[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/algorithm/fill.cu

@@ -0,0 +1,93 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#include "common.cuh"
+
+C2H_TEST("Fill", "[data_manipulation]")
+{
+  cuda::stream _stream{cuda::device_ref{0}};
+  SECTION("Host resource")
+  {
+    cuda::mr::legacy_pinned_memory_resource host_resource;
+    cudax::uninitialized_buffer<int, cuda::mr::device_accessible> buffer(host_resource, buffer_size);
+
+    cuda::fill_bytes(_stream, buffer, fill_byte);
+
+    check_result_and_erase(_stream, cuda::std::span(buffer));
+  }
+
+  SECTION("Device resource")
+  {
+    cuda::device_memory_pool_ref device_resource = cuda::device_default_memory_pool(cuda::device_ref{0});
+    cudax::uninitialized_buffer<int, cuda::mr::device_accessible> buffer(device_resource, buffer_size);
+    cuda::fill_bytes(_stream, buffer, fill_byte);
+
+    std::vector<int> host_vector(42);
+    REQUIRE_CUDART(cudaMemcpyAsync(
+      host_vector.data(), buffer.data(), buffer.size() * sizeof(int), cudaMemcpyDefault, _stream.get()));
+
+    check_result_and_erase(_stream, host_vector);
+  }
+  SECTION("Launch transform")
+  {
+    cuda::mr::legacy_pinned_memory_resource host_resource;
+    cudax::weird_buffer buffer(host_resource, buffer_size);
+
+    cuda::fill_bytes(_stream, buffer, fill_byte);
+    check_result_and_erase(_stream, cuda::std::span(buffer.data, buffer.size));
+  }
+}
+
+C2H_TEST("Mdspan Fill", "[data_manipulation]")
+{
+  cuda::stream stream{cuda::device_ref{0}};
+  {
+    cuda::std::dextents<size_t, 3> dynamic_extents{1, 2, 3};
+    auto buffer = make_buffer_for_mdspan(dynamic_extents, 0);
+    cuda::std::mdspan<int, decltype(dynamic_extents)> dynamic_mdspan(buffer.data(), dynamic_extents);
+
+    cuda::fill_bytes(stream, dynamic_mdspan, fill_byte);
+    check_result_and_erase(stream, cuda::std::span(buffer.data(), buffer.size()));
+  }
+  {
+    cuda::std::extents<size_t, 2, cuda::std::dynamic_extent, 4> mixed_extents{1};
+    auto buffer = make_buffer_for_mdspan(mixed_extents, 0);
+    cuda::std::mdspan<int, decltype(mixed_extents)> mixed_mdspan(buffer.data(), mixed_extents);
+
+    cuda::fill_bytes(stream, cuda::std::move(mixed_mdspan), fill_byte);
+    check_result_and_erase(stream, cuda::std::span(buffer.data(), buffer.size()));
+  }
+  {
+    cuda::mr::legacy_pinned_memory_resource host_resource;
+    using static_extents = cuda::std::extents<size_t, 2, 3, 4>;
+    auto size            = cuda::std::layout_left::mapping<static_extents>().required_span_size();
+    cudax::weird_buffer<cuda::std::mdspan<int, static_extents>> buffer(host_resource, size);
+
+    cuda::fill_bytes(stream, buffer, fill_byte);
+    check_result_and_erase(stream, cuda::std::span(buffer.data, buffer.size));
+  }
+}
+
+C2H_TEST("Non exhaustive mdspan fill_bytes", "[data_manipulation]")
+{
+  cuda::stream stream{cuda::device_ref{0}};
+  {
+    auto fake_strided_mdspan = create_fake_strided_mdspan();
+
+    try
+    {
+      cuda::fill_bytes(stream, fake_strided_mdspan, fill_byte);
+    }
+    catch (const ::std::invalid_argument& e)
+    {
+      CHECK(e.what() == ::std::string("fill_bytes supports only exhaustive mdspans"));
+    }
+  }
+}