[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/uninitialized_fill.cu

@@ -0,0 +1,205 @@
+#include <thrust/execution_policy.h>
+#include <thrust/uninitialized_fill.h>
+
+#include <unittest/unittest.h>
+
+#ifdef THRUST_TEST_DEVICE_SIDE
+template <typename ExecutionPolicy, typename Iterator, typename T>
+__global__ void uninitialized_fill_kernel(ExecutionPolicy exec, Iterator first, Iterator last, T val)
+{
+  thrust::uninitialized_fill(exec, first, last, val);
+}
+
+template <typename ExecutionPolicy>
+void TestUninitializedFillDevice(ExecutionPolicy exec)
+{
+  using Vector = thrust::device_vector<int>;
+  using T      = Vector::value_type;
+
+  Vector v{0, 1, 2, 3, 4};
+  T sub(7);
+
+  uninitialized_fill_kernel<<<1, 1>>>(exec, v.begin() + 1, v.begin() + 4, sub);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  Vector ref{0, sub, sub, sub, 4};
+  ASSERT_EQUAL(v, ref);
+
+  sub = 8;
+
+  uninitialized_fill_kernel<<<1, 1>>>(exec, v.begin() + 0, v.begin() + 3, sub);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ref = {sub, sub, sub, 7, 4};
+  ASSERT_EQUAL(v, ref);
+
+  sub = 9;
+
+  uninitialized_fill_kernel<<<1, 1>>>(exec, v.begin() + 2, v.end(), sub);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ref = {8, 8, sub, sub, 9};
+  ASSERT_EQUAL(v, ref);
+
+  sub = 1;
+
+  uninitialized_fill_kernel<<<1, 1>>>(exec, v.begin(), v.end(), sub);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ref = Vector(5, sub);
+  ASSERT_EQUAL(v, ref);
+}
+
+void TestUninitializedFillDeviceSeq()
+{
+  TestUninitializedFillDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestUninitializedFillDeviceSeq);
+
+void TestUninitializedFillDeviceDevice()
+{
+  TestUninitializedFillDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestUninitializedFillDeviceDevice);
+#endif
+
+void TestUninitializedFillCudaStreams()
+{
+  using Vector = thrust::device_vector<int>;
+  using T      = Vector::value_type;
+
+  Vector v{0, 1, 2, 3, 4};
+  T sub(7);
+
+  cudaStream_t s;
+  cudaStreamCreate(&s);
+
+  thrust::uninitialized_fill(thrust::cuda::par.on(s), v.begin(), v.end(), sub);
+  cudaStreamSynchronize(s);
+
+  Vector ref(v.size(), sub);
+  ASSERT_EQUAL(v, ref);
+
+  cudaStreamDestroy(s);
+}
+DECLARE_UNITTEST(TestUninitializedFillCudaStreams);
+
+#ifdef THRUST_TEST_DEVICE_SIDE
+template <typename ExecutionPolicy, typename Iterator1, typename Size, typename T, typename Iterator2>
+__global__ void uninitialized_fill_n_kernel(ExecutionPolicy exec, Iterator1 first, Size n, T val, Iterator2 result)
+{
+  *result = thrust::uninitialized_fill_n(exec, first, n, val);
+}
+
+template <typename ExecutionPolicy>
+void TestUninitializedFillNDevice(ExecutionPolicy exec)
+{
+  using Vector = thrust::device_vector<int>;
+  using T      = Vector::value_type;
+
+  Vector v{0, 1, 2, 3, 4};
+  T sub(7);
+
+  thrust::device_vector<Vector::iterator> iter_vec(1);
+
+  uninitialized_fill_n_kernel<<<1, 1>>>(exec, v.begin() + 1, 3, sub, iter_vec.begin());
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  Vector::iterator iter = iter_vec[0];
+
+  Vector ref{0, sub, sub, sub, 4};
+  ASSERT_EQUAL(v, ref);
+  ASSERT_EQUAL_QUIET(v.begin() + 4, iter);
+
+  sub = 8;
+
+  uninitialized_fill_n_kernel<<<1, 1>>>(exec, v.begin() + 0, 3, sub, iter_vec.begin());
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  iter = iter_vec[0];
+
+  ref = {sub, sub, sub, 7, 4};
+  ASSERT_EQUAL(v, ref);
+  ASSERT_EQUAL_QUIET(v.begin() + 3, iter);
+
+  sub = 9;
+
+  uninitialized_fill_n_kernel<<<1, 1>>>(exec, v.begin() + 2, 3, sub, iter_vec.begin());
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  iter = iter_vec[0];
+
+  ref = {8, 8, sub, sub, 9};
+  ASSERT_EQUAL(v, ref);
+  ASSERT_EQUAL_QUIET(v.end(), iter);
+
+  sub = 1;
+
+  uninitialized_fill_n_kernel<<<1, 1>>>(exec, v.begin(), v.size(), sub, iter_vec.begin());
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  iter = iter_vec[0];
+
+  ref = Vector(5, sub);
+
+  ASSERT_EQUAL(v, ref);
+  ASSERT_EQUAL_QUIET(v.end(), iter);
+}
+
+void TestUninitializedFillNDeviceSeq()
+{
+  TestUninitializedFillNDevice(thrust::seq);
+}
+DECLARE_UNITTEST(TestUninitializedFillNDeviceSeq);
+
+void TestUninitializedFillNDeviceDevice()
+{
+  TestUninitializedFillNDevice(thrust::device);
+}
+DECLARE_UNITTEST(TestUninitializedFillNDeviceDevice);
+#endif
+
+void TestUninitializedFillNCudaStreams()
+{
+  using Vector = thrust::device_vector<int>;
+  using T      = Vector::value_type;
+
+  Vector v{0, 1, 2, 3, 4};
+  T sub(7);
+
+  cudaStream_t s;
+  cudaStreamCreate(&s);
+
+  thrust::uninitialized_fill_n(thrust::cuda::par.on(s), v.begin(), v.size(), sub);
+  cudaStreamSynchronize(s);
+
+  Vector ref(5, sub);
+  ASSERT_EQUAL(v, ref);
+
+  cudaStreamDestroy(s);
+}
+DECLARE_UNITTEST(TestUninitializedFillNCudaStreams);