[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/mode.cu

@@ -0,0 +1,80 @@
+#include <thrust/device_vector.h>
+#include <thrust/extrema.h>
+#include <thrust/functional.h>
+#include <thrust/host_vector.h>
+#include <thrust/iterator/zip_iterator.h>
+#include <thrust/random.h>
+#include <thrust/reduce.h>
+#include <thrust/sort.h>
+#include <thrust/unique.h>
+
+#include <cuda/iterator>
+
+#include <iostream>
+#include <iterator>
+
+// This example compute the mode [1] of a set of numbers.  If there
+// are multiple modes, one with the smallest value it returned.
+//
+// [1] http://en.wikipedia.org/wiki/Mode_(statistics)
+
+int main()
+{
+  const size_t N = 30;
+  const size_t M = 10;
+  thrust::default_random_engine rng;
+  thrust::uniform_int_distribution<int> dist(0, M - 1);
+
+  // generate random data on the host
+  thrust::host_vector<int> h_data(N);
+  for (auto& e : h_data)
+  {
+    e = dist(rng);
+  }
+
+  // transfer data to device
+  thrust::device_vector<int> d_data(h_data);
+
+  // print the initial data
+  std::cout << "initial data" << '\n';
+  thrust::copy(d_data.begin(), d_data.end(), std::ostream_iterator<int>(std::cout, " "));
+  std::cout << '\n';
+
+  // sort data to bring equal elements together
+  thrust::sort(d_data.begin(), d_data.end());
+
+  // print the sorted data
+  std::cout << "sorted data" << '\n';
+  thrust::copy(d_data.begin(), d_data.end(), std::ostream_iterator<int>(std::cout, " "));
+  std::cout << '\n';
+
+  // count number of unique keys
+  size_t num_unique = thrust::unique_count(d_data.begin(), d_data.end());
+
+  // count multiplicity of each key
+  thrust::device_vector<int> d_output_keys(num_unique);
+  thrust::device_vector<int> d_output_counts(num_unique);
+  thrust::reduce_by_key(
+    d_data.begin(), d_data.end(), cuda::constant_iterator<int>(1), d_output_keys.begin(), d_output_counts.begin());
+
+  // print the counts
+  std::cout << "values" << '\n';
+  thrust::copy(d_output_keys.begin(), d_output_keys.end(), std::ostream_iterator<int>(std::cout, " "));
+  std::cout << '\n';
+
+  // print the counts
+  std::cout << "counts" << '\n';
+  thrust::copy(d_output_counts.begin(), d_output_counts.end(), std::ostream_iterator<int>(std::cout, " "));
+  std::cout << '\n';
+
+  // find the index of the maximum count
+  thrust::device_vector<int>::iterator mode_iter;
+  mode_iter = thrust::max_element(d_output_counts.begin(), d_output_counts.end());
+
+  int mode        = d_output_keys[cuda::std::distance(d_output_counts.begin(), mode_iter)];
+  int occurrences = *mode_iter;
+
+  std::cout << "Modal value " << mode << " occurs " << occurrences << " times " << '\n';
+
+  return 0;
+}