runmat-accelerate 0.4.5

Pluggable GPU acceleration layer for RunMat (CUDA, ROCm, Metal, Vulkan/Spir-V)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

use super::common::submit;

pub fn run(
    device: &wgpu::Device,
    queue: &wgpu::Queue,
    pipeline: &wgpu::ComputePipeline,
    bind_group: &wgpu::BindGroup,
    groups_x: u32,
    groups_y: u32,
) {
    let mut enc = device.create_command_encoder(&wgpu::CommandEncoderDescriptor {
        label: Some("runmat-matmul-encoder"),
    });
    {
        let mut pass = enc.begin_compute_pass(&wgpu::ComputePassDescriptor {
            label: Some("runmat-matmul-pass"),
            timestamp_writes: None,
        });
        pass.set_pipeline(pipeline);
        pass.set_bind_group(0, bind_group, &[]);
        pass.dispatch_workgroups(groups_x, groups_y, 1);
    }
    submit(device, queue, enc);
}