[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/cudax/test/common/utility.cuh

@@ -0,0 +1,151 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDA Experimental in CUDA C++ Core Libraries,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __COMMON_UTILITY_H__
+#define __COMMON_UTILITY_H__
+
+#include <cuda_runtime_api.h>
+// cuda_runtime_api needs to come first
+
+#include <cuda/__runtime/api_wrapper.h>
+#include <cuda/__stream/stream_ref.h>
+#include <cuda/atomic>
+#include <cuda/std/utility>
+
+#include <cuda/experimental/launch.cuh>
+
+#include <new> // IWYU pragma: keep (needed for placement new)
+
+#include "testing.cuh"
+
+namespace
+{
+namespace test
+{
+constexpr auto one_thread_dims = cuda::make_config(cuda::block_dims<1>(), cuda::grid_dims<1>());
+
+struct _malloc_pinned
+{
+private:
+  void* pv = nullptr;
+
+public:
+  explicit _malloc_pinned(std::size_t size)
+  {
+    cuda::__ensure_current_context guard(cuda::device_ref{0});
+    _CCCL_TRY_CUDA_API(::cudaMallocHost, "failed to allocate pinned memory", &pv, size);
+  }
+
+  ~_malloc_pinned()
+  {
+    cuda::__ensure_current_context guard(cuda::device_ref{0});
+    [[maybe_unused]] auto status = ::cudaFreeHost(pv);
+  }
+
+  template <class T>
+  T* get_as() const noexcept
+  {
+    return static_cast<T*>(pv);
+  }
+};
+
+template <class T>
+struct pinned
+{
+private:
+  _malloc_pinned _mem;
+
+public:
+  explicit pinned(T t)
+      : _mem(sizeof(T))
+  {
+    ::new (_mem.get_as<void>()) T(std::move(t));
+  }
+
+  ~pinned()
+  {
+    get()->~T();
+  }
+
+  T* get() noexcept
+  {
+    return _mem.get_as<T>();
+  }
+  const T* get() const noexcept
+  {
+    return _mem.get_as<T>();
+  }
+
+  T& operator*() noexcept
+  {
+    return *get();
+  }
+  const T& operator*() const noexcept
+  {
+    return *get();
+  }
+};
+
+template <int N>
+struct assign_n
+{
+  __device__ constexpr void operator()(int* pi) const noexcept
+  {
+    *pi = N;
+  }
+};
+
+template <int N>
+struct verify_n
+{
+  __device__ void operator()(int* pi) const noexcept
+  {
+    REQUIRE_DEVICE(*pi == N);
+  }
+};
+
+using assign_42 = assign_n<42>;
+using verify_42 = verify_n<42>;
+
+struct atomic_add_one
+{
+  __device__ void operator()(int* pi) const noexcept
+  {
+    cuda::atomic_ref atomic_pi(*pi);
+    atomic_pi.fetch_add(1);
+  }
+};
+
+struct atomic_sub_one
+{
+  __device__ void operator()(int* pi) const noexcept
+  {
+    cuda::atomic_ref atomic_pi(*pi);
+    atomic_pi.fetch_sub(1);
+  }
+};
+
+struct spin_until_80
+{
+  __device__ void operator()(int* pi) const noexcept
+  {
+    cuda::atomic_ref atomic_pi(*pi);
+    while (atomic_pi.load() != 80)
+      ;
+  }
+};
+
+struct empty_kernel
+{
+  __device__ void operator()() const noexcept {}
+};
+} // namespace test
+} // namespace
+#endif // __COMMON_UTILITY_H__