[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/merge.cu

@@ -0,0 +1,92 @@
+#include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
+#include <thrust/merge.h>
+#include <thrust/sort.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 merge_kernel
+{
+  template <typename ExecutionPolicy, typename Input1, typename Input2, typename Size, typename Output>
+  __device__ void operator()(ExecutionPolicy exec, Input1 a, Input2 b, Size b_size, Output result) const
+  {
+    const auto end = thrust::merge(exec, a.begin(), a.end(), b.begin(), b.begin() + b_size, result.begin());
+    TEST_ASSERT_DEVICE(end == result.end());
+  }
+};
+
+template <typename ExecutionPolicy>
+void TestMergeDevice(ExecutionPolicy exec)
+{
+  const auto device = test_runtime::current_test_device();
+  cuda::stream stream{device};
+
+  const size_t n         = 10000;
+  const size_t sizes[]   = {0, 1, n / 2, n, n + 1, 2 * n};
+  const size_t num_sizes = sizeof(sizes) / sizeof(size_t);
+
+  const auto max_size = static_cast<size_t>(*thrust::max_element(sizes, sizes + num_sizes));
+  auto h_a            = test_runtime::random_integers_buffer<int, unittest::int8_t>(stream, n);
+  auto h_b            = test_runtime::random_integers_buffer<int, unittest::int8_t>(stream, max_size, n);
+
+  thrust::stable_sort(h_a.begin(), h_a.end());
+  thrust::stable_sort(h_b.begin(), h_b.end());
+
+  auto d_a = cuda::make_device_buffer<int>(stream, device, h_a);
+  auto d_b = cuda::make_device_buffer<int>(stream, device, h_b);
+
+  for (size_t i = 0; i < num_sizes; i++)
+  {
+    const size_t size = sizes[i];
+
+    auto h_result = test_runtime::make_host_buffer<int>(stream, n + size);
+    stream.sync();
+
+    const auto h_end = thrust::merge(h_a.begin(), h_a.end(), h_b.begin(), h_b.begin() + size, h_result.begin());
+    ASSERT_EQUAL_QUIET(h_result.end(), h_end);
+
+    auto result = cuda::make_device_buffer<int>(stream, device, h_result.size(), cuda::no_init);
+
+    cuda::launch(stream, test_runtime::single_thread_config(), merge_kernel{}, exec, d_a, d_b, size, result);
+    stream.sync();
+
+    test_runtime::assert_equal(stream, result, h_result);
+  }
+}
+
+void TestMergeDeviceSeq()
+{
+  TestMergeDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestMergeDeviceSeq);
+
+void TestMergeDeviceDevice()
+{
+  TestMergeDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestMergeDeviceDevice);
+#endif
+
+void TestMergeCudaStreams()
+{
+  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, 7, cuda::no_init);
+
+  const auto end =
+    thrust::merge(thrust::cuda::par.on(stream.get()), 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, 0, 2, 3, 3, 4, 4});
+}
+DECLARE_UNITTEST(TestMergeCudaStreams);