deepcrayon
/
tinyrocs
1
0
Fork 0
Code Issues Activity

ucx, openmp, openmpi notes

main
Jeff Moe 2024-02-07 14:46:32 -07:00
parent da5f1a33bc
commit 55cbdd2c66
6 changed files with 224 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
docs/_source/_static/benchmarks/build-ucx.sh 100644
View File

@ -0,0 +1,37 @@
# XXX Fail
mkdir ucx # extra dir, dpkg writes to parent dir
cd ucx/
git clone https://github.com/openucx/ucx
cd ucx/
git checkout v1.15.0
# XXX cruft
cat > debian/rules.in <<EOF
%:
dh \$@
override_dh_auto_configure:
./contrib/configure-release-mt --prefix=/usr \
--enable-examples --with-java=no \
--enable-mt --with-rocm=/opt/rocm
chmod +x debian/rules
override_dh_auto_install:
dh_auto_install --destdir=debian/tmp
override_dh_shlibdeps:
dh_shlibdeps --dpkg-shlibdeps-params=--ignore-missing-info
if [ -e debian/ucx-cuda.substvars ]; then \
sed -i -e 's/libnvidia-compute-\([0-9]\+\)/& | libnvidia-ml1/' \
debian/ucx-cuda.substvars \
; fi
EOF
CC="clang" CXX="clang++" CFLAGS="-Wno-error" ./autogen.sh
CC="clang" CXX="clang++" CFLAGS="-Wno-error" ./configure --enable-mt --with-rocm=/opt/rocm
CC="clang" CXX="clang++" CFLAGS="-Wno-error" dpkg-buildpackage -rfakeroot -b -uc
# sudo apt purge --autoremove libucx-dev libucx0 ucx-utils
sudo dpkg -i ucx_1.15.348d14f_amd64.deb
exit
# configure: WARNING: HIP Runtime not found

3
docs/_source/os.rst → docs/_source/debian.rst
View File

@ -47,7 +47,8 @@ Install dependencies from Debian repositories.
environment-modules python3-numpy pybind11-dev libopengl-dev zip zsh \
hpcc gawk googletest libdw-dev libgtest-dev libsigsegv2 \
libbabeltrace-dev libbabeltrace1 libbison-dev libncurses5-dev \
libtext-unidecode-perl tex-common texinfo
libtext-unidecode-perl tex-common texinfo ucx-utils libucx-dev \
librdmacm-dev
# Packages like this aren't used from Debian's repository.
# Make sure they are gone.

2
docs/_source/index.rst
View File

@ -16,7 +16,7 @@ tinyrocs is currently working with AMD 7900 XTX GPUs and ``tinygrad``!
about
hardware
firmware
os
debian
kernel
toolchain-6.0.2
benchmarks

123
docs/_source/locale/en/LC_MESSAGES/debian.po 100644
View File

@ -0,0 +1,123 @@
# SOME DESCRIPTIVE TITLE.
# Copyright (C) 2023, 2024 Jeff Moe
# This file is distributed under the same license as the tinyrocs: Direct to
# Chip Liquid Cooled GPU AI Cluster package.
# FIRST AUTHOR <EMAIL@ADDRESS>, 2024.
#
#, fuzzy
msgid ""
msgstr ""
"Project-Id-Version: tinyrocs: Direct to Chip Liquid Cooled GPU AI Cluster 0\n"
"Report-Msgid-Bugs-To: \n"
"POT-Creation-Date: 2024-02-07 14:45-0700\n"
"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
"Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
"Language: en\n"
"Language-Team: en <LL@li.org>\n"
"Plural-Forms: nplurals=2; plural=(n != 1);\n"
"MIME-Version: 1.0\n"
"Content-Type: text/plain; charset=utf-8\n"
"Content-Transfer-Encoding: 8bit\n"
"Generated-By: Babel 2.14.0\n"
#: ../../../_source/debian.rst:3
msgid "Debian Operating System"
msgstr ""
#: ../../../_source/debian.rst:4
msgid "The tinyrocs operating system (OS) is based on Debian GNU/Linux."
msgstr ""
#: ../../../_source/debian.rst:8
msgid "Debian"
msgstr ""
#: ../../../_source/debian.rst:9
msgid "Debian stable (bookworm/12) is the base installation."
msgstr ""
#: ../../../_source/debian.rst:13
msgid "Installation"
msgstr ""
#: ../../../_source/debian.rst:14
msgid "Do a minimal Debian server installation, no GUI, just ssh."
msgstr ""
#: ../../../_source/debian.rst:18
msgid "Dependencies"
msgstr ""
#: ../../../_source/debian.rst:19
msgid "Install dependencies from Debian repositories."
msgstr ""
#: ../../../_source/debian.rst:62
msgid "OS Configuration"
msgstr ""
#: ../../../_source/debian.rst:63
msgid "Operating system configuration."
msgstr ""
#: ../../../_source/debian.rst:79
msgid "User Configuration"
msgstr ""
#: ../../../_source/debian.rst:80
msgid ""
"Set up the user account. Configure to use various caching services already "
"available in the cluster."
msgstr ""
#: ../../../_source/debian.rst:85
msgid "ccache"
msgstr ""
#: ../../../_source/debian.rst:87
msgid ""
"There is a ``redis`` ``ccache`` server on the tinyrocs network. Edit ``~/."
"config/ccache/ccache.conf`` thusly:"
msgstr ""
#: ../../../_source/debian.rst:98
msgid "PATH"
msgstr ""
#: ../../../_source/debian.rst:100
msgid "Add the ROCm binary path and ccache (XXX) to ``~/.bashrc``:"
msgstr ""
#: ../../../_source/debian.rst:108
msgid "Python pip cache"
msgstr ""
#: ../../../_source/debian.rst:110
msgid ""
"Set up to use LAN ``pip`` cache ``pydev`` if available, by editing ``~/."
"config/pip/pip.conf``, such as:"
msgstr ""
#: ../../../_source/debian.rst:124
msgid "Monitoring and Control"
msgstr ""
#: ../../../_source/debian.rst:125
msgid "Applications to monitor and control the hardware."
msgstr ""
#: ../../../_source/debian.rst:127
msgid "Most require dependenices built first."
msgstr ""
#: ../../../_source/debian.rst:131
msgid "nvtop"
msgstr ""
#: ../../../_source/debian.rst:132
msgid "``nvtop`` is nice to quickly visualize the GPUs in a text console."
msgstr ""
#: ../../../_source/debian.rst:134
msgid "Something like:"
msgstr ""

29
docs/_source/locale/en/LC_MESSAGES/toolchain-6.0.2.po
View File

@ -8,7 +8,7 @@ msgid ""
msgstr ""
"Project-Id-Version: tinyrocs 0\n"
"Report-Msgid-Bugs-To: \n"
"POT-Creation-Date: 2024-02-07 13:20-0700\n"
"POT-Creation-Date: 2024-02-07 14:44-0700\n"
"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
"Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
"Language: en\n"
@ -347,3 +347,30 @@ msgstr ""
#: ../../../_source/toolchain-6.0.2.rst:270
msgid "``hipBLAS`` plz."
msgstr ""
#: ../../../_source/toolchain-6.0.2.rst:287
msgid "OpenMP"
msgstr ""
#: ../../../_source/toolchain-6.0.2.rst:288
msgid ""
"OpenMP can be built as a part of LLVM, but it fails in a first pass build. "
"It can be built (perhaps) a rebuild of LLVM with LLVm."
msgstr ""
#: ../../../_source/toolchain-6.0.2.rst:291
msgid "OpenMP repos, check rebuild."
msgstr ""
#: ../../../_source/toolchain-6.0.2.rst:295
msgid "OpenMPI"
msgstr ""
#: ../../../_source/toolchain-6.0.2.rst:296
msgid ""
"OpenMPI, not to be confused with OpenMP. They can be used independently of "
"each other, or used together. Briefly, OpenMP parallelizes across one "
"machine (GPUs/CPUs), and OpenMPI parallelizes across multiple machines "
"(network). Both rebuilt for ROCm is ideal. Then applications need to be "
"built against them."
msgstr ""

33
docs/_source/toolchain-6.0.2.rst
View File

@ -273,6 +273,39 @@ hipBLAS
:language: bash
.. UCX
.. ---
.. UCX is in Debian, but it doesn't support ROCm for ``gfx1100``,
.. so it needs re-building.
.. OpenMPI needs UCX.
..
.. .. literalinclude:: _static/toolchain/rocm-6.0.2/build-ucx.sh
.. :language: bash
OpenMP
------
OpenMP can be built as a part of LLVM, but it fails in a first pass build.
It can be built (perhaps) a rebuild of LLVM with LLVm.
OpenMP repos, check rebuild.
OpenMPI
-------
OpenMPI, not to be confused with OpenMP.
They can be used independently of each other, or used together.
Briefly, OpenMP parallelizes across one machine (GPUs/CPUs),
and OpenMPI parallelizes across multiple machines (network).
Both rebuilt for ROCm is ideal. Then applications need to be
built against them.
.. rccl
.. ----
.. rccl, TODO.
.. aomp
.. ----
.. ``aomp``.