[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/group/synchronizer/barrier_synchronizer.cu

@@ -0,0 +1,147 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2026 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#include <cuda/barrier>
+#include <cuda/devices>
+#include <cuda/hierarchy>
+#include <cuda/launch>
+#include <cuda/std/cstddef>
+#include <cuda/std/type_traits>
+#include <cuda/stream>
+
+#include <cuda/experimental/group.cuh>
+
+#include "group_testing.cuh"
+
+namespace
+{
+template <class Level, class Config>
+__device__ void test_barrier_synchronizer(const Level& level, Config config)
+{
+  constexpr cuda::std::size_t nbarriers = 8;
+
+  // Test constructor from static span of barriers.
+  {
+    auto& barriers = get_barriers<nbarriers, 0>(level);
+    using Barrier  = cuda::std::remove_all_extents_t<cuda::std::remove_reference_t<decltype(barriers)>>;
+
+    cuda::std::span<Barrier, nbarriers> barriers_span{barriers, nbarriers};
+    cudax::barrier_synchronizer synchronizer{barriers_span};
+
+    static_assert(cuda::std::is_same_v<cudax::barrier_synchronizer<Barrier, nbarriers>, decltype(synchronizer)>);
+    static_assert(cuda::std::is_nothrow_constructible_v<decltype(synchronizer), decltype(barriers_span)>);
+
+    CHECK(synchronizer.barriers().data() == barriers);
+    CHECK(synchronizer.barriers().size() == nbarriers);
+  }
+
+  // Test constructor from dynamic span of barriers.
+  {
+    auto& barriers = get_barriers<nbarriers, 1>(level);
+    using Barrier  = cuda::std::remove_all_extents_t<cuda::std::remove_reference_t<decltype(barriers)>>;
+
+    cuda::std::span<Barrier> barriers_span{barriers, nbarriers};
+    cudax::barrier_synchronizer synchronizer{barriers_span};
+
+    static_assert(
+      cuda::std::is_same_v<cudax::barrier_synchronizer<Barrier, cuda::std::dynamic_extent>, decltype(synchronizer)>);
+    static_assert(cuda::std::is_nothrow_constructible_v<decltype(synchronizer), decltype(barriers_span)>);
+
+    CHECK(synchronizer.barriers().data() == barriers);
+    CHECK(synchronizer.barriers().size() == nbarriers);
+  }
+
+  // Test constructor from array of barriers.
+  {
+    auto& barriers = get_barriers<nbarriers, 2>(level);
+    using Barrier  = cuda::std::remove_all_extents_t<cuda::std::remove_reference_t<decltype(barriers)>>;
+
+    cudax::barrier_synchronizer synchronizer{barriers};
+
+    static_assert(cuda::std::is_same_v<cudax::barrier_synchronizer<Barrier, nbarriers>, decltype(synchronizer)>);
+    static_assert(cuda::std::is_nothrow_constructible_v<decltype(synchronizer), decltype(barriers)>);
+
+    CHECK(synchronizer.barriers().data() == barriers);
+    CHECK(synchronizer.barriers().size() == nbarriers);
+  }
+
+  // Test barriers().
+  {
+    auto& barriers = get_barriers<nbarriers, 3>(level);
+    using Barrier  = cuda::std::remove_all_extents_t<cuda::std::remove_reference_t<decltype(barriers)>>;
+
+    const cudax::barrier_synchronizer synchronizer{barriers};
+
+    static_assert(cuda::std::is_same_v<cuda::std::span<Barrier, nbarriers>, decltype(synchronizer.barriers())>);
+    static_assert(noexcept(synchronizer.barriers()));
+
+    CHECK(synchronizer.barriers().data() == barriers);
+    CHECK(synchronizer.barriers().size() == nbarriers);
+  }
+
+  // Test make_instance(...).
+  {
+    auto& barriers = get_barriers<nbarriers, 4>(level);
+    using Barrier  = cuda::std::remove_all_extents_t<cuda::std::remove_reference_t<decltype(barriers)>>;
+
+    const auto parent_group = cudax::make_this_group(level, config);
+    const ThreadsInWarpMappingResult prev_mapping_result;
+
+    const cudax::group_by mapping{4};
+    const cudax::barrier_synchronizer synchronizer{barriers};
+
+    const auto mapping_result = mapping.map(cuda::gpu_thread, parent_group, prev_mapping_result);
+    const auto synchronizer_instance =
+      synchronizer.make_instance(cuda::gpu_thread, parent_group, mapping, mapping_result);
+
+    // Test do_sync(...).
+    static_assert(cuda::std::is_same_v<void, decltype(synchronizer_instance.do_sync(mapping_result, synchronizer))>);
+    static_assert(noexcept(synchronizer_instance.do_sync(mapping_result, synchronizer)));
+    synchronizer_instance.do_sync(mapping_result, synchronizer);
+
+    // Test do_sync_aligned(...).
+    static_assert(
+      cuda::std::is_same_v<void, decltype(synchronizer_instance.do_sync_aligned(mapping_result, synchronizer))>);
+    static_assert(noexcept(synchronizer_instance.do_sync_aligned(mapping_result, synchronizer)));
+    synchronizer_instance.do_sync_aligned(mapping_result, synchronizer);
+  }
+}
+
+struct TestKernel
+{
+  template <class Config>
+  __device__ void operator()(const Config& config)
+  {
+    test_barrier_synchronizer(cuda::warp, config);
+    test_barrier_synchronizer(cuda::block, config);
+    test_barrier_synchronizer(cuda::cluster, config);
+    test_barrier_synchronizer(cuda::grid, config);
+  }
+};
+} // namespace
+
+C2H_TEST("Barrier synchronizer", "[group]")
+{
+  const auto device = cuda::devices[0];
+
+  const cuda::stream stream{device};
+
+  {
+    const auto config = cuda::make_config(cuda::grid_dims<1>(), cuda::block_dims<8, 4>(), cuda::cooperative_launch{});
+    cuda::launch(stream, config, TestKernel{});
+  }
+  {
+    const auto config =
+      cuda::make_config(cuda::grid_dims<1>(), cuda::block_dims(dim3{8, 4}), cuda::cooperative_launch{});
+    cuda::launch(stream, config, TestKernel{});
+  }
+
+  stream.sync();
+}

cccl_upstream/cudax/test/group/synchronizer/lane_synchronizer.cu

@@ -0,0 +1,87 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2026 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#include <cuda/devices>
+#include <cuda/hierarchy>
+#include <cuda/launch>
+#include <cuda/std/bit>
+#include <cuda/std/type_traits>
+#include <cuda/stream>
+
+#include <cuda/experimental/group.cuh>
+
+#include "group_testing.cuh"
+
+namespace
+{
+template <class Level, class Config>
+__device__ void test_lane_synchronizer(const Level& level, Config config)
+{
+  using Synchronizer = cudax::lane_synchronizer;
+  static_assert(cuda::std::is_empty_v<Synchronizer>);
+
+  // Test default constructor.
+  static_assert(cuda::std::is_trivially_default_constructible_v<Synchronizer>);
+
+  // Test make_instance(...).
+  {
+    const auto parent_group = cudax::make_this_group(level, config);
+    const ThreadsInWarpMappingResult prev_mapping_result;
+
+    const cudax::group_by mapping{2};
+    const Synchronizer synchronizer{};
+
+    const auto mapping_result = mapping.map(cuda::gpu_thread, parent_group, prev_mapping_result);
+    const auto synchronizer_instance =
+      synchronizer.make_instance(cuda::gpu_thread, parent_group, mapping, mapping_result);
+
+    // Test do_sync(...).
+    static_assert(cuda::std::is_same_v<void, decltype(synchronizer_instance.do_sync(mapping_result, synchronizer))>);
+    static_assert(noexcept(synchronizer_instance.do_sync(mapping_result, synchronizer)));
+    synchronizer_instance.do_sync(mapping_result, synchronizer);
+
+    // Test do_sync_aligned(...).
+    static_assert(
+      cuda::std::is_same_v<void, decltype(synchronizer_instance.do_sync_aligned(mapping_result, synchronizer))>);
+    static_assert(noexcept(synchronizer_instance.do_sync_aligned(mapping_result, synchronizer)));
+    synchronizer_instance.do_sync_aligned(mapping_result, synchronizer);
+  }
+}
+
+struct TestKernel
+{
+  template <class Config>
+  __device__ void operator()(const Config& config)
+  {
+    test_lane_synchronizer(cuda::warp, config);
+    test_lane_synchronizer(cuda::block, config);
+    test_lane_synchronizer(cuda::cluster, config);
+    test_lane_synchronizer(cuda::grid, config);
+  }
+};
+} // namespace
+
+C2H_TEST("Lane synchronizer", "[group]")
+{
+  const auto device = cuda::devices[0];
+
+  const cuda::stream stream{device};
+
+  {
+    const auto config = cuda::make_config(cuda::grid_dims<1>(), cuda::block_dims<8, 4>());
+    cuda::launch(stream, config, TestKernel{});
+  }
+  {
+    const auto config = cuda::make_config(cuda::grid_dims<1>(), cuda::block_dims(dim3{8, 4}));
+    cuda::launch(stream, config, TestKernel{});
+  }
+
+  stream.sync();
+}