Updated to Bevy 0.7. Here are PRs which I used as a reference for update:

• https://github.com/bevyengine/bevy/pull/3959
• https://github.com/bevyengine/bevy/pull/3979
• https://github.com/bevyengine/bevy/pull/4181
• https://github.com/bevyengine/bevy/pull/3989

But it's a draft since I faced the upstream issue: https://github.com/bevyengine/bevy/issues/4507 To make it work I need MaterialPipelineKey to have public fields as it was before (before it was a simple tuple). Should be fixed with https://github.com/bevyengine/bevy/pull/4508

Windows 10 64bit, nVidia GTX 3080Ti, Vulkan

Running the city example results in:

2022-10-25T23:52:46.783547Z ERROR bevy_render::render_resource::pipeline_cache: failed to process shader: 
error: Function [57] 'retreive_surface' is invalid
     ┌─ wgsl:1172:2
     │  
1172 │   }
     │ ╭─^
1173 │ │ 
1174 │ │ fn retreive_surface(material_index: u32, uv: vec2<f32>) -> Surface {
1175 │ │     var surface: Surface;
     · │
1181 │ │         surface.base_color *= textureSampleLevel(textures[id], samplers[id], uv, 0.0);
     │ │                                                  ^^^^^^^^^^^^ naga::Expression [64]
     · │
1205 │ │     return surface;
1206 │ │ }
     │ ╰─^ naga::Function [57]
     │  
     = Expression [64] is invalid
     = Shader requires capability SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING

Full console output: full.log

None of the examples work, I guess because of the above.

Now with

suppress_temporal_accum: true,
spatial_reuse: false,
direct_spatial_denoise: false,
indirect_spatial_denoise: false,
temporal_anti_aliasing: false,

we see pure path tracing results

In Cargo.toml you need:

bvh = "0.7.1"

because of cargo tree:

├── bvh v0.7.0
│   ├── approx v0.5.1 (*)
│   ├── glam v0.21.3

Two different glam versions (0.21 and 0.22) broke the compilation.

Currently we can toggle bevy-hikari using camera_render_graph. But the plugin continues to load the system very heavily due to the the building of accelerating structures. It would be great to have an option in config to disable it. This will allow to use debug mode with this plugin disabled for prototyping gameplay. Currently I experiencing low FPS in debug mode when this plugin is disabled only via camera_render_graph.

Sources & implementation details used from: https://github.com/GPUOpen-Effects/FidelityFX-FSR2

Its only active when setting in config upscale_ratio other than 1.0, for example 2 or 1.5.

The results are not good. Very blurry. At this moment I am not sure, if FSR1 just doesn't work well with path tracing or there are some issues with my implementation. fakeTextureGather could be the problem.

I will continue to improve this and maybe others can help.

Previously used render feature is a group that contains unneeded stuff, like bevy_ui or bevy_text. Specified features in this PR is the required minimum. You can read more about in in this issue: https://github.com/bevyengine/bevy/issues/4202

Also bevy_atmosphere has bumped its version, so now it contains your PR.

Any time reservoir is reused, its lifetime increases. Reaching the max lifetime, its replace by new one. The max lifetime if a configuration parameter and can be adjusted by needs.

Run any of the examples, like simple.rs.

In the console log there will be a huge, about 30s long delay during startup. It happens while the window is already visible, but still blank.

Console log:

2022-11-28T22:08:41.604666Z  INFO bevy_render::renderer: AdapterInfo { name: "NVIDIA GeForce RTX 3080 Ti", vendor: 4318, device: 8712, device_type: DiscreteGpu, driver: "NVIDIA", driver_info: "526.98", backend: Vulkan }
2022-11-28T22:08:45.744495Z  INFO bevy_hikari::mesh_material::mesh: Loaded mesh 0
2022-11-28T22:08:45.744551Z  INFO bevy_hikari::mesh_material::mesh: Loaded mesh 1
2022-11-28T22:08:45.744694Z  INFO bevy_hikari::mesh_material::mesh: Loaded mesh 2

Freeze is here.

2022-11-28T22:09:16.117321Z  INFO naga::back::spv::writer: Skip function Some("load_previous_reservoir")    
2022-11-28T22:09:16.117416Z  INFO naga::back::spv::writer: Skip function Some("store_reservoir")    
2022-11-28T22:09:16.117518Z  INFO naga::back::spv::writer: Skip function Some("store_previous_spatial_reservoir")  

This delay slows down development considerably.

I run the sample inside CLion with the Rust plugin installed on Windows 10. It is just plain "Run", not a "Debug" run.

(May be related to #11? Not sure)

thread 'main' panicked at 'internal error: entered unreachable code: metal requires all arrays be constant sized', /Users/ivan/.cargo/registry/src/github.com-1ecc6299db9ec823/naga-0.9.0/src/back/msl/writer.rs:235:21

  15:        0x1033378d8 - core::panicking::unreachable_display::h9fe587038728a68b
  16:        0x103338938 - <naga::back::msl::writer::TypeContext as core::fmt::Display>::fmt::h830f7fce5b7d1939
  18:        0x103338db0 - naga::back::msl::writer::TypedGlobalVariable::try_fmt::he6647d568d76e87b
  19:        0x103346afc - naga::back::msl::writer::Writer<W>::write_functions::hb2fcf614440db836
  20:        0x103344b40 - naga::back::msl::writer::Writer<W>::write::hbf0c1c30c083a2be
  21:        0x1033dae84 - naga::back::msl::write_string::h48019c2e606a918a
  22:        0x103308f60 - wgpu_hal::metal::device::<impl wgpu_hal::metal::Device>::load_shader::h06204f483b121b6c
  23:        0x10330cdcc - wgpu_hal::metal::device::<impl wgpu_hal::Device<wgpu_hal::metal::Api> for wgpu_hal::metal::Device>::create_compute_pipeline::hd744e8db82390d76
  24:        0x1032ca094 - wgpu_core::device::Device<A>::create_compute_pipeline::h0e0b9a7aa36ce61b
  25:        0x1032558c0 - wgpu_core::device::<impl wgpu_core::hub::Global<G>>::device_create_compute_pipeline::h1eecbb8c46938a60
  26:        0x1032dda20 - <wgpu::backend::direct::Context as wgpu::Context>::device_create_compute_pipeline::hd73131fee9b7c4cf
  27:        0x1032a1b7c - wgpu::Device::create_compute_pipeline::hdde1293e61f9ab4b
  28:        0x10315330c - bevy_render::render_resource::pipeline_cache::PipelineCache::process_queue::h2895d4c6fa89eb85
  29:        0x103125e00 - 

Currently when Rapier is enabled I have constantly printed warning Encounter an error when loading mesh: MissingAttributeNormal and no debug rendering is present. Could we have an integration with Rapier via an optional feature or at least do not have such warning? Minimal example project to show how this debug rendering looks (hikari is commented out): rapier.zip

Device: 13in 2020 Mac M1 16GB Source: Version 0.2.8 - Built from Source Code

Problem: Running 'cargo run' command on the M1 Mac devices produces a 'main' thread panic exit of type 'wgpu validation error'. The error produced was:

Caused by:
In Device::create_bind_group_layout
too many bindings of type SampledTextures in stage COMPUTE, limit is 16, count was 64