[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/cub/test/catch2_test_warp_mask.cu

@@ -0,0 +1,84 @@
+// SPDX-FileCopyrightText: Copyright (c) 2023, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: BSD-3
+
+#include <cub/util_arch.cuh>
+#include <cub/util_ptx.cuh>
+
+#include <cuda/__cmath/pow2.h>
+
+#include <c2h/catch2_test_helper.h>
+
+template <int logical_warp_threads>
+struct total_warps_t
+{
+private:
+  static constexpr unsigned int total_warps =
+    (::cuda::is_power_of_two(logical_warp_threads)) ? cub::detail::warp_threads / logical_warp_threads : 1;
+
+public:
+  static constexpr unsigned int value()
+  {
+    return total_warps;
+  }
+};
+
+bool is_lane_involved(unsigned int member_mask, unsigned int lane)
+{
+  return member_mask & (1 << lane);
+}
+
+using logical_warp_threads      = c2h::iota<1, 32>;
+using power_of_two_warp_threads = c2h::enum_type_list<int, 1, 2, 4, 8, 16, 32>;
+
+C2H_TEST("Warp mask ignores lanes before current logical warp", "[mask][warp]", power_of_two_warp_threads)
+{
+  constexpr int logical_warp_thread  = c2h::get<0, TestType>::value;
+  constexpr unsigned int total_warps = total_warps_t<logical_warp_thread>::value();
+
+  for (unsigned int warp_id = 0; warp_id < total_warps; warp_id++)
+  {
+    const unsigned int warp_mask  = cub::WarpMask<logical_warp_thread>(warp_id);
+    const unsigned int warp_begin = logical_warp_thread * warp_id;
+
+    for (unsigned int prev_warp_lane = 0; prev_warp_lane < warp_begin; prev_warp_lane++)
+    {
+      REQUIRE_FALSE(is_lane_involved(warp_mask, prev_warp_lane));
+    }
+  }
+}
+
+C2H_TEST("Warp mask involves lanes of current logical warp", "[mask][warp]", logical_warp_threads)
+{
+  constexpr int logical_warp_thread  = c2h::get<0, TestType>::value;
+  constexpr unsigned int total_warps = total_warps_t<logical_warp_thread>::value();
+
+  for (unsigned int warp_id = 0; warp_id < total_warps; warp_id++)
+  {
+    const unsigned int warp_mask  = cub::WarpMask<logical_warp_thread>(warp_id);
+    const unsigned int warp_begin = logical_warp_thread * warp_id;
+    const unsigned int warp_end   = warp_begin + logical_warp_thread;
+
+    for (unsigned int warp_lane = warp_begin; warp_lane < warp_end; warp_lane++)
+    {
+      REQUIRE(is_lane_involved(warp_mask, warp_lane));
+    }
+  }
+}
+
+C2H_TEST("Warp mask ignores lanes after current logical warp", "[mask][warp]", logical_warp_threads)
+{
+  constexpr int logical_warp_thread  = c2h::get<0, TestType>::value;
+  constexpr unsigned int total_warps = total_warps_t<logical_warp_thread>::value();
+
+  for (unsigned int warp_id = 0; warp_id < total_warps; warp_id++)
+  {
+    const unsigned int warp_mask  = cub::WarpMask<logical_warp_thread>(warp_id);
+    const unsigned int warp_begin = logical_warp_thread * warp_id;
+    const unsigned int warp_end   = warp_begin + logical_warp_thread;
+
+    for (unsigned int post_warp_lane = warp_end; post_warp_lane < cub::detail::warp_threads; post_warp_lane++)
+    {
+      REQUIRE_FALSE(is_lane_involved(warp_mask, post_warp_lane));
+    }
+  }
+}