d5ff7138d95488b0dce44dfe1c60a426f6a37b51,src/graph_transpiler/webdnn/backend/webgpu/kernels/flatten.py,,flatten,#Any#Any#Any#,28

Before Change

    // assert x.variable.order == y.variable.order

    meta_injector = MetaInjector()
    meta_injector.register({
        "flatten_x_offset": x.offset,
        "flatten_y_offset": y.offset,
        "flatten_N": y.variable.size,
    })

    name_injector = KernelNameInjector(op)

    source = template
    source = meta_injector.inject(source)
    source = name_injector.inject(source)

    kernel = Kernel(
        {name_injector.name: source},
        name_injector.name,
        GPUSize(8, 1, 1),
        GPUSize(1024, 1, 1),
        meta_injector.buffer
    )

    return [kernel]

After Change

    const int N = %%META_LOAD(flatten_N)%%;

    for (int gid = index; gid < N; gid += num_threads) {
        y[gid] = x[gid];
    }
}



def flatten(op: Flatten,
            memory_layout: MemoryLayout) -> List[Kernel]:
    x = memory_layout[op.inputs["x"]]
    y = memory_layout[op.outputs["y"]]

    // assert x.variable.order == y.variable.order

    meta_injector = MetaInjector()
    meta_injector.register({
        "flatten_x_offset": x.offset,
        "flatten_y_offset": y.offset,
        "flatten_N": y.variable.size,
    })

    name_injector = KernelNameInjector(op)

    source = template
    source = meta_injector.inject(source)
    source = name_injector.inject(source)

    kernel = Kernel(

Instances