From 19654480ad7c828d39ed886d03125d0ee37d6e31 Mon Sep 17 00:00:00 2001
From: Nicolas Richart <networms@gmail.ch>
Date: Tue, 9 Jul 2024 16:07:13 +0200
Subject: [PATCH] Addapting version of rdma-core and adding l40s definitions
---
dockerfiles/rhel9-kuma/Dockerfile | 2 +-
.../pinot-noir-gcc/definitions_kuma_l40s.yaml | 7 ++
stacks/pinot-noir-gcc/packages_kuma_h100.yaml | 2 +-
stacks/pinot-noir-gcc/packages_kuma_l40s.yaml | 75 ++++++++++++++++++-
4 files changed, 80 insertions(+), 6 deletions(-)
create mode 100644 stacks/pinot-noir-gcc/definitions_kuma_l40s.yaml
diff --git a/dockerfiles/rhel9-kuma/Dockerfile b/dockerfiles/rhel9-kuma/Dockerfile
index 2b6da77..c3cc960 100644
--- a/dockerfiles/rhel9-kuma/Dockerfile
+++ b/dockerfiles/rhel9-kuma/Dockerfile
@@ -10,7 +10,7 @@ FROM registry.c4science.ch/scitas-stack/rhel9-base:9.0
ARG APPSTREAM_REPO=rhel-9-for-x86_64-appstream-rpms
ARG SLURM_REPO=SCITAS_product-slurm_repo-slurm-2311
ARG PMIX_REPO=SCITAS_product-scitas_repo-scitas-cluster
-ARG MLNX_REPO=SCITAS_product-mlnx_repo-mlnx-ofed-24_01-0_3_3_1-5_14_0-70_30_1-rhel90
+ARG MLNX_REPO=SCITAS_product-mlnx_repo-mlnx-ofed-24_04-0_6_6_0-5_14_0-70_30_1-rhel90
ARG CUDA_REPO=SCITAS_product-cuda_repo-cuda-535_154_05-x86_64-rhel9
ARG GPFS_REPO=SCITAS_product-gpfs_repo-gpfs-519
diff --git a/stacks/pinot-noir-gcc/definitions_kuma_l40s.yaml b/stacks/pinot-noir-gcc/definitions_kuma_l40s.yaml
new file mode 100644
index 0000000..0f156fa
--- /dev/null
+++ b/stacks/pinot-noir-gcc/definitions_kuma_l40s.yaml
@@ -0,0 +1,7 @@
+definitions:
+ - cuda_system_codes:
+ - cuda
+ - cudnn
+
+ - cuda_serial_codes:
+ - nccl
diff --git a/stacks/pinot-noir-gcc/packages_kuma_h100.yaml b/stacks/pinot-noir-gcc/packages_kuma_h100.yaml
index 07fab46..9dd1e2a 100644
--- a/stacks/pinot-noir-gcc/packages_kuma_h100.yaml
+++ b/stacks/pinot-noir-gcc/packages_kuma_h100.yaml
@@ -62,7 +62,7 @@ packages:
rdma-core:
buildable: false
externals:
- - spec: 'rdma-core@47.1'
+ - spec: 'rdma-core@51.1'
prefix: /usr
slurm:
diff --git a/stacks/pinot-noir-gcc/packages_kuma_l40s.yaml b/stacks/pinot-noir-gcc/packages_kuma_l40s.yaml
index ad6b8fb..e73666a 100644
--- a/stacks/pinot-noir-gcc/packages_kuma_l40s.yaml
+++ b/stacks/pinot-noir-gcc/packages_kuma_l40s.yaml
@@ -2,10 +2,77 @@ packages:
all:
target: ["zen4"]
require:
- - spec: ^openmpi +cuda
- when: '%gcc ^mpi'
+ - "~rocm"
- - spec: cuda_arch=89
+ - spec: cuda_arch=89 target=zen4
when: '+cuda'
- - '+cuda'
+ - spec: '+cuda'
+ when: '%gcc'
+
+ # ---------------------------------------------------------------------------
+ hypre:
+ require:
+ - spec: '+cuda cuda_arch=89 +unified-memory'
+ when: '%gcc'
+
+ kokkos:
+ require:
+ - spec: '+cuda cuda_arch=89 +cuda_uvm +wrapper ~openmptarget'
+ when: '%gcc'
+
+ openmpi:
+ require:
+ - spec: '+cuda cuda_arch=89 target=zen4'
+
+ petsc:
+ require:
+ - spec: '+cuda cuda_arch=89'
+ when: '%gcc'
+
+ py-tensorflow:
+ require:
+ - spec: '+cuda cuda_arch=89 +nccl'
+
+ py-torch:
+ require:
+ - spec: '+cuda cuda_arch=89 +nccl +cudnn'
+
+ quantum-espresso:
+ require:
+ - spec: '~cuda'
+
+ suite-sparse:
+ require:
+ - spec: '+cuda'
+
+ ucx:
+ require:
+ - spec: '+xpmem +cma +rdmacm +rc +ud +dc +verbs +ib_hw_tm +gdrcopy +cuda cuda_arch=89 ~rocm target=zen4'
+
+ # ---------------------------------------------------------------------------
+ # Externals
+ # ---------------------------------------------------------------------------
+ pmix:
+ buildable: false
+ externals:
+ - spec: 'pmix@5.0.2'
+ prefix: /usr
+
+ rdma-core:
+ buildable: false
+ externals:
+ - spec: 'rdma-core@51.1'
+ prefix: /usr
+
+ slurm:
+ buildable: false
+ externals:
+ - spec: 'slurm@24-05-0-2'
+ prefix: /usr
+
+ xpmem:
+ buildable: false
+ externals:
+ - spec: 'xpmem@2.7.3'
+ prefix: /usr
--
GitLab