Fixes GPU crashes when releasing Indirect draw and query related dx12 objects
Fixes imgui profiler related to viewing buffer allocations
Signed-off-by: moudgils <moudgils@amazon.com>
* Moved PlatformLimits to setreg. Removed Device PostInit. Some clean up.
Signed-off-by: jiaweig <jiaweig@amazon.com>
* Move setreg loading the PlatformLimitsDescriptor super class.
Signed-off-by: jiaweig <jiaweig@amazon.com>
* Apply same implementation for fake device used in unit tests
Signed-off-by: jiaweig <jiaweig@amazon.com>
* Add implementation for Null renderer. Swap order for register RHI interface in initialization.
Signed-off-by: jiaweig <jiaweig@amazon.com>
* Move back setreg from PlatfromLimitsDescriptor to Device, due to Linux dependency issue.
Signed-off-by: jiaweig <jiaweig@amazon.com>
* Changed the function to take in RHI backend name
Signed-off-by: jiaweig <jiaweig@amazon.com>
* Re-add support for UI Elements that use Render Targets
* Move LyShine pass request from Atom's MainPipeline.pass to project's
* Make all dynamic draw contexts in LyShine draw to pass directly without the need of draw list tags
* Remove local RPI changes that are no longer needed
* Prevent crash if LyShine gem is enabled but its custom pass hasn't been added to the main render pipeline
* Revert to default UI pass if the LyShine pass has not been added to project's main render pipeline
Signed-off-by: abrmich <abrmich@amazon.com>
I had to update the ModelLod class to take in both the ModelLodAsset and ModelAsset for initialization so it can fetch the slots for each mesh.
Signed-off-by: santorac <55155825+santorac@users.noreply.github.com>
Here we inttroduce a new struct ModelMaterialSlot which formalizes the concept of material slot, with an ID, display name, and default material assignment. The ID still comes from the MaterialUid like before. The display name is built-in, rather than being parsed out from the asset file name. And the default material assignment can be any material asset, it doesn't have to come from the FBX (or other scene file).
This commit is just the preliminary set of changes. Cursory testing shows that it works pretty well but more testing is needed (and likely some fixes) before merging.
Here is what's left to do...
Add serialization version converters to preserve prior prefab data.
See if we can get rid of GetLabelByAssetId function only rely on the display name inside ModelMaterialSlot.
I'm not sure if the condition for enabling the "Edit Material Instance..." context menu item is correct.
Test actors
Lots more testing in general
Signed-off-by: santorac <55155825+santorac@users.noreply.github.com>
Adds new instrumentation macros throughout the codebase, using the
visualizer to find where current instrumentation is lacking using the
shadowed sponza sample + editor. Some notes from exploring:
- We spend ~5ms in CullingScene: BeginCulling
- PipelineStateCache: Compact usually 1ms
- CompileImageBarriers takes most of the time in CompileResourceBarriers
Signed-off-by: Jacob Hilliard <jhlliar@amazon.com>
The DynamicDrawContext can output to different scopes: Scene, RenderPipeline or RasterPass.
Updated RasterPass so it can handle not only draw calls from Views but also from DynamicDrawContexts.
* Metal sync interval support
Add shader name to SRGPool for debugging purposes
Set default value of rpi_vsync_interval to 1
Signed-off-by: moudgils <moudgils@amazon.com>
* Expose vsync interval with new cvar rpi_vsync_interval
On change, the rpi_vsync_interval is broadcasted to a new event on
the WindowNotificationBus and all swapchains are recreated with the
new vsync value.
Signed-off-by: Jeremy Ong <jeremycong@gmail.com>
* Add vsync interval support to the Vulkan RHI
Vsync intervals are not intrinsically supported using the Vulkan
swapchain extension. Instead, extra presents are enqueued for each
extra vblank requested past 1. Swapchain recreation is triggered
when transitioning to and from the FIFO presentation mode (when
rpi_vsync_interval transitions from and to 0 respectively).
Signed-off-by: Jeremy Ong <jeremycong@gmail.com>
* Rollback vsync > 1 implementation on Vulkan and leverage *Internal pattern
Signed-off-by: Jeremy Ong <jeremycong@gmail.com>
* ATOM-15939 Add support to capture attachment for ParentPass
- Moved the attachment read back support to Pass class so it supports both ParentPass and RenderPass.
- Added support to output input or output state of an InputOutput attachment.
- Enabled showing ParentPass attachments in PassTree tool.
Signed-off-by: Tao <qingtao@amazon.com>
-This includes implementations of the API for the Editor and Windows, all other platforms will have a 1.0 scale for now
Signed-off-by: nvsickle <nvsickle@amazon.com>
* Final update copyright headers to reference license files at the repo root
Signed-off-by: spham <spham@amazon.com>
* Fix copyright validator unit tests to support the stale O3DE header scenario
Signed-off-by: spham <spham@amazon.com>
Skip a shader item before creating a shader instance if it's not going to be rendered based on the draw tag. This avoids creating and releasing the shader instance over and over again, which results in a disk write each time during release.
-Re-purposed an unused boolean in RPI::Cullable for previous frame's visiblity to instead represent objects that are hidden in the simulation.
-Updated MeshFeatureProcessor::SetVisible to set this value on the cullable.
-Updated the MeshComponent to handle visiblity changes by not rendering the mesh instead of deactivating and/or reactivating the component.
-Updated the AtomActorInstance to handle changes to the visibility from the ActorComponent.
Tested by creating two entities with static mesh components, on entity hidden and the other visible. Plus three entities with actor components, one where the actor is visible, one where the entity is visible but the 'render character' setting on the actor component is disabled, and one where the 'render character' setting is enabled, but the entity is not visible.
For each of these 5 entities, I added them as 5 loose entities, 5 entities that were children to a parent entity, and a slice with all 5 as children to a parent entity, and tested toggling visibility of the parent entities.
For each of these 3 sets of 5 entities, I added them directly to the level, added them all to a layer where the layer was visible, and added them all to a layer where the layer was not visible, and tested toggling the visibility of the layers.
Removed unused shader "2" files
Removed unused shader "2" files that were part of a redesign effort, which is now complete on the development branch. These files are not needed for the upcoming release and could cause confusion for other developers.
ATOM-15837 Remove Unused Shader System Related *2 Classes From Stabilization Branch.
Testing: ASV full test suite on dx12 and vulkan. Only saw known issues.
See also https://github.com/aws-lumberyard/o3de-atom-sampleviewer/pull/142
Merge pull request #1585 from aws-lumberyard-dev/santorac/stabilization/2106/RemoveShader2-ATOM-15837
* ATOM-15859 AuxGeom rendering in editor is too expensive
- The OrphanBuffer calls is the main reason that AuxGeom FP render is slow.
- Switched to use DynamicBuffer for buffers used in DynamicPrimitiveProcessor
- Added some profiling marks.
- Removed DynamicPrimitiveProcessor per view which was added because of OrphanBuffer can only be called once per frame.
ATOM-15837 Remove Unused Shader System Related *2 Classes From Stabilization Branch.
Testing: ASV full test suite on dx12 and vulkan. Only saw known issues.
Currently, the first time a raycast is attempted for a model, the raycast will fail and the model's kdtree will asynchronously get built. This breaks the vegetation system, which expects the queries to always work. This adds in a brute-force fallback for use while the kdtree is building. However, other use cases like the Editor mouse cursor selection raycast still should get the current "silent failure" behavior, because otherwise the Editor will lock up for several seconds the first time the mouse moves over an extremely complex model.
* Enable 16 byte alignment for root constants
* Switching to R8G8B8A8_UINT for dummy stream buffers
* Clamping min values for scissor/viewport for fullscreen pass
[ATOM-14344][ATOM-13908] Merging skinned and static mesh motion shaders using an optional vertext stream. This removes the log spam "Mesh does not have all the required input streams. Missing 'POSITIONT0'." and allows Material::GetShaderCollection() to be const only as it was intended. The MeshFeatureProcessor also no longer needs to decide which motion vector shader to use, and therefore no longer has m_skinnedMeshWithMotion in the descriptor to acquire a mesh.
* - Fix the second window related tabbing issue
- Merge calls to UseResoources acrooaa all the SRGs
- Move SamplerCache to the device to ensure only one cache to reduce duplication
- Fixes to compute threading numbers getting reset to 0,0,0
- Cleanup withing BufferPoolResolver
- Argument buffers are now queued to be cleaned up upon shutdown
Added a ShaderSystem supervariant to provide a system-wide supervariant name.
Changed ShaderAsset to append the system-wide supervariant name when searching for supervariants.
Added the NoMSAA supervariant to several shaders.
* Fixed organization of the AssetProcessor SourceAssetBrowser
Assets within the Engine Root were grouped under a '/' entry.
That has been fixed to use the relative path within the engine root for
those assets
Assets outside of the Engine Root, but on the same drive were using
absolute paths before. Now there are child entries that navigate up the
directory hierarchy to those asset locations
* Added ly_enable_gems call to Atom gems targets that are required
The DefaultLevel.prefab contains several Atom components, that require
the Atom RHI, RPI, Common_Feature, ShaderBuilder and AtomLyIntegration CommonFeatures
gems to be enabled in order to successfully process in the
AssetProcessor.
* Added ly_enable_gems call to make the Camera gem required in Tools,
Builders and Clients.
This is needed as the DefaultLevel.prefab contains an Editor Camera
Component
* Adding the ly_enable_gem call to make the Maestro gem required
CrySystem currently requires Maestro to be enabled in order to
initialize
* Added ly_enable_gems call to the SceneProcessing gem to make it required
The SceneCore and SceneData libraries that are part of the core engine
Code folder requires the SceneProcessing gem to be enabled in order to
invoke the InitializeDynamicModule hooks in DllMain.cpp in order to
initialize those libraries.
* Fixed bad argument in comment for Prefab CMakeLists.txt
* Fixed Assert in Asset Builders due to the Atom RPI Builder
The Atom RPI Builder was enabling the Asset Catalog for the ScriptAsset a second time
The Atom Feature Common EditorSystemCommonComponent.cpp which also loads
in the AssetBuilder is enabling the Asset Catalog for the ScriptAsset
Added BehaviorContext reflection to the OutputDeviceTransformType enum
to fix the BehaviorContext errors about reflecting a method that returns
such an enum
* Added TypeId output to the JsonDeserializer report message about missing
ClassData
Previously the report callback would indicate that the target type was
missing Serialization class data, but didn't indicate the TypeId of the
target type
* Added support to the ly_enable_gems function to be able to support
0 gems being enabled.
Updated the Install step for CMake to propagate any ly_enable_gems
within a CMakeLists.txt for a target into the generated CMakeLists.txt
that is made for each installed IMPORTED target
* Adding newline to the end of the Camera Gem CMakeLists.txt
* Fixing target TYPE parameter for actual Gem Modules to use the GEM_MODULE tag instead of MODULE
* Reverting change to the DESTINATION directory for the installed CMakeLists.txt to use the relative path to the installed directory
* Adding the Atom_Bootstrap gem as a required gem
The Client and GameLaunchers required the Atom_Bootstrap gem in order to create the NativeWindow
Added Atom_Feature_Common client module as a runtime dependency of the AtomLyIntegration CommonsFeature client module
* Fixed register.py --all-projects-path and --all-gems-path arguments to
NOT register projects or gems that are within a template folder
Fixed reading of old pre-1.0 o3de_manifest.json files where the
"engines" key was a json array
* Changed how the relative target source directory is calculated when that source directroy resides outside of the engine root.
The final dirname component is used with a unique SHA256 has to form a <dirname>-<8 char SHA256> folder for installing files into
* Adding newline to the end of Atom_Bootstrap CMakeLists.txt
* Moving ly_enable_gems variants for Tools and Builders inside of PAL_TRAIT_BUILD_HOST_TOOLS block
* Adding a comment to AWSCore.ResourceMappingTool target to indicate that it is not a GEM_MODULE.
Furthermore it cannot be loaded with the Gem system because the library is in a different directory the executable
[ATOM-14544] Add Reset() function to IShaderVariantFinder
ShaderVariantAsyncLoader::Reset() now Shutdown and Init(). It clears
the cache of ShaderVariantAssets it keeps in memory.
Signed-off-by: garrieta <garrieta@amazon.com>
The problem was...
After a MaterialAsset reload, there could be two different versions of the MaterialAsset in memory: the old one and the reloaded one. The old one is still connected to buses and can send reinitialization messages when other things reload or reinitialize. So when the shader asset reloaded, both the old and new MaterialAsset were sending reinitialization messages. Material::OnMaterialAssetReinitialized was using the materialAsset parameter to initialize the Material, and the latest call to OnMaterialAssetReinitialized was for the *old* MaterialAsset.
The solution is to use the m_materialAsset member when reinitializing the Material. I also added checks in a couple places to skip unnecessary reinitialization, and added comments in the bus headers to warn developers about this issue.
Testing: Added a new step to ASV's MaterialHotReloadTest.bv.lua script for the error scenario, and this now passes. Ran ASV full test suite, both dx12 and vulkan, only known issues occurred.
- Added a pause button in imgui cpu profiler.
- Added a capture button to save cpu profiling data to a data file.
- Added some profile marks in both RPI and RHI.
This was done while working on "ATOM-15728 Shader Hot Reload Fails in Debug Build", but it turned out these changes did not actually fix the issue (or any other known hot-reload issue). Still, these improvements are appropriate as they correct logical oversights.
ShaderVariant was not listening to asset reloads. It needs to know when the ShaderVariantAsset reload happens so it can reinitialize it's members as well as propagate reinitialization messages. I added a member for the ShaderAsset as the class needs this to reinitialize itself. So now the class listens for reloads of both the ShaderVariantAsset and the ShaderAsset.
Shader was not listening for ShaderAsset reinitialization events.
Updated the API for ShaderReloadNotificationBus's OnShaderVariantReinitialized to include the ShaderVariant which is the most relevant information (the other information wasn't really being used anyway).
* Extend MeshFeatureProcessor to allow changing the mesh bbox, which requires re-compute the culling data for that mesh
* Update actor mesh bbox when EMFX actor instance bbox changes.
Also use the actor instance global bbox to compute the local bbox for the skinned render mesh, instead of the using the static bounds based bbox, as not every actor instance is going to be using the static bounds bbox.
* Store per-instance mesh AABB in the right place.
In the MeshInstanceData, which is unique per instance, instead of in the Model, which is shared between all instances.
For greater clarity, also remove Model::m_aabb and the corresponding getter and setter, as it isn't immediately obvious whether this gets the model asset bbox or the mesh instance bbox. Callers should instead be explicit about which bbox they want.
* Bug fix: model asset is not necessarily ready in AcquireMesh
* Remove now-unused forward declaration
* Update MockMeshFeatureProcessor with SetLocalAabb/GetLocalAabb
ATOM-15728 Shader Hot Reload Fails in Debug Build
The main change was to add OnAssetReady handlers to each of the asset classes. See comments in ShaderAsset::OnAssetReady for a detailed explanation. In short, OnAssetReloaded gets missed while assets are being reloaded at the same time on multiple threads, but OnAssetReady is always called whenever connecting to the AssetBus because of its AssetConnectionPolicy.
The above change required the addition of a new AssetInitBus to call the PostLoadInit() functions. Because OnAssetReady connects to buses that are not mutex-protected, they have to be connected on the main thread. AssetInitBus::PostLoadInit is called every frame in RPISystem::SimulationTick. All Atom's asset handlers that need to do post-load initialization must connect to the AssetInitBus, and the asset will disconnect itself after initialization is complete.
We also need the Shader class to handle OnShaderAssetReinitialized to properly handle the shader reload.
With these changes I can click back and forth between "Blending On" and "Blending Off" many times (like 20 times) without issue.
This is specifically in support of:
ATOM-14613 Baseviewer MatertialHotReloadTest fails to change the color after turning blending on and off
ATOM-15728 Shader Hot Reload Fails in Debug Build
Esteban Papp
Guthrie Adams
chcurran
moudgils
Kyle Birnbaum
jiaweig
Jeremy Ong
pappeste
michabr
antonmic
Chris Santora
Jacob Hilliard
Qing Tao
Steve Pham
AMZN-tpeng
Jeremy Ong
Chris Galvan
dmcdiar
nvsickle
Tommy Walton
Gene Walters
Chris Santora
Mike Balfour
Ken Pruiksma
hershey5045
mbalfour
lumberyard-employee-dm
galibzon
AMZN-mnaumov
yuriy0
mnaumov