add unit test

Signed-off-by: 400Ping <jiekaichang@apache.org>

Author: 400Ping

qdp/qdp-core/tests/dlpack.rs

@@ -18,7 +18,10 @@
 
 #[cfg(test)]
 mod dlpack_tests {
+    use std::ffi::c_void;
+
     use cudarc::driver::CudaDevice;
+    use qdp_core::dlpack::{synchronize_stream, CUDA_STREAM_LEGACY};
     use qdp_core::gpu::memory::GpuStateVector;
 
     #[test]
@@ -82,4 +85,29 @@ mod dlpack_tests {
             }
         }
     }
+
+    /// synchronize_stream(null) is a no-op and returns Ok(()) on all platforms.
+    #[test]
+    fn test_synchronize_stream_null() {
+        unsafe {
+            let result = synchronize_stream(std::ptr::null_mut::<c_void>());
+            assert!(
+                result.is_ok(),
+                "synchronize_stream(null) should return Ok(())"
+            );
+        }
+    }
+
+    /// synchronize_stream(CUDA_STREAM_LEGACY) syncs the legacy default stream (Linux + CUDA).
+    #[test]
+    #[cfg(target_os = "linux")]
+    fn test_synchronize_stream_legacy() {
+        unsafe {
+            let result = synchronize_stream(CUDA_STREAM_LEGACY);
+            assert!(
+                result.is_ok(),
+                "synchronize_stream(CUDA_STREAM_LEGACY) should succeed on Linux with CUDA"
+            );
+        }
+    }
 }

testing/qdp/test_bindings.py

@@ -125,6 +125,25 @@ def test_dlpack_single_use():
         qtensor2.__dlpack__()
 
 
+@requires_qdp
+@pytest.mark.gpu
+@pytest.mark.parametrize("stream", [1, 2], ids=["stream_legacy", "stream_per_thread"])
+def test_dlpack_with_stream(stream):
+    """Test __dlpack__(stream=...) syncs CUDA stream before returning capsule (DLPack 0.8+)."""
+    import torch
+    from _qdp import QdpEngine
+
+    engine = QdpEngine(0)
+    data = [1.0, 2.0, 3.0, 4.0]
+    qtensor = engine.encode(data, 2, "amplitude")
+
+    # stream=1 (legacy default) or 2 (per-thread default) should sync and return capsule
+    capsule = qtensor.__dlpack__(stream=stream)
+    torch_tensor = torch.from_dlpack(capsule)
+    assert torch_tensor.is_cuda
+    assert torch_tensor.shape == (1, 4)
+
+
 @requires_qdp
 @pytest.mark.gpu
 def test_pytorch_integration():