Before:
Total CPU Time: 53783.640625 s
Total time in Parallel executor: 558.66 seconds
After:
Total CPU Time: 47886.140625 s
Total time in Parallel executor: 498.81 seconds
#jira
[CL 22173145 by bryan sefcik in ue5-main branch]
#rb trivial
#jira UE-164210, UE-164236
#preflight 632a15d1826e0c2fe962a399
[Backout] - CL21983605
[FYI] Jason.Nadro
Original CL Desc
-----------------------------------------------------------------
Fix bug where we were not blocking on shader compilation before executing a material update context.
- Depending on timing could cause issues with missing shaders where we would update the component before its shaders were complete.
- By default just make the function that submits jobs always block.
- Renable some tests, which this change should fix.
#rb Arciel.Rekman, Chris.Kulla
#jira UE-155888
#preflight 631f7368f927bf0cbaf4e378
[CL 22105123 by jason nadro in ue5-main branch]
- Depending on timing could cause issues with missing shaders where we would update the component before its shaders were complete.
- By default just make the function that submits jobs always block.
- Renable some tests, which this change should fix.
#rb Arciel.Rekman, Chris.Kulla
#jira UE-155888
#preflight 631f7368f927bf0cbaf4e378
[CL 21989440 by jason nadro in ue5-main branch]
- Ported remaining ISceneViewExtension calls to RDG.
- Removed several empty override passes being added to the graph.
- Merged two graphs into one.
#preflight 62c4c61a2a05d4f55bd5912f
#rb luke.thatcher
[CL 20971717 by zach bethel in ue5-main branch]
Implement new function for switching view mode and visualize buffer before taking screenshots;
Screenshot individual buffers instead of using lighting on screenshot tests;
Fix import tests using whatever render context to import was serialized into the config for the import tests;
Disable geometry cache notification on other geometry cache tests;
Increase tolerance on test_pie_streaming_level to give more time for levels to stream in and out before checking for loaded actors;
#jira UE-138139
#rb Anousack.Kitisa, Jerome.Delattre
#preflight 620a448cd332bab853185b48
[CL 18977675 by daniel coelho in ue5-main branch]
When running with ODSC in the editor Materials can have incomplete or partial shader maps. In other words some shaders _might_ be missing. If a render system requires all shaders to be present and can't use ODSC functionality to compile the necessary shaders we need a mechanism for compiling just the missing shaders.
- Added UMaterialInterface::CacheShaders() which is implemented for UMatetrial and UMaterialInstance. The call is just forwarded along to CacheResourceShadersForRendering and InitStaticPermutation respectively.
- You can call CacheResourceShadersForRendering multiple times (even with a completed shader map) and it will not recache uniform expressions. If the Resource's shader map is complete the function is a no-op.
- FObjectCacheContext can accept a UWorld and will only return primitives that are in that world.
- Added a function SubmitRemainingJobsForWorld(UWorld* World, EMaterialShaderPrecompileMode CompileMode) that will iterate all materials used by primitives in the world and compile the missing shaders.
- SubmitRemainingJobsForWorld() is used in various scenarios in the editor to ensure _all_ shaders are compiled for all materials before proceeding.
- Explicitly try to compile UI and PP materials. Since we don't know which ones are used in the given world we just have to iterate all loaded ones. This is a potential area of optimization.
- If there are already complete shader maps (if you are not running with the shader job cache), SubmitRemainingJobsForWorld is fast and results in a no-op for each material.
#rb Arciel.Rekman, Danny.Couture
#jira UE-136442, UE-136447
#preflight 61fc0e9c176256ec4f764206
[CL 18848560 by Jason Nadro in ue5-main branch]
1) Editor viewports now display at high DPI by default for better Gen5-console feeling in editor on high DPI monitors.
2) Realtime editor viewports now display at screen percentage based on the display pixel count. Fix screen percentage independent of the pixel range could either render not enough pixel at low display resolution or be too slow at high display resolution. Given the level editor viewport can scale in size based on editor user layout preference, monitor, the screen percentage automatically scalling based on display pixel count gives a better experience of quality consistency/GPU cost for rendering frame, especially when going level editor fullscreen with F11 or not. The curve to compute rendering resolution from display resolution is in BaseEngine.ini
3) Non-realtime editor viewports still display at a screen percentage based on OS DPI scale, independent of the display pixel count because use spatial upscaler and therefore need consistent sharpness not worst than editor fonts, but keep same rendering resolution as before to not increase out of GPU memory errors. The performance of potentially rendering at high resolution than a realtime viewport is ignored given the non real-time viewports are not rendering every frame.
4) PIE viewports now override r.ScreenPercentage by default to have consistent resolution quality and performance between realtime editor viewport and gameplay viewport by default without risk of introducing out of GPU memory errors loosing possibly unsaved editor work if the game's graphic settings default to a screen percentage higher. To advertise discoverability of this change, any update to the r.ScreenPercentage cvar will lead to a message display in console/log if overriden by editor settings or dynamic resolution settings.
5) As much as a project can change the default anti-aliasing method, can also configure on a per project basis the default screen percentage behavior in project preference that the editor user can then override locally for his/her own use with his editor settings. For instance mobile only or forward renderer MSAA only video games.
6) When working very high end work-in progress rendering tech project in a econemic context where lattest GPUs are hard to get, the range of GPU performance can vary greatly between contributor of the same projects. To provide more robustness for lowerend GPU of collegues with tech not yet scaling, the project settings now offer a MaxRenderingResolution for editor viewports that can also be overriden on a per user basis.
7) Automated screenshot already takes care of forcing r.ScreenPercentage.
#rb none
#preflight 61e58d4a873f2ea48f33facd
#ROBOMERGE-AUTHOR: guillaume.abadie
#ROBOMERGE-SOURCE: CL 18633512 in //UE5/Release-5.0/... via CL 18633532 via CL 18633538
#ROBOMERGE-BOT: UE5 (Release-Engine-Test -> Main) (v899-18417669)
[CL 18637384 by guillaume abadie in ue5-main branch]
This represents UE4/Main @18073326, Release-5.0 @18081140 and Dev-PerfTest @18045971
[CL 18081471 by aurel cordonnier in ue5-release-engine-test branch]
Now only r.AntiAliasingMethod prevails to selects the anti-aliasing method explicitly for the deferred shading renderer. Each AA technic have it's own cvars to control their quality independently
FXAA now have its own r.FXAA.Quality. The FXAA shader permutations where mis configured with r.PostProcessAAQuality=1 & 2 both mapping to the FXAA_PC_CONSOLE=1 that is now r.FXAA.Quality=0. Instead r.FXAA.Quality now offer more mid-quality settings for FXAA_PC=1 with r.FXAA.Quality=1 & 2 & 3. Backward comaptible migration is as followed:
- r.AntiAliasingMethod=1 r.PostProcessAAQuality=0 -> r.FXAA.Quality=0
- r.AntiAliasingMethod=1 r.PostProcessAAQuality=1 -> r.FXAA.Quality=0
- r.AntiAliasingMethod=1 r.PostProcessAAQuality=2 -> r.FXAA.Quality=0
- r.AntiAliasingMethod=1 r.PostProcessAAQuality=3 -> r.FXAA.Quality=3 but with some neighborhood search setting changed in FXAAShader.usf
- r.AntiAliasingMethod=1 r.PostProcessAAQuality=4 -> r.FXAA.Quality=4
- r.AntiAliasingMethod=1 r.PostProcessAAQuality=5 -> r.FXAA.Quality=5
TAA now have its own r.TemporalAA.Quality. Backward compatible migration is as followed:
- r.AntiAliasingMethod=2 r.PostProcessAAQuality=0 -> r.AntiAliasingMethod=0
- r.AntiAliasingMethod=2 r.PostProcessAAQuality=1 -> r.AntiAliasingMethod=1 r.FXAA.Quality=0
- r.AntiAliasingMethod=2 r.PostProcessAAQuality=2 -> r.AntiAliasingMethod=1 r.FXAA.Quality=0
- r.AntiAliasingMethod=2 r.PostProcessAAQuality=3 r.TemporalAAUpsampleFiltered=0 -> r.TemporalAA.Quality=0
- r.AntiAliasingMethod=2 r.PostProcessAAQuality=3 r.TemporalAAUpsampleFiltered=1 -> r.TemporalAA.Quality=1
- r.AntiAliasingMethod=2 r.PostProcessAAQuality=4 -> r.TemporalAA.Quality=2
MSAA r.AntiAliasingMethod=3 & TSR r.AntiAliasingMethod=4 remains unchanged.
sg.AntiAliasingQuality now maps to different r.TemporalAA.Quality or r.FXAA.Quality
Automated tests on base CL: https://horde.devtools.epicgames.com/job/60d5b8410123b700014f9db5
Automated tests on change CL: https://horde.devtools.epicgames.com/job/60d47cde57b302000114bebf
#rb none
[FYI] jack.porter, wei.liu
#lockdown michal.valient
#ROBOMERGE-SOURCE: CL 16823623 via CL 16823646
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v836-16769935)
[CL 16823657 by guillaume abadie in ue5-release-engine-test branch]
