Function test_peer_copy_from_device

src/core/test/comp_node.cpp:434–464

namespace {
void test_peer_copy_from_device(const char* comp_node) {
    REQUIRE_THREAD();
    auto cn_gpu = CompNode::load(comp_node);
    auto cn_cpu = CompNode::load("cpux");

    HostTensorGenerator<> gen;
    auto a = gen({20, 3, 112, 112});
    auto b = gen({20, 3, 112, 112});
    auto c = gen({20, 3, 112, 112});
    DeviceTensorND dev_a{cn_gpu}, dev_b{cn_cpu}, dev_c{cn_gpu};
    dev_a.copy_from(*a).sync();
    dev_b.copy_from(*b).sync();
    dev_c.copy_from(*c).sync();

    auto wait_event = cn_gpu.create_event();

    opr::Sleep::sleep(cn_gpu, 0.1);
    dev_a.copy_from(dev_c);
    wait_event->record();
    cn_cpu.device_wait_event(*wait_event);
    dev_b.copy_from(dev_a);

    dev_b.sync();

    HostTensorND result;
    result.copy_from(dev_b);

    CompNode::sync_all();

    MGB_ASSERT_TENSOR_EQ(result, *c);
}
}  // namespace

TEST(TestCompNodeCPU, PeerCopyFromCUDA) {