This broke asyncGetLoadProgress() and caused WebGL to crash reliably
because ResourceLoader got destroyed too soon.
Bug was introduced with de7dfc2ea6.
I intend to cherry pick this to rc/1.27.0, which is where it was
introduced, so there's no need to update the release notes.
For each "cgltf_material", we now create one "MaterialInstance", even
if a given asset has multiple instances. In the future, we might make
this behavior configurable to make better use of Filament's
auto-instancing feature.
This change is a feature request from Google, but also this behavior is
more consistent with the code comments.
Also some related cleanup:
(1) Use FixedCapacityVector instead of robin_map.
(2) Move the cache to move out of the asset and into the loader,
because it is only used at load time.
in a few places filamat loses the value of targetApi and targetLanguage
and attempt to guess it from other values, this lead to inconsistencies.
we now keep both targetApi and targetLanguage and use them appropriately.
in particular we don't use the optimization level to "guess" what
targetLangage we're on.
In the end this resolve a prior unit test failure and allows us to not
have to use "spirv" rules when generating only GLSL.
We use TargetLanguage::SPIRV instead of GLSL which affects how the
source GLSL is generated, in particular which version is used.
This change is not a problem for these unit tests, but is probably
exposing another problem -- we will address that separately.
We were keeping the list if uniform names in Material and passing
pointers to the backend. Unfortunately, HwProgram can outlive
Material because it is legal to destroy a Material once it's not
needed on the client side. e.g. just after rendering something.
This was happening with the IBLPrefilter code, but probably elsewhere
too.
We now just store CStrings in HwProgram.
The immutable inverse bind matrices can be shared among instances, so
they are now stored in Asset, not in Instance.
Also, there are now two "load" phases for skinning data:
(1) storing the inverse bind matrices
(2) building the Entity mappings (for animation efficiency)
Phase 1 is done in `ResourceLoader` because inverse bind matrices can
live in an external bin file.
Phase 2 is done during Instance creation, because that's when entities
are created.
Prior to this change, `recomputeBoundingBoxes` was an opt-in config
parameter in ResourceLoader. It is now a method on FilamentInstance.
The old API did not work for dynamically created instances. Since this
is a relatively obscure feature, we considered removing it completely,
especially since the computation requires the presence of CPU-side
vertex data combined with the transform hierarchy.
Instead of removing the feature, we decided to move it to a better
place. This paves the way for some upcoming improvements, which include
reducing the memory footprint for assets. It also improves overall code
organization and separation of concerns.
* Fix warning C4146: unary minus operator applied to unsigned type, result still unsigned
* Fix warning C4068: unknown pragma 'nounroll'
* Fix warning C4068: unknown pragma 'unroll'
* Fix warning C4068: unknown pragma 'clang'
* Fix warning C4305: 'initializing': truncation from 'double' to 'float'
* Fix warning C4267: 'argument': conversion from 'size_t' to 'utils::FixedCapacityVector<filament::uberz::WritableArchive::Material,std::allocator<T>,true>::size_type', possible loss of data
* Fix warning C4267: 'argument': conversion from 'size_t' to 'uint32_t', possible loss of data
* Fix warning C4244: 'initializing': conversion from 'A' to 'T', possible loss of data
* Fix warning C4334: '<<': result of 32-bit shift implicitly converted to 64 bits (was 64-bit shift intended?)
* Fix warning C4293: '>>': shift count negative or too big, undefined behavior
* Fix diagnostic warning C4189: 'channels': local variable is initialized but not referenced
* Use [[maybe_unused]] where possible and revert aa79bd6fa8.
* Revert unary minus for non-MSVC compilers
* Add macro for enabling warnings temporarily
* Get rid of UTILS_HAS_CXX17
* Revisit warning related macros
Co-authored-by: Levente Koncz <levente.koncz@shapr3d.com>
this case could cause an assert in debug builds. There was a few bugs:
- when computing the intersection between a box and a frustum, a special
case test had a typo which would cause extra work in some case. Also,
we were not detecting correctly when there was no intersection at all.
This could result in some intersection point being found.
- Finally, we reject the intersection volume when it has only 3 vertices
because that's not a volume.
Fixes#5933
The main goal of this change is to have "ugly" code only in filamat.
In particular, the sampler binding information is now recorded
into the material file and used on the filament side to inform the
backend.
This means that on the filament side we only have generic code with
all the "uglyness" in filamat. In particular SamplerBindingMap and
SibGenerator now only exist in filamat.
Files of greater interest:
Material.cpp, MaterialBuilder.cpp
This is a workaround for the following Chrome issues:
https://crbug.com/1348017 Compiling GLSL is very slow with struct arrays
https://crbug.com/1348363 Lighting looks wrong with D3D11 but not OpenGL
This requires a rebuild of materials for WebGL users.
Fixes#5859.
partially revert a recent change where Program needs a DriverApi in its
ctor. This was a misguided change. Backend API never take a DriverApi
as parameter.
Checking the feature level, for instance, should happen as a
precondition, or in the backend as postcondition.
GLES3.0 (and GL4.1) need informations about the uniform block bindings
because they don't allow to do that setup in the shaders. So we store
that information in the material blob and retrieve it on the filament
side in Material. This information is passed to Program so it can
create the bindings.
Prior to this change, this information was set by Material itself, but
this ment we had two places where these bindings were hardcoded.
The bulk of the change is to add a new material chunk which stores an
array of {name, binding}, retrieve it on the Material side and pass it
to Program.
- remove unused SamplerBindingMap parameter
In fact, the SamplerBindingMap is used, but it is passed inside the
MaterialInfo structure, so it doesn't need to be passed as parameter.
- Don't create unneeded SamplerBindingMap temporaries
This change was motivated by some internal work at Google and has the
benefit of simplifying the gltfio API and implementation. There are 2
major API changes:
(1) Consolidate separate loader entry points for GLB and GLTF.
The distinction between GLB and GLTF can be made from the file content
alone, because GLB has a 4-byte magic string in its header. There is no
need for separate entry points. Clients do not (and should not) need
to check the file name extension.
(2) Remove the distinction between "instanced" and "non-instanced"
glTF assets.
In the new scheme, all assets have at least 1 instance.
Broadly speaking, in gltfio an "asset" is a collection of Filament
objects like textures and vertex buffers, while an "instance" is a
collection of entities and components (e.g. the transform hierarchy).
This API change makes life easier for clients because they no longer
need to decide a priori if they will ever need to add instances.
This change also moves some public-facing methods from FilamentAsset to
FilamentInstance:
- getSkinCount, getSkinNameAt
- getJointCountAt, getJointsAt
- attachSkin, detachSkin
* Begin Sorting SubProjects into Folders
* Add more subprojects to folders
* Add even more subprojects to folders
* Add further subprojects to folders
* Move the last two projects
* Move Resources to a Resources subfolder
* Remove spaces to be stylistically coherent
* Revert Improper CMake Modifications
* Revert erroneous line removals
* Only specify sdl2's folder on WIN32
* Add the shader subprojects to a Generated folder
* Move shaders to Filament/Shaders
* new feature level API for backends
backend can now return a "feature level", each level corresponds to a
"bundle" of features.
Level1: ES3.0 capabilities
Level2: ES3.1 capabilities + 31 textures + cubemap arrays
Currently metal always returns level 1, GL and Vulkan return level 2
if 31 textures or more are supported.
* Add public APIs for feature levels
* Add infrastructure to check feature levels in materials
* validate material feature level on use
The validation is done when creating a renderable. If the engine doesn't
support the material's feature level, an exception is thrown (or assert
if exceptions are not enabled).
* material documentation
* activate ESSL 3.10 for feature level 2
also generate #defines to identify available feature levels
* support for cubemap arrays in the public API
if feature level 2 is supported, cubemap arrays can be used from the
public API.
* add release notes
With VERTEX_DOMAIN_DEVICE the vertex shader doesn't apply the
projection (since the vertices are already in clip space), however,
both TAA and guard bands need to jitter/offset the clip space, and
it is done at the projection level.
We now store the clip space offset separately so that it can be applied
to VERTEX_DOMAIN_DEVICE vertices.
We also introduce a new material parameter, vertexDomainDeviceJittered,
a boolean that controls whether clip space offset (above) is applied.
This is because a VERTEX_DOMAIN_DEVICE material that uses the built-in
projection matrices generally ends-up with the jitter already applied,
this is the case with the Skybox for instance.
Fixes#5917
Fixed a comment that said "We store 64 bytes per bone.". The actual
number is 32.
Developers who know their users have powerful devices may wish to exceed
ES3.0 minspec constraints to allow more bones and / or morph targets.
Fixes#5785.
The custom release callback that I provided was in the wrong place; it
was only being used for a path string, not the actual buffer content.
The cgltf API is a bit awkward in this area.
No need to update RELEASE_NOTES because this will be cherry picked to
the RC branch, which is where the bug introduced.
Fixes#5918.
Filament does not yet fully support threads with WASM, but this is a
baby step in that direction.
To enable experimental pthreads support, enable the WEBGL_PTHREADS CMake
option. This will enable pthreads support in `gltfio` and `utils`, which
is known to work, but not when served with GitHub Pages.
The web server must emit COOP, COEP and CORP headers, so our build
instructions now recommend the use of `emrun` for local testing.
This also changes our demos so that they do not use unpkg, which
does not work when using `emrun`, due to cross-origin restrictions.
Filament does not yet fully support threads with WASM, but this is a
baby step in that direction.
To enable experimental pthreads support, enable the WEBGL_PTHREADS CMake
option. This will enable pthreads support in `gltfio` and `utils`, which
is known to work, but not when served with GitHub Pages.
The web server must emit COOP, COEP and CORP headers, so our build
instructions now recommend the use of `emrun` for local testing.
This also changes our demos so that they do not use unpkg, which
does not work when using `emrun`, due to cross-origin restrictions.
This feature can improve load time when textures are downloaded from the
web on non-filesystem platforms like Android.
More specifically, this allows downloaded texture assets to arrive after
the user calls asyncBeginLoad(), which means that decoding and
downloading can occur concurrently.
Prior to this PR, we already used JobSystem for decoding, but we did not
kick off any jobs until after all assets were downloaded.
Still TBD: add this feature for external vertex data.
Partial fix for #5909.
Earlier versions of cgltf did not support file reader customization.
This was fixed back in Dec 2019 but at the time I did not notice, so
we never bothered cleaning up our usage.
In the future we would like WebGL + Android builds to permit texture
downloads to occur concurrently with the texture decoder jobs. This PR
is basically a preparatory refactoring, but with the nice side effect of
removing a memcpy.
it was a duplicate state from the true gl state which could easily cause
problems. It only existed to optimize state updates when nothing changed
but we already have fine-grained tests for this.
It was culled by the frame-graph because it didn't have "read" from
any of its resource, only writes. However, it does set uniforms, so need
to be called.
fix#5874
StructureOfArray always aligned each array to the same alignment as
malloc (usually 8 bytes), but that was not enough if one of its type
has stricter alignment requirements.
StructureOfArray now always honors at least the alignment requirement
of each array.
Also removed dependency on EntityInstance.h
Fixes#5727
We used to have a mix of fatal and non-fatal assertions for precondtions
errors. From now on, we always throw if exceptions are enabled or
crash with a log otherwise.
In most places this is simply replaced by `std::string_view`.
We also change a few internal/private headers so they accept
`std::string_view` instead of `utils::CString`.
Internally we use std::string_view and the public API is augmented
with APIs that take a length for strings, which makes them useable
with string_view parameters.
Also fix exception specification for all setParameter methods, which
can throw if exceptions are enabled (or exit the program otherwise).
When FTransformManager computes accurate world transforms, it adds downcasted leftover to translation part to restore precision.
However, both transltation and leftover are given as float3 - simply adding them resut float3, so addling leftover become meaningless.
one or both should be upcasted first.
I have disabled building SDL with headless EGL, because
SDL_config_minimal.h doesn't work with EGL, and I don't know how to
implement a an SDL config that would work with EGL.
I have verified that this works in a separate project, and that it
compiles in this project.
```
$ ./build.sh -e release
$ find ./out/ -name "*EGL*"
./out/cmake-release/filament/backend/CMakeFiles/backend.dir/src/opengl/platforms/PlatformEGL.cpp.o
./out/cmake-release/filament/backend/CMakeFiles/backend.dir/src/opengl/platforms/PlatformEGLHeadless.cpp.o
./out/cmake-release/libs/bluegl/CMakeFiles/bluegl.dir/src/BlueGLLinuxEGL.cpp.o
```
I have disabled building SDL with headless EGL, because
SDL_config_minimal.h doesn't work with EGL, and I don't know how to
implement a an SDL config that would work with EGL.
I have verified that this works in a separate project, and that it
compiles in this project.
```
$ ./build.sh -e release
$ find ./out/ -name "*EGL*"
./out/cmake-release/filament/backend/CMakeFiles/backend.dir/src/opengl/platforms/PlatformEGL.cpp.o
./out/cmake-release/filament/backend/CMakeFiles/backend.dir/src/opengl/platforms/PlatformEGLHeadless.cpp.o
./out/cmake-release/libs/bluegl/CMakeFiles/bluegl.dir/src/BlueGLLinuxEGL.cpp.o
```
GL_ES_30 becomes MOBILE
GL_CORE_41 becomes DESKTOP
ShaderModel controls which flavor of GLSL (GL or ES) is used both when
reading and outputting materials.
It's also used to set default quality settings and the default precision
of all fragment shaders.
Technically, Vulkan and Metal don't need this distinction, but the GL
backend does.
this is more in line with other APIs of the backend. This change is
complete for the GL backend, but for the metal/vulkan backend we still
use a SamplerGroup as the internal data structure, which is just
temporary.
Eventually, SamplerGroup should becomes a "filament" only API (not
a backend API).
SamplerGroup itself is now a more "traditional" class where copy and
move ctor do what you'd expect. This actually removes a few copy of
the internal data in some cases and uses less memory, at the cost of
doing some heap allocations, but they should be rare and outside of
the main loop.
we mow do most of the work of validating the sampler parameters and
resolving to an actual GL sampler, when updating a SamplerGroup,
instead of doing this each time a program is made active.
A lot of cpu work is saved when the same samplergroup is reused with
multiple programs.
this is the start at making HwSamplerGroup more "real". It shouldn't
keep a reference to the filament::SamplerGroup, which is just an object
to pass the data.
for now, we move the SamplerGroup reference into the concrete classes to
keep the same implementation. But now it becomes an "implementation
detail" of the respective backends.
- remove the 2 argument version of setSampler
- use a bool instead of bitfield for the dirty state
- try to avoid setting dirty bit if setting same sampler
Starting with 3.1.14, embind started to support for `noexcept`
which caused multiple definition errors since we have a workaround
in place that alreadys supplies template instantiations for `noexcept`.
This change should not affect G3 since our JS bindings are not used
in G3.
The upstream fix is here:
https://github.com/emscripten-core/emscripten/pull/17140Fixes#5789.
This is not intended to be used yet because we're not currently
checking that cubemap array are actually supported, however, if they
are, they should work.
the depth of the texture wasn't taken into account. In practice this
would restrict the number of mip levels of a 3d texture that would have
a larger dimension in depth.
The main goal is to allow more flexibility, allow cubemap arrays in
the future and better match vk and metal apis.
Main changes:
- remove updateCubeImage
- remove update2DImage
- update3DImage is now the only texture upload backend API
cubemaps are now treated just like a 2D array of 6 layers.
For this reason, Texture::setImage(..., FaceOFfsets) is deprecated.
Additionally, the 2D versions of Texture::setImage() become inline
helpers.
A side effect of this change is that it is now possible to update only
a single face of a cubemap, but also a region of a face (or faces).
This PR changes how Metal handles buffer updates. Previously, Metal allocated a full new buffer each time an `updateBufferObject` command was issued; however, it did not copy the previous contents of the buffer over to the new buffer, so partial updates did not work correctly.
Now, Metal allocates a single, private GPU buffer and employs temporary staging buffers whose contents get blitted to the private buffer at each update.
There's still some room for optimizations, and I need to give more thought to how I want to implement `updateBufferObjectUnsynchronized`.
- move all froxel configuration constants out of Engine.h, unlike
the previous todo/comment, these shouldn't be part of Engine::Config.
- same for irradiance map
For some android gpu drivers, some uniform arrays should be initialized to be used in the shader, even if not used.
Co-authored-by: Mathias Agopian <mathias@google.com>
This code adds a config object, "ConfigParams," to Engine creation parameters for setting memory buffer sizes for command buffers and driver handle arena sizes. It will use #define values (FILAMENT_PER_FRAME_COMMANDS_SIZE_IN_MB, etc.) as defaults (and as minimum acceptable values) if the user does not provide values. It attempts to validate the values given to prevent the user from creating a unusable state.
BlobDictionary and LineDictionary were storing blobs as map keys to
achieve simple compression, but they also stored duplicates of
in a vector for index-based lookup.
Now, the vector is storage and the map has keys that are string_view.
- getNear() and getCullingFar() now return doubles
- updated documentation
- all setProjection() calls can now throw (when enabled) and will
do so if preconditions are not met (instead of setting a default
projection).
- Frustum can now be logged on debug builds
This patch enables user scissor in Metal backend; There was no implementation for it.
While the absence of the feature in Metal does not incur serious problem, many rendering glitches were found in apps using ImGui due to it.
Co-authored-by: Benjamin Doherty <bendoherty@google.com>
auto-instancing can have some overhead, so when it is known that the
scene doesn't have identical primitives, it is better to disable it.
(disabled by default).
Also add some missing bindings for `enableAccurateTranslations`.
Reflects a change from Betty and should be cherry picked to v1.23.3
Users could customize the ImGuiHelper camera, but they had no control
over the scissor coordinates. This allows them to use vertically flipped
coordinates, which, unfortunately, is required for MediaPipe
integration.
Reflects a change from Betty and should be cherry picked to v1.23.3
Users could customize the ImGuiHelper camera, but they had no control
over the scissor coordinates. This allows them to use vertically flipped
coordinates, which, unfortunately, is required for MediaPipe
integration.
This was too tricky, it highlights that we definitely need refactor this
to share code with the flattener / unflattener pipeline. I'll look at
that next.
a 64 bytes pool seems to work with both clang and msvc, unfortunately,
c++ doesn't let us know the allocator object size at compile time
for map containers, so we have to guess.
* BlueVKDarwin: Simplify the build and loading process.
Some of our CMake logic was not necessary because we do not staticly
link against anything for Vulkan. All entry points are dynamically
loaded.
Some of the load-time code was also needlessly complicated. In fact the
ICD environment variable that we were setting is now deprecated.
I tested this PR with:
- macOS + June LunarG SDK
- Android on a Pixel 6
* Fix up, further simplification.
This lets us avoid the following validation error.
Attempting to create a VkDevice from a VkPhysicalDevice which is
from a portability driver without the
VK_INSTANCE_CREATE_ENUMERATE_PORTABILITY_BIT_KHR bit in the
VkInstanceCreateInfo flags being set...
This will allow us to fix the validation error related to
VK_KHR_portability_enumeration.
Also, macOS comes with Python 3 nowadays so we're upgrading the script
from Python 2.
Unsynchronized update just means that the buffer is updated regardless
if what the gpu is doing, the synchronization is the responsibility of
the caller.
Also added resetBufferObject(), which essentially destroys and reallocate
a buffer for the same handle.
- rename PerRenderableUibBones to PerRenderableBonesUib to be more
consistant with other interface block naming.
- make sure the C++ struct and the UBO definition match.
PerRenderableBonesUib is a UBO with one bones[256] field, but the
C++ struct has a single bone and treated the UBO as an array.
This makes things less confusing.
We can now use sizeof(PerRenderableBonesUib) in many places which
better expresse what we are doing.
- Fix an off-by-one assert when setting a skinning buffer to a renderable
- Asserts that no more than 256 bones are associated to a renderable
- improve documentation and fix some typos
We no longer use the hardcoded objectUniform UBO name, instead, we
access the object uniforms through the new getObjectUniforms().
Because it's not possible to return a "Uniform Block" from a function
in GLSL, a large part of this change, is to replace the interface block
by a structure.
Note: getObjectUniforms() is not public, but it is used by gltfio to
access the userData field, which is also not public at the moment.
DEPTH commands are now sorted just like COLOR commands, that is,
they use a 10 bits z-bucket and the material id in the sorting key.
This allows instance-able primitives to be sorted next to each other.
We still use resgen for convience, but the archive is now passed in
from the client application.
This will allow us to shrink the gltfio Android library (stay tuned).
This boolean parameter is used with materials that need to access
`getInstanceIndex()` (filament's equivalent to `gl_InstanceIndex`).
It is false by default, and getInstanceIndex() is not accessible.
This is intended to be used in concert with
`RenderableManager::Builder::instances()`.
a 64 bytes pool seems to work with both clang and msvc, unfortunately,
c++ doesn't let us know the allocator object size at compile time
for map containers, so we have to guess.
An ubershader archive is a bundle of filamat packages with some metadata
that conveys which glTF features each material supports.
This PR does three things:
1. Adds a new command line tool called `uberz` that consumes a list
of filamat files and metadata text files and produces a single
ubershader archive.
2. Adds a new library (also called `uberz`) that is used by `gltfio`
to read ubershader archives, and used by the above command line
tool to write ubershader archives.
3. Enhances `UbershaderLoader` so that it no longers uses a hardcoded
set of materials, and instead takes an ubershader archive.
Ubershader archives have a simple binary layout that can be memcpy'd
directly into a C struct. The metadata is specified using a text file
with key-value pairs. These two file formats have formal desriptions in
the README in `libs/uberz`.
In a subsequent PR, we will remove the `gltfio_resources` target and
change the signature of `createUbershaderLoader` so that it takes
an archive.
We were re-building various C++ files in three different targets
(basis_encoder, basis_transcoder, basisu executable), it's probably
better just to set up proper dependencies between the libraries. Note
that I'm not sure if this actually improves build times.
More importantly, this creates a CMake target for the zstd library
because we want to use zstd in other places.
blendOrder was used to control the draw order of blended render
primitive within a Renderable. We now have an option to make the
blend order global, in this case those primitives with a global blend
order are always sorted solely using the blend order value (i.e. the
distance from the camera is not take into account).
This change includes a fix for an issue we filed against spirv-tools:
https://github.com/KhronosGroup/SPIRV-Tools/issues/4371
So, it should reduce the number of extraneous matrix copies in finalized
GLSL code.
I did quick size total of all filamat files (*) before and after this
update, it went from 11.3 MB to 11.1 MB.
(*) All backends are enabled, includes built-in Filament shaders and
gltfio ubershaders.
* Add JS bindings for Texture class methods
This change introduces bindings for the following methods:
- getWidth
- getHeight
- getDepth
- getLevels
Closes 4492
* Mark level arguments as optional
* Revert docs defs
* Revert docs/webgl/filament.js
if we have more than 16K of UBO data, we now use a "out of band" buffer
on the heap instead of the command stream, which has limited space.
to minimize the heap allocation we use a simple pool allocator that
recycles a few heap buffers (per Scene).
This optimizes and cleans up some code from a 3P contributor.
When computing a bounding box, there was no need for an inner loop
through the entire skins array.
Tested using the torus model in #4973 and the `-r` flag in gltf_viewer.
I think the most important thing here is that we now let users know that
"offset" is not a byte count. This also fixes some small typos, e.g.
"weights" was used in the docstring instead of "positions".
To avoid making API changes, this does not fix a few weird things I
noticed in the Java API. For example, there is a redundant "count"
argument in methods that take an array. Also some methods do not
provide an "offset" argument, which is not consistent with C++.
Our `morphNormal` GLSL function treats each normal target as a
displacement from the base normal (similar to the glTF spec) but the
normal that is extracted from the tangent frame is actually an absolute
normal.
In other words, if b is the base normal, we were doing:
b = b + w0 * n0 + w1 * n1 + ...
This is obviously wrong, since the base normal has an overpowering
influence.
The fixed math looks like this:
b = b + w0 * (n0 - b) + w1 * (n1 - b) + ...
Fixes#5584.
By design we shouldn't access the scene SOA when executing the
high level commands.
This change increases the size of a command to 64 bytes and
PrimitiveInfo to 48 bytes (both with some small padding left).
This removes usage of regexps from the parsing phase and splits the
program into three sub-packages: parser, database, and emitters.
The parser now generates an AST according to the formal grammar
described in the README:
https://github.com/google/filament/blob/pr/beamsplitter_rewrite2/tools/beamsplitter/README.md#grammar
This allows for nice readable error messages. More importantly, it
permits the C++ syntax to be less restrictive and paves the way for
possible expansion of the tool beyond `Options.h`.
I looked at the C++ AST generated by clang but it is huge and unwieldy.
For our purposes this simplified AST is much easier to work with.
The new lexer is inspired by the following Rob Pike talk.
- https://www.youtube.com/watch?v=HxaD_trXwRE
Beamsplitter does not use the state machine described in the above
prezo, but it does use a Go channel for separating the parser from the
lexer. In our case, the lexer is actually a recursive descent parser
with simple lookahead functionality. This made it easy for the "real"
parser to create an ergonomic coarse-grained AST.
This is a big change but it is a no-op in terms of the generated code.
* gltfio: Clear texture caches before loading resources
* iOS gltf-viewer: add double-tap to reload model for debugging
* iOS samples: add instructions on ASan / UBSan debugging
- Ensure C++17 is used to compile sample projects
- Upgrade to the latest Xcodegen and re-generate projects
- Add a bundle ID disambiguator so users don't need create unique bundle Ids when opening sample projects
- Add a simple generate-samples.sh script
The API is the same but there are some minor differences, which include:
- Ordering of fields in AmbientOcclusionOptions
- More @NonNull annotations that we forgot
- Javadoc formatting is different
This is a temporary workaround for a memory corruption issue observed on
some devices from a specific vendor. We will try to make this workaround
more targeted in a subequent change.
Partial revert for b2cdf9f2b4.
There were two places where we were doing unaligned reads: one when
computing the hash for the material identifier, and one when parsing
the chunk in ShaderReplacer.
We also had a potential overflow since civetweb does not add a trailing
null to incoming WebSockets messages.
This moves parser into a subpackage to shield it from the various
emitters (this folder structure is similar to what Rob Pike used for his
text templating library). This also adds a WIP Java code generator,
currently disabled.
We now have 3 public APIS:
- opaqueBlit() which is used for basic blitting and scaling and
supports sub-resources.
- blit() which is used for blitting + composition
- can optionally perform basic scaling
- doesn't support sub-resources
- upscale() which is used for high quality upscaling
- can do composition (blending)
- doesn't support sub-resources
This is a temporary workaround for a memory corruption issue observed on
some devices from a specific vendor. We will try to make this workaround
more targeted in a subequent change.
Partial revert for b2cdf9f2b4.
The most important change here is that we no longer apply the
backend (e.g. Metal/Vulkan) UV transformation before calling the
user's vertex shader. This is an API change for post-process materials
(but they're not public).
Now the user is responsible for using uvToRenderTargetUV() in their
shaders (either in the fragment or vertex as appropriate).
This makes it easier to handle offsets/transforms with all APIs.
Conceptually, the only thing that is needed is to call
uvToRenderTargetUV() just before making a texture call.
This fixes a couple of issues:
- some image shifting in metal/vk
- flare in metal/vk was upside down
We also simplify the DoF code quite a bit now that we can rely on the
rendertargets begin multiple of 16.
We also "fix" SSAO that was working by accident on metal/vk. The UV
correction was applied 3 times 2 of which were canceling each other.
This adds a code generator, implemented in Go.
This PR also prepares `Options.h` (the ground truth) by simplifying its
syntax just a bit. The ground truth file must have very simple C++
syntax, which is described in the README:
https://github.com/google/filament/blob/pr/codegen2/tools/codegen-options/README.md
This process revealed a small bug: `Filter.MEDIAN` was bound to the
incorrect value in Java.
The generated code is not yet used, stay tuned.
* Add multi-scene support to gltf_viewer.
To test this, I generated a multi-scene asset as follows.
```
gltf-transform merge Avocado/glTF/Avocado.gltf \
BarramundiFish/glTF/BarramundiFish.gltf \
~/Desktop/Merged.gltf
```
* Specify c++17 in pbxproj files.
When the guard band is enabled, the effective rendering area is
increased by a certain amount (currently 16 pixels on each side) so
that screen-space effects (e.g. SSAO, SSR, DoF) behave better around
the edges of the screen. Of course, this comes at a performance and
memory cost.
Currently the guard band is not configurable other than being optional.
Added c++, java and js bindings.
This also removes clipping of the blit rectangle that occurred in the VK
backend that is no longer necessary. There's also no need to assert
since validation will catch out-of-bounds blitting.
Reverts the clipping in acdb8addfd.
Also deprecate some fields of DisplayInfo this shouldn't be a problem
because they were not actually used.
This change should allow an external-to-filament code to manage
frame pacing (e.g. something like swappy).
- For completion, FilamentAsset now has getRenderableEntities(). This is
similar to sister methods getLightEntities() and getCameraEntities()
except there is no need to store a separate array.
- The web helmet demo does not need to enable shadows, they are already
enabled.
- The ViewerGui populateAsset() method was doing two things that are
now decoupled for clarity: setAsset() and populateAsset().
- The updateRootTransform() method is now called only when the autoscale
checkbox is toggled, instead of every frame.
- The getFooEntities() methods in Java now skip doing work for empty
lists. Actually these should not return arrays at all, but let's fix
that later, since it will break backwards compatibility.
We can't call prepareProgram() from any thread, since it is calling
into the backend. When we had more than a certain number of commands,
we would use the jobsystem to generate them and so prepareProgram()
could end up being called from a jobsystem thread.
This is fixed by looping through all the commands once they're generated
and calling prepareProgram() then.
The main changes are:
- ColorPass() and RefractionPass() are now static and moved into a
new RendererUtils file. These methods don't need FRenderer and are
more like a big "script".
- Cleaned-up includes
- Reformatted/reordered methods
The SimpleViewer C++ class was not viewer component like `ModelViewer`
and `<filament-viewer>`. It was actually just the UI builder used
in the `gltf_viewer` desktop app and the remote Android interface.
This adds a new implementation of the TextureProvider interface called
Ktx2Provider.
Tested using the KTX2 variant of the StainedGlassLamp model in the
Khronos samples repo.
Tested on WebGL 2.0 (Chrome v100), Android (Pixel 6 Pro), and Desktop
(Metal, OpenGL, and Vulkan via MoltenVK).
A "safe" mip-level is one that has exactly half the dimensions of
the current level. In other words, we guarantee that we can halve the
dimensions 4 times without loss (i.e. odd dimension).
This is achieved by effectively padding the viewport (i.e. making the
viewport's dimensions multiple of 16). The post-processing passes
eventually take care of cropping the extra padding so the final
image is unchanged.
This is a more generic solution to dynamic resolution padding, helps
with memory allocations and helps all the algorithms that depend on
mipmaping (e.g. DoF, SSAO, Bloom).
One important improvement brought by this change is that all "final"
post-processing effects are now able to handle a sub-region of the
source, including: color-gradding, fxaa and upscaling.
It was assumed that ETC2 is supported on all ES3 devices, but WebGL on
desktop machines does not support ETC2.
Note that WebGL has its own special extension strings for compressed
textures.
Some uniforms are needed during shadow map generation, other
during shadow sampling (color pass). We didn't have a clear separation
between them and ended up setting the UBOs that were needed only for
sampling multiple times during generation.
- rename common_math and common_shadowing from .fs to .glsl
because they were both included in both vertex and fragment
shaders
- add getFragCoord() which always returns coordinates in the
glsl convention. This is useful when used with RNGs to get
consistant results accross backends.
this is because this uniform is not well defined -- it's unclear if
it is set to the viewport or the buffer size.
also, it's not needed, and removes a couple multiplies from the shader.
- add getUserViewMatrix() on CameraInfo, the "user" view matrix is the
view matrix before we apply the world origin transform, it is needed in
a few places, so we make it a method so that in the future we could
precompute it if we wanted to.
- remove worldOffset which is just the last column of worldOrigin
it made it actually more confusing to have two version of the uv
(normalized and not). Also the unnormalized version was computed from
frameUniforms which is not great in the case of post-processing
(it just happened to work).
This allows clients to provide their own asynchronous texture decoders
for various mime-typed images. This is a plug-in component for gltfio,
in some ways similar to MaterialProvider.
There are two motivations for this: to decouple gltfio from STB and
to make it easier to integrate support for BasisU textures.
This also has the side effect of simplifying ResourceLoader, since the
texture decoding jobs have been moved out.
As part of this work, I made the "stb" CMake target into a traditional
static library. Previously we had several files called `Image.cpp`
whose sole purpose was to enable STB_IMAGE_IMPLEMENTATION.
If there is an error during the driver initialization, Engine::create()
returns nullptr, but the FEngine was allocated and would then never
be destroyed.
mTerminated was useless because by construction it would always have
the expected value.
This version is used only on qualcomm devices. It's now forbidden to
create a program inside a renderpass. Unfortunately there was still
one codepath doing so.
this state only exists during Renderer::render(), having it outside
of view is more flexible as it give us an opportunity to adjust
camera parameter globally before drawing.
Also greatly reduced the size of RenderPass (which is a Value object).
mipgen can now emit basis-encoded KTX2 files. Both the desktop and
web "suzanne" samples use this as a test for compressed textures.
This PR does not add KTX2 support to glTF, but it's on the way.
`BasisEncoder` has a builder style API that calls the basis encoder to
create KTX2 files. This hides some low-level BasisU features that we are
not using, like file I/O and mipmap generation.
`Ktx2Reader` is an easy-to-use API for creating Filament textures from
KTX2 files. Its API primarily consists of these two methods:
bool requestFormat(Texture::InternalFormat format);
Filament::Texture* load(const uint8_t* data, size_t size);
The first method is used to build an ordered list of formats that are
supported by your hardware. The second method consumes the contents of a
basis-encoded KTX2 file and attempts to produce a Filament texture with
a preferred format.
IMPORTANT: Our tools still let you use KTX1 for non-compressed images
because it is useful for HDR, but you can no longer use KTX1 for
block-compressed data.
Partial fix for #4771.
This is only a good place for a flush *if* we're going to draw another
view (so the GPU can start working). Otherwise, it can be costly
on some drivers and unnecessary since swapbuffers will do the same.
The original article we used for our bloom (siggraph'14) called for
a radius parameter for the upsample stage. We're always using 1.0 right
now, but it is useful to keep this possibility for experimentations.
a bug was introduced in 8/2020 where the downsampling filter was wrong,
due to a mistake using textureLodOffset, which obviously uses offsets
in source texel -- we were assuming destination.
Path resolution was broken with #5285, which only worked when the glTF
happened to be at root level. This new fix has been tested against:
- My Sponza zip file that hass all files in a "sponza" folder.
- My Bistro zip file, which has resources in a child folder relative to
the gltf file
- A self-created flat zip using IridescentDishWithOlives, with
everything at root level.
Side note: Bistro still fails on a Pixel 6 Pro Mali but for an unrelated
reason: (null deref in backend::TimerQueryNative::queryResultAvailable)
- don't use __PRETTY_FUNCTION__ and try to parse it, __func__ is
standardized and in most case returns what we want (the function name).
- add native support for string_view in our ostream.
- uninline string support from ostream
When creating a KTX file (e.g. with mipgen), you can of course choose
whether or not you want to use an sRGB format.
At load time, your app has an expectation about whether a particular
texture has an sRGB format. e.g. the glTF convention is that albedo
textures should be sRGB, while roughness textures are linear.
Before this commit, our KTX loader would look at the app's expectation,
then silently re-interpret the bits in the glTF texture in order to
match this expectation. This seems very wrong.
Instead, if the expectated format does not match the format in the KTX
file, we now print a warning.
Fixes#2422, which will also be obsoleted by our upcoming KTX2 support.
The `g_linearized` variable was being used for two purposes: to denote
the linearity of the source format AND the linearity of the destination
format. This was confusing and is now split is `sourceIsLinear` and
`destIsLinear`.
This change has no effect on the the look of the suzanne demo.
We do not support these formats and they add bloat to the basis tables.
In a release x86 build, this reduces the transcoder size from 774 KiB to
475 KiB.
Allocation and destruction of handles generates a fair amount of code
because they are synchronized and the spinlock is inlined.
So we fix this by generating the 3 (de)allocation methods we really
need, we only need 3 because we're using 3 pools. These are now
out-of-line, which reduces size and reduces pressure on the i-cache.
Performance is not affected much because in the end it's just a
static branch (and in most case the return is elided).
The main change here is that instead of allocating a Dispatcher on the
heap and storing a pointer to it in the CommandStream, we now have
a Dispatcher directly inside the CommandStream.
This avoids a heap allocation and a dereference when executing a command.
Changes:
- Remove the strange header-only variant of libs/image
- Rename KtxBundle => Ktx1Bundle
- Move image/KtxUtility => ktxreader/Ktx1Reader
- Add unit test for Ktx1Reader and test stub for KTX2.
Notes:
- Ktx1Bundle does not depend on Filament and is used by cmgen
and mipgen to generate KTX files.
- Ktx1Reader has a dependency on Filament and should therefore live in
a separate library, which it now does.
The `demo_*` web samples were using unpkg to refer to the Filament build
rather than using the local copy, which was inconsistent with the other
web samples.
We now use the local copy of Filament for everything except the
drag-and-drop viewer, which always uses the latest version of Filament.
With this change we are now generating refraction and reflections
buffer into their own layer of a 2D array -- so they can coexist.
This change doesn't actually enable both at the same time yet.
There are downsides to this approach:
- a 2d array of 2 layers is created even if only one of reflection or
refraction is active. This could be fixed at the cost of more
complexity.
- if reflections are active then refraction must use a RGBA16F texture
instead of RGB_11_11_10, because they share the texture
unlike for refraction, the reflection buffer is distinct from "the"
color buffer, this allows an optimization (compared to ssr) when we
generate the mipmap chain -- we can allocate the mipmaped texture
and render the SSR pass directly into its level 0. This achieved
use the framegraph's forward resource feature.
With this change, shaders are compiled at program creation, programs
are linked at the start of the next render pass and queries are
deferred until first use.
This should improve overall performance on some drivers.
This reduces the size of OpenGLProgram from 64 to 32 bytes and
improves the code that sets new samplers.
The main simplification is that we're not trying to "pack" the
texture units used in the program. For e.g. we could be using
units 0,3,4, where before it would be 0,1,2.
This allows us to do less tracking.
Another optimization is that we're discarding entirely a SamplerGroup
if the program doesn't use any of its samplers. This helps remove
a test in updateSampler which in turn remove somem tracking data.
Instead of recomputing the Frustum from its 8 vertices, we compute it
from the projection matrix directly. This ensures this works with any
projection matrix. Building a Frustum object from 8 vertices required
these vertices to be given in a certain order which wasn't correct if
the projection was flipped for instance.
Removed the API to construct a Frustum from vertices.
fixes b/225975881
move its content to DriverEnums.h, we do this because it is used
in libfilabridge, which has an unspoken rule that it only includes
DriverEnums.h.
filabridge also only has a header dependency, so it can't call any
method, not even operator<<.
Also cleanups and simplifications.
The main simplification is that we don't need an actual map to retrieve
the global offset of a sampler, it can be easily calculated as:
global_offset = offset_of_the_block + local_offset
Instead of passing the level and layer explicitly to gaussianBlurPass,
we rely on the information already contained in the handle. i.e. we
honor the "subresource" part of the handle (if it has one).
In other words, it's now possible to pass a subresource (e.g. a layer)
to generateGaussianMipmap or gaussianBlurPass and things should work as
expected.
The problem here would happen when forwarding a subresource (i.e.
replacing an existing resource by a subresource) and the existing
resource had been written to. In this situation, the subresource's
parent needs to depend on the subresource (i.e. as if it was written to
by it). This would normally happen during write() but in this situation,
the write had already happened.
the dependency graph would be incorrectly severed in the situation where
a subresource was written to and later its parent was read from.
In this situation, the parent node is in fact a new version of the
original parent node, so we need a "read" dependency of the orginal to
the copy (otherwise the orginal node is not a dependent of anyone
and gets culled).
Some clients need to link against both Filament and OpenGL. This can lead to symbol clashes under certain circumstances. As a fix, prefix all of our `bluegl` symbols with `bluegl_`. Previously we made this naming change for Windows, but now we'll do it on all platforms for simplicity.
The existing behavior was surprising for users who draw opaque objects
into a semitransparent views, and this was especially evident with
the labels in `TextureLinearInterpolationTest`.
We now disable blending for MASKED, which is what our existing materials
documentation already says. We also now set the fragment alpha to 1 when
the fragment is not discarded, which prevents the "punch through"
effect. This is consistent with ThreeJS.
Fixes#4576.
We were decrementing activeJobCount after removing the job from the
queue, which could cause other threads in the pool to preempt us before
the decrement, causing them to spin forever trying to get a non-existant
job, until the decrement actually happened.
Now we always decrement first and fix-up the count if we couldn't get
a job from the queues. The race is inverted, and doesn't cause threads
to spin a long time.
fixes b/201100123
This removes some special handling of the default render target lifetime
that is no longer needed, and renames "surface" to "swapChain" where it
is appropriate to do so.
This is a gltfio API change (!)
Jave clients now need to call destroy on MaterialProvider.
Previously, the Java AssetLoader took over ownership of a native
material provider upon construction, but this was neither documented
nor consistent with the C++ layer. (This is historical; in the past we
did not expose MaterialProvider to Java.)
One motivation for this (aside from API consistency) is that users may
wish to preserve the material cache from one run to another.
Fixes#5132.
We now have a single varyings.glsl containing our varyings for both
the vertex and fragment shader. This file is included with the right
definition of VARYING as `in` or `out`.
inputs.vs is now renamed attributes.vs
This changes the WebGL and desktop glTF viewers so that they always
call `updateBoneMatrices`, even when there are 0 animations. Note
that our Android sample was already doing this correctly.
When combined with #5301, this fixes#5299, although users would need to
use the `recomputeBoundingBoxes` feature in gltfio (`-r` in the desktop
viewer) for this model to be scaled to fill the viewport, since its
embedded bounds are quite large.
In the past there was an API gotcha because users had to "get" the
animator before releasing the glTF source data. This could have been
surprising because it was a getter method, not a factory method.
This was due to overeager optimization on my part, I wanted to avoid
animator overhead for non-animated models, when in fact it has very
little overhead.
Moreover, the animator is conceivably useful even when there are no
pre-supplied animations (e.g. for applying skins), so let's just create
it unconditionally.
Motivated by #5299.
Two changes:
1) Khronos has moved from "master" to "main".
2) BlueVK was attempting to generate reasonable "operator<<" support for
bit flags but this was buggy due to type aliases. Since we weren't using
it anyway, I simply removed this functionality from BlueVK.
Instead of having 2 fields we're following the vulkan convention of
having only the layer field.
The layer/face convention is still maintained in the public APIs.
fixes#5273
Dragging a zip file into the Remote Viewer page would fail if it
contained a glTF with `./` prefixes in the `images` section. The best
fix is to use Java's proper URI object rather than trying to parse the
string.
The JSON response to /api/active became malformed after #4465 because
raw hex strings need to be enclosed by quotes.
This commit changes the variant format in the /api/materials response
to be consistent with one used for /api/active. By using integers
instead of strings, we're avoiding the need to parse integers at run
time.
The JSON error did not appear in the Chrome console because it was being
silenced as a hack to appease "matinfo --web-server". I fix this by
removing the hack and simply emitting a valid response when there's
no live backend.
Also fixed the display of materials, which were always being marked
as active even when they had no active variants.
We held a reference to a cache item after potentially clearing it due to
pool growth. I was able to trigger an error only after enabling ASAN and
loading a large model. Many thanks to @jeanlemotan for catching this.
When we added per-layout arenas for each of the 3 descriptor types, we
did not account for them when determining how much of the Vk Pool is in
use. Therefore the "growth" (really a re-creation) of the Vk Pool wasn't
always occurring when necessary, causing descriptor set allocation to
fail with large models.
In the current version of the library, it looks like the DYNAMIC config
macro has been replaced with a STATIC config macro.
Also disabled the "unused-private-field" warning since this is a 3rd
party library and its not our fault.
The main change here is from ResourceList which ended-up generating
a lot of code due to inlining. This class is only used for tracking
user resources and is not in the performance path.
This saves another ~5K or so of code.
This caused a unit test failure because a StaticString constructed
from "make" had a different hash than one constructed from a literal,
even though the two strings were lexigraphically equivalent.
Having MorphTargetBuffer in FRenderPrimitive is a concept of glTF. Filament doesn't have to follow this concept. Actually the responsibility of FRenderPrimitive is to define vertex input stream because FRenderPrimitive maps to HwRenderPrimitive in Filament. If we store MorphTargetBuffer into FRenderPrimitive then it exceeds the responsibility of FRenderPrimitive. Because MorphTargetBuffer doesn't have any information on how to set up vertex input stream. So handling morphing like skinning will be better.
We were not unrolling the == loop because the generated code with
unrolling was very branchy; but that's not a good call either because
the loop has the same number of unpredicted branches.
Theoretically the compiler should be able to emit a branchless == but
it doesn't. At least we give it a chance.
When we introduced texture-based morphing, our custom depth resolver
needed to be updated to set up samplers correctly. This fixes a
validation error about shader stage access.
Every sidecar texture should have the same dimensions as its parent
texture.
Prior to this change, the sidecar texture would have the dimensions of
the RenderTarget wrapper rather than the parent texture. This was wrong
because the RenderTarget might be slightly smaller than the texture.
This caused an issue when the same texture was attached to two render
targets with different sizes, because there's only 1 cached sidecar per
texture.
We were passing the wrong texture object to VulkanDisposer::acquire().
MSAA depth sidecar textures are not managed by VulkanDisposer, but their
parent textures are.
Fixes#5217.
When we overhauled image layout transitions, we moved as many
transitions as possible into the render pass. However the transition
that we perform for "depth read only" layout needs to be applied to
multiple miplevels, so we cannot use the render pass for that.
We now handle "too many bone weights" in the same way that we handle
"too many UV sets". i.e. emit an error message and do not panic.
This change allows us to render the model in #5234 quite reasonably.
Fixes#5234.
This fixes the emscripten binding errors that we've been seeing
with the <filament-viewer> test page, which prevented us from
including web in the last few Filament releases.
The binding errors were caused by double-initializing the emscripten
module.
I fixed this by allowing clients (e.g. FilamentViewer) to call
Filament.init() more than once. We now accumulate a list of "on ready"
callbacks that get triggered after the emscripten module becomes ready.
As far as I can tell, multiple canvases were actually always broken, and
the viewer test page worked in the past only because we got lucky.
This fixes the emscripten binding errors that we've been seeing
with the <filament-viewer> test page, which prevented us from
including web in the last few Filament releases.
The binding errors were caused by double-initializing the emscripten
module.
I fixed this by allowing clients (e.g. FilamentViewer) to call
Filament.init() more than once. We now accumulate a list of "on ready"
callbacks that get triggered after the emscripten module becomes ready.
As far as I can tell, multiple canvases were actually always broken, and
the viewer test page worked in the past only because we got lucky.
Previously, the Vulkan backend made an educated guess about the depth
write state during the start of the render pass. The logic for this
did not make any sense, so now we're replacing it with an explicit
hint that PostProcessManager sets up. In a sense this gives backends
a sneak peek into the raster state.
The existing unit test had an incorrect expectation in what happens
after splicing an existing range. Fixing the unit test revealed an
actual bug in the RangeMap implementation. It should never modify the
lower bound of an existing interval, because that invalidates the
iterator. It should also never retain a pointer or reference to a range
when the range is potentially removed by subsequent codelines.
We can now specify an "instance count" per renderable during creation.
This instance count applies to all render primitives in the
Renderable.
This will simply draw the renderable "instance count" times. A new
method available to vertex shaders, getInstanceIndex(), allows the
material to discriminate the instances and adjust the position/transform
accordingly.
Typically this is used for particle systems.
Two big changes to simplify and fix layout transitions:
(1) The RenderPassKey struct has been reduced from 88 bytes to 56 bytes
by using custom enum types and bitmasks. It also now has separate fields
for "final" and "initial" image layouts.
(2) VulkanSwapChain now leverages VulkanTexture, which enables it to
share tracking functionality, avoid duplicated state, and have less code
overall.
In a future PR we will clean up the ReadPixels and Blit paths, and also
remove some hacks that appease Vulkan when SSAO is enabled.
The main change here is that for screen-space reflections, the lod
calculation takes the camera distance into account.
We also now correctly take into account that we blur each lod
successively which increases the standard deviation by 10%. This applies
to both refraction and reflection.
This works by first generating a reflection buffer which gets blurred,
then the color pass samples from this buffer according to the roughness
of the surface being rendered.
A lot of the changes here involve utilizing "reserved" variants for
the new "SSR" pass and all the fallout from that.
Variants are no longer just a bit mask, but rather a combination of
some bits, depending on the variant. Because of that the variant filer
must be updated.
Basically we now make a distinction between the "variants" as a public
material API and the actual `Variant` data type.
This change does the impedance match between the two.
The previous code would convert each element of the source data
into 8 bit-per-element, but we wnat to preserve the original format
that the user provides.
The new solution is to use `slice()` which is a robust way to clone
all the data in a typed array.
This fixes the new regression with Triangle that Ben caught.
The previous code would convert each element of the source data
into 8 bit-per-element, but we wnat to preserve the original format
that the user provides.
The new solution is to use `slice()` which is a robust way to clone
all the data in a typed array.
This fixes the new regression with Triangle that Ben caught.
If emscripten grows the heap inside one of our BufferDescriptor binding
functions, then the old heap becomes "detached" and an error can
occur.
This fixes the issue seen with the Parquet demo that Ben caught.
If emscripten grows the heap inside one of our BufferDescriptor binding
functions, then the old heap becomes "detached" and an error can
occur.
This fixes the issue seen with the Parquet demo that Ben caught.
filament/src/Renderer.cpp
- make code mode readable, hopefully
- SSR mipmap generation is now more or less standalone
- when possible, resolve color buffer directly into
mimpmapped texture's base level.
All three types of caches (descriptor sets, pipelines, and pipeline
layouts) are now managed in exactly the same way. They all use an LRU
eviction scheme that is based on a count of command buffer flush
events.
Vulkan objects can only be destroyed if there are no in-flight command
buffers that reference them, so an easy way to know when it is safe to
evict a given entry is to wait for "N" flushes after its last use, where
"N" is the number of command buffers in the command buffer ring.
Another big simplification is that there are no more dirty flags,
instead we store two sets of state vectors for each type of cache: the
"currently bound" state, and the "current requirements" state.
All three types of caches (descriptor sets, pipelines, and pipeline
layouts) are now managed in exactly the same way. They all use an LRU
eviction scheme that is based on a count of command buffer flush
events.
Vulkan objects can only be destroyed if there are no in-flight command
buffers that reference them, so an easy way to know when it is safe to
evict a given entry is to wait for "N" flushes after its last use, where
"N" is the number of command buffers in the command buffer ring.
Another big simplification is that there are no more dirty flags,
instead we store two sets of state vectors for each type of cache: the
"currently bound" state, and the "current requirements" state.
- Rename projectToPixelMatrix to uvFromViewMatrix, this match our
existing style better.
- slightly more efficient direction transform
- limit roughness to 0.1 instead of 0.01, because in a lot of cases,
the roughness increases with the LOD, for instance if it comes from
a texture (to combat aliasing).
The current Variant is need to calculate BindingIndexMap correctly.
But we can't use the current Variant when calculate BindingIndexMap
because SamplerBindingMap uses only the dummy Variant. Both are
should be same.
The morphing sampler group can be added if skinning or morphing is
enabled. However SamplerBindingMap can't know skinning or morphing
is enabled because only dummy Variant is used. It causes generating
the different sampler bindings between SamplerBindingMap and
program's sampler groups.
Fixes#5139
- Make sure to disabled SSR when it's... disabled.
- Fade the reflection when the ray direction is towards de camera,
where it will likely hit a back face
- Simplify the fading code, for near identical results.
Fixes#5142 by replacing unsafe pointers with map keys.
One of the differences between robin_map and unordered_map is the
following:
pointers to keys or values in the map are invalidated in the same
way as iterators to these keys-values
Therefore it is unsafe to track the pointer to a value that is stored
in a robin_map.
This is mostly just code cleanup. One actual bug was the fact that the
dummy sampler was re-created every time a new pipeline layout was
created.
It also felt strange to use `auto&` to refer to a C-style array. I
changed this into a `std::array` which is more consistent with other
fixed size arrays in this class.
Fixes#5142 by replacing unsafe pointers with map keys.
One of the differences between robin_map and unordered_map is the
following:
pointers to keys or values in the map are invalidated in the same
way as iterators to these keys-values
Therefore it is unsafe to track the pointer to a value that is stored
in a robin_map.
This is mostly just code cleanup. One actual bug was the fact that the
dummy sampler was re-created every time a new pipeline layout was
created.
It also felt strange to use `auto&` to refer to a C-style array. I
changed this into a `std::array` which is more consistent with other
fixed size arrays in this class.
This fixes the gltf_viewer automated testing mode, invoked via
`--batch=default --headless`.
ReadPixels was being issued with an out-of-bounds size on high DPI
platforms, which caused weird backend errors.
Since we create an invisible SDL window while in headless mode, the
presence of the SDL window does *not* indicate headless vs non-headless.
Partial fix for #5111
Because the morphing buffer is actually a texture, it's possible that
the last line is not a full width when updating the buffer.
Because texture are updated via a rectangle, it was possible to read
past the malloc'ed buffer during the update.
Moreover, because the count can be specified, we can't just blindly send
more "dummy" data.
Finally, added an offset to match the skinning API, and in that case,
the first line can also be incomplete.
We fix this by issuing up to 3 texture update calls if needed.
The java/js API update will come later.
Support legacy morphing (morphing with targets supplied via VertexAttributes) for older clients. This gives clients more time to transition over to the new MorphTargetBuffer API.
Support legacy morphing (morphing with targets supplied via VertexAttributes) for older clients. This gives clients more time to transition over to the new MorphTargetBuffer API.
Instead of building the Variant based on flags in RenderPass, we just
pass the Variant itself to RenderPass. The caller has more knowledge
of what it wants to do.
This allows `MorphStressTest` to work on Vulkan.
However, `Horse` is still broken because it provides positions but not
tangents. Separate fix for that is coming.
Partial fix for #5109.
This fixes validation errors and makes a first pass at simplification.
VulkanTexture now tracks image layout using RangeMap, which paves the
way for further simplification.
This will allow the Vulkan backend to efficiently track the subresource
image layouts for each texture.
This is a sparse container for a series of ordered non-overlapping
integer intervals, where each interval maps to a concrete value.
This is because we're using the same program variant for skinning
and morphing, in the skinning-only case, the buffer won't be accessed
in the shader, but it must be present.
fixes#5085
This allows the `Horse.glb` model to work with Vulkan without any
validation errors. As a reminder, the glTF spec says:
```
When normals are not specified, client implementations MUST calculate
flat normals
```
Fixes#5115.
* Get the morph target count from the renderable
* Set the morph target count by Builder
* Update morph target weights partially
* Remove noexcept at functions could throw exceptions
* Update Java bindings at RenderableManager
* Add missing comments
* Update morph target weights from offset
Co-authored-by: Mathias Agopian <mathias@google.com>
This allows `MorphStressTest` to work on Vulkan.
However, `Horse` is still broken because it provides positions but not
tangents. Separate fix for that is coming.
Partial fix for #5109.
This fixes validation errors and makes a first pass at simplification.
VulkanTexture now tracks image layout using RangeMap, which paves the
way for further simplification.
There is no need to go through all primitive and call updateSamplerGroup,
because the SamplerGroup never changes -- it just hold the texture that
holds the data.
This is because we're using the same program variant for skinning
and morphing, in the skinning-only case, the buffer won't be accessed
in the shader, but it must be present.
fixes#5085
This will allow the Vulkan backend to efficiently track the subresource
image layouts for each texture.
This is a sparse container for a series of ordered non-overlapping
integer intervals, where each interval maps to a concrete value.
This fixes a double-initialization problem for one of the function
pointers, and also updates the BlueVK generator to fix issues with the
latest Khronos XML file.
The corollary C++ changes will be made in a subsequent PR since they
will be difficult to review.
See #5026.
2D array textures are initialized with .depth = 1 and layerCount = depth (see line 58 in VulkanTexture.cpp). By setting layerCount = 1 only the first layer was copied over, leaving the other layers uninitialized.
This functionality (in the Automation section) was violating a
constraint that was landed in November.
1) Call postRender inside beginFrame / endFrame.
2) Use RGBA when issuing ReadPixels (required by some backends) and
convert to RGB when dumping out the PPM file.
screen-space reflections are integrated in the indirect light computation
(similarly to screen-space refraction), this is because it can be seen as
the specular term of screen-space GI.
However, because this effect is screen-space, the data it accesses has
the ibl luminance (and exposure) applied.
Besides some minor refactoring, this change just applies the
screen-space reflections *after* the ibl's specular term is exposed.
We can now do screen-space refraction and reflections at the same time.
This change mostly removes the checks that were preventing it from
working.
The real change in the PR is to adjust the reprojection matrix based
on whether the SSR buffer (both for refractions and reflections) comes
from this or the previous frame.
what's not allowed at this point is an object which has both screen
space refraction and reflection.
this is accomplished by just always disabling screen-space reflections
for the translucent step of the refraction pass.
The `uint8_t` here needs to be cast to an integer to be printed
correctly. Also these warnings seem to be quite common so we should
hide them in Release builds.
These warnings probably indicate an actual problem that I do not fully
understand yet.
Move the command generation of the structure pass into its execute
block, meaning that if the structure pass is culled, the command
generation is now skipped too.
The same can be achieved with VSM + blur=3 however, this is noticeably
slower. Also when using transparent shadows, VSM needs a much larger
blur to iron out the dithering pattern.
Fixes#5048, #5039
* Replace generic tonemap curve with a simpler one
This change removes the shoulder parameter which had inconsistent
behaviors. The new curve is simpler and designed to still match
by default the gray point and the contrast of ACES in a bright
surround.
* Fix build
ostream internals are now protected by a lock, which ensures the
internal state stays consistant, however, this won't prevent
multiple threads to have their output ominterfering with each other,
which is no different from the stl behavior.
This fixes#4992
The version of Texture::setBitmap() that tool an Android bitmap handled
callbacks differently and in certain case caused a "double release" of
the callback object. This didn't actually cause problems though.
We now use the same mechanism used elsewhere (i.e. JniCallback).
it was incorrect to store bone matrices as quaternions because it is
allowed for these matrices to have a skew component.
We now store the 4x4 matrix -- which is the same amount of data as
before, however we compute the cofactor matrix in the vertex shader when
transforming normals.
fixes#4887
- the hammersley sequence was completely wrong because of a missing
highp precision qualifier
- roughness 0 was also wrong because of divide-by-0
Fixes#4668
- FILAMENT_SKIP_SDL2=ON is used to skip dependencies of sdl2 and sdl2 CMake
- FILAMENT_LINUX_IS_MOBILE is used for Mobile build
Signed-off-by: Joel Winarske <joel.winarske@gmail.com>
__ANDROID__ is always set by the toolchain and less likely to cause
conflicts than ANDROID. This change also removes the -DANDROID flag
we set ourselves in our toolchain CMake files since we don't need
it anymore.
* Use locale-independent string->float conversion
strtof and friends are locale aware and won't parse decimal numbers
with a period ("12.6" for instance) in locales that use another
character for the decimal period ("," in French for instance).
This change introduces a new function called strtof_c that forces
the use of a specific locale (called "C") to make sure we always
parse floats in the desired "C" format ("12.6").
With C++17 we should be able to use std::from_chars but this API
is not implemented in clang for floats at the moment.
* Fix Linux
A recent refactor was causing the following error when the vertex domain
was set to `device`:
```
ERROR: main.vs:23: 'material' : undeclared identifier
ERROR: main.vs:23: 'materialVertex' : no matching overloaded function found
```
A recent refactor was causing the following error when the vertex domain
was set to `device`:
```
ERROR: main.vs:23: 'material' : undeclared identifier
ERROR: main.vs:23: 'materialVertex' : no matching overloaded function found
```
We're currently only using the positive layer of EVSM which works
work small blurs and anti-aliasing. This just rearranges the code to
make it easy to test the full formulation.
This is possible because, in fact, we know if we're dealing with a
spot or directional light when we invoke shadow(). So a conditional
inside it is resolved at compile time.
This will allow more flexibility in the future. Also now the shadow()
function only needs the shadow index and cascade -- which is more
future proof (e.g. if we want to handle more than one directional
shadow or spot lights with cascades).
this effectively undoes a recent change, where we had a method for
spot lights and one for the directional light. instead, we can
calculate the bias at the correct Z on the caller side -- which is
needed only for spotlights.
The faulty commit was:
commit 23dab0a013
Author: Mathias Agopian <mathias@google.com>
Date: Fri Oct 29 12:03:14 2021 -0700
Fix very narrow spotlight lighting and other minor issues
Shadow Receiver Plane Depth Bias is only needed when using large PCF
filters, which we are no longer doing. Large filter kernels are now
supported through VSM sampling.
This changes how we select the "optimized depth vertex shader", but
this shouldn't change the current behavior.
We now select the "optimized depth vertex shader" if and only if the
user code is empty. In that case, we can safely assume we can remove
all the code that's not necessary for the depth, since it is all
controlled by filament.
- remove PCF "low" quality, we only use "HARD" now, when using PCF.
Higher quality levels are achieved by using VSM.
- added a version of PCF that doesn't use a shadowSampler for future
use.
The normal bias is now computed correctly, this requires to compute
the z in lightspace in the shader.
Note that this would not work as well if we used LISPSM, but we'll
cross that bridge when we get there.
- constant bias and normal bias default values in java didn't match
C++ or the documentation
- stable shadows were enabled by default in java
- polygon offset biases were missing from the java API
- document and don't apply polygon offset to VSM
- remove unused code
We now cull the shadow casters before computing the near/far plane
for spotlights -- we can do that because we know the light's frustum.
So only these casters that contribute to the shadow are accounted for
when calculating the near/far plane.
This PR also include more cleanup and simplifications.
Near/far plane is now computed dynamically, this significantly
improves the shadowmap depth resolution.
Quite a bit of refactoring was necessary to get all the right data
in the right place, in particular SceneInfo is now passed around for
both directional and spot shadows.
The froxel code was failing when a spotlight cone was too narrow. This
is fixed here by never using a cone smaller than 0.5 degrees during
forxelization.
Additionally we now silently clamp the cone angles to 0.5 degrees at
the API level because, the falloff was also failing due to floating
point precision in the shader.
Finally, we clamp the inner cone upper value to the outer cone's
instead of the other way around. i.e. the outer cone defines the
spotlight while the inner cone just controls the falloff.
When calculating the linear depth for VSM, we were using the whole
range between 0 and the far plane, the near plane wasn't taken into
account.
This can be a problem is the light is very far, but it's near plane is
closer to the camera/scene, in this case the depth precision wasn't
used optimally.
Note that we don't hit this problem currently, because the directional
light is constructed such that its origin is at the near plane, and the
spotlights have a fixed near plane (which is a problem and will be
fixed at a later time).
The bug that prevented them to work has been worked around, but we
don't know when or how. The original bug still exists as demonstrated
by our standalone test apk.
For now, we reenable MSAA VSM on Adreno.
* Perform sanity check when combining static libraries
This is a small modification that simplifies Debian packaging when one
or more vendored dependencies are replaced by system shared libraries
and no longer need to be combined with a Filament library.
* Verify vendoring before including license texts
This is another small modification to simplify Debian packaging when one
or more vendored dependencies are removed and replaced by system
libraries, so they no longer need to be considered when compiling the
license texts.
- remove PCF MEDIUM and HIGH quality -- these were never used in
practice, and "quality" really meant "blurriness", now we have VSM
for doing that with much more control and more efficiently.
PCF_HARD is now the default. PCF_LOW is still available at compile time
for now. It shouldn't be needed though because VSM can achieve the same.
Moved the divide-by-w of light space into the shadow() method -- which
known if it should be done or not.
The codebase needs only a few additional standard #includes to compile
against libstdc++ on Linux; presumably those headers are implicitly
included with MSVC's and Clang's standard C++ library, but even if
libstdc++ compatibility is not a goal, it is advisable to have them
included directly.
in particular add a terminate() method to FrameSkipper instead of
destroying driver objects in the destructor -- this matches all other
objects.
also remove dependency on FEngine which wasn't needed.
- move the control code out of FrameInfo and into
View itself. FrameInfo now just gathers information about the
frames instead of also doing part of the control.
- the control code has been refactored into a more formal PIDController.
- a few bugs were fixed in the control loop and default parameters
tuned.
The control loop itself now outputs a relative scale factor instead
of an absolute one. The relative scale factor seems easier to
control and is less jittery, and doesn't have to rely on the "integral"
term of the PID.
It's possible that more tuning is needed, but the scaling is now
more stable.
- remove getProperties() we'll put it back if we need it some day
- use unordered_map because it generates less code than robin_map and
here performance is not critical
- defaults to 31 history points instead of 3, this gives a much better
estimate of the frame rate.
- make sure the calculations work for any definition of our internal
duration<> (it's useful to use milliseconds when dealing with
refresh periods).
- don't round to 8 pixels if the scale factor is exactly 1
- fix a few comments and warnings
* mipgen: Linearized PNG bitmaps (such as normal maps) no longer perform gamma transform on read.
* mipgen: Using "-k normals" argument now works for all file types, not just KTX
* Web: Exposed SurfaceOrientation functions getQuatsHalf4() and getQuatsFloat()
Unfortunately this bug appeared in recent drivers on some devices, it's
virtually impossible to know which device/drivers have this issue, so
the workaround is active for all Adreno GPUs.
glInvalidateFramebuffer is too aggressive and prevents loading of the
GMEM even if it's only called at the end of the render pass.
Statically link SDL2 to make our samples (glTF Viewer, etc.) completely
standalone, and include missing resources (fonts and default environment
map for the IBL).
When supported and enabled, resolve MSAA using an HDR aware filter,
which improves anti-aliasing quality.
This is only supported with backends that support multi-sampled
framebuffer-fetch. Currently, only Metal on iOS and GLES.
Update remote ui.
The DSR options were carried all the way to the final blit that is
sometimes needed to impedence-match our rendering features with the
swapchain. This blit is never a scaling blit, but we were using the
FSR option, which would sometimes trigger the FSR scaling code.
FSR assumes modern OpenGL. The functions that pose problem are in fact
never used in filament, so it doesn't matter how they're implemented.
We use the HLSL version instead.
Java callbacks are now directly dispatched to their handler, instead
of first going through filament's opportunistic dispatch, reducing
callback latency.
All APIs that take a callback as argument now also takes a
CallbackHandler* which is a new interface used to dispatch the
callback.
CallbackHandler* can always be nullptr in which case the behavior is
unchanged.
When CallbackHandler* is specified however, it is used to dispatch the
callback: its post() method is called from a service thread as soon
as possible (this will NEVER be the main thread), CallbackHandler::post
is then responsible for scheduling the callback onto the thread the
user desires. This is intended to make callbacks interoperate with
the platform/OS's own messaging system.
The main issue that gets fixed here are:
- VK_IMAGE_LAYOUT_UNDEFINED cannot be used if the image content is to
be preserved after the transition. Unfortunately, there is more than a
few places in the VK backend where we use VK_IMAGE_LAYOUT_UNDEFINED on
images that are later used as sources.
- accessing an image memory from the host is only supported if the the
layout is PREINITIALIZED or GENERAL. However, images in the staging
pool were transitioned to other layouts and put back into the pool,
later they would be mapped/memcpy'ed into. Instead we now always use
the GENERAL layout and never change it.
- images that where in the pool where transitioned using the miplevel
of the texture we were copying into, however this didn't make sense
because images from the pool don't have mip levels.
As part of these fixes, we also refactor imageLayoutTranstion(),
because there was two copies of this function, one specific to texture
and another one more generic. We now always use the more generic one,
with a new helper for textures.
Note: didn't fix "readPixels" which has at least problem (1), because
it's a tricky implementation, and it requires more toughts.
A new Engine::pumpMessageQueues() method can be used to trigger all
pending user callbacks right away. This can help reducing latency
of certain callbacks such as picking queries.
This is typically called once before calculating the next frame's
parameters (e.g. object positions or appearance that may depend on a
callback result).
Fix some comments and documentation.
This is a pixel accurate implementation of picking. Picking queries
can be created on view, and upon completion a user provided callback
is called with the Entity of the renderable at the queried coordinates
in the viewport.
Picking queries typically have 1 or 2 frame of latency and may impact
performance on some drivers.
It is mostly intended for use by editors, or when latency is not a major
concern. This api should not be used for dragging/moving objects, it is
intended for initial picking only.
Picking is implemented in the structure pass.
The depth buffer value is retrieved and the fragment coordinate is
reconstructed and passed to PickingQueryResult. This can be used in
turn to calculate the view and/or world space position.
This is just a minimal change to get the R_INTEGER format to work.
- if src/dst types are identical, just copy (this avoids a mul and a div)
- add R,RG and INTEGER types
There a just new inline helpers to more easily use lambdas, functors and
method callbacks with {Pixel}BufferDescriptor.
A BufferDescriptor can be constructed like this:
auto bd = BufferDescriptor::make(buffer, size, [](void* b, size_t s){ });
or
auto bd = BufferDescriptor::make(buffer, size, &Foo::bar, *foo);
the code generating the commands for the structure pass (called DEPTH)
in the code was different that from the color. This made sense, but
isn't great from an architectural standpoint.
Moreover, the special filtering flags DEPTH_FILTER_*_OBJECTS were
useless because part of their functionality was handled in the DEPTH
code directly.
Instead, we now generate the DEPTH commands the same way than we do for
COLOR, but we honor the DEPTH_FILTER_*_OBJECTS flags properly.
This should help Filament properly support more glTF assets with
transparency, and should make Filament behave more like other
renderers such as three.js.
The finalizer can't be used to dispose of filament resource, because
it runs on its own thread.
Added a destroy() method to all IBLPrefilter related classes that must
be called once the object is no longer needed.
note: Resources will be freed eventually anyways when the Engine itself is
destroyed.
fixes#4633
A lod bias can be applied to textures when dynamic resolution is
enabled. A proper lod bias is calculated and applied for quality levels
HIGH and ULTRA.
Lod Bias is implemented in glslang, which mean it's ignored when
glslang is not used, which currently happens with OpenGL backends with
optimization turned off (which includes filament LITE). This may be
fixed in the future.
We were decrementing activeJobCount after removing the job from the
queue, which could cause other threads in the pool to preempt us before
the decrement, causing them to spin forever trying to get a non-existant
job, until the decrement actually happened.
Now we always decrement first and fix-up the count if we couldn't get
a job from the queues. The race is inverted, and doesn't cause threads
to spin a long time.
the gc usually runs for a very short amount of time, so the jobsystem
overhead is very significant compared to the actual work.
additionally this works around a scheduling issue that happens sometimes
on some devices where a thread could be "runnable but not running" for
many milliseconds.
On some GPU such as Adreno, the shader compiler flattens the early
exit condition, which becomes useless. We work around this by using a
two pass version of the algorithm using the depth buffer for
skipping "fast" fragments that have already been rendered.
This improves performance significantly, e.g. from 5ms to 1.3ms on
Pixel4.
Currently this mode is always enabled, which hurst performance on GPUs
that don't have this problem. This will be remediated later.
Now that FSR scaling is cheap enough on mobile, we use it for quality
levels MEDIUM to ULTRA.
MEDIUM and HIGH use the mobile optimized version, which has a slightly
lower quality, ULTRA uses the original version.
Additionally the RCAS sharpening pass can always be disabled by setting
the sharpen value to 0.
We used the "float" code with the "half" texture coords, turns out FSR
uses slightly different code for both.
We're now using the "half" code (using mediump float).
filament cannot use fp16, so we make the necessary changes for the
float version. Note that we are using mediump float, which ends up
using 16 bits floats anyways.
This provides 78% reduction in execution time (i.e. 4.6x performance
improvement). We see the EASU go from 4.2ms to 0.9ms.
The quality is slightly impacted.
Non sampleable buffers (Renderbuffers) are always rounded to 32 pixels
currently and on the backend side this buffer can end up being cached
as a sidecar buffer for MSAA. When it is created, it needs to be created
with the size of the texture it represents, not the size of the FBO,
which might be smaller.
In low-light conditions, peak luminance sensitivity of the eye shifts
toward the blue end of the visible spectrum. This effect called the
Purkinje effect occurs during the transition from photopic (cone-based)
vision to scotopic (rod-based) vision. Because the rods and cones use the
same neural pathways, a color shift is introduced as the rods take over to
improve low-light perception.
This function aims to (somewhat) replicate this color shift and peak
luminance sensitivity increase to more faithfully reproduce scenes in
low-light conditions as they would be perceived by a human observer
(as opposed to an artificial observer such as a camera sensor).
The night adaptation can be controlled using a 0..1 factor for artistic
reasons: `ColorGrading::Builder::nightAdaptation()`.
* Use wide gamut color grading and gamut mapping
This change introduces a new API on ColorGrading to enable or disable gamut mapping at will. This feature is currently off by default and while useful and recommended, it isn't cheap and we need to figure out what we want our defaults to be especially on mobile targets.
Gamut mapping can be combined with luminance scaling to properly control hue skews.
* Small cleanup
* More precise Rec.2020 conversion
- don't use a separate array for "reversedWindingOrder"
it's just one bit of data and we had 7 bits left in the Visibility structure.
- use Culler::result_type instead of hardcoded value everywhere
- other minor cleanups
The main changes are:
- shadowmap command generation is now done in the framegraph pass
itself, which allows it to be culled should the pass be culled.
- CameraInfo is created only once instead of twice / shadowmap.
It's not super heavy but also not cheap, as several matrices are
computed.
- simplified code.
src/details now only has the private headers of public types/classes.
Private/implementation classes have their header next to them under src.
This makes things more consistant.
This used to reside in View and it is now its own class. The reality is
that "per view" UBO is a bit of a misnomer -- currently we indeed have
one per View, but that's not required and in fact, in the future we might
want to have a separate one for shadow maps for instance.
A better name would be "per camera". Currently it is dynamically updated
when rendering the View and shadow maps. In the future this probably need
to be separated.
There was a confusion on utils::bitset<> API, it specifies the number
of words to use, not then number of bits in the set.
VariantList was sized to store 8192 bits instead of 128.
model/view matrices on Camera now store their translation as double
double precision. This allows filament to handle a very large world
space (thousands of kilometers).
Augment the TransformManager so it can maintain the 4th column as
double precision; we accomplish this by storing an extra float3 per matrix.
It's calculated as `doubleTranslation - float3{doubleTranslation}`.
We also add methods to set and get a transform as mat4, internally only
the 4th column is kept as double precision.
Finally when we calculate the worldTransform, we take this extra data into
account, but only for calculating the new 4th column, so the extra work
is small.
The new default values match almost exactly the filmic tone mapper which
is itself a close approximation of ACES. This result in a slightly more
contrasty and pleasing image out of the box.
- Images with an alpha channel are now properly supported. Blending is implemented directly in the image material by blending against the known opaque background color.
- Non-EXR/HDR images were not applying the proper transfer function and were therefore displayed incorrectly.
* Properly address the color grading LUT
With this change we properly sample from the texel centers when we should, thus avoiding a bias toward the low values. This was particularly problematic in the darks, resulting in darker results than they should have been.
* Save 3 scalar muls
We made this wrong assumption that views with a custom rendertarget
are never blended, but it's not true -- e.g. two views can use the
same rendertarget.
Fixes#4430
May impact #2802
* Use a clip space that matches our screen space for the z axis
We now set the projection matrix so that the clip space for the z
axis follows the inverted DX convention, i.e.: z_clip between 1 (near)
and 0 (far), which matches our screen space.
Note that the actual clip space in the GPU can revert to the "inverted
GL convention", i.e.: z_clip between -1 and 1, if the clip_control
extension is not present. In that case the precision benefits are
mostly lost.
Also note that there shouldn't be any user facing changes.
The projection matrix at the API level, still follows the GL convention.
Also, currently, the VERTEX_DOMAIN_DEVICE for materials also keeps
the GL convention (this is corrected in the vertex shader).
The gist of this change is to move the "reversed DX" mapping from the
vertex shader to the projection matrix.
* Update shaders/src/getters.vs
Co-authored-by: Ben Doherty <bendoherty@google.com>
Co-authored-by: Ben Doherty <bendoherty@google.com>
Camera::getProjectionMatrix and Camera::getCullingProjectionMatrix used
to return the projection matrix modified by scaling and shift, instead
they should return the value set by the user, especially when
setCustomProjection() is used.
Bring color grading back into the Rec.709 color space to match
previous behaviors. This change also implements an exact inverse
tone map function for the "Filmic" operator.
Bring color grading back into the Rec.709 color space to match
previous behaviors. This change also implements an exact inverse
tone map function for the "Filmic" operator.
* Fix specular AO for the sheen and clear coat layers
We used a single specular AO term for the sheen and clear coat layers,
computed for the base layer. This was creating occlusion that was way
too strong when the roughness from the top layers was very different from
the roughness of the base layer. The effect was particularly obvious
on clear coat materials (car paint for instance).
With this change we can also remove the "grazing angle fade" artistic
hack that was previously introduced to combat this effect.
* Safer code
Currently there are 8 light channels, both renderables and lights can
be associated to one or several light channel, lighting occurs when
at least one channel is active on both the light and the renderable.
Fixes#4275
When we associate a sidecar renderbuffer (for MSRTT emulation) to a
texture, the msaa count comes from the rendertarget -- if this texture
is later used with a rendertarget with a different MS count, it should
still work -- in this case we have to reallocate the sidecar buffer.
This helps prevent over-darkening but it's an artistic hack.
We need to check how we compute the bent normals as it seems we
find occlusion where there is no occlusion. The effect is
particularly visible on clear coat materials like a car's body.
When calculating the froxel z slice, a depth buffer value of exactly
1 (i.e. far plane) would map to after the last slice, this value
must be clamped.
made the code more readable, hopefully.
Fixes#4388
We now have an option for evaluating the bent normals from SSAO.
This is used to improve specular AO.
Currently these bent normals are not used for diffuse lighting because
they're "flat" (face normals), which is too distracting.
The bent normal buffer is stored in the SSAO texture, in a separate
layer. This is the first post processing effect that uses 2D array
in order to limit our sampler usage.
This CL also changes how we handle SPECULAR_AO_BENT_NORMALS when
material bent normals are not available: we're now using the "Cones"
algorithm with the regular normal, instead of Lagarde's approximation.
- depthUtils.fs and geometry.fs are actually completely generic
at this point, and could be moved out of the SSAO folder.
- saoImpl and ssctImpl are just a way to be able make several versions
of the SAO material easily (which will be needed soon)
This all started because the gl backend version of blit() didn't
handle color attachments other than COLOR0 *AND* would actually
corrupt the destination attachments other than COLOR0 if present.
This is due to how very stateful the gl API is.
Since this is a case we're not running into currently, the GL backend will
now assert in that case.
Updated the Vulkan and Metal backends to also reject blits where
color buffers other than COLOR0 are requested.
Ultimately, we need to change entirely our backend blit API so it
is less GL centric.
It looks like they are not working, at least with the current version
of the drivers and/or MoltenVK. It is not impossible that we have a
bug in how we set the subpasses as well. This will need more investigation
later.
SamplerGroup was comparing texture handles to decide if a texture needed
to be updated, however, texture handles are (quickly) recycled and
therefore can't be used for that purpose. e.g. if a texture is destroyed,
its handle could be reused quickly by another texture, if that texture
is now set on the SamplerGroup, it will ignore it, thinking it's not
different.
This fixes two bugs that combined in a way that prevented a double-free
error.
The destructor of the impl class for AssetLoader was never called, which
was sort of good, because it should not destroy the material provider.
The asset loader was never meant to have ownership over the provider.
SamplerGroup was comparing texture handles to decide if a texture needed
to be updated, however, texture handles are (quickly) recycled and
therefore can't be used for that purpose. e.g. if a texture is destroyed,
its handle could be reused quickly by another texture, if that texture
is now set on the SamplerGroup, it will ignore it, thinking it's not
different.
Previously, if you accidentally had two matdbg tabs open in Chrome, the
web app would hang. This fixes two bugs that prevented multiple
simultaneous clients:
1) The websocket handler broadcasted edit events to only the most
recent connection.
2) The number of server threads was only 2, but the actual web app
requires at least 2 threads for each instance, due to the
externally linked CSS file.
This can have a dramatic effect on SSAO and needs
to be adjusted based on the scene's scale/units.
The default value used to be about 6cm, and has
been changed to 5mm.
this allows the sidecar buffer to have the same lifetime than the
texture it represents, and therefore participate to the caching of
that texture.
Fixes#4338
* WIP New tone mapper API
* Implement tone mapper constructors and destructors
* Add new genertic tone mapper
* Make the generic tone mapper available in our sample UIs
* Fix warnings and crashes
* Fix generic tone mapper and graph mappers in the UI
* Add Java APIs for ToneMapper
* Implement copy/move operators for GenericToneMapper
This lets you see highly readable GLSL when transpiling the generated
SPIRV back to GLSL, with the original variable names preserved.
For now this is only useful when using `matinfo`, but it will soon be
useful in `matdbg` too.
The API now allows to create a standalone SkinningBuffer containing
Bones for skinning. This SkinningBuffer can even be very large and
contain bone data for several renderables.
New APIs on RenderableManager allow to switch to skinning buffer mode
and to set the SkinningBuffer (and window into it) to use.
Fixes#3238
* Generalize luminance scaling in color grading
Previously the luminance scaling step was tied to a specific tone
mapping operator, called "EVILS" in the API. This was however
somewhat misleading as EVILS really is a color handling system
that's independent of what we call tone mapping.
This change splits the luminance scaling step from the tone mapping
step, allowing the use of luminance scaling (the LICH part of EVILS)
with any tone mapping operator.
As a result, the operator known as EVILS is currently mapped to the
linear operator and renamed to "RESERVED" until it is replaced by
a proper — configurable — operator.
* Update release notes
* Fix web build
Saturation and vibrance were using the wrong luminance weights,
and this change switches EVILS to HK weighted luminance weights
to give better results.
What was happening is that we were doing two blits at the end of the
frame instead of one, those blits are setup so that they only work
when the destination is the an imported target (it doesn't have to be
this way, but currently that's the only case we should have). Because
of the 2nd blit, the first one's destination would not be the
imported target and it would fail.
The real problem is that we shouldn't be doing two blits, just one
is enough. That was happening because the logic to decide if we
needed the last blit was flawed.
Fixes#4306
Previously we were zero filling the data blob used to store all Vulkan
handles. This occured in VulkanDriver::alloc_handle().
There were several places that depended on the zero fill. These
have now been fixed up so that the handle can be constructed in a
more natural way.
Also: remove the OPENGL specific constant from HandleAllocator. This
breaks in non-CMake build systems (e.g. Bazel) that do not provide the
constant at a global level.
We now have:
allocate<> which just allocates the handle's memory
construct<> which constructs the handle allocated with allocate<>
allocateAndConstruct<> which allocates AND constructs the handle
destroyAndConstruct<> which destroys and constructs a handle in place
Normally you'd use allocate/construct but with synchronous calls, it
might be necessary to have an initialized handle immediately.
Previously, Metal was never performing any texture conversions during upload. For example, `iblprefilter` loads `FLOAT` image data into the `R11F_G11_B10F` texture format. This PR fixes the issue, refactors the upload pipeline, and adds a new backend test for coverage.
Image loading errors are handled more gracefully by showing a
background color.
Added UI to set the background color.
Replaced fragment discards with solid background color.
* Add new sample app: image_viewer
This app shows how to load and visualize images. It's also useful to work
on color grading and other post-processing effects by just loading an HDR
or EXR image as a reference.
This change also adds a new exposure setting to the color grading LUT.
* Fix comments and build
* Fix Metal
* Use high precision for the UV coordinates
* Fix refactoring issue
We were inserting the colorgrading subpass command between the
refracted and blended objects, instead of after all of them.
Another bad side effect of this was to trigger the refraction pass for
no reason.
* Add support for KHR_materials_volume
* Add ubershader support for KHR_materials_volume
* Update readmes
* User highp on anything that affects UVs
* Add missing precision qualifiers
* Fix transmission
* gltfio: fix precision in KHR_texture_transform
Fixes#4253.
* Update to the new order of parameters
Co-authored-by: Romain Guy <romainguy@curious-creature.com>
* API CHANGE: honor user-defined precision, rename SamplerPrecision.
matc now honors 'precision' on non-samplers in mat files, and
the filamat API now allows clients to specify precision for
non-samplers.
This involved flattening a union that is internal to filamat.
* MaterialBuilder: Add Java bindings for precision.
We were inserting the colorgrading subpass command between the
refracted and blended objects, instead of after all of them.
Another bad side effect of this was to trigger the refraction pass for
no reason.
We use a hashmap instead, which won't change the behavior of the
HandleAllocator (due to the increased size of the handles) between
debug and release modes.
We always may need to discard lights because we have a limited
number we can use in the shader. We now do this earlier, in particular
before we set up the shadowmaps.
Because we also have a limited (smaller) number of shadowmap, this
allows us to automatically select the ones that are closer to the
camera, which gives better results.
The most recently loaded model is now automatically re-loaded if you
re-start the sampler-gltf-viewer app.
Super useful in my case for `rgb-lights.glb` :)
* Add support for transparent shadows
This solution uses an 8x8 Bayer matrix. We could find a better
noise. We also need better filtering of the shadow maps. A PCF
is barely enough and VSM fails without blurring.
* Use a fixed Bayer matrix, fix pre-caching of depth shaders
* Use gradient noise instead of fixed bayer pattern
* Update docs
This is a trivial change since we moved the lightspace computation
in the fragment shader. The limitation to 6 is due to how we're storing
the culling result.
* Remove support for Java/desktop builds
These builds are never tested nor used on our end. We cannot
guarantee their proper support. It should also be possible
for an app to handle this itself.
* Remove Tungsten since it cannot be compiled anymore
When HandleArena is full, we now revert to a slower system-heap
based mechanism. The problem is logged once. This allows the
system to continue to run.
This means it can easily be used by our other backend.
HandleAllocator uses 3 pool (which currently can't be configured),
and is very efficient in terms of memory overhead and handle to
pointer conversion.
A Handle<> is just an offset into a large heap managed by the 3 pools.
* Add simple ReadPixels performance test.
This simply lets the gtest framework report the total test time, which
includes startup and teardown. However we perform 100 runs of the
render+readback code to reduce the run-to-run variance. For 100 runs of
a 2000x2000 buffer, I see the following results on my mac:
- Vulkan: 1881 ms
- Metal: 1625 ms
- OpenGL: 978 ms
The test can be built & invoked like this:
```
./build.sh release backend_test_mac && \
./out/cmake-release/filament/backend/backend_test_mac -a opengl --gtest_filter="*ReadPixelsPerformance"
```
See also issue #4246.
* ReadPixels test: minor refactor.
The operator!= in std::array compares SPLIT_COUNT elements, which
is potentially greater than cascadeCount, which was the number of
initialized elements in splitPercentages.
The operator!= in std::array compares SPLIT_COUNT elements, which
is potentially greater than cascadeCount, which was the number of
initialized elements in splitPercentages.
- the ShadowMap cache is now inline and doesn't use unique_ptr<>
- reduced the size of ShadowMap significantly, it used some temporary
storage as member, but now that we have many of those, we shouldn't
do that.
- removed some duplication of fields
- fixed some headers inclusion
Instead of managing a GrowingSlice<> more or less shared between
several RenderPass, RenderPass now works directly with an Arena which
owns the commands. We do this so RenderPass can be copiable, which is
useful for creating a RenderPass from a "template" RenderPass
(usually empty, but with some scene parameters set).
RenderPass is now a lot more self contained.
Execution of the commands is now done with a RenderPass::Execcutor
which can be obtained from a RenderPass.
This change led to some code clean-up as well.
We also had to improve Arena<> to be able to take a custom Area.
This adds some significant JNI glue in order to make the Java API for
gltfio materials much more consistent with C++, and create feature
parity between the two languages.
The motivation is to allow Java clients to provide their own
ubershader-based solution or filamat-based solution.
- The `MaterialProvider` Java class is now an interface.
- The `UbershaderLoader` Java class is now a concrete implementation.
- Users can now write their own provider class using Java / Kotlin.
To streamline the amount of JNI work that occurs at run time, the
MaterialProvider Java interface uses ints in a couple spots instead
of enums. This sacrafices some readability for Java clients who wish
to write their own provider, but makes the JNI much simpler.
This replaces #3975.
We have a limited number of interpolants, and 2 were used for spotlights,
but this is a fairly uncommon case, so we move the calculation to the
fragment shader. A side effect of this is that we are not limited to
two spot-light anymore from the shader's point of view.
This essentially adds a vector transform per fragment, but on the other
hand, it removes communication between the vertex and fragment shader,
as well as interpolation of the position.
This basically expresses the fact that we won't need an "auto-resolve",
indeed the temporary depth buffer is always discarded. In the GLES
backend, this is likely (if MSRTT is supported) to take a less
complicated code path.
Improve some comments.
- the clear color was broken since we switched dot EVSM
- use the scissor instead of the viewport, it makes it easier to compute
the uv-scale.
- the min/max lod level wasn't reset properly
- also clamp the minimum shadowmap size to 8, and document it.
We make better use of the framegraph, by using a pass per shadowmap,
instead of generating all shadowmaps in a single framegraph pass.
This allows us to use less temporary memory when blurring, but also
simplifies the code.
This is still somewhat work-in-progress, but good enough to be useful.
Some caveats:
- large blurs don't work well with cascades
- the implementation uses more memory than it could
This allows clients to obtain the extras strings for the asset and
for individual nodes. Note that extras for buffer views, accessors, etc
are not supported, although C++ clients can access the raw cgltf data
directly if they need to.
Filament uses a handful of known UBO, the code for creating their
interface block is not needed in filament (via filabridge), so
reorganize things so that we can move that code to filamat.
On the other hand, filament needs the C structure corresponding to
the UBO, to find the fields offset and ubo size and name.
We now have a UibStructs.h in filabridge which only defines C struct
with a static name for the interface block's.
This should slightly reduce the size of filament proper.
We also now have a new helper TypedUniformBuffer<> which simply holds
one of these structs (or an array), and allows to set the fields
directly and more naturally from C++. std140 alignment is now left to
the caller, when going through that class.
- use EVSM (exponential VSM), helps a lot with light bleeding
- don't apply shadow biases with VSM
- improve VSM user settings: added more controls, expose to java
- minor code cleanups:
- rename some structures
- remove unused parameters
- remove unused uniforms
VSM requires shadow receivers to be in the shadowmap, instead of
handling this in the shadow mapping code, we're handling it in
FScene::prepare(), which guarantees it'll ripple everywhere it needs to.
When creating a texture with a swizzle, the swizzle should be applied to
the image view only when sampling from that texture. When a swizzled
texture is used as a render target attachment, the image view should not
have any swizzle applied.
- OpenGL ES on Android
- Metal on MacOS/iOS
- Vulkan everywhere else (Linux/Windows)
The main change here is that OpenGL is no longer the default on
desktop.
Most of our samples have a `-a` option that can be use to specify the
desired backend.
1) Add a missing destroyTexture to the test, which I believe fixed
the Metal assert that I was seeing, as well as some Vulkan validation
warnings.
2) Add a flushAndWait() utility to BackendTest, which is useful for
diagnostics.
3) Fix a layout validation issue in VulkanSwapChain for empty render
passes.
This reworks how we compute the bloom buffer size. The main changes are:
- The BloomOptions.resolution is changed to height instead of minor axis.
This keeps a consistant look of the bloom regardless of the viewport.
- The anamorphism setting now dilates the bloom rather than to compress
it (in the other direction), which maintains the same amount of bloom.
However, the resolution decreases meaning that the user might need to
adjust resolution and/or levels.
- limit the resolution do 2048 which is gles minspec.
We were disabling the color pass render target clear when a skybox
was present. This optimization doesn't work when the viewport is not
full. Currently there is no way to know that, so we disable this
optimization for now. Performance loss is not expected to be significant
if present at all.
We were automatically clearing the color pass render target when
the view was translucent -- the idea was that when it gets blended at
the end it wouldn't override the content of the render target.
Unfortunately, this doesn't work when the rendertarget is imported,
because in that case, it's not actually blended, we're just rendering
into it directly. The clear would then erase the previous content.
This is fixed by moving the clearing decision in the execute closure and
piggy-backing on the computed discard flag. Discard means clear.
The discard flag will be set for newly created buffers and not for an
imported target that needs to keep its content.
An imported target shouldn't specify the discardStart flags, those
are calculated from the graph. However, it needs to specify "keep" flags,
to take into account constaints that the graph cannot see (due to the fact
that it's imported into the graph). i.e. the imported target may have
content that needs to be preserved.
An imported target can specify a clear color and clear flags that will
override the flags from the logical Descriptor -- this is a way to
delay the swapchain clear to when we're actually rendering into it
(which usually happens in the very last pass).
When the source of a blit refers to an imported render target,
we can't create a HwRenderTarget from it because the resource
is not a real texture and doesn't have a HwTexture.
However, it can be converted back to the HwRenderTarget it
represents, using the FrameGraph's declareRenderPass API.
It's a slightly strange usage of this API, but not technically
wrong.
This is the main fix for bug #4085
* Materials can now provide custom lighting/shading
When a material uses the "lit" shading model, customSurfaceShading can
be enabled to replace Filament's lighting implementation. When this
feature is enabled, the material *must* provide the following function
in the fragment shader block:
vec3 surfaceShading(
const MaterialInputs materialInputs,
const ShadingData shadingData,
const LightData lightData
) {
return vec3(1.0); // custom lighting here
}
Please refer to the docs in Materials.html for more information about
this feature and the different values provided by the data structures
passed to the function.
* Update docs
* Update docs/Materials.md.html
Co-authored-by: Philip Rideout <philiprideout@gmail.com>
Co-authored-by: Philip Rideout <philiprideout@gmail.com>
This makes the car load in 3 seconds instead of 30 seconds on Mali.
On devices that have bufferImageGranularity set to 1 in the
VkPhysicalDeviceLimits caps structure, we observe that vmaCreateBuffer
can be two orders of magnitude faster than with devices that have a
higher value for their bufferImageGranularity.
According to the Vulkan spec, the bufferImageGranularity aligment
constraint needs to be honored only when linear and non-linear resources
live next to each other in memory. Since all VkBuffer objects are
linear, we can allocate them all from the same pool and ignore the
alignment constraint.
VK debug markers do not show up in an AGI system trace, but we can sneak
them in by using debug_utils to assign labels to VkFramebuffer.
Note that the debug_utils extension is enabled only when launching your
app from AGI.
More precisely, this logs the time that the Filament backend takes
to process all the commands that were issued in the gltfio synchronous
loading phase, which mostly consists of creating and populating vertex
buffers.
This can be quite slow for vertex-heavy (CAD-style) models, and seems to
vary by several orders of magnitude from device to device on Vulkan.
This constant was previously set to 3 because of triple buffering, which
does not account for the fact that Filament may issue multiple commit()
calls during a single frame.
This new number was determined via experimentation.
On a Pixel 4, this reduces a 4.5 ms gap in "GPU Queue 0" to 420 us.
We reuse the current job for the right side instead of spawning a new
job, which has overhead. This also cuts in half the number of times we
need to copy the user functor.
The overhead of the jobsystem / job is about 50-100us, so we should
try to use jobs about an order of magnitude bigger -- which is actually
quite rare.
In particular the culling is so efficient that we'd need to process
10000 or more primitives for the jobsystem to be of any help, so for
now we never use the jobsystem for culling.
A side effect of this flag was that we had to always use notify_all
when starting a job, instead of notify_one -- in practice, we found
experimentally that more calls to notify_one is cheaper than fewer
calls to notify_all.
If your build system does not provide a definition for this macro, then
you cannot compile CommandBufferQueue.cpp.
Since Allocators.h provides a backup value, it needs to be included.
std::allocator::deallocate() expects the same value that was given
during allocate().
Interestingly, this bug did not manifest any issues (even with ASAN) on
some platforms.
std::allocator::deallocate() expects the same value that was given
during allocate().
Interestingly, this bug did not manifest any issues (even with ASAN) on
some platforms.
This fixes an issue caught by a memory sanitizer, whereby padding in the
cache key structure was being consumed by a hashing function. In
practice this oversight was fairly harmless, but this change makes the
code more safe and predictable.
Fixes the "triangle count is zero" assertion with the big car model.
We do have nominal support for points and lines, but we should not
attempt to compute tangents for these.
As a reminder, glTF also supports loops, strips, and fans; this is the
only way in which Filament is non-conformant.
The CommandStream hold backend commands, but it has a limited size.
With some large models, we can run out of space especially during
initialization.
We can combat this by flushing the command stream regularly. With this
change, we do this every 128 renderable creation.
We also flush the CommandStream just before generating the draw
commands for the frame.
If the buffer is not sized correctly (currently at compile time), this
could cause stalls, which will be logged in DEBUG builds.
On Android, set `debug.filament.backend` to the desired backend:
0: default
1: gl/gles
2: vulkan
3: metal
4: noop
example:
`adb shell setprop debug.filament.backend 2`
This will affect all filament apps on the device.
This sets the backed as if it was set from the public API.
* fix runtime warnings about certain NSView methods being called off the main thread
* address code style issues
* fix code style and threading concerns
* Remove unneeded __strong qualifiers, put __weak after the star as recommended by Apple, convert weak to strong before testing for nil to avoid a race
* convert tabs to spaces
All (and only) git-ignored files under android/ are deleted.
This is to prevent a recurring problem where "clean" builds were not
actually clean in some cases.
This reverts commit cc6201b0db.
On Android the loggers are not always initialized when several .so are
used, leading to crashes. It's not completely clear what is exactly
happening.
This fixes most of the warnings that are emitted by the relatively new
SYNCHRONIZATION_VALIDATION feature that LunarG provides.
This commit does not fix a warning that is emitted when SSAO is enabled,
when we sample from the currently bound depth attachment. This one is
tricky to fix.
the CommandStream is limited in size and can get corrupted if
that size (default 1 MiB) is exceeded. We used to catch this
only on Debug builds, but this is too serious and unrecoverable, so
it must be caught in Release builds too.
At the point of detection the stream is corrupted, which can cause
the backend to get in an unknown state. But worse, program execution
could end up anywhere because the stream contains function pointers.
- libutil's slog, cout, cerr, cwarn and cinfo were not thread safe
at all.
- they are now thread_local
- for that reason the log buffer (which now exists per thread) is not
allocated until it's needed and is aggressively shrunk.
- removed our thread_local emulation which was incomplete.
- don't expose LogStream publicly.
This is (almost) drop-in replacement for std::vector, the main
difference is that FixedCapacityVector has a fixed (at runtime)
capacity, which makes it a lot more efficient and small for operations
like push_back();
The capacity can be set (and re-set) as usual with reserve(),
but exceeding the capacity when adding items will throw.
The capacity is also set when creating the vector with a size.
In addition a FixedCapacityVector<> instance is only 16 bytes instead
of 24 for std::vector<>.
The aim here is to reduce code bloat.
Newer versions of MoltenVK seem to behave similarly to Android,
so there's no need to separate codepaths. I tested this by not only
resizing the window, but also dragging between displays with different
pixel ratios, and rotating on Android with & without `orientation` in
`configChanges`. I also tried our headless unit tests.
The sample would crash when removing `orientation` from `configChanges`
and rotating the device, because the old RemoteServer would still be
alive when creating the second one, so the WebSocket port would be
unavailable.
This has no behavioral changes, it's mostly just moving code around,
except that VulkanDriver::createSwapChain() is simpler now, and some
of the logic in beginRenderPass() makes more sense too.
This removes the useless `VulkanSurfaceContext` structure by moving
its data directly into `VulkanSwapChain`. Also created a new cpp file
just for the swap chain wrapper, which makes `VulkanContext` much less
unwieldy.
This change allows access to the Engine's underlying Platform object. This is useful for Metal clients to access a new method on MetalPlatform, createAndEnqueueCommandBuffer.
* a way to keep the content of custom RenderTarget attachments
Until now, all attachments of RenderTarget beyond COLOR0 were discarded
after rendering.
With this change, attachment whose texture has the SAMPLEABLE usage
bit set will be kept.
Fixes#3962
This adds support for decoding HDR files, dropping them into the remote
web page, and sending them over the wire. However the decoded results
are not yet passed into IBLPrefilterContext, that will be subsequent
PR.
Rename KtxLoader to KTXLoader.
The remote UI has been tweaked a bit, so that it shows a more helpful
message when disconnected.
(1)
The C++ RenderTarget API might be used by an external Android library
such as filament-utils-android and therefore needs to be in the map
file.
(2)
Texture, Skybox, and IndirectLight each had two "nativeObject"
constructors which exist for the benefit of filament-utils-android,
but each of these classes only need one nativeObject constructor.
Vulkan does not permit the GENERAL layout when the depth buffer that is
currently being used for depth testing is also sampled from.
This (finally) fixes the last validation error that we were silencing,
which would occur only when SSAO was enabled.
deprecate Stream::stream(intptr_t).
This method was needed for ARCore back in the days, but there is now a
zero-copy way to achieve the same thing. This API shouldn't be used anymore.
The code that determined the "base slot" (i.e. the buffer object index
for the first morphed attribute) was very fragile and buggy.
MorphHelper now simply looks up the slot index in a stored mapping,
rather than trying to mimic the AssetLoader to determine the base slot.
Until now, most of the models we tested against only morphed the
position attribute, and never had a large set of target weights that
could be non-zero in a given time slice.
MATERIAL BREAKAGE: this change breaks materials
In order to use one less sampler, we now use a UBO instead of a texture
to store the "froxel records". Currently this UBO is limited to 16KiB
vs. 64KiB before with the texture.
We also handle running out of record space better, by using a predefined
record that has all the lights in the scene. This way the scene will be
rendered properly, albeit at a potentially large performance cost.
* Allow using specific NDK version, for CI don't use latest but required
* Default to highest available NDK version within same major release sequence
* Refactor NDK version selection
This fixes the "descriptor set fragmentation" error that we would see on
Mali.
The pipeline cache now maintains an arena of descriptors that can be
re-used. We call this an "arena" to avoid confusion with
VkDescriptorPool.
Descriptors are returned to the arena only when they are not in use
by any pending command buffer. This allows them to be mutated safely
without worrying about synchronization.
The API of the pipeline cache is now easier to use. Internally, it now
maintains separate pipeline and descriptor bindings for each command
buffer, since that's how Vulkan works.
We should take care not to call glVertexAttribPointer when there is
no bound ARRAY_BUFFER (i.e. when its binding is zero).
This fixes the black screen seen with some WebGL samples after
the recent memory leak fix related to the new BufferObject API.
If expecting a bool and receiving anything other than true or false,
this is invalid and should fail. This change triggered a unit test
failure which uncovered several actual problems.
We should take care not to call glVertexAttribPointer when there is
no bound ARRAY_BUFFER (i.e. when its binding is zero).
This fixes the black screen seen with some WebGL samples after
the recent memory leak fix related to the new BufferObject API.
It did improve the look of the blur on far way geometry in some case,
but overall it hurts a lot more.
The case that falls appart is when everything is in the background layer
but there is a large depth difference anyways.
This creates a halo around edges.
Fixes#3221
* BackendTest: add blit cubemap test.
This test is similar to manual mipmap generation for cubemaps and
can uncover issues that are provked by libiblprefilter. It simply
dumps out a PNG that can be manually inspected.
Also, this test does not do makeCurrent / commit and is therefore
similar to renderStandaloneView.
* Update filament/backend/test/test_Blit.cpp
Co-authored-by: Ben Doherty <bendoherty@google.com>
Co-authored-by: Ben Doherty <bendoherty@google.com>
This leak was introduced in the following PR on April 7.
https://github.com/google/filament/pull/3775
The guilty party has been contacted and properly admonished for his
transgression.
This was tested by adding the following code after applyAnimation in
gltf_viewer.cpp
static int nframes = 0;
if (!gpath.empty() && nframes++ > 100) {
static int count = 0;
printf("reloading %d\n", count++);
nframes = 0;
app.resourceLoader->asyncCancelLoad();
app.resourceLoader->evictResourceData();
app.viewer->removeAsset();
app.assetLoader->destroyAsset(app.asset);
loadAsset(gpath, app);
loadResources(gpath, app);
}
This leak was introduced in the following PR on April 7.
https://github.com/google/filament/pull/3775
The guilty party has been contacted and properly admonished for his
transgression.
This was tested by adding the following code after applyAnimation in
gltf_viewer.cpp
static int nframes = 0;
if (!gpath.empty() && nframes++ > 100) {
static int count = 0;
printf("reloading %d\n", count++);
nframes = 0;
app.resourceLoader->asyncCancelLoad();
app.resourceLoader->evictResourceData();
app.viewer->removeAsset();
app.assetLoader->destroyAsset(app.asset);
loadAsset(gpath, app);
loadResources(gpath, app);
}
- remove areFeedbackLoopsSupported from backend API, since this is only
a problem on WebGL
- binary only include either the regular or webgl version of the bloom
code, never both, which should save some binary size
- bloom pass is now called outside of the colorGrading pass, which is
cleaner and makes it available for other pass, should they need it
(not the case right now).
- always pass ColorGradingConfig to all color grading related methods,
instead of individual parameters
- very minor optimization in bloom shader
- move kernel weight computation to the gpu side as a small cost, in
order to simplify the CPU side and reduce the binary size.
- skip samples with a weight of 0, which is a completely coherent
check.
- fix a typo in move ctor
We chose 8 renderbuffer because it seems to be commonly supported
on mobile gpu and is more than 6 which is needed for generating
cubemaps. 4 was the minimum guaranteed by ES3.0.
The main changes here are:
- TargetBufferFlags is now a uint32_t instead of uint8_t
- The number of render buffer supported by filament and supported by
the hardware can now be different. Applications/users now must
query this value if more than 4 is needed.
- added a backend query (used gl terminology)
This is a redo of a previous attempt. There was an issue in the GL
backend that caused MRT to fail on some devices.
This removes a lot of caveats that we had in the Vulkan backend. Many
driver API calls were unsupported due to a lack of command buffer, but
in the new scheme, a command buffer can *always* be made available,
and there's no weird separate "work and swap chain command buffers".
Here's how various driver API entry points now map into Vulkan actions:
```
makeCurrent => Acquire Swap Chain.
beginFrame => Do nothing.
endFrame => Submit command buffer and create a new one.
commit => Submit commit buffer again if necessary, Present Swap Chain.
```
This handles both copyFrame (which does not do beginFrame / endFrame)
and renderStandaloneView (which does not do makeCurrent / commit).
This makes release builds consistent with continuous builds and default
local builds. It also makes it easier to test the Vulkan backend.
Clients who need a smaller APK size can always make their own build.
This also makes filament-android-release.aar go from 1.8M to 2.3M.
Note that we are now including bluevk, vkshaders (for the blit shader),
smol-v (for SPIRV decompression) and vk_mem_alloc (for mem allocation).
Simply tweak the "spiralTurns" parameter, empirically to improve the
noise pattern produced by the gradient noise. In turn, this helps the
gaussian filter.
We used to have that before, but it was hardcoded and depended on
mobile vs. desktop.
With this change, users can set the quality on the material itself.
The default match the current settings. There are 3 quality levels:
low: enables optimization that might not be 100% correct
(e.g. abs(x) instead of sqrt(x*x))
normal: like low, but doesn't sacrifices correctness.
high: could be improved upsampling, etc...
- create a new FILAMENT_HAS_FEATURE_TEXTURE_GATHER, which can be used
in the shaders to use textureGather.
- A new TARGET_GLES_ENVIRONMENT and TARGET_GL_ENVIRONMENT in addition
to TARGET_METAL_ and TARGET_VULKAN_, for completeness.
- Rename TARGET_LANGUAGE_SPIRV to FILAMENT_VULKAN_SEMANTICS and add
FILAMENT_OPENGL_SEMANTICS for completeness.
- expose the shader model to shaders as: FILAMENT_SHADER_MODEL_ES30 and
FILAMENT_SHADER_MODEL_GL41
Use these new #define where appropriate.
The main change here is that TARGET_MOBILE is now only used for dealing
with the `mediump` default on mobile.
This change also enables textureGather on Metal.
We chose 8 renderbuffer because it seems to be commonly supported
on mobile gpu and is more than 6 which is needed for generating
cubemaps. 4 was the minimum guaranteed by ES3.0.
The main changes here are:
- TargetBufferFlags is now a uint32_t instead of uint8_t
- The number of render buffer supported by filament and supported by
the hardware can now be different. Applications/users now must
query this value if more than 4 is needed.
- added a backend query (used gl terminology)
* New experimental tone mapper
The new tone mapper proposed in this change replaces Uchimura (see
below) and is composed of two distinct algorithms:
- The first is the customizable compression curve from Timothy Lotte
(from GDC 2015). The curve's default parameters were chosen to
match the overall appearance (contrast) of the ACES curves used
as our current default tone mapper. This was done to provide a
good migration path to making this new solution the default in
a future version of Filament.
The compression curve can be customized via its contrast, shoulder,
input middle gray, output middle gray, and maximum "HDR value"
(scene referred).
It is important to note that the compression is only applied to
the luminance of the input scene referred value.
- The second is a novel approach by Troy Sobotka called Exposure
Value Invariant Luminance Scaling (or EVILS). This solution works
on the largest chromatic channel of the input value as opposed
to mapping all RGB channels as we do in our other tone mappers.
The algorithm effectively scales all channel by a ratio that
depends on both the compressed luminance from the first stage
and the luminance of input value normalized by the largest
channel.
This new solution offers important benefits as it manages to
preserve chromaticity for high input values. It is also designed
to provide a desirable rolloff to white on overexposure. This
means that even without gamut mapping, this solution avoids many
hue skews (red to orange, blue to purple, etc.). The results
provided by this solution are much more natural and make using
other color grading tools much easier.
* Use std::pow instead of pow
* Fix Web builds
- move public headers from filament/foo to filament-foo/
to avoid confusion with libfilament's headers
- add support for equirectangular to cubemap conversion
- add support for using an .hdr file directly in all our samples
Fixes issue seen with Vulkan + the quantized version of Avocado.
We already support the quantized glTF extension, which is when this
comes up. Technically this is only needed for Vulkan but gltfio does
not (and should not) know which backend is being used, so we simply
apply these transformations assuming the worst.
In practice this code usually won't be activated, and if even when it
is, it won't hurt loading time as much as other things like large
textures.
* Add Transcoder API for C++ clients.
We will add JavaScript and Java support in a later PR.
* Transcoder: use fixed inner loop in some cases.
This lets clang produce better ARM code for common cases.
* Transcoder: stronger typing in Config, etc.
* Transcoder: add UTILS_RESTRICT and noexcept.
This is the first step at trying to un-hardcode the number of drawbuffer
we support. The ultimate goal is to go from currently 4 to 8.
Until now the maximum render buffer supported was the same than the actual
number of render buffer supported. At some point in the future this will
change, it this will have to be dynamic.
This change is to allow rendering a cubemap in a single pass on h/w that
support it. Many mobile GPU support 8 draw buffers.
- the miplevel wasn't computed correctly
- the random rotation was incorrect
- the default HDR compression was too aggressive
- better validation of the input cubemap
we now store the kernel for each lod in a texture, this has several
advantages:
- the texture can be reused, so we can amortize that cost a bit
- we could generate the texture with a shader, so that we don't
have to do any texture uplaod (TBD).
Until now we were using TARGET_MOBILE in our shaders for both
quality setting and correctness.
With this PR we introduce a new define FILAMENT_QUALITY which currently
can take two value FILAMENT_QUALITY_LOW and FILAMENT_QUALITY_HIGH and
we use that instead of TARGET_MOBILE for quality settings.
Currently, FILAMENT_QUALITY_LOW is selected when TARGET_MOBILE is set.
New Api:
- Renderer::renderStandaloneView() is a new method that can be used
outside of beginFrame/endFrame on Views that have a RenderTarget
associated. This can be used as a poor man's compute API.
Improvements to Renderer:
- more aggressively skip lighting/shadowing code that won't be
needed by the current view
- Allow up to 64 KiB to be allocated from the render stream, up from
1 KiB
Even though the vertex shader comes before the fragment shader,
both bits needs to be present in the barrier, since a Vulkan
implementation might partition its uniform cache between stages.
Thanks to Chris Forbes for pointing this out.
After eyeballing this barrier I realized it was configured incorrectly,
but it did not seem to manifest as an actual error anywhere.
Basically, it needs to ensure that the UBO is fully updated before it
is read by the vertex shader associated with the next draw call.
This will allow folks to drop a GLB into the following page to get a
WebGL preview. It always uses the latest version of Filament via unpkg.
https://google.github.io/filament/viewer
After the next release we can also use the `<filament-viewer>` component
itself from unpkg and remove the local copy.
- If a resource file is missing, show toast instead of crashing.
- Allow glb files in zip files.
- Run blocking calls like createTempFile() on IO thread.
It's an error to ask for a higher MSAA sample count than supported by
the gl driver, and results in corrupted rendering.
We now always clamp the requested sample count to the maximum allowed
but the gl driver.
includes some minor refactoring.
Some drivers claim to support anisotropic filtering but don't allow
setting it via glSamplerParameter. Until now we were disabling
anisotropic filtering entirely on these drivers at compile time.
With this change, we should detect at runtime which drivers have this
issue and use a workaround -- instead of disabling anisotropic
filtering entirely.
- it was called on the main thread side, which didn't allow to,
for instance, systrace commands on the backend thread side.
- moved to a begin/end callback model
- implemented basic logging/systracing for all backends in Driver.cpp
(and removed this functionality for the metal driver).
note: refactored CommandStream class attributes only to work around
a clang bug we running into in CI.
This includes changes to OpenGL, Metal, and Vulkan backends.
At the backend level, vertex buffers are now always composed of buffer
objects. This lets us simplify the Driver API and some bookkeeping in
the backends.
This change also splits MAX_VERTEX_ATTRIBUTE_COUNT into two constants
because the maximum number of bound buffers is a separate concept from
the maximum number of attribute semantics (e.g. consider interleaving).
For now these two constants are set to the same value.
We also now store a byte count in HwBufferObject, which allows us to
remove the byte count from the Metal-specific handle, and to add some
asserts to debug builds to prevent size overflow.
- DoF: don't convert gl_FragCoord to mediump when generating noise
- SSAO: add a bunch of missing highp qualifiers
- all matrices and UV should be highp
- Also fix some typo in comments.
The focal length slider now works over a remote connection. Also, the
slider for focus distance now controls the value in `ViewerSettings`
(which gets applied to Camera) rather than the value in
`DepthOfFieldOptions` (which is deprecated).
according to the gl spec, we can't mix the EXT versions of
glRenderbufferStorageMultisample/glFramebufferMultisample with
the regular versions. Other drivers are more permissive.
When allocating a renderbuffer and the extention is supported, we always
use it.
It looks like the layout(location) is needed for output parameters of
the fragment shader in GLES. Adreno dirver didn't seem to care, but
Mali drivers do.
This fixes shaders that use multiple out buffers in debug mode.
This adds the 565 enum to a switch block (which is an actual bug),
and it fixes a JavaDoc warning about missing summary fragments (which
is just a silly warning).
Just for fun, this is an example of a web component that allows you to
embed a Filament viewer with a single HTML element. It does not have
many features, but is much easier to use than our raw JavaScript API.
* ImGuiHelper supports client-provided context
this lets impress load >1 embedded font into the font table
* ImGuiHelper supports client-provided context
this lets impress load >1 embedded font into the font table
Co-authored-by: Philip Rideout <philiprideout@gmail.com>
We now annotate the JS wrapper for Engine with a GL context handle.
This is fairly hacky because emscripten hides its internal GL object
from JavaScript, and it does not provide a mechanism to pass JS objects
into inline assembly.
This script makes it easy to update the web page at
https://google.github.io/filament/remote/ even if you do not have the
emscripten SDK. It works by downloadinug the latest continuous web
build, then placing the wasm file in the right place.
After running the script, the user must make a git commit to finish the
operation.
This is a big hackish, but is intended to help us debug scaling issues,
this adds a couple of sliders under [Debug] to control dynamic
scaling manually.
Shockingly there were no unit tests that did a simple write-then-read.
This is important because it's easy to forget details such as trailing
commas and the fact that uint8_t is a char.
NOTE: after Metal and Vulkan support lands, we can potentially simplify
the Driver API by removing some of the old VertexBuffer entry points.
This introduces a new API-level object called `BufferObject`, and a new
backend-level object called `HwBufferObject`. This encapsulates a single
VBO. It's especially useful when clients need to share data between
multiple VertexBuffer objects. (e.g. for morphing).
In the future, buffer objects could perhaps be used for non-vertex data
(e.g. for compute).
The existing `VertexBuffer` class can now be configured to use buffer
objects by calling `enableBufferObjects` on the builder. By default its
API is unchanged. If buffer objects are enabled, then clients should use
`setBufferObjectAt` (which takes a buffer object) rather than
`setBufferAt` (which takes CPU data).
OpenGL is the trickiest backend for this, due to existence of VAOs.
The implementation in this PR uses a version-tracking scheme to
figure out when to update the VAO.
We use the technique used by Jimenez for the foreground applied to the
background, where we "guess" the missing background information from
neighboring pixels. In several papers it's not deemed useful for the
background, however, it does improve the blur quality a lot very shallow
DoF, i.e. when a moderately blurred background sits over a strongly
blured one.
Before we were keeping a foreground and background version of the
low resolution source, but in practice it really doesn't make a big
difference and we save a fair amount of bandwidth this way.
The bulk of the change is to remove a texture/target which mostly
affects the downsample and mipmap passes.
Also renamed some variables to be less confusing (MaxMin to MinMax for e.g.)
and fixed a few typos.
there has been a typo in the DoF code a while, but fixing it looks
much worse. we need to understand why. so for now we at least
document the typo in the code.
The gist of it is that we can't always accumulate samples naively, and
we have to use different strategies with different tile types.
For the fast tiles it seems enough to reorder the sum, accumulating each
ring first, and then accumulating that into the output, because all
samples have the same weight.
For the foreground tiles, for some reason I couldn't find any major
issue and I'm not sure why.
For the background tiles, we need to use highp intermediates because
every sample has different magnitudes (weights)
* move the focus distance from DofOptions to Camera
The DofOption parameter is now deprecated, but will still work if the
focus distance is not set on Camera.
* A few new helper APIs on Camera
- getFocalLength() which returns the focal length used internally for
DoF computations.
- computeEffectiveFov() and computeEffectiveFocalLength() which can be
used to better simulate a real camera's FOV changes with the
focus distance.
Modified gltf_viewer so it uses the effective fov/focallendth when
DoF is active.
* simplify setLensProjection
We can call setProjection directly instead of converting to a fov
first, which ends up performing atan(tan(x)).
Also improve parameters names.
This adjustment is conservative, i.e. we limit the number of rings in
the kernel based on the kernel size in a way that it won't impact the
result. For instance, if the kernel size is 2 or lower, we don't need 4
rings.
This improves performance significantly, especially with large maximum
right counts.
As far as I can tell, there is no need for the `enabledAttributes`
argument in `setRenderPrimitiveBuffer`. This helps pave the way for
the upcoming BufferObject API.
- setImage() used the width of the texture for calculating the required
user buffer size, but this was too large when only updating a
portion of the texture. It would result in a BufferOverFlowException.
- setImage3D() had the opposite problem where it didn't take into
account the depth of the texture.
- the offset in the user buffer was called "bottom" on the C++ JNI side,
but it is called "top" on the java and filament side.
Fixes#3685
Currently the only NONE and MEDIAN are available options, at some
point we may add a "MAX" option which is cheaper.
There is no real practical uses for this option with these only two
choices, other than for debugging.
This is mostly used for debugging and having somewhat of a reference.
Change to the shader are minimal because all CoC values were always
treated as full-resolution. We just need to replace the downsampling
shader by one that just calculates the CoC. And we need to introduce
a CoC scale factor to compute the LOD to use (this was hardcoded
before).
On the C++ side, it's mostly a matter of replacing hardcoded buffer size
scales by a runtime or compile time variable.
* CString can construct from a string_view
(string_views lack a null terminator)
* Update libs/utils/src/CString.cpp
Co-authored-by: Mathias Agopian <pixelflinger@gmail.com>
We're implementing the two-layer technique described in
Jimenez DoF paper, where the foreground field is split in two layers, front
and back. In addition we compute accurately the foreground field alpha.
From a performance standpoint, most of the extra work is increased
register pressure and a bit of ALU.
Always store min CoC in the red channel and max Coc in the green
channel for both the CoC and Tiles textures.
(The tiles texture was doing the opposite).
this would happen only when a fast tile had a CoC < 2, because it
would then have an opacity, but we were ignoring it. It's color then
didn't match the color of the surrounding layer and the median filter
would spread a pixel from the surrounding layer into the fast tile.
we simply fix this by (1) generating the alpha/color for fast tiles,
which is easy to do and on the combine pass, skip the fast tile
combine when the tile has a an opacity.
before we were not doing that because it caused some aliasing
artifacts with mipmapping, however it turns out that there are artifacts
anyways, and using the proper weights actually reduces the most severe
ones.
the artifacts are due -- it seems -- to the low density of the kernel.
The hang was caused by a subtle race. When a job is completed, its
thread must signal all the threads that might be waiting on this job.
The signaling code was attempting to signal only the minimum number
of threads -- this was important especially in the case where no threads
were waiting, then the call to notify() could be avoided.
Unfortunately, for performance reasons we're not calling notify() with
the condition lock held, this meant that between the time the number of
waiting threads was latched and the time of the notify() call, more
threads could enter their condition variable wait(), and it would
then be possible for these threads to wake up, instead of the thread
we were trying to wake up (the one waiting on the job).
It would then get stuck forever.
This bug was introduced in 2df639133b
Also add some debugging code for this kind of failure (disabled)
* Fix blanking windows not being drawn into on macOS
* Refactored to create swap chain info object; added fixes for resizing; added sample.
* CocoaGLSwapChain should inherit from Platform::SwapChain
* Link to SDL directly for multiple_windows example
* SDL apparently redefines main to SDL_main and requires C++ linkage
* Fix main to have SDL-approved signature
* gltfio: Add internal MorphHelper, enable up to 255 targets.
Previously, gltfio could not animate models with more than 4 morph
targets. Now we can handle up to 255 targets, as long as no more than
4 targets are ever used simultaneously. This allows us to handle
many more models, such as those exported by Modo, and the horse model
in #1852.
We do this by picking the highest 4 weights. The indices for these
weights are called the "primary indices", and we construct a new
VertexBuffer for each unique set of primary indices.
In the near future this will be made more efficient by adding a new
VertexBuffer API to the Filament core.
Fixes#1487.
* Code review fixups.
The hang was caused by a subtle race. When a job is completed, its
thread must signal all the threads that might be waiting on this job.
The signaling code was attempting to signal only the minimum number
of threads -- this was important especially in the case where no threads
were waiting, then the call to notify() could be avoided.
Unfortunately, for performance reasons we're not calling notify() with
the condition lock held, this meant that between the time the number of
waiting threads was latched and the time of the notify() call, more
threads could enter their condition variable wait(), and it would
then be possible for these threads to wake up, instead of the thread
we were trying to wake up (the one waiting on the job).
It would then get stuck forever.
This bug was introduced in 2df639133b
Also add some debugging code for this kind of failure (disabled)
we can't use a projection with infinite far as a custom
projection matrix, because this is not suited for culling.
recently a new version of setCustomProjection was added which allows
to specify both the rendering and culling projections.
This wasn't very useful in the first place because we're recycling
handles very quickly. Additionally there was a race condition
which cause false positives.
This reverts commit bc6acd5c5a.
This reverts commit 3a15756c78.
This wasn't very useful in the first place because we're recycling
handles very quickly. Additionally there was a race condition
which cause false positives.
This reverts commit bc6acd5c5a.
This reverts commit 3a15756c78.
The vertex data for glTF morph targets are deltas rather than final
values, so we were passing incorrect data into the SurfaceOrientation
helper.
This has been a long-standing bug that affected models with morphing,
including Buster Drone. However in practice we have observed visual
issues only with AnimatedMorphCube.
Tested against AnimatedMorphCube, Buster Drone, Cesium Man, Fox, and
Littlest Tokyo.
Fixes#1609
This fixes two different logcat messages that would occur when scrubbing
a slider in the UI:
- "Discarding message, message queue overflow."
- "Downloading model..." would repeat itself needlessly
- Add a "shift" parameter to Camera. This has the effect of translating
the viewport, without changing its size.
This is an effect similar to using a shift lens.
- Camera::setScaling() now takes a double2 instead of double4, this is
because scaling the result of the projection in the Z direction
can lead to very confusing problems -- it will essentially move the
near/far planes, and we don't want to expose that as a public API.
- setCustomProjection() now allows to set a different projection for
rendering and culling (useful for e.g. for using an infinite far
rendering projection matrix).
- generously document the CoC computation
- rewrite the code so that the computation will work for any
perspective projection matrix, instead of assuming
an "infinite-far" projection.
MaterialInstance: generate versions of setParameter() methods that
are templated on the parameter sizeof(), which allows us to
coalesce implementations, e.g. vec4<float> and vec4<int> are now
the same. We make the compiler generate non-inline version of all
these methods, the typed version just call through, taking
advantage of the tail-call optimization in clang.
MaterialParser: improve code to avoid calling multiple methods that
all did more or less the same thing and were inlined.
PostProcessManager: use a loop to initialize the materials
ToneMapping: de-dup ACES implementation
And other more minor optimizations.
Fix alignment issues with setting booleans in UBOs
With multithreaded variant compilation, compiled materials differ based on which variants finish first. This change ensures that compiled materials are always bit-for-bit identical.
This is helpful when comparing two compiled materials. Hashing the material files can be used to check for equality. It's also convenient for variants to occupy the same index when using matinfo's --print-glsl flag, for example.
This adds a WebSockets server to filament-utils-android.
Unlike my first attempt (#3599), this does not use a piping server and
is therefore much faster to use over a LAN. It also has lower user
friction because there is no need to touch a button in the app or
scan a QR code.
The main goal of this rewrite was to make the code
simpler and easier to maintain.
The API is mostly unchanged, however there are some differences.
- we now have the concept of subresources, e.g. for Textures, as
subresource is a mip level or layer.
- RenderTargets are no longer resources, instead they are transiant
objects associated to a pass and are now called RenderPasses.
- RenderPasses take subresources for attachments.
- We have better validation of graph building.
We should also compute discard flags more accurately.
This adds libs/viewer to filament-utils-android and provides JNI
bindings for AutomationEngine.
This commit also changes the AutomationEngine to use PPM for screenshots
rather than PNG. The reasoning is that it is wasteful to include stb and
our two image-related libraries just to dump screenshots, especially for
a feature we might not use on Android.
Mathias noticed that the `setAcquiredImage` method was not well
documented on the C++ side. While fixing this I figured I'd try to write
up a docstring for the class itself.
assert() is now replaced by assert_invariant() which has the same
prototype and (currently) behaves the same than assert().
<assert.h> should not be used anymore, and is replaced by
<utils/debug.h>, which is where assert_invariant() is found.
The main motivation for this is to be able to set a breakpoint on the
assertion as lldb doesn't handle abort() very well, and doesn't
permit to inspect the stack trace.
A secondary motivation is to be able (at some point) to enable
assertions without necessarily doing a debug build.
Since we cannot upload a depth buffer, this test works by first drawing
a triangle into COLOR + DEPTH, copying them both to a new render target,
then drawing a slightly larger triangle just underneath the first one.
One of Filament's Linux clients at Google wishes to specify a special
path to the SwiftShader library that proffers Vulkan entry points, and
this path can only be determined at run time.
An environment variable seems like the easiest way to support this
functionality. To prevent a security hazard, the env var is checked only
if a special build-time flag is enabled.
According to the portability spec: "If this extension is supported by
the Vulkan implementation, the application must enable this extension."
On my 2019 MacBook, enabling this extension triggers errors relating to
depth comparison samplers which I believe to be bogus, so I filed a
LunarG bug. For now we silence this specific error message.
When running semantic analysis on a material, we were arbitrarily choosing the first code gen permutation to analyze. So, running matc with arguments --api metal versus --api all would run analysis on slightly different shader code. This causes bugs when flags passed to glslang differ during semantic analysis. This change updates all semantic analysis to always use the same shader code.
When running semantic analysis on a material, we were arbitrarily choosing the first code gen permutation to analyze. So, running matc with arguments --api metal versus --api all would run analysis on slightly different shader code. This causes bugs when flags passed to glslang differ during semantic analysis. This change updates all semantic analysis to always use the same shader code.
This fixes some OpenGL non-compliance (as per "Rendering Feedback
Loops" in the spec), as well as validation errors with Vulkan.
I'm hopeful this can be automatic in the new frame graph. :)
This prevents an "Invalid VkImageView Object" error that would occur
after disabling bloom.
This was caused by the fact that we do not bother zeroing out all the
descriptor slots that are not used by the current draw call.
This fixes `GL_FRAMEBUFFER_INCOMPLETE_MULTISAMPLE` caused by the
following spec clause being violated:
```
The value of RENDERBUFFER_SAMPLES [must be] the same for all
attached renderbuffers and, if the attached images are a mix of
renderbuffers and textures, the value of RENDERBUFFER_SAMPLES
is zero.
```
Tested on Chrome with Triangle and Helmet demos (after enabling MSAA).
Also tried running the automation suite in a debug build on macOS.
Fixes#3128.
This adds setMinMaxLevels to the DriverAPI and calls it from
PostProcessManager and the new unit test. The allows us to be compliant
to the OpenGL specification. The new unit test now looks much better on
Intel GPUs. Chrome still gives an INVALID_OPERATION.
Unfortunately the actual bloom effect still has issues on my Intel GPU,
even though the test passes. So, we need to keep the ping-pong
implementation both for Chrome and for Intel.
I noticed that our slerp function sometimes produces a jolt in
animation, but only when the time delta is very small, and only when the
two operands have completely opposing signs.
For example, let's say you are slerping from <0.76, 0.39, 0.51, 0.19>
to <-0.72, -0.45, -0.49, -0.17>.
These quats are actually quite near to each other because the total
negation of the second quat is similar to the first quat.
We were already doing the short path check in the proper slerp path, but
not when falling back to lerp due to a small angle.
MaterialGenerator selects a custom blend mode for transmissive models,
therefore the AssetLoader should look at the material's blend mode when
it sets up properties, not the one requested by the glTF file.
Fixes#3444.
Previously this test was actually finding a ReadPixels failure on Intel
GPU's, rather than a failure of feedback loop behavior. This was due to
the fact that it was calling ReadPixels on a non-zero miplevel, which
seems to be unsupported on Intel GPU's.
The test now follows the downsample phase with an upsample phase, just
like the real bloom effect. This allows us to issue ReadPixels on the
base level, thus making the test more targetted to rendering feedback
loops.
* Stop using members as globals between methods
* Multi-thread shaders generation with JobSystem
* Pass JobSystem to MaterialBuilder::build()
* Fix MeshAssimp to use the new API
* Allow the Java API to pass a job system via Engine
* Update docs
* Apply suggestions from code review
Co-authored-by: Philip Rideout <philiprideout@gmail.com>
Recall that we now use shared_ptr to simplify lifetime management of
cgltf source data (#3413).
Before this commit, we were only using shared_ptr to retain the source
data during buffer uploads. After this commit, we are using it to retain
source data during texture decoding.
Fixes#3428.
Second attempt after abandoning #3451.
Feedback loops do not work with Chrome or Intel GPU's, so we now have an
optional ping-pong implementation of Bloom.
Issue #2338.
This fixes "illegal hardware instruction" seen on my MacBook, which
is the same as #1839, but triggered only in a test environment where
the main thread is the same as the driver thread, and filament is built
with HAS_THREADING enabled.
* Add swap() methods to Allocator.h
This allows StructureOfArraysBase's move constructor to compile.
Removed an apparent workaround in SingleInstanceComponentManager.
* swap can be noexcept
* Add swap() methods to Allocator.h
This allows StructureOfArraysBase's move constructor to compile.
Co-authored-by: Mathias Agopian <pixelflinger@gmail.com>
Fix a possible out-of-bound access when reading the x-planes array.
Also account for the fact that the loops filling the froxels assume
stl-like iterators (1-past the end on the right), while the indices
themselves were right-inclusive.
fixes#3142
I tested this by viewing CesiumMan with our `gltf_instances` sample app.
There is now a need for some general refactoring in gltfio. I feel that
the high-level objects have become too tightly coupled to each other.
Fixes#3137.
Only items older than a certain value were purged from the cache,
this allowed the cache to grow unbounded before this age limit was
reached.
We now purge the cache at each frame if it's above the size threshold.
Fixes#3381
don't use slog from a static variable, because it might not have been
constructed yet, due to static initialized order.
For some reason, this happens on aarch64 macs, but this was actually
a real problem.
By default, the demo now reflects the camera instead of the renderable.
However this can be changed via the command line. Users can now change
the backend too.
- add darwin-aarch64 support to bluegl
- remove all aarch64 conditions in our cmakefiles, the architecture
is better handled in the source files (to allow fat binaries later)
When we removed the old "ghetto bloom" demo, we ceased to have any
samples demonstrating the RenderTarget API. This demo is a fairly simple
demonstration of planar reflection.
The gltfio API allows users to destroy ResourceLoader or FilamentAsset
even when various asynchronous work (e.g. uploading buffers to the GPU)
has not yet been completed. This was achieved in an error-prone manner
using manual reference counting and an internal management object called
AssetPool.
This PR refactors gltfio by wrapping cgltf_data in shared_ptr, which I
usually try to avoid. However in this case it provides the precisely the
functionality that is needed.
I tested this PR for memory leaks and crashes by hacking gltf_viewer
and monitoring memory usage in Activity Monitor.
This fixes#3383 and makes it easier to implement some missing features,
such as animation support for instanced assets.
Bad texture files would cause the background decoder thread to never
finish and would generate textures with 0x0 dimensions, which eventually
triggers GL_INVALID_VALUE.
To recap, gltfio texture loading has two phases: the first phase is
synchronous and uses stbi_info to gather texture info. The second phase
is asynchronous and uses stbi_load to do the actual decoding.
During the first phase, we now check the return code from stbi_info
rather than assuming that it worked. :)
Fixes#3344
This makes the tgz consistent with the npm package. Fixes#3268.
Note that we already stopped packaging the old HTML docs. Going forward,
we would like to use the TS declarations as documentation because they
must be kept up to date anyway, and they already contain type
information. This was discussed in #3250.
* Add support for sheenColor and sheenRoughness
This work is necessary to support the glTF extension KHR_materials_sheen.
This change effectively adds the specular lobe from the cloth material
model to the base material model. The cloth model remains useful for
its extra subsurface color feature but also because it's cheaper.
* Add support for KHR_materials_sheen to gltfio
* Update documentation
* Document default shading values
This tool makes it easy to analyze the composition of Android
applications that use Filament.
- Filament materials are shown in the treemap if `resgen --json` was
used in the build.
- The input path can be a so file, folder, or zip archive (apk or aar).
- If the path is a zip or folder, helps you find the file to analyze.
- The generated web report is a self-contained HTML file.
- Reports the gzipped size of all Filament materials.
For example, from a mac, you can generate a self-contained HTML file by
typing this from the Filament repo root:
./tools/zbloat/zbloat.py ./android/filament-android
open index.html
This tool uses Python, `nm`, and `objdump`.
The `nm` tool works slightly differently between macOS and Linux, so a
`Dockerfile` is provided that installs dependencies inside a Linux
container. This is also convenient if you do not have both versions of
Python on your system.
The easy way to use docker is to invoke the helper bash script. Simply
type `zbloat.sh [args...]` instead of `zbloat.py [args...]`. The first
time you run it, it will be slow but subsequent times will be fast.
Many thanks to Evan Martin for his interactive treemap widget.
This option embeds a small JSON string inside the resgen blob that lists
all materials and their respective sizes. This is useful when analyzing
the footprint of the binary executable.
This fixes a bug Renderer clearing a render target multiple times when the first time and back again.
Usually the default RenderTarget is unintentionally cleared/discarded in the current behavior.
1) Renderer renders a view without custom RenderTarget having full-screen viewport,
2) Renderer renders a view with custom RenderTarget
3) And Renderer renders an another view without custom RenderTarget having different viewport from 1)
You may see junk or empty pixels in outside of the viewport of the view 3), depending on ClearOptions and the backend.
This fix make Renderer put cleared RenderTargets into a set and check them to be cleared only for the first time, instead of comparing the previous RenderTarget to the current RenderTarget.
We now export only two groups of symbols: (1) JNI and (2) dynamic linker
symbols used by the companion libraries.
This commit has the following impact:
- Shrinks the arm64 so from 1024 KiB to 954 KiB.
- Shrinks the arm64 aar from 560 KiB to 538 KiB.
Note that the patterns in the map file are tested against mangled names
and therefore should not be used to match the double-colon operator.
When passing only 1 fence to vkWaitForFences, the `waitAll` argument
should not have any effect, but SwiftShader seems to skip the wait
when this argument is set to VK_FALSE.
More specifically, the failure to wait in `acquireWorkCommandBuffer`
causes the subsequent destruction of an in-use fence, which causes
a TSAN failure with Google's internal tests.
I am consulting with the SwiftShader team on a real fix, meanwhile
we can commit this easy workaround.
We have 5 usages of vkWaitForFences, one of which uses multiple fences
and should have used VK_TRUE anyway.
When passing only 1 fence to vkWaitForFences, the `waitAll` argument
should not have any effect, but SwiftShader seems to skip the wait
when this argument is set to VK_FALSE.
More specifically, the failure to wait in `acquireWorkCommandBuffer`
causes the subsequent destruction of an in-use fence, which causes
a TSAN failure with Google's internal tests.
I am consulting with the SwiftShader team on a real fix, meanwhile
we can commit this easy workaround.
We have 5 usages of vkWaitForFences, one of which uses multiple fences
and should have used VK_TRUE anyway.
This prevents a SwiftShader crash and/or a slew of "no texture bound"
warnings that would appear when the client provides an IBL without
providing reflections texture, which should be a valid thing to do.
Note that it is okay to declare a sampler in GLSL that never gets bound,
as long as it is never sampled from. Since we always sample from the
IBL specular texture, we should always bind something to it.
This prevents a SwiftShader crash and/or a slew of "no texture bound"
warnings that would appear when the client provides an IBL without
providing reflections texture, which should be a valid thing to do.
Note that it is okay to declare a sampler in GLSL that never gets bound,
as long as it is never sampled from. Since we always sample from the
IBL specular texture, we should always bind something to it.
With Vulkan, this warning would sometimes be a false positive. It could
trigger for internal samplers like `ssao` and `structure`, even though
they were not declared in SPIR-V.
With OpenGL, this warning would never be a false positive because it has
the luxury of calling `glGetUniformLocation`.
This adds a private attribute to our samplers called `strict` that
indicates whether or not a sampler should always have a bound texture.
For now the only strict samplers are the custom ones declared in the
user's material.
At some point I think we should consider adding `spirv-reflect` to our
tree to help with problems like this.
With Vulkan, this warning would sometimes be a false positive. It could
trigger for internal samplers like `ssao` and `structure`, even though
they were not declared in SPIR-V.
With OpenGL, this warning would never be a false positive because it has
the luxury of calling `glGetUniformLocation`.
This adds a private attribute to our samplers called `strict` that
indicates whether or not a sampler should always have a bound texture.
For now the only strict samplers are the custom ones declared in the
user's material.
At some point I think we should consider adding `spirv-reflect` to our
tree to help with problems like this.
This fixes a bug seen with client applications that use ClearOptions
instead of Skybox, and one or more offscreen RenderTarget objects.
These apps would see junk pixels because Filament would only clear the
first render target in the frame.
The fix is to factor some the flag-setting logic in `beginFrame()` into
a private method, and call this method from `render()` each time
the user-level RenderTarget has been changed.
I wrote a simple C++ demo to reproduce the issue and to verify that
this fix works.
This fixes a bug seen with client applications that use ClearOptions
instead of Skybox, and one or more offscreen RenderTarget objects.
These apps would see junk pixels because Filament would only clear the
first render target in the frame.
The fix is to factor some the flag-setting logic in `beginFrame()` into
a private method, and call this method from `render()` each time
the user-level RenderTarget has been changed.
I wrote a simple C++ demo to reproduce the issue and to verify that
this fix works.
Fence.cpp:
My previous fix to the fix was insufficient because a signed integer
overflow could still occur as follows (this was observed in an internal
test at Google).
If the conditional on line 87 is true, then a clamped value is passed
into `wait`, which causes the conditional on line 133 to fail, which
causes `wait_for` to be called inappropriately, which is where an
overflow would occur.
MetalHandles.mm:
This is more theoretical, but to prevent potential integer overflow
within STL, I think we should call `wait()` rather than `wait_for()`
when an infinite timeout is desired.
This is our fourth fix to fences in since November 2; see also dcca236,
e9c9e45, and 8969416. :)
FFence::wait() has some logic that tests if "timeout" is equal to
FENCE_WAIT_FOR_EVER, which is actually impossible because it would
be modified by the clamp in this situation.
In practice I think this oversight would not cause any problems. I
noticed it while reviewing the code.
To recap why the clamp exists: according to the ISO C++ spec, the
"nanoseconds" duration in chrono is at least a signed 64-bit type, but
our sentinel constant might not (and does not) fit within this range.
Note that this API is on the loader rather than the asset. This is
because the loader knows how to create Filament entities by traversing
a cgltf node hierarchy.
Animation on dynamically added instances is not yet supported.
We did not add destroyInstance() because gltfio favors flat arrays for
long term storage of entity lists and instance lists, which would be
slow to shift. We also wish to discourage create/destroy churn since it
is more efficient to pre-allocate instances and selectively add them
into the scene.
Fixes#3137.
Release builds do not call cgltf_validate() so it was possible to
read out-of-bounds animation data when encountering a badly formed glTF
file with mismatched counts between sampler inputs and outputs.
This fixes a slew of validation warnings and errors seen when multiple
subpasses are enabled, starting with:
Attachment 1 not written by fragment shader; undefined values will
be written to attachment
The Vulkan driver was using the same color attachment list for both
subpasses, so the first subpass had unused attachments.
Note that Vulkan makes a distinction between color attachments and input
attachments and requires separate lists to be supplied for each subpass,
but our Driver API consolidates everything into a single list. This
should perhaps be refactored at a later date.
This prepares for #3137 by moving the mesh cache and material instance
cache out of the loader (where they were transient anyway) and into the
actual asset. This paves the way for a `createInstance()` API.
This makes it so that the PixelBufferDescriptor callback triggers at a
time consistent with other backends.
ReadPixels is still asynchronous in the sense that the callback is
triggered on the main thread. However, it is now guaranteed to trigger
during (or before) flushAndWait(), which is less surprising behavior.
You can now build Filament with support for both X11 APIs, or neither.
If both are supported, run-time selection is achieved using a SwapChain
flag.
Supporting only one API at build time (or neither) is useful because our
list of "required" VkInstance extensions can vary according to which
API's are supported. During VkInstance creation, we do not have a priori
knowledge about what kinds of swap chains will be created. (headless vs
non-headless, XCB vs XLIB, etc)
Note that some Vulkan implementation (e.g. some builds of SwiftShader)
only support XCB.
This avoids the following validation error when clients create then
immediately destroy vertex buffers, index buffers, and uniform buffers.
VUID-vkDestroyBuffer-buffer-00922
Cannot free VkBuffer that is in use by a command buffer.
Buffers that are used across multiple frames were fine, but
create-and-destroy scenarios were problematic.
This fixes a regression introduced by 548b28c6e manifesting as
intermittent assertions. The VulkanDisposer gc() should only be called
from the driver thread.
This change fixes an issue where ImGuiHelper would crash on destruction if you created more than one instance at the same time.
This crash occurred when ImGui::DestroyContext() was called, because when the second instance was destroyed there was no current context causing it to crash.
This fixes it by making ImGuiHelper store and manage it's own context.
Having two instances of ImGuiHelper is useful in cases where you want to use imgui with multiple views.
This adds support for more format conversions and removes a bogus
assert that prevented ReadPixels within beginFrame / endFrame.
This was tested with:
backend_test_mac --api vulkan --gtest_filter=BackendTest.ReadPixels
This adds a Dockerfile and a new bash script that makes it
east to invoke the appopriate Docker commands.
This does not yet enable a GitHub Action because of intermittent
issues that we have not yet ironed out.
This allows clients to build the Vulkan backend with either XLIB or XCB
support. I did a quick smoke test of the XCB option, but for now we are
continuing to default to XLIB. Note that the native window type used to
create the swap chain differs between these two API's.
The non-const getSkybox() method is especially useful to client code
that needs to change the clear color, but does not have direct
access to the Renderer object.
If hardware_concurrency() returned 2 while UTILS_HAS_HYPER_THREADING was
enabled, `mThreadCount` was resulting in zero, so jobs (such as texture
decoding) would simply never start. This problem was noticed with
GitHub Actions.
I tested this fix locally by replacing `hardware_concurrency()` with
fixed values like 1 or 2, and verified that the hang went away.
This adds support for headless rendering as well as readPixels. These
features are necessary for automated testing.
NOTE: invoking readPixels with a non-headless SwapChain seems to fail
with MoltenVK. I will investigate this later.
The Filament View that gets passed to tick() is retained briefly when
screenshots are enabled, but it may be destroyed by the time the
PixelBufferDescriptor callback is invoked. So, we needed a way of
notifying the AutomationEngine.
This adds a `material` key to `Settings`, as a sister to `view`.
Here's an example of an automation spec that manipulates material
parameters:
[{
"name": "metallic_vs_roughness",
"permute": {
"material.scalar.roughnessFactor": [0.0, 0.5, 1.0],
"material.scalar.metallicFactor": [0.0, 1.0]
}
}]
Currently this is limited to float, vec3, and vec4 parameters.
This adds `AutomationEngine` to libs/viewer, which iterates through
`Settings` instances that were generated from a JSON spec and applies
them to a Filament `View`. It can be configured to sleep between tests
using a time delay or a frame count.
This also adds command line arguments and user-interface elements to
`gltf_viewer` for automated testing.
This lets us generate long lists of settings permutations by writing a
simple JSON spec. For example, the following spec would generate six
`Settings` objects, all of which have SSAO enabled. See the unit test
for a larger example.
```
{
"name": "viewopts",
"base": { "view.ssao.enabled": true }
"permute": {
"view.dithering": ["NONE", "TEMPORAL"],
"view.sampleCount": [1, 4, 8]
}
}
```
- Use lower depth LOD as the cone radius increases.
This is the same technique we use for SAO.
The better cache access significantly improve performance.
On a test on Pixel4 at 585MHz, SAO pass improves by 30%.
This also helps the algorithm scale with the shadow distance.
- Shadow direction/length no longer dependent on aspect-ratio and camera
orientation.
This is fixed by doing all the computations in screen-space instead of
normalized screen space.
- Shadow parameters are no longer dependent on the field of view.
- Also rename dominantLightShadowing.fs to ssct.fs
- remove zoom parameter, it wasn't very useful.
This introduces the `viewer::Settings` struct, and a JSON reader /
writer.
This will be used for automated testing and for client / server
communication.
Note that `viewer::Settings` is closely associated with the
`filament::View` API; when updating the latter we will often need to
update the former, as well as some serialization code. This increases
the maintenance burden and I think we should consider using a parser
library like libclang or a macro-based reflection utility.
This PR also migrates SimpleViewer into libs/viewer and un-inlines its
implementation. It does not belong in gltfio because it has an imgui
dependency.
This is a technique from Naughty Dog used in TLOU2, where we compute
the "ambient" shadowing of a dominant light in screen-space.
This is currently integrated to the SSAO pass.
The image barrier must wait until all write operations are complete;
this was not set up correctly but somehow did not cause problems.
Also tried to document this better, since using an image barrier for
the swap chain seems to be an unusual thing to do (e.g. the Sascha
Willems demos do not do this).
There's no really good solution, so this just keeps it simple and makes
is so that embind consumes an integer rather than an embind enum object.
Tested by hacking _createTextureFromImageFile.
Fixes#3028.
For example, this makes it so that the primary material in Sponza
is *always* "untitled 0", which makes it easier to figure out which
material to modify.
Recall that Ctrl + Up / Down navigates between materials while
Left / Right navigates between variants.
- Make this behavior more explicit in the documentation.
- Shortcuts now use the same ordering as the displayed list.
- Shortcuts now use Cmd on macOS instead of Ctrl to avoid conflict with
Mission Control defaults.
If the websocket message exceeds a certain size, we might receive it
in chunks and civetweb does not auto-consolidate for us.
Fixed this by adding a length prefix to the edit command.
This change also makes it so that matdbg shows a numeric suffix to help
with duplicated names, and fixes a bug with names that start with
non-alpha characters.
Block compression is currently only used when we invoke mipgen for the
suzanne demo. This PR makes it easy to disable.
Block compression is still enabled in our CMake build, but other build
systems can simply omit BlockCompression.{h,cpp} from imageio (as well
as the two third_party libs) and it will "just work".
This was tested by replacing the node 0 scale in BusterDrone with
[-1, 1, 1].
For future reference, commit f728776 shows when we switched from
transpose(inverse()) to cof(). This was a good change, but before that
particular change, we had a "two wrongs made a right" situation for
mirrored normals.
Fixes#3001.
The equality operator for MaterialKey was missing some fields (also, its
ordering was inconsistent with the struct), which caused faulty behavior
in the material cache.
I ran into this bug when testing against a Blender-exported model that
has some materials with texture transforms, and some without.
We borrow a page from the HBAO book here and allow to limit the angle
with the horizon of SSAO samples. This can help reducing the effects
of low tessellation, which tend to create unwanted creases with SSAO.
contact shadows wouldn't work if shadowmap-based shadows were not
enabled, in this case the (unneeded) shadow pass would run and
attempt to create the shadow-map with an incorrect size (0), which
would later crash/assert in the backend.
This is a bit of a hack but it helps a lot -- for some reason the
background layer often has a sharp transition to the in-focus plane and
because we're calculating the DoF at 1/4 resolution these transitions
are quite visible. We work around this by fading out the transition over
a 1 CoC unit, specifically between a CoC of 2 to 1 pixel. Below 1 pixel,
we currently clamp to 0.
The sidecar buffer for multisampled_render_to_texture emulation was
stored with the texture, instead of the rendertarget, this prevented
a single texture to be used as an attachment of multiple render targets
with different sample counts.
There was an inconsistency in the gles backend where non-sampleable,
non-MS textures failed (assert) if they were used with a render target
with MS (i.e. "magic resolve" case).
We fix this by taking the same code path for textures and renderbuffers,
the only difference being the calls to glFrameBufferRenderbuffer
instead of glFramebufferTexture2D.
resources structures where not cleared when destroyed, and some code
relies in the handles being set or not to decide if a texture should
be used. in this case, the texture would be set in the shader after
it was destroyed, because its handle wasn't cleared.
- inScatteringSize actually cannot be set to zero because it produces
a 0^0 in the shader. So with this change, in-scattering must be
strictly > 0 to enable.
- inScatteringSize on the java side had an incorrect default value of
zero (which of course, now doesn't matter anymore).
- also clamp the fog altitude to 1mm which simplifies the shader quite
a bit for the same result.
Fixes#3069
.blurScale was in fact a scale factor applied to the circle of
confusion (which makes it indeed a "blur scale", but let's use
a more precise language here).
Update comments to show how to use .cocScale to control the DoF
effect independently from the camera aperture, which can be useful for
artistic reasons.
- simulate a standard deviation of 8 by skipping every other pixel
when blurring the AO buffer at LOW or MEDIuM quality. This creates
a faint checker pattern, but does look much better than the low
frequency noise that is introduced otherwise.
- tweak AO parameters to give a more uniform and smooth AO effect by
default. This also makes AO even more consistant between quality
modes.
- add a simple dithering stage to AO blur to help eliminate banding
artifacts at higher sample counts.
e.g. `F_PI` is also available as `d::PI` and `f::PI` as `double` or
`float` respectively.
Most of our uses of those constants are `float` and it was error prone
and cumbersome to have to cast to that each time we used them.
The `F_` constants are intended to mimic `math.h`, however these new
constants are not, and all the "n over" constants are named
`N_OVER_` instead of `N_`, which helps avoiding the confusion that e.g.:
`F_2_PI` is actually `2/pi` and not `2*pi`.
Also added `F_TAU` which is 2 * PI.
all parameters of the bilateral blur can now be set from the CPU,
instead of being hardcoded -- this probably prevent the shader compiler
from unrolling the filter, but it's also probably not a big deal.
additionally, we now use a kernel size of 23 (12 samples) when the
SSAO quality is set to HIGH or more.
ssaogen is no longer needed
Higher quality levels used to darken the ssao a lot, now they all have
a consistant look.
Add quality control to material sandbox.
Slightly simplify shader code.
We add noise do the radius of the samples on the spiral
(instead of adding noise only to the angle), this reduces a visible
"echo", especially at higher radii.
- don't use frameUniform.time as the temporal noise seed because
this has a cycle to it. It's not very visible on dithering but
obvious when used for other things.
- don't use fract(time) since time was already in [0,1] -- though it's
moot now.
- instead of time pass a random number in [0,1] generated on the CPU
each frame.
This will also allow to control temporal noise more easily if
needed in the future (e.g. for accessibility or rendering tests)
- fix a recent typo that affected SSAO quality
- add a upsampling quality checkbox in material_sandbox
- make use of textureLodOffset instead of texelFetch, so that
we emulate more closely textureGather
- don't hardcode bilateral filter edge-distance in the shader
(still hardcoded on the cpu side)
this broke recently with the reverse-z, the froxel calculation relies
on the depth, which is now inverted.
we actually fix our getNormalizedViewportCorrd() public API, which the
froxel code uses.
Because initializing the backend can be relatively slow, creating an
Engine can block the main thread, is this is a problem, this new API
can be used instead; while a bit more complex, it allow the engine to
be initialized asynchronously.
This significantly improve the depth-buffer resolution utilization
through the near-infinity range on Metal and Vulkan.
On OpenGL, this benefit is only seen when glClipControl is available.
The user-facing clip plane is unchanged [-1,1], the conversion
happens in filament's vertex shaders.
The bulk of this change consists in:
- invert the clip space's z in the vertex shader
- clear the depth buffer with 0
- invert all depth function comparisons
- fix all screen-space effects, e.g. ssao, DoF
- fix shadows and shadow biases
- use floating-point depth
- add a driver API to query which clip-space is used
- add glClipControl support to the gl backend
the depth buffer wasn't resolved in this case. We fix this by always
creating a depth buffer resolve pass -- which is a lot more robust.
this pass automatically culled if the depth is not used (i.e. TAA or
DoF are off). The pass is a no-op if resolving is not needed.
- don't apply SSAO to blended objects
these are not drawn into the structure buffer, so they don't
participate in SSAO.
- remove unused code that was needed for the depth prepass, which
simplifies handling of blended objects.
- always treat alpha-masked objects as opaque.
This also adds a -t flag to the easy build script for convenience.
SwiftShader on Vulkan can now be used as follows.
git clone https://github.com/google/swiftshader.git
cd swiftshader/build
cmake .. && make -j && export SWIFTSHADER_LD_LIBRARY_PATH=`pwd`
cd ../filament
./build.sh -t debug && ./out/cmake-debug/samples/gltf_viewer -a vulkan
Note that third_party/swiftshader does not need to exist since we
already have Vulkan headers. I will remove it in a subsequent PR.
FrameGraphTexture::Descriptor uses "0" to mean "auto" but the
sample count that is passed to the backend should always be 1 or
greater.
See also #2957.
The depth buffer was always set with a sample count of 1, this was
mostly moot because we almost never need the depth buffer after the
color pass; except for refraction and DoF.
In the case of refraction, we need to preserve the MS depth buffer
(just like we preserve the MS color buffer), moreover "backward resolve"
is not supported. So in that case, the depth buffer needed to be set to
MSAA samples.
In the case of DoF, we need to access a resolved depth texture, and
there is 2 cases here:
1) without refraction, a magic resolve is adequate -- in this case
the depth MS count must be 1.
2) with refraction, we are forced in to a MS depth, so it needs a
special resolve pass to be accessible from DoF.
depth/stencil attachments must have the same MS sample count
than the FBO (rt) they're attached to.
this is because EXT_multisampled_render_to_texture[2] doesn't
resolve depth attachments. In this case, we can't use this extension.
This regressed when we added a callback argument, because the callback
cannot be bound.
This commit also modifies the helmet example to use the low-level API
for testing purposes.
Note that this omits the optional nanoseconds argument. JavaScript
does not support overloading and this argument probably cannot be
accurately provided anyway in a web environment.
Fixes#3002.
- more flexible linearization of depth, doesn't impact performance
at all in the shader, but takes into account 3 parameters of
the projection matrix instead of true. Mathematically identical
to before.
- use depth test "less or equal" for depth only passes to be consistant
with the color pass. Unsure if using "less" would be better.
- use textureLod() when sampling the skybox
- use mediump for all samplers in the shadowing.fs, which correspond to
their actual definition.
Filament calls createRenderTarget very frequently, so this PR has the
side effect of reducing a huge amount of create-destroy churn in other
Vulkan objects such as VkRenderPass and VkFramebuffer.
Marking a view with a custom render target will now disable blending
for that view but require an alpha channel for intermediate buffers.
Blending is disabled because it is meaningless.
fixes#2802
This fixes the long-standing "black rectangle" issue in our shadowtest
sample.
We were mistakenly constraining the clear to the viewport, which
is meant only to affect the vertex pipeline.
Also, we were needlessly setting the scissor rect in BeginRenderPass(),
since we already do so in Draw().
This is more similar to the Vulkan shader pipeline and less magical.
There are 3 places in our shader code where we perform fixups like this:
- main.vs
- depth_main.vs
- post_process_getters.vs
That last one needs no change since it does not involve Z.
Also cleanup all setImage() error reporting. All error checks are
changed from silent to NON_FATAL_ASSERT, which means they'll throw
if exceptions are enabled or do a no-op and log a message otherwise.
Also added more precondition checks.
* Add back proguard file for Android and update README.md
The proguard-rules.pro file provided by Jeff McGlynn (9fe60a636)
works perfectly for Android projects with minifyEnabled.
Add the config file back to the repository and update README.md
to help other Android users setup release build.
* Refine android proguard file and remove proguard section in README.md
The stride passed to Vulkan for query pools is the stride between
queries, not the stride between individual values. I realized this after
seeing the SwiftShader implementation of queries.
I also used log statements to confirm that this fixes bogus frame
skipping on my Pixel 4.
This fixes the visual hang that would occur after going to the task
switcher and back.
After recreating the swap chain, all of its associated fences and sync
objects become invalid. Since we were not tracking this, FrameSkipper
was going into an infinite wait after the swap chain was created.
This removes the old getClientExtent stuff that I added a while back,
and replaces it with the strategy espoused by the MoltenVK author found
here:
https://github.com/KhronosGroup/MoltenVK/issues/428
I tested this PR on macOS with Vulkan and Metal. I tested both on high
and low DPI displays. I also tried dragging the window between the two
displays.
Android still has issues with rotation, I will deal with that next.
cleanup
This fixes a lot of flickering when SSAO and TAA are enabled.
SSAO TAA shouldn't be done based on a jittered depth buffer, but
rather on jittered samples.
while this started as a sample to demonstrate how to use the new
frame history feature, this is now solid enough to become a feature.
this version of TAA doesn't use motion vectors, so it is prone to
various artifacts (e.g. ghosting). It also has the typical issues
TAA has (e.g. transparents).
with that said, performace is good and quality is overall superior
to FXAA or MSAAx4.
In retrospect, I should've exposed ints to JS instead of entities.
Making this fix is already a TODO, but it will be a tedious change. For
now, this PR fixes two issues that were causing tiny but frequent memory
leaks, according to the emscripten memory leak widget.
(1) Do not include Entity in the list of classes that should only be
passed around by pointer.
(2) Clients should always call delete() on returned Entities.
This example has fallen out of date, I tried to resurrect it but it
turned out to be non-trivial, given the special nature of emscripten's
generated JavaScript wrappers.
It might be better to provide an example that uses rollup, since that's
the bundler that model-viewer uses, or a simple TypeScript example.
For now, I think our simple HTML + raw JS samples are sufficient.
We can now save an arbitrary (but decided at compile time) number of
entries containing information about previous frames and it for while
drawing the current frame. Typically this is used for such effects as
TAA, screen-space reflection, building probes over time, etc...
Currently the frame history consists of one frame per View. FView
provides a new (internal) getFrameHistory() method used to access
previous frames entries as well as to save the current's frame
information.
The FrameGraphPassResources (which is available in the execute closure
of a FrameGraph pass) has a new detach() method which can be used to
remove that resource from the FrameGraph lifetime management, at which
point it becomes the responsibility of the caller. Typically this
resource will be stored in the frame history.
The FrameGraph cannot store pointers internally because all objects
are stored in vectors. Unfortunately we were storing the writer of
resources as a pointer.
This is fixed by storing an index instead.
Sets of post-process materials are in their own folder now.
Also moved fxaa.fs from shaders/ to materials/ so it's next to the
material that includes it. This file is not shared with any other
material.
This prevents strange behavior with semi-transparent render targets,
which the model viewer team discovered when testing against the Khronos
alpha test conformance model.
* attempt to reduce code size in PostProcessManager
The two main changes are:
- materials are now stored in a hash map which allows us to
automatically destroy all of them using one loop -- instead of
having to call terminate() for each of them.
- factored a very common sequence of code:
commit()
use()
beginRenderPass()
draw()
endRenderPass()
This makes the code easier to write and also results in less code.
There are also minor other changes, e.g. RenderPass constructor is not
inline anymore.
This fixes the depth fighting in the column of red spheres in the test
model while in filamat mode. Thanks Romain for catching this.
It also makes filamat mode more consistent with ubershader mode.
So...we use integers for bone indices, and that makes them special. :)
On May 26, the emsdk WebGL 2.0 wrapper for the integer variant of
VertexAttribPointer was broken due to an incomplete renaming operation
from an emscripten contributor.
I have a pending upstream fix to the emscripten repo (#11742) but for
now we can workaround this with a simple monkeypatch.
Fixes#2856.
We now validate supported combinations of internalFormat vs. format/type.
We currently follow the GLES specification, but we could be more
restrictive if needed for vulkan/metal.
The validation terminates in debug builds and logs (and becomes a no-op)
in release builds.
- we were not clamping the color grading output to [0,1]
- we were not using a supported input (cpu) pixel format for the
3d LUT in low and medium quality
fast tiles (i.e. where all pixels have the same circle-of-confusion)
are allowed to use bilinear filtering.
this results in better looking DoF in these areas.
Using a ring density of 4 was wrong -- apparently I can't add.
With a density of 8 we can guarantee we don't have gaps between samples
when using mips properly (or at least gaps are consistant and
predictable).
* Add quality option to ColorGrading
Low: 16x16x16 LUT in 10 bit
Medium: 32x32x32 LUT in 10 bit
High: 32x32x32 LUT in 16 bit
Ultra: 64x64x64 LUT in 16 bit
* Move comment
- Use the random offset to calculate the correct intersection radius.
For the center sample, the ring radius is otherwise zero, which means
any sample in the random disk with a CoC radius > 0 would intersect,
and that seems wrong.
- The selected mip was off by one.
We were always picking a mip too high because the CoC are in
high resolution, but our mip-chain is 1/4 res from that.
This improves rendering quality quite a lot.
For testing purposes, sample-gltf-viewer reloads the model when the
user double-taps the screen. This operation was causing a buildup of
blank entities in the Scene, because we were merely destroying the
components, not removing them from the Scene.
This chooses type=THIN, mode=SCREENSPACE, and blending=TRANSPARENT.
Due to sampler overload, we print a friendly error message and enable a
fallback if the user tries to simultaneously enable ubershader mode, the
clearcoat extension, and the transmission extension.
- make sure all downsampling is exactly correct and use proper
uv values to access these textures. This stabilizes the DoF a bit.
- for now only use NEAREST filtering, as bilinear can't be used naively.
- noise is now mandatory, fix noise radius
- add a median pass (performance is not impacted because DoF is faster
due to using NEAREST filtering)
- fix be consistant in how the kernel radius is computed, fix a couple
errors in that area.
- change ring density from 8 to 4 (but starting with 8), so each new
ring is 4 samples larger than the previous. This reduces the number
of samples needed from 25 to 21 on mobile, and 49 to 37 on desktop.
It also makes it conceivable to use 5 rings (57 samples). Quality is
not visibly impacted because the center ring is still 8 samples.
This density, matches the number of samples required for "square"
rings.
- fix scaling issue with diaphragm simulation
- fix typos in background accumulation code -- this fixes a lot of
quality issues.
- improve performance of "fast tiles" during both DoF and combine passes
Overall the effect runs about 0.5ms faster and quality is greatly
improved.
This makes two changes to the Driver API:
1. Adds the beginSubpass() method.
2. Adds subpassMask to RenderPassParams.
Locally I've proven that this PR is sufficient by making a series of
Vulkan changes; sneak peek is available at commit
bfef6914a7d288bbd25926c02ec93d231a8c0d07.
Note that this PR moves the new usage flag to a different texture;
I had placed it on the wrong one.
See also #2780.
The new way is to follow the instructions at the top of
`web/docs/build.py`.
The option in build.sh has been unused for a while. Invoking Python from
CMake was pointless, especially since I don't want to force other
developers to install a specific version of Python etc.
Since postRun can no longer be set from within the extern post JS, we
now use the promise rather than postRun.
This upgrades emsdk to 1.39.19 (see note in #2800).
This makes the tutorials more consistent with the live demos and lets us
more easily ensure that the tutorials are kept up to date without
errors. In fact it caught an error in the Suzanne tutorial.
Since this affects source rather than served files, this PR is intended
for main. The master branch will be updated in a subsequent PR.
The existing API was fine for FilamentApp but it prevented clients from
making ImGui calls from arbitrary places, which forced them in some
cases to queue up their ImGui calls using vector<function<>>.
This PR publicizes the low-level method that consumes the ImGui draw
list and removes calls to global functions.
This is step 1 for supporting subpasses (aka framebuffer fetch).
Vulkan requires us to explicitly bind the shader to any framebuffers
that we read from. These are called input attachments in Vulkan
parlance, and each binding is called a descriptor.
For simplicity we only create homogeneous descriptor sets where each
descriptor set is comprised of only one type of binding (i.e. all
uniforms or all samplers). I see no reason to break with that
architecture right now, so input attachments live in their own set.
In the long term, we should probably think about partitioning
descriptors differently (e.g. according to their frequency of change).
This PR also renames some internal fields in order to avoid nomenclature
confusion, such as `mDescriptorSets` => `mDescriptorBundles` . Each
entry in this map is a set of descriptor sets, or what we are now
calling a "bundle".
This fix has the side effect of causing a validation warning due to
a declared fragment shader output that has no corresponding attachment.
We can fix that in a future PR.
Fixes#2752
The UI was in a weird state after the user selected a new material but
before they had chosen a particular variant. The highlighted
material did not match the content of the source editor.
We now download shader source only after the user selects a variant.
Previously we tried to download all shaders at once. This caused Chrome
to throw ERR_INSUFFICIENT_RESOURCES.
Fixes#2759
We now get away with a sin/cos per fragment instead of per-sample,
this increases a little bit the ALU load but decreases the EFU's.
This is not necessarily faster, but sin/cos may cost more power or
have longer latencies.
glslang generate much less code when function parameters are marked
const (it also works around a bug where highp intermediate variables
are introduced -- but this will be fixed soon).
Also remove non-trivial uses of ?: as it also triggers a similar problem
in glslang.
This prevents the following validation error:
Cannot get query results on queryPool 0x2 with index 0 as data has
not been collected for this index.
This was occasionally triggered when we attempted to query a timer that
had not yet been inserted into any command stream. (The error seems
bogus to me since we've enabled the AVAILABILITY flag, but we might as
well fix it anyway.)
This implementation is an order of magnitude more efficient, running
in about 6ms on Pixel 4.
We're using the ring binning technique described in "Life of a Bokeh",
SIGGRAPH 2018.
This would happen when using math -E -a vulkan. GLSLangCleaner lifetime
must be shorter than TProgram/TShader.
I think we can fix this by using another scoped GLSLangCleaner.
Tested successfully with ASAN.
- Allow JavaScript and Java / Kotlin clients to configure this setting.
- Opt in by default (this was disabled in #2621)
- Optimize large assets like Bistro by iterating through prims only when
the asset actually has skins.
Fixes#2714.
The Animation selector is now a top-level group (collapsed by default),
previously it was in the Hierarchy group, which was non-intuitive.
Also:
- Use indent() for consistency with other UI groups.
- Remove un-necessary usage of a lambda.
This repairs the `animation.html` sample. Using a JS array is more
natural for web developers and more consistent with the latest
JavaScript bindings in gltfio.
* Improve JS API for SurfaceOrientation helper.
Web developers should not be expected to call malloc on the emscripten
heap. This new API is better aligned with how we handle vertex buffers.
Fixes#2702.
* Update web/filament-js/extensions.js
Co-authored-by: Wojciech Czerniak <wojciech.czerniak@gmail.com>
Co-authored-by: Wojciech Czerniak <wojciech.czerniak@gmail.com>
Reproduced the issue by hacking the TexturedCube glTF file.
Verified the fix by viewing TexturedCube in ubershader mode, then
hacking the shader with matdbg in order to visualize texture
coordinates.
Fixes#2659
matdbg
- Materials are now sorted by name and the sha id is hidden.
- PostProcess materials now separated from Surface materials.
- PostProcess details now hide the non-existing properties.
- Larger default pane size for the material list.
- Use nicer font.
matinfo
- Print out the Material Domain.
filament
- get*Slow methods in FMaterial are now private.
The blit DriverAPI command was not enforcing correct synchronization.
On Android, this manifested as an intermittent issue with bad mipmap
levels.
This did not trigger any validation warnings and therefore took a long
time to track down.
Fixes#2643
- don't crash when a material doesn't have a name
- don't cache the program in PostProcessManager so we can do live
editing. Not needed anyways, because Material has a cache.
When invoking MaterialBuilder.samplerParameter with Java enum SAMPLER_EXTERNAL it was wrongly recast as native enum SAMPLE_CUBEMAP in the JNI interface because the Java enum was missing SAMPLE_2D_ARRAY.
- assert that shutdown() is called from the main thread
- attempt to catch whether CommandBufferQueue::mExitRequested gets corrupted
(which could explain the ANR).
* Add transforms for ACEScct and enabling/disabling color grading in sample
* Add saturation and contrast adjustments
* Rename ACES to ACES_LEGACY and introduce ACES
ACES_LEGACY is ACES with a brightness boost to match our old
tone mapping operator.
Tested in Android Studio by starting sample-gltf-viewer and typing the
following into the debugger:
modelViewer.asset!!.getFirstEntityByName("Drone_Body")
modelViewer.asset!!.getEntitiesByName("Drone_Body")
modelViewer.asset!!.getEntitiesByPrefix("Drone")
* Add new color grading feature: shadows/mid-tones/highlights
This feature can be used to color correct specific tonal ranges.
The tonal zones can be precisely controlled.
glTF assets are required to provide min/max attributes for POSITION,
so in theory we never need to compute the AABB. However the option is
still there.
This PR also makes it so that Java clients can choose this option,
previously it was hardcoded for all non-native clients.
* Add the ability to set a ColorGrading LUT per View
In this change, material_sandbox allows to change the tone mapping
operator at runtime.
* Add Mathematica notebook used to explore white balance
The implementation is meant to reproduce the temperature/tint sliders
founds in Adobe Lightroom. The temperature can be offset from 2,000K
to 50,000K using a slider between -100 and +100 (-1.0 to 1.0 in our
API). The range of tint was modelled after the range used for the
temperature.
* Fix various issues
- A refactoring wrongly remamed the color grading pass
- Setting a View's ColorGrading to null selects the
default color grading options
* Implement white balance in ColorGrading
* Introduce ColorGrading API
A ColorGrading object will eventually be settable on a View to
choose the tone mapping and other color transformations.
There is currently only a single default ColorGrading instance
used by all views.
* Use raw function pointers
* Prep work for proper color grading
* Bring back local colorGrading bool to drive that pass
* Fix formatting issue
* Add missing case for the color grading pass
* Fix formatting issues
* Formatting issues
When animation is applied to the master asset, all instances are
animated.
Instances can also be individually animated via the Animator in
FilamentInstance.
Fixes#1513.
This adds createInstancedAsset() to AssetLoader, which creates a master
asset and a set of slave instances. Vertex buffers, index buffers,
textures, and material instances are shared. Entities and components are
duplicated.
Instances have their own API object that is very simple. Light sources
and material instances are not instanced, so are not accessible through
the instance API object.
The master-slave ownership model lets us avoid complex shared ownership
semantics. The existing cache structures in AssetLoader allow the
implementation of this feature to be fairly simple.
The master asset exposes the union of all entities and allows clients to
modify all instances en masse if they wish. This design also works
naturally with ResourceLoader, which does not need to know about
instancing. For example, asynchronous loading is completely unchanged;
the dependency graph simply contains the union of entities across all
instances.
Support for animation is added in a subsequent commit.
Fixes#1513.
To enable some basic profiling of the gl thread using performance
counters, the "filament.perfcounters" property must be set to 1
(before the Filament Engine is created):
adb shell setprop filament.perfcounters 1
This is so that filament doesn't interfere with simpleperf by default.
This is now the default, and the tone mapper used
is now the same on desktop and mobile.
In the future this will allow us to implement many features such as
color grading at no cpu cost.
OpenGL allows vertex shaders to consume attributes that are never
assigned values from a vertex buffer. When this occurs, GL provides a
default zeroed-out value, or a value from the most recent call to
`glVertexAttrib*v`.
Vulkan does not allow this, so our Vulkan backend provides dummy data by
re-using the position buffer, which we know exists. This was an existing
hack, but it did not always assign a compatible type, which resulted in
validation spew with gltf_viewer + BusterDrone.
This fixes one of the two bugs reported in #2573.
Vulkan complains if a pipeline declares an integer vec2 attribute but
the shader consumes it as "vec2" rather than "ivec2". gltfio was doing
this when supplying dummy texture coordinates.
This fixes one of the two bugs reported in #2573.
Since AO is computed at 1/4 resolution, it is necessary to upsample
the AO buffer. Until now this was done with a bilinear tap, which is
less than ideal as it can creates jaggies at edges.
High quality upsampling can now be enabled and uses a bilateral filter.
The cost is about 2.0 ms at 250MHz on Pixel 4. ES3.1 is required.
SSAO starts at 1/4 resolution, and because we used derivatives
to calculate a cheap pre-blur, the output was mostly 1/4 resolution
of that, leading to 1/4 of the destination pixels being mostly identical.
This, in turn caused sampling issues when reading the SSAO buffer during
the color pass.
We fix this by getting rid of the pre-blur, and increasing slightly the
size of the gaussian blur from 9 samples to 13, which increases the
SSAO pass time by 20%, or 0.3ms at 430 MHz on Pixel 4.
With this change and bilinear filtering in the color pass, we get rid
of all the sampling artifacts.
until now we allowed any resolution for SSAO, but it didn't make
much sense, especially that the depth pass is now used for other things.
To keep things more manageable, we only allow 0.5 and 1.0 scale
factor settings (respectively quarter and full resolution).
Adding this for completeness, although the feature triggers a known
issue with Chrome's WebGL implementation that emits an error when
source and destination textures of a draw call are the same.
Refracted objects could be drawn on top of opaque objects that were
closer to the camera.
Du to a typo, we were not reusing the existing depth buffer during the
refraction pass.
Fixes#2576
This contains the following two fixes, but the second fix is more
important because it prevents an actual validation error on Android.
1. The query pool bitset was written in the main thread but read
in the driver thread, this is now protected with a mutex.
2. Querying a timestamp before it has been written into the command
buffer was still causing validation errors, despite the fact
that we are now properly using the AVAILABILTY flag. This is a
CPU-side synchronization problem and can therefore be fixed using
std::atomic<bool>.
This was sometimes causing a cascade of log messages and swap chain
re-creations on Android.
To support resizing, we still re-create the swap chain when encountering
VK_ERROR_OUT_OF_DATE_KHR.
Suboptimal means: "A swapchain no longer matches the surface properties
exactly, but can still be used to present to the surface successfully."
It is convenient to control the debug_info extension and the validation
layers from one central place. This makes it easier to force-enable
validation in Release builds to validate optimized shaders.
Since we are still on Vulkan 1.0, we cannot use SPIR-V 1.3.
If we try to do so, this error is generated:
Invalid SPIR-V binary version 1.3 for target environment SPIR-V 1.0
In non-optimized builds, we were already generating spirv 1.0, but
when we enabled the shader optimizer, we generated spirv 1.3.
This bug has actually been around forever, but we did not notice because
we were only invoking the optimizer in release builds, which does not
enable validation.
We cannot upgrade Vulkan 1.1 because the latest LunarG SDK for macOS
does not support it.
The value we pass to spvContextCreate() is not SPV_ENV_UNIVERSAL_1_3,
which is what we use in GLSLPostProcessor.
Also add a call to spvValidateBinary(), which would have caught this
oversight.
This allows FrameSkipper to work on desktop.
Similar to the GL backend, we can only query the fence from tick().
The Vulkan spec stipulates that a fence object cannot be accessed by
more than one thread simultaneously. My initial implementation violated
this, which was caught by the validation layer.
This allows to never store/load to main memory the HDR color buffer,
saving precious bandwidth. This mode relies on the
ext_framebuffer_fetch extension being present and can't be enabled
with certain features (currently msaa, Bloom, DoF).
This will eventually be supported on Metal and Vulkan as well.
this is intended to be temporary until we properly add support
for vulkan's subpasses. this PR just helps us activate
ext_framebuffer_fetch on select materials.
The presence of a special Gradle property is now used to exclude Vulkan
support from the build. By making Vulkan "always on" for local
development, we can avoid stale CMake cache issues that arise from
toggling the Gradle property. It also lets you avoid adding the flag
to Android Studio preferences.
This is motivated by testing and does not indicate production readiness.
Clients still need to pass VULKAN into the Engine constructor to select
the Vulkan backend.
To keep APK size down and keep CI fast, we are continuing to exclude
Vulkan from official Android builds.
After syncing this change, you might need to use `./build.sh -c` to
clobber various build caches.
These changes make it easier to test headless rendering in macOS.
I verified that this works using the frame_generator tool, however
it requires the fixes in PR #2529.
This is done for consistency with the Android samples. Tested with
gltf_viewer against Vulkan, Metal, and OpenGL. Dragged window between
high DPI and low DPI displays.
Since resizing does not trigger the "NSView has changed" handler in
PlatformCocoaGL, it now must to detect a resize in order to make the
required call to NSOpenGLContext::update() as per the following
documentation.
https://developer.apple.com/documentation/appkit/nsopenglcontext/1436135-updateFixes#2528
We were not holding a global ref on more recent version of android,
this made the Stream API inconsistant.
Filament now always keeps a global ref on the surfacetexture while in
use.
this was a misguided optimization, on low-end devices this actually
has a large overhead (presumably because of the smaller cache and
increased branch missprediction as well as extra synchronization).
on some low-end devices we see the gl thread go from ~40ms to ~10ms
with the gltf sample.
- only triangular noise needs to be scaled between +/-1, other noises
have a uniform distribution and need to be scaled between +/-0.5
- all dither routines work in RGBA
- fixed FXAA in opaque mode when dithering modified the alpha channel
(which is used by FXAA). This fixes flickering when FXAA and dithering
was enabled.
- use triangular noise dithering on mobile and desktop. The cost in
not measurable on a pixel 4 / 1080p, and the quality is better.
- refactor dithering code a bit such that:
- noise methods are not temporal
- all dither functions have the same structure
* debug option to track Entities
Set FILAMENT_UTILS_TRACK_ENTITIES to true when building libutils to
activate entity tracking. This adds two public methods:
getActiveEntities() and dumpActiveEntities() the later displays the
stack trace of where the remaining entities were allocated.
This is useful for tracking leaks.
* Update libs/utils/include/utils/EntityManager.h
Co-authored-by: Philip Rideout <philiprideout@gmail.com>
Co-authored-by: Philip Rideout <philiprideout@gmail.com>
* Very basic SwiftShader support
- we only provide SwiftShader's khornos headers in third_party
- swiftshader itself must be available on the host (for instance it
can be compiled from source).
- to enable pass -DFILAMENT_USE_SWIFTSHADER=ON option to cmake
- only GLES 3.0 is supported. Vulkan is not yet supported.
CMake should find the swiftshader libraries automatically, but if
they're installed in a non-standard place, the environment variable
SWIFTSHADER_LD_LIBRARY_PATH can be set to that place.
We also use the GLES 3.0 headers everywhere, since we don't rely on 3.1
at this point.
* add a tnt folder with a README
This removes (rather than deprecates) all public constructors that
take a native pointer without an accompanying Engine.
Most notably, MaterialInstance had a public pointer constructor which
is now package private. This means that FilamentAsset needs an Engine,
so it now takes the one from AssetLoader.
- When creating a Camera component, it is the responsibility of the caller
to destroy the Entity. ShadowMap didn't do that, so it would leak two
entities.
- When destroying a Camera component with the legacy (and deprecated)
API, Engine::destroyCamera() should destroy the associated Entity
(and was documented as doing so). However, it actually didn't.
- don't use the static EntityManager is ShadowMap.
* fix compile error when vulkan
* matc：print shaders support
* fix link error
* ussage for matc print flag
Co-authored-by: brian.wang <brian.wang@noitom.com>
Tested locally by hacking gltf-viewer on Android, added the following
lines to MainActivity:
```kotlin
for (mat in modelViewer.asset!!.materialInstances) {
Log.d("gltf-viewer", mat.name)
}
```
* Fix filament-utils-android build on Windows
* Update build.gradle
* nit: use char version instead
* Fix No signature of method: java.lang.String.replace() is applicable for argument types: (Character, String) values: [\, /]
deprecated Texture(long nativeTexture) in favor of
Texture(Engine, long nativeTexture), there is no difference
currently, but in the future Texture(Engine, long nativeTexture)
will be able to validate the native pointer.
The JNI layer already does this, but can only track objects it created,
sometimes developers might create filament objects on the native side
and wrap them into java objects and this might cause a failure to
detected when objects are double-destroyed.
However, this can often be caught by the native code -- so, when the
native side is asked to destroy an object that doesn't exist, we now
return an error (exception if enabled) and we throw an exception
on the java side.
Filament typically doesn't do this kind of tests, however these bugs
can be very hard to find, and the cost is small.
- all backend return nullptr when handle_cast<> is called with
an invalid handle. this ensures a prompt crash on all backends rather
than different behaviors, some including potential memory corruptions.
assert() in debug builds. The proper way to check is to cast the
handle to bool.
- be more consistent in the GL backend about where we check for the
validity of handle -- which is basically nowhere, except for destroy
(because we have the same semantic than free(nullptr) -- i.e. no-op.
Other places are treated like bad pointers, and as per above
handle_cast<> will return nullptr.
- make sure we don't call FScene::updateUBOs with an invalid handle
when the scene is empty
To see proof that this works, use gltf_viewer with -u and examine the
node hierarchy. You will now see the correct material names instead of
the ubershader names.
When no name is provided during instance creation, it does NOT inherit
the name of its parent material. This is because instances should be
lightweight and users can already do instance->getMaterial()->getName().
In a subsequent PR, this feature will be exercised and tested via the
gltfio AssetLoader.
Fixes#2485.
- we were allocating objects with a destructor in the command stream
which is always invalid because there is no guarantee that when
the callback is called, the underlaying memory is still valid
(and it wasn't).
- AutoBuffer move-ctor wasn't moving some of its state, which would
lead to destroying the same ref several times.
This updates the tutorial markdown, fixes up the literate programming
Python script, and updates the web site itself.
The doc build script now uses a Pipfile instead of "requirements.txt",
which I find less frustrating since it does not interfere with other
Python projects on your machine.
Fixes#2483.
- bokeh rotates with the aperture diameter
- match sample count on cpu and cpu sides which affects CoC calculation
- feather blur radius to avoid visible "steps" in bokeh size
This fixes a crash with `lightbulb -m`.
- During VkImage creation, the subresource depth should be 1 since it is
meant for 3D textures.
- Render targets should select a single level from an array texture, not
all of them.
- Add debug_trap to validation errors to make them easier to debug.
This fixes dynamic lighting and SSAO when a viewport is not in 0,0.
In practice this currently happens only when all post-processing is
disabled.
Instead of using gl_FragCoord we introduce a new API,
getNormalizedViewportCoord(), which as the name implies returns
normalized [0, 1] viewport coordinates with origin at the bottom-left,
on all platforms.
This is implemented in this PR by interpolating gl_Position.
We keep very the same (or very close) quality by exploiting h/w
filtering. On the flip side, this is not compatible with dithering,
which is removed here.
In OpenGL with use glInvalidateFramebuffer() for hinting controlling
the load/store of tiles. However, glInvalidateFramebuffer() conceptually
destroys the content of the framebuffer attachment.
This causes a problem when we want for instance to use a buffer for
reading only (e.g. a depth buffer) in multiple passes. In such
scenario, the attachment will be marked as "discard" (which is a
misnomer for storeOp==DONT_CARE in vulkan parlance), without the
intention of making the buffer invalid.
We fix this by ignoring the discardEnd flags entirely if a buffer
has not been written. In this case, we relying on the driver to not
write the tiles out -- but we don't have any other way to express
this in GL.
This issue cannot be encountered currently because the framegraph
is not aggressive enough setting the discardEnd flags.
As per the spec:
```
vkCmdWriteTimestamp latches the value of the timer when all previous
commands have completed executing as far as the specified pipeline
stage, and writes the timestamp value to memory.
```
the Instance id was off-by-one and didn't match the
given entity.
in gltf_viewer this caused contact shadows to not work
when there was only a single directional light
- Added an extra README file detailing what the scripts are for
- The scripts now display a warning and a prompt before requiring user confirmation to continue
- The system install commands are only executed in CI environments (ninja and cmake being installed as root without checking for the environment was a leftover from a previous non-GitHub based CI setup)
instead of storing two counters for point and spot lights, we now
only have a single counter for both, and the light data structure
has a type, the shader can use to decide what to do -- instead of
using two loops. In practice spot and point light code is very similar
anyways.
This simplify a lot the CPU side (as in make it less complex) and frees
up 8-bits in the per-froxel GPU data structure.
In Vulkan, the arrayness of a texture is an aspect of the "image view",
not the image itself.
In my earlier attempt I used VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT on
the image, which MoltenVK does not support. Confusingly, that flag is
actually meant for rendering into a slice of a 3D texture, which we are
not doing.
Vulkan is hard.
we can calculate each plane direction from its clip-space equation
(which is trivial), instead of back-projecting 2 points and computing
the cross-product
This came up when trying to use a very small number of froxels like
1x1x1 (just one!).
Some froxels where not detected as used when scanning horizontally,
this would happen when a sphere was fully inside a froxel (i.e. not
actually intersecting any planes), and this was the very last
column on the right.
Now we use the same code we use for Z and Y (which where working fine)
for the X loop.
this could lead to w/h froxel count to be 0, later leading to
divide-by-zero. There could also be a case where the viewport itself
had w/h of zero, also leading to divide-by-zero exception.
- there was a case where setting a clear color on the Renderer
would clear a blended view with that color instead of transparent
pixels.
- move configuring the ui view into ImGuiHelper
- fix gltf_viewer sidebar so its background is not uninitialized
sometimes this happens because the shader actually doesn't use the
texture (e.g. in a uber shader model), however this is an error in
metal and vulkan.
This adds a new "backend_headers" target because small changes to any
backend cpp file was resulting in a huge build, causing materials to be
rebuilt, etc. Note that filabridge only needs DriverEnums, nothing else.
it's not good enough to clear the SSAO buffer, we also need to clear
*all* buffers involved in the final SSAO buffer, in particular the
bilateral blur buffers.
all this clearing is needed because we ignore skybox pixels based on the
depth buffer test, yet we need an initialized buffer.
Emissive was previously defined in exposure compensation stops, which
was confusing to many. It is now a value in nits, with the alpha
channel controlling how much the camera exposure affects the emissive.
At 0, the emissive value is just added to the final pixel color, at
1 the emissive value is multiplied by the exposure just like with
regular lights.
The intensity of the emissive property can be computed from an
exposure value (EV) easily with the following formula:
emissive.rgb = emissive.rgb * pow(2.0, EV - 3.0);
This formula is available as Exposure::luminance(float) already
in Filament.
- contact shadows are now supported for point and spot lights
- and are now independent from regular shadows, that is they can
be enabled without enabling regular shadows
The only limitation currently is that the distance and step count for
ALL contact shadow are taken from the directional light options.
Our enum-to-string converter was throwing an exception rather than
printing a friendly error message in some cases.
To be specific, I hit this when writing:
vec3 worldTangent;
toTangentFrame(mesh_tangents, material.worldNormal, worldTangent);
Static JNI lookups were causing issues with multiple libraries.
We now do the lookups when we need them as they are effectively
just hashmap lookups and we do them only in places where the
work we need to perform will be much larger than a simple hash
map lookup anyway.
This chane also manually registers filament-utils JNI bindings
to get rid of unnecessary symboles. We should do the same for
other Filament libraries (the symbols are pretty long and
we now have many of them).
It looks like using `half` during static initialization causes issues,
where sometimes these values are not initialized correctly.
The problem goes away when some other random static initializer
executes, which suggests a race, except there is only a single thread
at that point.
I think for some reason `half4` are not put in the data section where
float4 are.
note: we don't understand the exact issue yet
Originally it was used for validating the Engine* used, but this
makes no sense (no reason to validate this pointer more than any
other one).
The list also made sure that Engine instances were automatically
destroyed when the filament library was unloaded, but again, there
is no reason to do this -- it's C++ after all!
the problem was that the actual viewport's origin can be different
from the viewport set by the user, for instance when the view is
rendered in an intermediate buffer.
* Add new tool cso-lut
This tool is for internal usage only. It computes a 3D lookup table
for cone/sphere occlusion computations. These can be used to implement
specular occlusion, directional ambient occlusion, etc.
* Improve generated comment
* Fix Windows build
ignoring beginFrame() return value is allowed, but doing so would
leave filament in an invalid state.
we now make sure that we don't execute any code after determining whether
the frame should be skipped, and execute the remainder of beginFrame()
upon the first call to render(View*).
We also make sure the FrameSkipper can deal with its endFrame() being
called when the current fence hasn't signaled.
* API BREAKAGE: this change aims to fix multi-view support
What has changed:
- View doesn't have a notion of clear color anymore
- View doesn't have a notion of discard flags anymore
- The clear color and color-buffer discard/clear flags are moved to Renderer
- Skybox can now be set to a constant color
- View have a blend-mode
What does is all mean:
"Clearing" (i.e.) setting its background is now handled by Skybox, by
setting a constant color to the Skybox. This should take care of
drawing views side by side.
When a view needs to be drawn on top of another, it's BlendMode needs to
be set to TRANSLUCENT and of course and, generally, it wither won't have
a skybox, or will have one that sets some translucent pixels.
As an optimization, a View with BlendMode::OPAQUE will have its
background cleared with the color specified in Renderer.
If the SwapChain already has some content, it's now possible to set
the Renderer to not discard the content, together with TRANSLUCENT views,
it's possible to draw on top of that content.
It is NOT possible to share depth/stencil buffers between views.
Fixes: #2369, #2372, #2364
* Address reviewers comments.
Note: WebGL is still broken with this PR
* Expose Java API to control discard flags
This change also adds the new sample sample-multi-view that shows
how to use discard flags to render multiple views.
* Disable clears
we now flush the backend (e.g. GL) after the color pass, this a
good time because a lot of GPU work has been queued by that point, this
allows the GPU to start while we queue the post-processing commands.
This reduces latency by about 2.5ms on pixel4.
gltfio-lite now has its own template file rather than re-using the
ubershader one, which allows us to optimize the materials by removing
support for various glTF extensions like texture matrices,
specular-glossiness, and clear coat.
in OpenGL we need to check the status of queries on the gl thread,
so we did this in when beginFrame was called -- however, this is
too late, because beginFrame is asynchronous, so we were always
behind by 1 frame.
we fix this by running the callbacks more often.
The user can now choose amongst 3 specular AO methods:
- None, specAO is off
- Simple, specAO is inferred from roughness and diffuse AO
- Bent normals, specAO is computed accurately from cone intersections
The last method is more expensive but produces the best results.
This change also fixes a few issues:
- Rename materialRefraction() and materialRefractionType() for
consistency
- Fixes user time in shaders
On drivers that don't implement timer queries properly, we use
fences instead, however we can't use a fence to determine the start
time of the GPU work (only the end time). To workaround this we
use the last call to glFlush() as a proxy.
At least on Pixel it works very well.
We also remove all mid-frame calls to flush(), we did this to kick
cpu work early, but it's not as efficient as overlapping the work
with the previous frame, it increases latency a bit though.
A sync object for us is like a fence that cannot be waited on. It can
only be queried.
This is useful for filament to know if (not when) some previous
commands have finished executing.
In OpenGL this is different from "fences" which only exist in EGL and
can be waited on from any thread.
"sync" objects correspond to GLsync object in OpenGL and are widely
supported, as opposed to EGL fences.
So we use the term "fence" to refer to synchronization object that are
external to OpenGL and the term "sync" for the native gl synchronization
objects. This distinction may or may not exist in other APIs.
rewrite the FrameSkipper to use this, which enables it on MacOS.
Stores font texture as RGBA instead of R. This is because the ui material will not be compatible for RGBA textures, as it is expecting an R texture. I was unable get a texture with ALPHA format to function as I wanted.
This also removes material instance caching by scissor rect. Instead it just assigns the scissor rect to each new primitive, as well as an optional texture.
- now the DynamicResolutionOptions don't specify the target frame rate,
they only specify how to scale this given view
- there is a new FrameRateOptions setting on Renderer, which is used
to specify the desired target frame rate for the whole Renderer
- the frame rate is now specified as a "frame interval" in units of
the display frame period.
- the display frame period is (indirectly) set via DisplayInfo.
In other words, the use must set (and update) DisplayInfo properly
(the default are reasonable, but assume 60 Hz). They must also set the
desired interval (default is 1, which is probably what you want).
Finally they must enable dynamic scaling per View, the defaults are
also reasonable.
Currently Renderer doesn't attempt (yet) to actually target the
requested frame rate, so it's up to the caller to push frames at the
desired speed. however, dynamic resolution, like before, will attempt
to shrink work to fit the target.
Renderer can now be given a DisplayInfo that contains some
important information about the current display. This will be used
for frame-pacing and dynamic-resolution.
This is most relevant on Android, where we can accurately query these
parameters. Added support for that in our samples.
This also sets the minSDK to 19 (including gltf viewer's), since this is
the version we support. We were cheating before by under-reporting our
minSdk.
Until now, only the memory of Hw handes was allocated synchronously,
handles were constructed later asynchronously. This was error prone
with the few (but growing) synchronous calls we have.
Now we have an init<> call that allocates and constructs handles synchronously,
the asynchronous construction if needed, is still performed with construct<>,
which is now implemented as "destroy+construct", this is okay because
most of our handles have trivial dtors.
This fixes a race with TimerQuery and Fence. Fences are still not fully
working though, but at least won't access uninitialized memory.
Currently backend fences MUST be used trough FFence.
On metals (made only of specular light), specular AO can
create large black spots. This change simply kills specular
AO to avoid this artifact. In practice we should take into
account multiple bounces but this is rasterization...
This adds a simple query API to the backend.
There are 2 methods to create/destroy queries, which are
essentially futures. And 3 methods to mesure elapsed time:
beginTimerQuery/endTimerQuery and getTimerQueryValue.
The begin/end pair is not nestable.
On the GL backend side, there are 2 implementations of this, one uses
arb_timer_query or disjoint_timer_query, the other uses fences.
We need both implementations because on some GPUs, including
qualcomm's elapsed-time timer query is useless, as it measures
cpu time.
Metal/Vulkan implementations will be part of subsequent PRs. Vulkan
will be able to implement this with vkWriteTimestamp which is
reported to be accurate.
An immediate benefit is that we can now get frame times on MacOS, which
should allow it to use dynamic-resolution.
the view clear flags would not be honored when rendering directly
into an imported render target.
note: now that this works this could cause an issue when using
multiple views side-by-side and one of them doesn't have a skybox,
that view will trigger a clear of the whole buffer (because clears
clear everything, not just the viewport).
A solution will come in a later PR.
Fixes#2312
* Add support for bent normals
Bent normals can be enabled via the bentNormal property of a material.
When specular occlusion is enabled, bent normals improve the quality
of the computation.
* Save a couple of multiplications in bent specular AO
This needs a bit more testing before merging, but I wanted to put up the
PR for review. I will look into supporting Android after we land this.
Fixes#1932.
- handle heightFalloff=0 (i.e. fog doesn't depend on height) correctly,
previously, a divide-by-zero on the cpu side would get in the way.
The fix is to clamp heightFalloff to a small-enough value, and to
make sure that this is handled correctly in the shader.
- default fog distance is 0 instead of 1m
- inScatteringSize parameter should be allowed to be large in samples
This replaces the "bindings" structures with simple types that re-use
data structures from cgltf.
This also moves some trivial work from ResourceLoader to AssetLoader,
which simplifies the communication between these two classes.
Motivated by #1932.
In some situations the shadow map would be too aggressively clipped.
This was due to a bug in the box-frustum intersection algorithm, which
could fail when some vertices of the frustum were inside the box, and
some vertices of the box inside the frustum and the total number
of vertices such classified was 8.
* Add support for screen-space contact shadows
This CL adds support and always enables it.
toggles and setting in the next CL (same PR).
* Plumb settings for screen-space contact shadows
screen-space contact shadows is handled like a shadow option,
parameters (including on/off) are set in the LightManager, using
the existing ShadowOptions API.
Additionally there is a per-renderable toggle.
Both toggles are off by default.
* Allow contact shadows when shadowing is auto-disabled
Shadowing can be auto-disabled when for instance there are no
shadow casters in the scene. We still allow contact shadows in
that case.
This would allow for instance, to make the vegetation on a terrain
not shadow-casting, and still get some shadowing there by using
contact shadows instead.
* apply micro-shadowing after contact shadows
also, don't compute contact shadows when we know we're fully shadowed.
- decouple the depth and ssao passes, so that the depth
pass could be used by another client, down the framegraph
- set uniforms/samplers from the execute closure of the color pass,
instead of using a separate pass. This is more correct and it
potentially reduces calls to commitUniforms.
- make setSampler check that the value set is different,
the idea here is that it's much more costly to have to commit the
samplers rather than having to compare them. Also realistically we
don't have a lot of samplers in a frame, so that's not a lot of
compares.
- one downside of this change is that commands for the SSAO pass are
always generated, we will fix that in an upcoming change.
Imported render targets are only partially implemented, which
caused a discard flags bug. It would happen when an imported
render target is used more than once like when using SSrefr
without tone mapping. In that case, the 2nd pass would use the same
discard flags as the first pass.
We make sure to only (and always) clear newly allocated buffers,
but only clear existing buffers based on the user request. This
should allow to share the depth buffer between drawing two
Views (if post-process is disabled). This should also fix
a bug where the intermediate buffer wasn't cleared when blending
a view on to of another.
This should fix#2138
This change adds new commands to matinfo to print dictionaries.
This feature is useful to debug dictionaries, and I just used
it to identify that using a shared dictionary would save ~12 KiB
in the Android build. The macOS build (GLSL + Metal) would go
down to 71 KiB from 136 KiB.
The SPIRV to Metal conversion leaves leading spaces and some
comments. This change also runs the shrinker on optimized Metal
shaders to further reduce the size of shaders.
We were not using the correct offsets, however because everything
was so blurry, it wasn't obvious that something was wrong.
This should result in a bit smoother bloom effect.
It controls the quality of the upsampling filter. This can help a lot
if heavy scaling is needed to maintain performance, it also helps if
MSAA is not enabled.
There are 3 quality levels, low, medium and high. 1, 4 and 9 bilinear
taps are used for each level respectively. The high quality setting
employs a tent filter.
we've had issues with some devices running in arm mode, where they
would throw spurious SIGILL on the WFE instruction used in spin locks.
Since on Android Mutexes are very efficient, we just use that instead.
This might fix#2197
libmath itself doesn't expose any stream operators anymore. However,
libutils is able to automatically print libmath types into its
io::ostream -- however matrices are not formatted nicely.
Added a new optional library, libmathio, that provides std::ostream
operators for all libmath types. libmathio does a better job at
formating matrices.
Also removed apply() and map() from libmath because there were not used
anywhere and they forced us to depend on <functional> in public headers.
The scissor is now always ignored -- i.e. when we clear, the whole
surface is cleared, which echoes how discard flags work.
Geometry must be used to clear only a sub-region of the screen.
This simplifies a lot of things because we had workaround for Adreno
GPUs, which were slow when clearing with the scissor enabled.
Major changes are:
- RenderTarget are now real resources as opposed to ad-hoc objects.
This allows to remove a lot of code and simply a lot of things.
- ResouceNode now have a level of indirection (i.e. we store a pointer
to them), which allow fast and more correct implementation of
moveResource().
A new unit test checks that all scenarios of moveResource with a
generic resource are working.
- We have a special version of moveResource that moves a RenderTarget
resource into a TextureResource -- this shouldn't be allowed, but
this helps make the API better. What it means is that
"all rendertargets that use the specified texture as attachment
are replaced with the given rendertarget".
- In this version there are a lot of limitations when moving
a rendertarget into a texture resource; be we cover the case that
is useful to us.
This fixes the validation error that occurs when rendering to one
miplevel while sampling from another. It might even be faster since
fewer layout transitions are required.
Closes#2159
There was two issues:
1) Fence::wait would return prematurely when the the timeout was
set to WAIT_FOREVER.
2) (At least) when running from the command line, events were not
read from the main queue. Instead we have to use NSRunLoop instead
of NSApplication APIs.
* Add ground shadow option to gltf_viewer
The shadow is always positioned at the bottom of the loaded object's
bounding box to ground it perfectly.
* Remove unused include
The ultimate goal is to get rid of the concept of render targets in
the framegraph, and this is the first step. Here RenderTarget doesn't
store the discard flags anymore. These flags are now calculated
per-resource, per-pass.
A side effect of this change is that the discard flags are now slightly
better, as they don't contain attachments that don't exist.
i.e. we won't see the STENCIL flag set for instance.
- this simply changes the createRenderTarget() API to take an
'MRT' structure for the color attachment. The MRT structure is
just a wrapper around an array of four TargetBufferInfo.
- in this PR only implemented for OpenGL
- the tuple<> where we store the parameters should only remove the
reference, but not decay, so that arrays could theoretically
be stored (more on this below).
- use std::forward<> instead of std::move<> in Command's ctor, this is
more appropriate, we just want to forward the arguments unchanged.
- add std::move() to all parameters on the command stream stub, it is
appropriate here, because we receive parameters by value or r-value
references.
All this was an attempt to support C-style arrays as parameters, but
it doesn't work because they are decay'ed to pointers when they
appear as parameters.
-DFILAMENT_SKIP_SAMPLES=ON with CMake
-Pfilament_skip_samples with gradle
This change also renames CMake options specific to Filament
to avoid clashes with subprojects.
- minor uniform optimizations (saves ~0.1ms)
- fix some missing highp precision qualifers
- make it easier to tweak the blur filter
- make sure samples's radius goes from 0 to 1
- update comments
After the native async functionality landed, there was no way for web
clients to be notified that the decoding has finished. This PR changes
the existing `onDone` callback so that it gets called after all textures
have been decoded. (Previously it was called after downloading rather
than decoding.)
This has the side effect of simplifying the API because clients no
longer need to call a finalize function.
During the SSAO pass, we pack the decoded depth to the GB channels of
the AO texture, which reduces our bandwidth requirements during the
blurring pass, as well as some ALU usage.
It puts SSAO at about 2ms on pixel4/1080p.
Currently it only selects how many samples are used.
Low,medium,high and ultra respectively map to 7,11,16 and 32 samples.
The default is "low", which is sufficient for most mobile applications.
This is achieved by computing a small 2x2 box blur in the AO pass
taking advantage of quad shading, which allows to halve the size
of the bilateral blur kernel.
Reducing the size of the blur kernel has a side effect to kick Adreno
gpu into direct mode, which apparently is much faster here.
Overall we go from 3.4ms to 2.4ms on Pixel4.
The quality is impacted, but not severely. This probably assumes
GPU that have working derivatives.
- got rid of all the precomputed samples for now, it makes things
much more simple, and actually doesn't slow things down. Use the
analytic offsets (spiral) instead.
- use textureLod instead of texelFetch in places where we need to do
how own clamping, turns out it's faster (as measured on pixel4).
- use a interleaved gradient noise instead of a hardcoded blue noise,
the difference isn't big.
Overall we're actually 0.2ms faster out of 3.5ms. It's not immediately
clear why, might just be noise.
* Add support for ccache
To benefit from ccache, first install ccache (for instance with
brew install ccache on macOS). On my machine a clean build takes
~5 minutes the first time. A second clean build with ccache enabled
takes 34 seconds.
* Enable more aggressive caching
This is 466 KB insted of the usual 850 KB. It assumes that users do not
need spec-gloss, non-lit, or a special transparency mode.
(uncompressed SO size on arm64)
We can add a new maven project for this next week.
This doesn't affect quality significantly, but saves 30% of gpu time
on the upsampling, overall it's about 0.1 ms. This brings The bloom
effect below 2ms on Pixel4.
We untonemap/tonemap the first level of blur to reduce the very high
frequencies due to HDR highlights in the image, this produce a softer
image at higher roughness. This can programmatically be turned off,
but that setting is not exposed.
The very first downsample stage costs us a lot because it reads from
a full-res texture. We mitigate this by always doing a blit to
1/4 res. Blits are quite a bit faster, this saves about 1ms on
Pixel4.
math/mathwfd.h forward declares all {mat|vec}{2|3|4}<> classes,
which allows us to remove their respective #include in a lot of
our public headers.
Our math headers are full of templates, so this should help build times
a bit.
Also we want to keep the public headers as minimalist as possible.
To trigger an exception users could either shrink the window or enlarge
the sidebar. Constraining both of these is somewhat tricky so let's
just clamp the viewport width at a low level.
There was actually no need to give special treatment to leading drive
designators since they effectively form the first path segment anyway.
To help prevent regressions, I added a few unit tests in a previous CL.
The motivation for the CL is to remove a dependency on `locale.cpp`,
which can result in shorter build times and reduced binary sizes.
For Android and wasm we do not use the filesystem and therefore do not
need `utils::Path`.
```
libgltfio-jni.so BEFORE 1.9 MB (693 KB gzipped)
libgltfio-jni.so AFTER 1.4 MB (542 KB gzipped)
```
Threshold can be either on or off, and only thresholds at 1.0 when
activated (in pre-exposed mode).
This allows to use very high strengths values for the bloom without
softening the rest of the image, and is useful for artistic considerations.
This is a temporary (hopefully) hack to make the skybox sun look
better when bloom is enabled (this doesn't affect how the "sun" lights
objects, only how it looks in the skybox).
This scale factor was adjusted empirically.
Note that libgeometry is already included in `filament-android`, this
simply exposes more of its existing functionality.
The Java version of SurfaceOrientation is similar to the JavaScript
version because we are bundling it into the main Filament package, even
though it is a separate library in C++. This is much simpler than
creating a brand new Java package.
New Android sample that tests this is forthcoming.
Fixes#1729.
because the intermediate buffer could have non-square pixels, blurring
the buffer would result in an anamorphic blur.
We fix this by properly scaling the buffer before blurring.
Fulfills a request from tirichards@.
This is similar in some ways to dynamic backface culling but simpler
because there is no interaction with double-sided lighting.
For reference, dynamic backface culling was implemented with #1936, #1877, #1641.
This "fixes" the new animated Fox model in the glTF conformance suite.
As per the glTF spec, we now generate per-face normals for the case
where normals are not specified in the model.
Note that true flat shading (i.e. flat interpolation) is not a
requirement in the glTF spec, since some Khronos members advocate for
WebGL 1.0 compatibility.
Fixes#2088.
It turns out that textureLod() only reads the base layer if the sampler
is not set to xxx_MIPMAP_xxx. This prevented the gaussian-blur code
to work properly.
Specifically, the horizontal pass would always use the base layer. Not
only this made the blur unstable, but also hurt performance significantly.
With this change, the untonamap/tonemap is no longer needed and the
results are much more convincing with objects with highlights.
- always take the "blit" path (rather than quad path) when MSAA
and is on. we now rely on the render target to have been resolved
into its texture (implicitly or explicitly). This means in some cases
there will be 2 bliss in a row -- but that's still better than a
blit and a quad. These are unavoidable without using a custom resolve.
- removed creating renderbuffer with the EXT_multisampled_render_to_texture
extension because we didn't support it fully and didn't support
the fallback mode (for desktop or if the extension is not there).
Also this case never happens currently.
- added more assert()
- added a bunch of comments explaining the restrictions of using MSAA
regarding blitting.
This feature adds one new method to `FilamentAsset` and uses it in our
Kotlin, JavaScript, and C++ helpers:
utils::Entity popRenderable() noexcept;
This pops a ready renderable off an internal queue, or returns 0 if no
renderables have become ready. It provides a simple way for clients to
gradually add renderables to the scene as they become ready. Previously
clients could only get the entire list of entities, regardless of
whether they had Renderable components or complete textures.
To facilitate this feature, this PR adds a new internal-only class to
gltfio called `DependencyGraph`, which is a temporary object used for
bookkeeping during the asynchronous load.
`DependencyGraph` discovers ready-to-render entities by tracking the
textures that each entity depends on. This is a graph because
renderables connect to a set of material instances, which in turn
connect to a set of parameter names, which in turn connect to a set of
texture objects. These relationships are not easily inspectable using
the Filament API or ECS.
If a client application selects a backend that is not supported by the
build, Engine::create() now returns null, which allows the Java wrapper
to throw IllegalStateException. Prevously, we would crash in the native
layer by hitting a null pointer in Engine::loop().
We now use a fixed-size kernel (right now 17), for each mip-level.
And we workout the mapping from roughness to such mip.
This improves performance by reducing significantly the number of taps
per mip, however, we don't control how much blur is actually applied
at each level. This gives a mip-chain that is proportional to
'roughness'.
It's now easier to tune quality vs. speed by tweaking the kernel's width
as well as how we map 'sigma' to a the kernel width.
We were already using jobs for decoding PNG and JPEG files, but we were
doing a join. This add three methods to ResourceLoader that allow
clients to amortize the decoding process across multiple frames, even on
single-threaded platforms like WebGL.
This PR adds async loading to the following demos:
- samples/gltf_viewer (now shows a progress bar in the UI)
- android/sample-gltf-viewer
- web/samples/helmet.html
Fixes#1876.
Clients should be prevented from allocating unused buffer slots because
uploading to an unused slot triggers undefined behavior in the backend.
Motivated by #2042
With this PR screen-space refraction is functional.
Caveats:
The blur passes for supporting rough refractive materials is quite
heavy and increases with the resolution.
SSR uses a gaussian approximation for the brdf and therefore doesn’t
match perfectly cubemap-based refraction and IBL.
The use of MSAA with screen-space effects, while working, is going
to incur a large cost, especially on tilers.
This gives us a compile time error instead of link-time error, when
using a parameter type not supported. The code is also a bit more
self-documenting.
In my first attempt I kept the multiple project-level dependencies in
each sample's `build.gradle` but that resulted in a gradle error
concerning multiple copies of the same SO file. To work around this
we now append to "srcDirs" to bring in Java dependencies rather than
using project-level dependencies.
The new APK contents are now much more reasonably sized:
libfilament-jni.so 981 KB
libgltfio-jni.so 817 KB
libfilament-utils-jni.so 96 KB
Previously this was:
libfilament-utils-jni.so 1.8 MB
libgltfio-jni.so 1.8 MB (unused)
libfilament-jni.so 1 MB (unused)
Fixes#2070
Replace with forward declarations if needed and includes in .cpp that
now need them.
The idea here is to have our headers have the least amount of impact as
possible on our clients (e.g. compilation time).
If you need this, you know who you are.
The new Texture::Builder::import() allows to specify a OpenGLES texture
id from a shared GL context, effectively importing the texture into
filament.
The texture is NOT destroyed when calling Engine::destroy(Texture*), only
its filament handle.
The NOOP backend didn’t call user-API callbacks from the main thread
(instead calling them from the “backend thread”). This caused problems
with unit tests for instance.
This reduces the LOC of our gltf-viewer app from 260 to 146. This could
have been more, but I feel this provides good balance between
flexibility and convenience.
The raw gltfio API involves multiple objects (FilamentAsset,
AssetLoader, ResourceLoader) whereas the ModelViewer has a simple facade
consisting of two methods, one for GLB and one for GLTF. The underlying
gltfio objects are available via non-settable properties.
Add bindings for releaseSourceAsset() and fix a double-free issue.
According to the Android Studio profiler, this makes Java memory usage
go from 18.6 MB to 6.5 MB.
glTF specifies that client implementations must support at least two UV
texture coordinate sets but does not seem to have an upper limit.
We use a mapping table in order to honor high-numbered UV set indices,
but our mapping table only accomodates 8 entries.
Fixes#2042
We currently use static linking and will continue to do so. This means
that libfilament-jni is a subset of of libgltfio-jni, and libgltfio-jni
will become a subset of the upcoming libutils-jni.
This is a simple layer cake, but requires adding some CMake logic to the
gltfio-android CMakeLists, similar to what we already did for the
filament-android CMakeLists. This prevents an obscure CMake error about
duplicated build artifacts.
Also, previously we required clients to call AssetLoader#init(), this
is now renamed to Gltfio#init(), which is more consistent with
Filament#init().
When UiHelper is set to translucent with setOpaque(false), the
target SurfaceView's surface is set to be composited on top of
the activity's window. There are situations however where one
may want to render regular Views on top of the SurfaceView.
The new setMediaOverlay() API allows exactly that: when this
property is enabled, rendering to a translucent target will
cause the SurfaceView's surface to be composited above other
surfaces, but below the activity's surface.
* Added sRGB support for DXT compressed textures - OpenGL
* Added sRGB support for DXT compressed textures - Vulkan and Metal
* Added detection if OpenGL driver supports sRGB compressed textures (EXT_texture_sRGB extension)
* Fixed whitespace to use spaces rather than tabs
* Test for GL_EXT_texture_sRGB or GL_EXT_texture_compression_s3tc_srgb support for sRGB DXT compressed textures
* GL ES uses GL_EXT_texture_compression_s3tc_srgb instead of GL_EXT_texture_sRGB
FrameGraph always keeps a default Blackboard, but additional ones
can be created if needed.
A blackboard is just a registry of FrameGraphHandles, queryable by name.
This makes the upcoming Java bindings simpler and improves parity with
existing API conventions.
Note that camutils does not depend on filament and therefore cannot use
BuilderBase.
Note that the Manipulator Builder state is public which breaks with
convention but makes implementation simple and allows C++ clients to
continue using modern initializer syntax if they wish.
The main purpose of this change is to fix a problem where a resource
marked as sampleable by a *culled* pass would stay sampleable.
This happened because the sampleable state was tracked on the resource
itself.
We fix this by treating sample() just like read(), such resources are
added to an array living in the pass, which ensures the state is correct
when passes are culled.
Also fixed:
- simplified read() code a bit (minor)
- simplified resource aliasing code
- don't resolve() renderTargets of culled passes
setParent() is not supposed to reorder child/parents in the array
because it doesn't rely on them being orderer. This also has the
advantage of not invalidating Instances.
However, commitLocalTransformTransaction() does rely on child/parent
ordering and the previous change did fix that.
The current code only updates the order in the transaction commit if there's a single out-of-order node in the scene graph, but if multiple are out of order (or one swap doesn't fully fix the ordering), then we end up with out of order scenes, and incorrect world transforms.
Additionally, when nodes are reparented outside of a transaction, the order is not updated.
The FrameGraph would erroneously specify a sample count on textures
that are sampleable and used as an attachment to a multisampled
target -- however, that's not allowed.
We didn't run into issues, because had a workaround for this when
creating the texture. This is augmented by an assertion.
- we were not unbinding the renderbuffer int the right place,
when it's created, instead we were doing that when using it the
first time. Unbinding soon makes sense because we never need to
rebind it after that.
- in debug builds we were clearing render targets at creation time,
but this was wrong because a render target won't necessarily be written
(it could be used as the source of the blit for instance).
This didn't cause any trouble in the framegraph, by chance, it was smart
enough to reuse existing render targets in the case of blitting.
Ideally we'd clear textures/renderbuffer at creation time,
merge frameBufferRenderbuffer() and framebufferTexture() because
the filament API doesn't have a distinction. This actually removes
quite a bit of code too.
- write() is marked as [[nodiscard]] since it's
generally an error to discard it
- graphviz export not display wether a texture resource is
an actual texture or renderbuffer
* Rewrite Maven publish tasks to support multiple flavors
* Don't build filamat with an unused CMake option
* Don't compiler debug in release mode
* ... for real
* Link to BUILDING.md
This was untested because our only glTF Android sample uses glb instead
of JSON. We will soon be adding a new Android demo that uses an actual
gltf file.
This fixes#1841.
Instead we use the clear flags that the user specified. This is needed
in the case where we're rendering in the target in two passes, the
first pass might want to clear the buffers, while the 2nd one will not.
If we were combining the clear flags, the first pass's flags would be
sticky and the view would always be cleared.
If combining is needed it must be specified by the user.
This is in preparation to supporting screen-space effects.
There are two major changes:
- RenderPass is now copiable and intended to be passed by copy to
the execute stage of frame graph passes
- The color pass is in its own function now
This actually simplify RenderPass api.
We have one very important unit test that is supposed to make sure
the framegraph reuses the same rendertarget between 2 passes, this
unit test would always succeed when using the NoopDriver.
Now that we can mock the ResourceAllocator, we can ensure this unit
test is doing its job.
Rotate the sample pattern at every light direction, which trades
aliasing for noise (which is arguably more pleasant) for IBLs with
very high frequency (i.e. high level of HDR-ness).
* Re-organize Android Gradle files
Clean up our Gradle files, share versioning, etc. and prepare for
publication to sonatype.
* Use androix annotations for desktop
* Add samples as subprojects to root Gradle
* Fix build script
* Don't break when samples aren't compiled
* Better organize the Android Studio project
We were conflating two concepts in our comments about the roughness
remapping, which was confusing. Also changed the remapping so it
matches a log2 mapping in our default configuration.
The C definition was changed to include METAL, moving NOOP further down.
The Java definition never caught up, making NOOP mean METAL and making the
real NOOP unreachable.
This patch puts them back in sync. Also, updated the Java comments to match
the comments from the C definition.
This moves the camera manipulator into its own library and adds new
functionality including a new "Google Maps" manipulator and a bookmark
feature to facilitate camera animation.
Java bindings and an Android sample will land later this week.
In glTF, multiple nodes can refer to the same mesh, and each node can be
assigned a unqiue string label. Previously we used NameComponentManager
to annotate entities with mesh labels rather than node labels. This was
wrong because Filament entities are 1:1 with nodes, not meshes.
We still fall back to mesh labels when node labels are not provided.
Fixes#1989 by fetching extension function pointers that were refactored
in #1977. Moved string-splitting utility to a common location so that
the base class and concrete classes can both use it.
With KTX, we should simply create the specified output folder and put
the KTX file(s) in there. There is no need to create a subfolder of the
output folder.
This change also allows users to pick a custom name for their KTX file,
because we only use the "deployment" name when generating filenames.
Previous to this change, we suffered from a flash-of-red followed by a
flash-with-wrong-projection every time we resized an OpenGL window.
SDL throttles the window resize events, so FilamentApp updates the View
only on the final low-level resize event. However we were calling
Cocoa's update method on every low-level resize event. Thus we were
showing uninitialized content during the resize.
FilamentApp recreates the SwapChain after the final resize event; one
reason for this is the fact that Vulkan swap chains must be recreated.
The fix is to defer the Cocoa update until the swap chain gets
recreated. This is the perfect time because FilamentApp has already set
the new Viewport at this point.
We now should be able to use PlatformEGL for all generic EGL platforms
(not tested though). Most texture streaming operations are not supported
in that case, since they're platform specific.
An version of clang is smarter about generating warnings for template code. `assert` here is not included, so clang generates: `error: use of undeclared identifier 'assert'`
When doing the depth path for ssao, we need to remove the blended
objects, i.e. produce a depth path similar to the regular depth prepass.
We also need to clear the SAO buffer. It's not needed for opaque
objects because they are written in the SAO's depth buffer. This fixes
"ship in a bottle" scenario.
If spherical harmonics are not specified in IndirectLight (i.e.
omitted entirely), we instead use the roughness==1 LOD in the shader.
This produce much better results with absolute HDR environment, and
generally works better more often.
This is now the preferred solution.
For cases where spherical harmonicas are better or preferred, they
can simply be supplied as before.
The thin roughness remapping only makes sense for thin surfaces, it was
a mistake to apply it to solids. It assumes that a ray will go
undisturbed though the thin surface with an ior of 1.
In PR #1906 we enabled depth for all backends, but we forgot web.
Note that this required 0 new codelines whereas Vulkan required 116
additional codelines. :)
This cleans up 3 related issues, but only the first of these caused an
actual user-visible bug:
(1) When the IBL was null, prepareLighting was failing to bind the black
1x1 texture.
(2) If you added and removed an IBL from a Scene, then Scene was not
returned to its initial state.
(3) The default 1x1 IBL used an intensity of 1 rather than 30000, which
is inconsistent with the null IBL.
Fixes#1940.
We now take into account the light bouncing between the two interfaces
of the materials. We can do this because we assume thin surfaces,
meaning that the light will exit where it entered and with the same
direction that the specular.
This strengthen the specular at grazing angles (and reduces the transmission).
When dealing with thin objects we really have two thicknesses to
consider, the thickness in the direction of the normal, which
corresponds to the value used for solid objects, and the thickness of
the object's walls, which generally is a constant.
Reusing thickness in the later case is problematic for assets that have
a thickness map, but are rendered hollow.
In the future we can even imagine handling double-sided hollow objects
by using the thickness information -- e.g. a hollow cube.
this is because for solids we're also handling both interfaces in one
go (as opposed to calculate the actual btdf), and roughness should
depend on the ior.
Also rename some variables for less confusion (we now use
incident(I)/refracted(R)).
We also now assume a thickness of 0, if it's not set in the material.
We don't want to assume any particular thickness, because it depends
on the scene's scale.
Also clamp the remapped roughness.
Normally the IOR is deduced from the reflectance, but now we allow to
specify the IOR instead or in addition to the reflectance.
In the later case, it's possible to create physically impossible
materials, but this can be useful for artistic reasons.
Filament will now allow incomplete mipmap chains for the IBL and
map the roughness to the available chain.
This allows to create roughness cubemaps with a minim size for
roughness==1, instead of always mapping it to 1x1.
cmgen will now use 16x16 cubemaps for roughness==1, which
improves the quality of rough objects significantly.
The beauty of this is that there is no asset or API change. Old assets
will continue to work like before.
This cleans up 3 related issues, but only the first of these caused an
actual user-visible bug:
(1) When the IBL was null, prepareLighting was failing to bind the black
1x1 texture.
(2) If you added and removed an IBL from a Scene, then Scene was not
returned to its initial state.
(3) The default 1x1 IBL used an intensity of 1 rather than 30000, which
is inconsistent with the null IBL.
Fixes#1940.
This removes a lot of code and will make it easier to migrate into a new
library with Java bindings, etc.
- The clipping planes and projection matrix were unused, removed them.
- In split view mode, our demos were using only 2 of the 3 manipulators,
removed one.
- Instead of callbacks and update methods, users can simply ask for the
current matrix at every frame.
- The manipulator's job is to consume mouse events and maintain an
"look at" basis, there is no need to store a reference to a Filament
Camera.
some drivers declare supporting anisotropic filtering, but don't
support calling glSamplerParameter() to set the max anisotropy, presumably because they only support glTexParameter().
We simply turn off anisotropic filtering on these drivers.
Sometimes CMake decides to use the JNI bits included with XCode instead
of those from the installed JDK. The header files are located in different
directories. This does the proper include.
This new sample differs from "Hello Camera" in that it exercises all
three ways of using Stream. It also uses Canvas instead of Camera2 to
draw into the external texture. It shows two sets of stripes, one
animated using the shader and the other animated using Canvas. If the
two stripes are aligned, then the stream is perfectly synchronized.
Users can tap the screen to cycle between the three modes.
Unfortunately we require RTTI to be disabled for web builds
due to the way in which we handle array buffers. This may change
in the future but for now I think it's an acceptable constraint,
especially given the fact that RTTI adds bloat anyway.
A texture resource can end-up with not usage bit set if it is used as
an attachment of a render target that gets replaced by a moveResource()
call. The framegraph should take care of culling our this resource,
in that case, but currently doesn't.
We worked around this issue by not declaring the resources, but it's
better to do this in the frame graph code, as to not affect the
user facing API.
Fix#1165
In order to use transparent views, post-processing had to be disabled
because we were not able to blit post-processed buffers with blending.
This is now fixed by simply reverting to a quad.
A side effect of this, is improved performance when a scaling blit is
needed when MSAA is active.
This also introduced new "bugs", or at least weird behaviors: the
system needs to know when blending is needed, and this has to be
based on heuristics (unless we add a new api). Currently the heuristic
is that the COLOR buffer is not discarded and the view is cleared with
an alpha value (or not cleared).
1) When scaling is enabled, we're using a blit at the end of the frame,
however, if MSAA is also enabled, this blit is not allowed, so we
need an explicit MSAA resolve.
2) When rendering directly into the default render target with MSAA
enabled, we need an explicit resolve ONLY IF there is a format
conversion, because those are also not allowed.
Before, we were doing this resolve regardless.
3) The test for "rendering directly into the default target" had
become wrong because, the Render target can now be a user texture
(which is assumed to have the proper MSAA-ness and a depth buffer),
so no intermediate buffer should be needed for those. We now
check that render target is actually the default render target
We used to have this functionality, but it we removed it because we
didn't have a use for it and it was costly.
This new implementation is only internal, and would (will?) be useful
if we wanted to generate a special buffer for SSAO for instance (instead
of just depth). This implementation, overrides the material at the time
the commands are executed, which is much cheaper than the previous
implementation.
For the street light glb from the Khronos suite, this reduces texture
loading time from 290 ms to 110 ms.
Stay tuned for another feature: notification callbacks.
Related to #1876.
- we add an 'intensity' parameter that allows to control the strength
of the AO effect. This is useful for aesthetic reasons.
- the default intensity is now 2x that of before this change, which
makes the intensity parameter match AO papers this implementation
is inspired of.
- bias is now z-dependent, which reduces some self shadowing wrt the
depth.
This is needed because filament doesn't make strong assumptions about
the units used (e.g. meters), so we can't hardcode a distance in the
shader. But also, this is a cheap change.
ssaogen now produces better formatted tables, directly usable in
shader code.
SAO shader now squares the sample radius outside of `tapLocation`, also
added some comments.
Previously, Streams had two modes (native and texid), this adds a third
mode called "acquired", which allows for copy-free synchronized external
textures in OpenGL and paves the way for Vulkan.
The native mode is now deprecated but texid mode needs to stay around
until all clients can be upgraded.
In an early prototype, this functionality was added directly into
Texture but required quite a bit of additional state tracking, so the
Stream API seemed like a better fit.
This API is probably not necessary for Metal due to Metal's shared
ownership semantics.
This has been tested with a new Android sample that will be added in a
subsequent PR.
RenderPass's API now doesn't need to know about FScene, instead we
pass the geometry info as parameters.
This is one step towards keeping the "scene" concept more away from
rendering.
This fixes an oversight in #1641 for scenes that have only 1 primitive
when the depth prepass is enabled. In such cases, the same prim is
rendered twice in a row: once for depth and once for color.
We had enhanced the main render loop to override the primitive's
backface culling state when the material instance changed, but that was
not sufficient since the same material instance can be used for depth
and color passes.
Fixes#1872.
New internal API to add custom commands (lambda) before or after each
pass. These commands can be added at any point and will be sorted
properly.
e.g.:
appendCustomCommand(Pass::COLOR, CustomCommand::PROLOG, 0, [](){
// custom code
});
Also split appendSortedCommands() in two functions to make it easier
to insert custom commands.
On Debug builds, HeapAllocatorArena needs LockingPolicy::Mutex because
it uses a TrackingPolicy, which needs to be synchronized.
On Release builds, HeapAllocatorArena doesn't need a LockingPolicy
because HeapAllocator is intrinsically synchronized as it relies on
heap allocations (i.e.: malloc/free)
- this is to allow the use of C++17 std::invoke
- don't std::move() each argument of the tuple<> instead
std::move() the tuple itself and perfectly-forward each argument.
- handle return values
- handle synchronous commands
The "clampNoV" shader function is ignoring the passed dot product argument and recomputing the dot product itself. This will result in the wrong NoV value being used for clear coat IBL computations.
TrackingPolicy::Debug didn't store the base pointer of the Area, and
instead relied on the first allocation to discover it, however, because
of alignment, the first allocation may not match the base pointer.
Because of that there could be an overflow in onRewind(), i.e. we
could rewind to a pointer before the (wrongly computed) base. This
overflow caused the debug memset to go awry and stomped on memory.
This is fixed by passing the base pointer to the constructor of the
TrackingPolicy. This base pointer could be nullptr with certain
allocators, but in that case, onReset/onRewind should never be called;
and this is enforced at compile time.
Also fixed a (luckily) harmless buffer overflow when preparing the
dynamic lights, if the number of lights wasn't a multiple of 4. This
was harmless because we use a linear allocator, so overflows are not
really overflows.
Clients who do not yet have this fix can usually work around this issue
by calling the non-array overload of `MaterialInstance::setParameter`
when the array size is 1.
This was caught by ASAN.
Our algorithm header has many one-liners that compute the "next power of
two length / 2" but they all have the caveat that if the input is
already POT, then the "/ 2" part does not occur.
Usually we deal with this by testing the difference against zero.
However in `partition_point` we were skipping the test, thus causing a
potential out-of-bounds access.
I fixed `partition_point` and added a few more tests for non-POT cases.
This doesn't change the user facing API because it already used a
callback.
With this change we implement readPixels() with a PBO, which won't
stall the GPU. Then we periodically check a fence to see when the
command is completed, at which point we map the PBO and copy the data
to the user buffer.
This is useful for unit tests. The content of the swapchain can be
read to main memory with Renderer::readPixels().
This PR doesn't implement this new feature in the following backends and
platforms (TODO):
VulkanDriver.cpp
MetalDriver.mm
PlatformWGL.cpp
PlatformWebGL.cpp
PlatformCocoaTouchGL.mm
Optimized macOS builds can trigger "stack_not_16_byte_aligned_error"
when building with clang 11. We happen to see this with resgen but it
could happen anywhere.
https://forums.developer.apple.com/thread/121887Fixes#1823
Filament keeps a global static list of engines for verification purposes
which was being recreated multiple times, causing a panic. After a lot
of experimentation with the build, I was not able to fix this while
keeping the current scheme of dynamically loading both libraries.
As a "solution", we now staticly link filament into gltfio-jni, and ask
clients to load only that.
Fixes#1811.
This enhances the assert added in 2548da9 in the following ways:
- Adds console output at startup to help detect "double load" bugs.
- Adds PRETTY_FUNCTION to show where the invalid usage is coming from.
- Tweaks the wording since use-after-destroy is not always the cause.
- Consolidates usage of static globals into a helper class.
Motivated by #1811.
This enhances the assert added in 2548da9 in the following ways:
- Adds console output at startup to help detect "double load" bugs.
- Adds PRETTY_FUNCTION to show where the invalid usage is coming from.
- Tweak the wording since use-after-destroy is not always the cause.
Motivated by #1811.
This extension dumps out some interesting vendor-specific information
to the log, which could be useful for debugging. Interestingly, some
of Android demos produce this output on Pixel 3:
KHR_debug: EsxBufferObject::Map - Ignoring EsxBufferMapUnsyncedBit
Enabled for debug builds only.
This simple Kotlin app projects a live feed on the side of a cube.
It looks very similar to the `lit-cube` demo but there is a new Kotlin
class called `CameraHelper` that creates Camera2 objects, starts
the capture session, creates the Filament texture, etc.
For now, this app uses SurfaceTexture. This is a starting point and
provides a testing ground for tentative improvements to our Stream API.
- we use by-value parameters in some places, which helps the compiler
(not sure why, but assuming references can be aliased and the
compiler is cautious about that).
This helps matrix multiplies (which is almost never inlined) and
transpose (but that's almost always inlined and merged with
something else, so the gain here is not completely real)
- remove support for matrices-of-matrices because this wasn't complete
anyways.
- improve matrix * scalar by inlining the loop manually instead of
going through *=, this saves some extra writes.
We were not checking Camera's preconditions properly -- we assumed that
near < 1/1024 meant "near == 0".
Now we check preconditions properly and revert to the same hardcoded
default if they're not met -- doesn't matter what the default is,
it's there only to avoid NaNs and Infs elsewhere -- rendering will
be bogus eitherway.
Fixes#1766
Sparse accessors are easy to punt gracefully, we can emit a warning
and instead use the dense "base" data.
Also, after cgltf support for this lands, we will need to apply sparse
data in ResourceLoader, so this adds a bit of plumbing to prep for that.
Relates to #1727.
We now assert if the Engine given is invalid,
which would happen if it was used after being
destroyed.
Builder::build() is a reasonable place for this since it's generally
not in the critical path.
The vertex shader in the material variant for skinning-and-morphing
declares a `uvec4` attribute that can be unused and disabled, but still
needs to be typed correctly.
The fix is motivated by WebGL, but the issue might occur with other
OpenGL implementations that have strict drivers.
Fixes#1726
These constants are not part of the standard. We instead use our own
constexpr definitions in the filament::math namespace, as part of
the scalar.h include.
* Add the ability to modify clip space coordinates in the vertex shader
This introduces MaterialVertexInputs.clipSpaceTransform, a mat4 that
is applied to gl_Position before exiting the vertex stage.
* Address code review comments
* Lower limit from API 21 to API 19
This was requested by an internal application. API 19 is when OpenGL
ES 3.0 support was added so there is no good reason for us to not
support this API level. The only trick is to avoid referring to the
glTexStorage2DMultisample symbol directly as it only exists in 3.1.
* Compile out code we never use
* Use reflection to handle shared EGL contexts pre-API 21.
* Remove comment
* More fixes required to run on API level 19
- dlym() fails for ashmem on API 19, so we only try on API 26+ instead.
- Older emulation for OpenGL ES 3.0 returns error states for valid API calls so we need to clear the GL error bit before we create the GL driver.
- Activity lifecycle changes since API 19 would cause animations to keep running and to reference destroyed objects.
- EGLContext.getNativeHandle() is new in API 21, we need to use reflection on API 19. The new code path uses the class loading trick to avoid a bytecode verification error on API 19.
* Filament now runs properly on API level 19
This commit adds a new api_level() API to libutils which can be used to
query the platform's API level. On Android it works as expected, other
platforms currently return 0.
This is to allow designated aggregate initialization in C++20.
We do this because we have been relying on C99 designated
initialization, which is only supported by clang, other compilers need
c++20 to get a similar functionality. Therefore, we now try to limit
Ourselves to C++20 rules.
This could break existing code, since now SamplerParams needs to be
zero-initialized:
SamplerParams p{};
This is generally not an issue because the intended use is:
doSomethingWithSamplerParams({
.filterMag = LINEAR,
});
the C++ standard says:
if T is a class type with a default constructor that is neither
user-provided nor deleted (that is, it may be a class with an
implicitly-defined or defaulted default constructor), the object
is zero-initialized and then it is default-initialized if it has a
non-trivial default constructor
Unfortunately, MSVC always calls the default constructor, even if it
is trivial, which breaks constexpr-ness.
To workaround this, we're always zero-initializing TVecN<>
Also removed constexpr from default constructors, since they never can
be constexpr as they're not initializing the vector.
* Add new setExposure(float) API on Camera
This API can be used to set a specific exposure value. For instance:
camera.setExposure(1.0)
Will set the exposure to 1.0 (or ISO 100, apeture 1.0 and shutter speed 1.2s).
This can be useful to match the lighting setup of other engines/tools. Setting
the exposure to 1.0 can for instance be used to treat lights as being unitless
(a directional light could have a strength of 2.0 for instance).
Fixes#1667
* Update release notes
* Fix Java build error
* Fix Web bindings
We need to document the backend headers that are public.
Added documentation for:
- DriverEnums.h
- BufferDescriptor.h
- PixelBufferDescriptor.h
Disabled doxygen for public headers of backend that don't need to
be public for filament's API.
Also improved documentation for:
- LightManager.h
- RenderTarget.h
- TextureSampler.h
This is a similar fix to the previous vector fix. Commutative
operations now always return the same type and the operations are
carried out in the precision resulting from following traditional
C++ rules.
It used to be that operations e.g. like:
float3{} + double{} would be computed as
float3{} + float3{double{}} instead of
float3{} + double3{double{}}
I other words, when an implicit conversion was involved on the right
it would be converted to the left side’s type, possibly losing
precision.
Another problem was that swiping the operands could produce different
Results, e.g.:
float3{1} * 5.0 -> float3{5.0f}
5.0 * float3{1} -> double3{5.0}
This is no longer the case, now both expressions would return a double3.
Note:
float3 r{};
r *= 5;
Is now equivalent to:
r[0] *= 5;
r[1] *= 5;
r[2] *= 5;
Instead of before:
r[0] *= 5.0f;
r[1] *= 5.0f;
r[2] *= 5.0f;
We now follow the same rules than for basic types, when performing
mixed-precision operations, e.g.:
float3 + double3 -> double3
double3 + float3 -> double3
dot(float3, double3) -> double
In particular, e.g. swapping the arguments to arithmetic operations
now always yields to the same result type (before, float3 + double3
would not return the same type than double3 + float3).
To prevent an excess of variants, sometimes our vertex shaders declare
inputs that they never read from. For example, our skin-and-morph
variant might never read from the skinning attribute, but still declares
the input.
This PR is a sister to #1663, which is a fix for #1525.
See also #1279.
We now use GENERAL layout for depth-sampled images in order to support
the SSAO use case of simultaneous binding + sampling.
We may be able to optimize this in the future by enhancing DriverAPI
so that it feeds more information into the render pass.
Fixes#1627, #1649.
We now inverse the face winding order based on the determinant of
each render primitive world's transform. This prevents front/back faces
to be reversed when using a left-handed basis, after transform.
Fix#1520
Inspired by our recent scissor change (#1639) but motivated by gltfio,
which was forced to create a ton of ubershaders (see #1562).
Tested with:
gltf_viewer -u ../glTF-Sample-Models/2.0/TextureSettingsTest/glTF/TextureSettingsTest.gltf
This is nicer because this way the methods that don't match don't
even exist. Also make it better when editing code with a C++ aware
ide.
Also use the less verbose std::enable_if_t<>
We now have OpenGLContext which only tracks the OpenGL state.
It exposes an API very similar to OpenGL and does all the state
tracking. Currently, it doesn't expose non state-changing APIs
(e.g. draw methods) and only implements the state-changing API we
need (to track the state of).
If you tried removing `culling: none` from any post-process materials,
everything would be fine in GL but Vulkan would be black because
our full-screen triangle was back-facing.
Continue reading only if you thrive in absurd situations.
- Metal and OpenGL define clip space as Y-up whereas Vulkan is Y-down.
- Metal and Vulkan define tex coords as Y-down whereas OpenGL is Y-up.
* Add Android presubmit
* Build Android on Linux, macOS targets have NDK 18 only
* Consolidate workflows and jobs
* Try to build Android on macOS
* Fix typo
* Cleanup scripts
* Try NDK side by side if NDK bundle has the wrong version
* Simplify logic
* Fix NDK logic
* Only update the NDK when necessary
* Fix typo
* Update required NDK version
* Favor NDK side-by-side
* Remove support for obsolete NDK bundle, require NDK side-by-side
* Install the NDK side by side when missing
This prevents values like NONE or COLOR which are very generic
names to collide with other classic enums. It also forces us to be
specific about what we're doing when converting to bitfields.
Also added a utility to easily enable any enum or enum class to
support binary operation (i.e. Bitmask contract), e.g.:
template<> struct utils::EnableBitMaskOperators<Foo> : public std::true_type{};
gives all binary operators to 'Foo'. Foo must be an enum.
Also added any() and none() convenience to convert an enum to a boolean.
This simplifies its usage at call sites at lot.
The C++ ctor was needed before to create uninitialized objects, which
was needed in SamplerGroup. Instead we use a custom 'static_vector'
(instead of std::array), this actually makes SamplerGroup's
implementation clearer.
static_vector<> is just a fixed-capacity, fixed-storage dynamic
vector. we currently don't expose it and implement only what we
need.
The suzanne Vulkan demo was asserting because `setViewportScissor` can
only be called within a render pass, which means that
`MaterialInstance::use` can only be called within a render pass.
We probably need a better way to enforce these rules...
Every time ninja invoked CMake, we were unconditionally regenerating
all of our "licenses.inc" files which caused a ton of re-compiles.
Before the change, building after touching `ubershader.mat.in` took over
a minute, now it takes 20 seconds.
The CMake "file" commands are bad. The CMake documentation says:
If the file is a build input, use the configure_file() command
to update the file only when its content changes.
Some buffers (like animation data) do not contain any vertex data
but web-based clients still need to know about this.
This interface is much simpler than the bindings lists, and in fact
we might remove the binding lists in the future to simplify the API.
Fixes#1593.
some driver don't support all float buffers as rendertarget, so we
improve the backends (opengl in this PR) to return what it actually
supports and we update the client (filament) to take that into
account when making decisions.
by using variadic macros we can greatly simplify the macros
definitions for declaring the backend API.
in particular we don't need a macro per number-of-argument, this is
now done automatically.
Unfortunately, we still can't handle zero-argument declarations, so
we keep the _0 variants.
This performs surgical modification of the IFF chunks rather than
invoking filamat from within matdbg. Low-level direct manipulation
(bypassing optimization passes etc) allows us to diagnose issues with
the shading pipeline.
This CL also adds keystroke bindings to the Monaco editor. You can
press Cmd+S to rebuild the current materials, or use Ctrl+Arrow to
navigate between shader variants and materials.
In a subsequent CL I will add a README that describes how this works in
detail, it will include a list of limitations and a feature wishlist.
We now check for FILAMENT_ENABLE_MATDBG and simply skip the draw call
when the program is invalid. The initial error message is still dumped
to the console, but we will not terminate or dump an infinite cascade of
error messages.
This only adds Engine-side support for edits. The debugger client will
be enhanced in another PR. This works by simply clearing the program
cache and swapping out the MaterialParser. Neat!
However, there are some gotchas:
- The DebugServer callback is triggered on another thread so to be safe
we defer the swap until the subsequent call to getProgram().
- The getProgram() method is const but in reality it can insert entries
into a mutable cache. This CL makes the constness of it even more
misleading by applying edits.
- The edit operation will leak the old HwProgram but I feel this is
fine because simplicity is crucial for injected behavior and this is
a developer-only feature.
Texture are special resources that can be read in different ways; either as
attachment (read) or sampled (sample), the user must now declare explicitly if she intends
to sample from a texture.
sample() implies read() so there is no need to call both.
This was already the case, as read() used to take a boolean, essentially meaning the same
thing. However, the bool didn't make sense for non-texture resources.
This also simplifies the implementation.
the framegraph will figure it out automatically from the
rendertarget sample count. this is done during resolve(), and after
moveResource is taken into account (so this works with imported and
moved resources).
when creating a rendertarget, if the sample count is not specified,
any existing target with the same attachment will match.
this simplifies both the user and implementation.
getRenderTargetDescriptor() now takes a RenderTargetHandle instead
of a texture handle.
This can be enabled by using the Tracking::Debug policy,
currently this just fills allocation on alloc() and free(),
which is useful to help detect access to uninitialized memory
and use after free.
Now enabled by default in libfilament allocators on debug builds.
This caught uninitialized access in the froxelizer.
I tested this with the problem model and a couple other models.
This bug is inherited from cgltf_node_transform_local, so I will
upstream the fix. There are two ways to see that the new math is
correct:
1) Look at decomposeMatrix and note that scales belong to rows,
not columns.
2) Look at the Brandon Jones implementation:
http://glmatrix.net/docs/mat4.js.html#line1079Fixes#1519.
We were storing pointer to RenderTarget objects that were allocated
in a vector<>, which obviously is not safe.
This didn't cause any issue most of the time because our allocator is
a linear allocator.
We've been using STB for a while now and we already include a JPEG
decoder in our wasm bundle. This removes some vestigial code that was
getting in the way.
Fixes#1565.
Performing `base + offset` pointer arithmetic is only allowed when `base` itself is not nullptr. In other words, the compiler is assumed to allow that `base + offset` is always non-null, which an upcoming compiler release will do in this case. The result is that CommandStream.cpp, which calls this in a loop until the result is nullptr, will never terminate (until it runs junk data and crashes).
Avoid this by using intptr_t instead of actual pointers (i.e. char*), which seems to avoid this failure.
* Improve materials under white furnace test
Two major changes:
- Mobile target now implements a cheaper variant of the off specular
peak bias (which moves the reflected vector towards the normal).
This greatly helps with rough surfaces that may otherwise point
toward a bright part of the IBL.
- The indirect diffuse love is now properly attenuated to avoid adding
the energy reflected by the specular layer. This allows dielectrics
to be correctly energy conserving under a white furnace.
- Tweak the (hacky) clear coat layer attenuation to behave properly
under a white furnace.
* Use the same reflected vector modification everywhere
* Add #include preprocessing to filamat and matc
* Update RELEASE_NOTES
* Fix RELEASE_NOTES
* Use final instead of virtual / override
* Clarify comments
* Use pure virtual for includer functions
* Use a callback instead of an interface
* Rename Includer.h to IncludeCallback.h
* Update comment
matdbg is now linked into the Filament Engine in debug config (allowing
live inspection of GLSL / SPIRV) and into the matinfo tool (to support
the --web-server option).
In both cases, the library spins up a small web server that listens to
http://localhost:8080. You can run any Filament app and attach to it.
The web client caches all material information. This allows the user to
close an atttached Filament app, and the web app will continue to
function properly (useful for crash diagnosis). Moreover the user can
launch a second Filament app and the web client will add its materials
to the existing list (useful when comparing two Filament apps).
For now this only supports inspection, not editing. Some of the material
info such as required attributes is not yet displayed but this will be
easy to flesh out in a subsequent PR.
This is a dependency of our upcoming web-based material debugger. This
CL also includes `tnt/CMakeLists.txt`, which builds a static lib in a
minimal configuration that enables WebSocket support. The entire library
is built from only 4 files:
${PUBLIC_HDR_DIR}/CivetServer.h
${PUBLIC_HDR_DIR}/civetweb.h
${SRC_DIR}/civetweb.c
${SRC_DIR}/CivetServer.cpp
access to resources now needs to be done explicitly before calling
useRenderTarget(). This simplifies things and is needed for more
simplifications to come.
This enhances the exisiting matc option "--debug" so that it includes
debug information in the resulting SPIR-V.
Previously we tied this to the build configuration which isn't very
flexible. Moreover this needs to be orthogonal to shader opts, because
we often need to debug problems that arise from optimization.
This CL also gives meaningful names to the ubershader materials and
fixes up the new matinfo analysis so that it prints the entire SPIRV
chunk that corresponds to each potentially problematic GLSL codeline.
Motivated by #1516.
This is an improved workaround for a known limitation with Vulkan
drivers on Android P. The offending SPIR-V sequence was introduced
during shader optimization and can be avoided by using full precision
for certain variables.
This adds a new option to matinfo that does some very dumb regex-based
analysis on the disassembled SPIR-V, then dumps out transpiled GLSL
that has annotations added to the end of some codelines. For example:
```glsl
float luminance(vec3 linear) // POTENTIAL MIXED PRECISION %49 = OpDot %float %linear %48; relaxed = %49 %linear
{
return dot(linear, vec3(0.2125999927520751953125, 0.715200006961822509765625, 0.072200000286102294921875));
}
```
The code currently tries to find potential mixed precision but does not
do the right thing yet for pointers and structs. This is just a starting
point to help diagnose problem areas in our generated SPIR-V.
* Update Froxelizer.h
Fix this error when building with msvc from vs2019
error C2926: 'filament::details::Froxelizer::FroxelEntry::<unnamed-tag>::offset': a default member initializer is not allowed for a member of an anonymous struct within a union
* Fix some compilation issues with vs2019/msvc
Program.cpp:
1>C:\greg\github\filament\filament\backend\src\Program.cpp(28,42): error C2610: 'filament::backend::Program::Program(void) noexcept': is not a special member function or comparison operator which can be defaulted
1>C:\greg\github\filament\filament\backend\src\Program.cpp(28,42): message : exception specification does not match the implicitly declared specification.
GLUtils.h: __PRETTY_FUNCTION__ macro is clang specific. Use MSVC equivalent
Color.h: fix warning
* #1493 - inline constructor in definition as requested by @romainguy
* #1493 "move this #define inside the #else below" as requested
* #1493 revert last change which causes compilation failures on other platforms.
provide empty implementation of Program::Program() in Program.cpp
* More fixes for building with vs2019/msvc
* #1500 use consistent macro definition syntax (@bejado)
* #1500 simplify DEBUG_COMMAND macro as requested by @pixelflinger
* #1500 use `{ 0 }` which is accepted by Visual Studio (` = 0 ` is not accepted)
* #1500 remove incorrect UTILS_RESTRICT alltogether
We're only caching textures (not render targets yet), and we're
evicting cache entries older than 30 allocations.
This should cut down on texture allocation / gl calls.
The cache should get in a stable state very quickly (less than half
second).
* Update Froxelizer.h
Fix this error when building with msvc from vs2019
error C2926: 'filament::details::Froxelizer::FroxelEntry::<unnamed-tag>::offset': a default member initializer is not allowed for a member of an anonymous struct within a union
* Fix some compilation issues with vs2019/msvc
Program.cpp:
1>C:\greg\github\filament\filament\backend\src\Program.cpp(28,42): error C2610: 'filament::backend::Program::Program(void) noexcept': is not a special member function or comparison operator which can be defaulted
1>C:\greg\github\filament\filament\backend\src\Program.cpp(28,42): message : exception specification does not match the implicitly declared specification.
GLUtils.h: __PRETTY_FUNCTION__ macro is clang specific. Use MSVC equivalent
Color.h: fix warning
* #1493 - inline constructor in definition as requested by @romainguy
* #1493 "move this #define inside the #else below" as requested
* #1493 revert last change which causes compilation failures on other platforms.
provide empty implementation of Program::Program() in Program.cpp
This removes the samples/envs folder and replaces these KTX files with a
build step that is driven by CMake / gradle / bash, depending on
platform.
This makes it easier to use IBL files that are generated by the latest
and greatest version of cmake.
Fix this error when building with msvc from vs2019
error C2926: 'filament::details::Froxelizer::FroxelEntry::<unnamed-tag>::offset': a default member initializer is not allowed for a member of an anonymous struct within a union
Some drivers have performance issues when repeatedly (once per frame)
allocating and destroying render targets, so we will
re-introduce a cache for that.
this just adds the hooks for the cache.
We use R11F_G11F_B10F for HDR so we should stop checking for compressed
IBL files.
Also, make the glTF sample on web more consistent with the default
settings used by gltf_viewer on native and use venetian_crossroads
instead of syferfontein.
We should not use the address of a stdlib function for BufferDescriptor
callbacks. With emcc this compiles without warnings or errors, but
produces incorrect code that causes intermittent out-of-memory errors at
run time.
* Use GL_EXTENSION for GL_OES_EGL_image_external_essl3
A prior change that reworked some code incorrectly attempted to
use the existing EGL extensions to set a GL extension boolean.
Added additional GL extension code.
* Fix typo
* Fix typo
Avoid enabling scissors when the viewport is the same size as the
buffer. On Adreno devices, running clearWithGeometryPipe adds
three additional unresolves for each bin.
VkFramebuffer requires its attachments to have only one miplevel each,
but our VulkanRenderTarget wrapper was sharing the VkImageView that
associated with the underlying texture object.
This fixes up VulkanRenderTarget so that it owns a unique VkImageView
that is pinned to a specific miplevel.
We now build three gltfio libraries:
- gltfio_core ....... lightweight library with ubershaders
- gltfio ............ uses filamat to generate materials at runtime
- gltfio_pipeline ... depends on path tracer functionality
This fixes one of the bugs seen with `gltf-bloom` on Android.
We already dynamically map Filament's DEPTH24 format to either
VK_FORMAT_D32_SFLOAT or VK_FORMAT_X8_D24_UNORM_PACK32, depending on
the platform. Therefore we should return `true` when asked if the
backend supports DEPTH24. Otherwise the client will not be able to
create a depth texture.
(1) Sometimes (but not always!) gltfio-android was failing to build due
to two missing ANativeWindow functions. Linking in "android" seems to
fix this, and is consistent with libfilament-jni.
(2) Improve our "build.sh -c" utility by clobbering some additional
Android build directories. This is especially useful after
adding "-Pextra_cmake_args=-DFILAMENT_SUPPORTS_VULKAN=ON", otherwise
Gradle will try to use a cached CMake configuration.
By design, VulkanRenderTarget has weak references to its attachments and
does not manage any hardware resources. Therefore, when we start
rendering to a particular render target, we need to inform
VulkanDisposer that the command buffer has acquired a reference to the
underlying textures. This prevents the possibility of the texture being
destroyed while still in use.
This fixes#1450 although we should also add a render target pool for
best performance.
This works by aliasing CUSTOM0 - CUSTOM7 to morphing attributes, and by
extending our existing skinning variant.
This PR was tested against some upcoming changes to gltfio.
Issue #1149, #1417
When processing very high dynamic range environments, the importance
sampling code falls appart, it becomes a user choice to decide if
prefilter importance sampling is better or worse than just regular
importance sampling -- both are usually bad.
--ibl-no-prefilter gives the user this choice.
In the case where we have 2 cores, we would spawn only one thread in
the thread pool. If that thread got to try to steal() from another
thread before the main thread was adopted, it would end-up always
trying to steal from itself and enter an infinite loop.
This seems to happen during windows builds.
This combines two constants into one, and changes it into a value that
is actually correct. :)
Technically the movement of the CUSTOM attributes will change the layout
qualifier and therefore merits a materials version bump, however custom
attributes were only recently introduced so this seems unnecessary.
Did some quick testing with 3 samples: gltf_viewer, lucy_bloom, and
point_sprites.
Clamping is now disabled by default in cmgen, there is a new option
to enable it "--clamp".
Automatic SH windowing is also enabled by default and can be controled
with the "--sh-window" option. Accepted parameters are "no" to disable
windowing, "auto" for automatic windowing or a number to specify the
cutoff band.
auto windowing only works for 1, 2, and 3 bands.
Currently we're arbitrarily clamping environments to 16384 because both
the pre-filtering and SH algorithms can't handle very larger dynamic
ranges. Instead of clamping, we now tonemap, which is a little bit better.
It turns out that most of libmath couldn't be used in constexpr
expression due to our use of union{}. The C++ standard requires that
all accesses to a union{} in a constexpr expression be the same
element.
Also because libm and cmath are not constexpr some functions such
as length() or normalize() can't be constexpr. The same is true for
anything needing things like sqrt, cos, sin, ceil, floor.
This change mainly does the following:
- replace all accesses to vector elements by operator[]
(this ensure all of libmath uses the same union element)
- avoid use of std::min / std::max / std::abs
- avoid uninitialized variables, which can't be constexpr
- remove 'constexpr' keyword on functions that can never be
It is now possible to write things like:
constexpr mat4f I = inverse(
transpose(mat4f::translation(float3{ 1, 2, 3 })
* mat4f::scaling(4)));
--sh-window=band, -w band : this low-pass-filters the environment
such that bands above 'band' are zero. This can be used to reduce ringing
when the source environment has high frequencies
--noclamp : turns off clamping before processing the cube map
This is still work in progress.
The conversion factors from radiance to irradiance where wrong.
The bug above was found while refactoring the code to be clearer. Now
the method that computes the coefficients for the shader calls the
regular SH code and applies all the appropriate factors on that.
With this change the options "--sh-shader" and "-sh=3 -i" won't
produce the same result because --sh-shader includes the lambertian
diffuse. "--sh" now always produces actual SH coefficients.
This is a prep step for the upcoming morph feature and does not require
a bump to our material version number.
Stay tuned for a new sample app that demonstrates this feature.
this is because the JobSystem's queue works as a LIFO, by creating
jobs in reverse (memory) order, we attempt to help streaming to
the d-cache on that threads -- until the point where
jobs are stolen.
we also execute the last job immediately instead of creating a job
for it -- since we're already in a job.
- parallel_for doesn't use recursion anymore to create the "leaf"
jobs, this is now done linearly on N thread (one thread per CPU).
This uses less stack space, and reduces miss-predicted branches.
- remove almost all SYSTRACE calls because they have a huge impact
on things like parallel_for() and are misleading. They can be
enabled again by setting HEAVY_SYSTRACE to true.
- we simplify the waiting code by using only a single
condition variable instead of two.
- wait() now behaves just like a looper, it will process jobs until
the one it's waiting for finishes -- before it could just sit there
(the idea was that the job would finish quickly, but that's not always
the case).
- we also make sure to never call notify_n() when it's not needed.
We track how many waiters we have and use that to decide if we need
to notify().
notify is pretty slow on all architectures, even on linux it's always
a syscall, so it's better to avoid it.
- don't use stand-alone fences, makes things ugly for no real benefit
- refactored the code a bit, hopefully it's more clear.
If the client opts in to "recomputeBoundingBoxes", then we manually
compute a bounding box that ignores the glTF min / max annotations.
This computation was erroneously including the transform of the injected
root node, which is not part of the model.
This could cause a problem when creating the asset, then immediately
positioning it with its injected root node before the ResourceLoader
is done downloading vertex buffers.
The Work-stealing dequeue indices could wrap around after ~2 billion
calls to steal(). This could probably be achieved in a few hours.
By using 64-bits indices, we avoid the problem entirely.
This uses strtof rather than stringstream and provides a common location
that can be leveraged by the upcoming JNI bindings.
Note that this new method lives in KtxBundle rather than KtxUtility. The
latter creates Filament textures and therefore does not get built into
libimage.
Here are all the libraries available in the group `com.google.android.filament`:
| Artifact | Description |
| ------------- | ------------- |
| [](https://maven-badges.herokuapp.com/maven-central/com.google.android.filament/filament-android) | The Filament rendering engine itself. |
| [](https://maven-badges.herokuapp.com/maven-central/com.google.android.filament/gltfio-android) | A glTF 2.0 loader for Filament, depends on `filament-android`. |
| [](https://maven-badges.herokuapp.com/maven-central/com.google.android.filament/gltfio-android-lite) | Trimmed version of `gltfio` that does not support some glTF extensions. |
| [](https://maven-badges.herokuapp.com/maven-central/com.google.android.filament/filament-utils-android) | KTX loading, Kotlin math, and camera utilities, depends on `gltfio-android`. |
| [](https://maven-badges.herokuapp.com/maven-central/com.google.android.filament/filamat-android) | A runtime material builder/compiler. This library is large but contains a full shader compiler/validator/optimizer and supports both OpenGL and Vulkan. |
| [](https://maven-badges.herokuapp.com/maven-central/com.google.android.filament/filamat-android-lite) | A much smaller alternative to `filamat-android` that can only generate OpenGL shaders. It does not provide validation or optimizations. |
### iOS
iOS projects can use CocoaPods to install the latest release:
```
pod 'Filament', '~> 1.27.0'
```
### Snapshots
If you prefer to live on the edge, you can download a continuous build by following the following
steps:
1. Find the [commit](https://github.com/google/filament/commits/main) you're interested in.
2. Click the green check mark under the commit message.
3. Click on the _Details_ link for the platform you're interested in.
4. On the top left click _Summary_, then in the _Artifacts_ section choose the desired artifact.
## Documentation
@@ -36,26 +76,21 @@ badges above.
- [Material Properties](https://google.github.io/filament/Material%20Properties.pdf), a reference
sheet for the standard material model.
## Samples
## Examples
Here are a few sample materials rendered with Filament:
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Benjamin Doherty
Philip Rideout
Mathias Agopian
Balazs Vegh
gaborvalasekandLevente Koncz
Johnathon Selstad
Filip Henningsson
Romain Guy
MasTraER
Alan Eneev
hzzengxiaoqiandMathias Agopian
BStringhamVRSK
MasTraERandBenjamin Doherty
daemyung jang
LaiJF
Rahul Sheth
Josh Faust
Dom Corvasce
dependabot[bot]
Pei Jia
MinWoo Kim
Pavel Korolev
Alexander Biggs
chubei-oppen
chubei-oppen
Elie Michel
HanYunChenLuo
Alexey Pelykh
Ondrej Stava
hzzengxiaoqi
Jared Fowler
Ankbzpx
jeanlemotan
abwood
Joel Winarske
Timo Röhling
clayly
BStringhamVRSK
Rupert Rawnsley
jeanlemotan
Ian Brown
xjdkc
James Walker
James Walker
Adrian PerezandPhilip Rideout
Thomas Gorisse
prewettg
Seth Vargo
MirsfangandMathias Agopian
ritwik
cjsliuj
dsternfeld7
Yixing Lao
daichen
toppa102
Yadikaer Yasheng
MorphTail
Richard Chang
ruitaocc
@roxlu ☾
Tim Psiaki
Aleksi Sapon
Mingwei Samuel
Max Rebuschatis
Michael LouandMichael Lou
Jaime Blasco
neshumeandneshume
brian.wangandbrian.wang
Kostiantyn Zghirovskyi
KJ Liu
Nakshathru Ajay
jaynus
Logan Lawrence
piggie
Kai Chieh Liu (KJ) 劉凱傑
Matthias Moulin
brian.wang
John Davis
Philip Rideout
Daniel Lehmann
Jordan Rupprecht
Russell Chou
flashk
Nicklas Ansman Giertz
Gregory Popovitch
Hendrik Wagenaar
Ben Doherty