Veldrid supports compute shaders, which are a unique type of shader that can perform highly-parallel computations and output arbitrary types of data. Although compute shaders are generally used to produce large amounts of data to be fed into subsequent rendering operations, it is also possible to use compute shaders as a tool for general-purpose parallel computation.
Compute shaders execute in parallel, 3-dimensional thread groups which are dispatched from a CommandList (see CommandList.Dispatch(uint, uint, uint). The number of threads in this work group is determined by the X, Y, and Z counts provided to CommandList.Dispatch, and by the work group size defined in the compute shader code's size attribute and in the ComputePipelineDescription.
Inputs and Outputs
Unlike shaders in a graphics Pipeline, a compute shader has no "vertex" inputs, and no Framebuffer outputs. All data in a compute shader must be read from bound shader resources such as DeviceBuffers and Textures. To output data, compute shaders make use of read-write buffers and textures.
The state tracked in a CommandList is separate for graphics and compute Pipelines. Setting a compute Pipeline does not alter the currently-bound graphics Pipeline, and vice-versa. Setting compute ResourceSets does not alter any of the currently-bound graphics ResourceSets, and vice-versa.