Fix building cuml with CCCL main

.devcontainer/Dockerfile

@@ -7,6 +7,11 @@ FROM ${BASE} as pip-base
 
 ENV DEFAULT_VIRTUAL_ENV=rapids
 
+RUN apt update -y \
+ && DEBIAN_FRONTEND=noninteractive apt install -y \
+ libblas-dev liblapack-dev \
+ && rm -rf /tmp/* /var/tmp/* /var/cache/apt/* /var/lib/apt/lists/*;
+
 FROM ${BASE} as conda-base
 
 ENV DEFAULT_CONDA_ENV=rapids

.devcontainer/cuda11.8-conda/devcontainer.json

@@ -11,7 +11,7 @@
  "runArgs": [
  "--rm",
  "--name",
- "${localEnv:USER}-rapids-${localWorkspaceFolderBasename}-24.06-cuda11.8-conda"
+ "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-24.06-cuda11.8-conda"
  ],
  "hostRequirements": {"gpu": "optional"},
  "features": {

.devcontainer/cuda11.8-pip/devcontainer.json

@@ -5,19 +5,16 @@
  "args": {
  "CUDA": "11.8",
  "PYTHON_PACKAGE_MANAGER": "pip",
- "BASE": "rapidsai/devcontainers:24.06-cpp-cuda11.8-ubuntu22.04"
+ "BASE": "rapidsai/devcontainers:24.06-cpp-cuda11.8-ucx1.15.0-openmpi-ubuntu22.04"
  }
  },
  "runArgs": [
  "--rm",
  "--name",
- "${localEnv:USER}-rapids-${localWorkspaceFolderBasename}-24.06-cuda11.8-pip"
+ "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-24.06-cuda11.8-pip"
  ],
  "hostRequirements": {"gpu": "optional"},
  "features": {
- "ghcr.io/rapidsai/devcontainers/features/ucx:24.6": {
- "version": "1.15.0"
- },
  "ghcr.io/rapidsai/devcontainers/features/cuda:24.6": {
  "version": "11.8",
  "installcuBLAS": true,
@@ -28,7 +25,6 @@
  "ghcr.io/rapidsai/devcontainers/features/rapids-build-utils:24.6": {}
  },
  "overrideFeatureInstallOrder": [
- "ghcr.io/rapidsai/devcontainers/features/ucx",
  "ghcr.io/rapidsai/devcontainers/features/cuda",
  "ghcr.io/rapidsai/devcontainers/features/rapids-build-utils"
  ],

.devcontainer/cuda12.2-conda/devcontainer.json

@@ -11,7 +11,7 @@
  "runArgs": [
  "--rm",
  "--name",
- "${localEnv:USER}-rapids-${localWorkspaceFolderBasename}-24.06-cuda12.2-conda"
+ "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-24.06-cuda12.2-conda"
  ],
  "hostRequirements": {"gpu": "optional"},
  "features": {

.devcontainer/cuda12.2-pip/devcontainer.json

@@ -5,19 +5,16 @@
  "args": {
  "CUDA": "12.2",
  "PYTHON_PACKAGE_MANAGER": "pip",
- "BASE": "rapidsai/devcontainers:24.06-cpp-cuda12.2-ubuntu22.04"
+ "BASE": "rapidsai/devcontainers:24.06-cpp-cuda12.2-ucx1.15.0-openmpi-ubuntu22.04"
  }
  },
  "runArgs": [
  "--rm",
  "--name",
- "${localEnv:USER}-rapids-${localWorkspaceFolderBasename}-24.06-cuda12.2-pip"
+ "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-24.06-cuda12.2-pip"
  ],
  "hostRequirements": {"gpu": "optional"},
  "features": {
- "ghcr.io/rapidsai/devcontainers/features/ucx:24.6": {
- "version": "1.15.0"
- },
  "ghcr.io/rapidsai/devcontainers/features/cuda:24.6": {
  "version": "12.2",
  "installcuBLAS": true,
@@ -28,7 +25,6 @@
  "ghcr.io/rapidsai/devcontainers/features/rapids-build-utils:24.6": {}
  },
  "overrideFeatureInstallOrder": [
- "ghcr.io/rapidsai/devcontainers/features/ucx",
  "ghcr.io/rapidsai/devcontainers/features/cuda",
  "ghcr.io/rapidsai/devcontainers/features/rapids-build-utils"
  ],

cpp/bench/CMakeLists.txt

@@ -1,5 +1,5 @@
 #=============================================================================
-# Copyright (c) 2019-2023, NVIDIA CORPORATION.
+# Copyright (c) 2019-2024, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -61,6 +61,10 @@ if(BUILD_CUML_BENCH)
  set_target_properties(
  ${CUML_CPP_BENCH_TARGET}
  PROPERTIES INSTALL_RPATH "\$ORIGIN/../../../lib"
+ CXX_STANDARD 17
+ CXX_STANDARD_REQUIRED ON
+ CUDA_STANDARD 17
+ CUDA_STANDARD_REQUIRED ON
  )
 
  install(

cpp/cmake/modules/ConfigureCUDA.cmake

@@ -1,5 +1,5 @@
 #=============================================================================
-# Copyright (c) 2018-2022, NVIDIA CORPORATION.
+# Copyright (c) 2018-2024, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -31,8 +31,8 @@ endif()
 list(APPEND CUML_CUDA_FLAGS -Xcompiler=-Wall,-Werror,-Wno-error=deprecated-declarations,-Wno-error=sign-compare)
 
 if(DISABLE_DEPRECATION_WARNINGS)
- list(APPEND CUML_CXX_FLAGS -Wno-deprecated-declarations)
- list(APPEND CUML_CUDA_FLAGS -Wno-deprecated-declarations -Xcompiler=-Wno-deprecated-declarations)
+ list(APPEND CUML_CXX_FLAGS -Wno-deprecated-declarations -DRAFT_HIDE_DEPRECATION_WARNINGS)
+ list(APPEND CUML_CUDA_FLAGS -Wno-deprecated-declarations -Xcompiler=-Wno-deprecated-declarations -DRAFT_HIDE_DEPRECATION_WARNINGS)
 endif()
 
 # make sure we produce smallest binary size

cpp/src/arima/batched_arima.cu

@@ -84,7 +84,7 @@ struct is_missing {
  typedef T argument_type;
  typedef T result_type;
 
- __thrust_exec_check_disable__ __device__ const T operator()(const T& x) const { return isnan(x); }
+ __device__ const T operator()(const T& x) const { return isnan(x); }
 }; // end is_missing
 
 bool detect_missing(raft::handle_t& handle, const double* d_y, int n_elem)

cpp/src/hdbscan/condensed_hierarchy.cu

@@ -26,6 +26,7 @@
 #include <rmm/exec_policy.hpp>
 
 #include <cub/cub.cuh>
+#include <cuda/functional>
 #include <thrust/copy.h>
 #include <thrust/device_ptr.h>
 #include <thrust/execution_policy.h>
@@ -156,7 +157,7 @@ void CondensedHierarchy<value_idx, value_t>::condense(value_idx* full_parents,
  thrust::cuda::par.on(stream),
  full_sizes,
  full_sizes + size,
- [=] __device__(value_idx a) { return a != -1; },
+ cuda::proclaim_return_type<bool>([=] __device__(value_idx a) -> bool { return a != -1; }),
  0,
  thrust::plus<value_idx>());
 

cpp/src/hdbscan/detail/select.cuh

@@ -216,13 +216,14 @@ void excess_of_mass(const raft::handle_t& handle,
  value_t subtree_stability = 0.0;
 
  if (indptr_h[node + 1] - indptr_h[node] > 0) {
- subtree_stability = thrust::transform_reduce(
- exec_policy,
- children + indptr_h[node],
- children + indptr_h[node + 1],
- [=] __device__(value_idx a) { return stability[a]; },
- 0.0,
- thrust::plus<value_t>());
+ subtree_stability =
+ thrust::transform_reduce(exec_policy,
+ children + indptr_h[node],
+ children + indptr_h[node + 1],
+ cuda::proclaim_return_type<value_t>(
+ [=] __device__(value_idx a) -> value_t { return stability[a]; }),
+ 0.0,
+ thrust::plus<value_t>());
  }
 
  if (subtree_stability > node_stability || cluster_sizes_h[node] > max_cluster_size) {

cpp/src/hdbscan/detail/utils.h

@@ -114,7 +114,7 @@ Common::CondensedHierarchy<value_idx, value_t> make_cluster_tree(
  thrust_policy,
  sizes,
  sizes + condensed_tree.get_n_edges(),
- [=] __device__(value_idx a) { return a > 1; },
+ cuda::proclaim_return_type<bool>([=] __device__(value_idx a) -> bool { return a > 1; }),
  0,
  thrust::plus<value_idx>());
 

cpp/src/tsne/distances.cuh

@@ -33,6 +33,7 @@
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
 
+#include <cuda/functional>
 #include <thrust/functional.h>
 #include <thrust/transform_reduce.h>
 
@@ -162,7 +163,8 @@ void get_distances(const raft::handle_t& handle,
 template <typename value_t>
 void normalize_distances(value_t* distances, const size_t total_nn, cudaStream_t stream)
 {
- auto abs_f = [] __device__(const value_t& x) { return abs(x); };
+ auto abs_f = cuda::proclaim_return_type<value_t>(
+ [] __device__(const value_t& x) -> value_t { return abs(x); });
  value_t maxNorm = thrust::transform_reduce(rmm::exec_policy(stream),
  distances,
  distances + total_nn,

cpp/src/tsne/utils.cuh

@@ -39,6 +39,7 @@
 #include <sys/time.h>
 #include <unistd.h>
 
+#include <cfloat>
 #include <chrono>
 #include <iostream>
 

cpp/test/CMakeLists.txt

@@ -85,6 +85,10 @@ function(ConfigureTest)
  set_target_properties(
  ${_CUML_TEST_NAME}
  PROPERTIES INSTALL_RPATH "\$ORIGIN/../../../lib"
+ CXX_STANDARD 17
+ CXX_STANDARD_REQUIRED ON
+ CUDA_STANDARD 17
+ CUDA_STANDARD_REQUIRED ON
  )
 
  set(_CUML_TEST_COMPONENT_NAME testing)