Raytracing shader binding table records are addressed through Vulkan
alignment rules that distinguish between the raw shader group handle size,
the per-record handle alignment, and the base alignment required for SBT
region addresses.
The previous layout treated the shader group handle size as the record
stride everywhere. This worked only on devices where the handle size
already satisfied all relevant alignment requirements. It broke down once
multiple ray generation shaders were placed in the same SBT: selecting a
non-zero raygen index advanced the raygen region address by the raw handle
size, which could produce an address that was not aligned to the required
shader group base alignment.
Separate the SBT layout into the appropriate strides:
- ray generation records use a base-aligned stride, since selecting a
raygen shader changes the SBT region start address for the dispatch.
- miss and hit records use the handle-aligned shader group stride, since
their regions start at a base-aligned address and individual records are
selected through the region stride.
This keeps the CPU-side SBT population and the dispatch-time SBT regions
using the same layout, so shader group handles are written to the same
aligned slots that Vulkan later addresses.
This will allow decoupling `display_server.h` from a number of headers in the
codebase which only require those enums and not all the DisplayServer API.
- Vulkan implementations in `RenderingDeviceDriverVulkan`.
- Raytracing instruction list in `RenderingDeviceGraph`.
- Functions to create acceleration structures and raytracing pipelines
in `RenderingDevice`.
- Raygen, Miss, and ClosestHit shader stages support.
- GDScript bindings.
- Update classes documentation.
- Unimplemented placeholders for Metal and D3D12.
- Build acceleration structure command.
- Expose a shader preprocessor define.
- Align build scratch address.
- Create STB after creating the pipeline.
- Separate acceleration structure barriers.
- Use transforms in TLAS instances.
- AnyHit and Intersection stages.
- Optionally set acceleration structure build input buffer usage.
- Introduce instances buffer.
- Move scratch buffers to RenderingDevice.
- Rename AccelerationStructureGeometryBits.
- Use regular buffer creation and free routines for scratch buffer.
- Store trackers in acceleration structures.
- Bump Shader SPIR-V version to 1.4
- Add Position attribute location in blas_create.
- Encapsulate MoltenVK check in preprocessor macro.
- Split SUPPORTS_RAYTRACING for pipeline and query.
- Add support for vertex bindings and UMA vertex buffers in D3D12.
- Simplify 2D instance params and move more into per-batch data to save
bandwidth
Co-authored-by: Skyth <19259897+blueskythlikesclouds@users.noreply.github.com>
Co-authored-by: Clay John <claynjohn@gmail.com>
Co-authored-by: A Thousand Ships <96648715+athousandships@users.noreply.github.com>
This work is a heavily refactored and rewritten from TheForge's initial
code.
TheForge's original code had too many race conditions and was
fundamentally flawed as it was too easy to incur into those data races
by accident.
However they identified the proper places that needed changes, and the
idea was sound. I used their work as a blueprint to design this work.
This PR implements:
- Introduction of UMA buffers used by a few buffers
(most notably the ones filled by _fill_instance_data).
Ironically this change seems to positively affect PC more than it does
on Mobile.
Updates D3D12 Memory Allocator to get GPU_UPLOAD heap support.
Metal implementation by Stuart Carnie.
Co-authored-by: Stuart Carnie <stuart.carnie@gmail.com>
Co-authored-by: TheForge team
Introduce a specialised `texture_get_data` for `RenderDeviceDriver`,
which can retrieve the texture data from the GPU driver for shared
textures (`TEXTURE_USAGE_CPU_READ_BIT`).
Closes#108115
