A high-performance inference engine for AI models
use std::convert::Infallible;
use crate::backends::{
common::{
Backend,
allocator::{
bump::BumpAllocation,
pool::{PoolAllocation, PoolAllocator},
},
},
webgpu::{
buffer::{
dense::{WebGPUDenseBufferCpuGpu, WebGPUDenseBufferGpu},
sparse::WebGPUSparseBuffer,
},
command_buffer::{WebGPUCommandBuffer, WebGPUCommandBufferEncoding},
context::WebGPUContext,
error::WebGPUError,
kernel::WebGPUKernels,
},
};
#[derive(Debug, Clone)]
pub struct WebGPU;
impl Backend for WebGPU {
type Context = WebGPUContext;
type CommandBuffer = WebGPUCommandBuffer;
type DebugGroupSpan<'a> = &'a mut WebGPUCommandBufferEncoding;
type GlobalGpuBuffer = WebGPUDenseBufferGpu;
type GlobalBufferCpuGpu = WebGPUDenseBufferCpuGpu;
type ConstantBuffer = BumpAllocation<WebGPUDenseBufferCpuGpu>;
type ScratchBuffer = PoolAllocation<WebGPUDenseBufferGpu>;
type AllocationPool = PoolAllocator<WebGPUDenseBufferGpu>;
type CaptureSpan = Infallible;
type SparseBuffer = WebGPUSparseBuffer;
type Kernels = WebGPUKernels;
type Error = WebGPUError;
const NAME: &'static str = "webgpu";
}