Reorganising build.gradle for CUDA 12

Signed-off-by: brian <brian@brutex.de>

.docker/Dockerfile

@@ -1,15 +1,37 @@
-FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04
+FROM nvidia/cuda:12.2.0-devel-ubuntu20.04
 
-RUN apt-get update &&  \
-    DEBIAN_FRONTEND=noninteractive apt-get install -y openjdk-11-jdk wget build-essential checkinstall zlib1g-dev libssl-dev git
+ENV OS=ubuntu2004
+ENV cudnn_version=8.9.4.25
+ENV cuda_version=cuda12.2
+ENV CMAKE_VER=3.27.4
+
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y wget software-properties-common
+
+RUN wget https://developer.download.nvidia.com/compute/cuda/repos/${OS}/x86_64/cuda-${OS}.pin
+
+RUN mv cuda-${OS}.pin /etc/apt/preferences.d/cuda-repository-pin-600
+RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/${OS}/x86_64/7fa2af80.pub
+RUN add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/${OS}/x86_64/ /"
+
+
+RUN apt-get update && apt-get upgrade -y &&  \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y openjdk-11-jdk \
+    build-essential checkinstall zlib1g-dev libssl-dev git libpthread-stubs0-dev \
+    libcudnn8=${cudnn_version}-1+${cuda_version} libcudnn8-dev=${cudnn_version}-1+${cuda_version} \
+    cuda-drivers
+
+
+#RUN apt-get install libcudnn8-samples=${cudnn_version}-1+${cuda_version}
     #Build cmake version from source \
 #RUN wget https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2.tar.gz && \
 #    tar -xvf cmake-3.24.2.tar.gz && cd cmake-3.24.2 && \
 #    ./bootstrap && make && make install
-RUN wget -nv https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2-linux-x86_64.sh && \
-    mkdir /opt/cmake && sh ./cmake-3.24.2-linux-x86_64.sh --skip-license --prefix=/opt/cmake && ln -s /opt/cmake/bin/cmake /usr/bin/cmake && \
-    rm cmake-3.24.2-linux-x86_64.sh
+RUN wget -nv https://github.com/Kitware/CMake/releases/download/v${CMAKE_VER}/cmake-${CMAKE_VER}-linux-x86_64.sh && \
+    mkdir -p /opt/cmake && sh ./cmake-${CMAKE_VER}-linux-x86_64.sh --skip-license --prefix=/opt/cmake && ln -s /opt/cmake/bin/cmake /usr/bin/cmake && \
+    rm cmake-${CMAKE_VER}-linux-x86_64.sh
+RUN ln -s /usr/bin/make /usr/bin/gmake
 
 
-RUN echo "nameserver 8.8.8.8" >> /etc/resolv.conf
+#RUN echo "nameserver 8.8.8.8" >> /etc/resolv.conf
+#RUN echo "nameserver 9.9.9.9" >> /etc/resolv.conf
 

.gitea/workflows/build-windows.yaml

@@ -0,0 +1,69 @@
+name: Gitea Actions Demo
+run-name: ${{ gitea.actor }} is testing out Gitea Actions 🚀
+on: [push]
+
+env:
+  OS: windows
+  cudnn_version: 8.9.4.25
+  cuda_version: cuda12.2
+  CMAKE_VER: 3.27.4
+
+jobs:
+  Explore-Gitea-Actions:
+    runs-on: windows
+    #container:
+    defaults:
+      run:
+        shell: msys2 {0}
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v3
+
+      - name: Install MSYS2
+        uses: msys2/setup-msys2@v2
+        with:
+            msystem: UCRT64
+            update: true
+            install: git tar gzip mingw-w64-ucrt-x86_64-gcc
+
+     # - run: Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy Unrestricted
+     # - name: Check for CUDA
+     #   run: |
+     #     echo "Path: $env:PATH"
+
+     # - name: Install CUDA
+     #   uses: Jimver/cuda-toolkit@v0.2.11
+     #   id: cuda-toolkit
+     #   with:
+     #     cuda: '12.2.0'
+     # - run: nvcc -V
+
+      - name: Install CMake and Ninja
+        uses: lukka/get-cmake@dev/fix91
+        with:
+          useLocalCache: false
+          useCloudCache: false
+          #cmakeVersion: "~3.27.0"
+          cmakeVersion: latest
+          ninjaVersion: latest
+
+      - name: Execute Gradle build
+        run: |
+          cmd.exe /C ./gradlew.bat build \
+          --stacktrace \
+          -Pmavenuser=${{ secrets.MAVENUSER }} \
+          -Pmavenpass=${{ secrets.MAVENPASS }} \
+          -PossrhUsername=${{ secrets.OSSRHUSERNAME }} \
+          -PossrhPassword=${{ secrets.OSSRHPASSWORD }} \
+          -PCAVIS_CHIP=cpu,cuda -Pskip-native=false \
+          >buildlog.txt 2>&1 
+
+      - name: Upload log
+        uses: actions/upload-artifact@v3
+        if: success() || failure()    # run this step even if previous step failed
+        with:
+          name: Build-Log
+          path: buildlog.txt
+
+      - run: echo "This job's status is ${{ job.status }}."

.gitea/workflows/demo.yaml

@@ -0,0 +1,79 @@
+name: Gitea Actions Demo
+run-name: ${{ gitea.actor }} is testing out Gitea Actions 🚀
+on: [push]
+
+env:
+  OS: ubuntu2004
+  cudnn_version: 8.9.4.25
+  cuda_version: cuda12.2
+  CMAKE_VER: 3.27.4
+
+jobs:
+  Explore-Gitea-Actions:
+    runs-on: ubuntu-20.04:docker://nvidia/cuda:12.2.0-devel-ubuntu20.04
+    steps:
+      - run: echo "The job was automatically triggered by a ${{ gitea.event_name }} event."
+      - run: echo "This job is now running on a ${{ runner.os }} server hosted by Gitea!"
+      - run: echo "The name of your branch is ${{ gitea.ref }} and your repository is ${{ gitea.repository }}."
+
+      - name: Check out repository code
+        uses: actions/checkout@v3
+      - run: echo "💡 The ${{ gitea.repository }} repository has been cloned to the runner."
+      - run: echo "🖥️ The workflow is now ready to test your code on the runner."
+
+      - name: List files in the repository
+        run: |
+          ls ${{ gitea.workspace }}
+
+      - name: Update initial docker image with apt-get
+        run: |
+          apt-get -qq update && DEBIAN_FRONTEND=noninteractive apt-get -qq install -y wget software-properties-common && \
+
+          wget https://developer.download.nvidia.com/compute/cuda/repos/$OS/x86_64/cuda-$OS.pin && \
+          mv cuda-$OS.pin /etc/apt/preferences.d/cuda-repository-pin-600 && \
+          apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/$OS/x86_64/7fa2af80.pub && \
+          apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/$OS/x86_64/3bf863cc.pub && \
+          add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/$OS/x86_64/ /" && \
+
+          apt-get -qq update && apt-get -qq upgrade -y &&  \
+              DEBIAN_FRONTEND=noninteractive apt-get -qq install -y \
+              build-essential checkinstall zlib1g-dev libssl-dev git libpthread-stubs0-dev \
+              libcudnn8=$cudnn_version-1+$cuda_version libcudnn8-dev=$cudnn_version-1+$cuda_version \
+              libblas{3,-dev} liblapack{3,-dev} libatlas-base-dev libopenblas-dev && \
+          wget -q https://developer.download.nvidia.com/compute/cuda/12.2.1/local_installers/cuda_12.2.1_535.86.10_linux.run && \
+          sh cuda_12.2.1_535.86.10_linux.run --silent --toolkit
+
+      - name: Setup Java
+        uses: actions/setup-java@v3
+        with:
+          distribution: 'temurin' # See 'Supported distributions' for available options
+          java-version: '11'
+          cache: 'gradle'
+
+      - name: Install CMake and Ninja
+        uses: lukka/get-cmake@latest
+        with:
+          useLocalCache: true
+          useCloudCache: false
+          cmakeVersion: "~3.27.0"
+          ninjaVersion: latest
+
+      - name: Execute Gradle build
+        run: |
+          sh ./gradlew build \
+          --stacktrace \
+          -Pmavenuser=${{ secrets.MAVENUSER }} \
+          -Pmavenpass=${{ secrets.MAVENPASS }} \
+          -PossrhUsername=${{ secrets.OSSRHUSERNAME }} \
+          -PossrhPassword=${{ secrets.OSSRHPASSWORD }} \
+          -PCAVIS_CHIP=cpu,cuda -Pskip-native=false \
+          >buildlog.log 2>&1 
+
+      - name: Upload log
+        uses: actions/upload-artifact@v3
+        if: success() || failure()    # run this step even if previous step failed
+        with:
+          name: my-artifact
+          path: buildlog.log
+
+      - run: echo "This job's status is ${{ job.status }}."

.gitignore

@@ -96,3 +96,4 @@ bruai4j-native-common/cmake*
 /cavis-dnn/cavis-dnn-core/build/resources/test/logback-test.xml
 /cavis-dnn/cavis-dnn-core/build/test-results/cudaTest/TEST-org.deeplearning4j.gradientcheck.AttentionLayerTest.xml
 /cavis-dnn/cavis-dnn-core/build/tmp/jar/MANIFEST.MF
+/.metadata/

.jenkins/linux-x86_64-docker-cuda-build.jenkinsfile

@@ -26,7 +26,7 @@ pipeline {
             dir '.docker'
             label 'linux && docker && cuda'
             //additionalBuildArgs  '--build-arg version=1.0.2'
-            //args '--gpus all' --needed for test only, you can build without GPU
+            //args '--gpus all' //--needed for test only, you can build without GPU
         }
     }
 
@@ -57,19 +57,31 @@ pipeline {
             }
         }
         stage('test-linux-cuda') {
+           /* agent {
+                dockerfile {
+                    filename 'Dockerfile'
+                    dir '.docker'
+                    label 'linux && docker && cuda && cudart'
+                    //additionalBuildArgs  '--build-arg version=1.0.2'
+                    args '--gpus all' //--needed for test only, you can build without GPU
+                }
+            }
+            */
             environment {
                 MAVEN = credentials('Internal_Archiva')
                 OSSRH = credentials('OSSRH')
             }
 
-            steps {
+            steps {/*
                 withGradle {
                     sh 'sh ./gradlew test --stacktrace -PexcludeTests=\'long-running,performance\' -Pskip-native=true -PCAVIS_CHIP=cuda \
                                 -Pmavenuser=$MAVEN_USR -Pmavenpass=$MAVEN_PSW \
                                 -PossrhUsername=$OSSRH_USR -PossrhPassword=$OSSRH_PSW'
                 }
                 //stash includes: '/cavis-native/cavis-native-lib/build/lib/*.jar', name: 'cuda-build'
+                */
             }
+
         }
     }
 

brutex-extended-tests/src/main/java/net/brutex/ai/LoaderTest.java

@@ -19,7 +19,12 @@
  *
  */
 
-package org.nd4j.nativeblas;
+package net.brutex.ai;
 
-public class Dummy {
+public class LoaderTest {
+
+    public static void main(String[] args){
+        System.load("C:\\Users\\brian\\_projects\\deeplearning4j\\cavis-native\\cavis-native-lib"
+                + "\\build\\generated\\sources\\javacpp\\cuda\\windows-x86_64-avx2\\jnind4jcuda.dll");
+    }
 }

brutex-extended-tests/src/test/java/net/brutex/ai/nd4j/tests/LoadBackendTests.java

@@ -22,7 +22,9 @@
 package net.brutex.ai.nd4j.tests;
 
 import lombok.extern.slf4j.Slf4j;
+import org.bytedeco.javacpp.Loader;
 import org.junit.jupiter.api.Test;
+import org.nd4j.linalg.api.environment.Nd4jEnvironment;
 import org.nd4j.linalg.api.ndarray.INDArray;
 import org.nd4j.linalg.factory.Nd4j;
 
@@ -37,12 +39,7 @@ public class LoadBackendTests {
 
     @Test
     public void loadBackend() throws ClassNotFoundException, NoSuchFieldException, IllegalAccessException {
-        // check if Nd4j is there
-        //Logger.getLogger(LoadBackendTests.class.getName()).info("System java.library.path: " + System.getProperty("java.library.path"));
-        final Field sysPathsField = ClassLoader.class.getDeclaredField("sys_paths");
-        sysPathsField.setAccessible(true);
-        sysPathsField.set(null, null);
-        //System.loadLibrary("jnind4jcpu");
+        log.info("get number of GPUs {}", Nd4jEnvironment.getEnvironment().getNumGpus());
         log.info("Backend: {}", Nd4j.getBackend().buildInfo());
         double d1 = 2.0;
         double d2 = 5.0;
@@ -52,4 +49,10 @@ public class LoadBackendTests {
         Number n = res.sumNumber();
         assertEquals(n.doubleValue(), 7.0, String.format("Addition of two scalar values %g and %g", d1, d2));
     }
+
+    @Test
+    public void loadCudaDLL() {
+    System.load(
+        "C:\\Users\\brian\\_projects\\deeplearning4j\\cavis-native\\cavis-native-lib\\build\\generated\\sources\\javacpp\\cuda\\windows-x86_64-avx2\\jnind4jcuda.dll");
+    }
 }

build.gradle

@@ -23,7 +23,7 @@ buildscript {
         mavenCentral()
     }
     dependencies {
-        classpath "com.vanniktech:gradle-dependency-graph-generator-plugin:0.6.0"
+        classpath "com.vanniktech:gradle-dependency-graph-generator-plugin:0.8.0"
         classpath 'com.google.gradle:osdetector-gradle-plugin:1.7.0'
     }
 }

cavis-common-platform/build.gradle

@@ -48,9 +48,9 @@ dependencies {
 
 
     constraints {
-        api enforcedPlatform("io.netty:netty-bom:${netty.version}")
-        api enforcedPlatform("com.fasterxml.jackson:jackson-bom:${jackson.version}")
-        api enforcedPlatform("com.squareup.okhttp3:okhttp-bom:${okhttp3.version}")
+        api ("io.netty:netty-bom:${netty.version}")
+        api ("com.fasterxml.jackson:jackson-bom:${jackson.version}")
+        api ("com.squareup.okhttp3:okhttp-bom:${okhttp3.version}")
 
         api 'com.google.guava:guava:30.1-jre'
         api "com.google.protobuf:protobuf-java:3.15.6"
@@ -105,18 +105,19 @@ dependencies {
         api "org.bytedeco:javacv:${javacv.version}"
         api "org.bytedeco:opencv:${opencv.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:openblas:${openblas.version}-${javacpp.presetsVersion}"
+        api "org.bytedeco:openblas-platform:${openblas.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:leptonica-platform:${leptonica.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:hdf5-platform:${hdf5.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:hdf5:${hdf5.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:hdf5:${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:${tensorflow.version}-1.5.8"
-        api "org.bytedeco:tensorflow-lite:${tensorflow_lite.version}-${javacpp.presetsVersion}"
-        api "org.bytedeco:tensorflow-lite-platform:${tensorflow_lite.version}-${javacpp.presetsVersion}"
+        //api "org.bytedeco:tensorflow:${tensorflow.version}-1.5.8" //not available for javacpp 1.5.9 ?
+        //api "org.bytedeco:tensorflow-platform:${tensorflow.version}-1.5.8"
+        //api "org.bytedeco:tensorflow-lite:${tensorflow_lite.version}-${javacpp.presetsVersion}"
+        //api "org.bytedeco:tensorflow-lite-platform:${tensorflow_lite.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:tensorrt:${tensorrt.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:tensorrt-platform:${tensorrt.version}-${javacpp.presetsVersion}"
         api "org.bytedeco:cpython:${cpython.version}-${javacpp.presetsVersion}:${javacppPlatform}"

cavis-native/cavis-native-cpu-presets/build.gradle

@@ -0,0 +1,10 @@
+plugins {
+    id 'java-library'
+}
+
+dependencies {
+    implementation platform(projects.cavisCommonPlatform)
+    implementation projects.cavisNative.cavisNativeBlas
+    implementation "org.bytedeco:javacpp"
+    implementation group: "org.bytedeco", name: 'openblas-platform'
+}

cavis-native/cavis-native-cpu-presets/src/main/resources/META-INF/services/org.nd4j.linalg.factory.Nd4jBackend

@@ -19,5 +19,4 @@
 #
 #
 
-#org.nd4j.linalg.jcublas.JCublasBackend
 org.nd4j.linalg.cpu.nativecpu.CpuBackend

cavis-native/cavis-native-cpu/build.gradle

@@ -13,7 +13,7 @@ dependencies {
     implementation projects.cavisNative.cavisNativeCommon
     implementation projects.cavisDnn.cavisDnnApi
     implementation projects.cavisDnn.cavisDnnCommon
-
+    implementation projects.cavisNative.cavisNativeCpuPresets
 
     implementation (projects.cavisNative.cavisNativeLib) {
         capabilities {

cavis-native/cavis-native-cuda-presets/build.gradle

@@ -0,0 +1,9 @@
+plugins {
+    id 'java-library'
+}
+
+dependencies {
+    implementation platform(projects.cavisCommonPlatform)
+    implementation projects.cavisNative.cavisNativeBlas
+    implementation "org.bytedeco:javacpp"
+}

cavis-native/cavis-native-cuda-presets/src/main/java/org/nd4j/nativeblas/cuda/Nd4jCudaPresets.java

@@ -37,7 +37,10 @@ import java.util.List;
  * @author saudet
  */
 @Properties(target = "org.nd4j.nativeblas.Nd4jCuda", helper = "org.nd4j.nativeblas.cuda.Nd4jCudaHelper",
-                value = {@Platform(define = "LIBND4J_ALL_OPS", include = {
+                value = {
+        @Platform(
+                define = "LIBND4J_ALL_OPS",
+                include = {
                         "array/DataType.h",
                         "array/DataBuffer.h",
                         "array/PointerDeallocator.h",
@@ -125,12 +128,16 @@ import java.util.List;
                                 "cnpy/cnpy.h"
                         },
                                 compiler = {"cpp11", "nowarnings"},
-                                library = "jnind4jcuda", link = "nd4jcuda", preload = "nd4jcuda"),
-                @Platform(value = "linux", preload = "gomp@.1", preloadpath = {"/lib64/", "/lib/", "/usr/lib64/", "/usr/lib/"}),
+                                library = "jnind4jcuda",
+                                link = {"nd4jcuda"}),
+                                //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", "nd4jcuda"}) })
+                @Platform(value = "windows", preload = {"libwinpthread-1", "libgcc_s_seh-1", "libgomp-1", "libstdc++-6"})
+      })
 public class Nd4jCudaPresets implements LoadEnabled, InfoMapper {
 
     @Override public void init(ClassProperties properties) {
@@ -143,14 +150,19 @@ public class Nd4jCudaPresets implements LoadEnabled, InfoMapper {
             return;
         }
         int i = 0;
+        /*
         String[] libs = {"cudart", "cublasLt", "cublas", "curand", "cusolver", "cusparse", "cudnn",
                          "cudnn_ops_infer", "cudnn_ops_train", "cudnn_adv_infer",
                          "cudnn_adv_train", "cudnn_cnn_infer", "cudnn_cnn_train"};
+
+         */
+        // test no preload
+        String[] libs = {};
         for (String lib : libs) {
             if (platform.startsWith("linux")) {
-                lib += lib.startsWith("cudnn") ? "@.8" : lib.equals("curand") ? "@.10" : lib.equals("cudart") ? "@.11.0" : "@.11";
+                lib += lib.startsWith("cudnn") ? "@.8" : lib.equals("curand") ? "@.10" : lib.equals("cufft") ? "@.11" : "@.12";
             } else if (platform.startsWith("windows")) {
-                lib += lib.startsWith("cudnn") ? "64_8" : lib.equals("curand") ? "64_10" : lib.equals("cudart") ? "64_110" : "64_11";
+                lib += lib.startsWith("cudnn") ? "64_8" : lib.equals("cufft") ? "64_11" : lib.equals("cusolver") ? "64_11" : lib.equals("curand") ? "64_10" : "64_12";
             } else {
                 continue; // no CUDA
             }
@@ -158,9 +170,9 @@ public class Nd4jCudaPresets implements LoadEnabled, InfoMapper {
                 preloads.add(i++, lib);
             }
         }
-        if (i > 0) {
+        //if (i > 0) {
             resources.add("/org/bytedeco/cuda/");
-        }
+        //}
     }
 
     @Override

cavis-native/cavis-native-cuda-presets/src/main/resources/META-INF/services/org.nd4j.linalg.compression.NDArrayCompressor

@@ -0,0 +1,23 @@
+#
+#
+#    ******************************************************************************
+#    *
+#    * This program and the accompanying materials are made available under the
+#    * terms of the Apache License, Version 2.0 which is available at
+#    * https://www.apache.org/licenses/LICENSE-2.0.
+#    *
+#    *  See the NOTICE file distributed with this work for additional
+#    *  information regarding copyright ownership.
+#    * Unless required by applicable law or agreed to in writing, software
+#    * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+#    * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+#    * License for the specific language governing permissions and limitations
+#    * under the License.
+#    *
+#    * SPDX-License-Identifier: Apache-2.0
+#    *****************************************************************************
+#
+#
+
+
+org.nd4j.linalg.cpu.nativecpu.compression.CpuThreshold

cavis-native/cavis-native-cuda-presets/src/main/resources/META-INF/services/org.nd4j.linalg.factory.Nd4jBackend

@@ -0,0 +1,23 @@
+#
+#
+#    ******************************************************************************
+#    *
+#    * This program and the accompanying materials are made available under the
+#    * terms of the Apache License, Version 2.0 which is available at
+#    * https://www.apache.org/licenses/LICENSE-2.0.
+#    *
+#    *  See the NOTICE file distributed with this work for additional
+#    *  information regarding copyright ownership.
+#    * Unless required by applicable law or agreed to in writing, software
+#    * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+#    * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+#    * License for the specific language governing permissions and limitations
+#    * under the License.
+#    *
+#    * SPDX-License-Identifier: Apache-2.0
+#    *****************************************************************************
+#
+#
+
+org.nd4j.linalg.jcublas.JCublasBackend
+#org.nd4j.linalg.cpu.nativecpu.CpuBackend

cavis-native/cavis-native-cuda-presets/src/main/resources/function_threads.properties

@@ -0,0 +1,21 @@
+#
+# /* ******************************************************************************
+#  *
+#  *
+#  * This program and the accompanying materials are made available under the
+#  * terms of the Apache License, Version 2.0 which is available at
+#  * https://www.apache.org/licenses/LICENSE-2.0.
+#  *
+#  *  See the NOTICE file distributed with this work for additional
+#  *  information regarding copyright ownership.
+#  * Unless required by applicable law or agreed to in writing, software
+#  * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+#  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+#  * License for the specific language governing permissions and limitations
+#  * under the License.
+#  *
+#  * SPDX-License-Identifier: Apache-2.0
+#  ******************************************************************************/
+#
+
+iamax_strided = 1

cavis-native/cavis-native-cuda-presets/src/main/resources/native.properties

@@ -0,0 +1,22 @@
+#
+# /* ******************************************************************************
+#  *
+#  *
+#  * This program and the accompanying materials are made available under the
+#  * terms of the Apache License, Version 2.0 which is available at
+#  * https://www.apache.org/licenses/LICENSE-2.0.
+#  *
+#  *  See the NOTICE file distributed with this work for additional
+#  *  information regarding copyright ownership.
+#  * Unless required by applicable law or agreed to in writing, software
+#  * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+#  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+#  * License for the specific language governing permissions and limitations
+#  * under the License.
+#  *
+#  * SPDX-License-Identifier: Apache-2.0
+#  ******************************************************************************/
+#
+
+org.nd4j.linalg.api.resources.maxallocated= 2000000000
+org.nd4j.linalg.api.resources.memoryratio=0.5

cavis-native/cavis-native-jcublas/build.gradle

@@ -10,22 +10,22 @@ ext {
 
 dependencies {
     implementation platform(projects.cavisCommonPlatform)
-
-    //implementation project(":cavis-native:cavis-native-blas")
     implementation projects.cavisNative.cavisNativeBlas
 
     implementation group: "org.bytedeco", name: "cuda"
     implementation group: "org.bytedeco", name: "cuda", classifier: buildTarget
-    implementation group: "org.bytedeco", name: "cuda", classifier: "${buildTarget}-redist"
+    //implementation group: "org.bytedeco", name: "cuda", classifier: "${buildTarget}-redist"
 
     implementation group: "org.bytedeco", name: "javacpp"
     implementation group: "org.bytedeco", name: "javacpp", classifier: buildTarget
 
-    implementation(project(path: ":cavis-native:cavis-native-lib")) {
+    implementation projects.cavisNative.cavisNativeCudaPresets
+    implementation(project(":cavis-native:cavis-native-lib")) {
         capabilities {
-            it.requireCapability(group: "net.brutex.cavis.cavis-native", name: "cavis-native-lib-cuda-support", version:project.version)
+            requireCapability("${project.group}:cavis-native-lib-cuda-support:${project.version}")
         }
     }
+
     implementation project(":cavis-native:cavis-native-common")
     implementation project(":cavis-dnn:cavis-dnn-api")
     implementation project(":cavis-dnn:cavis-dnn-common")
@@ -36,3 +36,9 @@ dependencies {
     implementation "org.apache.commons:commons-lang3"
 }
 
+tasks.named("compileJava").configure {
+    dependsOn ":cavis-native:cavis-native-lib:javacppCudaSupportBuildParser",
+            ":cavis-native:cavis-native-lib:cudaJar"
+
+}
+

cavis-native/cavis-native-jcublas/src/main/java/org/nd4j/jita/constant/CudaConstantHandler.java

@@ -20,7 +20,6 @@
 
 package org.nd4j.jita.constant;
 
-import lombok.extern.log4j.Log4j2;
 import lombok.extern.slf4j.Slf4j;
 import org.nd4j.linalg.api.buffer.DataBuffer;
 import org.nd4j.linalg.api.buffer.DataType;

cavis-native/cavis-native-lib/CMakeLists.txt

@@ -3,13 +3,14 @@ cmake_minimum_required(VERSION 3.20)
 
 project(libnd4j)
 set(CMAKE_VERBOSE_MAKEFILE ON)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 
 set (CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}")
 message("CMAKE MODULE PATH IS ${CMAKE_MODULE_PATH}")
 
 #ensure we create lib files
-set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS OFF)
+#set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS OFF)
 
 
 option(SD_NATIVE "Optimize for build machine (might not work on others)" OFF)
@@ -25,6 +26,12 @@ set(FLATBUFFERS_BUILD_FLATC "OFF" CACHE STRING "Hack to disable flatc build" FOR
 
 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:
@@ -120,8 +127,8 @@ endfunction()
 
 
 if (SD_CUDA)
-    #enable_language(CUDA)
-     find_package(CUDAToolkit 11.4 REQUIRED)
+       find_package(CUDAToolkit 12.2 REQUIRED)
+       enable_language(CUDA)
        message(STATUS "CUDAToolkit_VERSION: ${CUDAToolkit_VERSION}")
        message(STATUS "CUDAToolkit_VERSION_MAJOR: ${CUDAToolkit_VERSION_MAJOR}")
        message(STATUS "CUDAToolkit_VERSION_MINOR: ${CUDAToolkit_VERSION_MINOR}")
@@ -136,8 +143,7 @@ else()
     set(DEFAULT_ENGINE "samediff::ENGINE_CPU")
 endif()
 
-# MSVC runtime lib can be either "MultiThreaded" or "MultiThreadedDLL", /MT and /MD respectively
-#set(MSVC_RT_LIB "MultiThreadedDLL")
+
 
 set(SD_X86_BUILD false)
 
@@ -155,10 +161,10 @@ elseif (APPLE)
     set(CMAKE_CXX_FLAGS_RELEASE  "-O3 -fPIC -Wno-braced-scalar-init -Wno-delete-non-virtual-dtor -Wno-unused-command-line-argument -Wno-dangling-else -D__APPLE_OS__=true -D_RELEASE=true")
     set(CMAKE_CXX_FLAGS_DEBUG  " -O0 -g -fPIC -Wno-braced-scalar-init -Wno-delete-non-virtual-dtor -Wno-unused-command-line-argument -Wno-dangling-else -D__APPLE_OS__=true")
 elseif(WIN32)
-    set(SD_X86_BUILD true)
+    set(SD_X86_BUILD false)
     if (SD_CUDA)
         set(CMAKE_CXX_FLAGS_RELEASE  "-D_RELEASE=true")
-        set(CMAKE_CXX_FLAGS_DEBUG  "  /FS /EHsc")
+        #set(CMAKE_CXX_FLAGS_DEBUG  "  /FS /EHsc")
     else()
         set(CMAKE_CXX_FLAGS_RELEASE  "-O3 -fPIC -D_RELEASE=true")
         set(CMAKE_CXX_FLAGS_DEBUG  " -g -O2 -fPIC")
@@ -362,7 +368,7 @@ if(SD_BUILD_TESTS)
     # tests are always compiled with all ops included
     set(SD_ALL_OPS true)
     set(SD_BUILD_MINIFIER true)
-    add_subdirectory(tests_cpu)
+    add_subdirectory(src/test/cpp/tests_cpu)
 endif()
 
 
@@ -370,7 +376,6 @@ if (MSVC_DEV)
     set(SD_BUILD_MINIFIER false)
 endif ()
 
-set (CMAKE_INSTALL_PREFIX $ENV{ND4J_HOME}/bruai4j-native/bruai4j-native-common/src/main/resources)
 
 # Set package information
 set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "Native operations for nd4j.")

cavis-native/cavis-native-lib/build.gradle

@@ -1,3 +1,5 @@
+import org.gradle.api.internal.java.DefaultJavaPlatformExtension
+import org.gradle.api.plugins.internal.DefaultJavaPluginExtension
 import org.gradle.api.publish.maven.internal.publisher.MavenRemotePublisher
 import org.gradle.language.nativeplatform.internal.Dimensions
 
@@ -44,8 +46,7 @@ buildscript {
             logger.info("Setting properties for task '{}' to '{}'", tsk.getName(), pf)
             return pf
         }
-
-    }
+    } // End of ext block
 
 
     dependencies {
@@ -64,104 +65,106 @@ buildscript {
 
 plugins {
     id 'java-library'
-    id 'org.bytedeco.gradle-javacpp-build' version "1.5.9"
+    id 'org.bytedeco.gradle-javacpp-build' version "1.5.9" //version "1.5.10-SNAPSHOT"
     id 'maven-publish'
     id 'signing'
 }
 
-chipList.each {thisChip ->
-    sourceSets.register("${thisChip}Support") {
+chipList.each {String thisChip ->
+    /*sourceSets.register(thisChip) {
         java {
-            srcDirs = ['src/main/java', "${buildDir}/generated/sources/javacpp/${thisChip}//${javacppPlatform}${javacppPlatformExtension}/"]
+            srcDirs = ["${projectDir}/src/main/java/"]
             include "org/nd4j/nativeblas/${thisChip}/Nd4j${thisChip.capitalize()}Helper.java"
             include "org/nd4j/nativeblas/${thisChip}/Nd4j${thisChip.capitalize()}Presets.java"
+        }
+    }*/
+    sourceSets.register("${thisChip}").configure {
+        java {
+            srcDirs = ["${buildDir}/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/"]
             include "org/nd4j/nativeblas/Nd4j${thisChip.capitalize()}.java"
         }
-        it.compiledBy("javacpp${thisChip.capitalize()}SupportBuildCommand",
-                "javacpp${thisChip.capitalize()}SupportBuildCompiler")
+        compiledBy "javacpp${thisChip.capitalize()}SupportBuildCompiler"
     }
 }
 
-
-//if(osdetector.os.startsWith("windows")) {
+/*
 sourceSets {
     main {
         java {
-                srcDirs = ['src/main/java']
+            srcDirs = new HashSet<>();
             include 'org/nd4j/nativeblas/Dummy.java'
         }
     }
 }
-//}
+
+ */
 
 
+
+// This block registers the cpu and cuda features and creates
+// i. e. the {chip}Implementation
 java {
     chipList.each {thisChip ->
         registerFeature("${thisChip}Support") {
-            usingSourceSet(sourceSets.findByName("${thisChip}Support"))
+            usingSourceSet(sourceSets.findByName("${thisChip}"))
             capability(project.group, "cavis-native-lib-${thisChip}-support", project.version)
             //withJavadocJar()
             //withSourcesJar()
-        }
-    }
-}
-
+        }}}
 
 dependencies {
-    api platform(project(':cavis-common-platform'))
-    implementation "org.bytedeco:javacpp"
-    implementation group: "org.bytedeco", name: "javacpp", classifier: "${javacppPlatform}"
-
     if(withCuda()) {
-        cudaSupportImplementation platform(project(':cavis-common-platform'))
-        cudaSupportImplementation project(":cavis-dnn:cavis-dnn-api")
-        cudaSupportImplementation project(":cavis-dnn:cavis-dnn-common")
-        cudaSupportImplementation project(":cavis-native:cavis-native-blas")
-        cudaSupportImplementation project(":cavis-native:cavis-native-common")
-        cudaSupportImplementation "commons-io:commons-io"
-        cudaSupportImplementation group: "org.bytedeco", name: "openblas"
-        cudaSupportImplementation group: "org.bytedeco", name: "openblas", classifier: "${javacppPlatform}"
-        cudaSupportImplementation group: "org.bytedeco", name: "cuda"
-        cudaSupportImplementation group: "org.bytedeco", name: "cuda", classifier: "${javacppPlatform}"
-        cudaSupportImplementation "org.apache.logging.log4j:log4j-core:2.17.0"
-        cudaSupportImplementation "com.google.guava:guava:14.0.1"
-        cudaSupportImplementation "org.apache.commons:commons-lang3"
-        cudaSupportImplementation "org.apache.commons:commons-math3"
-        cudaSupportImplementation "com.google.flatbuffers:flatbuffers-java"
-        cudaSupportImplementation 'javax.mail:javax.mail-api:1.6.2'
+        cudaImplementation platform(project(':cavis-common-platform'))
+
+        //cudaImplementation project(":cavis-dnn:cavis-dnn-api")
+        //cudaImplementation project(":cavis-dnn:cavis-dnn-common")
+        cudaImplementation project(":cavis-native:cavis-native-blas")
+        //cudaImplementation project(":cavis-native:cavis-native-common")
+        //cudaImplementation "commons-io:commons-io"
+        //cudaImplementation "org.bytedeco:openblas"
+        //cudaImplementation "org.bytedeco:openblas::${javacppPlatform}"
+        //cudaImplementation "org.bytedeco:cuda"
+        //cudaImplementation "org.bytedeco:cuda::${javacppPlatform}"
+        //cudaImplementation "org.apache.logging.log4j:log4j-core:2.17.0"
+        //cudaImplementation "com.google.guava:guava:14.0.1"
+        //cudaImplementation "org.apache.commons:commons-lang3"
+        //cudaImplementation "org.apache.commons:commons-math3"
+        //cudaImplementation "com.google.flatbuffers:flatbuffers-java"
+        //cudaImplementation 'javax.mail:javax.mail-api:1.6.2'
+        cudaImplementation "org.bytedeco:javacpp"
+        cudaImplementation "org.bytedeco:javacpp::${javacppPlatform}"
+        cudaImplementation project(":cavis-native:cavis-native-cuda-presets")
+
+        //cudaGeneratedImplementation platform(project(':cavis-common-platform'))
+        //cudaGeneratedImplementation project(":cavis-native:cavis-native-blas")
+        //cudaGeneratedImplementation "org.bytedeco:javacpp"
+        //cudaGeneratedImplementation "org.bytedeco:javacpp::${javacppPlatform}"
+        //cudaGeneratedImplementation project(":cavis-native:cavis-native-cuda-presets")
     }
 
     if(withCpu()) {
-        cpuSupportImplementation platform(project(':cavis-common-platform'))
-        cpuSupportImplementation project(":cavis-dnn:cavis-dnn-api")
-        cpuSupportImplementation project(":cavis-dnn:cavis-dnn-common")
-        cpuSupportImplementation project(":cavis-native:cavis-native-blas")
-        cpuSupportImplementation project(":cavis-native:cavis-native-common")
-        cpuSupportImplementation "commons-io:commons-io"
-        cpuSupportImplementation group: "org.bytedeco", name: "openblas"
-        cpuSupportImplementation group: "org.bytedeco", name: "openblas", classifier: "${javacppPlatform}"
-        cpuSupportImplementation group: "org.bytedeco", name: "opencv"
-        cpuSupportImplementation group: "org.bytedeco", name: "opencv", classifier: "${javacppPlatform}"
-        cpuSupportImplementation "org.apache.logging.log4j:log4j-core:2.17.0"
-        cpuSupportImplementation "com.google.guava:guava:14.0.1"
-        cpuSupportImplementation "org.apache.commons:commons-lang3"
-        cpuSupportImplementation "org.apache.commons:commons-math3"
-        cpuSupportImplementation "com.google.flatbuffers:flatbuffers-java"
-        cpuSupportImplementation 'javax.mail:javax.mail-api:1.6.2'
-    }
+        cpuImplementation platform(project(':cavis-common-platform'))
+        //cpuImplementation project(":cavis-dnn:cavis-dnn-api")
+        //cpuImplementation project(":cavis-dnn:cavis-dnn-common")
+        cpuImplementation project(":cavis-native:cavis-native-blas")
+        //cpuImplementation project(":cavis-native:cavis-native-common")
+        //cpuImplementation "commons-io:commons-io"
+        //cpuImplementation "org.bytedeco:opencv"
+        //cpuImplementation "org.bytedeco:opencv::${javacppPlatform}"
+        //cpuImplementation "org.apache.logging.log4j:log4j-core:2.17.0"
+        //cpuImplementation "com.google.guava:guava:14.0.1"
+        //cpuImplementation "org.apache.commons:commons-lang3"
+        //cpuImplementation "org.apache.commons:commons-math3"
+        //cpuImplementation "com.google.flatbuffers:flatbuffers-java"
+        //cpuImplementation 'javax.mail:javax.mail-api:1.6.2'
+        cpuImplementation "org.bytedeco:javacpp"
+        cpuImplementation "org.bytedeco:javacpp::${javacppPlatform}"
+        // https://mvnrepository.com/artifact/org.bytedeco/openblas
+        cpuImplementation 'org.bytedeco:openblas:0.3.23-1.5.9'
 
-    implementation projects.cavisDnn.cavisDnnApi
-    implementation projects.cavisDnn.cavisDnnCommon
-    implementation project(":cavis-native:cavis-native-blas")
-    implementation project(":cavis-native:cavis-native-common")
-    implementation "commons-io:commons-io"
-    implementation "org.bytedeco:openblas"
-    implementation group: "org.bytedeco", name: "openblas", classifier: "${javacppPlatform}"
-    implementation "org.apache.logging.log4j:log4j-core"
-    implementation "com.google.guava:guava:14.0.1"
-    implementation "org.apache.commons:commons-lang3"
-    implementation "org.apache.commons:commons-math3"
-    implementation "com.google.flatbuffers:flatbuffers-java"
+
+        cpuImplementation project(":cavis-native:cavis-native-cpu-presets")
+    }
 }
 
 
@@ -183,40 +186,34 @@ task deepClean(type: Delete) {
 }
 
 
-tasks.withType(org.bytedeco.gradle.javacpp.BuildTask) {
-    buildResource = [ "/org/bytedeco/openblas/${javacppPlatform}/",
+tasks.withType(org.bytedeco.gradle.javacpp.BuildTask).configureEach { org.bytedeco.gradle.javacpp.BuildTask it ->
+    /*
+    it.buildResource = ["/org/bytedeco/openblas/${javacppPlatform}/",
                         "/org/bytedeco/mkldnn/${javacppPlatform}/"]
 
-    includeResource = ["/org/bytedeco/openblas/${javacppPlatform}/include/"]
+    it.includeResource = ["/org/bytedeco/openblas/${javacppPlatform}/include/"]
 
-    linkResource = ["/org/bytedeco/openblas/${javacppPlatform}/",
+    it.linkResource = ["/org/bytedeco/openblas/${javacppPlatform}/",
                        "/org/bytedeco/openblas/${javacppPlatform}/lib/"]
 
-   //buildPath = [ org.bytedeco.javacpp.Loader.getCacheDir() ]
-
-
-
+     */
 }
 
 
 // Disable the standard javacpp generated tasks and use own
 // versions below. This allows to build for each variant
+
 [javacppBuildParser, javacppBuildCommand, javacppCompileJava, javacppBuildCompiler].each {
     it.enabled false
 }
 
-chipList.each { thisChip ->
+chipList.each { String thisChip ->
 
     // 1)
     //Run the C++ compile first
-    tasks.register("javacpp${thisChip.capitalize()}SupportBuildCommand", org.bytedeco.gradle.javacpp.BuildTask) {
-        if (project.hasProperty("skip-native") && project.getProperty("skip-native").equals("true")) {
-            enabled = false
-        }
-        dependsOn "processResources"
+    tasks.register("javacpp${thisChip.capitalize()}SupportBuildCommand", org.bytedeco.gradle.javacpp.BuildTask) {org.bytedeco.gradle.javacpp.BuildTask it ->
         properties = getBuildPlatform( thisChip, it )
 
-
         includePath = ["${projectDir}/src/main/cpp/blas/",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/src/main/include/",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/flatbuffers-src/include",
@@ -226,19 +223,16 @@ chipList.each { thisChip ->
         //No idea why this is here, but it looks like even for the javacppBuildCommand task,
         //there is a javacpp Loader actively determining platform etc.
         classOrPackageNames = ["org.nd4j.nativeblas.${thisChip}.Nd4j${thisChip.capitalize()}Presets"]
-        workingDirectory = projectDir
+        //workingDirectory = projectDir
         //if the classpath is not set here, the javacpp classloader starts to look around
         //everywhere and causes java.io.IOExceptions:  because files is being used by another process
-        classPath = [:]
-        classPath += ["${buildDir}/classes/java/${thisChip}Support/"]
-        //classPath += ["${buildDir}/classes/java/main/"]
+        //logger.quiet("Using compile classpath from configuration named '{}'", sourceSets.named(thisChip).get().getCompileClasspathConfigurationName())
+
+        classPath = sourceSets.named(thisChip).get().compileClasspath.collect()
 
         /* Get VCVARS in case we want to build CUDA
         * MinGW64 g++ on MSYS is used otherwise */
-        if (thisChip.equals('cuda') && osdetector.os.startsWith("win")
-        && project.hasProperty("skip-native")
-                && !project.getProperty("skip-native").equals("true")
-                && !VISUAL_STUDIO_INSTALL_DIR.isEmpty()) {
+        if (thisChip.equals('cuda') && osdetector.os.startsWith("win") && !VISUAL_STUDIO_INSTALL_DIR.isEmpty()) {
             def proc = ["cmd.exe", "/c", "${VISUAL_STUDIO_VCVARS_CMD} > nul && set"].execute()
             it.environmentVariables = it.environmentVariables ?: [:]
             def lines = proc.text.split("\\r?\\n")
@@ -246,14 +240,15 @@ chipList.each { thisChip ->
                 if (line.contains("=")) {
                     def parts = line.split("=")
                     it.environmentVariables.put(parts[0], parts[1])
+                    logger.debug("Added variable to environment: {} = {}", parts[0], parts[1])
                 }
             }
         }
-
+        workingDirectory projectDir
         if (thisChip.equals('cuda') && osdetector.os.startsWith("windows")) { //cuDNN requires CUDA
             it.buildCommand = ['sh', 'buildnativeoperations.sh',
                             '-V',
-                            '--build-type', 'release',
+                            '--build-type', 'debug',
                             '--chip', thisChip,
                             '--plattform', 'x86_64',
                             '--chip-extension', avxExtension,
@@ -280,24 +275,13 @@ chipList.each { thisChip ->
                             '-j', "${host_cores}",
                             '--helper', 'mkldnn']
         }
-    }
-
-
-    //Create a task to (pre)compile the java presets (required for javacppBuildParser)
-    tasks.register("compile${thisChip.capitalize()}Support", JavaCompile) {
-        def thisSS = sourceSets.findByName("${thisChip}Support")
-        it.source = thisSS.allSource
-        it.classpath = thisSS.compileClasspath
-        it.destinationDirectory = file("${buildDir}/classes/java/${thisChip}Support/")
+        if(project.hasProperty("nativeTests")) it.buildCommand += "--tests"
     }
 
     //Run the parser on the InfoMap in Nd4j$ChipPresets and listed header files in @Platform
     //Generates Nd4jCpu.java and/ or Nd4jCuda.java Java JNI code
     tasks.register("javacpp${thisChip.capitalize()}SupportBuildParser", org.bytedeco.gradle.javacpp.BuildTask) {
-        if (project.hasProperty("skip-native") && project.getProperty("skip-native").equals("true")) {
-            enabled = false
-        }
-        dependsOn "compile${thisChip.capitalize()}Support"
+
 
         includePath = ["${projectDir}/src/main/cpp/blas/",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/src/main/include/",
@@ -305,43 +289,35 @@ chipList.each { thisChip ->
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/cpu_features-src/include",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/mkldnn-src/include"]
 
-
-
         classOrPackageNames = ["org.nd4j.nativeblas.${thisChip}.Nd4j${thisChip.capitalize()}Presets"]
-        outputDirectory = file("${buildDir}/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/")
-
-        classPath = sourceSets.getByName("${thisChip}Support").getRuntimeClasspath()
-        classPath += ["${buildDir}/classes/java/${thisChip}Support/"]
+        classPath = sourceSets.named(thisChip).get().compileClasspath.collect()
+        outputDirectory file("${buildDir}/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/")
     }
 
 
     // Generates jnijavacpp.cpp and jniNativeLibrary.cpp, compiles and links it
-    tasks.register("javacpp${thisChip.capitalize()}SupportBuildCompiler", org.bytedeco.gradle.javacpp.BuildTask) {
-        if (project.hasProperty("skip-native") && project.getProperty("skip-native").equals("true")) {
-            enabled = false
-        }
-        def thisTask = (org.bytedeco.gradle.javacpp.BuildTask) it
-        thisTask.dependsOn = ["javacpp${thisChip.capitalize()}SupportBuildParser"]
+    tasks.register("javacpp${thisChip.capitalize()}SupportBuildCompiler", org.bytedeco.gradle.javacpp.BuildTask) {org.bytedeco.gradle.javacpp.BuildTask it ->
 
-        thisTask.linkPath = ["${projectDir}/blasbuild/${thisChip}/${avxExtension}/output"]
-        thisTask.includePath = ["${projectDir}/src/main/cpp/blas/",
+        linkPath = ["${projectDir}/blasbuild/${thisChip}/${avxExtension}/output"]
+        includePath = ["${projectDir}/src/main/cpp/blas/",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/src/main/include/",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/flatbuffers-src/include",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/cpu_features-src/include",
                        "${projectDir}/blasbuild/${thisChip}/${avxExtension}/mkldnn-src/include"]
 
-        thisTask.properties = getBuildPlatform( thisChip, thisTask )
+        properties = getBuildPlatform( thisChip, it )
 
-        if(thisChip.equals('cuda') && osdetector.os.startsWith("win") && project.hasProperty("skip-native")
-                && !project.getProperty("skip-native").equals("true") && !VISUAL_STUDIO_INSTALL_DIR.isEmpty()) {
+        if(thisChip.equals('cuda') && osdetector.os.startsWith("win") && !VISUAL_STUDIO_INSTALL_DIR.isEmpty()) {
             def proc = ["cmd.exe", "/c", "${VISUAL_STUDIO_VCVARS_CMD} > nul && where.exe cl.exe"].execute()
             def outp = proc.text
-            def cl = outp.replace("\\", "\\\\").trim()
+            def cl = "\"" + outp.replace("\\", "\\\\").trim() + "\""
             def currentCompiler = ""
             doFirst{
                 currentCompiler = System.getProperty("org.bytedeco.javacpp.platform.compiler")
                 System.setProperty("org.bytedeco.javacpp.platform.compiler", cl)
-                logger.quiet("Task ${thisTask.name} overrides compiler '${currentCompiler}' with '${cl}'.")
+                System.setProperty("platform.compiler.cpp11", cl)
+                logger.quiet("Task ${name} overrides compiler '${currentCompiler}' with '${cl}'.")
+
             }
             doLast {
                 //restore compiler
@@ -350,12 +326,12 @@ chipList.each { thisChip ->
             //System.setProperty("org.bytedeco.javacpp.platform.compiler.cpp11", cl)
 
             proc = ["cmd.exe", "/c", "${VISUAL_STUDIO_VCVARS_CMD} > nul && set"].execute()
-            thisTask.environmentVariables = thisTask.environmentVariables ?: [:]
+            environmentVariables = environmentVariables ?: [:]
             def lines = proc.text.split("\\r?\\n")
             for (def line in lines) {
                 if (line.contains("=")) {
                     def parts = line.split("=")
-                    thisTask.environmentVariables.put(parts[0], parts[1])
+                    environmentVariables.put(parts[0], parts[1])
                 }
             }
 
@@ -364,32 +340,27 @@ chipList.each { thisChip ->
         }
 
 
-        thisTask.buildPath = ["$buildDir/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/"]
-        thisTask.copyLibs = true
-        thisTask.deleteJniFiles(false)
-        outputName = "jnind4j${thisChip}"
-        thisTask.outputDirectory = file("$buildDir/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/")
-        thisTask.classOrPackageNames= ["org.nd4j.nativeblas.Nd4j${thisChip.capitalize()}"]
+        buildPath = ["${buildDir}/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/"]
+        copyLibs = true
+        deleteJniFiles(false)
+        //outputName = "jnind4j${thisChip}"
+        outputDirectory = file("${buildDir}/generated/sources/javacpp/${thisChip}/${javacppPlatform}${javacppPlatformExtension}/")
+        classOrPackageNames= ["org.nd4j.nativeblas.Nd4j${thisChip.capitalize()}"]
 
-        thisTask.configDirectory = file("${buildDir}/classes/java/${thisChip}Support/META-INF/native-image/${javacppPlatform}")
-
-        //Need to set the classpath, so that external jars from the dependency list are resolved by the ClassLoader as well
-        thisTask.classPath = [:]
-        thisTask.classPath = ["${buildDir}/classes/java/${thisChip}Support"]
-        thisTask.classPath += sourceSets.findByName("${thisChip}Support").runtimeClasspath
-        //sourceSets.findByName("${thisChip}Support").runtimeClasspath.each{ s ->
-        //    thisTask.classPath += s
-        //}
+        configDirectory = file("${buildDir}/classes/java/${thisChip}Support/META-INF/native-image/${javacppPlatform}")
+        classPath = sourceSets.named("${thisChip}").get().compileClasspath.collect()
+        classPath += "${buildDir}/classes/java/${thisChip}/"
     }
 
     // Create Jar with classifier
-    tasks.getByName("${thisChip}SupportJar") { Jar thisTask ->
+    tasks.named("${thisChip}Jar").configure { Jar thisTask ->
         dependsOn "javacpp${thisChip.capitalize()}SupportBuildCompiler"
         dependsOn "javacpp${thisChip.capitalize()}SupportBuildCommand"
 
-        //it.from sourceSets.getByName("${thisChip}Support").getOutput()
+
         def spec = copySpec {
-           from(tasks.getByName("javacpp${thisChip.capitalize()}SupportBuildCompiler")) {
+
+           from(tasks.named("javacpp${thisChip.capitalize()}SupportBuildCompiler").get()) {
                exclude { f ->
                    def exclude = f.file.isDirectory()
                    if(exclude) {
@@ -401,8 +372,8 @@ chipList.each { thisChip ->
                }
                into "${javacppPlatform}/" //path within jar, we need it in a platform, that javacpp Loader understands
            }
-            from(sourceSets.getByName("${thisChip}Support").getOutput()) {
-
+            from(sourceSets.named(thisChip).get().getOutput()) {
+                into "${javacppPlatform}/" //path within jar, we need it in a platform, that javacpp Loader understands
             }
             duplicatesStrategy DuplicatesStrategy.EXCLUDE
         }
@@ -414,34 +385,43 @@ chipList.each { thisChip ->
 
 //Before we can compile the whole java part, we
 //need to generate the Nd4jXXX.java files first
-chipList.each { thisChip ->
-    tasks.findByName("compile${thisChip.capitalize()}SupportJava").each { t ->
-        t.dependsOn "javacpp${thisChip.capitalize()}SupportBuildParser"
+tasks.named("compileJava").configure {enabled false}
+
+chipList.each { String thisChip ->
+    //ensure full build process is running on "build"
+    tasks.named("build").configure {
+        dependsOn "javacpp${thisChip.capitalize()}SupportBuildCompiler"
+    }
+    //Compiles and links the generated jni code with the underlying native library
+    tasks.named("javacpp${thisChip.capitalize()}SupportBuildCompiler").configure {
+        dependsOn "javacpp${thisChip.capitalize()}SupportBuildParser"
+    }
+    //Generates the jni interface sources
+    tasks.named("javacpp${thisChip.capitalize()}SupportBuildParser").configure {
+        dependsOn "javacpp${thisChip.capitalize()}SupportBuildCommand"
+    }
+    //Compiles the c++ and cuda sources
+    tasks.named("javacpp${thisChip.capitalize()}SupportBuildCommand").configure {
+
+    }
+    //Compile the generates jni interface (java portion)
+    tasks.named("compile${thisChip.capitalize()}Java").configure {
+        dependsOn "javacpp${thisChip.capitalize()}SupportBuildParser"
+    }
+
+    tasks.named("${thisChip}Jar").configure {
+        dependsOn "javacpp${thisChip.capitalize()}SupportBuildCompiler"
     }
 }
 
-tasks.withType(JavaCompile) {
+tasks.withType(JavaCompile).configureEach {
     // options.setCompilerArgs(Arrays.asList("-Xlint:unchecked"))
 }
 
-tasks.withType(Javadoc) {
+tasks.withType(Javadoc).configureEach {
     options.addStringOption('Xdoclint:none', '-quiet')
 }
 
-/*
-jar {
-    manifest {
-        attributes 'Class-Path': configurations.runtimeClasspath.collect { it.getName() }.join(' '),
-                'Implementation-Title': 'Brutex AI - Native Components',
-                'Implementation-Vendor': 'Brutex Network',
-                'Implementation-Version': archiveVersion,
-                'Specification-Title': 'Brutex AI - Native Components',
-                'Specification-Vendor': 'Brutex Network',
-                'Specification-Version': archiveVersion
-    }
-    //archiveClassifier = "${javacppPlatform}${javacppPlatformExtension}-${chip}"
-}
-*/
 javadoc {
     dependsOn "javacppPomProperties"
     failOnError = false
@@ -451,10 +431,6 @@ javadoc {
 }
 
 
-
-
-
-
     tasks.getByName("generatePomFileForMavenJavaPublication") {
         enabled = true
     }
@@ -464,32 +440,14 @@ javadoc {
 
 artifacts {
     //implementation(jar)
-    chipList.each { thisChip ->
-        implementation(tasks.getByName("${thisChip}SupportJar"))
-    }
+
+    chipList.each { String thisChip ->
+        implementation tasks.getByName("${thisChip}Jar")
     }
 
-/*
-artifacts {
-        archives jar
-        chipList.each { thisChip ->
-            archives tasks.getByName("${thisChip}SupportJar")
-        }
+
 }
 
- */
-/*
-publishing {
-    publications {
-        mavenJava(MavenPublication) {
-            artifact jar
-            chipList.each { thisChip ->
-                artifact tasks.getByName("${thisChip}SupportJar")
-            }
-        }
-    }
-}
-*/
 /*
 
 if( osdetector.os.startsWith("windows")) {
@@ -515,48 +473,6 @@ if( osdetector.os.startsWith("windows")) {
 }
 */
 
-task printDeps {
-    doLast {
-        configurations.apiElements.dependencies.each { dep ->
-            println "${dep.group} - ${dep.name} - ${dep.version}"
-            dep.artifacts.each { art ->
-                println "    ${art.extension} - ${art.classifier}"
-            }
-        }
-    }
-}
-
-
-/*
-def pomClosure = {
-    name = 'Brutex AI - Native Components'
-    delegate.description = 'Underlying native components for the Brutex AI deeplearning framework for Java'
-    url = 'https://ai.brutex.net'
-    licenses {
-        license {
-            name = 'Apache License, Version 2.0'
-            url = 'http://www.apache.org/licenses/LICENSE-2.0'
-            distribution = 'repo'
-        }
-    }
-    developers {
-        developer {
-            id = 'irnbrux'
-            name = 'Brian Rosenberger'
-            email = 'bru@brutex.de'
-        }
-    }
-    scm {
-        url = 'https://brutex.net/svn/'
-        connection = 'scm:svn:https://brutex.net/svn/bruai4j/'
-    }
-}
-*/
-
-//tasks.getByName("publishMavenJavaPublicationToOSSRHRepository") { MavenRemotePublisher pub ->
- //   logger.quiet(pub.dump());
-//}
-
 signing {
     useGpgCmd()
     if (!version.endsWith('SNAPSHOT')) {

cavis-native/cavis-native-lib/buildnativeoperations.sh

@@ -19,6 +19,7 @@
 #  ******************************************************************************/
 #
 
+
 #env
 
 set -eu

cavis-native/cavis-native-lib/src/main/cpp/blas/CMakeLists.txt

@@ -102,16 +102,18 @@ ENDIF()
 
 IF(${SD_EXTENSION} MATCHES "avx2")
     message("Extension AVX2 enabled.")
-    set(ARCH_TUNE "${ARCH_TUNE} -mmmx -msse -msse2 -msse3 -msse4.1 -msse4.2 -mavx -mavx2 -mfma -mf16c -mprefetchwt1 -DSD_F16C=true -DF_AVX2=true")
+    #-mmmx -msse -msse2 -msse3 -msse4.1 -msse4.2 -mavx -mavx2 -mfma -mf16c -mprefetchwt1
+    set(ARCH_TUNE "${ARCH_TUNE} -DSD_F16C=true -DF_AVX2=true")
 ELSEIF(${SD_EXTENSION} MATCHES "avx512")
         message("Extension AVX512 enabled.")
         # we need to set flag here, that we can use hardware f16 conversion + tell that cpu features should be tracked
-        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mmmx -msse -msse2 -msse3 -msse4.1 -msse4.2 -mavx -mavx2 -mfma -mf16c -mavx512f -mavx512vl -mavx512bw -mavx512dq  -mavx512cd -mbmi -mbmi2 -mprefetchwt1 -mclflushopt -mxsavec -mxsaves -DSD_F16C=true -DF_AVX512=true")
+        #-mmmx -msse -msse2 -msse3 -msse4.1 -msse4.2 -mavx -mavx2 -mfma -mf16c -mavx512f -mavx512vl -mavx512bw -mavx512dq  -mavx512cd -mbmi -mbmi2 -mprefetchwt1 -mclflushopt -mxsavec -mxsaves
+        set(ARCH_TUNE "${ARCH_TUNE} -DSD_F16C=true -DF_AVX512=true")
 ENDIF()
 
 if (NOT WIN32)
         # we don't want this definition for msvc
-        set(ARCH_TUNE "-march=${SD_ARCH} -mtune=${ARCH_TYPE}")
+        set(ARCH_TUNE "${ARCH_TUNE} -march=${SD_ARCH} -mtune=${ARCH_TYPE}")
 endif()
 
 if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "AppleClang" AND SD_X86_BUILD)
@@ -125,7 +127,7 @@ elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Intel")
     SET( CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} ${ARCH_TUNE} -O3 -fp-model fast")
 elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
     # using Visual Studio C++
-    set( CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} ${ARCH_TUNE}")
+    set( CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} ${ARCH_TUNE} /Ox")
 elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
     # using GCC
     SET( CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} ${ARCH_TUNE} -fmax-errors=2 -fdiagnostics-show-caret ")
@@ -159,15 +161,10 @@ if(HAVE_ARMCOMPUTE)
     file(GLOB_RECURSE CUSTOMOPS_ARMCOMPUTE_SOURCES false ops/declarable/platform/armcompute/*.cpp ops/declarable/platform/armcompute/*.h)
 endif()
 
-if(SD_CUDA)
-    message("Build cublas")
-    if(NOT DEFINED ${CMAKE_CUDA_ARCHITECTURES})
-        set(CMAKE_CUDA_ARCHITECTURES 75)
-    endif()
-    message(STATUS "CUDA architectures set to ${CMAKE_CUDA_ARCHITECTURES}")
 
-    find_package(CUDAToolkit)
-    enable_language(CUDA)
+if(SD_CUDA)
+    #find_package(CUDAToolkit)
+    #enable_language(CUDA)
 
     set(CMAKE_CUDA_STANDARD 17)
     set(CMAKE_CXX_STANDARD 14)
@@ -176,6 +173,9 @@ if(SD_CUDA)
     #Enable features prio C++17
     add_definitions(-D_HAS_AUTO_PTR_ETC=1)
 
+    set(CMAKE_CUDA_RUNTIME_LIBRARY "shared")
+    set(CMAKE_CUDA_ARCHITECTURES "61") #set(CMAKE_CUDA_ARCHITECTURES "62;75")
+
     #This basically kills instrinsic activated through SD_F16C=true
     #if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
     #    set (CMAKE_CXX_FLAGS "")
@@ -203,47 +203,29 @@ if(SD_CUDA)
            set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=-fPIC")
         endif()
 
-        if(WIN32)
-         message("In windows, setting cublas library and cusolver library")
-         if(NOT DEFINED CUDA_cublas_LIBRARY)
-                  set(CUDA_cublas_LIBRARY ${CUDA_HOME}/lib/x64/cublas.lib)
-          endif()
 
-          if(NOT DEFINED CUDA_cusolver_LIBRARY)
-                set(CUDA_cusolver_LIBRARY ${CUDA_HOME}/lib/x64/cusolver.lib)
-          endif()
-        endif()
-
-#
-        #string( TOLOWER "${COMPUTE}" COMPUTE_CMP )
-#        if ("${COMPUTE_CMP}" STREQUAL "all")
-#            CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Common")
-#        elseif("${COMPUTE_CMP}" STREQUAL "auto")
-#            CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Auto")
-#        elseif(COMPUTE_CMP MATCHES "^[0-9]+$")
-#            #matches USER COMPUTE old way
-             #set(CUDA_ARCH_FLAGS "-gencode arch=compute_${COMPUTE},code=sm_${COMPUTE} ")
-#        else()
-#            #matches numbers NAME | NUM.NUM | NUM.NUM(NUM.NUM) | NUM.NUM+PTX
-#            #NAME: Fermi Kepler Maxwell Kepler+Tegra Kepler+Tesla Maxwell+Tegra Pascal
-#            #NUM:   2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2 et cetera
-#            CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "${COMPUTE}")
+#        if(WIN32)
+#         message("In windows, setting cublas library and cusolver library")
+#         if(NOT DEFINED CUDA_cublas_LIBRARY)
+#                  set(CUDA_cublas_LIBRARY ${CUDA_HOME}/lib/x64/cublas.lib)
+#          endif()
+
+#          if(NOT DEFINED CUDA_cusolver_LIBRARY)
+#                set(CUDA_cusolver_LIBRARY ${CUDA_HOME}/lib/x64/cusolver.lib)
+#          endif()
 #        endif()
-        # list to spaces
-        #string (REPLACE ";" " " CUDA_ARCH_FLAGS "${CUDA_ARCH_FLAGS}")
 
         #set(CMAKE_CUDA_FLAGS " ${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=${CUDA_VERSION_MAJOR} ${EXPM} -w --cudart=static --expt-extended-lambda -Xfatbin -compress-all ")
-        set(CMAKE_CUDA_ARCHITECTURES OFF)
         #set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_53,code=[compute_53,sm_53]\" " )
-        set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_61,code=[compute_61,sm_61]\" " )
+        #set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_61,code=[compute_61,sm_61]\" " )
         #set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_75,code=[compute_75,sm_75]\" " )
         set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --extended-lambda ")
         set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr ")
-        set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=11 -w --cudart=static -Xfatbin -compress-all")
+        set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=12 -w -Xfatbin -compress-all")
+        set(CUDAHOSTCXX "${CMAKE_CXX_COMPILER}")
         if(WIN32)
-            set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=/EHsc")
+            set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=/EHsc -Xcompiler=/bigobj")
         endif()
-        #set(GPU_ARCH)
 
         message("CMAKE_CUDA_FLAGS = ${CMAKE_CUDA_FLAGS}")
         message("CMAKE_CXX_FLAGS = ${CMAKE_CXX_FLAGS}")
@@ -253,6 +235,9 @@ if(SD_CUDA)
         message("CUDA_NVCC_FLAGS = ${CUDA_NVCC_FLAGS}")
         message("CUDA_PROPAGATE_HOST_FLAGS = ${CUDA_PROPAGATE_HOST_FLAGS}")
         message("CUDA_ARCH_FLAGS = ${CUDA_ARCH_FLAGS}")
+        message("CUDAHOSTCXX = ${CUDAHOSTCXX}")
+        message("CMAKE_CUDA_ARCHITECTURES = ${CMAKE_CUDA_ARCHITECTURES}")
+        message("CMAKE_CUDA_RUNTIME_LIBRARY = ${CMAKE_CUDA_RUNTIME_LIBRARY}")
 
         file(GLOB_RECURSE PERF_SOURCES false performance/*.cpp performance/*.h)
         file(GLOB_RECURSE EXCEPTIONS_SOURCES false exceptions/*.cpp exceptions/*.h)
@@ -299,33 +284,36 @@ if(SD_CUDA)
 
         # build shared library by default or when it's explicitly requested
         if(NOT SD_STATIC_LIB OR SD_SHARED_LIB)
+            message("Will build a shared library '${SD_LIBRARY_NAME}'.")
             add_library(${SD_LIBRARY_NAME} SHARED $<TARGET_OBJECTS:samediff_obj>)
         endif()
 
         if (SD_STATIC_LIB AND SD_SHARED_LIB)
             # if both static and shared library are going to be built - static library will have special suffix
+            message("Will build a static library '${SD_LIBRARY_NAME}static'.")
             add_library(${SD_LIBRARY_NAME}static STATIC $<TARGET_OBJECTS:samediff_obj>)
-            set_property(TARGET ${SD_LIBRARY_NAME}static PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
+            #set_property(TARGET ${SD_LIBRARY_NAME}static PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RUNTIME_LIBRARY}$<$<CONFIG:Debug>:Debug>")
             install(TARGETS ${SD_LIBRARY_NAME}static  DESTINATION .)
         elseif(SD_STATIC_LIB)
             # if we only build static library - use this name
             add_library(${SD_LIBRARY_NAME} STATIC $<TARGET_OBJECTS:samediff_obj>)
-            set_property(TARGET ${SD_LIBRARY_NAME} PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
+            #set_property(TARGET ${SD_LIBRARY_NAME} PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RUNTIME_LIBRARY}$<$<CONFIG:Debug>:Debug>")
             install(TARGETS ${SD_LIBRARY_NAME}  DESTINATION .)
         endif()
 
         # on windows we want to make sure we use MT or MD, but since we use it in one lib, we must use it everywhere to avoid conflicts
-        set_property(TARGET samediff_obj PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
-        set_property(TARGET ${SD_LIBRARY_NAME} PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
+        #set_property(TARGET samediff_obj PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RUNTIME_LIBRARY}$<$<CONFIG:Debug>:Debug>")
+        #set_property(TARGET ${SD_LIBRARY_NAME} PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RUNTIME_LIBRARY}$<$<CONFIG:Debug>:Debug>")
 
 # Done by nvcc as default on windows
         if(WIN32)
-            message("CUDA on Windows: enabling /EHsc")
+            message("CUDA on Windows: enabling /EHsc and /bigobj")
             SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc /bigobj")
         endif()
 
         #target_link_libraries(${SD_LIBRARY_NAME} ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_cusolver_LIBRARY} ${CUDNN} ${MKLDNN})
-        target_link_libraries(${SD_LIBRARY_NAME} CUDA::cudart CUDA::cublas CUDA::cusolver ${CUDNN} ${MKLDNN})
+        target_link_libraries(${SD_LIBRARY_NAME} CUDA::cudart CUDA::cublas CUDA::cusolver CUDA::cublasLt Threads::Threads ${CUDNN} ${MKLDNN})
+        #target_link_libraries(${SD_LIBRARY_NAME} ${CUDNN} ${MKLDNN})
 
         set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/cuda/${SD_EXTENSION})
         install(TARGETS ${SD_LIBRARY_NAME} DESTINATION .)
@@ -437,13 +425,13 @@ elseif(SD_CPU)
             # if both static and shared library are going to be built - static library will have special suffix
             message("Adding a static library for ${SD_LIBRARY_NAME} as ${SD_LIBRARY_NAME}static")
             add_library(${SD_LIBRARY_NAME}static STATIC $<TARGET_OBJECTS:samediff_obj>)
-            set_property(TARGET ${SD_LIBRARY_NAME}static PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
+            #set_property(TARGET ${SD_LIBRARY_NAME}static PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
             install(TARGETS ${SD_LIBRARY_NAME}static  DESTINATION .)
         elseif(SD_STATIC_LIB)
             # if we only build static library - use this name
             message(Only building a static library for ${SD_LIBRARY_NAME})
             add_library(${SD_LIBRARY_NAME} STATIC $<TARGET_OBJECTS:samediff_obj>)
-            set_property(TARGET ${SD_LIBRARY_NAME} PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
+            #set_property(TARGET ${SD_LIBRARY_NAME} PROPERTY MSVC_RUNTIME_LIBRARY "${MSVC_RT_LIB}$<$<CONFIG:Debug>:Debug>")
             install(TARGETS ${SD_LIBRARY_NAME}  DESTINATION .)
         endif()
     endif()
@@ -460,13 +448,13 @@ elseif(SD_CPU)
 
     #This breaks the build. Normally you want to run tests anyways.
     if(NOT "$ENV{CLION_IDE}")
-        target_link_libraries(${SD_LIBRARY_NAME} ${MKLDNN}  ${MKLDNN_LIBRARIES} ${ARMCOMPUTE_LIBRARIES} ${OPENBLAS_LIBRARIES} ${BLAS_LIBRARIES} ${CPU_FEATURES})
+        target_link_libraries(${SD_LIBRARY_NAME} Threads::Threads ${MKLDNN} ${MKLDNN_LIBRARIES} ${ARMCOMPUTE_LIBRARIES} ${OPENBLAS_LIBRARIES} ${BLAS_LIBRARIES} ${CPU_FEATURES})
     endif()
 
     if ("${SD_ALL_OPS}" AND "${SD_BUILD_MINIFIER}")
         message(STATUS "Building minifier...")
         add_executable(minifier ../minifier/minifier.cpp ../minifier/graphopt.cpp)
-        target_link_libraries(minifier samediff_obj ${MKLDNN_LIBRARIES} ${ARMCOMPUTE_LIBRARIES} ${OPENBLAS_LIBRARIES} ${MKLDNN} ${BLAS_LIBRARIES} ${CPU_FEATURES})
+        target_link_libraries(minifier samediff_obj Threads::Threads ${MKLDNN_LIBRARIES} ${ARMCOMPUTE_LIBRARIES} ${OPENBLAS_LIBRARIES} ${MKLDNN} ${BLAS_LIBRARIES} ${CPU_FEATURES})
     endif()
 
     if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND "${CMAKE_CXX_COMPILER_VERSION}" VERSION_LESS 4.9)

cavis-native/cavis-native-lib/src/test/cpp/tests_cpu/layers_tests/CMakeLists.txt

@@ -1,5 +1,5 @@
 include_directories(${gtest_SOURCE_DIR}/include ${gtest_SOURCE_DIR})
-include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include)
+include_directories(../../../../../src/main/cpp/blas)
 if(LINUX)
     link_directories(/usr/local/lib)
     link_directories(/usr/lib)
@@ -21,10 +21,18 @@ if(WIN32)
     endforeach()
 endif()
 
+set(CMAKE_THREAD_PREFER_PTHREAD TRUE)
+set(THREADS_PREFER_PTHREAD_FLAG TRUE)
+find_package(Threads REQUIRED)
+
 if (SD_CUDA)
-	find_package(CUDA)
-	message("Tests CUDA include directory: ${CUDA_INCLUDE_DIRS}")
-	include_directories(${CUDA_INCLUDE_DIRS})
+	find_package(CUDAToolkit 12.2 REQUIRED)
+	enable_language(CUDA)
+
+    set(CMAKE_CUDA_STANDARD 17)
+    set(CMAKE_CXX_STANDARD 14)
+	message("Tests CUDA include directory: ${CUDAToolkit_INCLUDE_DIRS}")
+	include_directories(${CUDAToolkit_INCLUDE_DIRS})
 	add_definitions(-D__CUDABLAS__=true)
 
     if(WIN32)
@@ -34,17 +42,14 @@ if (SD_CUDA)
 
     string( TOLOWER "${COMPUTE}" COMPUTE_CMP ) 
     if ("${COMPUTE_CMP}" STREQUAL "all")
-        CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Common")
+        set(CMAKE_CUDA_ARCHITECTURES "all")
     elseif("${COMPUTE_CMP}" STREQUAL "auto")
-        CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Auto")
-    elseif(COMPUTE_CMP MATCHES "^[0-9]+$")
-        #matches USER COMPUTE old way 
-        set(CUDA_ARCH_FLAGS "-gencode arch=compute_${COMPUTE},code=sm_${COMPUTE} ")
+        set(CMAKE_CUDA_ARCHITECTURES "all-major")
     else()
         #matches numbers NAME | NUM.NUM | NUM.NUM(NUM.NUM) | NUM.NUM+PTX
         #NAME: Fermi Kepler Maxwell Kepler+Tegra Kepler+Tesla Maxwell+Tegra Pascal
         #NUM:   2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2 et cetera
-        CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "${COMPUTE}")
+        set(CMAKE_CUDA_ARCHITECTURES "all")
     endif()
     # list to spaces
     string (REPLACE ";" " " CUDA_ARCH_FLAGS "${CUDA_ARCH_FLAGS}")
@@ -149,7 +154,7 @@ if (SD_CPU)
     endif()
 
 	add_executable(runtests ${TEST_SOURCES})
-	target_link_libraries(runtests samediff_obj ${MKLDNN_LIBRARIES} ${OPENBLAS_LIBRARIES} ${MKLDNN} ${BLAS_LIBRARIES} ${CPU_FEATURES}  ${ARMCOMPUTE_LIBRARIES}  gtest gtest_main)
+	target_link_libraries(runtests samediff_obj Threads::Threads ${MKLDNN_LIBRARIES} ${OPENBLAS_LIBRARIES} ${MKLDNN} ${BLAS_LIBRARIES} ${CPU_FEATURES}  ${ARMCOMPUTE_LIBRARIES}  gtest gtest_main)
 elseif(SD_CUDA)
 
 	add_executable(runtests ${TEST_SOURCES})
@@ -167,5 +172,5 @@ elseif(SD_CUDA)
         message("CUDNN library: ${CUDNN}")
     endif()
 
-	target_link_libraries(runtests samediff_obj ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_cusolver_LIBRARY} ${CUDNN} ${MKLDNN} gtest gtest_main)
+	target_link_libraries(runtests samediff_obj Threads::Threads ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_cusolver_LIBRARY} ${CUDNN} ${MKLDNN} gtest gtest_main)
 endif()