From 1b33d3d5862707b230711311a4848c9b279734ef Mon Sep 17 00:00:00 2001
From: David Rohr <drohr@jwdt.org>
Date: Thu, 7 Nov 2024 14:35:03 +0100
Subject: [PATCH] Update SLC9 GPU container

---
 slc9-gpu-builder/amdgpu.repo  |  2 +-
 slc9-gpu-builder/packer.json  |  4 ++--
 slc9-gpu-builder/provision.sh | 14 +++++++++-----
 slc9-gpu-builder/rocm.repo    |  2 +-
 4 files changed, 13 insertions(+), 9 deletions(-)

diff --git a/slc9-gpu-builder/amdgpu.repo b/slc9-gpu-builder/amdgpu.repo
index 079e61b..b68005c 100644
--- a/slc9-gpu-builder/amdgpu.repo
+++ b/slc9-gpu-builder/amdgpu.repo
@@ -1,5 +1,5 @@
 [amdgpu]
 name=amdgpu
-baseurl=http://repo.radeon.com/amdgpu/6.2/rhel/9.4/main/x86_64/
+baseurl=http://repo.radeon.com/amdgpu/6.2.4/rhel/9.4/main/x86_64/
 enabled=1
 gpgcheck=0
diff --git a/slc9-gpu-builder/packer.json b/slc9-gpu-builder/packer.json
index 9f22297..01b45cd 100644
--- a/slc9-gpu-builder/packer.json
+++ b/slc9-gpu-builder/packer.json
@@ -1,9 +1,9 @@
 {
-  "_comment": "Alma 9.4 GPU builder X-enabled CUDA12.6-enabled AMD ROCm 6.2.0-enabled",
+  "_comment": "Alma 9.4 GPU builder X-enabled CUDA12.6-enabled AMD ROCm 6.2.4-enabled",
   "variables": {
     "REPO": "registry.cern.ch/alisw/slc9-gpu-builder",
     "TAG": "latest",
-    "CUDA_PKG_VERSION": "12-6-12.6.*",
+    "CUDA_PKG_VERSION": "12-6",
     "NVIDIA_GPGKEY_SUM": "d0664fbbdb8c32356d45de36c5984617217b2d0bef41b93ccecd326ba3b80c87"
   },
   "builders": [
diff --git a/slc9-gpu-builder/provision.sh b/slc9-gpu-builder/provision.sh
index d6df49c..0d47bfb 100755
--- a/slc9-gpu-builder/provision.sh
+++ b/slc9-gpu-builder/provision.sh
@@ -26,11 +26,15 @@ echo "${NVIDIA_GPGKEY_SUM}  /etc/pki/rpm-gpg/RPM-GPG-KEY-NVIDIA" | sha256sum -c
 # rpm --import https://repo.almalinux.org/almalinux/RPM-GPG-KEY-AlmaLinux
 dnf update -y
 # Install requirements for GPU event display, NVIDIA CUDA and AMD ROCm stacks
-dnf install -y freeglut-devel lsof "cuda-cudart-$CUDA_PKG_VERSION" 'cuda-compat-12-0-*'           \
-               "cuda-libraries-$CUDA_PKG_VERSION" "cuda-nvtx-$CUDA_PKG_VERSION"                   \
-               "cuda-libraries-devel-$CUDA_PKG_VERSION" "cuda-nvml-devel-$CUDA_PKG_VERSION"       \
-               "cuda-minimal-build-$CUDA_PKG_VERSION" "cuda-command-line-tools-$CUDA_PKG_VERSION" \
-               hip-rocclr ocl-icd ocl-icd-devel hipcub rocthrust rocm-dev hipify-clang
+CUV=${CUDA_PKG_VERSION}-${CUDA_PKG_VERSION/-/.}.*
+dnf install -y freeglut-devel lsof                                                            \
+               "cuda-cudart-$CUV" 'cuda-compat-12-0-*' "cuda-libraries-$CUV" "cuda-nvtx-$CUV" \
+               "cuda-libraries-devel-$CUV" "cuda-nvml-devel-$CUV" "cuda-minimal-build-$CUV"   \
+               "cuda-command-line-tools-$CUV"                                                 \
+               "cudnn9-cuda-$CUDA_PKG_VERSION"                                                \
+               hip-rocclr ocl-icd ocl-icd-devel hipcub rocthrust rocm-dev hipify-clang        \
+               hiprand-devel hipblas-devel hipsparse-devel rocblas-devel rocrand-devel        \
+               miopen-hip-devel hipfft-devel rccl-devel migraphx-devel
 # ROCm: Notice we do not need the version for ROCM because we target a specific distribution in rocm.repo
 
 # Set up NVIDIA CUDA stack
diff --git a/slc9-gpu-builder/rocm.repo b/slc9-gpu-builder/rocm.repo
index 58dc483..65568b0 100644
--- a/slc9-gpu-builder/rocm.repo
+++ b/slc9-gpu-builder/rocm.repo
@@ -1,5 +1,5 @@
 [ROCm]
 name=ROCm
-baseurl=http://repo.radeon.com/rocm/rhel9/6.2/main/
+baseurl=http://repo.radeon.com/rocm/rhel9/6.2.4/main/
 enabled=1
 gpgcheck=0