Function test_gather_other

src/core/test/mem_alloc.cpp:573–591

}

void test_gather_other(CompNode cn0, CompNode cn1) {
    if (cn0.get_mem_status_bytes().second > cn1.get_mem_status_bytes().second) {
        std::swap(cn0, cn1);
    }
    size_t elems = cn0.get_mem_status_bytes().second * 2 / 5 / sizeof(dt_int32);
    auto xv = std::make_shared<DeviceTensorND>(cn0, TensorShape{elems}, dtype::Int32());
    auto graph = ComputingGraph::make();
    auto x = opr::SharedDeviceTensor::make(*graph, xv), x1 = x + 1,
         x2 = opr::MarkDynamicVar::make(x), y = opr::Copy::make(x1, {cn1});
    // x1 must be released (which requires y to finish) before x2 succeeds

    set_priority(x1, -10);
    set_priority(y, -10);
    graph->options().var_sanity_check_first_run = false;
    graph->options().async_exec_level = 0;
    auto func = graph->compile({{x2, {}}, {y, {}}});
    opr::Sleep::sleep(cn1, 0.7);
    func->execute();
}
}  // namespace

#if MGB_CUDA