[INFRA] Import NVIDIA/CCCL upstream as optimization reference library

CCCL (CUDA C++ Core Libraries) provides:
- CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk)
- Thrust: high-level parallel algorithms (transform_reduce, sort, scan)
- libcudacxx: CUDA C++ standard library (atomics, barriers, memory)
- cudax: experimental features (memory resources, allocators)
- Tuning policies: per-SM hardware-specific algorithm parameters

Competition optimization vectors mapped to CCCL:
- Output TPS (83% weight): warp_reduce, block_reduce, device_topk
- Input TPS (14% weight): device_scan, block_load, prefetch
- Cache TPS (3% weight): prefix caching strategy patterns
- Memory (0.9 util): pooled/cached/buddy allocators

Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only)
License: Apache-2.0

cccl_upstream/thrust/testing/cuda/set_intersection.cu

@@ -0,0 +1,86 @@
+#include <thrust/execution_policy.h>
+#include <thrust/set_operations.h>
+
+#include <cuda/buffer>
+#include <cuda/cccl_runtime_test_helper.cuh>
+#include <cuda/launch>
+#include <cuda/stream>
+
+#include <unittest/unittest.h>
+
+#ifdef THRUST_TEST_DEVICE_SIDE
+struct set_intersection_kernel
+{
+  template <typename ExecutionPolicy, typename Input1, typename Input2, typename Output>
+  __device__ void operator()(ExecutionPolicy exec, Input1 a, Input2 b, Output result) const
+  {
+    const auto end = thrust::set_intersection(exec, a.begin(), a.end(), b.begin(), b.end(), result.begin());
+    TEST_ASSERT_DEVICE(end == result.end());
+  }
+};
+
+template <typename ExecutionPolicy>
+void TestSetIntersectionDevice(ExecutionPolicy exec)
+{
+  const auto device = test_runtime::current_test_device();
+  cuda::stream stream{device};
+
+  auto a      = cuda::make_device_buffer<int>(stream, device, {0, 2, 4});
+  auto b      = cuda::make_device_buffer<int>(stream, device, {0, 3, 3, 4});
+  auto result = cuda::make_device_buffer<int>(stream, device, 2, cuda::no_init);
+
+  cuda::launch(stream, test_runtime::single_thread_config(), set_intersection_kernel{}, exec, a, b, result);
+  stream.sync();
+
+  test_runtime::assert_equal(stream, result, {0, 4});
+}
+
+void TestSetIntersectionDeviceSeq()
+{
+  TestSetIntersectionDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestSetIntersectionDeviceSeq);
+
+void TestSetIntersectionDeviceDevice()
+{
+  TestSetIntersectionDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestSetIntersectionDeviceDevice);
+
+void TestSetIntersectionDeviceNoSync()
+{
+  TestSetIntersectionDevice(thrust::cuda::par_nosync);
+}
+DECLARE_UNITTEST(TestSetIntersectionDeviceNoSync);
+#endif
+
+template <typename ExecutionPolicy>
+void TestSetIntersectionCudaStreams(ExecutionPolicy policy)
+{
+  const auto device = test_runtime::current_test_device();
+  cuda::stream stream{device};
+
+  auto a      = cuda::make_device_buffer<int>(stream, device, {0, 2, 4});
+  auto b      = cuda::make_device_buffer<int>(stream, device, {0, 3, 3, 4});
+  auto result = cuda::make_device_buffer<int>(stream, device, 2, cuda::no_init);
+
+  const auto streampolicy = policy.on(stream.get());
+
+  const auto end = thrust::set_intersection(streampolicy, a.begin(), a.end(), b.begin(), b.end(), result.begin());
+  stream.sync();
+
+  ASSERT_EQUAL_QUIET(result.end(), end);
+  test_runtime::assert_equal(stream, result, {0, 4});
+}
+
+void TestSetIntersectionCudaStreamsSync()
+{
+  TestSetIntersectionCudaStreams(thrust::cuda::par);
+}
+DECLARE_UNITTEST(TestSetIntersectionCudaStreamsSync);
+
+void TestSetIntersectionCudaStreamsNoSync()
+{
+  TestSetIntersectionCudaStreams(thrust::cuda::par_nosync);
+}
+DECLARE_UNITTEST(TestSetIntersectionCudaStreamsNoSync);