[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/scatter.cu

@@ -0,0 +1,178 @@
+#include <thrust/execution_policy.h>
+#include <thrust/scatter.h>
+
+#include <algorithm>
+
+#include <unittest/unittest.h>
+
+#ifdef THRUST_TEST_DEVICE_SIDE
+template <typename ExecutionPolicy, typename Iterator1, typename Iterator2, typename Iterator3>
+__global__ void
+scatter_kernel(ExecutionPolicy exec, Iterator1 first, Iterator1 last, Iterator2 map_first, Iterator3 result)
+{
+  thrust::scatter(exec, first, last, map_first, result);
+}
+
+template <typename ExecutionPolicy>
+void TestScatterDevice(ExecutionPolicy exec)
+{
+  size_t n                 = 1000;
+  const size_t output_size = std::min((size_t) 10, 2 * n);
+
+  thrust::host_vector<int> h_input(n, 1);
+  thrust::device_vector<int> d_input(n, 1);
+
+  thrust::host_vector<unsigned int> h_map = unittest::random_integers<unsigned int>(n);
+
+  for (size_t i = 0; i < n; i++)
+  {
+    h_map[i] = h_map[i] % output_size;
+  }
+
+  thrust::device_vector<unsigned int> d_map = h_map;
+
+  thrust::host_vector<int> h_output(output_size, 0);
+  thrust::device_vector<int> d_output(output_size, 0);
+
+  thrust::scatter(h_input.begin(), h_input.end(), h_map.begin(), h_output.begin());
+
+  scatter_kernel<<<1, 1>>>(exec, d_input.begin(), d_input.end(), d_map.begin(), d_output.begin());
+  cudaError_t const err = cudaDeviceSynchronize();
+  ASSERT_EQUAL(cudaSuccess, err);
+
+  ASSERT_EQUAL(h_output, d_output);
+}
+
+void TestScatterDeviceSeq()
+{
+  TestScatterDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestScatterDeviceSeq);
+
+void TestScatterDeviceDevice()
+{
+  TestScatterDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestScatterDeviceDevice);
+
+template <typename ExecutionPolicy,
+          typename Iterator1,
+          typename Iterator2,
+          typename Iterator3,
+          typename Iterator4,
+          typename Function>
+__global__ void scatter_if_kernel(
+  ExecutionPolicy exec,
+  Iterator1 first,
+  Iterator1 last,
+  Iterator2 map_first,
+  Iterator3 stencil_first,
+  Iterator4 result,
+  Function f)
+{
+  thrust::scatter_if(exec, first, last, map_first, stencil_first, result, f);
+}
+
+template <typename T>
+struct is_even_scatter_if
+{
+  _CCCL_HOST_DEVICE bool operator()(const T i) const
+  {
+    return (i % 2) == 0;
+  }
+};
+
+template <typename ExecutionPolicy>
+void TestScatterIfDevice(ExecutionPolicy exec)
+{
+  size_t n                 = 1000;
+  const size_t output_size = std::min((size_t) 10, 2 * n);
+
+  thrust::host_vector<int> h_input(n, 1);
+  thrust::device_vector<int> d_input(n, 1);
+
+  thrust::host_vector<unsigned int> h_map = unittest::random_integers<unsigned int>(n);
+
+  for (size_t i = 0; i < n; i++)
+  {
+    h_map[i] = h_map[i] % output_size;
+  }
+
+  thrust::device_vector<unsigned int> d_map = h_map;
+
+  thrust::host_vector<int> h_output(output_size, 0);
+  thrust::device_vector<int> d_output(output_size, 0);
+
+  thrust::scatter_if(
+    h_input.begin(), h_input.end(), h_map.begin(), h_map.begin(), h_output.begin(), is_even_scatter_if<unsigned int>());
+
+  scatter_if_kernel<<<1, 1>>>(
+    exec,
+    d_input.begin(),
+    d_input.end(),
+    d_map.begin(),
+    d_map.begin(),
+    d_output.begin(),
+    is_even_scatter_if<unsigned int>());
+  cudaError_t const err = cudaDeviceSynchronize();
+  ASSERT_EQUAL(cudaSuccess, err);
+
+  ASSERT_EQUAL(h_output, d_output);
+}
+
+void TestScatterIfDeviceSeq()
+{
+  TestScatterIfDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestScatterIfDeviceSeq);
+
+void TestScatterIfDeviceDevice()
+{
+  TestScatterIfDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestScatterIfDeviceDevice);
+#endif
+
+void TestScatterCudaStreams()
+{
+  using Vector = thrust::device_vector<int>;
+
+  Vector map{6, 3, 1, 7, 2}; // scatter indices
+  Vector src{0, 1, 2, 3, 4}; // source vector
+  Vector dst(8, 0); // destination vector
+
+  cudaStream_t s;
+  cudaStreamCreate(&s);
+
+  thrust::scatter(thrust::cuda::par.on(s), src.begin(), src.end(), map.begin(), dst.begin());
+
+  cudaStreamSynchronize(s);
+
+  Vector ref{0, 2, 4, 1, 0, 0, 0, 3};
+  ASSERT_EQUAL(dst, ref);
+
+  cudaStreamDestroy(s);
+}
+DECLARE_UNITTEST(TestScatterCudaStreams);
+
+void TestScatterIfCudaStreams()
+{
+  using Vector = thrust::device_vector<int>;
+
+  Vector flg{0, 1, 0, 1, 0}; // predicate array
+  Vector map{6, 3, 1, 7, 2}; // scatter indices
+  Vector src{0, 1, 2, 3, 4}; // source vector
+  Vector dst(8); // destination vector
+
+  cudaStream_t s;
+  cudaStreamCreate(&s);
+
+  thrust::scatter_if(thrust::cuda::par.on(s), src.begin(), src.end(), map.begin(), flg.begin(), dst.begin());
+  cudaStreamSynchronize(s);
+
+  Vector ref{0, 0, 0, 1, 0, 0, 0, 3};
+  ASSERT_EQUAL(dst, ref);
+
+  cudaStreamDestroy(s);
+}
+DECLARE_UNITTEST(TestScatterIfCudaStreams);