Method exec

dnn/src/cuda/cumsum/opr_impl.cpp:36–53

}  // anonymous namespace

void CumsumForwardImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_in dst, _megdnn_workspace workspace) {
    check_exec(src.layout, dst.layout, workspace.size);
    size_t A, B, C;
    reduce::get_ABC(src.layout, A, B, C, param().axis);
    auto stream = cuda_stream(handle());
#define cb(DType)                                                                  \
    if (src.layout.dtype == DType()) {                                             \
        using ctype = DTypeTrait<DType>::ctype;                                    \
        dispatch<ctype, SumOp<ctype>>(                                             \
                dst.ptr<ctype>(), workspace.ptr<ctype>(), workspace.size, A, B, C, \
                param().exclusive, param().reverse, src.ptr<ctype>(), stream);     \
        return;                                                                    \
    }
    MEGDNN_FOREACH_COMPUTING_DTYPE(cb)
#undef cb
    megdnn_assert_internal(false);
}

size_t CumsumForwardImpl::get_workspace_in_bytes(
        const TensorLayout& src, const TensorLayout&) {