This adds HLSL backend support for `ImageClass::External` (ie WGSL's `external_texture` texture type). For each external texture global variable in the IR, we declare 3 `Texture2D` globals as well as a `cbuffer` for the params. The additional bindings required by these are found in the newly added `external_texture_binding_map`. Unique names for each can be obtained using `NameKey::ExternalTextureGlobalVariable`. For functions that contain ImageQuery::Size, ImageLoad, or ImageSample expressions for external textures, ensure we have generated wrapper functions for those expressions. When emitting code for the expressions themselves, simply insert a call to the wrapper function. For size queries, we return the value provided in the params struct. If that value is [0, 0] then we query the size of the plane 0 texture and return that. For load and sample, we sample the textures based on the number of planes specified in the params struct. If there is more than one plane we additionally perform YUV to RGB conversion using the provided matrix. Unfortunately HLSL does not allow structs to contain textures, meaning we are unable to wrap the 3 textures and params struct variables in a single variable that can be passed around. For our wrapper functions we therefore ensure they take the three textures and the params as consecutive arguments. Likewise, when declaring user-defined functions with external texture arguments, we expand the single external texture argument into 4 consecutive arguments. (Using NameKey::ExternalTextureFunctionArgument to ensure unique names for each.) Thankfully external textures can only be used as either global variables or function arguments. This means we only have to handle the `Expression::GlobalVariable` and `Expression::FunctionArgument` cases of `write_expr()`. Since in both cases we know the external texture can only be an argument to either a user-defined function or one of our wrapper functions, we can simply emit the names of the variables for each three textures and the params struct in a comma-separated list.
wgpu_hal: a cross-platform unsafe graphics abstraction
This crate defines a set of traits abstracting over modern graphics APIs, with implementations ("backends") for Vulkan, Metal, Direct3D, and GL.
wgpu_hal is a spiritual successor to
gfx-hal, but with reduced scope, and
oriented towards WebGPU implementation goals. It has no overhead for
validation or tracking, and the API translation overhead is kept to the bare
minimum by the design of WebGPU. This API can be used for resource-demanding
applications and engines.
The wgpu_hal crate's main design choices:
-
Our traits are meant to be portable: proper use should get equivalent results regardless of the backend.
-
Our traits' contracts are unsafe: implementations perform minimal validation, if any, and incorrect use will often cause undefined behavior. This allows us to minimize the overhead we impose over the underlying graphics system. If you need safety, the
wgpu-corecrate provides a safe API for drivingwgpu_hal, implementing all necessary validation, resource state tracking, and so on. (Note thatwgpu-coreis designed for use via FFI; thewgpucrate provides more idiomatic Rust bindings forwgpu-core.) Or, you can do your own validation. -
In the same vein, returned errors only cover cases the user can't anticipate, like running out of memory or losing the device. Any errors that the user could reasonably anticipate are their responsibility to avoid. For example,
wgpu_halreturns no error for mapping a buffer that's not mappable: as the buffer creator, the user should already know if they can map it. -
We use static dispatch. The traits are not generally object-safe. You must select a specific backend type like
vulkan::Apiormetal::Api, and then use that according to the main traits, or call backend-specific methods. -
We use idiomatic Rust parameter passing, taking objects by reference, returning them by value, and so on, unlike
wgpu-core, which refers to objects by ID. -
We map buffer contents persistently. This means that the buffer can remain mapped on the CPU while the GPU reads or writes to it. You must explicitly indicate when data might need to be transferred between CPU and GPU, if
wgpu_halindicates that the mapping is not coherent (that is, automatically synchronized between the two devices). -
You must record explicit barriers between different usages of a resource. For example, if a buffer is written to by a compute shader, and then used as and index buffer to a draw call, you must use [
CommandEncoder::transition_buffers] between those two operations. -
Pipeline layouts are explicitly specified when setting bind group. Incompatible layouts disturb groups bound at higher indices.
-
The API accepts collections as iterators, to avoid forcing the user to store data in particular containers. The implementation doesn't guarantee that any of the iterators are drained, unless stated otherwise by the function documentation. For this reason, we recommend that iterators don't do any mutating work.
Unfortunately, wgpu_hal's safety requirements are not fully documented.
Ideally, all trait methods would have doc comments setting out the
requirements users must meet to ensure correct and portable behavior. If you
are aware of a specific requirement that a backend imposes that is not
ensured by the traits' documented rules, please file an issue. Or, if you are
a capable technical writer, please file a pull request!
Primary backends
The wgpu_hal crate has full-featured backends implemented on the following
platform graphics APIs:
-
Vulkan, available on Linux, Android, and Windows, using the
ashcrate's Vulkan bindings. It's also available on macOS, if you install MoltenVK. -
Metal on macOS, using the
metalcrate's bindings. -
Direct3D 12 on Windows, using the
windowscrate's bindings.
Secondary backends
The wgpu_hal crate has a partial implementation based on the following
platform graphics API:
- The GL backend is available anywhere OpenGL, OpenGL ES, or WebGL are
available. See the
glesmodule documentation for details.
You can see what capabilities an adapter is missing by checking the
DownlevelCapabilities in [ExposedAdapter::capabilities], available
from [Instance::enumerate_adapters].
The API is generally designed to fit the primary backends better than the secondary backends, so the latter may impose more overhead.
Debugging
Most of the information on the wiki Debugging wgpu Applications
page still applies to this API, with the exception of API tracing/replay
functionality, which is only available in wgpu-core.