[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/examples/minmax.cu

@@ -0,0 +1,91 @@
+#include <thrust/device_vector.h>
+#include <thrust/extrema.h>
+#include <thrust/functional.h>
+#include <thrust/host_vector.h>
+#include <thrust/random.h>
+#include <thrust/transform_reduce.h>
+
+#include <iostream>
+
+// compute minimum and maximum values in a single reduction
+
+// minmax_pair stores the minimum and maximum
+// values that have been encountered so far
+template <typename T>
+struct minmax_pair
+{
+  T min_val;
+  T max_val;
+};
+
+// minmax_unary_op is a functor that takes in a value x and
+// returns a minmax_pair whose minimum and maximum values
+// are initialized to x.
+template <typename T>
+struct minmax_unary_op
+{
+  __host__ __device__ minmax_pair<T> operator()(const T& x) const
+  {
+    minmax_pair<T> result;
+    result.min_val = x;
+    result.max_val = x;
+    return result;
+  }
+};
+
+// minmax_binary_op is a functor that accepts two minmax_pair
+// structs and returns a new minmax_pair whose minimum and
+// maximum values are the min() and max() respectively of
+// the minimums and maximums of the input pairs
+template <typename T>
+struct minmax_binary_op
+{
+  __host__ __device__ minmax_pair<T> operator()(const minmax_pair<T>& x, const minmax_pair<T>& y) const
+  {
+    minmax_pair<T> result;
+    result.min_val = thrust::min(x.min_val, y.min_val);
+    result.max_val = thrust::max(x.max_val, y.max_val);
+    return result;
+  }
+};
+
+int main()
+{
+  // input size
+  size_t N = 10;
+
+  // initialize random number generator
+  thrust::default_random_engine rng;
+  thrust::uniform_int_distribution<int> dist(10, 99);
+
+  // initialize data on host
+  thrust::host_vector<int> host_data(N);
+  for (auto& e : host_data)
+  {
+    e = dist(rng);
+  }
+  thrust::device_vector<int> data = host_data;
+
+  // setup arguments
+  minmax_unary_op<int> unary_op;
+  minmax_binary_op<int> binary_op;
+
+  // initialize reduction with the first value
+  minmax_pair<int> init = unary_op(data[0]);
+
+  // compute minimum and maximum values
+  minmax_pair<int> result = thrust::transform_reduce(data.begin(), data.end(), unary_op, init, binary_op);
+
+  // print results
+  std::cout << "[ ";
+  for (auto& e : host_data)
+  {
+    std::cout << e << " ";
+  }
+  std::cout << "]" << '\n';
+
+  std::cout << "minimum = " << result.min_val << '\n';
+  std::cout << "maximum = " << result.max_val << '\n';
+
+  return 0;
+}