[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/testing/cuda/fill.cu

@@ -0,0 +1,207 @@
+#include <thrust/execution_policy.h>
+#include <thrust/fill.h>
+
+#include <algorithm>
+
+#include <unittest/unittest.h>
+
+#ifdef THRUST_TEST_DEVICE_SIDE
+template <typename ExecutionPolicy, typename Iterator, typename T>
+__global__ void fill_kernel(ExecutionPolicy exec, Iterator first, Iterator last, T value)
+{
+  thrust::fill(exec, first, last, value);
+}
+
+template <typename T, typename ExecutionPolicy>
+void TestFillDevice(ExecutionPolicy exec, size_t n)
+{
+  thrust::host_vector<T> h_data   = unittest::random_integers<T>(n);
+  thrust::device_vector<T> d_data = h_data;
+
+  thrust::fill(h_data.begin() + std::min((size_t) 1, n), h_data.begin() + std::min((size_t) 3, n), (T) 0);
+
+  fill_kernel<<<1, 1>>>(exec, d_data.begin() + std::min((size_t) 1, n), d_data.begin() + std::min((size_t) 3, n), (T) 0);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  thrust::fill(h_data.begin() + std::min((size_t) 117, n), h_data.begin() + std::min((size_t) 367, n), (T) 1);
+
+  fill_kernel<<<1, 1>>>(
+    exec, d_data.begin() + std::min((size_t) 117, n), d_data.begin() + std::min((size_t) 367, n), (T) 1);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  thrust::fill(h_data.begin() + std::min((size_t) 8, n), h_data.begin() + std::min((size_t) 259, n), (T) 2);
+
+  fill_kernel<<<1, 1>>>(
+    exec, d_data.begin() + std::min((size_t) 8, n), d_data.begin() + std::min((size_t) 259, n), (T) 2);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  thrust::fill(h_data.begin() + std::min((size_t) 3, n), h_data.end(), (T) 3);
+
+  fill_kernel<<<1, 1>>>(exec, d_data.begin() + std::min((size_t) 3, n), d_data.end(), (T) 3);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  thrust::fill(h_data.begin(), h_data.end(), (T) 4);
+
+  fill_kernel<<<1, 1>>>(exec, d_data.begin(), d_data.end(), (T) 4);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+}
+
+template <typename T>
+void TestFillDeviceSeq(size_t n)
+{
+  TestFillDevice<T>(thrust::seq, n);
+}
+DECLARE_VARIABLE_UNITTEST(TestFillDeviceSeq);
+
+template <typename T>
+void TestFillDeviceDevice(size_t n)
+{
+  TestFillDevice<T>(thrust::device, n);
+}
+DECLARE_VARIABLE_UNITTEST(TestFillDeviceDevice);
+
+template <typename ExecutionPolicy, typename Iterator, typename Size, typename T>
+__global__ void fill_n_kernel(ExecutionPolicy exec, Iterator first, Size n, T value)
+{
+  thrust::fill_n(exec, first, n, value);
+}
+
+template <typename T, typename ExecutionPolicy>
+void TestFillNDevice(ExecutionPolicy exec, size_t n)
+{
+  thrust::host_vector<T> h_data   = unittest::random_integers<T>(n);
+  thrust::device_vector<T> d_data = h_data;
+
+  size_t begin_offset = std::min<size_t>(1, n);
+
+  thrust::fill_n(h_data.begin() + begin_offset, std::min((size_t) 3, n) - begin_offset, (T) 0);
+
+  fill_n_kernel<<<1, 1>>>(exec, d_data.begin() + begin_offset, std::min((size_t) 3, n) - begin_offset, (T) 0);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  begin_offset = std::min<size_t>(117, n);
+
+  thrust::fill_n(h_data.begin() + begin_offset, std::min((size_t) 367, n) - begin_offset, (T) 1);
+
+  fill_n_kernel<<<1, 1>>>(exec, d_data.begin() + begin_offset, std::min((size_t) 367, n) - begin_offset, (T) 1);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  begin_offset = std::min<size_t>(8, n);
+
+  thrust::fill_n(h_data.begin() + begin_offset, std::min((size_t) 259, n) - begin_offset, (T) 2);
+
+  fill_n_kernel<<<1, 1>>>(exec, d_data.begin() + begin_offset, std::min((size_t) 259, n) - begin_offset, (T) 2);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  begin_offset = std::min<size_t>(3, n);
+
+  thrust::fill_n(h_data.begin() + begin_offset, h_data.size() - begin_offset, (T) 3);
+
+  fill_n_kernel<<<1, 1>>>(exec, d_data.begin() + begin_offset, d_data.size() - begin_offset, (T) 3);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+
+  thrust::fill_n(h_data.begin(), h_data.size(), (T) 4);
+
+  fill_n_kernel<<<1, 1>>>(exec, d_data.begin(), d_data.size(), (T) 4);
+  {
+    cudaError_t const err = cudaDeviceSynchronize();
+    ASSERT_EQUAL(cudaSuccess, err);
+  }
+
+  ASSERT_EQUAL(h_data, d_data);
+}
+
+template <typename T>
+void TestFillNDeviceSeq(size_t n)
+{
+  TestFillNDevice<T>(thrust::seq, n);
+}
+DECLARE_VARIABLE_UNITTEST(TestFillNDeviceSeq);
+
+template <typename T>
+void TestFillNDeviceDevice(size_t n)
+{
+  TestFillNDevice<T>(thrust::device, n);
+}
+DECLARE_VARIABLE_UNITTEST(TestFillNDeviceDevice);
+#endif
+
+void TestFillCudaStreams()
+{
+  thrust::device_vector<int> v{0, 1, 2, 3, 4};
+
+  cudaStream_t s;
+  cudaStreamCreate(&s);
+
+  thrust::fill(thrust::cuda::par.on(s), v.begin() + 1, v.begin() + 4, 7);
+  cudaStreamSynchronize(s);
+
+  thrust::device_vector<int> ref{0, 7, 7, 7, 4};
+  ASSERT_EQUAL(v, ref);
+
+  thrust::fill(thrust::cuda::par.on(s), v.begin() + 0, v.begin() + 3, 8);
+  cudaStreamSynchronize(s);
+
+  ref = {8, 8, 8, 7, 4};
+  ASSERT_EQUAL(v, ref);
+
+  thrust::fill(thrust::cuda::par.on(s), v.begin() + 2, v.end(), 9);
+  cudaStreamSynchronize(s);
+
+  ref = {8, 8, 9, 9, 9};
+  ASSERT_EQUAL(v, ref);
+
+  thrust::fill(thrust::cuda::par.on(s), v.begin(), v.end(), 1);
+  cudaStreamSynchronize(s);
+
+  ref = {1, 1, 1, 1, 1};
+  ASSERT_EQUAL(v, ref);
+
+  cudaStreamDestroy(s);
+}
+DECLARE_UNITTEST(TestFillCudaStreams);