diff --git a/.docker/Dockerfile b/.docker/Dockerfile index a06385207..59b91db9e 100644 --- a/.docker/Dockerfile +++ b/.docker/Dockerfile @@ -17,7 +17,8 @@ RUN add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/r RUN apt-get update && apt-get upgrade -y && \ DEBIAN_FRONTEND=noninteractive apt-get install -y openjdk-11-jdk \ build-essential checkinstall zlib1g-dev libssl-dev git libpthread-stubs0-dev \ - libcudnn8=${cudnn_version}-1+${cuda_version} libcudnn8-dev=${cudnn_version}-1+${cuda_version} + libcudnn8=${cudnn_version}-1+${cuda_version} libcudnn8-dev=${cudnn_version}-1+${cuda_version} \ + cuda-drivers #RUN apt-get install libcudnn8-samples=${cudnn_version}-1+${cuda_version} diff --git a/.jenkins/linux-x86_64-docker-cuda-build.jenkinsfile b/.jenkins/linux-x86_64-docker-cuda-build.jenkinsfile index 25085b1ad..1fd08178d 100644 --- a/.jenkins/linux-x86_64-docker-cuda-build.jenkinsfile +++ b/.jenkins/linux-x86_64-docker-cuda-build.jenkinsfile @@ -57,7 +57,7 @@ pipeline { } } stage('test-linux-cuda') { - agent { + /* agent { dockerfile { filename 'Dockerfile' dir '.docker' @@ -66,11 +66,12 @@ pipeline { args '--gpus all' //--needed for test only, you can build without GPU } } + */ environment { MAVEN = credentials('Internal_Archiva') OSSRH = credentials('OSSRH') } - +/* steps { withGradle { sh 'sh ./gradlew test --stacktrace -PexcludeTests=\'long-running,performance\' -Pskip-native=true -PCAVIS_CHIP=cuda \ @@ -79,6 +80,8 @@ pipeline { } //stash includes: '/cavis-native/cavis-native-lib/build/lib/*.jar', name: 'cuda-build' } + + */ } } diff --git a/cavis-common-platform/build.gradle b/cavis-common-platform/build.gradle index 8904ba793..5b64c61b2 100644 --- a/cavis-common-platform/build.gradle +++ b/cavis-common-platform/build.gradle @@ -112,7 +112,7 @@ dependencies { api "org.bytedeco:hdf5:${hdf5.version}-${javacpp.presetsVersion}:${javacppPlatform}" api "org.bytedeco:cuda:${cuda.version}-${cudnn.version}-${javacpp.presetsVersion}" - api "org.bytedeco:cuda-platform-redist:${cuda.version}-${cudnn.version}-${javacpp.presetsVersion}" + //api "org.bytedeco:cuda-platform-redist:${cuda.version}-${cudnn.version}-${javacpp.presetsVersion}" api "org.bytedeco:mkl:${mkl.version}-${javacpp.presetsVersion}" api "org.bytedeco:tensorflow:${tensorflow.version}-1.5.8" //not available for javacpp 1.5.9 ? api "org.bytedeco:tensorflow-platform:${tensorflow.version}-1.5.8" diff --git a/cavis-native/cavis-native-cuda-presets/src/main/java/org/nd4j/nativeblas/cuda/Nd4jCudaPresets.java b/cavis-native/cavis-native-cuda-presets/src/main/java/org/nd4j/nativeblas/cuda/Nd4jCudaPresets.java index 4784a2ac2..9e9f2541f 100644 --- a/cavis-native/cavis-native-cuda-presets/src/main/java/org/nd4j/nativeblas/cuda/Nd4jCudaPresets.java +++ b/cavis-native/cavis-native-cuda-presets/src/main/java/org/nd4j/nativeblas/cuda/Nd4jCudaPresets.java @@ -130,13 +130,13 @@ import java.util.List; compiler = {"cpp11", "nowarnings"}, library = "jnind4jcuda", link = "nd4jcuda", - preload = "libnd4jcuda"), + preload = "nd4jcuda"), @Platform(value = "linux", preload = "gomp@.1", preloadpath = {"/lib64/", "/lib/", "/usr/lib64/", "/usr/lib/", "/usr/local/cuda/lib64"}), @Platform(value = "linux-armhf", preloadpath = {"/usr/arm-linux-gnueabihf/lib/", "/usr/lib/arm-linux-gnueabihf/"}), @Platform(value = "linux-arm64", preloadpath = {"/usr/aarch64-linux-gnu/lib/", "/usr/lib/aarch64-linux-gnu/"}), @Platform(value = "linux-ppc64", preloadpath = {"/usr/powerpc64-linux-gnu/lib/", "/usr/powerpc64le-linux-gnu/lib/", "/usr/lib/powerpc64-linux-gnu/", "/usr/lib/powerpc64le-linux-gnu/"}), - @Platform(value = "windows", preload = {"libwinpthread-1", "libgcc_s_seh-1", "libgomp-1", "libstdc++-6", "libnd4jcuda"}) + @Platform(value = "windows", preload = {"libwinpthread-1", "libgcc_s_seh-1", "libgomp-1", "libstdc++-6", "nd4jcuda"}) }) public class Nd4jCudaPresets implements LoadEnabled, InfoMapper { @@ -170,9 +170,9 @@ public class Nd4jCudaPresets implements LoadEnabled, InfoMapper { preloads.add(i++, lib); } } - if (i > 0) { + //if (i > 0) { resources.add("/org/bytedeco/cuda/"); - } + //} } @Override diff --git a/cavis-native/cavis-native-lib/CMakeLists.txt b/cavis-native/cavis-native-lib/CMakeLists.txt index 8ca0b94ca..5339e1ab4 100644 --- a/cavis-native/cavis-native-lib/CMakeLists.txt +++ b/cavis-native/cavis-native-lib/CMakeLists.txt @@ -28,6 +28,9 @@ set(CMAKE_CXX_STANDARD 14) set(CMAKE_THREAD_PREFER_PTHREAD TRUE) find_package(Threads REQUIRED) +# MSVC runtime lib can be either "MultiThreaded" or "MultiThreadedDLL", /MT and /MD respectively +set(CMAKE_MSVC_RUNTIME_LIBRARY "MultiThreaded") + #/////////////////////////////////////////////////////////////////////////////// # genCompilation: Generates cpp, cu files # INPUT: @@ -139,8 +142,7 @@ else() set(DEFAULT_ENGINE "samediff::ENGINE_CPU") endif() -# MSVC runtime lib can be either "MultiThreaded" or "MultiThreadedDLL", /MT and /MD respectively -set(CMAKE_MSVC_RUNTIME_LIBRARY "MultiThreadedDLL") + set(SD_X86_BUILD false) diff --git a/cavis-native/cavis-native-lib/src/main/cpp/blas/CMakeLists.txt b/cavis-native/cavis-native-lib/src/main/cpp/blas/CMakeLists.txt index de942d441..c804420e8 100644 --- a/cavis-native/cavis-native-lib/src/main/cpp/blas/CMakeLists.txt +++ b/cavis-native/cavis-native-lib/src/main/cpp/blas/CMakeLists.txt @@ -174,6 +174,7 @@ if(SD_CUDA) add_definitions(-D_HAS_AUTO_PTR_ETC=1) set(CMAKE_CUDA_RUNTIME_LIBRARY "shared") + set(CMAKE_CUDA_ARCHITECTURES "62;75") #This basically kills instrinsic activated through SD_F16C=true #if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC") @@ -191,9 +192,7 @@ if(SD_CUDA) set(EXPM " -D__ND4J_EXPERIMENTAL__=true") endif() - if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES) - set(CMAKE_CUDA_ARCHITECTURES 75) - endif() + # the only difference for debug mode here is host/device debug symbols set(CMAKE_CUDA_FLAGS_DEBUG " -G -g")