[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/pair_reduce.cu

@@ -0,0 +1,58 @@
+#include <thrust/reduce.h>
+#include <thrust/transform.h>
+
+#include <cuda/std/utility>
+
+#include <unittest/unittest.h>
+
+struct make_pair_functor
+{
+  template <typename T1, typename T2>
+  _CCCL_HOST_DEVICE cuda::std::pair<T1, T2> operator()(const T1& x, const T2& y)
+  {
+    return cuda::std::make_pair(x, y);
+  } // end operator()()
+}; // end make_pair_functor
+
+struct add_pairs
+{
+  template <typename Pair1, typename Pair2>
+  _CCCL_HOST_DEVICE Pair1 operator()(const Pair1& x, const Pair2& y)
+  {
+    // Need cast to undo integer promotion, decltype(char{} + char{}) == int
+    using P1T1 = typename Pair1::first_type;
+    using P1T2 = typename Pair1::second_type;
+    return cuda::std::make_pair(static_cast<P1T1>(x.first + y.first), static_cast<P1T2>(x.second + y.second));
+  } // end operator()
+}; // end add_pairs
+
+template <typename T>
+struct TestPairReduce
+{
+  void operator()(const size_t n)
+  {
+    using P = cuda::std::pair<T, T>;
+
+    thrust::host_vector<T> h_p1 = unittest::random_integers<T>(n);
+    thrust::host_vector<T> h_p2 = unittest::random_integers<T>(n);
+    thrust::host_vector<P> h_pairs(n);
+
+    // zip up pairs on the host
+    thrust::transform(h_p1.begin(), h_p1.end(), h_p2.begin(), h_pairs.begin(), make_pair_functor());
+
+    thrust::device_vector<T> d_p1    = h_p1;
+    thrust::device_vector<T> d_p2    = h_p2;
+    thrust::device_vector<P> d_pairs = h_pairs;
+
+    P init = cuda::std::make_pair(T{13}, T{13});
+
+    // reduce on the host
+    P h_result = thrust::reduce(h_pairs.begin(), h_pairs.end(), init, add_pairs());
+
+    // reduce on the device
+    P d_result = thrust::reduce(d_pairs.begin(), d_pairs.end(), init, add_pairs());
+
+    ASSERT_EQUAL_QUIET(h_result, d_result);
+  }
+}; // end TestPairReduce
+VariableUnitTest<TestPairReduce, SignedIntegralTypes> TestPairReduceInstance;