June 20, 2020
In this second post of the series we’ll learn about Bind Groups, which let us pass buffers and textures to our shaders. When writing a renderer, we typically have inputs which do not make sense as vertex attributes (e.g., transform matrices, material parameters), or simply cannot be passed as vertex attributes (e.g., textures). Such parameters are instead passed as uniforms in GLSL terms, or root parameters in HLSL terms. The application then associates the desired buffers and textures with the parameters in the shader. In WebGPU, the association of data to parameters is made using Bind Groups. In this post, we’ll use Bind Groups to pass a uniform buffer containing a view transform to our vertex shader, allowing us to add camera controls to our triangle from the previous post. If you haven’t read the first post in this series I recommend reading that first, as we’ll continue directly off the code written there.
This post has been updated to reflect changes to the WebGPU API! Please see the updated post.
At a high level, bind groups follow a similar model to vertex buffers in WebGPU. Each bind group specifies an array of buffers and textures which it contains, and the parameter binding indices to map these too in the shader. Each pipeline specifies that it will use zero or more such bind groups. During rendering, the bind groups required by the pipeline are bound to the corresponding bind group slots specified when creating the pipeline layout, to bind data to the shader parameters. The bind group layout and bind groups using the layout are treated as separate objects, allowing parameter values to be changed without changing the entire rendering pipeline. By using multiple bind group sets, we can swap out per-object parameters without conflicting with bind groups specifying global parameters during rendering. The bind group parameters can be accessible in both the vertex and fragment stages (or, compute). An example pipeline using two bind group sets is illustrated in the figure below.
To add camera controls to our triangle from before, we’ll pass a combined projection and view matrix through to our vertex shader and use it to transform the triangle’s vertices. The transform matrix will be stored in a uniform buffer, read by the shader. Uniform buffers are used to pass small to medium size buffers of constant data to shaders, WebGPU also supports storage buffers, which can be used to pass larger or shader writeable buffers.
In GLSL, the different input slots to which bind groups are bound are referred
to as sets. Each bind group bound to a set contains some number of parameter bindings,
which are mapped to the corresponding parameters in the shader.
Uniform shader parameters in GLSL are annotated with the corresponding set and binding which will
provide the underlying data for the parameter, using the
Here we’ll have a single set, set 0, which binds the uniform buffer containing our
transform matrix to binding 0.
The updated vertex shader that applies the transform passed through the uniform
buffer to our vertices is shown below.
The layouts of the bind groups used by a pipeline are part of the pipeline’s layout.
As discussed last time, by fixing the
configuration of inputs to the pipeline, the GPU can better optimize execution of
The layouts of the bind groups which will be used in the pipeline
are specified through an array of
objects which are passed to
The index of a bind group layout in the pipeline layout’s
is its set number in GLSL. The actual data associated with the binding points
of each layout is specified separately, allowing re-use of the pipeline and
layout with different inputs.
The bind group and pipeline layout which we’ll use in this post is illustrated below.
The bind group layout is created through
which takes a list of entries that the corresponding bind group used during rendering must contain.
Each entry specifies the type
of the parameter, which shader stages will be able to
access it, and the binding number to place the parameter at in the shader.
After creating the bind group layout we use it to create our pipeline layout,
specifying that pipelines created using the layout (i.e., our rendering pipeline)
will use bind groups matching the bind group layouts specified.
We’ll create a uniform buffer to store our view transform, which will be used as a uniform buffer (in our bind group), and a copy destination buffer. The latter usage will allow us to update the buffer’s data each frame by copying in a new view transform.
GPUBindGroup specifies the
actual buffers or textures which will be passed to the shaders. During
rendering, each bind group is bound to the index of its corresponding layout in
The bind group using our bind group layout passes our view parameter buffer at binding 0, matching the layout.
For bind groups with multiple entries, the order of the entries specified in the
layout and bind group do not
need to match; however, for each binding index specified in the layout,
a matching binding must be specified in the bind group.
Next, we need a camera controller to update the view transform based on the user’s input. I’ll be using my implementation of the Shoemake Arcball Camera, which you can get from my webgl-utils library. The library uses gl-matrix as its math library, which you’ll need to download separately. WebGL Utils also provides a simple controller object which can be bound to the canvas to more easily apply user inputs to the camera. If you have your own preferred camera and controller code, you can also use that, all we’ll need in the end is to update the view transform based on the user’s inputs.
With everything set up, all that’s left to do each frame is upload
the new view transform, set our bind group, and render!
After computing the combined projection and view matrix using
the updated view transform (stored in
camera.camera), we create a
mapped staging buffer to upload the transform to the GPU.
In WebGPU, the
MAP_WRITE usage modes cannot be combined
with any modes other than
and thus we cannot map back our view parameters buffer to write to it directly.
For simplicity, each frame we’ll just create a new mapped buffer to
upload our view transform. We set this buffer’s usage as
meaning that we’ll use it to copy data from it into another buffer,
i.e., into our uniform buffer.
We then copy the uploaded view matrix into our uniform buffer
by enqueuing a
command on the command encoder before beginning our rendering pass.
In the rendering pass, we set the bind group to be used before rendering our triangle
on the render pass, specifying the bind group set number to place our bind group at.
After the command buffer is run we’ll have our triangle rendered with our
view transform applied, as shown below!
Now that we’ve learned about bind groups, we’ll be able to use them to pass view and object transform matrices, material parameters, and textures, to our shaders. In the next post, we’ll take a look at the glTF Binary format to load and render our first real triangle mesh. If you run into issues getting the example to work, check out the code for rendering the triangle in Figure 3, or get in touch via Twitter or email.
device.createBuffer with the
mappedAtCreation parameter set.
See the corresponding Dawn PSA about these changes in Chrome and WebGPU.