Basic CUDA support

.github/workflows/test.yml

@@ -34,4 +34,4 @@ jobs:
 
     - name: Pytest
       run: |
-        pytest tests/ --cov
+        pytest tests/ --cov -k "not cuda"

Jenkinsfile

@@ -0,0 +1,91 @@
+pipeline {
+  agent none
+  options {
+    disableConcurrentBuilds()
+    buildDiscarder(logRotator(numToKeepStr: '8', daysToKeepStr: '20'))
+    timeout(time: 1, unit: 'HOURS')
+  }
+  stages {
+    stage('CUDA Tests') {
+      agent {
+         dockerfile {
+            filename 'ci/docker/Dockerfile-cuda11.8'
+            args '--gpus 2'
+            label 'docker && v100'
+         }
+      }
+      environment {
+    HOME = "$WORKSPACE"
+    PYBIN = "/opt/python/cp39-cp39/bin"
+    LIBRARY_PATH = "$WORKSPACE/finufft/build"
+    LD_LIBRARY_PATH = "$WORKSPACE/finufft/build"
+      }
+      steps {
+
+    // TODO - reconsider install strategy once finufft/cufinufft 2.2 is released
+  checkout scmGit(branches: [[name: '*/master']],
+                  extensions: [cloneOption(noTags: true, reference: '', shallow: true),
+                               [$class: 'RelativeTargetDirectory', relativeTargetDir: 'finufft'],
+                               cleanAfterCheckout()],
+                  userRemoteConfigs: [[url: 'https://github.com/flatironinstitute/finufft']])
+
+    sh '''#!/bin/bash -ex
+      nvidia-smi
+    '''
+    sh '''#!/bin/bash -ex
+      echo $HOME
+      ls
+    '''
+    sh '''#!/bin/bash -ex
+        cd finufft
+        # v100 cuda arch
+        cuda_arch="70"
+
+        cmake -B build . -DFINUFFT_USE_CUDA=ON \
+                         -DFINUFFT_USE_CPU=OFF \
+                         -DFINUFFT_BUILD_TESTS=OFF \
+                         -DCMAKE_CUDA_ARCHITECTURES="$cuda_arch" \
+                         -DBUILD_TESTING=ON
+        cd build
+        make -j4
+    '''
+
+    sh '${PYBIN}/python3 -m venv $HOME'
+    sh '''#!/bin/bash -ex
+      source $HOME/bin/activate
+      python3 -m pip install --upgrade pip
+      # we could also move pytorch install inside docker
+      python3 -m pip install "torch~=2.1.0" --index-url https://download.pytorch.org/whl/cu118
+      python3 -m pip install finufft/python/cufinufft
+
+      python3 -m pip install -e .[dev]
+
+      python3 -m pytest -k "cuda" tests/ --cov -v
+    '''
+      }
+    }
+  }
+  post {
+    failure {
+      emailext subject: '$PROJECT_NAME - Build #$BUILD_NUMBER - $BUILD_STATUS',
+           body: '''$PROJECT_NAME - Build #$BUILD_NUMBER - $BUILD_STATUS
+
+Check console output at $BUILD_URL to view full results.
+
+Building $BRANCH_NAME for $CAUSE
+$JOB_DESCRIPTION
+
+Chages:
+$CHANGES
+
+End of build log:
+${BUILD_LOG,maxLines=200}
+''',
+           recipientProviders: [
+         [$class: 'DevelopersRecipientProvider'],
+           ],
+           replyTo: '$DEFAULT_REPLYTO',
+           to: '[email protected]'
+    }
+  }
+}

ci/docker/Dockerfile-cuda11.8

@@ -0,0 +1,58 @@
+# Based on https://github.com/flatironinstitute/finufft/blob/master/tools/cufinufft/docker/cuda11.2/Dockerfile-x86_64
+
+FROM quay.io/pypa/manylinux2014_x86_64
+LABEL maintainer "Brian Ward"
+
+ENV CUDA_MAJOR 11
+ENV CUDA_MINOR 8
+ENV CUDA_DASH_VERSION ${CUDA_MAJOR}-${CUDA_MINOR}
+ENV CUDA_DOT_VERSION ${CUDA_MAJOR}.${CUDA_MINOR}
+
+# ---- The following block adds layers for CUDA --- #
+# base
+RUN NVIDIA_GPGKEY_SUM=d0664fbbdb8c32356d45de36c5984617217b2d0bef41b93ccecd326ba3b80c87 && \
+    curl -fsSL https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/D42D0685.pub | sed '/^Version/d' > /etc/pki/rpm-gpg/RPM-GPG-KEY-NVIDIA && \
+    echo "$NVIDIA_GPGKEY_SUM  /etc/pki/rpm-gpg/RPM-GPG-KEY-NVIDIA" | sha256sum -c --strict -
+
+COPY ci/docker/cuda.repo /etc/yum.repos.d/cuda.repo
+
+# For libraries in the cuda-compat-* package: https://docs.nvidia.com/cuda/eula/index.html#attachment-a
+RUN yum install -y \
+        cuda-cudart-${CUDA_DASH_VERSION} \
+        cuda-compat-${CUDA_DASH_VERSION} && \
+    ln -s cuda-${CUDA_DOT_VERSION} /usr/local/cuda
+
+# nvidia-docker 1.0
+RUN echo "/usr/local/nvidia/lib" >> /etc/ld.so.conf.d/nvidia.conf && \
+    echo "/usr/local/nvidia/lib64" >> /etc/ld.so.conf.d/nvidia.conf
+
+ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:${PATH}
+ENV LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
+
+# nvidia-container-runtime
+ENV NVIDIA_VISIBLE_DEVICES all
+ENV NVIDIA_DRIVER_CAPABILITIES compute,utility
+ENV NVIDIA_REQUIRE_CUDA "cuda>=${CUDA_DOT_VERSION} brand=tesla,driver>=418,driver<419 brand=tesla,driver>=440,driver<441"
+
+# runtime
+RUN yum install -y \
+        cuda-libraries-${CUDA_DASH_VERSION} \
+        cuda-nvtx-${CUDA_DASH_VERSION} \
+        cuda-cudart-devel-${CUDA_DASH_VERSION} \
+        cuda-libraries-devel-${CUDA_DASH_VERSION} \
+        cuda-nvprof-${CUDA_DASH_VERSION} \
+        cuda-nvcc-${CUDA_DASH_VERSION}
+
+ENV LIBRARY_PATH /usr/local/cuda/lib64/stubs
+
+# /CUDA #
+
+# CUDA 11 doesn't work on gcc/g++ newer than v9
+RUN yum install -y \
+        devtoolset-9-gcc \
+        devtoolset-9-gcc-c++ \
+        cmake && \
+    rm -rf /var/cache/yum/*
+
+ENV PATH /opt/rh/devtoolset-9/root/usr/bin:${PATH}
+

ci/docker/cuda.repo

@@ -0,0 +1,6 @@
+[cuda]
+name=cuda
+baseurl=https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64
+enabled=1
+gpgcheck=1
+gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-NVIDIA