[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/host_device.cuh

@@ -0,0 +1,101 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2023 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __COMMON_HOST_DEVICE_H__
+#define __COMMON_HOST_DEVICE_H__
+
+#include "utility.cuh"
+
+template <typename Dims, typename Lambda>
+void __global__ lambda_launcher(const Dims dims, const Lambda lambda)
+{
+  lambda(dims);
+}
+
+template <typename Comparator, unsigned int FilterArch>
+bool arch_filter(const cudaDeviceProp& props)
+{
+  int act_arch = props.major * 10 + props.minor;
+  if (Comparator()(act_arch, FilterArch))
+  {
+    return true;
+  }
+  return false;
+}
+
+static bool skip_host_exec(bool (* /* filter */)(const cudaDeviceProp&))
+{
+  return false;
+}
+
+static bool skip_device_exec(bool (*filter)(const cudaDeviceProp&))
+{
+  cudaDeviceProp props;
+  REQUIRE_CUDART(cudaGetDeviceProperties(&props, 0));
+  return filter(props);
+}
+
+template <typename Dims, typename Lambda, typename... Filters>
+void test_host_dev(const Dims& dims, const Lambda& lambda, const Filters&... filters)
+{
+  SECTION("Host execution")
+  {
+    if ((... && !skip_host_exec(filters)))
+    {
+      // host testing
+      lambda(dims);
+    }
+  }
+
+  SECTION("Device execution")
+  {
+    // Asymmetrical but cleaner
+    if ((... || skip_device_exec(filters)))
+    {
+      return;
+    }
+
+    cudaLaunchConfig_t config = {};
+    config.gridDim            = {0};
+    cudaLaunchAttribute attrs[1];
+    config.attrs = &attrs[0];
+
+    config.blockDim = dims.extents(cuda::gpu_thread, cuda::block);
+    config.gridDim  = dims.extents(cuda::block, cuda::grid);
+
+    if constexpr (Dims::has_level(cluster))
+    {
+      dim3 cluster_dims                            = dims.extents(cuda::block, cuda::cluster);
+      config.attrs[config.numAttrs].id             = cudaLaunchAttributeClusterDimension;
+      config.attrs[config.numAttrs].val.clusterDim = {cluster_dims.x, cluster_dims.y, cluster_dims.z};
+      config.numAttrs                              = 1;
+    }
+    else
+    {
+      config.numAttrs = 0;
+    }
+
+    // device testing
+    REQUIRE_CUDART(cudaLaunchKernelEx(&config, lambda_launcher<Dims, Lambda>, dims, lambda));
+    REQUIRE_CUDART(cudaDeviceSynchronize());
+  }
+}
+
+template <typename Fn, typename Tuple>
+void apply_each(const Fn& fn, const Tuple& tuple)
+{
+  cuda::std::apply(
+    [&](const auto&... elems) {
+      (fn(elems), ...);
+    },
+    tuple);
+}
+
+#endif // __COMMON_HOST_DEVICE_H__