[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/thrust/benchmarks/bench/adjacent_difference/basic.cu

@@ -0,0 +1,33 @@
+// SPDX-FileCopyrightText: Copyright (c) 2011-2023, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: BSD-3
+
+#include <thrust/adjacent_difference.h>
+#include <thrust/device_vector.h>
+#include <thrust/execution_policy.h>
+
+#include "nvbench_helper.cuh"
+
+template <typename T>
+static void basic(nvbench::state& state, nvbench::type_list<T>)
+{
+  const auto elements = static_cast<std::size_t>(state.get_int64("Elements"));
+
+  thrust::device_vector<T> input = generate(elements);
+  thrust::device_vector<T> output(elements);
+
+  state.add_element_count(elements);
+  state.add_global_memory_reads<T>(elements);
+  state.add_global_memory_writes<T>(elements);
+
+  caching_allocator_t alloc;
+  state.exec(nvbench::exec_tag::gpu | nvbench::exec_tag::sync, [&](nvbench::launch& launch) {
+    thrust::adjacent_difference(policy(alloc, launch), input.cbegin(), input.cend(), output.begin());
+  });
+}
+
+using types = nvbench::type_list<int8_t, int16_t, int32_t, int64_t, float, double>;
+
+NVBENCH_BENCH_TYPES(basic, NVBENCH_TYPE_AXES(types))
+  .set_name("base")
+  .set_type_axes_names({"T{ct}"})
+  .add_int64_power_of_two_axis("Elements", nvbench::range(16, 28, 4));

cccl_upstream/thrust/benchmarks/bench/adjacent_difference/custom.cu

@@ -0,0 +1,52 @@
+// SPDX-FileCopyrightText: Copyright (c) 2011-2023, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: BSD-3
+
+#include <thrust/adjacent_difference.h>
+#include <thrust/device_vector.h>
+#include <thrust/execution_policy.h>
+
+#include "nvbench_helper.cuh"
+
+template <typename T>
+struct custom_op
+{
+  T val;
+
+  custom_op() = delete;
+
+  explicit custom_op(T val)
+      : val(val)
+  {}
+
+  __device__ T operator()(const T& lhs, const T& rhs)
+  {
+    return lhs * rhs + val; // Hope to gen mad
+  }
+};
+
+template <typename T>
+static void basic(nvbench::state& state, nvbench::type_list<T>)
+{
+  const auto elements = static_cast<std::size_t>(state.get_int64("Elements"));
+
+  thrust::device_vector<T> input = generate(elements);
+  thrust::device_vector<T> output(elements);
+
+  state.add_element_count(elements);
+  state.add_global_memory_reads<T>(elements);
+  state.add_global_memory_writes<T>(elements);
+
+  caching_allocator_t alloc;
+  state.exec(nvbench::exec_tag::gpu | nvbench::exec_tag::no_batch | nvbench::exec_tag::sync,
+             [&](nvbench::launch& launch) {
+               thrust::adjacent_difference(
+                 policy(alloc, launch), input.cbegin(), input.cend(), output.begin(), custom_op<T>{42});
+             });
+}
+
+using types = nvbench::type_list<int8_t, int16_t, int32_t, int64_t, float, double>;
+
+NVBENCH_BENCH_TYPES(basic, NVBENCH_TYPE_AXES(types))
+  .set_name("base")
+  .set_type_axes_names({"T{ct}"})
+  .add_int64_power_of_two_axis("Elements", nvbench::range(16, 28, 4));

cccl_upstream/thrust/benchmarks/bench/adjacent_difference/in_place.cu

@@ -0,0 +1,33 @@
+// SPDX-FileCopyrightText: Copyright (c) 2011-2023, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: BSD-3
+
+#include <thrust/adjacent_difference.h>
+#include <thrust/device_vector.h>
+#include <thrust/execution_policy.h>
+
+#include "nvbench_helper.cuh"
+
+template <typename T>
+static void basic(nvbench::state& state, nvbench::type_list<T>)
+{
+  const auto elements = static_cast<std::size_t>(state.get_int64("Elements"));
+
+  thrust::device_vector<T> vec(elements, 0);
+
+  state.add_element_count(elements);
+  state.add_global_memory_reads<T>(elements);
+  state.add_global_memory_writes<T>(elements);
+
+  caching_allocator_t alloc;
+  state.exec(nvbench::exec_tag::gpu | nvbench::exec_tag::no_batch | nvbench::exec_tag::sync,
+             [&](nvbench::launch& launch) {
+               thrust::adjacent_difference(policy(alloc, launch), vec.begin(), vec.end(), vec.begin());
+             });
+}
+
+using types = nvbench::type_list<int8_t, int16_t, int32_t, int64_t, float, double>;
+
+NVBENCH_BENCH_TYPES(basic, NVBENCH_TYPE_AXES(types))
+  .set_name("base")
+  .set_type_axes_names({"T{ct}"})
+  .add_int64_power_of_two_axis("Elements", nvbench::range(16, 28, 4));