[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/tabulate.cu

@@ -0,0 +1,95 @@
+#include <thrust/execution_policy.h>
+#include <thrust/functional.h>
+#include <thrust/tabulate.h>
+
+#include <unittest/unittest.h>
+
+#ifdef THRUST_TEST_DEVICE_SIDE
+template <typename ExecutionPolicy, typename Iterator, typename Function>
+__global__ void tabulate_kernel(ExecutionPolicy exec, Iterator first, Iterator last, Function f)
+{
+  thrust::tabulate(exec, first, last, f);
+}
+
+template <typename ExecutionPolicy>
+void TestTabulateDevice(ExecutionPolicy exec)
+{
+  using Vector = thrust::device_vector<int>;
+  using namespace thrust::placeholders;
+  using T = typename Vector::value_type;
+
+  Vector v(5);
+
+  tabulate_kernel<<<1, 1>>>(exec, v.begin(), v.end(), ::cuda::std::identity{});
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  Vector ref{0, 1, 2, 3, 4};
+  ASSERT_EQUAL(v, ref);
+
+  tabulate_kernel<<<1, 1>>>(exec, v.begin(), v.end(), -_1);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ref = {0, -1, -2, -3, -4};
+  ASSERT_EQUAL(v, ref);
+
+  tabulate_kernel<<<1, 1>>>(exec, v.begin(), v.end(), _1 * _1 * _1);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ref = {0, 1, 8, 27, 64};
+  ASSERT_EQUAL(v, ref);
+}
+
+void TestTabulateDeviceSeq()
+{
+  TestTabulateDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestTabulateDeviceSeq);
+
+void TestTabulateDeviceDevice()
+{
+  TestTabulateDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestTabulateDeviceDevice);
+#endif
+
+void TestTabulateCudaStreams()
+{
+  using namespace thrust::placeholders;
+  using Vector = thrust::device_vector<int>;
+  using T      = Vector::value_type;
+
+  Vector v(5);
+
+  cudaStream_t s;
+  cudaStreamCreate(&s);
+
+  thrust::tabulate(thrust::cuda::par.on(s), v.begin(), v.end(), ::cuda::std::identity{});
+  cudaStreamSynchronize(s);
+
+  Vector ref{0, 1, 2, 3, 4};
+  ASSERT_EQUAL(v, ref);
+
+  thrust::tabulate(thrust::cuda::par.on(s), v.begin(), v.end(), -_1);
+  cudaStreamSynchronize(s);
+
+  ref = {0, -1, -2, -3, -4};
+  ASSERT_EQUAL(v, ref);
+
+  thrust::tabulate(thrust::cuda::par.on(s), v.begin(), v.end(), _1 * _1 * _1);
+  cudaStreamSynchronize(s);
+
+  ref = {0, 1, 8, 27, 64};
+  ASSERT_EQUAL(v, ref);
+
+  cudaStreamSynchronize(s);
+}
+DECLARE_UNITTEST(TestTabulateCudaStreams);