Require Kokkos 4.0

.gitlab-ci.yml

@@ -30,7 +30,7 @@ BuildKokkos:
   extends: .LoadModules
   stage: buildDependencies
   script:
-    - git clone --depth=1 --branch 3.7.01 https://github.com/kokkos/kokkos.git &&
+    - git clone --depth=1 --branch 4.0.00 https://github.com/kokkos/kokkos.git &&
       cd kokkos &&
       mkdir build &&
       cd build &&

.jenkins/continuous.groovy

@@ -42,7 +42,7 @@ pipeline {
                         dockerfile {
                             filename "Dockerfile"
                             dir "docker"
-                            additionalBuildArgs '--build-arg BASE=nvidia/cuda:11.5.2-devel-ubuntu20.04 --build-arg KOKKOS_VERSION=3.7.01 --build-arg KOKKOS_OPTIONS="-DCMAKE_CXX_EXTENSIONS=OFF -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ARCH_VOLTA70=ON" --build-arg CUDA_AWARE_MPI=1'
+                            additionalBuildArgs '--build-arg BASE=nvidia/cuda:11.5.2-devel-ubuntu20.04 --build-arg KOKKOS_VERSION=4.1.00 --build-arg KOKKOS_OPTIONS="-DCMAKE_CXX_EXTENSIONS=OFF -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ARCH_VOLTA70=ON" --build-arg CUDA_AWARE_MPI=1'
                             args '-v /tmp/ccache:/tmp/ccache --env NVIDIA_VISIBLE_DEVICES=${NVIDIA_VISIBLE_DEVICES}'
                             label 'NVIDIA_Tesla_V100-PCIE-32GB && nvidia-docker'
                         }
@@ -161,7 +161,7 @@ pipeline {
                         dockerfile {
                             filename "Dockerfile"
                             dir "docker"
-                            additionalBuildArgs '--build-arg BASE=nvidia/cuda:11.0.3-devel-ubuntu18.04 --build-arg KOKKOS_VERSION="3.7.01" --build-arg KOKKOS_OPTIONS="-DCMAKE_CXX_EXTENSIONS=OFF -DCMAKE_CXX_COMPILER=clang++ -DKokkos_ENABLE_THREADS=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ARCH_VOLTA70=ON -DCMAKE_PREFIX_PATH=/usr/lib/x86_64-linux-gnu"'
+                            additionalBuildArgs '--build-arg BASE=nvidia/cuda:11.0.3-devel-ubuntu18.04 --build-arg KOKKOS_VERSION="4.1.00" --build-arg KOKKOS_OPTIONS="-DCMAKE_CXX_EXTENSIONS=OFF -DCMAKE_CXX_COMPILER=clang++ -DKokkos_ENABLE_THREADS=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ARCH_VOLTA70=ON -DCMAKE_PREFIX_PATH=/usr/lib/x86_64-linux-gnu"'
                             args '-v /tmp/ccache:/tmp/ccache --env NVIDIA_VISIBLE_DEVICES=${NVIDIA_VISIBLE_DEVICES}'
                             label 'NVIDIA_Tesla_V100-PCIE-32GB && nvidia-docker'
                         }

CMakeLists.txt

@@ -1,13 +1,7 @@
 cmake_minimum_required(VERSION 3.16)
 project(ArborX CXX)
 
-# Workaround to compile against Trilinos 14.0. That version
-# of Trilinos does not set the compatibility mode for Kokkos correctly.
-# FIXME: remove once we require Kokkos 4.0.1 or later.
-find_package(Kokkos QUIET 3.7.01 CONFIG)
-if (NOT Kokkos_FOUND)
-  find_package(Kokkos 4.0.00 REQUIRED CONFIG)
-endif()
+find_package(Kokkos 4.0 REQUIRED CONFIG)
 message(STATUS "Found Kokkos: ${Kokkos_DIR} (version \"${Kokkos_VERSION}\")")
 if(Kokkos_ENABLE_CUDA AND Kokkos_VERSION VERSION_LESS 4.1)
   # CUDA_LAMBDA is deprecated starting 4.1

benchmarks/bvh_driver/bvh_driver.cpp

@@ -107,8 +107,7 @@ void register_bvh_benchmarks(Spec const &spec)
 
 #ifdef KOKKOS_ENABLE_HIP
   if (spec.backends == "all" || spec.backends == "hip")
-    BVHBenchmarkRegistration<Kokkos::Experimental::HIP>(spec,
-                                                        "ArborX::BVH<HIP>");
+    BVHBenchmarkRegistration<Kokkos::HIP>(spec, "ArborX::BVH<HIP>");
 #else
   if (spec.backends == "hip")
     throw std::runtime_error("HIP backend not available!");

benchmarks/distributed_tree_driver/distributed_tree_driver.cpp

@@ -602,8 +602,7 @@ int main(int argc, char *argv[])
     if (node == "hip")
     {
 #ifdef KOKKOS_ENABLE_HIP
-      using Node = Kokkos::Device<Kokkos::Experimental::HIP,
-                                  Kokkos::Experimental::HIPSpace>;
+      using Node = Kokkos::Device<Kokkos::HIP, Kokkos::HIPSpace>;
       main_<Node>(pass_further, comm);
 #else
       throw std::runtime_error("HIP node type is disabled");

src/geometry/ArborX_Box.hpp

@@ -18,9 +18,7 @@
 #include <ArborX_Point.hpp>
 
 #include <Kokkos_Macros.hpp>
-#if __has_include(<Kokkos_ReductionIdentity.hpp>) // FIXME requires Kokkos 4.0
 #include <Kokkos_ReductionIdentity.hpp>
-#endif
 
 namespace ArborX
 {

src/geometry/ArborX_HyperBox.hpp

@@ -18,9 +18,7 @@
 #include <ArborX_HyperPoint.hpp>
 
 #include <Kokkos_Macros.hpp>
-#if __has_include(<Kokkos_ReductionIdentity.hpp>) // FIXME requires Kokkos 4.0
 #include <Kokkos_ReductionIdentity.hpp>
-#endif
 
 namespace ArborX::ExperimentalHyperGeometry
 {

src/kokkos_ext/ArborX_DetailsKokkosExtSort.hpp

@@ -116,7 +116,7 @@ void sortByKey(
 #if defined(KOKKOS_ENABLE_CUDA)
     Kokkos::Cuda const &space,
 #else
-    Kokkos::Experimental::HIP const &space,
+    Kokkos::HIP const &space,
 #endif
     Keys &keys, Values &values)
 {

test/CMakeLists.txt

@@ -13,7 +13,7 @@ if(Kokkos_ENABLE_THREADS)
   list(APPEND ARBORX_DEVICE_TYPES Kokkos::Threads::device_type)
 endif()
 if(Kokkos_ENABLE_HIP)
-  list(APPEND ARBORX_DEVICE_TYPES Kokkos::Experimental::HIPSpace::device_type)
+  list(APPEND ARBORX_DEVICE_TYPES Kokkos::HIPSpace::device_type)
 endif()
 if(Kokkos_ENABLE_OPENMPTARGET)
   list(APPEND ARBORX_DEVICE_TYPES Kokkos::Experimental::OpenMPTarget::device_type)

test/tstUnionFind.cpp

@@ -37,18 +37,7 @@ build_representatives(ExecutionSpace const &space, UnionFind union_find)
       "Test::find_representatives",
       Kokkos::RangePolicy<ExecutionSpace>(space, 0, n), KOKKOS_LAMBDA(int i) {
         auto r = union_find.representative(i);
-#if KOKKOS_VERSION >= 30799
         Kokkos::atomic_min(&map2smallest(r), i);
-#else
-         // Workaround for undefined
-        //   desul::atomic_min(int*, int, desul::MemoryOrderRelaxed,
-        //   desul::MemoryScopeDevice)
-        // in older Kokkos versions.
-        auto v = map2smallest(r);
-        while (v > i) {
-          v = Kokkos::atomic_compare_exchange(&map2smallest(r), v, i);
-        }
-#endif
         representatives(i) = r;
       });
   // We want the representative values to not depend on a specific