Skip to content

Commit 53139a0

Browse files
Artemy-MellanoxArtemy-Mellanox
committed
Merge remote-tracking branch 'origin/master' into topic/collapsed_cq-2
2 parents 5133a6b + 1be106e commit 53139a0

File tree

21 files changed

+393
-45
lines changed

21 files changed

+393
-45
lines changed

‎buildlib/az-distro-release.yml‎

Lines changed: 7 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -43,6 +43,12 @@ jobs:
4343
ubuntu24_cuda13_${{ parameters.arch }}:
4444
build_container: ubuntu24_cuda13_${{ parameters.arch }}
4545
artifact_name: $(POSTFIX)-ubuntu24.04-mofed5-cuda13-${{ parameters.arch }}.tar.bz2
46+
rocky8_cuda13_${{ parameters.arch }}:
47+
build_container: rocky8_cuda13_${{ parameters.arch }}
48+
artifact_name: $(POSTFIX)-rocky8-mofed24.10-cuda13-${{ parameters.arch }}.tar.bz2
49+
rocky9_cuda13_${{ parameters.arch }}:
50+
build_container: rocky9_cuda13_${{ parameters.arch }}
51+
artifact_name: $(POSTFIX)-rocky9-mofed24.10-cuda13-${{ parameters.arch }}.tar.bz2
4652
# x86 only
4753
${{ if eq(parameters.arch, 'x86_64') }}:
4854
centos7_cuda11_${{ parameters.arch }}:
@@ -84,7 +90,7 @@ jobs:
8490
cd ../../..
8591
tar -tjf "${AZ_ARTIFACT_NAME}"
8692
displayName: Build RPM package
87-
condition: and(succeeded(), contains(variables['artifact_name'], 'centos'))
93+
condition: and(succeeded(), or(contains(variables['artifact_name'], 'centos'), contains(variables['artifact_name'], 'rocky')))
8894
env:
8995
AZ_ARTIFACT_NAME: $(artifact_name)
9096

‎buildlib/azure-pipelines-release-drp.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -48,6 +48,12 @@ resources:
4848
image: $(REPO_MIRROR)/ucx/x86_64/ubuntu22.04-mofed5-cuda13:1
4949
- container: ubuntu24_cuda13_x86_64
5050
image: $(REPO_MIRROR)/ucx/x86_64/ubuntu24.04-mofed24.10-cuda13:1
51+
- container: rocky8_cuda13_x86_64
52+
image: $(REPO_MIRROR)/ucx/x86_64/rocky8-mofed24.10-cuda13:1
53+
options: $(DOCKER_OPT_VOLUMES)
54+
- container: rocky9_cuda13_x86_64
55+
image: $(REPO_MIRROR)/ucx/x86_64/rocky9-mofed24.10-cuda13:1
56+
options: $(DOCKER_OPT_VOLUMES)
5157

5258
# aarch64
5359
- container: centos8_cuda11_aarch64
@@ -69,6 +75,12 @@ resources:
6975
image: $(REPO_MIRROR)/ucx/aarch64/ubuntu22.04-mofed5-cuda13:1
7076
- container: ubuntu24_cuda13_aarch64
7177
image: $(REPO_MIRROR)/ucx/aarch64/ubuntu24.04-mofed24.10-cuda13:1
78+
- container: rocky8_cuda13_aarch64
79+
image: $(REPO_MIRROR)/ucx/aarch64/rocky8-mofed24.10-cuda13:2
80+
options: $(DOCKER_OPT_VOLUMES)
81+
- container: rocky9_cuda13_aarch64
82+
image: $(REPO_MIRROR)/ucx/aarch64/rocky9-mofed24.10-cuda13:2
83+
options: $(DOCKER_OPT_VOLUMES)
7284

7385
stages:
7486
- stage: Prepare

‎buildlib/azure-pipelines-release.yml‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -44,6 +44,12 @@ resources:
4444
image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu22.04-mofed5-cuda13:1
4545
- container: ubuntu24_cuda13_x86_64
4646
image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu24.04-mofed24.10-cuda13:1
47+
- container: rocky8_cuda13_x86_64
48+
image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/rocky8-mofed24.10-cuda13:1
49+
options: $(DOCKER_OPT_VOLUMES)
50+
- container: rocky9_cuda13_x86_64
51+
image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/rocky9-mofed24.10-cuda13:1
52+
options: $(DOCKER_OPT_VOLUMES)
4753

4854
# aarch64
4955
- container: centos8_cuda11_aarch64
@@ -65,6 +71,12 @@ resources:
6571
image: rdmz-harbor.rdmz.labs.mlnx/ucx/aarch64/ubuntu22.04-mofed5-cuda13:1
6672
- container: ubuntu24_cuda13_aarch64
6773
image: rdmz-harbor.rdmz.labs.mlnx/ucx/aarch64/ubuntu24.04-mofed24.10-cuda13:1
74+
- container: rocky8_cuda13_aarch64
75+
image: rdmz-harbor.rdmz.labs.mlnx/ucx/aarch64/rocky8-mofed24.10-cuda13:2
76+
options: $(DOCKER_OPT_VOLUMES)
77+
- container: rocky9_cuda13_aarch64
78+
image: rdmz-harbor.rdmz.labs.mlnx/ucx/aarch64/rocky9-mofed24.10-cuda13:2
79+
options: $(DOCKER_OPT_VOLUMES)
6880

6981
stages:
7082
- stage: Prepare

‎buildlib/dockers/docker-compose-aarch64.yml‎

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -117,3 +117,29 @@ services:
117117
CUDA_VERSION: 13.0.0
118118
NV_DRIVER_VERSION: 580
119119
ARCH: aarch64
120+
rocky8-mofed5-cuda13:
121+
image: rocky8-mofed24.10-cuda13:2
122+
build:
123+
context: .
124+
network: host
125+
dockerfile: rocky-release.Dockerfile
126+
args:
127+
MOFED_VERSION: 24.10-3.2.5.0
128+
MOFED_OS: rhel8.9
129+
OS_VERSION: 8
130+
CUDA_VERSION: 13.0.0
131+
NV_DRIVER_VERSION: 580
132+
ARCH: aarch64
133+
rocky9-mofed5-cuda13:
134+
image: rocky9-mofed24.10-cuda13:2
135+
build:
136+
context: .
137+
network: host
138+
dockerfile: rocky-release.Dockerfile
139+
args:
140+
MOFED_VERSION: 24.10-3.2.5.0
141+
MOFED_OS: rhel9.6
142+
OS_VERSION: 9
143+
CUDA_VERSION: 13.0.0
144+
NV_DRIVER_VERSION: 580
145+
ARCH: aarch64

‎buildlib/dockers/docker-compose-x86_64.yml‎

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -153,3 +153,29 @@ services:
153153
CUDA_VERSION: 13.0.0
154154
NV_DRIVER_VERSION: 580
155155
ARCH: x86_64
156+
rocky8-mofed5-cuda13:
157+
image: rocky8-mofed24.10-cuda13:1
158+
build:
159+
context: .
160+
network: host
161+
dockerfile: rocky-release.Dockerfile
162+
args:
163+
MOFED_VERSION: 24.10-3.2.5.0
164+
MOFED_OS: rhel8.9
165+
OS_VERSION: 8
166+
CUDA_VERSION: 13.0.0
167+
NV_DRIVER_VERSION: 580
168+
ARCH: x86_64
169+
rocky9-mofed5-cuda13:
170+
image: rocky9-mofed24.10-cuda13:1
171+
build:
172+
context: .
173+
network: host
174+
dockerfile: rocky-release.Dockerfile
175+
args:
176+
MOFED_VERSION: 24.10-3.2.5.0
177+
MOFED_OS: rhel9.6
178+
OS_VERSION: 9
179+
CUDA_VERSION: 13.0.0
180+
NV_DRIVER_VERSION: 580
181+
ARCH: x86_64

‎buildlib/dockers/rocky-release.Dockerfile‎

Lines changed: 62 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,62 @@
1+
ARG CUDA_VERSION
2+
ARG OS_VERSION
3+
FROM nvidia/cuda:${CUDA_VERSION}-devel-rockylinux${OS_VERSION}
4+
5+
ARG MOFED_VERSION
6+
ARG MOFED_OS
7+
ARG ARCH
8+
9+
RUN yum install -y \
10+
autoconf \
11+
automake \
12+
environment-modules \
13+
ethtool \
14+
file \
15+
fuse-libs \
16+
gcc-c++ \
17+
git \
18+
glibc-devel \
19+
libtool \
20+
libusbx \
21+
lsof \
22+
make \
23+
maven \
24+
numactl-devel \
25+
pciutils \
26+
perl \
27+
pinentry \
28+
python3 \
29+
rdma-core-devel \
30+
rpm-build \
31+
tcl \
32+
tcsh \
33+
tk \
34+
valgrind-devel \
35+
wget \
36+
&& yum clean all \
37+
&& rm -rf /var/cache/yum
38+
39+
ENV MOFED_DIR=MLNX_OFED_LINUX-${MOFED_VERSION}-${MOFED_OS}-${ARCH} \
40+
MOFED_SITE_PLACE=MLNX_OFED-${MOFED_VERSION} \
41+
CPATH=/usr/local/cuda/include:${CPATH} \
42+
LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} \
43+
LIBRARY_PATH=/usr/local/cuda/lib64:${LIBRARY_PATH}
44+
45+
RUN wget --no-verbose http://content.mellanox.com/ofed/${MOFED_SITE_PLACE}/${MOFED_DIR}.tgz \
46+
&& tar -xzf ${MOFED_DIR}.tgz \
47+
&& ${MOFED_DIR}/mlnxofedinstall --basic -q \
48+
--user-space-only \
49+
--without-fw-update \
50+
--skip-distro-check \
51+
--without-ucx \
52+
--without-hcoll \
53+
--without-openmpi \
54+
--without-sharp \
55+
--distro ${MOFED_OS} \
56+
# MOFED sets memlock unlimited (required for RDMA runtime), but this breaks su in
57+
# unprivileged containers. Safe to remove for CI build containers.
58+
&& sed -i '/memlock/d' /etc/security/limits.conf \
59+
&& rm -rf ${MOFED_DIR} *.tgz \
60+
&& cd /usr/lib64 \
61+
&& ln -s libudev.so.1 libudev.so \
62+
&& ln -s libz.so.1 libz.so

‎buildlib/pr/cuda/cuda.yml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -96,6 +96,10 @@ jobs:
9696
CONTAINER: ubuntu22_cuda_13_0
9797
ubuntu24_cuda_13_0:
9898
CONTAINER: ubuntu24_cuda_13_0
99+
rocky8_cuda_13_0:
100+
CONTAINER: rocky8_cuda_13_0
101+
rocky9_cuda_13_0:
102+
CONTAINER: rocky9_cuda_13_0
99103

100104
container: $[ variables['CONTAINER'] ]
101105
timeoutInMinutes: 35

‎buildlib/pr/main.yml‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -177,6 +177,12 @@ resources:
177177
- container: ubuntu24_cuda_13_0
178178
image: nvidia/cuda:13.0.0-devel-ubuntu24.04
179179
options: $(DOCKER_OPT_ARGS) $(DOCKER_OPT_GPU)
180+
- container: rocky8_cuda_13_0
181+
image: nvidia/cuda:13.0.0-devel-rockylinux8
182+
options: $(DOCKER_OPT_ARGS) $(DOCKER_OPT_GPU)
183+
- container: rocky9_cuda_13_0
184+
image: nvidia/cuda:13.0.0-devel-rockylinux9
185+
options: $(DOCKER_OPT_ARGS) $(DOCKER_OPT_GPU)
180186
- container: centos8_cuda11
181187
image: rdmz-harbor.rdmz.labs.mlnx/ucx/centos8-mofed5-cuda11:1
182188
options: $(DOCKER_OPT_ARGS) $(DOCKER_OPT_VOLUMES) $(DOCKER_OPT_GPU)

‎config/cuda.am‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -31,3 +31,9 @@ endef
3131

3232
CUDA_DEP_FILES := $(shell find $(DEPDIR)/cuda/ -type f -name *.d 2>/dev/null)
3333
-include $(CUDA_DEP_FILES)
34+
35+
clean-local:
36+
-rm -rf $(DEPDIR)/cuda
37+
38+
distclean-local:
39+
-rm -rf $(DEPDIR)/cuda

‎configure.ac‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -10,7 +10,7 @@
1010
AC_PREREQ([2.63])
1111

1212
define([ucx_ver_major], 1) # Major version. Usually does not change.
13-
define([ucx_ver_minor], 20) # Minor version. Increased for each release.
13+
define([ucx_ver_minor], 21) # Minor version. Increased for each release.
1414
define([ucx_ver_patch], 0) # Patch version. Increased for a bugfix release.
1515
define([ucx_ver_extra], ) # Extra version string. Empty for a general release.
1616

0 commit comments

Comments
 (0)