From 17168e6f0bdf92f00cd26674031d265da492e071 Mon Sep 17 00:00:00 2001
From: PRUVOST Florent <florent.pruvost@inria.fr>
Date: Fri, 28 Feb 2020 11:21:36 +0100
Subject: [PATCH] Update to new kibana server and scripts factorization
---
.gitlab/bench_plafrim.yml | 67 ++++++-----
.gitlab/common.yml | 20 ++++
tools/bench/chameleon_guix.sh | 13 +--
tools/bench/chameleon_guix_nmad.sl | 22 ----
tools/bench/chameleon_guix_openmpi.sl | 22 ----
tools/bench/jube/add_result.py | 50 ++++----
tools/bench/plafrim/bora/chameleon_nmad.xml | 108 ------------------
.../bench/plafrim/bora/chameleon_openmpi.xml | 108 ------------------
tools/bench/plafrim/bora/run.sh | 70 ------------
tools/bench/plafrim/chameleon.xml | 56 +++++++++
tools/bench/plafrim/miriel/chameleon_nmad.xml | 108 ------------------
.../plafrim/miriel/chameleon_openmpi.xml | 108 ------------------
tools/bench/plafrim/miriel/run.sh | 70 ------------
.../plafrim/parameters/bora/parameters.xml | 54 +++++++++
.../plafrim/parameters/miriel/parameters.xml | 54 +++++++++
.../plafrim/parameters/sirocco/parameters.xml | 54 +++++++++
tools/bench/plafrim/run.sh | 83 ++++++++++++++
tools/bench/plafrim/sirocco/chameleon.xml | 108 ------------------
tools/bench/plafrim/sirocco/run.sh | 70 ------------
tools/bench/plafrim/slurm.sh | 59 ++++++++++
20 files changed, 450 insertions(+), 854 deletions(-)
delete mode 100644 tools/bench/chameleon_guix_nmad.sl
delete mode 100644 tools/bench/chameleon_guix_openmpi.sl
delete mode 100644 tools/bench/plafrim/bora/chameleon_nmad.xml
delete mode 100644 tools/bench/plafrim/bora/chameleon_openmpi.xml
delete mode 100755 tools/bench/plafrim/bora/run.sh
create mode 100644 tools/bench/plafrim/chameleon.xml
delete mode 100644 tools/bench/plafrim/miriel/chameleon_nmad.xml
delete mode 100644 tools/bench/plafrim/miriel/chameleon_openmpi.xml
delete mode 100755 tools/bench/plafrim/miriel/run.sh
create mode 100644 tools/bench/plafrim/parameters/bora/parameters.xml
create mode 100644 tools/bench/plafrim/parameters/miriel/parameters.xml
create mode 100644 tools/bench/plafrim/parameters/sirocco/parameters.xml
create mode 100755 tools/bench/plafrim/run.sh
delete mode 100644 tools/bench/plafrim/sirocco/chameleon.xml
delete mode 100755 tools/bench/plafrim/sirocco/run.sh
create mode 100755 tools/bench/plafrim/slurm.sh
diff --git a/.gitlab/bench_plafrim.yml b/.gitlab/bench_plafrim.yml
index 24d9f5a7e..4c9cb4394 100644
--- a/.gitlab/bench_plafrim.yml
+++ b/.gitlab/bench_plafrim.yml
@@ -1,29 +1,42 @@
-bench_plafrim_bora:
- stage: test
- tags: ['plafrim']
- before_script:
- - git submodule update --init --recursive
- script:
- - ./tools/bench/plafrim/bora/run.sh
- only:
- - schedules
+---
+bench_plafrim_bora_openmpi:
+ variables:
+ NODE: bora
+ MPI: openmpi
+ SLURM_NP: 9
+ JUBE_ID: "000001"
+ extends: .bench_plafrim_common
-.bench_plafrim_miriel:
- stage: test
- tags: ['plafrim']
- before_script:
- - git submodule update --init --recursive
- script:
- - ./tools/bench/plafrim/miriel/run.sh
- only:
- - schedules
+bench_plafrim_miriel_openmpi:
+ variables:
+ NODE: miriel
+ MPI: openmpi
+ SLURM_NP: 9
+ JUBE_ID: "000002"
+ extends: .bench_plafrim_common
-.bench_plafrim_sirocco:
- stage: test
- tags: ['plafrim']
- before_script:
- - git submodule update --init --recursive
- script:
- - ./tools/bench/plafrim/sirocco/run.sh
- only:
- - schedules
\ No newline at end of file
+# Nmad segfaults for now, waiting for a stable version
+.bench_plafrim_bora_nmad:
+ variables:
+ NODE: bora
+ MPI: nmad
+ SLURM_NP: 9
+ JUBE_ID: "000003"
+ extends: .bench_plafrim_common
+
+.bench_plafrim_miriel_nmad:
+ variables:
+ NODE: miriel
+ MPI: nmad
+ SLURM_NP: 9
+ JUBE_ID: "000004"
+ extends: .bench_plafrim_common
+
+# Chameleon+CUDA in Guix is not working yet
+.bench_plafrim_sirocco_openmpi:
+ variables:
+ NODE: sirocco
+ MPI: openmpi
+ SLURM_NP: 1
+ JUBE_ID: "000005"
+ extends: .bench_plafrim_common
\ No newline at end of file
diff --git a/.gitlab/common.yml b/.gitlab/common.yml
index 5960f8c8b..ed2298842 100644
--- a/.gitlab/common.yml
+++ b/.gitlab/common.yml
@@ -63,3 +63,23 @@
- (cd build-$VERSION && lcov --directory . --capture --output-file ../bis_${LOGNAME}.lcov)
except:
- schedules
+
+.bench_plafrim_common:
+ only:
+ - schedules
+ stage: test
+ tags: ['plafrim']
+ before_script:
+ - git submodule update --init --recursive
+ script:
+ - ./tools/bench/plafrim/run.sh
+ artifacts:
+ name: "$CI_JOB_NAME"
+ expire_in: 1 week
+ paths:
+ - "chameleon-$NODE-$MPI-$SLURM_NP.err"
+ - "chameleon-$NODE-$MPI-$SLURM_NP.out"
+ - "build-$NODE-$MPI/chameleon.csv"
+ - "tools/bench/plafrim/results/$JUBE_ID"
+ variables:
+ PLATFORM: plafrim
diff --git a/tools/bench/chameleon_guix.sh b/tools/bench/chameleon_guix.sh
index a120649c1..5c1c641f6 100755
--- a/tools/bench/chameleon_guix.sh
+++ b/tools/bench/chameleon_guix.sh
@@ -2,24 +2,23 @@
set -x
-# this script depends on a MPI vendor: openmpi or nmad
-MPI=$1
-
# Configure and Build Chameleon
mkdir -p $CI_PROJECT_DIR/build-$NODE-$MPI
cp $CI_PROJECT_DIR/guix.json $CI_PROJECT_DIR/build-$NODE-$MPI/
cd $CI_PROJECT_DIR/build-$NODE-$MPI
rm CMake* -rf
-cmake $BUILD_OPTIONS ..
+cmake $CHAMELEON_BUILD_OPTIONS ..
make -j20 VERBOSE=1
export CHAMELEON_BUILD=$PWD
+# clean old benchmarks
+jube remove --id $JUBE_ID
# Execute jube benchmarks
-jube run $CI_PROJECT_DIR/tools/bench/$PLATFORM/$NODE/chameleon_$MPI.xml --tag gemm potrf geqrf
+jube run $CI_PROJECT_DIR/tools/bench/$PLATFORM/chameleon.xml --tag gemm potrf geqrf --include-path $CI_PROJECT_DIR/tools/bench/$PLATFORM/parameters/$NODE --id $JUBE_ID
# jube analysis
-jube analyse $CI_PROJECT_DIR/tools/bench/$PLATFORM/$NODE/results/
+jube analyse $CI_PROJECT_DIR/tools/bench/$PLATFORM/results --id $JUBE_ID
# jube report
-jube result $CI_PROJECT_DIR/tools/bench/$PLATFORM/$NODE/results/ -i last > chameleon.csv
+jube result $CI_PROJECT_DIR/tools/bench/$PLATFORM/results --id $JUBE_ID > chameleon.csv
# send results to the elasticsearch server
export PYTHONPATH=$GUIX_ENVIRONMENT/lib/python3.7/site-packages
diff --git a/tools/bench/chameleon_guix_nmad.sl b/tools/bench/chameleon_guix_nmad.sl
deleted file mode 100644
index b3b952de2..000000000
--- a/tools/bench/chameleon_guix_nmad.sl
+++ /dev/null
@@ -1,22 +0,0 @@
-#!/usr/bin/env bash
-#SBATCH --exclusive
-#SBATCH --ntasks-per-node=1
-#SBATCH --threads-per-core=1
-
-echo "######################### Chameleon benchmarks #########################"
-
-# to avoid a lock during fetching chameleon branch in parallel
-export XDG_CACHE_HOME=/tmp/guix-$$
-
-# save guix commits
-guix describe --format=json > guix.json
-
-# Submit jobs
-
-# Nmad version
-exec guix environment --pure --preserve=^CI --preserve=^SLURM --preserve=^STARPU --preserve=^PADICO --preserve=^PIOM --preserve=PLATFORM --preserve=NODE --preserve=BUILD_OPTIONS chameleon --with-input=openblas=mkl --with-input=openmpi=nmad --ad-hoc slurm jube python python-click python-gitpython python-elasticsearch python-certifi coreutils inetutils util-linux procps grep tar sed gzip which gawk perl zlib openssh hwloc nmad starpu mkl -- /bin/bash --norc $CI_PROJECT_DIR/tools/bench/chameleon_guix.sh nmad
-
-echo "####################### End Chameleon benchmarks #######################"
-
-# clean tmp
-rm -rf /tmp/guix-$$
diff --git a/tools/bench/chameleon_guix_openmpi.sl b/tools/bench/chameleon_guix_openmpi.sl
deleted file mode 100644
index fd84573af..000000000
--- a/tools/bench/chameleon_guix_openmpi.sl
+++ /dev/null
@@ -1,22 +0,0 @@
-#!/usr/bin/env bash
-#SBATCH --exclusive
-#SBATCH --ntasks-per-node=1
-#SBATCH --threads-per-core=1
-
-echo "######################### Chameleon benchmarks #########################"
-
-# to avoid a lock during fetching chameleon branch in parallel
-export XDG_CACHE_HOME=/tmp/guix-$$
-
-# save guix commits
-guix describe --format=json > guix.json
-
-# Submit jobs
-
-# OpenMPI version
-exec guix environment --pure --preserve=^CI --preserve=^SLURM --preserve=^STARPU --preserve=PLATFORM --preserve=NODE --preserve=BUILD_OPTIONS chameleon --with-input=openblas=mkl --ad-hoc slurm jube python python-click python-gitpython python-elasticsearch python-certifi sed coreutils grep gawk openssh perl hwloc openmpi starpu mkl -- /bin/bash --norc $CI_PROJECT_DIR/tools/bench/chameleon_guix.sh openmpi
-
-echo "####################### End Chameleon benchmarks #######################"
-
-# clean tmp
-rm -rf /tmp/guix-$$
diff --git a/tools/bench/jube/add_result.py b/tools/bench/jube/add_result.py
index 63a68d6d2..4f27dc483 100755
--- a/tools/bench/jube/add_result.py
+++ b/tools/bench/jube/add_result.py
@@ -92,37 +92,35 @@ def main(
):
"""Add a result to an elasticsearch database."""
es = Elasticsearch(elastic_url)
- es_index = team + "_" + project + "_" + "perf"
+ es_index = team + "-" + project + "_" + "perf"
if not es.indices.exists(es_index):
es.indices.create(es_index)
mapping_input = {
- "result": {
- "properties": {
- "Commit_date_chameleon": {"type": "date", "format": "yyyy-MM-dd' 'HH:mm:ss"},
- "Commit_sha_chameleon": {"type": "keyword"},
- "Commit_sha_guix": {"type": "keyword"},
- "Commit_sha_guix_hpc": {"type": "keyword"},
- "Commit_sha_guix_hpcnonfree": {"type": "keyword"},
- "Hostname": {"type": "keyword"},
- "MPIvendor": {"type": "keyword"},
- "Algorithm": {"type": "keyword"},
- "Precision": {"type": "keyword"},
- "Nmpi": {"type": "integer"},
- "P": {"type": "integer"},
- "Q": {"type": "integer"},
- "Nthread": {"type": "integer"},
- "Ngpu": {"type": "integer"},
- "M": {"type": "integer"},
- "N": {"type": "integer"},
- "K": {"type": "integer"},
- "Cputime": {"type": "float"},
- "Gflops": {"type": "float"}
- }
+ "properties": {
+ "Commit_date_chameleon": {"type": "date", "format": "yyyy-MM-dd HH:mm:ss"},
+ "Commit_sha_chameleon": {"type": "keyword"},
+ "Commit_sha_guix": {"type": "keyword"},
+ "Commit_sha_guix_hpc": {"type": "keyword"},
+ "Commit_sha_guix_hpcnonfree": {"type": "keyword"},
+ "Hostname": {"type": "keyword"},
+ "MPIvendor": {"type": "keyword"},
+ "Algorithm": {"type": "keyword"},
+ "Precision": {"type": "keyword"},
+ "Nmpi": {"type": "integer"},
+ "P": {"type": "integer"},
+ "Q": {"type": "integer"},
+ "Nthread": {"type": "integer"},
+ "Ngpu": {"type": "integer"},
+ "M": {"type": "integer"},
+ "N": {"type": "integer"},
+ "K": {"type": "integer"},
+ "Cputime": {"type": "float"},
+ "Gflops": {"type": "float"}
}
+
}
- # es.indices.put_mapping(index=es_index, doc_type="result" , body=mapping_input, include_type_name=True)
- es.indices.put_mapping(index=es_index, doc_type="result" , body=mapping_input)
+ es.indices.put_mapping(index=es_index, body=mapping_input)
repo = Repo(directory, search_parent_directories=True)
commit_chameleon = repo.head.commit
@@ -147,7 +145,7 @@ def main(
)
]
for request in requests:
- es.index(index=es_index.lower(), doc_type="result", body=request)
+ es.index(index=es_index.lower(), body=request)
if __name__ == "__main__":
diff --git a/tools/bench/plafrim/bora/chameleon_nmad.xml b/tools/bench/plafrim/bora/chameleon_nmad.xml
deleted file mode 100644
index 522da430c..000000000
--- a/tools/bench/plafrim/bora/chameleon_nmad.xml
+++ /dev/null
@@ -1,108 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<jube>
- <benchmark name="bora" outpath="results">
- <comment>benchmark chameleon on host plafrim bora</comment>
-
- <parameterset name="param_gemm">
- <parameter name="hostname" type="string">bora</parameter>
- <parameter name="algorithm" type="string">gemm</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >34</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- </parameterset>
-
- <parameterset name="param_potrf">
- <parameter name="hostname" type="string">bora</parameter>
- <parameter name="algorithm" type="string">potrf</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >34</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <parameterset name="param_geqrf">
- <parameter name="hostname" type="string">bora</parameter>
- <parameter name="algorithm" type="string">geqrf_hqr</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >34</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <!-- Operation -->
- <step name="run_gemm" tag="gemm">
- <use>param_gemm</use>
- <do>mpiexec -DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1 -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_potrf" tag="potrf">
- <use>param_potrf</use>
- <do>mpiexec -DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1 -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_geqrf_hqr" tag="geqrf">
- <use>param_geqrf</use>
- <do>mpiexec -DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1 -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
-
- <!-- Analyse -->
- <analyser name="analyse">
- <!-- use a pattern set -->
- <use from="../../jube/patterns.xml">chameleon</use>
- <analyse step="run_gemm" tag="gemm">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_potrf" tag="potrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_geqrf_hqr" tag="geqrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- </analyser>
-
-
- <!-- Create result table -->
- <result>
- <use>analyse</use> <!-- use existing analyser -->
- <!--<table name="result" style="csv" sort="number">-->
- <table name="result" style="csv">
- <column>hostname</column>
- <column>algorithm</column>
- <column>precision</column>
- <column>nmpi</column>
- <column>p</column>
- <column>q</column>
- <column>nthr</column>
- <column>ngpu</column>
- <column>m</column>
- <column>n</column>
- <column>k</column>
- <column>cputime</column>
- <column>gflops</column>
- </table>
- </result>
- </benchmark>
-</jube>
diff --git a/tools/bench/plafrim/bora/chameleon_openmpi.xml b/tools/bench/plafrim/bora/chameleon_openmpi.xml
deleted file mode 100644
index fe80ac54b..000000000
--- a/tools/bench/plafrim/bora/chameleon_openmpi.xml
+++ /dev/null
@@ -1,108 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<jube>
- <benchmark name="bora" outpath="results">
- <comment>benchmark chameleon on host plafrim bora</comment>
-
- <parameterset name="param_gemm">
- <parameter name="hostname" type="string">bora</parameter>
- <parameter name="algorithm" type="string">gemm</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >34</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- </parameterset>
-
- <parameterset name="param_potrf">
- <parameter name="hostname" type="string">bora</parameter>
- <parameter name="algorithm" type="string">potrf</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >34</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <parameterset name="param_geqrf">
- <parameter name="hostname" type="string">bora</parameter>
- <parameter name="algorithm" type="string">geqrf_hqr</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >34</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <!-- Operation -->
- <step name="run_gemm" tag="gemm">
- <use>param_gemm</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_potrf" tag="potrf">
- <use>param_potrf</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_geqrf_hqr" tag="geqrf">
- <use>param_geqrf</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
-
- <!-- Analyse -->
- <analyser name="analyse">
- <!-- use a pattern set -->
- <use from="../../jube/patterns.xml">chameleon</use>
- <analyse step="run_gemm" tag="gemm">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_potrf" tag="potrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_geqrf_hqr" tag="geqrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- </analyser>
-
-
- <!-- Create result table -->
- <result>
- <use>analyse</use> <!-- use existing analyser -->
- <!--<table name="result" style="csv" sort="number">-->
- <table name="result" style="csv">
- <column>hostname</column>
- <column>algorithm</column>
- <column>precision</column>
- <column>nmpi</column>
- <column>p</column>
- <column>q</column>
- <column>nthr</column>
- <column>ngpu</column>
- <column>m</column>
- <column>n</column>
- <column>k</column>
- <column>cputime</column>
- <column>gflops</column>
- </table>
- </result>
- </benchmark>
-</jube>
diff --git a/tools/bench/plafrim/bora/run.sh b/tools/bench/plafrim/bora/run.sh
deleted file mode 100755
index baf9e7424..000000000
--- a/tools/bench/plafrim/bora/run.sh
+++ /dev/null
@@ -1,70 +0,0 @@
-#!/bin/bash
-
-echo "######################### Chameleon benchmarks #########################"
-echo "HOSTNAME $HOSTNAME"
-echo "USERNAME $USERNAME"
-echo "GIT REPO $CI_REPOSITORY_URL"
-echo "GIT BRANCH $CI_COMMIT_REF_NAME"
-echo "GIT COMMIT $CI_COMMIT_SHA"
-echo "PROJECT DIR $CI_PROJECT_DIR"
-
-set -x
-
-function wait_completion {
- # Wait for completion of jobs
- echo "JOB_LIST $JOB_LIST"
- while [ "$NJOB" -gt 0 ]
- do
- for JOB in $JOB_LIST
- do
- IS_JOB_IN_QUEUE=`squeue |grep "$JOB"`
- if [[ -z "$IS_JOB_IN_QUEUE" ]]
- then
- NJOB=$[NJOB-1]
- JOB_LIST=`echo $JOB_LIST | sed "s#$JOB##"`
- echo "JOB $JOB finished"
- else
- echo "$IS_JOB_IN_QUEUE"
- fi
- done
- sleep 30
- done
-}
-
-# Parameters for scripts
-export PLATFORM=plafrim
-export NODE=bora
-export BUILD_OPTIONS="-DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
-export STARPU_SILENT=1
-#export STARPU_LIMIT_CPU_MEM=180000
-#export STARPU_LIMIT_MAX_SUBMITTED_TASKS=16000
-#export STARPU_LIMIT_MIN_SUBMITTED_TASKS=15000
-
-# Parameters of the Slurm jobs
-TIME=01:00:00
-PART=routage
-CONS=bora
-EXCL=
-NP=9
-
-# Submit jobs
-NJOB=0
-MPI_LIST="openmpi nmad"
-#MPI_LIST="nmad"
-for MPI in $MPI_LIST
-do
- JOB_ID=`JOB_NAME=chameleon\_$MPI\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL $CI_PROJECT_DIR/tools/bench/chameleon_guix\_$MPI.sl | sed "s#Submitted batch job ##"`
- #JOB_ID=`JOB_NAME=chameleon\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL chameleon_guix.sl | sed "s#Submitted batch job ##"`
- if [[ -n "$JOB_ID" ]]
- then
- JOB_LIST="$JOB_LIST $JOB_ID"
- NJOB=$[NJOB+1]
- fi
-done
-
-# Wait for completion of jobs
-wait_completion
-
-echo "####################### End Chameleon benchmarks #######################"
-
-exit 0
diff --git a/tools/bench/plafrim/chameleon.xml b/tools/bench/plafrim/chameleon.xml
new file mode 100644
index 000000000..1fd85508b
--- /dev/null
+++ b/tools/bench/plafrim/chameleon.xml
@@ -0,0 +1,56 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<jube>
+ <benchmark name="plafrim" outpath="results">
+ <comment>benchmark chameleon on plafrim</comment>
+
+ <!-- Operation -->
+ <step name="run_gemm" tag="gemm">
+ <use from= "parameters.xml">param_gemm</use>
+ <do>$command</do>
+ </step>
+ <step name="run_potrf" tag="potrf">
+ <use from= "parameters.xml">param_potrf</use>
+ <do>$command</do>
+ </step>
+ <step name="run_geqrf_hqr" tag="geqrf">
+ <use from= "parameters.xml">param_geqrf</use>
+ <do>$command</do>
+ </step>
+
+ <!-- Analyse -->
+ <analyser name="analyse">
+ <!-- use a pattern set -->
+ <use from="../jube/patterns.xml">chameleon</use>
+ <analyse step="run_gemm" tag="gemm">
+ <file>stdout</file> <!-- file which should be scanned -->
+ </analyse>
+ <analyse step="run_potrf" tag="potrf">
+ <file>stdout</file> <!-- file which should be scanned -->
+ </analyse>
+ <analyse step="run_geqrf_hqr" tag="geqrf">
+ <file>stdout</file> <!-- file which should be scanned -->
+ </analyse>
+ </analyser>
+
+ <!-- Create result table -->
+ <result>
+ <use>analyse</use> <!-- use existing analyser -->
+ <!--<table name="result" style="csv" sort="number">-->
+ <table name="result" style="csv">
+ <column>hostname</column>
+ <column>algorithm</column>
+ <column>precision</column>
+ <column>nmpi</column>
+ <column>p</column>
+ <column>q</column>
+ <column>nthr</column>
+ <column>ngpu</column>
+ <column>m</column>
+ <column>n</column>
+ <column>k</column>
+ <column>cputime</column>
+ <column>gflops</column>
+ </table>
+ </result>
+ </benchmark>
+</jube>
diff --git a/tools/bench/plafrim/miriel/chameleon_nmad.xml b/tools/bench/plafrim/miriel/chameleon_nmad.xml
deleted file mode 100644
index 28be9ef0b..000000000
--- a/tools/bench/plafrim/miriel/chameleon_nmad.xml
+++ /dev/null
@@ -1,108 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<jube>
- <benchmark name="miriel" outpath="results/miriel">
- <comment>benchmark chameleon on host plafrim miriel</comment>
-
- <parameterset name="param_gemm">
- <parameter name="hostname" type="string">miriel</parameter>
- <parameter name="algorithm" type="string">gemm</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >22</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- </parameterset>
-
- <parameterset name="param_potrf">
- <parameter name="hostname" type="string">miriel</parameter>
- <parameter name="algorithm" type="string">potrf</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >22</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <parameterset name="param_geqrf">
- <parameter name="hostname" type="string">miriel</parameter>
- <parameter name="algorithm" type="string">geqrf_hqr</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >22</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <!-- Operation -->
- <step name="run_gemm" tag="gemm">
- <use>param_gemm</use>
- <do>mpiexec -DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1 -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_potrf" tag="potrf">
- <use>param_potrf</use>
- <do>mpiexec -DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1 -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_geqrf_hqr" tag="geqrf">
- <use>param_geqrf</use>
- <do>mpiexec -DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1 -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
-
- <!-- Analyse -->
- <analyser name="analyse">
- <!-- use a pattern set -->
- <use from="../../jube/patterns.xml">chameleon</use>
- <analyse step="run_gemm" tag="gemm">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_potrf" tag="potrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_geqrf_hqr" tag="geqrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- </analyser>
-
-
- <!-- Create result table -->
- <result>
- <use>analyse</use> <!-- use existing analyser -->
- <!--<table name="result" style="csv" sort="number">-->
- <table name="result" style="csv">
- <column>hostname</column>
- <column>algorithm</column>
- <column>precision</column>
- <column>nmpi</column>
- <column>p</column>
- <column>q</column>
- <column>nthr</column>
- <column>ngpu</column>
- <column>m</column>
- <column>n</column>
- <column>k</column>
- <column>cputime</column>
- <column>gflops</column>
- </table>
- </result>
- </benchmark>
-</jube>
diff --git a/tools/bench/plafrim/miriel/chameleon_openmpi.xml b/tools/bench/plafrim/miriel/chameleon_openmpi.xml
deleted file mode 100644
index d6b70cc84..000000000
--- a/tools/bench/plafrim/miriel/chameleon_openmpi.xml
+++ /dev/null
@@ -1,108 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<jube>
- <benchmark name="miriel" outpath="results/miriel">
- <comment>benchmark chameleon on host plafrim miriel</comment>
-
- <parameterset name="param_gemm">
- <parameter name="hostname" type="string">miriel</parameter>
- <parameter name="algorithm" type="string">gemm</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >22</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- </parameterset>
-
- <parameterset name="param_potrf">
- <parameter name="hostname" type="string">miriel</parameter>
- <parameter name="algorithm" type="string">potrf</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >22</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <parameterset name="param_geqrf">
- <parameter name="hostname" type="string">miriel</parameter>
- <parameter name="algorithm" type="string">geqrf_hqr</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0, 1, 2</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >22</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
- <parameter name="b" type="int" >320</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <!-- Operation -->
- <step name="run_gemm" tag="gemm">
- <use>param_gemm</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_potrf" tag="potrf">
- <use>param_potrf</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_geqrf_hqr" tag="geqrf">
- <use>param_geqrf</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
-
- <!-- Analyse -->
- <analyser name="analyse">
- <!-- use a pattern set -->
- <use from="../../jube/patterns.xml">chameleon</use>
- <analyse step="run_gemm" tag="gemm">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_potrf" tag="potrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_geqrf_hqr" tag="geqrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- </analyser>
-
-
- <!-- Create result table -->
- <result>
- <use>analyse</use> <!-- use existing analyser -->
- <!--<table name="result" style="csv" sort="number">-->
- <table name="result" style="csv">
- <column>hostname</column>
- <column>algorithm</column>
- <column>precision</column>
- <column>nmpi</column>
- <column>p</column>
- <column>q</column>
- <column>nthr</column>
- <column>ngpu</column>
- <column>m</column>
- <column>n</column>
- <column>k</column>
- <column>cputime</column>
- <column>gflops</column>
- </table>
- </result>
- </benchmark>
-</jube>
diff --git a/tools/bench/plafrim/miriel/run.sh b/tools/bench/plafrim/miriel/run.sh
deleted file mode 100755
index d37974b03..000000000
--- a/tools/bench/plafrim/miriel/run.sh
+++ /dev/null
@@ -1,70 +0,0 @@
-#!/bin/bash
-
-echo "######################### Chameleon benchmarks #########################"
-echo "HOSTNAME $HOSTNAME"
-echo "USERNAME $USERNAME"
-echo "GIT REPO $CI_REPOSITORY_URL"
-echo "GIT BRANCH $CI_COMMIT_REF_NAME"
-echo "GIT COMMIT $CI_COMMIT_SHA"
-echo "PROJECT DIR $CI_PROJECT_DIR"
-
-set -x
-
-function wait_completion {
- # Wait for completion of jobs
- echo "JOB_LIST $JOB_LIST"
- while [ "$NJOB" -gt 0 ]
- do
- for JOB in $JOB_LIST
- do
- IS_JOB_IN_QUEUE=`squeue |grep "$JOB"`
- if [[ -z "$IS_JOB_IN_QUEUE" ]]
- then
- NJOB=$[NJOB-1]
- JOB_LIST=`echo $JOB_LIST | sed "s#$JOB##"`
- echo "JOB $JOB finished"
- else
- echo "$IS_JOB_IN_QUEUE"
- fi
- done
- sleep 30
- done
-}
-
-# Parameters for scripts
-export PLATFORM=plafrim
-export NODE=miriel
-export BUILD_OPTIONS="-DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
-export STARPU_SILENT=1
-export STARPU_LIMIT_CPU_MEM=120000
-export STARPU_LIMIT_MAX_SUBMITTED_TASKS=16000
-export STARPU_LIMIT_MIN_SUBMITTED_TASKS=15000
-
-# Parameters of the Slurm jobs
-TIME=01:00:00
-PART=routage
-CONS="miriel,omnipath"
-EXCL=
-NP=9
-
-# Submit jobs
-NJOB=0
-MPI_LIST="openmpi nmad"
-#MPI_LIST="nmad"
-for MPI in $MPI_LIST
-do
- JOB_ID=`JOB_NAME=chameleon\_$MPI\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL $CI_PROJECT_DIR/tools/bench/chameleon_guix\_$MPI.sl | sed "s#Submitted batch job ##"`
- #JOB_ID=`JOB_NAME=chameleon\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL chameleon_guix.sl | sed "s#Submitted batch job ##"`
- if [[ -n "$JOB_ID" ]]
- then
- JOB_LIST="$JOB_LIST $JOB_ID"
- NJOB=$[NJOB+1]
- fi
-done
-
-# Wait for completion of jobs
-wait_completion
-
-echo "####################### End Chameleon benchmarks #######################"
-
-exit 0
diff --git a/tools/bench/plafrim/parameters/bora/parameters.xml b/tools/bench/plafrim/parameters/bora/parameters.xml
new file mode 100644
index 000000000..e7e4f1486
--- /dev/null
+++ b/tools/bench/plafrim/parameters/bora/parameters.xml
@@ -0,0 +1,54 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<jube>
+ <parameterset name="param_gemm">
+ <parameter name="hostname" type="string">bora</parameter>
+ <parameter name="algorithm" type="string">gemm</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >34</parameter>
+ <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="b" type="int" >320</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*32*${b}][$i_mn]</parameter>
+ <parameter name="k" mode="python" type="int" >${m}</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+ <parameterset name="param_potrf">
+ <parameter name="hostname" type="string">bora</parameter>
+ <parameter name="algorithm" type="string">potrf</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >34</parameter>
+ <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="b" type="int" >320</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*32*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="k" type="int" >1</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+ <parameterset name="param_geqrf">
+ <parameter name="hostname" type="string">bora</parameter>
+ <parameter name="algorithm" type="string">geqrf_hqr</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >34</parameter>
+ <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="b" type="int" >320</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*32*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="k" type="int" >1</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+</jube>
diff --git a/tools/bench/plafrim/parameters/miriel/parameters.xml b/tools/bench/plafrim/parameters/miriel/parameters.xml
new file mode 100644
index 000000000..00622ac1a
--- /dev/null
+++ b/tools/bench/plafrim/parameters/miriel/parameters.xml
@@ -0,0 +1,54 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<jube>
+ <parameterset name="param_gemm">
+ <parameter name="hostname" type="string">miriel</parameter>
+ <parameter name="algorithm" type="string">gemm</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >22</parameter>
+ <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="b" type="int" >320</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*24*${b}][$i_mn]</parameter>
+ <parameter name="k" mode="python" type="int" >${m}</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+ <parameterset name="param_potrf">
+ <parameter name="hostname" type="string">miriel</parameter>
+ <parameter name="algorithm" type="string">potrf</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >22</parameter>
+ <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="b" type="int" >320</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*24*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="k" type="int" >1</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+ <parameterset name="param_geqrf">
+ <parameter name="hostname" type="string">miriel</parameter>
+ <parameter name="algorithm" type="string">geqrf_hqr</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >22</parameter>
+ <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="b" type="int" >320</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*24*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="k" type="int" >1</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+</jube>
diff --git a/tools/bench/plafrim/parameters/sirocco/parameters.xml b/tools/bench/plafrim/parameters/sirocco/parameters.xml
new file mode 100644
index 000000000..873195b22
--- /dev/null
+++ b/tools/bench/plafrim/parameters/sirocco/parameters.xml
@@ -0,0 +1,54 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<jube>
+ <parameterset name="param_gemm">
+ <parameter name="hostname" type="string">sirocco</parameter>
+ <parameter name="algorithm" type="string">gemm</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >29</parameter>
+ <parameter name="ngpu" type="int" >2</parameter>
+ <parameter name="b" type="int" >1600</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
+ <parameter name="k" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+ <parameterset name="param_potrf">
+ <parameter name="hostname" type="string">sirocco</parameter>
+ <parameter name="algorithm" type="string">potrf</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >29</parameter>
+ <parameter name="ngpu" type="int" >2</parameter>
+ <parameter name="b" type="int" >1600</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*32*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="k" type="int" >1</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+ <parameterset name="param_geqrf">
+ <parameter name="hostname" type="string">sirocco</parameter>
+ <parameter name="algorithm" type="string">geqrf_hqr</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
+ <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
+ <parameter name="nthr" type="int" >29</parameter>
+ <parameter name="ngpu" type="int" >2</parameter>
+ <parameter name="b" type="int" >1600</parameter>
+ <parameter name="i_mn" type="int" >0, 1, 2, 3, 4, 5</parameter>
+ <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*2*${b}, ${nmpi}*4*${b}, ${nmpi}*8*${b}, ${nmpi}*16*${b}, ${nmpi}*32*${b}][$i_mn]</parameter>
+ <parameter name="n" mode="python" type="int" >${m}</parameter>
+ <parameter name="k" type="int" >1</parameter>
+ <parameter name="command" type="string">mpiexec $MPI_OPTIONS -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</parameter>
+ </parameterset>
+</jube>
diff --git a/tools/bench/plafrim/run.sh b/tools/bench/plafrim/run.sh
new file mode 100755
index 000000000..e33126df5
--- /dev/null
+++ b/tools/bench/plafrim/run.sh
@@ -0,0 +1,83 @@
+#!/bin/bash
+
+echo "######################### Chameleon benchmarks #########################"
+
+set -x
+
+# to avoid a lock during fetching chameleon branch in parallel
+export XDG_CACHE_HOME=/tmp/guix-$$
+
+# save guix commits
+guix describe --format=json > guix.json
+
+# define env var depending on the node type
+export STARPU_SILENT=1
+export STARPU_CALIBRATE=0
+export STARPU_COMM_STATS=1
+export STARPU_WORKER_STATS=1
+if [ $NODE = "bora" ]
+then
+ export SLURM_CONSTRAINTS="bora,omnipath"
+ export CHAMELEON_BUILD_OPTIONS="-DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
+ export STARPU_HOSTNAME="bora"
+elif [ $NODE = "miriel" ]
+then
+ export SLURM_CONSTRAINTS="miriel,infinipath"
+ export CHAMELEON_BUILD_OPTIONS="-DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
+ export STARPU_HOSTNAME="miriel"
+elif [ $NODE = "sirocco" ]
+then
+ export SLURM_CONSTRAINTS="sirocco,omnipath,v100"
+ export CHAMELEON_BUILD_OPTIONS="-DCHAMELEON_USE_MPI=ON -DCHAMELEON_USE_CUDA=ON -DCMAKE_BUILD_TYPE=Release"
+ export STARPU_HOSTNAME="sirocco"
+else
+ echo "$0: Please set the NODE environnement variable to bora or miriel or sirocco."
+ exit -1
+fi
+
+# define env var and guix rule to use depending on the mpi vendor
+GUIX_ENV="chameleon --with-input=openblas=mkl"
+if [ $NODE = "sirocco" ]
+then
+ GUIX_ENV="chameleon-cuda --with-input=openblas=mkl"
+fi
+if [ $MPI = "openmpi" ]
+then
+ export MPI_OPTIONS=""
+ if [ $NODE = "miriel" ]
+ then
+ export MPI_OPTIONS="--mca mtl psm"
+ fi
+ GUIX_ENV_MPI=""
+ GUIX_ADHOC_MPI="openssh openmpi"
+elif [ $MPI = "nmad" ]
+then
+ export MPI_OPTIONS="-DPIOM_DEDICATED=1 -DPIOM_DEDICATED_WAIT=1"
+ GUIX_ENV_MPI="--with-input=openmpi=nmad"
+ GUIX_ADHOC_MPI="which gzip zlib tar inetutils util-linux procps openssh nmad"
+else
+ echo "$0: Please set the MPI environnement variable to openmpi or nmad."
+ exit -1
+fi
+GUIX_ADHOC="slurm jube python python-click python-gitpython python-elasticsearch python-certifi sed coreutils grep gawk perl"
+GUIX_RULE="$GUIX_ENV $GUIX_ENV_MPI --ad-hoc $GUIX_ADHOC $GUIX_ADHOC_MPI"
+
+# Submit jobs
+
+# OpenMPI version
+exec guix environment --pure \
+ --preserve=PLATFORM \
+ --preserve=NODE \
+ --preserve=^CI \
+ --preserve=^SLURM \
+ --preserve=^JUBE \
+ --preserve=^MPI \
+ --preserve=^STARPU \
+ --preserve=^CHAMELEON \
+ $GUIX_RULE \
+ -- /bin/bash --norc ./tools/bench/plafrim/slurm.sh
+
+echo "####################### End Chameleon benchmarks #######################"
+
+# clean tmp
+rm -rf /tmp/guix-$$
\ No newline at end of file
diff --git a/tools/bench/plafrim/sirocco/chameleon.xml b/tools/bench/plafrim/sirocco/chameleon.xml
deleted file mode 100644
index fda2c21c7..000000000
--- a/tools/bench/plafrim/sirocco/chameleon.xml
+++ /dev/null
@@ -1,108 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<jube>
- <benchmark name="sirocco" outpath="results/sirocco">
- <comment>benchmark chameleon on host plafrim sirocco</comment>
-
- <parameterset name="param_gemm">
- <parameter name="hostname" type="string">sirocco</parameter>
- <parameter name="algorithm" type="string">gemm</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >29</parameter>
- <parameter name="ngpu" type="int" >2</parameter>
- <parameter name="b" type="int" >1600</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- </parameterset>
-
- <parameterset name="param_potrf">
- <parameter name="hostname" type="string">sirocco</parameter>
- <parameter name="algorithm" type="string">potrf</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >29</parameter>
- <parameter name="ngpu" type="int" >2</parameter>
- <parameter name="b" type="int" >1600</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <parameterset name="param_geqrf">
- <parameter name="hostname" type="string">sirocco</parameter>
- <parameter name="algorithm" type="string">geqrf_hqr</parameter>
- <parameter name="precision" type="string">s, d</parameter>
- <parameter name="i_pq" type="int" >0</parameter>
- <parameter name="p" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
- <parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
- <parameter name="nthr" type="int" >29</parameter>
- <parameter name="ngpu" type="int" >2</parameter>
- <parameter name="b" type="int" >1600</parameter>
- <parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${nmpi}*${b}, ${nmpi}*5*${b}, ${nmpi}*10*${b}, ${nmpi}*20*${b}, ${nmpi}*40*${b}][$i_mn]</parameter>
- <parameter name="k" type="int" >1</parameter>
- </parameterset>
-
- <!-- Operation -->
- <step name="run_gemm" tag="gemm">
- <use>param_gemm</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_potrf" tag="potrf">
- <use>param_potrf</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
- <step name="run_geqrf_hqr" tag="geqrf">
- <use>param_geqrf</use>
- <do>mpiexec -np $nmpi $CHAMELEON_BUILD/new-testing/${precision}new-testing -o ${algorithm} -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b</do>
- </step>
-
- <!-- Analyse -->
- <analyser name="analyse">
- <!-- use a pattern set -->
- <use from="../../jube/patterns.xml">chameleon</use>
- <analyse step="run_gemm" tag="gemm">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_potrf" tag="potrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- <analyse step="run_geqrf_hqr" tag="geqrf">
- <file>stdout</file> <!-- file which should be scanned -->
- </analyse>
- </analyser>
-
-
- <!-- Create result table -->
- <result>
- <use>analyse</use> <!-- use existing analyser -->
- <!--<table name="result" style="csv" sort="number">-->
- <table name="result" style="csv">
- <column>hostname</column>
- <column>algorithm</column>
- <column>precision</column>
- <column>nmpi</column>
- <column>p</column>
- <column>q</column>
- <column>nthr</column>
- <column>ngpu</column>
- <column>m</column>
- <column>n</column>
- <column>k</column>
- <column>cputime</column>
- <column>gflops</column>
- </table>
- </result>
- </benchmark>
-</jube>
diff --git a/tools/bench/plafrim/sirocco/run.sh b/tools/bench/plafrim/sirocco/run.sh
deleted file mode 100755
index e082b5a79..000000000
--- a/tools/bench/plafrim/sirocco/run.sh
+++ /dev/null
@@ -1,70 +0,0 @@
-#!/bin/bash
-
-echo "######################### Chameleon benchmarks #########################"
-echo "HOSTNAME $HOSTNAME"
-echo "USERNAME $USERNAME"
-echo "GIT REPO $CI_REPOSITORY_URL"
-echo "GIT BRANCH $CI_COMMIT_REF_NAME"
-echo "GIT COMMIT $CI_COMMIT_SHA"
-echo "PROJECT DIR $CI_PROJECT_DIR"
-
-set -x
-
-function wait_completion {
- # Wait for completion of jobs
- echo "JOB_LIST $JOB_LIST"
- while [ "$NJOB" -gt 0 ]
- do
- for JOB in $JOB_LIST
- do
- IS_JOB_IN_QUEUE=`squeue |grep "$JOB"`
- if [[ -z "$IS_JOB_IN_QUEUE" ]]
- then
- NJOB=$[NJOB-1]
- JOB_LIST=`echo $JOB_LIST | sed "s#$JOB##"`
- echo "JOB $JOB finished"
- else
- echo "$IS_JOB_IN_QUEUE"
- fi
- done
- sleep 30
- done
-}
-
-# Parameters for scripts
-export PLATFORM=plafrim
-export NODE=v100
-export BUILD_OPTIONS="-DCHAMELEON_USE_CUDA=ON -DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
-export STARPU_SILENT=1
-export STARPU_LIMIT_CPU_MEM=370000
-export STARPU_LIMIT_MAX_SUBMITTED_TASKS=16000
-export STARPU_LIMIT_MIN_SUBMITTED_TASKS=15000
-
-# Parameters of the Slurm jobs
-TIME=01:00:00
-PART=routage
-CONS=v100
-EXCL=
-NP=1
-
-# Submit jobs
-NJOB=0
-#MPI_LIST="openmpi nmad"
-MPI_LIST="openmpi"
-for MPI in $MPI_LIST
-do
- JOB_ID=`JOB_NAME=chameleon\_$MPI\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL $CI_PROJECT_DIR/tools/bench/chameleon_guix\_$MPI\_cuda.sl | sed "s#Submitted batch job ##"`
- #JOB_ID=`JOB_NAME=chameleon\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL chameleon_guix.sl | sed "s#Submitted batch job ##"`
- if [[ -n "$JOB_ID" ]]
- then
- JOB_LIST="$JOB_LIST $JOB_ID"
- NJOB=$[NJOB+1]
- fi
-done
-
-# Wait for completion of jobs
-wait_completion
-
-echo "####################### End Chameleon benchmarks #######################"
-
-exit 0
diff --git a/tools/bench/plafrim/slurm.sh b/tools/bench/plafrim/slurm.sh
new file mode 100755
index 000000000..f64c4ff69
--- /dev/null
+++ b/tools/bench/plafrim/slurm.sh
@@ -0,0 +1,59 @@
+#!/bin/bash
+
+echo "######################### Chameleon benchmarks #########################"
+
+# Check the environment
+echo $PLATFORM
+echo $NODE
+env |grep ^CI
+env |grep ^SLURM
+env |grep ^JUBE_ID
+env |grep ^MPI
+env |grep ^STARPU
+env |grep ^CHAMELEON
+
+set -x
+
+function wait_completion {
+ # Wait for completion of jobs
+ echo "JOB_LIST $JOB_LIST"
+ while [ "$NJOB" -gt 0 ]
+ do
+ for JOB in $JOB_LIST
+ do
+ IS_JOB_IN_QUEUE=`squeue |grep "$JOB"`
+ if [[ -z "$IS_JOB_IN_QUEUE" ]]
+ then
+ NJOB=$[NJOB-1]
+ JOB_LIST=`echo $JOB_LIST | sed "s#$JOB##"`
+ echo "JOB $JOB finished"
+ else
+ echo "$IS_JOB_IN_QUEUE"
+ fi
+ done
+ sleep 30
+ done
+}
+
+# Parameters of the Slurm jobs
+TIME=02:00:00
+PART=routage
+NP=$SLURM_NP
+CONS=$SLURM_CONSTRAINTS
+EXCL=
+
+# Submit jobs
+NJOB=0
+JOB_ID=`JOB_NAME=chameleon\-$NODE\-$MPI\-$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS --exclude=$EXCL --exclusive --ntasks-per-node=1 --threads-per-core=1 $CI_PROJECT_DIR/tools/bench/chameleon_guix.sh | sed "s#Submitted batch job ##"`
+if [[ -n "$JOB_ID" ]]
+then
+ JOB_LIST="$JOB_LIST $JOB_ID"
+ NJOB=$[NJOB+1]
+fi
+
+# Wait for completion of jobs
+wait_completion
+
+echo "####################### End Chameleon benchmarks #######################"
+
+exit 0
--
GitLab