daniel.Wright Handled edge case where no furthest samples are found in precomputed visibility Change 3514852 by Rolando.Caloca DR - Fix -directcompile on SCW Change 3515049 by Rolando.Caloca DR - hlslcc dump crash fix Change 3515167 by Rolando.Caloca DR - hlslcc - Fix bogus string pointer - Allow reading from non-scalar UAVs Change 3515745 by Rolando.Caloca DR - Linux compile fix Change 3515862 by Rolando.Caloca DR - Remove old reference to CCT - Link with hlslcc debug libs on SCW debug config for easier debugging Change 3516292 by Rolando.Caloca DR - glslang exe fixes Change 3516568 by Rolando.Caloca DR - hlslcc - Copy fix for *Buffer as functionparameters Change 3516659 by Marcus.Wassmer Fix some d3derrors with distance fields Change 3516801 by Daniel.Wright Fixed crash when doing editor 'Force Delete' on a static mesh whose distance field is still being built. Any UObject reference that is to an asset can be NULL'ed by the editor. Change 3516825 by Rolando.Caloca DR - Some initial fixes for structured buffers Change 3516843 by Rolando.Caloca DR - Fix for Vulkan dist fields Change 3516869 by Marcus.Wassmer Add format to the createrendertarget blueprint node Change 3516957 by Daniel.Wright Fixed bUsesDistortion being editable Change 3516965 by Daniel.Wright Still mark the distance field task completed, even if the static mesh has been deleted Change 3517039 by Yujiang.Wang GitHub #2655: Optimization for shadow map resolution selection for spot lights * Use the radius of the inscribed sphere at the cone end as the spot light's screen radius Note: slight drop of shadow quality of spot lights may occur when they are far away from the camera. This is intended, since before this optimization they tend to be always rendered with the maximum shadow map resolution (2048), which is very costly #jira UE-33982 Change 3517069 by Yujiang.Wang Fix for ScissorRect settings in d3d11 being lost under certain scenarios * Scissor rectangle is always enabled in the low-level d3d11 pipeline, and it is expected that at least one ScissorRect is present no matter whether RHISetScissorRect is called with bEnable=false (when it is false we just use a big rect to make it effectively disabled) * However FD3D11StateCacheBase::ClearState() clears all the states, which removes scissor rectangles and causes problems for certain routines (FScene::UpdateSkyCaptureContents) * Now SetScissorRectIfRequiredWhenSettingViewport will always set a effectively disabled ScissorRect on each FD3D11DynamicRHI::RHISetViewport call, just like d3d12 does #jira UE-45465 UE-44760 Change 3517134 by Yujiang.Wang CIS fix Change 3517662 by Rolando.Caloca DR - Execute upload Vulkan cmds on the RHI thread - Fix crash with structured buffer Change 3517677 by Rolando.Caloca DR - Update/copy textures on RHI thread Change 3517680 by Rolando.Caloca DR - Copy texture bulk data on rhi thread Change 3517748 by Marcus.Wassmer temporary workaround for one class of GPU crashes Change 3518832 by Rolando.Caloca DR - Copy & extend 3518077 - Fix for movable skylight shader missing on simple forward (low lighting quality mode) Change 3519973 by Richard.Wallis Jittering in Engine Menu Dropdown Options. Jitter fix: Fix some areas that hadn't been changed from RoundToInt (from previous CL's) to CeilToInt. #jira UE-46505 Change 3520849 by Uriel.Doyon Fixed issue with investigate texture command and dynamic component entries. Change 3521064 by Guillaume.Abadie Returns absolute path of shader files on error to avoid work loss in visual studio that can't figure out that a sln relative and absolute path might leading to same file on disk. Change 3521834 by Rolando.Caloca DR - Fix decals on Vulkan Change 3521892 by Rolando.Caloca DR - Fix Vulkan texture streaming Change 3523181 by Rolando.Caloca DR - Copy from 3523176 UE4.17 - Fix Vulkan scissor causing text to not clip Change 3523534 by Yujiang.Wang UE-46631: Implement a scalable LongGPUTask to fix ProfileGPU * A new, scalable, platform-independent IssueLongGPUTask is now implemented in UtilityShaders * Removed IssueLongGPUTask and G*Vector4VertexDeclaration from RHI implementations * The measurement of the execution time of a basic LongGPUTask unit is kicked off on the very first frame #jira UE-46631 Change 3524552 by Yujiang.Wang Fix iteration number calculation of LongGPUTask Change 3524975 by Joe.Graf Moved the Hamming-weight function from StaticMeshDrawList.inl to FGenericPlatformMath Added SSE versions using _mm_popcnt_u64 for platforms that support it Added a SSE check to gracefully exit when missing the instruction and it was expected to be there #CodeReview: arciel.rekman, brian.karis Change 3525306 by Daniel.Wright Fixed ensure from LPV Change 3525346 by Rolando.Caloca DR - Fix linking issue Change 3525459 by Daniel.Wright Volumetric Lightmaps - higher quality precomputed GI on dynamic objects and GI on Volumetric Fog * Enabled by default on all maps, effective after a lighting build. This replaces the existing Precomputed Light Volume and Indirect Lighting Cache features. * New Lightmass World Settings: VolumeLightingMethod, VolumetricLightmapDetailCellSize and VolumetricLightmapMaximumBrickMemoryMb. * Lightmass computes lighting samples in an adaptive grid, with higher density around geometry inside the importance volume. Positions outside the importance volume get lit with the border texels. * Improved Lightmass volume solver to use importance photons and full adaptive final gather, so volume samples have similar quality to 2d lightmaps. * A static indirection texture is built covering the importance volume and flattening the brick tree by storing the offset to the highest density brick at each indirection cell. * Seamless and efficient GPU interpolation across density levels is achieved by adding a single row of padding to bricks, copied from neighbors, and stitching up bricks with lower density neighbors * The Volumetric lightmap stores Irradiance as a 3 band SH, which is 27 floats, quantized into 28 bytes, 7 texture lookups. * A full screen barebones material using Volumetric Lightmaps costs .42ms on 970 GTX, while Indirect Lighting Cache Point costs .32ms * Sky bent normal is also stored for stationary skylights and Directional Light Shadowing for Single Sample Shadow receiving. * Volumetric fog, Movable components, unbuilt Static Components, SingleSampleShadow receiving and Capsule Shadows use Volumetric Lightmaps if available * New Visualization show flag for Volumetric Lightmap sample points * Level streaming of volume light data is not currently supported with this method Change 3525461 by Daniel.Wright Lowered default r.Shadow.RadiusThreshold for Epic shadow settings as it was causing a lot of visible artifacts from small objects popping out. This will increase shadowmap cost slightly (13.5ms RT -> 14.3ms RT in Fortnite on PS4, no measurable GPU difference). Change 3526459 by Rolando.Caloca DR - Fix validation error Change 3526474 by Rolando.Caloca DR - Integrate from GV Change 3526487 by Daniel.Wright Disabled Volumetric Lightmap filtering with neighbors due to artifacts Fix linux compile errors Change 3526833 by Rolando.Caloca DR - Workaround for hlslcc Change 3526991 by Uriel.Doyon Integrated 3526859 : Texture mip bias is now reset whenever the streaming budget increases. This fixes an issue where textures persistently become low res after a memory spike. Change 3527574 by Rolando.Caloca DR - Added some missing resource entries for SCW direct mode Change 3527625 by Rolando.Caloca DR - Copy from 3527113 UE4.17 - Fix Vulkan not calling Present Change 3528461 by Brian.Karis Support larger hash sizes. Added uint list hashing function. Change 3528780 by Rolando.Caloca DR - Default Vulkan resources Change 3528818 by Rolando.Caloca DR - glslang - Added missing accessor Change 3528839 by Rolando.Caloca DR - Fix virtual path issue when using non-engine relative absolute paths Change 3528900 by Daniel.Wright Fixed variable shadowing Change 3529039 by Rolando.Caloca DR - Read Spirv reflection data (not used yet) Change 3529040 by Joe.Graf Fixed the 32bit compile failures for the popcnt optimization #CodeReview: arciel.rekman Change 3529060 by Rolando.Caloca DR - hlslcc - New flag for keeping resource names Change 3529344 by Rolando.Caloca DR - Delete unused file Change 3529723 by Brian.Karis Fixed static analysis cleaner. Change 3531357 by Michael.Trepka Updated Mac glslang libraries with latest changes. Also, updated the Xcode project (generated with CMake) and moved it to a different location so that it no longer uses hardcoded absolute paths. It should be easy to rebuild these libraries in the future. Change 3531517 by Joe.Graf Added support for ddx_fine, ddy_fine, ddx_coarse, ddy_coarse to hlslcc #CodeReview: arciel.rekman, mark.satterthwaite, rolando.caloca Change 3531626 by Joe.Graf Mac version of the popcount optimization Changed Linux version to use the same builtin #CodeReview: mark.satterthwaite, arciel.rekman Change 3531837 by Chris.Bunner SetScissorRectIfRequiredWhenSettingViewport sets the viewport size by default rather than disabling the scissor rect. #jra UE-46753 Change 3533415 by Joe.Graf Renamed the SSSE3 checks per feedback #CodeReview: arciel.rekman Change 3533480 by Michael.Lentine Use more accurate descriptions for shader recompile options Change 3533511 by Joe.Graf Updated the GenericPlatformMisc to match the SSSE3 name change #CodeReview: arciel.rekman Change 3533521 by Marcus.Wassmer Fix scenerenderer leak when updating out of view planar reflections Change 3533528 by Joe.Graf Updated comments #CodeReview: n/a Change 3533608 by Mark.Satterthwaite New manual Xcode project for glslang so that we include all the necessary code and can link again. Change 3534260 by Mark.Satterthwaite Fix the Xcode 9 Beta 3 compile errors in MetalRHI without breaking Xcode 8.3.3. Change 3535789 by Yujiang.Wang Fix for wrong hair shading in forward shading * IBL reflections should be turned off for hairs Change 3537059 by Ben.Marsh Fixing case of iOS directories, pt1 Change 3537060 by Ben.Marsh Fixing case of iOS directories, pt2 Change 3538297 by Michael.Lentine Add shader comparison test. Adding the basic test case. Adding logic to Common.ush to enable FP16 conditionally on a define (which is not set by default) Adding more exported functionality to automation for use in the shader test. Change 3538309 by Michael.Lentine Add missing file from Shader Test CL. Change 3538751 by Michael.Lentine Add missing pragma once. Change 3539236 by Michael.Lentine Do not ignore return values. Change 3539237 by Michael.Lentine Check in the correct file Change 3540343 by Rolando.Caloca DR - Added t.DumpHitches.AllThreads Change 3540661 by Yujiang.Wang Fix spot tube light direction * The tube direction for a spot light was pointing along the light direction, now it is along the local Z axis which is perpendicular to the light direction. Lightmass is also touched * A new LightTangent is added to FDeferredLightData * Packed all the values from LightSceneProxy->GetParameters into a single FLightParameters struct to avoid copy-pasting them everywhere Change 3541129 by Rolando.Caloca DR - vk - Copy all Vulkan fixes from 4.17 Change 3541347 by Yujiang.Wang Fix wrong ViewFlags being set between objects when rendering shadow depth maps * Bug caused by trying to share DrawRenderState between objects, but SetViewFlagsForShadowPass was designed to start from a fresh render state * Now SetViewFlagsForShadowPass recalculates and sets the flags on each call Change 3542603 by Rolando.Caloca DR - vk - Allow sharing samplers on Vulkan Change 3542639 by Jian.Ru Changed warning text to better indicate that global clip plane needs to be enabled for planar reflection #RB Marcus.Wassmer Change 3543167 by Michael.Lentine Fix naming for the shader comparison tests. Change 3543210 by Uriel.Doyon Fixed an issue when computing material scales where the default material ends up being used instead of the required material. In that case, we used the default settings for texture streaming (assuming a scale of 1). Change 3543221 by Brian.Karis Simplifier optimizations Change 3543239 by Arciel.Rekman hlslcc: remove FCustomStd* workarounds. - This was previous attempt to work around problems arising from different STL used for building libhlslcc (in the cross-toolchain) and possibly different STL used for building engine (on the system). - The same problem has been resolved by bundling libc++. Change 3543946 by Michael.Lentine Add comparison output. Change 3544277 by Brian.Karis Fixed uninitialized var error Change 3544404 by Rolando.Caloca DR - Fix broken textures Change 3544503 by Jian.Ru Ensure lighting failure delegates are always called #RB Marcus.Wassmer,Daniel.Wright #3689 Change 3545241 by Daniel.Wright Fixed spotlight whole scene shadows using a radius 2x too long Change 3545347 by Daniel.Wright Fixed shadow occlusion culling broken by shadowmap caching change. FProjectedShadowKey is now computed correctly for whole scene shadows and SDCM_StaticPrimitivesOnly shadowmaps will fall back to the query for a SDCM_MovablePrimitivesOnly, since the static primitives shadowmap's query is not issued every frame. Change 3546196 by Marcus.Wassmer Fix minor typo Change 3546459 by Daniel.Wright ULevel::PostEditChangeProperty recreates rendering resources if MapBuildData is modified - fixes a crash when Force Deleting the MapBuildData package. Change 3546469 by Jian.Ru Take into account CVarStaticMeshLODDistanceScale during static mesh LOD calculation Change 3546804 by Daniel.Wright [Copy] Added SendAllEndOfFrameUpdates draw event to wrap skin cache events Change 3546814 by Daniel.Wright [Copy] Only use skylight OcclusionMaxDistance for the global distance field if it casts shadows Change 3546815 by Daniel.Wright [Copy] Snap volumetric fog light function target resolution to a factor of 32 to avoid constant texture reallocation Change 3546817 by Daniel.Wright [Copy] Warmup time warning Change 3546828 by Daniel.Wright [Copy] Fixed UWorld::DestroyActor in PIE calling InvalidateLightingCacheDetailed which can do a FlushRenderingCommands and cause a large hitch Change 3546836 by Daniel.Wright [Copy] ULightComponent::InvalidateLightingCacheInner uses MarkRenderStateDirty instead of slow reregister + FlushRendingCommands, and only for lights which might have static lighting data Change 3546849 by Rolando.Caloca DR - vk - Fix missing samplerstates - Fixes for structured buffers - Add missing Draw and Dispatch Indirect Change 3547516 by Brian.Karis Linear time 5-coloring for planar graphs. Brought in the Planarity library written by John Boyer, heavily edited and trimmed down to only include code necesary for graph coloring. Put behind a simple wrapper. Change 3547542 by Brian.Karis Linear time 5-coloring for planar graphs. Brought in the Planarity library written by John Boyer, heavily edited and trimmed down to only include code necesary for graph coloring. Put behind a simple wrapper. Change 3547563 by Brian.Karis Fixed some compiler warnings and hopefully some errors. Change 3547610 by Brian.Karis Replaced macros with inlined functions Change 3547620 by Brian.Karis Clean up includes Change 3547770 by Marcus.Wassmer GPU Crash for MTBF analytics Change 3547773 by Marcus.Wassmer Updated doxygen comment for new analytic Change 3548244 by Rolando.Caloca DR - Fix for translucency Change 3548352 by Yujiang.Wang Added soft source radius for point and spot lights * Soft source radius controls how 'blurry' the shape of specular lighting looks * Implemented by LobeRoughness modification * Better approximation for spherical lights so that they don't look sharp when the radius is large using 'smoothed representative point' method * Suppoted LightTangent in forward shading Change 3548530 by Brian.Karis Fix for mac build Change 3548770 by Rolando.Caloca DR - vk - Prereq work for Vulkan parallel RHI contexts Change 3548772 by Jian.Ru Fixed an issue that caused an ensure when switching levels in D3D10. #rb Marcus.Wassmer Change 3548865 by Daniel.Wright With shadowmap caching of whole scene shadows, only one of the cache modes issues an occlusion query. Fixes a crash where the static primitive shadowmap is culled but the movable primitive shadowmap is visible, which is normally not possible. Change 3548952 by Rolando.Caloca DR - Allow separate samplers in the shaders on Vulkan Change 3549197 by Marcus.Wassmer Fix DX12 PIx not working in cooked builds Change 3549209 by Daniel.Wright Occlusion culling for CSM, from the main camera, controlled by 'r.Shadow.OcclusionCullCascadedShadowMaps'. Disabled by default as rapid view changes don't work well with latent occlusion queries. Change 3549943 by Ben.Marsh Include better diagnostic information when a modified build product is detected after running a build step. Change 3550546 by Rolando.Caloca DR - Fix merge issue Change 3550962 by Marcus.Wassmer EarlyZ Masking requires full depth prepass, so just force it to. Change 3551062 by Daniel.Wright Handle NULL skylight Change 3551104 by Rolando.Caloca DR - vk - Remove assert to match other platforms Change 3551221 by Rolando.Caloca DR - vk - Add mirror clamp to edge extension - Fix framebuffer deletion Change 3551224 by Daniel.Wright Volumetric lightmap increase density around static lights affecting a voxel brighter than LightBrightnessSubdivideThreshold. Change 3551495 by Rolando.Caloca DR - vk - Intiial support for async queue Change 3552101 by Rolando.Caloca DR - vk - Fix for async Change 3552102 by Rolando.Caloca DR - SkinCache - Fix potential leak on staging buffers for recompute tangents - Integrate changes from 4.17 for memory optimizations Change 3552104 by Rolando.Caloca DR - vk - Support for SRVs for index buffers Change 3552838 by Rolando.Caloca DR - vk - Enable debug markers if found Change 3553106 by Rolando.Caloca DR - vk - Fixes for index buffer SRVs Change 3553107 by Rolando.Caloca DR - vk - Enable recompute tangents on Vulkan Change 3553154 by Rolando.Caloca DR - vk - Fix crash with null uav Change 3553342 by Yujiang.Wang Fix redundant skylights in AdvancedPreviewScene * PreviewScene was changed to using a skylight instead of ambient cubemap to support forward shading * AdvancedPreviewScene originally had a skylight, now it is changed to using the one inherited from PreviewScene Change 3553481 by Rolando.Caloca DR - Integrate fix for D3D12 support of index buffers SRVs #jira UE-47674 Change 3553715 by Rolando.Caloca DR - Fix crash when launching PC with -featureleveles31 Change 3553725 by Rolando.Caloca DR - Redo fix Change 3553803 by Rolando.Caloca DR - Shader compile fixes for ES3.1 Change 3553963 by Rolando.Caloca DR - vk - Remove extra IRDump Change 3554741 by Ben.Marsh CIS fix. Change 3555222 by Rolando.Caloca DR - vk - static analysis fix Change 3555362 by Rolando.Caloca DR - vk - Prep work for separate present queue Change 3556800 by Daniel.Wright Fixed screenshot for simple volume material doc Change 3556942 by Brian.Karis Fixed Bokeh DOF regression. Change 3556959 by Rolando.Caloca DR - vk - Rework staging buffer peak usage Change 3557497 by Daniel.Wright Better display name for Unbound property on post process volume Change 3557499 by Daniel.Wright Disable r.GenerateLandscapeGIData by default, opt-in for kite demo. Projects that want to use heightfield GI need to opt-in to r.GenerateLandscapeGIData. Change 3557068 by Olaf.Piesche Configurable spawn rate scaling reference value; sets the zero-scale reference value (default: 2), so additional quality levels can be added and scaling customized further. IMPORTANT: This sets the reference to 3 in PS4Scalability.ini; effects on PS4 are again going to have reduced spawn rates versus PC and Neo, as intended by the FX artists starting with this change. #tests QAGame test maps Change 3558123 by Rolando.Caloca DR - vk - static analysis fix Change 3558685 by Yujiang.Wang Github #3323: Two sided foliage lightmap directionality fix * Subsurface is not intended to work with lightmaps that don't have directionality, however we still want it to look similar to a directional one * Now it uses a constant directionality value #jira UE-42523 Change 3559052 by Brian.Karis Hopefully fix static analysis Change 3559113 by Rolando.Caloca DR - Fix crash witrh planar reflections Change 3559275 by Yujiang.Wang Fix race condition on several scalability CVars between rendering thread and game thread Change 3559612 by Rolando.Caloca DR - vk - SM5 with uniform buffers backend support Change 3559716 by Rolando.Caloca DR - hlslcc - Fix linker warning on SCW debug Change 3559768 by Rolando.Caloca DR - vk - Keep ub names for bindings Change 3560195 by Rolando.Caloca DR - accessor Change 3560275 by Rolando.Caloca DR - vk - Support for uniform buffers Change 3560913 by Rolando.Caloca DR - vk - Fix static analysis Change 3561145 by Rolando.Caloca DR - Don't crash if out of resource table bits Change 3561194 by Rolando.Caloca DR - vk - Integrate timestamp fixes Change 3562009 by Rolando.Caloca DR - vk - Workaround for bad UTexture data Change 3563884 by Chris.Bunner VK_NULL_HANDLE fix. Change 3563885 by Jian.Ru Ignore a warning caused by enabling distance field generation so that test Cube_Blue and Cube_Section don't fail. #rb Chris.Bunner Change 3565943 by Jian.Ru Add extra warning log triggered when attempt to create FRWBuffer greater than 256MB in ComputeLightGrid() #rb Chris.Bunner Change 3569479 by Michael.Lentine Integrate rhino shader changes to dev-rendering Change 3569511 by Michael.Lentine Fix formating and string out on windows. Change 3569572 by Yujiang.Wang Fix MeasureLongGPUTaskExecutionTime crashing on AMD on Macs Change 3569614 by Yujiang.Wang Flush rendering commands before measuring the long GPU task's excution time to get accurate results Change 3570524 by Jian.Ru Add extra parentheses to avoid compilation warning #rb Chris.Bunner Change 3570722 by Chris.Bunner Static analysis workaround - same code, just validating compile-time assumptions a little further. Change 3570880 by Jian.Ru Add small depth offset to avoid depth test failing during velocity pass #jira UE-37556 Change 3572532 by Jian.Ru Disable a warning to let tests pass #jira UE-48021 Change 3573109 by Michael.Lentine Checkin Michael.Trepka's fix for external dynamic libraries on mac. This is needed to make the build go green on mac. Change 3573995 by Jian.Ru Move an include out of define to let nightly build pass Change 3574777 by Chris.Bunner Continued merge fixes. Change 3574792 by Rolando.Caloca DR - Rename todo Change 3574794 by Chris.Bunner Re-adding includes lost in a pre-merge merge. Change 3574879 by Michael.Trepka Disabled a couple of Mac deprecation warnings Change 3574932 by Chris.Bunner Merge fix. Change 3575048 by Michael.Trepka Fixed iOS compile warnings Change 3575530 by Chris.Bunner Duplicating static analysis fix CL 3539836. Change 3575582 by Chris.Bunner Fixed GetDimensions return type in depth resolve shaders. Compile error fix. Change 3576326 by Chris.Bunner Static analysis fixes. Change 3576513 by Michael.Trepka Updated Mac MCPP lib to be compatible with OS X 10.9 Change 3576555 by Richard.Wallis Metal Validation Errors. Dummy black volume texture is in the wrong format in the Metal shader for the VolumetricLightmapIndirectionTexture. Create a new dummy texture with pixel format PF_R8G8B8A8_UINT. #jira UE-47549 Change 3576562 by Chris.Bunner OpenGL SetStreamSource stride updates. Change 3576589 by Michael.Trepka Fixed Mac CIS warnings and errors in Dev-Rendering Change 3576708 by Jian.Ru Fix cascade preview viewport background color not changing #jira UE-39687 Change 3576827 by Rolando.Caloca DR - Minor fix for licensee Change 3576973 by Chris.Bunner Fixing up HLSLCC language spec mismatch (potential shader compile crashes in GL and Vulkan). Change 3577729 by Rolando.Caloca DR - Fix for info on SCW crashes Change 3578723 by Chris.Bunner Fixed issue where custom material attribute was using display name as hlsl function name. Change 3578797 by Chris.Bunner Fixed pixel inspector crashing on high-precision normals gbuffer format. #jira UE-48094 Change 3578815 by Yujiang.Wang Fix for UE-48207 Orion cooked windows server crash on startup * Crash caused by rendering features not available in a dedicated server build * Skip over MeasureLongGPUTaskExecutionTime when !FApp::CanEvenRender() #jira UE-48207 Change 3578828 by Daniel.Wright Disable volumetric lightmap 3d texture creation on mobile Change 3579473 by Daniel.Wright Added View.SharedBilinearClampSampler and View.SharedBilinearWrapSampler. Used these to reduce base pass sampler counts with volumetric lightmaps. Change 3580088 by Jian.Ru Fix QAGame TM-CharacterMovement crashing on PIE #jira UE-48031 Change 3580388 by Daniel.Wright Fixed shadowed light injection into volumetric fog fallout from Rhino merge Change 3580407 by Michael.Trepka Updated Mac UnrealPak binaries Change 3581094 by Michael.Trepka Fix for ScreenSpaceReflections not working properly on iOS 11 Change 3581242 by Michael.Trepka Fixed a crash on startup on Mac when launching TM-ShaderModels in QAGame #jira UE-48255 Change 3581489 by Olaf.Piesche Replicating CL 3578030 from Fortnite-Main to fix #jira UE-46475 #jira FORT-47068, FORT-49705 Don't inappropriaely touch game thread data on the render thread. Push SubUV cutout data into a RT side object owned by the sprite dynamic data. #tests FN LastPerfTest Change 3581544 by Simon.Tovey Fix for ensure accessing cvar from task thread. #tests no more ensure Change 3581934 by Chris.Bunner Fixed ConsoleVariables.ini break from merge. Change 3581968 by Jian.Ru Fix QAGame TM-ShaderModels PIE crash when resizing game viewport #jira UE-48251 Change 3581989 by Richard.Wallis Fix for NULL PrecomputedLightingBuffer. It is null for first frame request in forward rendering so should have the GEmptyPrecomputedLightingUniformBuffer set in these cases after it's been initially tried to be set not before. #jira UE-46955 Change 3582632 by Chris.Bunner Resolved merge error. Change 3582722 by Rolando.Caloca DR - Workaround for PF_R8G8B8A8_UINT on GL #jira UE-48208 Change 3584096 by Rolando.Caloca DR - Fix for renderdoc crashing in shipping #jira UE-46867 Change 3584245 by Jian.Ru Fix System.Promotion.Editor.Particle Editor test crash #jira UE-48235 Change 3584359 by Yujiang.Wang Fix for UE-48315 Wall behind base in Monolith is flickering white in -game Orion * Caused by dot(N, V) being negative * Clamp to (0, 1) #jira UE-48315 Change 3587864 by Mark.Satterthwaite Fix the GPU hang on iOS caused by changes to the Depth-Stencil MSAA handling: you can't store the MSAA stencil results on iOS < 10 unless you use the slower MTLStoreActionStoreAndMultisampleResolve which we don't need for the mobile renderer. #jira UE-48342 Change 3587866 by Mark.Satterthwaite Correctly fix iOS compilation errors against Xcode 9 Beta 5 and Xcode 8.3.3 - duplicating function definitions is guaranteed to be wrong. Change 3588168 by Mark.Satterthwaite Move the Xcode version into the Metal shader format header, not the DDC key, so that we can handle bad compiler/driver combinations in the runtime and don't force all users to recompile every time the Xcode version changes. Change 3588192 by Rolando.Caloca DR - Fix d3d12 linker error when EXECUTE_DEBUG_COMMAND_LISTS is enabled Change 3588291 by Rolando.Caloca DR - Fix for d3d12 command list crash: Commited resources can not have aliasing barriers #jira UE-48299 Change 3590134 by Michael.Trepka Copy of CL 3578963 Reset automation tests timer after shader compilation when preparing for screenshots taking to make sure tests don't time out. Change 3590405 by Rolando.Caloca DR - hlslcc - support for sqrt(uint) Change 3590436 by Mark.Satterthwaite Rebuild Mac hlslcc for CL #3590405 - without the various compiler workarounds left over from before the recent code changes. Change 3590674 by Rolando.Caloca DR - vk - Integration from working branch - Fixes distance field maps - Compute pipelines stored in saved file - Adds GRHIRequiresRenderTargetForPixelShaderUAVs for platforms that need dummy render targets Change 3590699 by Rolando.Caloca DR - Fix distance fields mem leak Change 3590815 by Rolando.Caloca DR - vk - Fixes for uniform buffers and empty resource tables Change 3590818 by Mark.Satterthwaite Temporarily switch back to OpenVR v1.0.6 for Mac only until I can clarify what to do about a required but missing API hook for Metal. Re-enabled and fixed compile errors with Mac SteamVR plugin code. Change 3590905 by Mark.Satterthwaite For Metal shader compilation where the bytecode compiler is unavailable force the debug compiler flag and disable the archiving flag because storing text requires this. #jira UE-48163 Change 3590961 by Mark.Satterthwaite Submitted on Richard Wallis's behalf as he's on holiday: Mac fixes for Compute Skin Cache rendering issues (resulting in incorrect positions and tangents) and for recomputing tangents. Problem sampling from buffers/textures as floats with packed data. Some of the data appears as denorms so get flushed to zero then reinterpreted as uints via asuint or in Metal as_type<uint>(). Fix here for Metal seems to be to use uint types for the skin cache SRV's and as_type<> to floats instead. There could be some other areas where we're unpacking via floats that could affect Metal and I'm not sure how this will impact on other platforms. #jira UE-46688, UE-39256, UE-47215 Change 3590965 by Mark.Satterthwaite Remove the Z-bias workaround from Metal MRT as it isn't required and actually causes more problems. Change 3590969 by Mark.Satterthwaite Make all Metal shader platforms compile such that half may be used, unless the material specifies full precision. Change 3591871 by Rolando.Caloca DR - Enable PCSS on Vulkan & Metal - Enable capsule shadows on Vulkan Change 3592014 by Mark.Satterthwaite Remove support for Mac OS X El Capitan (10.11) including the stencil view workaround. Bump the minimum Metal shader standard for Metal SM4, SM5 & Metal MRT to v1.2 (macOS 10.12 Sierra & iOS 10) so we can use FMAs and other newer shader language features globally. Enable the new GRHIRequiresRenderTargetForPixelShaderUAVs flag as Metal is like Vulkan and needs a target for fragment rendering. Also fix the filename for direct-compile & remove the old batch file generation in the Metal shader compiler. Change 3592171 by Rolando.Caloca DR - CIS fix Change 3592753 by Jian.Ru repeat Daniel's fix on xb1 profilegpu crash (draw events cannot live beyond present) Change 3594595 by Rolando.Caloca DR - Fix D3D shader compiling run time stack corruption failure on debug triggering falsely Change 3594794 by Michael.Trepka Call FPlatformMisc::PumpMessages() before attempting to toggle fullscreen on Mac to fix an issue on some Macs running 10.13 beta that would ignore the toggle fullscreen call freezing the app Change 3594999 by Mark.Satterthwaite Disable MallocBinned2 for iOS as on Rhino it worked but on iOS 10.0.2 there are bugs (munmap uses 64kb granularity, not the 4096 the code expects given the reported page-size). While we are here remove the spurious FORCE_MALLOC_ANSI from the iOS platform header. #jira UE-48342 Change 3595004 by Mark.Satterthwaite Disable Metal's Deferred Store Actions and combined Depth/Stencil formats on iOS < 10.3 as there are bugs on earlier versions of iOS 10. #jira UE-48342 Change 3595386 by Mark.Satterthwaite Silence the deprecation warning for kIOSurfaceIsGlobal until SteamVR switches to one of the newer IOSurface sharing mechanisms. Change 3595394 by Rolando.Caloca DR - Added function for tracking down errors in the hlsl parser - Added support for simple #if 0...#endif Change 3599352 by Rolando.Caloca DR - Fixes for HlslParser - Added missing attributes for functions - Fixed nested assignment Change 3602440 by Michael.Trepka Fixed Metal shader compilation from Windows with remote compilation disabled #jira UE-48163 Change 3602898 by Chris.Bunner Resaving assets. Change 3603731 by Jian.Ru fix a crash caused by a material destroyed before the decal component #jira UE-48587 Change 3604629 by Rolando.Caloca DR - Workaround for PF_R8G8B8A8_UINT on Android #jira UE-48208 Change 3604984 by Peter.Sauerbrei fix for orientation not being limited to that specified in the plist #jira UE-48360 Change 3605738 by Chris.Bunner Allow functional screenshot tests to request a camera cut (e.g. tests relying on temporal aa history). #jira UE-48748 Change 3606009 by Mark.Satterthwaite Correctly implement ClipDistance for Metal as an array of floats as required by the spec. and fix a few irritating issues from the merge that should not have. - When compiling a tessellation vertex shader in the SCW direct mode we can't evaluate non-existant defines and we don't actually need to. - The define names, values & shader file name are irrelevant to the Metal output key, but the shader format name & Metal standard really do matter - should speed up Metal shader compilation a bit. - Move the shader vertex layer clip-distance to index 2 to avoid conflicts. - Don't default initialise the debug code string for Metal shaders or it won't print out the actual code.... #jira UE-47663 Change 3606108 by Mark.Satterthwaite Temporary hack to avoid a crash in AVPlayer. #jira UE-48758 Change 3606121 by Mark.Satterthwaite Fix Windows compilation. Change 3606992 by Chris.Bunner Static analysis fix. [CL 3608256 by Marcus Wassmer in Main branch]
2017-08-24 15:38:57 -04:00
virtual uint32 GetVersion(FName Format) const override final
{
// If there's no compiler on this machine, this is irrelevant so just return 0
if (!FMetalCompilerToolchain::Get()->IsCompilerAvailable())
{
return 0;
}
bool bUseFullMetalVersion = false;
EShaderPlatform ShaderPlatform = FMetalCompilerToolchain::MetalShaderFormatToLegacyShaderPlatform(Format);
if (FMetalCompilerToolchain::IsMobile(ShaderPlatform))
{
GConfig->GetBool(TEXT("/Script/IOSRuntimeSettings.IOSRuntimeSettings"), TEXT("UseFullMetalVersionInShaderVersion"), bUseFullMetalVersion, GEngineIni);
}
else
{
GConfig->GetBool(TEXT("/Script/MacTargetPlatform.MacTargetSettings"), TEXT("UseFullMetalVersionInShaderVersion"), bUseFullMetalVersion, GEngineIni);
}
FMetalCompilerToolchain::PackedVersion MetalVersionNumber = FMetalCompilerToolchain::Get()->GetCompilerVersion(ShaderPlatform);
uint16 HashValue = MetalVersionNumber.Major;
if (bUseFullMetalVersion)
{
// Use entire Metal version if .ini settings instruct us to do so (e.g. p4 dev build)
HashValue ^= MetalVersionNumber.Minor;
HashValue ^= MetalVersionNumber.Patch;
}
else
{
// Only use Metal major version (e.g. Installed build)
// Since Metal minor/patch version changes every Xcode minor version, we don't want users to rebuild shaders for every minor version update
}
uint32 Result = GetTypeHash(HashValue);
Result = HashCombine(Result, GetTypeHash(HLSLCC_VersionMinor));
Result = HashCombine(Result, GetTypeHash(UE_SHADER_METAL_VER));
return Result;
}
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
virtual void GetSupportedFormats(TArray<FName>& OutFormats) const override final
{
OutFormats.Add(NAME_SF_METAL);
OutFormats.Add(NAME_SF_METAL_MRT);
OutFormats.Add(NAME_SF_METAL_TVOS);
OutFormats.Add(NAME_SF_METAL_MRT_TVOS);
OutFormats.Add(NAME_SF_METAL_SM5);
OutFormats.Add(NAME_SF_METAL_SM6);
OutFormats.Add(NAME_SF_METAL_SIM);
Merging //UE4/Release-4.11 to //UE4/Main (Up to 2874930) #lockdown Nick.Penwarden ========================== MAJOR FEATURES + CHANGES ========================== Change 2868448 on 2016/02/16 by Mark.Satterthwaite Bring MetalRHI & MetalShaderFormat mostly up to Dev-Platform CL #2867146 to address JIRA UE-26181 and as the first part of addressing JIRA UE-23208. Only CL #2854142 has been omitted for compatibility with the 4.11 branch. #jira UE-26181 Change 2868454 on 2016/02/16 by Mark.Satterthwaite Shader changes necessary to properly fix Mobile Preview under Mac Metal (JIRA UE-23208) - Mac ES 3.1 doesn't support framebuffer fetch & is in fact a Mobile Emulation shader platform. #jira UE-23208 Change 2868650 on 2016/02/16 by Matthew.Griffin Allow Developer modules to be pre-compiled when the target is editor. #jira UE-26802 Change 2868859 on 2016/02/16 by Nick.Whiting Merging StereoPanorama fixes from Main to Release-4.11 (CL 2811839) #jira UE-25066 Change 2868927 on 2016/02/16 by Mieszko.Zielinski Fixed some regular-use crashes in LogVisualizer #UE4 #rb Lukasz.Furman #jira UE-27003 Change 2868994 on 2016/02/16 by Lina.Halper Fix refresh UI issue with remove joint #jira : UE-26529 #rb: Martin.Wilson Change 2868996 on 2016/02/16 by Lina.Halper Fix node stop working when negative value of LODThreshold #jira: UE-26828 #rb:Martin.Wilson Change 2868998 on 2016/02/16 by Lina.Halper Fix with crash when invalid index has entered #jira : UE-26715 #rb : Martin.Wilson Change 2869003 on 2016/02/16 by Ori.Cohen - Fix thread safety issue when cloth child collision or environment collision is used. - Fix cloth bounds growing when stale transform data is used. #JIRA OR-14990 #rb James.Golding Change 2869109 on 2016/02/16 by mason.seay Updated test assets for Restitution testing #jira UE-24473 Change 2869223 on 2016/02/16 by Taizyd.Korambayil #jira UE-19083 Disabled LOD on SkySphere BP Change 2869558 on 2016/02/16 by Dan.Oconnor Conservative fix for crash that occurs when adding a weak object ptr to an array of object ptrs in a blueprint #jira UE-25893 Change 2869891 on 2016/02/17 by Thomas.Sarkanen Fix crash when re-compiling anim BPs that are dependencies of 'parent' Blueprints Force a re-initialzation of nodes when initializing the whole anim instance. The bInitialized flag was intended as a runtime optimization - we assume our function will not change for the lifetime of the UAnimInstance. While parts of the graph will be re-initialized we dont need to re-acquire our UFunction ptrs (etc.) unless the whole instance is getting re-initialized. #rb Martin.Wilson #jira UE-26642 - Switch Skeletal Mesh node crashes the editor if the blueprint is compiled. Change 2869956 on 2016/02/17 by Tim.Hobson #Jira UE-26550 - Added three new icons for Arrow, Locked, and Unlocked per request from IanS. Change 2869965 on 2016/02/17 by Gareth.Martin Fix check() being hit when loading KiteDemo x1_y1 (and other old foliage maps) #jira UE-26930 Change 2870007 on 2016/02/17 by Richard.TalbotWatkin Merging from //UE4/Dev-Editor CL 2867609 Fixed auto-generation of unique object/package names when duplicating, so they generate the same names if they the source object was named the same as its package. #jira UE-25769 - Crash when Copy+Pasting numbered umaps in the content browser #RB Bob.Tellez Change 2870072 on 2016/02/17 by Michael.Schoell Level Blueprints will no longer preload newly created Blueprints that they reference when they are reloaded. #jira UE-23637 - Use of Blueprint delegate in sub-level script causes errors on startup Change 2870087 on 2016/02/17 by Matthew.Griffin Removed Android and IOS from list of platforms to build only for UnrealMatch3, so that they can be cooked and packaged correctly. Change 2870141 on 2016/02/17 by mason.seay Test content for Copy Pose From Mesh #jira UE-24473 Change 2870195 on 2016/02/17 by Mieszko.Zielinski PR #2052: Added missing Super::BeginPlay() (Contributed by Skylonxe) #jira UE-26915 Change 2870325 on 2016/02/17 by Steve.Robb [CL 2882948 by Matthew Griffin in Main branch]
2016-02-26 05:49:37 -05:00
OutFormats.Add(NAME_SF_METAL_MACES3_1);
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
OutFormats.Add(NAME_SF_METAL_MRT_MAC);
}
void CheckShaderFormat(FName Format) const
{
check(Format == NAME_SF_METAL
|| Format == NAME_SF_METAL_MRT
|| Format == NAME_SF_METAL_TVOS
|| Format == NAME_SF_METAL_MRT_TVOS
|| Format == NAME_SF_METAL_SM5
|| Format == NAME_SF_METAL_SM6
|| Format == NAME_SF_METAL_SIM
|| Format == NAME_SF_METAL_MACES3_1
|| Format == NAME_SF_METAL_MRT_MAC);
}
virtual bool PreprocessShader(
const FShaderCompilerInput& Input,
const FShaderCompilerEnvironment& Environment,
FShaderPreprocessOutput& PreprocessOutput) const override final
{
CheckShaderFormat(Input.ShaderFormat);
return PreprocessMetalShader(Input, Environment, PreprocessOutput);
}
virtual void CompilePreprocessedShader(
const FShaderCompilerInput& Input,
const FShaderPreprocessOutput& PreprocessOutput,
FShaderCompilerOutput& Output,
const FString& WorkingDirectory) const override final
{
CheckShaderFormat(Input.ShaderFormat);
CompileMetalShader(Input, PreprocessOutput, Output);
}
virtual void OutputDebugData(
const FShaderCompilerInput& Input,
const FShaderPreprocessOutput& PreprocessOutput,
const FShaderCompilerOutput& Output) const override final
{
OutputMetalDebugData(Input, PreprocessOutput, Output);
}
virtual bool CanStripShaderCode(bool const bNativeFormat) const override final
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
{
return CanCompileBinaryShaders() && bNativeFormat;
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
}
virtual bool StripShaderCode( TArray<uint8>& Code, FString const& DebugOutputDir, bool const bNative ) const override final
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
{
return StripShader_Metal(Code, DebugOutputDir, bNative);
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 3388261) #lockdown Nick.Penwarden #rb None ========================== MAJOR FEATURES + CHANGES ========================== Change 3358140 on 2017/03/22 by Rolando.Caloca DR - Fix copy to cube face - Compile fix when using dump layer - Add new error enum Change 3358301 on 2017/03/22 by Mitchell.Wilson Initial check in of LODs in InfiltratorForward. First pass on optimization in level. Adding a visibility track for SceneCapture2D in tunnel section. Change 3358477 on 2017/03/22 by Mitchell.Wilson Updating Skeletal Mesh DPW_Robot_Export to resolve screen size being too low for LOD1. Cleaned up LOD1 which was showing some visible popping when transitioning. Change 3358529 on 2017/03/22 by Mark.Satterthwaite Globally disable clang's "constant-logical-operand" warning when running under Distcc - it is much easier and less invasive than constantly fixing the code. Change 3358745 on 2017/03/22 by Mark.Satterthwaite Disable another warning (parentheses-equality) under Distcc because again the separation of preprocessing from compilation means it turns up where it isn't expected. Change 3358837 on 2017/03/22 by Joe.Graf Merge of pull request #3214 for the RenderDocPlugin #CodeReview: matt.kuhlenschmidt, marcus.wassmer #rb: marcus.wassmer Change 3359112 on 2017/03/22 by Ben.Salem Update perf monitor to include frame time by default. Also, use only game/PIE world timers when in editor, instead of all worlds combined. #tests Ran several Showdown test runs with plugin! Change 3359363 on 2017/03/22 by Joe.Graf First pass at non-unity & no pch compilation Change 3359449 on 2017/03/22 by Joe.Graf Added missing null check when exporting a EXR on Linux (UE-40268) #CodeReview: dmitry.rekman #rb: n/a Change 3360349 on 2017/03/23 by Guillaume.Abadie Fixes TAA's AA_FORCE_ALPHA_CLAMP causing DOF layouts. #jira UE-42920 Change 3360405 on 2017/03/23 by Marcus.Wassmer Better method for detecting Kepler Change 3360718 on 2017/03/23 by Daniel.Wright Planar reflections handle views smaller than the render target in a general way * Fixes planar reflections with adaptive pixel density (ViewFamily size larger than actual views combined) * Planar reflections are now supported in splitscreen Change 3360758 on 2017/03/23 by Daniel.Wright [Copy] Added new light property bCastVolumetricShadow, which defaults to true for directional and sky lights, but false for point / spot lights as supporting volumetric fog shadowing has significant GPU overhead Change 3360762 on 2017/03/23 by Daniel.Wright [Copy] Texture flags are now properly routed to RHICreateTexture3D from the render target pool Change 3360768 on 2017/03/23 by Daniel.Wright [Copy] Disabled GPUProfiler histogram by default, controlled by r.ProfileGPU.ShowEventHistogram Change 3360770 on 2017/03/23 by Daniel.Wright [Copy] Disabled fast clears on CustomDepth, saves .2ms on xbox Change 3360771 on 2017/03/23 by Daniel.Wright [Copy] Particle lights no longer force tiled deferred lighting. Tiled deferred lighting is only used if enough unshadowed lights + particle lights are on screen. Saves 1.5ms Xbox with one particle light. Change 3360774 on 2017/03/23 by Daniel.Wright [Copy] Distance field cvar comments Change 3360782 on 2017/03/23 by Daniel.Wright [Copy] Disabled selection color on Volume materials Change 3360795 on 2017/03/23 by Daniel.Wright [Copy] Volume materials now specify Albedo and Extinction, which is more intuitive than Scattering and Absorption. Albedo is [0-1] reflectance, while Extinction is a world space density. Change 3360799 on 2017/03/23 by Daniel.Wright [Copy] Cinematic scalability levels get 2x volumetric fog resolution in x and y Change 3360806 on 2017/03/23 by Daniel.Wright [Copy] Fixed volumetric fog being offset when viewport min is not 0 Change 3360809 on 2017/03/23 by Daniel.Wright [Copy] Volumetric fog now adds a bias to the inverse squared light falloff denominator, prevents extreme aliasing from the hotspot. Can be controlled with r.VolumetricFog.InverseSquaredLightDistanceBiasScale. Change 3361651 on 2017/03/23 by Brian.Karis Higher quality sharp SSR at quality 4 Change 3361678 on 2017/03/23 by Brian.Karis Fresnel darkens diffuse for clearcoat. Change 3361683 on 2017/03/23 by Brian.Karis Fixed SSR artifact Change 3361691 on 2017/03/23 by Brian.Karis Chagned min roughness limit Change 3361707 on 2017/03/23 by Brian.Karis Added inverse film tone map Change 3361726 on 2017/03/23 by Brian.Karis Better precision inverse Change 3361758 on 2017/03/23 by Brian.Karis Material flag normal curvature to roughness is no longer forward only. Change 3361765 on 2017/03/23 by Brian.Karis Update ACES Change 3361774 on 2017/03/23 by Brian.Karis Cleaned up alpha support and disabled screen edge clipping. Change 3362478 on 2017/03/24 by Guillaume.Abadie Cherry pick 3316084's PostProcessing.cpp: Fixes a bug in Circle DOF where the apply pass was no longer using the downres DOF's TAA output. #author Brian.Karis #jira UE-42920 Change 3362738 on 2017/03/24 by Rolando.Caloca DR - Hide scene capture on IF Change 3362890 on 2017/03/24 by Guillaume.Abadie Renames r.SceneAlpha to r.PostProcessing.PropagateAlpha Change 3363665 on 2017/03/24 by Mark.Satterthwaite PR #3414: Add command line option "-noheartbeatthread" to disable heart beat thread (Contributed by JeffRous) Change 3363866 on 2017/03/24 by Arne.Schober DR - Updated NVAPI #RB Marcus.Wassmer Change 3364300 on 2017/03/24 by Brian.Karis SSR use dynamic velocity Change 3364372 on 2017/03/24 by Brian.Karis Fix changing off axis projection velocities. Change 3364373 on 2017/03/24 by Brian.Karis Enabled velocity drawing in scene captures Change 3365531 on 2017/03/27 by Guillaume.Abadie Computes the material's screen position material expression directly from the pixel shader SvPosition Change 3365764 on 2017/03/27 by Chris.Bunner Lowering severity of crash for missing values in scalability.ini. #jira UE-41331 Change 3365916 on 2017/03/27 by Guillaume.Abadie Exposes the viewport offset within the view property material expression Change 3365979 on 2017/03/27 by Brian.Karis Fixed skylight intensity from double applying Change 3365987 on 2017/03/27 by Brian.Karis Stopped post process indirect lighting intensity from scaling skylight reflections Change 3365991 on 2017/03/27 by Brian.Karis Fix for static analysis Change 3366028 on 2017/03/27 by Daniel.Wright Volumetric fog supports static shadowing from Stationary lights * Using bilinear on static shadowmap depths + 1 PCF to smooth out results Change 3366029 on 2017/03/27 by Daniel.Wright Static shadow depth maps for Stationary point and spot lights are 2x higher res by default (4x more texels), which is more appropriate for volumetric fog Change 3366055 on 2017/03/27 by Guillaume.Abadie Cherry picks 3251469: Implements scene capture component's CaptureSortPriority to control GPU execution order in order to manage inter dependencies. Change 3366447 on 2017/03/27 by Simon.Tourangeau Fix IES light profile importer. - Bug in the LM-63-1986 format importer. Change 3366836 on 2017/03/27 by Brian.Karis ClearUAV now supports int types Change 3367435 on 2017/03/28 by Benjamin.Hyder Submitting Decal Automation map for initial approval Change 3367572 on 2017/03/28 by Chris.Bunner Changed ClampedPow {max(abs(x),0.00001)} to PositiveClampedPow {max(x,0)} to give more expected results to Power node in material graphs. #jira UE-42989 Change 3367756 on 2017/03/28 by Olaf.Piesche Niagara material usage flags Change 3367835 on 2017/03/28 by Marcus.Wassmer Fix crash when TileRenderer runs before anything else. Make explicit behavior when rendering at a time when there is no valid scene. Change 3367837 on 2017/03/28 by Marcus.Wassmer Missed a file. Change 3367838 on 2017/03/28 by Richard.Wallis Updated items from original shelved version by Mark Satt: - Added MetalBackend.cpp to change main function string to have an initial crc + code length zero's **Description below taken from Mark Satt's original verison of this in CL3343280** Updated for Dev-Rendering's PSOs & integrates Richard's work on RHI shader libraries. Replace the FShaderCache's cook-time binary shader cache with Dmitriy Dyomin's standalone FShaderCodeLibrary that saves all shader byte-code arrays to files named by the FSHAHash. This de-duplicates shaders so we only ever store the byte code once. Includes optional support for generating a platform specific library file - which Metal implements to provide a single Metal library. The platform-native implementation can perform more de-duplication and in the case of Metal has lower file overheads and will compress more efficiently. - All of the support code for the FShaderCache's cook caching is gone, which affects all platforms. The FShaderCodeLibrary is currently supported by Cook-By-The-Book but can be used with iterate or child cookers - only DLC cooking requires further work. - With further modifications it should be possible to support Cook-on-the-Fly as well (output directories would be needed in FShaderCodeLibrary::InitForCooking) and the file-access pattern should be changed to use async. IO so that Material loading is not considered complete until all required byte-code arrays are loaded into the FShaderCodeLibrary. - For Metal archiving shaders this way will compile with debug information and the FShaderCodeLibrary, with some help from extensions to IShaderFormat, will save the debug information out into separate files during cooking - these can then be used to debug the game without having to locally recompile, recook & repackage but the shipped byte-code is stripped. Global shader caches are also subject to de-duplication in the library in order to support Metal's shader stripping. - File Move operations need to respect the 'Replace' flag - for FShaderCodeLibrary to work we need Move to be atomic. - This bumps the object version and will cause all content to recook. - Native library support is optional - only Metal currently implements one, but so could Vulkan and D3D12. For Metal the big advantages are further de-duplication where different materials generate the same MetalSL text but a different FSHAHash, that the single Metal library has lower overhead and that as a single file it all compresses far better (esp. with LZMA - 5x smaller). Change 3367854 on 2017/03/28 by Mark.Satterthwaite Don't track or record draw call resources for non-OpenGL shader platforms in the shader-cache as it is unnecessary and makes it slower on the CPU than it needs to be. Change 3367877 on 2017/03/28 by Brian.Karis Fixed linux build hopefully Change 3368001 on 2017/03/28 by Mark.Satterthwaite Compile fixes from Richard's checkin caused by not having visibility to all platforms from my original shelves. Change 3368019 on 2017/03/28 by Mark.Satterthwaite And another fix for Windows compilation of MetalShaderFormat. Change 3368042 on 2017/03/28 by Mark.Satterthwaite And a couple of simpler MSVC errors. Change 3368271 on 2017/03/28 by Mark.Satterthwaite Make SceneRenderTargets compile again. Change 3368691 on 2017/03/28 by Daniel.Wright [Copy from BenW] Renamed r.Shadow.MaxCSMShadowResolution to r.Shadow.MaxCSMResolution to match scalability inis Change 3369689 on 2017/03/29 by Marcus.Wassmer Fix non editor compile for now Change 3369862 on 2017/03/29 by Marcus.Wassmer Get the rest of the things compiling again. Change 3369896 on 2017/03/29 by Chris.Bunner Enabling AMD HDR support by default. #jira UE-42113 Change 3370535 on 2017/03/29 by Marcus.Wassmer DR - Fix template explicit instantiation for ClearUAV permutations #RB Brian.Karis, Arne.Schober Change 3370704 on 2017/03/29 by Rolando.Caloca DR - Rewrote GPU Skin Cache - Per section buffers - Limited memory per non-editor worlds (control with r.SkinCache.SceneMemoryLimitInMB) Copied from 3370529 Change 3371389 on 2017/03/30 by Richard.Wallis Remove temp working directories after archive packages built. Change 3371641 on 2017/03/30 by Rolando.Caloca DR - Copy 3371640 (fix mem leak) Change 3372436 on 2017/03/30 by Uriel.Doyon Added flags in UPrimitiveComponent to keep track of its state in the streaming manager. This allows to avoid unnecessary callback and processing in begin destroy reattach and being destroy logic. Removed the limitation of only processing UMeshComponent when handling spawed primitive. This releases the level manager from having to manage dynamic primitives. This improves performance by not having to manage dynamic references in the level manager. Primitives managed as dynamic now have a callback when ever their proxy is udpated, handling many cases automatically where previously a manual callback to notify would have been required. Fixed an issue where primitives with no reference to streaming textures would loose they dynamic state because of lack of references in the streamer. Change 3372740 on 2017/03/30 by Chris.Bunner [Experimental] Partial compute post process pipeline (r.PostProcess.PreferCompute). StencilSceneTexture added to deferred list. A few known issues to be fixed in a follow-up CL. Change 3372765 on 2017/03/30 by Uriel.Doyon Disabled concurrent call to NotifyPrimitiveUpdated while we don't have a safe concurrent update Change 3372979 on 2017/03/30 by Richard.Hinckley #jira UE-43501 The stencil buffer can now use single-channel bitmasks that ignore depth. This makes it possible to detect overlaps between stencil objects. Change 3373053 on 2017/03/30 by Simon.Tourangeau LPV Fade support - mostly integrated from CL 2959511 Change 3373272 on 2017/03/30 by Uriel.Doyon Added support for the concurrent update of dynamic primitives by the streaming manager. Change 3373450 on 2017/03/30 by Rolando.Caloca DR - FNT - Fix bad data for odd texcoord channels used on skin cache passthrough factory Copy 3373364 #jira UE-43492 Change 3373470 on 2017/03/30 by Marcus.Wassmer Nvidia Aftermath support Change 3374187 on 2017/03/31 by Chris.Bunner Volume texture support for CombineLUTs/Tonemap compute pass. Refactored common param code to shared sub-class in CombineLUTs and Tonemap PS/CS. Skip compute post process out-of-bounds writes. Unsigned type conversion fixes. Trimmed compute post process shader inputs. Change 3374233 on 2017/03/31 by Chris.Bunner Removed several redundant post process compute fences and resource transitions. Added testing CVar to force compute post processes to async (r.PostProcess.ForceAsyncDispatch). Change 3374412 on 2017/03/31 by Rolando.Caloca DR - Fix static analysis Change 3374544 on 2017/03/31 by Richard.Wallis FShaderCache Parallel-Context-Aware Merged with FShaderCache Single Library. Future Work - This was done before Engine PSO were in so this now needs a refector in the recording and playback on pipeline states instead an emulate PSO in OpenGL Driver. - Remove FShaderCacheState and replace the logic with FGraphicsPipelineStateInitializer which should be able to record from the RHI current pipeline state - This would reduce the Locking required as it's naturally per thread/context and only the final record would need a lock Change 3374588 on 2017/03/31 by Richard.Wallis Windows Compile Fixes Change 3374810 on 2017/03/31 by Benjamin.Hyder updating recommended GPU drivers Change 3375207 on 2017/03/31 by Rolando.Caloca DR - vk - Fixed swapchain format selection for some Linux platforms Change 3375248 on 2017/03/31 by Rolando.Caloca DR - vk - Prefer D32S8 Change 3375495 on 2017/03/31 by Rolando.Caloca DR - vk - Update to sdk 1.0.42.2 Change 3375496 on 2017/03/31 by Rolando.Caloca DR - Force compiling with updated Vulkan SDK Change 3375636 on 2017/03/31 by Mark.Satterthwaite Copying Metal improvements from task stream, with some modifications: - Off-by-default implementations for MTLFence & MTLHeap, including some small changes to the RHI interface for parallel contexts. - Support for Apple's Instruments "Points of Interest" tool. - Consolidation of some Mac & iOS compiler, memory and thread handling code. - Fixes for Metal not having implicit buffer SRV typecasting for DistanceField effects. - Improvements to the internal FMetalDebug layer, still off by default. - Limited support for Xcode automatic code-signing for iOS/tvOS. - Minimisation of render-target changes in some rendering code, esp. SceneOcclusion, DBufferDecals. - Added RHISetResourceAliasability_RenderThread to FDynamicRHI for RHIs to implement simple render-target aliasing. - Added FApplePlatformObject, a custom block allocator for Objective-C types (with NSZombie support) which is now used in MetalRHI to decrease allocation costs of Objective-C types. - Smattering of lesser fixes. Change 3375654 on 2017/03/31 by Mark.Satterthwaite Incremental Windows build fix. Change 3375656 on 2017/04/01 by Mark.Satterthwaite Correct extern declaration, including the module export macro which Mac unhelpfully doesn't enforce (for now...). Change 3375797 on 2017/04/01 by Mark.Satterthwaite Nullability qualifiers to fix Mac build-farm compilation: perversely this is not a problem for local builds... Change 3375798 on 2017/04/01 by Mark.Satterthwaite Fix the first mis-merge in ParticleGpuSimulation - these changes clearly weren't properly resolved in the task-stream. Change 3375835 on 2017/04/01 by Mark.Satterthwaite Try again with nullability and fix the occlusion changes as the PSO work wasn't merged correctly. Change 3376143 on 2017/04/02 by Mark.Satterthwaite Switch back to flat dSYMs for Dev-Rendering - they don't work with Instruments etc. but they are required by our build system. Change 3376324 on 2017/04/03 by Chris.Bunner Fixed cvar re-registration log spam and flagged a testing-only cvar as such. Change 3376726 on 2017/04/03 by Benjamin.Hyder Submitting initial HDR test map (WIP) Change 3376756 on 2017/04/03 by Guillaume.Abadie Fixes scene captures ordering's backward compatibility. Before, 2d scene captures were rendered before cube scene captures. The CaptureSortPriority broke backward compatibility by settings this new member to 0 in the USceneCaptureComponent's constructor. Since it is a higher come first policy, this CL set the default of this value to 1 in USceneCaptureComponent2D's constructor. Change 3377378 on 2017/04/03 by Arne.Schober DR - Fix ShaderRecompiling over and over again #RB Chris.Bunner Change 3377512 on 2017/04/03 by Daniel.Wright [Copy] Fixed profilegpu in d3d12 - initialize FLongGPUTaskPS when it is safe to do so, and fixed FSlateRHIRenderer's incorrect usage of draw events Change 3377518 on 2017/04/03 by Daniel.Wright [Copy] Distance field atlas coalesces updates to reduce RHIUpdateTexture3D memory overhead on d3d12 Change 3377526 on 2017/04/03 by Daniel.Wright [Copy] "Ran out of GPU queries!" log only happens once Change 3377535 on 2017/04/03 by Daniel.Wright [Copy] Fixed unreferenced local variable Change 3377539 on 2017/04/03 by Daniel.Wright [Copy] Xbox One RHIGetResourceInfo takes ESRAM into account - fixes render target pool 'VRamInKB request failed' messages Change 3377546 on 2017/04/03 by Daniel.Wright [Copy] Added r.LightMaxDrawDistanceScale for local light scalability Change 3377553 on 2017/04/03 by Daniel.Wright [Copy] Removed NEW_ESRAM_ALLOCATOR define and old unused path Change 3377560 on 2017/04/03 by Daniel.Wright [Copy] Fixed two d3d12 refcounting bugs causing -norhithread crashes Change 3377565 on 2017/04/03 by Daniel.Wright [Copy] Fixed Xbox One deleting GPU resources before the GPU is done reading from them (GRHINeedsExtraDeletionLatency was false) Change 3377572 on 2017/04/03 by Daniel.Wright [Copy] Disabled point / spot lights with MaxDrawDistance on LowPC Change 3377586 on 2017/04/03 by Daniel.Wright Fixed compile error Change 3377699 on 2017/04/03 by David.Hill FFT Code. Moved over from raven and refactored #review-3374589 @guillaume.abadie Change 3377910 on 2017/04/03 by David.Hill GPU FFT: Fix Linux Build adding a missing template<> to an IMPLEMENT_SHADER_TYPE Change 3378751 on 2017/04/04 by Marcus.Wassmer HQ particle lights now spawn attached to the same socket as their parent module. Change 3378819 on 2017/04/04 by Richard.Wallis Should be no need to protect shader cache against RHI thread now. Change 3378823 on 2017/04/04 by Richard.Wallis FRHIShaderLibrary Opaque Type - Base FRHIShaderLibrary has no Create*Shader functions and is passed to Overloaded RHICreate*Shader functions instead of creation directly through the library. - Assumed that only Native libraries will end up in the RHICreate*Shader functions. - ShaderCache and ShaderCode Libraries now inherit from a common factory interface. Change 3378883 on 2017/04/04 by Arne.Schober DR - Fix DCC build Change 3378885 on 2017/04/04 by Richard.Wallis Metal resource cast compile fix post merge. Change 3378946 on 2017/04/04 by Chris.Bunner SM4 assert fix. Change 3378953 on 2017/04/04 by Chris.Bunner Fixed type-correctness on legacy BreakMA material nodes and set more flexible formats to global attributes which should result in much more forgiving graphs for users. Allowed material nodes to opt out of mask-based pin coloration. #tests Compiled most Paragon materials + QAGame test maps. #jira UE-39885 Change 3379189 on 2017/04/04 by Arne.Schober DR - Fix aftermath staging Change 3379229 on 2017/04/04 by Arne.Schober DR - Fix missing include Change 3379374 on 2017/04/04 by Mark.Satterthwaite Revert an accidentally merged change in MacPlatformProcess that relies on further changes from the Metal task stream. Change 3379505 on 2017/04/04 by Rolando.Caloca DR - Fix mismatched interpolators Change 3379539 on 2017/04/04 by Mark.Satterthwaite No FFT for any hlslcc platform - the IR for one or more RWTexture2D isn't quite right... #jira UE-43626 Change 3379561 on 2017/04/04 by Rolando.Caloca DR - Fix root signature issues on D3D12 PC Change 3379590 on 2017/04/04 by Mark.Satterthwaite Back out changelist 3379539 & change the shader slightly instead, the HLSLCC library generates bogus IR when you have an inout RWTexture. #jira UE-43626 Change 3379917 on 2017/04/04 by Uriel.Doyon Fix to input mismatch Change 3380578 on 2017/04/05 by Chris.Bunner Shader type fixes. #jira UE-43652 Change 3380639 on 2017/04/05 by Rolando.Caloca DR - Expose GetOrCreate PSO and document Change 3380821 on 2017/04/05 by Guillaume.Abadie Fixes a crash in USceneCaptureComponent::UpdateDeferredCaptures() #jira UE-43642 Change 3381092 on 2017/04/05 by Guillaume.Abadie Cherry pick 3362517: Implements TAA's scene color unpremultiplication from alpha channel to reduce DOF alpha channel temporal ghosting. This CL take the oportunity to transform AA_ALPHA to an compile time enumeration, and add a basic TAA compile time configuration validation to improve readability of the different TAA passes' configurations. Change 3381300 on 2017/04/05 by Mark.Satterthwaite Quick fix for changes to MetalRHI's render-thread safe texture creation not correctly handling AVFoundation video player handing us an IOSurface. #jira UE-43597 Change 3381359 on 2017/04/05 by Guillaume.Abadie Back out changelist 3381092 Change 3381421 on 2017/04/05 by Mark.Satterthwaite Amended CL #3380995 from Richard Wallis to address crash in the Material Editor under the validation layer - when there are no textures bound the default pass descriptor assigns store actions, which means we can't override them with our deferred store actions. #jira UE-43689 Change 3381422 on 2017/04/05 by Mark.Satterthwaite Absolute time queries can't be batched in Metal but I also can't rely on them being started with a call to BeginQuery - only EndQuery. #jira UE-43691 Change 3381503 on 2017/04/05 by Daniel.Wright More intuitive controls for Volumetric Fog * Removed ScatteringScale / AbsorptionScale on Exponential Height Fog and added Albedo / Extinction * InscatteringColorCubemap is now supported by Volumetric Fog * Particle lights have a default VolumetricScatteringIntensity of 0 to avoid trailing * Tweaked GVolumetricFogDepthDistributionScale better for nearby details * Volume Materials have twice the interpolators available Change 3381527 on 2017/04/05 by Mark.Satterthwaite Disable Private GPU storage for PVRTC texture formats on iOS Metal - these require more changes to the blit-encoder usage as PVRTC has strange requirements. Change 3381671 on 2017/04/05 by Mark.Satterthwaite Better error message for failure to compile shaders remotely from PC for Metal. Change 3381769 on 2017/04/05 by Rolando.Caloca DR - Added lock texture array 2d on Vulkan Change 3382003 on 2017/04/05 by Mark.Satterthwaite Remove the automatic Metal aliasing/re-use when releasing some resource types as it doesn't work as intended. Change 3382030 on 2017/04/05 by Zachary.Wilson Fix compiling Metal text shaders from PC broken in merge from task stream. #submitter mark.satterthwaite #jira UE-43652 Change 3382880 on 2017/04/06 by Mark.Satterthwaite Michael Trepka's CL #3379927: VolumetricFogVoxelization implementation for Mac Change 3383315 on 2017/04/06 by Mark.Satterthwaite Partially revert CL #3382003 - the emulated Metal heaps require invoking makeAliasable in order to reclaim memory. #jira UE-43739 Change 3384639 on 2017/04/07 by Marcus.Wassmer Move ShaderResource version bump to RenderingObjectVersion Change 3384704 on 2017/04/07 by Mark.Satterthwaite Compile fix for merge. Change 3384933 on 2017/04/07 by Rolando.Caloca DR - Fix skin cache crash with BP (copy 3384714) Change 3385104 on 2017/04/07 by Mark.Satterthwaite Fix MetalRHI's abs(int2) handling - it can't be translated to fabs(int2) as that won't compile. Also rebuild hlslcc for my sanity. #jira UE-43783 Change 3385105 on 2017/04/07 by Mark.Satterthwaite Force a shader rebuild to ensure that everybody picks up the fix for #jira UE-43783 #jira UE-43783 Change 3385118 on 2017/04/07 by Arne.Schober DR - [OR-37359] - Fix disapearing Decals when StencilLod Fade is enabled #RB none Change 3385149 on 2017/04/07 by Marcus.Wassmer Fix skincache motion blur Change 3385189 on 2017/04/07 by Rolando.Caloca DR - Fix swapchain format for editor on Vulkan Change 3385287 on 2017/04/07 by Mark.Satterthwaite Enable SM5 on Intel as of 10.12.4 and later. Change 3385347 on 2017/04/07 by Rolando.Caloca DR - Temp fix for GL4 corruption on editor #jira UE-43785 Change 3385363 on 2017/04/07 by Rolando.Caloca DR - Actually fix all win platforms for GL bug #jira UE-43785 Change 3385557 on 2017/04/07 by Arne.Schober DR - [UE-43205] - Fix mesh paint #RB none Change 3385608 on 2017/04/07 by Daniel.Wright Fixed SampleCmp being used on a non-depth texture, causing a d3d error Change 3385980 on 2017/04/10 by Rolando.Caloca DR - Remove transition functions RHIClearColor* RHIClearDepthStencilTexture Change 3386042 on 2017/04/10 by Rolando.Caloca DR - Fix metal merge issue Change 3386157 on 2017/04/10 by Rolando.Caloca DR - Remove VS2013 libs generation off hlslcc & glslang (to match main) Change 3386356 on 2017/04/10 by Chris.Bunner Resolving merge errors. Change 3386414 on 2017/04/10 by Chris.Bunner Resolved merge issue in RendererScene.cpp. Change 3386700 on 2017/04/10 by Mark.Satterthwaite Silence documentation warnings. Change 3387178 on 2017/04/10 by Chris.Bunner Removed invalid mask correction on MakeMA material nodes. Change 3388177 on 2017/04/11 by Marcus.Wassmer Disable ensure that is no longer relevant now that we bind clear colors on texture creation Change 3388261 on 2017/04/11 by Chris.Bunner Static analysis fix. [CL 3388266 by Chris Bunner in Main branch]
2017-04-11 10:32:07 -04:00
}
Copying //UE4/Dev-Mobile to //UE4/Dev-Main (Source: //UE4/Dev-Mobile @ 3627362) #lockdown Nick.Penwarden #rb None ============================ MAJOR FEATURES & CHANGES ============================ Change 3600639 by Nick.Shin HTML5 remove old emscripten toolchain forgot to remove this (was checked out in another changelist) #jira UE-47813 Change 3600641 by Nick.Shin HTML5 TM-Core crash fixes new PhysX HTML5 libs #jira UE-47813 Index Out Of Bounds crash running "GC and Level Load Stress Test" in TM-Core on Firefox Change 3600644 by Nick.Shin HTML5 TM-Core crash fixes emscripten doesn't seem to know how to look at <PxRigidActor> from the PxActor class... #jira UE-47813 Index Out Of Bounds crash running "GC and Level Load Stress Test" in TM-Core on Firefox Change 3600647 by Nick.Shin HTML5 UInterpTrackInst::GetGroupActor() pre-null check #jira UE-47485 ( Pri:1 - 4.18 ) Crash running Stat Command test in TM-Core on Firefox Change 3601439 by Dmitriy.Dyomin Fixed: Protostar leads to device reboot in Galaxy S7 Adreno device. (SM-G930V) #jira UE-48323 Change 3601440 by Dmitriy.Dyomin Fixed: Texture streaming after world origin was rebased Change 3601456 by Dmitriy.Dyomin Added: Async loading bytecode from shader library - Shader library will be packed into single file - Shader library will be created only when packaging project by the book #jira UEMOB-381 Change 3601624 by Jack.Porter Fix landscape crash on HTML5 Exception !IsInRenderingThread() failed. #jira UE-48527 Change 3603890 by Dmitriy.Dyomin GitHub 3905 : Engine crash in LandscapeComponent when streaming levels #3905 #jira UE-48422 Change 3603933 by Dmitriy.Dyomin Fixed: Crash after Splash Screen on Android (ETC2) when Adding r.UseShaderCaching and r.UseShaderPredraw - Added r.SaveShaderCache command to save current cache on demand - Removed support for caching multiple platfroms at the same time, each platform now uses separate cache/file - Significantly reduced size of draw log on disk - Mobile platfroms support only basic caching, logging shaders and bound shader states without full gfx state #jira UE-47553 Change 3604050 by Sorin.Gradinaru #jira UE-47428 Android virtual keyboard polishing Done: Multiline should be disabled when not needed Hide suggestions and autocorrect "Done" / back button behavior to make it feel natural - Done/Enter and the Back key event sent to the engine If the keyboard is up and you click on the same control you're currently inputting into, it will hide the keyboard. Look into animation - not critical. S6 with Swiftkey. The numbers don't show up in our textedit but the string is updated (we see dots in the Slate control) with the password entry Could not reproduce/test: Make sure we don't use negative coords for the input box if the keyboard is at the top of the screen Change 3604081 by Allan.Bentham Reduce redundant log spam from SustainedPerformanceMode on android. Change 3604152 by Allan.Bentham Improved vulkanRHI availability and selection reporting. Change 3604186 by Dmitriy.Dyomin Vulkan: Write to buffers directly on UMA devices (no staging) Change 3604396 by Nick.Shin HTML5 - stats and multi-threading checks more null & multi-threading - and some functions flat out disabled for HTML5 platform the big "don't use on HTML5" is TLockFreeFixedSizeAllocator_TLSCacheBase's manual TLS.PartialBundle memory handler... i'm not sure it's working properly... switched on USE_NIEVE_TLockFreeFixedSizeAllocator_TLSCacheBase sections of code (i.e. basic malloc and free) for PLATFORM_HTML5 - it seems only the stats function was exacerbating the memory bug ... but, shutting this completely out for HTML5... don't know if it's an emscripten compiler/corruption that's causing this... - will send this to emscripten makers as another test case for them to help see what's going on... #jira UE-47485 ( Pri:1 - 4.18 ) Crash running Stat Command test in TM-Core on Firefox Change 3604752 by Allan.Bentham Reduce battery status log spam. Change 3604825 by Nick.Shin HTML5 emscripten 1.37.19 updated license file #jira UE-47813 Change 3606486 by Jack.Porter Enabled refraction on tvOS #jira UE-47229 Change 3606546 by Dmitriy.Dyomin Vulkan: Missed null check from CL# 3601439 Change 3606654 by Allan.Bentham mobile post process shaders will not attempt depth buffer fetch, instead they will always read from the depth texture. #jira UE-41919 Change 3606672 by Dmitriy.Dyomin Fixed: Vulkan mode crashes without error on Tegra K1 Nvidia Shield due to OOM - fixed r.MobileReduceLoadedMips has no effect - added r.MobileMaxLoadedMips to set a limit to a number of mips - reduced size of vulkan allocation pages on android #jira UE-42838 Change 3607204 by Allan.Bentham Do not attempt getprocaddress for GL_EXT_DEBUG_LABEL functions when the extension is not declared. Change 3607214 by Nick.Shin HTML5 - stats font crash fix #jira UE-47485 ( Pri:1 - 4.18 ) Crash running Stat Command test in TM-Core on Firefox Change 3609164 by Chris.Babcock More informative Android SDK license requirements and errors #jra UE-48837 #ue4 #android Change 3609175 by Chris.Babcock Linux build fix #jira none Change 3609516 by Dmitriy.Dyomin Fixed: Decals in TM-ShaderModels appearing and disappearing when viewed at certain angles on Android #jira UE-45342 Change 3609618 by Dmitriy.Dyomin Back out changelist 3603933, to help merge from main Change 3610979 by Peter.Sauerbrei addition of asset catalogs for icons on Xcode 9 for iOS 11 Change 3612145 by Dmitriy.Dyomin Resubmitting CL #3603933 (shader cache fixes) Change 3613981 by Chris.Babcock Fix issue with Android password keyboard input #jira WEX-7343 #ue4 #android Change 3614375 by Sorin.Gradinaru #jira UE-44656 StrategyGame Crashes on launch on KindleFire 5th Gen. Kindle Fire HD7 5th (CPU Mali 450) seems to have problems with thread rendering, crashing when returning to the main thread. Added new device profile [Android_Mali_4xx_KindleFire] with a new cvar r.AndroidDisableThreadedRenderingFirstLoad=1, disabling movie player rendering on the intial screen. For subsequent loading screens, the (same) code apparently runs Ok. A warning "Initial loading screen disabled from BaseDeviceProfiles.ini: r.AndroidDisableThreadedRenderingFirstLoad=1" should appear when the thread rendering is disabled Change 3614971 by Cosmin.Sulea UE-46769 - GitHub 3745 : bForcePVRTC4 was ignored. #jira UE-46769 Change 3616431 by Peter.Sauerbrei PR3657,3658 - fixes for InApp purchase bugs courtesy of nverenik Change 3617306 by Jack.Porter Fix issue where undocked tabs had no way to be made visible again when hidden #jira UE-12044 Change 3617312 by Jack.Porter Find a new best MobileDirectionalLights[channel] when then current one is removed from the world (eg by a level streaming operation) #jira UE-47135 #3785 Change 3617383 by Dmitriy.Dyomin Vk - only dynamic buffers in HostVisible on UMA devices Change 3617437 by Dmitriy.Dyomin Vk - SRV allocates one BufferView for each buffer in FVulkanResourceMultiBuffer, so they can be reused (contributed by Samsung) Change 3617474 by Dmitriy.Dyomin Vk - Don't use fences on image acquire on Android Change 3617483 by Dmitriy.Dyomin Vk - Remove redundant dynamic state setup (viewport, scissor, stencilref) (contributed by Samsung) Change 3617521 by Dmitriy.Dyomin Fix CIS warnings Change 3617574 by Dmitriy.Dyomin Vk - Optimized RenderPass pool and Framebuffer pool (reduced vkObject count) (contributed by Samsung) ProtoStar: Framebuffer count : 133 -> 108, Renderpass count : 21 -> 18 Change 3617585 by Dmitriy.Dyomin compile fix for CL# 3617574 Change 3617849 by Allan.Bentham Log UnsatisfiedLinkError's content when loadlibrary fails. Change 3617945 by Chris.Babcock Allow UPL variable expansion in addPermission, addFeature, and addLibrary #jira UE-47421 #ue4 #android Change 3618097 by Allan.Bentham Fix Y axis switch with android GLES when rendering triangles to canvas. #jira UE-44510 Change 3618733 by Peter.Sauerbrei fix from Dev-Rendering for tvOS shader compilation Change 3618761 by Peter.Sauerbrei fix for shader crash on startup on iOS Change 3618769 by Peter.Sauerbrei bump metal shader guid to force a rebuild of shaders Change 3620061 by Peter.Sauerbrei fix for resource directory on Asset catalogs #jira UE-49074 Change 3620520 by Peter.Sauerbrei remove the texture warning, the logic was incorrect and in the end we don't need the warning #jira UE-49057 Change 3621811 by Allan.Bentham Add mipindex and array slice index to framebuffer hashing code. #jira UE-49171 Change 3624410 by Jack.Porter Fix issue where the Shared Material Native Libraries checkbox causes packaging for Android to fail #jira UE-49105 Change 3627361 by Jack.Porter Fixing case on iOS files #jira None Change 3627362 by Jack.Porter Fixed case on IOS files #jira None [CL 3627373 by Jack Porter in Main branch]
2017-09-06 01:04:25 -04:00
virtual bool SupportsShaderArchives() const override
{
return CanCompileBinaryShaders();
Copying //UE4/Dev-Mobile to //UE4/Dev-Main (Source: //UE4/Dev-Mobile @ 3627362) #lockdown Nick.Penwarden #rb None ============================ MAJOR FEATURES & CHANGES ============================ Change 3600639 by Nick.Shin HTML5 remove old emscripten toolchain forgot to remove this (was checked out in another changelist) #jira UE-47813 Change 3600641 by Nick.Shin HTML5 TM-Core crash fixes new PhysX HTML5 libs #jira UE-47813 Index Out Of Bounds crash running "GC and Level Load Stress Test" in TM-Core on Firefox Change 3600644 by Nick.Shin HTML5 TM-Core crash fixes emscripten doesn't seem to know how to look at <PxRigidActor> from the PxActor class... #jira UE-47813 Index Out Of Bounds crash running "GC and Level Load Stress Test" in TM-Core on Firefox Change 3600647 by Nick.Shin HTML5 UInterpTrackInst::GetGroupActor() pre-null check #jira UE-47485 ( Pri:1 - 4.18 ) Crash running Stat Command test in TM-Core on Firefox Change 3601439 by Dmitriy.Dyomin Fixed: Protostar leads to device reboot in Galaxy S7 Adreno device. (SM-G930V) #jira UE-48323 Change 3601440 by Dmitriy.Dyomin Fixed: Texture streaming after world origin was rebased Change 3601456 by Dmitriy.Dyomin Added: Async loading bytecode from shader library - Shader library will be packed into single file - Shader library will be created only when packaging project by the book #jira UEMOB-381 Change 3601624 by Jack.Porter Fix landscape crash on HTML5 Exception !IsInRenderingThread() failed. #jira UE-48527 Change 3603890 by Dmitriy.Dyomin GitHub 3905 : Engine crash in LandscapeComponent when streaming levels #3905 #jira UE-48422 Change 3603933 by Dmitriy.Dyomin Fixed: Crash after Splash Screen on Android (ETC2) when Adding r.UseShaderCaching and r.UseShaderPredraw - Added r.SaveShaderCache command to save current cache on demand - Removed support for caching multiple platfroms at the same time, each platform now uses separate cache/file - Significantly reduced size of draw log on disk - Mobile platfroms support only basic caching, logging shaders and bound shader states without full gfx state #jira UE-47553 Change 3604050 by Sorin.Gradinaru #jira UE-47428 Android virtual keyboard polishing Done: Multiline should be disabled when not needed Hide suggestions and autocorrect "Done" / back button behavior to make it feel natural - Done/Enter and the Back key event sent to the engine If the keyboard is up and you click on the same control you're currently inputting into, it will hide the keyboard. Look into animation - not critical. S6 with Swiftkey. The numbers don't show up in our textedit but the string is updated (we see dots in the Slate control) with the password entry Could not reproduce/test: Make sure we don't use negative coords for the input box if the keyboard is at the top of the screen Change 3604081 by Allan.Bentham Reduce redundant log spam from SustainedPerformanceMode on android. Change 3604152 by Allan.Bentham Improved vulkanRHI availability and selection reporting. Change 3604186 by Dmitriy.Dyomin Vulkan: Write to buffers directly on UMA devices (no staging) Change 3604396 by Nick.Shin HTML5 - stats and multi-threading checks more null & multi-threading - and some functions flat out disabled for HTML5 platform the big "don't use on HTML5" is TLockFreeFixedSizeAllocator_TLSCacheBase's manual TLS.PartialBundle memory handler... i'm not sure it's working properly... switched on USE_NIEVE_TLockFreeFixedSizeAllocator_TLSCacheBase sections of code (i.e. basic malloc and free) for PLATFORM_HTML5 - it seems only the stats function was exacerbating the memory bug ... but, shutting this completely out for HTML5... don't know if it's an emscripten compiler/corruption that's causing this... - will send this to emscripten makers as another test case for them to help see what's going on... #jira UE-47485 ( Pri:1 - 4.18 ) Crash running Stat Command test in TM-Core on Firefox Change 3604752 by Allan.Bentham Reduce battery status log spam. Change 3604825 by Nick.Shin HTML5 emscripten 1.37.19 updated license file #jira UE-47813 Change 3606486 by Jack.Porter Enabled refraction on tvOS #jira UE-47229 Change 3606546 by Dmitriy.Dyomin Vulkan: Missed null check from CL# 3601439 Change 3606654 by Allan.Bentham mobile post process shaders will not attempt depth buffer fetch, instead they will always read from the depth texture. #jira UE-41919 Change 3606672 by Dmitriy.Dyomin Fixed: Vulkan mode crashes without error on Tegra K1 Nvidia Shield due to OOM - fixed r.MobileReduceLoadedMips has no effect - added r.MobileMaxLoadedMips to set a limit to a number of mips - reduced size of vulkan allocation pages on android #jira UE-42838 Change 3607204 by Allan.Bentham Do not attempt getprocaddress for GL_EXT_DEBUG_LABEL functions when the extension is not declared. Change 3607214 by Nick.Shin HTML5 - stats font crash fix #jira UE-47485 ( Pri:1 - 4.18 ) Crash running Stat Command test in TM-Core on Firefox Change 3609164 by Chris.Babcock More informative Android SDK license requirements and errors #jra UE-48837 #ue4 #android Change 3609175 by Chris.Babcock Linux build fix #jira none Change 3609516 by Dmitriy.Dyomin Fixed: Decals in TM-ShaderModels appearing and disappearing when viewed at certain angles on Android #jira UE-45342 Change 3609618 by Dmitriy.Dyomin Back out changelist 3603933, to help merge from main Change 3610979 by Peter.Sauerbrei addition of asset catalogs for icons on Xcode 9 for iOS 11 Change 3612145 by Dmitriy.Dyomin Resubmitting CL #3603933 (shader cache fixes) Change 3613981 by Chris.Babcock Fix issue with Android password keyboard input #jira WEX-7343 #ue4 #android Change 3614375 by Sorin.Gradinaru #jira UE-44656 StrategyGame Crashes on launch on KindleFire 5th Gen. Kindle Fire HD7 5th (CPU Mali 450) seems to have problems with thread rendering, crashing when returning to the main thread. Added new device profile [Android_Mali_4xx_KindleFire] with a new cvar r.AndroidDisableThreadedRenderingFirstLoad=1, disabling movie player rendering on the intial screen. For subsequent loading screens, the (same) code apparently runs Ok. A warning "Initial loading screen disabled from BaseDeviceProfiles.ini: r.AndroidDisableThreadedRenderingFirstLoad=1" should appear when the thread rendering is disabled Change 3614971 by Cosmin.Sulea UE-46769 - GitHub 3745 : bForcePVRTC4 was ignored. #jira UE-46769 Change 3616431 by Peter.Sauerbrei PR3657,3658 - fixes for InApp purchase bugs courtesy of nverenik Change 3617306 by Jack.Porter Fix issue where undocked tabs had no way to be made visible again when hidden #jira UE-12044 Change 3617312 by Jack.Porter Find a new best MobileDirectionalLights[channel] when then current one is removed from the world (eg by a level streaming operation) #jira UE-47135 #3785 Change 3617383 by Dmitriy.Dyomin Vk - only dynamic buffers in HostVisible on UMA devices Change 3617437 by Dmitriy.Dyomin Vk - SRV allocates one BufferView for each buffer in FVulkanResourceMultiBuffer, so they can be reused (contributed by Samsung) Change 3617474 by Dmitriy.Dyomin Vk - Don't use fences on image acquire on Android Change 3617483 by Dmitriy.Dyomin Vk - Remove redundant dynamic state setup (viewport, scissor, stencilref) (contributed by Samsung) Change 3617521 by Dmitriy.Dyomin Fix CIS warnings Change 3617574 by Dmitriy.Dyomin Vk - Optimized RenderPass pool and Framebuffer pool (reduced vkObject count) (contributed by Samsung) ProtoStar: Framebuffer count : 133 -> 108, Renderpass count : 21 -> 18 Change 3617585 by Dmitriy.Dyomin compile fix for CL# 3617574 Change 3617849 by Allan.Bentham Log UnsatisfiedLinkError's content when loadlibrary fails. Change 3617945 by Chris.Babcock Allow UPL variable expansion in addPermission, addFeature, and addLibrary #jira UE-47421 #ue4 #android Change 3618097 by Allan.Bentham Fix Y axis switch with android GLES when rendering triangles to canvas. #jira UE-44510 Change 3618733 by Peter.Sauerbrei fix from Dev-Rendering for tvOS shader compilation Change 3618761 by Peter.Sauerbrei fix for shader crash on startup on iOS Change 3618769 by Peter.Sauerbrei bump metal shader guid to force a rebuild of shaders Change 3620061 by Peter.Sauerbrei fix for resource directory on Asset catalogs #jira UE-49074 Change 3620520 by Peter.Sauerbrei remove the texture warning, the logic was incorrect and in the end we don't need the warning #jira UE-49057 Change 3621811 by Allan.Bentham Add mipindex and array slice index to framebuffer hashing code. #jira UE-49171 Change 3624410 by Jack.Porter Fix issue where the Shared Material Native Libraries checkbox causes packaging for Android to fail #jira UE-49105 Change 3627361 by Jack.Porter Fixing case on iOS files #jira None Change 3627362 by Jack.Porter Fixed case on IOS files #jira None [CL 3627373 by Jack Porter in Main branch]
2017-09-06 01:04:25 -04:00
}
virtual bool CreateShaderArchive(FString const& LibraryName,
FName ShaderFormatAndShaderPlatformName,
const FString& WorkingDirectory,
const FString& OutputDir,
const FString& DebugOutputDir,
const FSerializedShaderArchive& InSerializedShaders,
const TArray<TArray<uint8>>& ShaderCode,
TArray<FString>* OutputFiles) const override final
{
const int32 NumShadersPerLibrary = 10000;
check(LibraryName.Len() > 0);
TArray<FString> Components;
FString ShaderPlatform = ShaderFormatAndShaderPlatformName.ToString();
ShaderPlatform.ParseIntoArray(Components, TEXT("-"));
check(Components.Num() == 2);
FName ShaderFormatName(Components[0]);
check(ShaderFormatName == NAME_SF_METAL || ShaderFormatName == NAME_SF_METAL_MRT || ShaderFormatName == NAME_SF_METAL_TVOS || ShaderFormatName == NAME_SF_METAL_MRT_TVOS || ShaderFormatName == NAME_SF_METAL_SM5 || ShaderFormatName == NAME_SF_METAL_SM6 || ShaderFormatName == NAME_SF_METAL_SIM || ShaderFormatName == NAME_SF_METAL_MACES3_1 || ShaderFormatName == NAME_SF_METAL_MRT_MAC);
const FString ArchivePath = (WorkingDirectory / ShaderFormatAndShaderPlatformName.GetPlainNameString());
IFileManager::Get().DeleteDirectory(*ArchivePath, false, true);
IFileManager::Get().MakeDirectory(*ArchivePath);
FSerializedShaderArchive SerializedShaders(InSerializedShaders);
check(SerializedShaders.GetNumShaders() == ShaderCode.Num());
TArray<uint8> StrippedShaderCode;
TArray<uint8> TempShaderCode;
TArray<TSet<uint64>> SubLibraries;
for (int32 ShaderIndex = 0; ShaderIndex < SerializedShaders.GetNumShaders(); ++ShaderIndex)
{
SerializedShaders.DecompressShader(ShaderIndex, ShaderCode, TempShaderCode);
StripShader_Metal(TempShaderCode, DebugOutputDir, true);
uint64 ShaderId = AppendShader_Metal(ArchivePath, SerializedShaders.ShaderHashes[ShaderIndex], TempShaderCode);
uint32 LibraryIndex = ShaderIndex / NumShadersPerLibrary;
if (ShaderId)
{
if (SubLibraries.Num() <= (int32)LibraryIndex)
{
SubLibraries.Add(TSet<uint64>());
}
SubLibraries[LibraryIndex].Add(ShaderId);
}
FShaderCodeEntry& ShaderEntry = SerializedShaders.ShaderEntries[ShaderIndex];
ShaderEntry.Size = TempShaderCode.Num();
ShaderEntry.UncompressedSize = TempShaderCode.Num();
StrippedShaderCode.Append(TempShaderCode);
}
SerializedShaders.Finalize();
bool bOK = false;
FString LibraryPlatformName = FString::Printf(TEXT("%s_%s"), *LibraryName, *ShaderFormatAndShaderPlatformName.GetPlainNameString());
LibraryPlatformName.ToLowerInline();
volatile int32 CompiledLibraries = 0;
TArray<FGraphEventRef> Tasks;
for (uint32 Index = 0; Index < (uint32)SubLibraries.Num(); Index++)
{
TSet<uint64>& PartialShaders = SubLibraries[Index];
FString LibraryPath = (OutputDir / LibraryPlatformName) + FString::Printf(TEXT(".%d"), Index) + FMetalCompilerToolchain::MetalLibraryExtension;
if (OutputFiles)
{
OutputFiles->Add(LibraryPath);
}
// Enqueue the library compilation as a task so we can go wide
FGraphEventRef CompletionFence = FFunctionGraphTask::CreateAndDispatchWhenReady([ShaderFormatName, ArchivePath, LibraryPath, PartialShaders, DebugOutputDir, &CompiledLibraries]()
{
if (FinalizeLibrary_Metal(ShaderFormatName, ArchivePath, LibraryPath, PartialShaders, DebugOutputDir))
{
FPlatformAtomics::InterlockedIncrement(&CompiledLibraries);
}
}, TStatId(), NULL, ENamedThreads::AnyThread);
Tasks.Add(CompletionFence);
}
#if WITH_ENGINE
FGraphEventRef DebugDataCompletionFence = FFunctionGraphTask::CreateAndDispatchWhenReady([ShaderFormatAndShaderPlatformName, OutputDir, LibraryPlatformName, DebugOutputDir]()
{
//TODO add a check in here - this will only work if we have shader archiving with debug info set.
//We want to archive all the metal shader source files so that they can be unarchived into a debug location
//This allows the debugging of optimised metal shaders within the xcode tool set
//Currently using the 'tar' system tool to create a compressed tape archive
//Place the archive in the same position as the .metallib file
FString CompressedDir = (OutputDir / TEXT("../MetaData/ShaderDebug/"));
IFileManager::Get().MakeDirectory(*CompressedDir, true);
FString CompressedPath = (CompressedDir / LibraryPlatformName) + TEXT(".zip");
IPlatformFile& PlatformFile = FPlatformFileManager::Get().GetPlatformFile();
IFileHandle* ZipFile = PlatformFile.OpenWrite(*CompressedPath);
if (ZipFile)
{
FZipArchiveWriter* ZipWriter = new FZipArchiveWriter(ZipFile);
//Find the metal source files
TArray<FString> FilesToArchive;
IFileManager::Get().FindFilesRecursive(FilesToArchive, *DebugOutputDir, TEXT("*.metal"), true, false, false);
//Write the local file names into the target file
const FString DebugDir = DebugOutputDir / *ShaderFormatAndShaderPlatformName.GetPlainNameString();
for (FString FileName : FilesToArchive)
{
TArray<uint8> FileData;
FFileHelper::LoadFileToArray(FileData, *FileName);
FPaths::MakePathRelativeTo(FileName, *DebugDir);
ZipWriter->AddFile(FileName, FileData, FDateTime::Now());
}
delete ZipWriter;
ZipWriter = nullptr;
}
else
{
UE_LOG(LogShaders, Error, TEXT("Failed to create Metal debug .zip output file \"%s\". Debug .zip export will be disabled."), *CompressedPath);
}
}, TStatId(), NULL, ENamedThreads::AnyThread);
Tasks.Add(DebugDataCompletionFence);
#endif // WITH_ENGINE
// Wait for tasks
for (auto& Task : Tasks)
{
FTaskGraphInterface::Get().WaitUntilTaskCompletes(Task);
}
if (CompiledLibraries == SubLibraries.Num())
{
FString BinaryShaderFile = (OutputDir / LibraryPlatformName) + FMetalCompilerToolchain::MetalMapExtension;
BinaryShaderFile.ToLowerInline();
FArchive* BinaryShaderAr = IFileManager::Get().CreateFileWriter(*BinaryShaderFile);
if (BinaryShaderAr != NULL)
{
FMetalShaderLibraryHeader Header;
Header.Format = ShaderFormatName.GetPlainNameString();
Header.NumLibraries = SubLibraries.Num();
Header.NumShadersPerLibrary = NumShadersPerLibrary;
*BinaryShaderAr << Header;
*BinaryShaderAr << SerializedShaders;
*BinaryShaderAr << StrippedShaderCode;
BinaryShaderAr->Flush();
delete BinaryShaderAr;
if (OutputFiles)
{
OutputFiles->Add(BinaryShaderFile);
}
bOK = true;
}
}
return bOK;
//Map.Format = Format.GetPlainNameString();
}
[shaders] shader format preprocessing cleanup & refactoring - 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]
2023-11-30 15:56:34 -05:00
virtual void ModifyShaderCompilerInput(FShaderCompilerInput& Input) const override
{
// Work out which standard we need, this is dependent on the shader platform.
// TODO: Read from toolchain class
const bool bIsMobile = FMetalCompilerToolchain::Get()->IsMobile((EShaderPlatform)Input.Target.Platform);
if (bIsMobile)
{
Input.Environment.SetDefine(TEXT("IOS"), 1);
}
else
{
Input.Environment.SetDefine(TEXT("MAC"), 1);
}
Input.Environment.SetDefine(TEXT("COMPILER_METAL"), 1);
if (Input.ShaderFormat == NAME_SF_METAL || Input.ShaderFormat == NAME_SF_METAL_TVOS)
{
Input.Environment.SetDefine(TEXT("METAL_PROFILE"), 1);
}
else if (Input.ShaderFormat == NAME_SF_METAL_SIM)
{
Input.Environment.SetDefine(TEXT("METAL_PROFILE"), 1);
}
else if (Input.ShaderFormat == NAME_SF_METAL_MRT || Input.ShaderFormat == NAME_SF_METAL_MRT_TVOS)
{
Input.Environment.SetDefine(TEXT("METAL_MRT_PROFILE"), 1);
}
else if (Input.ShaderFormat == NAME_SF_METAL_MACES3_1)
{
Input.Environment.SetDefine(TEXT("METAL_PROFILE"), 1);
}
else if (Input.ShaderFormat == NAME_SF_METAL_SM5)
{
Input.Environment.SetDefine(TEXT("METAL_SM5_PROFILE"), 1);
Input.Environment.SetDefine(TEXT("USING_VERTEX_SHADER_LAYER"), 1);
}
else if (Input.ShaderFormat == NAME_SF_METAL_SM6)
{
Input.Environment.SetDefine(TEXT("METAL_SM6_PROFILE"), 1);
Input.Environment.SetDefine(TEXT("USING_VERTEX_SHADER_LAYER"), 1);
}
else if (Input.ShaderFormat == NAME_SF_METAL_MRT_MAC)
{
Input.Environment.SetDefine(TEXT("METAL_MRT_PROFILE"), 1);
}
Input.Environment.SetDefine(TEXT("COMPILER_HLSLCC"), 2);
#if UE_METAL_USE_METAL_SHADER_CONVERTER
const bool bUseMetalShaderConverter = Input.Target.GetPlatform() == EShaderPlatform::SP_METAL_SM6
&& RHIGetBindlessSupport(EShaderPlatform::SP_METAL_SM6) != ERHIBindlessSupport::Unsupported;
if (bUseMetalShaderConverter)
{
Input.Environment.SetDefine(TEXT("COMPILER_METAL_SHADER_CONVERTER"), 1);
}
#endif
#if !UE_METAL_USE_METAL_SHADER_CONVERTER
[shaders] shader format preprocessing cleanup & refactoring - 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]
2023-11-30 15:56:34 -05:00
if (Input.Environment.FullPrecisionInPS || (IsValidRef(Input.SharedEnvironment) && Input.SharedEnvironment->FullPrecisionInPS))
{
Input.Environment.SetDefine(TEXT("FORCE_FLOATS"), (uint32)1);
}
#else
// We can use 16bits types with Msc (since we do not use the frontend).
if (Input.Environment.CompilerFlags.Contains(CFLAG_AllowRealTypes))
{
Input.Environment.SetDefine(TEXT("PLATFORM_SUPPORTS_REAL_TYPES"), 1);
}
#endif
[shaders] shader format preprocessing cleanup & refactoring - 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]
2023-11-30 15:56:34 -05:00
if (Input.Environment.CompilerFlags.Contains(CFLAG_AvoidFlowControl)
|| Input.Environment.CompilerFlags.Contains(CFLAG_PreferFlowControl))
{
Input.Environment.SetDefine(TEXT("COMPILER_SUPPORTS_ATTRIBUTES"), (uint32)0);
}
else
{
Input.Environment.SetDefine(TEXT("COMPILER_SUPPORTS_ATTRIBUTES"), (uint32)1);
}
bool bUsesInlineRayTracing = Input.Environment.CompilerFlags.Contains(CFLAG_InlineRayTracing);
if (bUsesInlineRayTracing)
{
Input.Environment.SetDefine(TEXT("PLATFORM_SUPPORTS_INLINE_RAY_TRACING"), 1);
}
Input.Environment.SetDefine(TEXT("COMPILER_SUPPORTS_DUAL_SOURCE_BLENDING_SLOT_DECORATION"), (uint32)1);
}
virtual bool CanCompileBinaryShaders() const override final
{
#if PLATFORM_MAC
return FPlatformMisc::IsSupportedXcodeVersionInstalled();
#else
return FMetalCompilerToolchain::Get()->IsCompilerAvailable();
#endif
}
Copying //UE4/Dev-Rendering to //UE4/Dev-Main (Source: //UE4/Dev-Rendering @ 4358666) #lockdown Nick.Penwarden ============================ MAJOR FEATURES & CHANGES ============================ Change 4073167 by Krzysztof.Narkowicz Added subsurface profile for eye shading model. #jira none Change 4073422 by Krzysztof.Narkowicz Added dual specular for subsurface profile shading model. #jira none Change 4075278 by Krzysztof.Narkowicz Fixed forward reflection/refraction rendering issues, which caused ShaderModels.Material.Refraction to fail. #jira none Change 4084231 by Krzysztof.Narkowicz Dual specular - replace lobe spread with two separate roughness multipliers. Default material roughness is now replaced by an average lobe roughness in order to support non dual specular features. #jira none Change 4092798 by Matt.Collins Some HDR refactoring. Previously the DisplayOutput and ColorGamut were only set in GameUserSettings. I added a Sink that checks the HDR enable. If it's toggled we apply the correct DisplayOutput and ColorGamut for the current platform (this way we get good settings even if you toggle via the console). These settings are still exposed via the console and can be set independently if the user wants. Change 4096954 by Chris.Bunner Added ShaderModelID as scene texture option and renamed existing value to ShaderModelColor to better reflect the internal code. Change 4111285 by Brian.Karis Eye shading update. Added Iris normal (disabled). Removed wrap. Fixed contact shadows. Change 4155261 by Krzysztof.Narkowicz Planar reflection prefilter - use scene viewport size instead of reflection target size in order to keep filter size constant in screen space. This makes planar reflection filter more stable in case of dynamic resolution. #jira none Change 4167644 by Krzysztof.Narkowicz Global shader map is now stored in multiple DDC entries (one per shader filename) instead of keeping everything in a single one. This allows to skip recompilation of unchanged shader files. Change 4183727 by Yuriy.ODonnell Implemented auto-conversion from deferred to DBuffer decals in forward shading mode (when GBuffer is not available). Added support for specular and metallic channels for DBuffer decals, based on work by Chris Bunner. This requires DBufferC to be expanded from 2 to 4 channels, leading to slight increase in DBuffer bandwidth and memory requirements. Appearance of DBuffer decals is affected by this change, as specular and metallic channel values previously hard-coded in DBufferDecalShared.ush. Decals were forced to be non-metallic and have specular of 4% (0.5 numeric value). Now the authored decal material values will be used, which matches GBuffer decals. Added support for DBuffer decals with emissive component. Most decal types can now be automatically converted, with the exception of stain decals. Those are currently approximated as regular translucent decals. Change 4197684 by laz.matech Added a PostProcess Volume test to the map to test that Cinematic Depth of View can be achieved through PPVs as well. Changed the BP_DepthOfFeildPOV asset - I exposed Focus Method so that it can be disabled for the PPV test. Added a second Hair Model head to the InFocusHair test so that it tests in and out of focus hair models (changed the name of the test to FocusHair). #jira none Change 4225614 by Rolando.Caloca DR - Enable depth collision particles on Vulkan mobile Change 4235489 by Uriel.Doyon Removed r.DefaultFeature.PointLightUnits and r.DefaultFeature.SpotLightUnits and replaced them by a single r.DefaultFeature.LightUnits which also controls the units of newly placed rect lights. #jira UE-59525 Change 4260154 by Mark.Satterthwaite Parallelize the creation of Metal archives and libraries when they are broken up into smaller sub-libraries, this should reduce apparent cook time by going wide across threads on the host of the cooker. Change 4270594 by Brian.Karis Fix for textured rect light L pointing away from plane due to approximate diffuse integration. Change 4273361 by Daniel.Wright Particle Cutouts with 8 verts now always use stochastic approach. Circle textures with > 234 edges in the convex hull were overflowing the uint64 calculation of the total number of combinations, causing an infinite loop. Change 4309174 by Mark.Satterthwaite Graph device utilization from the driver monitor stats - really helps see how well the GPU is being used. Change 4310121 by Matt.Collins Optmizing RemoveUniformBuffersFromSource. Brings it from ~20% to ~1.5% in my testing. #jira none Change 4312960 by Daniel.Wright Fix from Stephen Hill for incorrect light grid culling near the near plane Change 4314169 by Richard.Wallis FShaderCache and associated public structures are now marked as deprecated. All FShaderCache code hooks removed from MetalRHI, OpenGLDrv and engine Launch/Shutdown logic. #jira none Change 4320760 by Arne.Schober DR - Remove SV_Coverage from basepass interpolants when running with Masked in early Depth with ForwardShading as otherwise earlyZ will be disabled (as the PS has to run). #jira UE-60992 Change 4334607 by Uriel.Doyon Added custom overrides to reset ULightComponent::Intensity to default (in FLightComponentDetails). Now settings a light Intesity to default resets the brightness to the archetype brightness. This handles correctly cases where the intensity units differs between the two objects. Also changed FLocalLightComponentDetails so that changing intensity units kepts the same brightness (by recomputing the Intensity). #jira UE-61401 Change 4336188 by Rolando.Caloca DR - Added -ReduceThreadUsage so programs can use less threads (for SCW ) Change 4337967 by Rolando.Caloca DR - Remove unused RHISupportsShaderCompression function #rb none [CL 4358751 by Rolando Caloca in Main branch]
2018-09-11 14:44:10 -04:00
virtual const TCHAR* GetPlatformIncludeDirectory() const
{
return TEXT("Metal");
}
};
/**
* Module for Metal shaders
*/
static IShaderFormat* Singleton = nullptr;
class FMetalShaderFormatModule : public IShaderFormatModule
{
public:
virtual ~FMetalShaderFormatModule()
{
Singleton = nullptr;
}
virtual IShaderFormat* GetShaderFormat()
{
return Singleton;
}
virtual void StartupModule() override
{
Singleton = new FMetalShaderFormat();
}
virtual void ShutdownModule() override
{
delete Singleton;
Singleton = nullptr;
}
};
IMPLEMENT_MODULE(FMetalShaderFormatModule, MetalShaderFormat);
static FMetalCompilerToolchain::EMetalToolchainStatus ParseCompilerVersionAndTarget(const FString& OutputOfMetalDashV, FString& VersionString, FMetalCompilerToolchain::PackedVersion& PackedVersionNumber, FMetalCompilerToolchain::PackedVersion& PackedTargetNumber)
{
/*
Output of metal -v might look like this:
Apple LLVM version 902.11 (metalfe-902.11.1)
Target: air64-apple-darwin19.5.0
Thread model: posix
InstalledDir: C:\Program Files\Metal Developer Tools\ios\bin
*/
TArray<FString> Lines;
OutputOfMetalDashV.ParseIntoArrayLines(Lines);
int32 VersionLineIndex = 0;
{
for (int32 Index = 0; Index < Lines.Num(); ++Index)
{
if (Lines[Index].StartsWith(TEXT("Apple ")) && Lines[Index].Contains(TEXT(" version ")) && Lines[Index].EndsWith(TEXT(")")))
{
VersionLineIndex = Index;
break;
}
}
VersionString = Lines[VersionLineIndex];
FString& Version = Lines[VersionLineIndex];
check(!Version.IsEmpty());
int32 Major = 0, Minor = 0;
int32 NumResults = 0;
#if !PLATFORM_WINDOWS
char AppleToolName[PATH_MAX] = { '\0' };
char SupplementaryVersionName[PATH_MAX] = { '\0' };
NumResults = sscanf(TCHAR_TO_ANSI(*Version), "Apple %s version %d.%d (metalfe-%s)", AppleToolName, &Major, &Minor, SupplementaryVersionName);
#else
TCHAR AppleToolName[WINDOWS_MAX_PATH] = { '\0' };
TCHAR SupplementaryVersionName[WINDOWS_MAX_PATH] = { '\0' };
NumResults = swscanf_s(*Version, TEXT("Apple %ls version %d.%d (metalfe-%ls)"), AppleToolName, WINDOWS_MAX_PATH, &Major, &Minor, SupplementaryVersionName, WINDOWS_MAX_PATH);
#endif
if (NumResults != 4)
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Metal version string format unrecoginzed"));
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Expecting: Apple LLVM version 902.11 (metalfe-902.11.1)"));
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Obtained: %s"), *Version);
}
PackedVersionNumber.Major = Major;
PackedVersionNumber.Minor = Minor;
// The version name in brackets is too irregular to extract a useful patch version
// Sometimes (metalfe-31001.667.2), sometimes (metalfe-31001.643.2.1), sometimes (metalfe-31001.362-windows)
PackedVersionNumber.Patch = 0;
}
if (PackedVersionNumber.Version == 0)
{
return FMetalCompilerToolchain::EMetalToolchainStatus::CouldNotParseCompilerVersion;
}
{
FString& FormatVersion = Lines[VersionLineIndex + 1];
int32 Major = 0, Minor = 0, Patch = 0;
int32 NumResults = 0;
#if !PLATFORM_WINDOWS
NumResults = sscanf(TCHAR_TO_ANSI(*FormatVersion), "Target: air64-apple-darwin%d.%d.%d", &Major, &Minor, &Patch);
#else
NumResults = swscanf_s(*FormatVersion, TEXT("Target: air64-apple-darwin%d.%d.%d"), &Major, &Minor, &Patch);
#endif
PackedTargetNumber.Major = Major;
PackedTargetNumber.Minor = Minor;
PackedTargetNumber.Patch = Patch;
}
if (PackedTargetNumber.Version == 0)
{
return FMetalCompilerToolchain::EMetalToolchainStatus::CouldNotParseTargetVersion;
}
return FMetalCompilerToolchain::EMetalToolchainStatus::Success;
}
static FMetalCompilerToolchain::EMetalToolchainStatus ParseLibraryToolpath(const FString& OutputOfMetalSearchDirs, FString& LibraryPath)
{
static FString LibraryPrefix(TEXT("libraries: =%s"));
TArray<FString> Lines;
OutputOfMetalSearchDirs.ParseIntoArrayLines(Lines);
{
int32 LibrariesLineIndex = 0;
for (int32 Index = 0; Index < Lines.Num(); ++Index)
{
if (Lines[Index].StartsWith(TEXT("libraries: =")))
{
LibrariesLineIndex = Index;
break;
}
}
FString& LibraryLine = Lines[LibrariesLineIndex];
LibraryPath = LibraryLine.RightChop(LibraryPrefix.Len());
if (!FPaths::DirectoryExists(LibraryPath))
{
return FMetalCompilerToolchain::EMetalToolchainStatus::CouldNotFindMetalStdLib;
}
FPaths::Combine(LibraryPath, TEXT("include"), TEXT("metal"));
if (!FPaths::DirectoryExists(LibraryPath))
{
return FMetalCompilerToolchain::EMetalToolchainStatus::CouldNotFindMetalStdLib;
}
}
return FMetalCompilerToolchain::EMetalToolchainStatus::Success;
}
FMetalCompilerToolchain* FMetalCompilerToolchain::Singleton = nullptr;
FString FMetalCompilerToolchain::MetalExtention(TEXT(".metal"));
FString FMetalCompilerToolchain::MetalLibraryExtension(TEXT(".metallib"));
FString FMetalCompilerToolchain::MetalObjectExtension(TEXT(".air"));
#if PLATFORM_WINDOWS
FString FMetalCompilerToolchain::MetalFrontendBinary(TEXT("metal.exe"));
FString FMetalCompilerToolchain::MetalArBinary(TEXT("metal-ar.exe"));
FString FMetalCompilerToolchain::MetalLibraryBinary(TEXT("metallib.exe"));
FString FMetalCompilerToolchain::AirPackBinary(TEXT("air-pack.exe"));
#else
FString FMetalCompilerToolchain::MetalFrontendBinary(TEXT("metal"));
FString FMetalCompilerToolchain::MetalArBinary(TEXT("metal-ar"));
FString FMetalCompilerToolchain::MetalLibraryBinary(TEXT("metallib"));
FString FMetalCompilerToolchain::AirPackBinary(TEXT("air-pack"));
#endif
FString FMetalCompilerToolchain::MetalMapExtension(TEXT(".metalmap"));
FString FMetalCompilerToolchain::XcrunPath(TEXT("/usr/bin/xcrun"));
FString FMetalCompilerToolchain::MetalMacSDK(TEXT("macosx"));
FString FMetalCompilerToolchain::MetalMobileSDK(TEXT("iphoneos"));
FString FMetalCompilerToolchain::DefaultWindowsToolchainPath(TEXT("c:/Program Files/Metal Developer Tools"));
// Static methods
void FMetalCompilerToolchain::CreateAndInit()
{
Singleton = new FMetalCompilerToolchain;
Singleton->Init();
}
void FMetalCompilerToolchain::Destroy()
{
Singleton->Teardown();
delete Singleton;
Singleton = nullptr;
}
EShaderPlatform FMetalCompilerToolchain::MetalShaderFormatToLegacyShaderPlatform(FName ShaderFormat)
{
if (ShaderFormat == NAME_SF_METAL) return SP_METAL;
if (ShaderFormat == NAME_SF_METAL_MRT) return SP_METAL_MRT;
if (ShaderFormat == NAME_SF_METAL_TVOS) return SP_METAL_TVOS;
if (ShaderFormat == NAME_SF_METAL_MRT_TVOS) return SP_METAL_MRT_TVOS;
if (ShaderFormat == NAME_SF_METAL_MRT_MAC) return SP_METAL_MRT_MAC;
if (ShaderFormat == NAME_SF_METAL_SM5) return SP_METAL_SM5;
if (ShaderFormat == NAME_SF_METAL_SM6) return SP_METAL_SM6;
if (ShaderFormat == NAME_SF_METAL_SIM) return SP_METAL_SIM;
if (ShaderFormat == NAME_SF_METAL_MACES3_1) return SP_METAL_MACES3_1;
return SP_NumPlatforms;
}
// Instance methods
FMetalCompilerToolchain::PackedVersion FMetalCompilerToolchain::GetCompilerVersion(EShaderPlatform Platform) const
{
if (this->IsMobile(Platform))
{
return this->MetalCompilerVersion[AppleSDKMobile];
}
return this->MetalCompilerVersion[AppleSDKMac];
}
FMetalCompilerToolchain::PackedVersion FMetalCompilerToolchain::GetTargetVersion(EShaderPlatform Platform) const
{
if (this->IsMobile(Platform))
{
return this->MetalTargetVersion[AppleSDKMobile];
}
return this->MetalTargetVersion[AppleSDKMac];
}
const FString& FMetalCompilerToolchain::GetCompilerVersionString(EShaderPlatform Platform) const
{
if (this->IsMobile(Platform))
{
return this->MetalCompilerVersionString[AppleSDKMobile];
}
return this->MetalCompilerVersionString[AppleSDKMac];
}
void FMetalCompilerToolchain::Init()
{
bToolchainAvailable = false;
bToolchainBinariesPresent = false;
bSkipPCH = true;
#if PLATFORM_MAC
EMetalToolchainStatus Result = DoMacNativeSetup();
#else
EMetalToolchainStatus Result = DoWindowsSetup();
#endif
if (Result != EMetalToolchainStatus::Success)
{
#if CHECK_METAL_COMPILER_TOOLCHAIN_SETUP
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Metal compiler not found. Shaders will be stored as text."));
#endif
bToolchainAvailable = false;
}
else
{
Result = FetchCompilerVersion();
if (Result != EMetalToolchainStatus::Success)
{
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Could not parse compiler version."));
}
Result = FetchMetalStandardLibraryPath();
if (Result != EMetalToolchainStatus::Success)
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Could not parse metal_stdlib path. Will not use PCH."));
bSkipPCH = true;
// This is not really an error since we can compile without the PCH just fine.
Result = EMetalToolchainStatus::Success;
}
else
{
// This is forced off for now. If we wish to re-enable it a lot of testing should be done.
//bSkipPCH = false;
}
bToolchainAvailable = true;
}
#if CHECK_METAL_COMPILER_TOOLCHAIN_SETUP
if (Result == EMetalToolchainStatus::Success)
{
check(IsCompilerAvailable());
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Metal toolchain setup complete."));
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Using Local Metal compiler"));
if (!MetalFrontendBinaryCommand[AppleSDKMac].IsEmpty())
{
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Mac metalfe found at %s"), *MetalFrontendBinaryCommand[AppleSDKMac]);
}
if (!MetalFrontendBinaryCommand[AppleSDKMobile].IsEmpty())
{
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Mobile metalfe found at %s"), *MetalFrontendBinaryCommand[AppleSDKMobile]);
}
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Mac metalfe version %s"), *MetalCompilerVersionString[AppleSDKMac]);
UE_LOG(LogMetalCompilerSetup, Log, TEXT("Mobile metalfe version %s"), *MetalCompilerVersionString[AppleSDKMobile]);
}
else
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Failed to set up Metal toolchain. See log above. Shaders will not be compiled offline."));
}
#endif
}
void FMetalCompilerToolchain::Teardown()
{
// remove temporaries
if (this->LocalTempFolder.IsEmpty())
{
return;
}
if (!FPaths::DirectoryExists(this->LocalTempFolder))
{
return;
}
bool bSuccess = IFileManager::Get().DeleteDirectory(*this->LocalTempFolder, false, true);
if (!bSuccess)
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Could not delete temporary %s"), *this->LocalTempFolder);
}
}
FMetalCompilerToolchain::EMetalToolchainStatus FMetalCompilerToolchain::FetchCompilerVersion()
{
EMetalToolchainStatus Result = EMetalToolchainStatus::Success;
{
int32 ReturnCode = 0;
FString StdOut;
// metal -v writes its output to stderr
// But the underlying (windows) implementation of CreateProc puts everything into one pipe, which is written to StdOut.
bool bResult = this->ExecMetalFrontend(AppleSDKMac, TEXT("-v --target=air64-apple-darwin18.7.0"), &ReturnCode, &StdOut, &StdOut);
check(bResult);
if (ReturnCode > 0)
{
return EMetalToolchainStatus::CouldNotParseCompilerVersion;
}
Result = ParseCompilerVersionAndTarget(StdOut, this->MetalCompilerVersionString[AppleSDKMac], this->MetalCompilerVersion[AppleSDKMac], this->MetalTargetVersion[AppleSDKMac]);
if (Result != EMetalToolchainStatus::Success)
{
return Result;
}
}
{
int32 ReturnCode = 0;
FString StdOut;
// metal -v writes its output to stderr
bool bResult = this->ExecMetalFrontend(AppleSDKMobile, TEXT("-v --target=air64-apple-darwin18.7.0"), &ReturnCode, &StdOut, &StdOut);
check(bResult);
if (ReturnCode > 0)
{
return EMetalToolchainStatus::CouldNotParseCompilerVersion;
}
Result = ParseCompilerVersionAndTarget(StdOut, this->MetalCompilerVersionString[AppleSDKMobile], this->MetalCompilerVersion[AppleSDKMobile], this->MetalTargetVersion[AppleSDKMobile]);
}
return Result;
}
FMetalCompilerToolchain::EMetalToolchainStatus FMetalCompilerToolchain::FetchMetalStandardLibraryPath()
{
// if we've already decided to skip compiling a PCH we don't need this path at all.
if (this->bSkipPCH)
{
return EMetalToolchainStatus::Success;
}
EMetalToolchainStatus Result = EMetalToolchainStatus::Success;
{
int32 ReturnCode = 0;
FString StdOut, StdErr;
bool bResult = this->ExecMetalFrontend(AppleSDKMac, TEXT("--print-search-dirs"), &ReturnCode, &StdOut, &StdErr);
check(bResult);
Result = ParseLibraryToolpath(StdOut, this->MetalStandardLibraryPath[AppleSDKMac]);
if (Result != EMetalToolchainStatus::Success)
{
return Result;
}
}
{
int32 ReturnCode = 0;
FString StdOut, StdErr;
bool bResult = this->ExecMetalFrontend(AppleSDKMobile, TEXT("--print-search-dirs"), &ReturnCode, &StdOut, &StdErr);
check(bResult);
Result = ParseLibraryToolpath(StdOut, this->MetalStandardLibraryPath[AppleSDKMobile]);
}
return Result;
}
#if PLATFORM_MAC
FMetalCompilerToolchain::EMetalToolchainStatus FMetalCompilerToolchain::DoMacNativeSetup()
{
FString ToolchainBase;
if (FParse::Value(FCommandLine::Get(), TEXT("-MetalToolchainOverride="), ToolchainBase))
{
const bool bUseOverride = (!ToolchainBase.IsEmpty() && FPaths::DirectoryExists(ToolchainBase));
if (bUseOverride)
{
MetalFrontendBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / MetalFrontendBinary;
MetalFrontendBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / MetalFrontendBinary;
const bool bIsFrontendPresent = FPaths::FileExists(MetalFrontendBinaryCommand[AppleSDKMac]) && FPaths::FileExists(MetalFrontendBinaryCommand[AppleSDKMobile]);
if (!bIsFrontendPresent)
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Missing Metal frontend in %s."), *ToolchainBase);
return EMetalToolchainStatus::ToolchainNotFound;
}
MetalArBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / MetalArBinary;
MetalArBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / MetalArBinary;
MetalLibBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / MetalLibraryBinary;
MetalLibBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / MetalLibraryBinary;
AirPackBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / AirPackBinary;
AirPackBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / AirPackBinary;
if (!FPaths::FileExists(MetalArBinaryCommand[AppleSDKMac]) ||
!FPaths::FileExists(MetalArBinaryCommand[AppleSDKMobile]) ||
!FPaths::FileExists(MetalLibBinaryCommand[AppleSDKMac]) ||
!FPaths::FileExists(MetalLibBinaryCommand[AppleSDKMobile]) ||
!FPaths::FileExists(MetalLibBinaryCommand[AppleSDKMobile]) ||
!FPaths::FileExists(AirPackBinaryCommand[AppleSDKMac]) ||
!FPaths::FileExists(AirPackBinaryCommand[AppleSDKMobile]))
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Missing toolchain binaries in %s."), *ToolchainBase);
return EMetalToolchainStatus::ToolchainNotFound;
}
this->bToolchainBinariesPresent = true;
return EMetalToolchainStatus::Success;
}
}
int32 ReturnCode = 0;
FString StdOut, StdErr;
bool bSuccess = this->ExecGenericCommand(*XcrunPath, *FString::Printf(TEXT("--sdk %s --find %s"), *this->MetalMacSDK, *this->MetalFrontendBinary), &ReturnCode, &StdOut, &StdErr);
bSuccess |= FPaths::FileExists(StdOut);
if(!bSuccess || ReturnCode > 0)
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Missing Mac Metal toolchain (macos SDK not found)."));
return EMetalToolchainStatus::ToolchainNotFound;
}
bSuccess = this->ExecGenericCommand(*XcrunPath, *FString::Printf(TEXT("--sdk %s --find %s"), *this->MetalMobileSDK, *this->MetalFrontendBinary), &ReturnCode, &StdOut, &StdErr);
bSuccess |= FPaths::FileExists(StdOut);
if(!bSuccess || ReturnCode > 0)
{
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Missing Mobile Metal toolchain (iphoneos SDK not found)."));
return EMetalToolchainStatus::ToolchainNotFound;
}
this->bToolchainBinariesPresent = true;
return EMetalToolchainStatus::Success;
}
#endif
#if PLATFORM_WINDOWS
FMetalCompilerToolchain::EMetalToolchainStatus FMetalCompilerToolchain::DoWindowsSetup()
{
int32 Result = 0;
FString ToolchainBase;
GConfig->GetString(TEXT("/Script/IOSRuntimeSettings.IOSRuntimeSettings"), TEXT("WindowsMetalToolchainOverride"), ToolchainBase, GEngineIni);
const bool bUseOverride = (!ToolchainBase.IsEmpty() && FPaths::DirectoryExists(ToolchainBase));
if (!bUseOverride)
{
ToolchainBase = DefaultWindowsToolchainPath;
}
// Look for the windows native toolchain
MetalFrontendBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / MetalFrontendBinary;
MetalFrontendBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / MetalFrontendBinary;
bool bUseLocalMetalToolchain = FPaths::FileExists(MetalFrontendBinaryCommand[AppleSDKMac]) && FPaths::FileExists(MetalFrontendBinaryCommand[AppleSDKMobile]);
if (!bUseLocalMetalToolchain)
{
#if CHECK_METAL_COMPILER_TOOLCHAIN_SETUP
UE_LOG(LogMetalCompilerSetup, Display, TEXT("Searching for Metal toolchain, but it doesn't appear to be installed."));
UE_LOG(LogMetalCompilerSetup, Display, TEXT("Searched for %s and %s"), *MetalFrontendBinaryCommand[AppleSDKMac], *MetalFrontendBinaryCommand[AppleSDKMobile]);
#endif
return EMetalToolchainStatus::ToolchainNotFound;
}
MetalArBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / MetalArBinary;
MetalArBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / MetalArBinary;
MetalLibBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / MetalLibraryBinary;
MetalLibBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / MetalLibraryBinary;
AirPackBinaryCommand[AppleSDKMac] = ToolchainBase / TEXT("macos") / TEXT("bin") / AirPackBinary;
AirPackBinaryCommand[AppleSDKMobile] = ToolchainBase / TEXT("ios") / TEXT("bin") / AirPackBinary;
if (!FPaths::FileExists(MetalArBinaryCommand[AppleSDKMac]) ||
!FPaths::FileExists(MetalArBinaryCommand[AppleSDKMobile]) ||
!FPaths::FileExists(MetalLibBinaryCommand[AppleSDKMac]) ||
!FPaths::FileExists(MetalLibBinaryCommand[AppleSDKMobile]) ||
!FPaths::FileExists(AirPackBinaryCommand[AppleSDKMac]) ||
!FPaths::FileExists(AirPackBinaryCommand[AppleSDKMobile]))
{
#if CHECK_METAL_COMPILER_TOOLCHAIN_SETUP
UE_LOG(LogMetalCompilerSetup, Warning, TEXT("Missing toolchain binaries."))
#endif
return EMetalToolchainStatus::ToolchainNotFound;
}
this->bToolchainBinariesPresent = true;
return EMetalToolchainStatus::Success;
}
#endif
bool FMetalCompilerToolchain::ExecMetalFrontend(EAppleSDKType SDK, const TCHAR* Parameters, int32* OutReturnCode, FString* OutStdOut, FString* OutStdErr) const
{
check(this->bToolchainBinariesPresent);
#if PLATFORM_MAC
if (this->MetalFrontendBinaryCommand[SDK].IsEmpty())
{
FString BuiltParams = FString::Printf(TEXT("--sdk %s %s %s"), *SDKToString(SDK), *this->MetalFrontendBinary, Parameters);
return ExecGenericCommand(*XcrunPath, *BuiltParams, OutReturnCode, OutStdOut, OutStdErr);
}
else
#endif
return ExecGenericCommand(*this->MetalFrontendBinaryCommand[SDK], Parameters, OutReturnCode, OutStdOut, OutStdErr);
}
bool FMetalCompilerToolchain::ExecMetalLib(EAppleSDKType SDK, const TCHAR* Parameters, int32* OutReturnCode, FString* OutStdOut, FString* OutStdErr) const
{
check(this->bToolchainBinariesPresent);
#if PLATFORM_MAC
if (this->MetalLibBinaryCommand[SDK].IsEmpty())
{
FString BuiltParams = FString::Printf(TEXT("--sdk %s %s %s"), *SDKToString(SDK), *this->MetalLibraryBinary, Parameters);
return ExecGenericCommand(*XcrunPath, *BuiltParams, OutReturnCode, OutStdOut, OutStdErr);
}
else
#endif
return ExecGenericCommand(*this->MetalLibBinaryCommand[SDK], Parameters, OutReturnCode, OutStdOut, OutStdErr);
}
bool FMetalCompilerToolchain::ExecAirPack(EAppleSDKType SDK, const TCHAR* Parameters, int32* OutReturnCode, FString* OutStdOut, FString* OutStdErr) const
{
check(this->bToolchainBinariesPresent);
#if PLATFORM_MAC
if (this->AirPackBinaryCommand[SDK].IsEmpty())
{
FString BuiltParams = FString::Printf(TEXT("--sdk %s %s %s"), *SDKToString(SDK), *this->AirPackBinary, Parameters);
return ExecGenericCommand(*XcrunPath, *BuiltParams, OutReturnCode, OutStdOut, OutStdErr);
}
else
#endif
return ExecGenericCommand(*this->AirPackBinaryCommand[SDK], Parameters, OutReturnCode, OutStdOut, OutStdErr);
}
bool FMetalCompilerToolchain::ExecMetalAr(EAppleSDKType SDK, const TCHAR* ScriptFile, int32* OutReturnCode, FString* OutStdOut, FString* OutStdErr) const
{
check(this->bToolchainBinariesPresent);
// WARNING: This phase may be run in parallel so we must not collide our scripts
// metal-ar is really llvm-ar, which acts like the standard ar. Since we usually end up with a ton of objects we are archiving we'd like to script it
// Unfortunately ar reads its script from stdin (when the -M arg is present) instead of being provided a file
// So on windows we'll spawn cmd.exe and pipe the script file into metal-ar.exe -M
#if PLATFORM_MAC
FString Command;
if (this->MetalArBinaryCommand[SDK].IsEmpty())
{
Command = FString::Printf(TEXT("-c \"%s -sdk %s '%s' -M < '%s'\""), *XcrunPath, *SDKToString(SDK), *this->MetalArBinary, ScriptFile);
}
else
{
Command = FString::Printf(TEXT("-c \"'%s' -M < '%s'\""), *this->MetalArBinaryCommand[SDK], ScriptFile);
}
bool bSuccess = ExecGenericCommand(TEXT("/bin/sh"), *Command, OutReturnCode, OutStdOut, OutStdErr);
#else
FString Command = FString::Printf(TEXT("/C type \"%s\" | \"%s\" -M"), ScriptFile, *this->MetalArBinaryCommand[SDK]);
bool bSuccess = ExecGenericCommand(TEXT("cmd.exe"), *Command, OutReturnCode, OutStdOut, OutStdErr);
#endif
if (!bSuccess)
{
UE_LOG(LogMetalShaderCompiler, Error, TEXT("Error creating .metalar. %s."), **OutStdOut);
UE_LOG(LogMetalShaderCompiler, Error, TEXT("Error creating .metalar. %s."), **OutStdErr);
}
return bSuccess;
}
bool FMetalCompilerToolchain::ExecGenericCommand(const TCHAR* Command, const TCHAR* Params, int32* OutReturnCode, FString* OutStdOut, FString* OutStdErr) const
{
#if PLATFORM_WINDOWS
{
// Why do we have our own implementation here? Because metal.exe wants to create a console window.
// So if we don't specify the options to CreateProc we end up with tons and tons of windows appearing and disappearing during a cook.
void* OutputReadPipe = nullptr;
void* OutputWritePipe = nullptr;
FPlatformProcess::CreatePipe(OutputReadPipe, OutputWritePipe);
FProcHandle Proc = FPlatformProcess::CreateProc(Command, Params, false, true, true, nullptr, -1, nullptr, OutputWritePipe, nullptr);
if (!Proc.IsValid())
{
FPlatformProcess::ClosePipe(OutputReadPipe, OutputWritePipe);
return false;
}
int32 RC;
FPlatformProcess::WaitForProc(Proc);
FPlatformProcess::GetProcReturnCode(Proc, &RC);
if (OutStdOut)
{
*OutStdOut = FPlatformProcess::ReadPipe(OutputReadPipe);
}
FPlatformProcess::ClosePipe(OutputReadPipe, OutputWritePipe);
FPlatformProcess::CloseProc(Proc);
if (OutReturnCode)
{
*OutReturnCode = RC;
}
return RC == 0;
}
#else
// Otherwise use the API
return FPlatformProcess::ExecProcess(Command, Params, OutReturnCode, OutStdOut, OutStdErr);
#endif
}
bool FMetalCompilerToolchain::CompileMetalShader(FMetalShaderBytecodeJob& Job, FMetalShaderBytecode& Output) const
{
// The local files
const FString& LocalInputMetalFilePath = Job.InputFile;
const FString& LocalOutputMetalAIRFilePath = Job.OutputObjectFile;
const FString& LocalOutputMetalLibFilePath = Job.OutputFile;
EAppleSDKType SDK = FMetalCompilerToolchain::MetalFormatToSDK(Job.ShaderFormat);
// .metal -> .air
FString IncludeArgs = Job.IncludeDir.Len() ? FString::Printf(TEXT("-I %s"), *Job.IncludeDir) : TEXT("");
{
// Invoke the metal frontend.
FString MetalParams = FString::Printf(TEXT("%s %s %s %s -Wno-null-character -fbracket-depth=1024 %s %s %s %s %s -o %s"), *Job.MinOSVersion, *Job.PreserveInvariance, *Job.DebugInfo, *Job.MathMode, TEXT("-c"), *Job.Standard, *Job.Defines, *IncludeArgs, *LocalInputMetalFilePath, *LocalOutputMetalAIRFilePath);
bool bSuccess = this->ExecMetalFrontend(SDK, *MetalParams, &Job.ReturnCode, &Job.Results, &Job.Errors);
if (!bSuccess || (Job.ReturnCode != 0))
{
Job.Message = FString::Printf(TEXT("Failed to compile %s to bytecode %s, code: %d, output: %s %s"), *LocalInputMetalFilePath, *LocalOutputMetalAIRFilePath, Job.ReturnCode, *Job.Results, *Job.Errors);
return false;
}
}
{
// If we have succeeded, now we can create a metallib out of the AIR.
// TODO do we want to do this in every case? Should be able to skip if we are using Shader Libraries at the high level
FString MetalLibParams = FString::Printf(TEXT("-o %s %s"), *LocalOutputMetalLibFilePath, *LocalOutputMetalAIRFilePath);
bool bSuccess = this->ExecMetalLib(SDK, *MetalLibParams, &Job.ReturnCode, &Job.Results, &Job.Errors);
if (!bSuccess || (Job.ReturnCode != 0))
{
Job.Message = FString::Printf(TEXT("Failed to package %s into %s, code: %d, output: %s %s"), *LocalOutputMetalAIRFilePath, *LocalOutputMetalLibFilePath, Job.ReturnCode, *Job.Results, *Job.Errors);
return false;
}
}
// At this point we have an .air file and a .metallib file
if (Job.bRetainObjectFile)
{
// Retain the .air. This usually means we are using shared native libraries.
bool bSuccess = FFileHelper::LoadFileToArray(Output.ObjectFile, *LocalOutputMetalAIRFilePath);
if (!bSuccess)
{
Job.Message = FString::Printf(TEXT("Failed to store AIR %s"), *LocalOutputMetalAIRFilePath);
return false;
}
}
{
// Retain the .metallib
Output.NativePath = LocalInputMetalFilePath;
bool bSuccess = FFileHelper::LoadFileToArray(Output.OutputFile, *LocalOutputMetalLibFilePath);
if (!bSuccess)
{
Job.Message = FString::Printf(TEXT("Failed to store metallib %s"), *LocalOutputMetalAIRFilePath);
return false;
}
}
return true;
}