Add support for CUDA 10.2

alpaka-group · Nov 27, 2019 · ce1dc4e · ce1dc4e
1 parent 219d09d
commit ce1dc4e
Show file tree

Hide file tree

Showing 7 changed files with 61 additions and 260 deletions.
diff --git a/.travis.yml b/.travis.yml
diff --git a/README.md b/README.md
@@ -56,7 +56,7 @@ Accelerator Back-ends
 | std::thread | std::thread |Host CPU (multi core)|sequential|parallel (preemptive multitasking)|
 | Boost.Fiber | boost::fibers::fiber |Host CPU (single core)|sequential|parallel (cooperative multitasking)|
 |TBB|TBB 2.2+|Host CPU (multi core)|parallel (preemptive multitasking)|sequential (only 1 thread per block)|
-|CUDA|CUDA 8.0-10.0|NVIDIA GPUs|parallel (undefined)|parallel (lock-step within warps)|
+|CUDA|CUDA 8.0-10.2|NVIDIA GPUs|parallel (undefined)|parallel (lock-step within warps)|
 |HIP(nvcc)|[HIP 1.5+](https://github.com/ROCm-Developer-Tools/HIP)|NVIDIA GPUs SM 2.0+|parallel (undefined)|parallel (lock-step within warps)|
 
 
@@ -74,7 +74,7 @@ This library uses C++11 (or newer when available).
 | std::thread |:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|
 | Boost.Fiber |:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:x:|:white_check_mark:|
 |TBB|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|:white_check_mark:|
-|CUDA (nvcc)|:white_check_mark: <br/> (CUDA 8.0-10.1)|:white_check_mark: <br/> (CUDA 9.0-10.1)|:white_check_mark: <br/> (CUDA 9.2-10.1) |:x:|:white_check_mark: <br/> (CUDA 9.1-10.1)|:white_check_mark: <br/> (CUDA 10.1)|:white_check_mark: <br/> (CUDA 10.1)|:white_check_mark: <br/> (CUDA 10.1)|:white_check_mark: <br/> (CUDA 10.1)|:x:|:x:|:white_check_mark: <br/> (CUDA 10.0-10.1)|
+|CUDA (nvcc)|:white_check_mark: <br/> (CUDA 8.0-10.2)|:white_check_mark: <br/> (CUDA 9.0-10.2)|:white_check_mark: <br/> (CUDA 9.2-10.2) |:x:|:white_check_mark: <br/> (CUDA 9.1-10.2)|:white_check_mark: <br/> (CUDA 10.1-10.2)|:white_check_mark: <br/> (CUDA 10.1-10.2)|:white_check_mark: <br/> (CUDA 10.1-10.2)|:white_check_mark: <br/> (CUDA 10.1)|:x:|:x:|:white_check_mark: <br/> (CUDA 10.0-10.2)|
 |CUDA (clang) | - | - | - | - | :white_check_mark: <br/> (CUDA 8.0)| :white_check_mark: <br/> (CUDA 8.0)| :white_check_mark: <br/> (CUDA 8.0-9.0) | :white_check_mark: <br/> (CUDA 8.0-9.2) | :white_check_mark: <br/> (CUDA 8.0-10.0) | :white_check_mark: <br/> (CUDA 9.2-10.1) | - | - |
 |[HIP](doc/markdown/user/implementation/mapping/HIP.md) (nvcc)|:white_check_mark: <br/> (nvcc 9.0+)|:x:|:x:|:x:|:x:|:x:|:x:|:x:|:x:|:x:|:x:|:x:|
 

diff --git a/alpakaConfig.cmake b/alpakaConfig.cmake
@@ -466,6 +466,10 @@ IF(ALPAKA_ACC_GPU_CUDA_ENABLE)
                     IF(CUDA_VERSION GREATER_EQUAL 10.1)
                         MESSAGE(FATAL_ERROR "Clang versions lower than 9 do not support CUDA 10.1 or greater!")
                     ENDIF()
+                ELSEIF(CMAKE_CXX_COMPILER_VERSION LESS 10.0)
+                    IF(CUDA_VERSION GREATER_EQUAL 10.2)
+                        MESSAGE(FATAL_ERROR "Clang versions lower than 10 do not support CUDA 10.2 or greater!")
+                    ENDIF()
                 ENDIF()
 
                 IF(ALPAKA_ACC_CPU_B_SEQ_T_FIBERS_ENABLE)
@@ -545,6 +549,10 @@ IF(ALPAKA_ACC_GPU_CUDA_ENABLE)
                         IF(CMAKE_CXX_COMPILER_VERSION GREATER_EQUAL 9.0)
                             MESSAGE(FATAL_ERROR "NVCC 10.1 does not support GCC 9+. Please use GCC 4.9, 5, 6, 7 or 8!")
                         ENDIF()
+                    ELSEIF(CUDA_VERSION VERSION_EQUAL 10.2)
+                        IF(CMAKE_CXX_COMPILER_VERSION GREATER_EQUAL 9.0)
+                            MESSAGE(FATAL_ERROR "NVCC 10.2 does not support GCC 9+. Please use GCC 4.9, 5, 6, 7 or 8!")
+                        ENDIF()
                     ENDIF()
                 ELSEIF("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
                     IF(CUDA_VERSION VERSION_EQUAL 8.0)
@@ -571,6 +579,10 @@ IF(ALPAKA_ACC_GPU_CUDA_ENABLE)
                         IF(CMAKE_CXX_COMPILER_VERSION GREATER_EQUAL 9.0)
                             MESSAGE(FATAL_ERROR "NVCC 10.1 does not support clang 9+. Please use clang 4, 5, 6, 7 or 8!")
                         ENDIF()
+                    ELSEIF(CUDA_VERSION VERSION_EQUAL 10.2)
+                        IF(CMAKE_CXX_COMPILER_VERSION GREATER_EQUAL 8.0)
+                            MESSAGE(FATAL_ERROR "NVCC 10.2 does not support clang 9+ and fails compiling with clang 8. Please use clang 4, 5, 6 or 7!")
+                        ENDIF()
                     ENDIF()
                 ENDIF()
 
@@ -678,15 +690,9 @@ IF(ALPAKA_ACC_GPU_CUDA_ENABLE)
                 # avoids warnings on host-device signature of 'std::__shared_count<>'
                 IF(CUDA_VERSION EQUAL 10.0)
                     LIST(APPEND CUDA_NVCC_FLAGS -Xcudafe --diag_suppress=2905)
-                ENDIF()
-
-                # avoids warnings on host-device signature of 'std::__shared_count<>'
-                IF(CUDA_VERSION EQUAL 10.1)
+                ELSEIF(CUDA_VERSION EQUAL 10.1)
                     LIST(APPEND CUDA_NVCC_FLAGS -Xcudafe --diag_suppress=2912)
-                ENDIF()
-
-                # avoids warnings on host-device signature of 'std::__shared_count<>'
-                IF(CUDA_VERSION EQUAL 10.2)
+                ELSEIF(CUDA_VERSION EQUAL 10.2)
                     LIST(APPEND CUDA_NVCC_FLAGS -Xcudafe --diag_suppress=2976)
                 ENDIF()
 

diff --git a/cmake/dev.cmake b/cmake/dev.cmake
@@ -144,3 +144,12 @@ ELSE()
     ENDIF()
   ENDIF()
 ENDIF()
+
+# nvcc
+IF(ALPAKA_ACC_GPU_CUDA_ENABLE)
+    IF(NOT ALPAKA_CUDA_COMPILER MATCHES "clang")
+        IF(CUDA_VERSION GREATER_EQUAL 10.2)
+            #LIST(APPEND CUDA_NVCC_FLAGS -Werror=all-warnings)
+        ENDIF()
+    ENDIF()
+ENDIF()
diff --git a/script/travis/install_cuda.sh b/script/travis/install_cuda.sh
@@ -60,8 +60,13 @@ then
         ALPAKA_CUDA_PKG_DEB_NAME=cuda-repo-ubuntu1804-10-1-local
         ALPAKA_CUDA_PKG_FILE_NAME="${ALPAKA_CUDA_PKG_DEB_NAME}"-10.1.168-418.67_1.0-1_amd64.deb
         ALPAKA_CUDA_PKG_FILE_PATH=https://developer.nvidia.com/compute/cuda/10.1/Prod/local_installers/${ALPAKA_CUDA_PKG_FILE_NAME}
+    elif [ "${ALPAKA_CUDA_VERSION}" == "10.2" ]
+    then
+        ALPAKA_CUDA_PKG_DEB_NAME=cuda-repo-ubuntu1804-10-2-local
+        ALPAKA_CUDA_PKG_FILE_NAME="${ALPAKA_CUDA_PKG_DEB_NAME}"-10.2.89-440.33.01_1.0-1_amd64.deb
+        ALPAKA_CUDA_PKG_FILE_PATH=http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/${ALPAKA_CUDA_PKG_FILE_NAME}
     else
-        echo CUDA versions other than 8.0, 9.0, 9.1, 9.2, 10.0 and 10.1 are not currently supported on linux!
+        echo CUDA versions other than 8.0, 9.0, 9.1, 9.2, 10.0, 10.1 and 10.2 are not currently supported on linux!
     fi
     if [ -z "$(ls -A ${ALPAKA_CI_CUDA_DIR})" ]
     then
@@ -96,8 +101,12 @@ then
     then
         ALPAKA_CUDA_PKG_FILE_NAME=cuda_10.1.168_425.25_win10.exe
         ALPAKA_CUDA_PKG_FILE_PATH=https://developer.nvidia.com/compute/cuda/10.1/Prod/local_installers/${ALPAKA_CUDA_PKG_FILE_NAME}
+    elif [ "${ALPAKA_CUDA_VERSION}" == "10.2" ]
+    then
+        ALPAKA_CUDA_PKG_FILE_NAME=cuda_10.2.89_441.22_win10.exe
+        ALPAKA_CUDA_PKG_FILE_PATH=http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/${ALPAKA_CUDA_PKG_FILE_NAME}
     else
-        echo CUDA versions other than 10.0 and 10.1 are not currently supported on Windows!
+        echo CUDA versions other than 10.0, 10.1 and 10.2 are not currently supported on Windows!
     fi
 
     curl -L -o cuda_installer.exe ${ALPAKA_CUDA_PKG_FILE_PATH}

diff --git a/test/unit/CMakeLists.txt b/test/unit/CMakeLists.txt
@@ -14,8 +14,6 @@
 
 CMAKE_MINIMUM_REQUIRED(VERSION 3.11.0)
 
-PROJECT("alpakaUnitTest")
-
 ################################################################################
 # Add subdirectories.
 ################################################################################

diff --git a/test/unit/queue/src/CollectiveQueue.cpp b/test/unit/queue/src/CollectiveQueue.cpp
@@ -25,7 +25,7 @@
 struct QueueCollectiveTestKernel
 {
     template<typename TAcc>
-    ALPAKA_FN_ACC auto operator()(
+    auto operator()(
         TAcc const & acc,
         int* resultsPtr) const
     -> void