Use rocm 6.2 in noble image

CExA-project · Aug 28, 2024 · d3f9daf · d3f9daf
1 parent 4938221
commit d3f9daf
Show file tree

Hide file tree

Showing 2 changed files with 23 additions and 20 deletions.
diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
@@ -41,9 +41,6 @@ jobs:
       matrix:
         image: ['oldest', 'latest']
         backend: ['cpu', 'cuda', 'hip']
-        exclude:
-        - image: 'latest' # no ppa yet for hip in noble image
-          backend: 'hip'
     needs: id_repo
     runs-on: ubuntu-latest
     steps:
@@ -98,8 +95,6 @@ jobs:
         - image: 'oldest' # nvcc 11 only supports C++-17
           backend: 'cuda'
           cxx_version: '23'
-        - image: 'latest' # no ppa yet for hip in noble image
-          backend: 'hip'
         - image: 'latest' # nvcc 12 only supports C++-20
           backend: 'cuda'
           cxx_version: '23'

diff --git a/docker/latest/Dockerfile b/docker/latest/Dockerfile
@@ -7,11 +7,15 @@ FROM ubuntu:noble
 LABEL "org.opencontainers.image.source"="https://github.com/Maison-de-la-Simulation/ddc"
 
 ARG BACKEND
+ARG AMDGPU_VERSION=6.2
+ARG ROCM_VERSION=6.2
 
 COPY bash_run /bin/
 ENV BASH_ENV=/etc/profile
 SHELL ["/bin/bash", "-c"]
 
+ENV PATH="${PATH:+$PATH:}/opt/rocm-${ROCM_VERSION}.0/bin"
+ENV CMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH:+$CMAKE_PREFIX_PATH:}/opt/rocm-${ROCM_VERSION}.0"
 
 RUN chmod +x /bin/bash_run \
  && export DEBIAN_FRONTEND=noninteractive \
@@ -26,12 +30,12 @@ RUN chmod +x /bin/bash_run \
     gpg \
     wget \
  && mkdir --parents --mode=0755 /etc/apt/keyrings \
-#  && if [ "xhip" = "x${BACKEND}" ] \
-#   ; then echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/amdgpu/5.7.3/ubuntu noble main" > /etc/apt/sources.list.d/amdgpu.list \
-#       && echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/5.7.3 noble main" > /etc/apt/sources.list.d/rocm.list \
-#       && wget -q -O - https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor --output /etc/apt/keyrings/rocm.gpg \
-#       && echo -e "Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600" > /etc/apt/preferences.d/rocm-pin-600 \
-#  ;  fi \
+ && if [ "xhip" = "x${BACKEND}" ] \
+  ; then echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/amdgpu/${AMDGPU_VERSION}/ubuntu noble main" > /etc/apt/sources.list.d/amdgpu.list \
+      && echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/${ROCM_VERSION} noble main" > /etc/apt/sources.list.d/rocm.list \
+      && wget -q -O - https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor --output /etc/apt/keyrings/rocm.gpg \
+      && echo -e "Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600" > /etc/apt/preferences.d/rocm-pin-600 \
+ ;  fi \
  && echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/pdidev-archive-keyring.gpg] https://raw.githubusercontent.com/pdidev/repo/ubuntu noble main" > /etc/apt/sources.list.d/pdi.list \
  && wget -q -O /etc/apt/keyrings/pdidev-archive-keyring.gpg https://raw.githubusercontent.com/pdidev/repo/ubuntu/pdidev-archive-keyring.gpg \
  && apt-get update -y \
@@ -57,19 +61,26 @@ RUN chmod +x /bin/bash_run \
       nvidia-cuda-toolkit \
       gcc-12 \
       g++-12 \
-#  ;; "hip") \
-#       apt-get install -y --no-install-recommends \
-#       rocm-hip-sdk \
- ;; esac \
+ ;; "hip") \
+      apt-get install -y --no-install-recommends \
+      rocm-hip-sdk \
+      && echo "/opt/rocm/lib" >> /etc/ld.so.conf.d/rocm.conf \
+      && echo "/opt/rocm/lib64" >> /etc/ld.so.conf.d/rocm.conf \
+      && ldconfig \
+;; esac \
  && git clone -b v1.8.0 https://github.com/ginkgo-project/ginkgo.git \
  && cd ginkgo \
  && case "${BACKEND}" in \
     "cpu") \
       cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo -DGINKGO_BUILD_OMP=ON -DGINKGO_BUILD_TESTS=OFF -DGINKGO_BUILD_EXAMPLES=OFF -DGINKGO_BUILD_BENCHMARKS=OFF \
  ;; "cuda") \
       cmake -S . -B build -DCMAKE_C_COMPILER=gcc-12 -DCMAKE_CXX_COMPILER=g++-12 -DCMAKE_CUDA_HOST_COMPILER=g++-12 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DGINKGO_CUDA_ARCHITECTURES=70 -DGINKGO_BUILD_CUDA=ON -DGINKGO_BUILD_TESTS=OFF -DGINKGO_BUILD_EXAMPLES=OFF -DGINKGO_BUILD_BENCHMARKS=OFF \
-#  ;; "hip") \
-#       cmake -S . -B build -DCMAKE_PREFIX_PATH=/opt/rocm -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_HIP_ARCHITECTURES=gfx90a -DGINKGO_BUILD_HIP=ON -DGINKGO_BUILD_TESTS=OFF -DGINKGO_BUILD_EXAMPLES=OFF -DGINKGO_BUILD_BENCHMARKS=OFF \
+ ;; "hip") \
+      # HIP_PATH needs to be set to avoid Ginkgo defining it (https://github.com/ginkgo-project/ginkgo/blob/develop/cmake/hip_path.cmake).
+      # The Ginkgo heuristic does not seem to be compatible with the new HIP v6 directory layout.
+      # This makes amdclang++ fail during cmake compiler test (https://releases.llvm.org/18.1.0/tools/clang/docs/HIPSupport.html#order-of-precedence-for-hip-path)
+      export HIP_PATH=`hipconfig --path` \
+      && cmake -S . -B build -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_HIP_ARCHITECTURES=gfx90a -DGINKGO_BUILD_HIP=ON -DGINKGO_BUILD_TESTS=OFF -DGINKGO_BUILD_EXAMPLES=OFF -DGINKGO_BUILD_BENCHMARKS=OFF \
  ;; esac \
  && cmake --build build \
  && cmake --install build \
@@ -89,9 +100,6 @@ RUN chmod +x /bin/bash_run \
   ; then echo 'CUDA_GCC=gcc-12' > /etc/profile.d/ddc-cuda.sh \
        ; echo 'CUDA_GXX=g++-12' >> /etc/profile.d/ddc-cuda.sh \
  ;  fi
-#  && if [ "xhip" = "x${BACKEND}" ] \
-#  ;  then echo 'export LD_LIBRARY_PATH="/opt/rocm/lib"' > /etc/profile.d/10-rocm.sh \
-#  ;  fi
 
 USER ci:ci
 WORKDIR /data