[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/stf/slice/pinning.cu

@@ -0,0 +1,59 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDASTF 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) 2022-2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#include <cuda/experimental/__stf/stream/stream_ctx.cuh>
+
+#include <iostream>
+
+using namespace cuda::experimental::stf;
+
+int main()
+{
+  stream_ctx ctx;
+
+  // Contiguous 1D
+  double* X     = new double[1024];
+  auto handle_X = ctx.logical_data(make_slice(X, 1024));
+
+  // Contiguous 2D
+  double* X2     = new double[1024 * 1024];
+  auto handle_X2 = ctx.logical_data(make_slice(X2, std::tuple{1024, 1024}, 1024));
+
+  // Contiguous 3D
+  double* X4     = new double[128 * 128 * 128];
+  auto handle_X4 = ctx.logical_data(make_slice(X4, std::tuple{128, 128, 128}, 128, 128 * 128));
+
+  // Discontiguous 2D
+  double* X3     = new double[128 * 8];
+  auto handle_X3 = ctx.logical_data(make_slice(X3, std::tuple{64, 8}, 128));
+
+  // Discontiguous 3D
+  double* X5     = new double[32 * 4 * 4];
+  auto handle_X5 = ctx.logical_data(make_slice(X5, std::tuple{16, 4, 4}, 32, 32 * 4));
+
+  double* X6     = new double[32 * 4 * 4];
+  auto handle_X6 = ctx.logical_data(make_slice(X6, std::tuple{32, 2, 4}, 32, 32 * 4));
+
+  double* X7 = new double[128 * 128 * 128];
+  cuda_safe_call(cudaHostRegister(X7, 128 * 128 * 128, cudaHostRegisterPortable));
+  auto handle_X7 = ctx.logical_data(make_slice(X7, std::tuple{128, 128, 128}, 128, 128 * 128));
+
+  // Detect that this was already pinned
+  double* X9 = new double[1024];
+  cuda_safe_call(cudaHostRegister(X9, 1024, cudaHostRegisterPortable));
+  auto handle_X9 = ctx.logical_data(make_slice(X9, 1024));
+
+  // Detect that this was already pinned
+  double* X8 = new double[4 * 4 * 4];
+  cuda_safe_call(cudaHostRegister(X8, 4 * 4 * 4, cudaHostRegisterPortable));
+  auto handle_X8 = ctx.logical_data(make_slice(X8, std::tuple{1, 4, 4}, 4, 4 * 4));
+
+  ctx.finalize();
+}