diff --git a/.gitlab/bench_plafrim.yml b/.gitlab/bench_plafrim.yml
index 0542e3018927c99988efeca00cec18035d47c7bb..24d9f5a7e02e26e4d77b99db22ea4350c4de756e 100644
--- a/.gitlab/bench_plafrim.yml
+++ b/.gitlab/bench_plafrim.yml
@@ -1,10 +1,29 @@
-bench_plafrim:
+bench_plafrim_bora:
stage: test
tags: ['plafrim']
- variables:
- BUILD_OPTIONS: "-DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
- VERSION: miriel
+ before_script:
+ - git submodule update --init --recursive
script:
- - (cd tools/bench/plafrim/miriel && ./run.sh)
+ - ./tools/bench/plafrim/bora/run.sh
only:
- schedules
+
+.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_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
diff --git a/tools/bench/chameleon_guix.sh b/tools/bench/chameleon_guix.sh
new file mode 100755
index 0000000000000000000000000000000000000000..a120649c163ee4143bfbeae8ccab26f0e8156e25
--- /dev/null
+++ b/tools/bench/chameleon_guix.sh
@@ -0,0 +1,26 @@
+#!/bin/bash
+
+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 $BUILD_OPTIONS ..
+make -j20 VERBOSE=1
+export CHAMELEON_BUILD=$PWD
+
+# Execute jube benchmarks
+jube run $CI_PROJECT_DIR/tools/bench/$PLATFORM/$NODE/chameleon_$MPI.xml --tag gemm potrf geqrf
+# jube analysis
+jube analyse $CI_PROJECT_DIR/tools/bench/$PLATFORM/$NODE/results/
+# jube report
+jube result $CI_PROJECT_DIR/tools/bench/$PLATFORM/$NODE/results/ -i last > chameleon.csv
+
+# send results to the elasticsearch server
+export PYTHONPATH=$GUIX_ENVIRONMENT/lib/python3.7/site-packages
+python3 $CI_PROJECT_DIR/tools/bench/jube/add_result.py -e https://elasticsearch.bordeaux.inria.fr -t hiepacs -p "chameleon" -h $NODE -m $MPI chameleon.csv
diff --git a/tools/bench/chameleon_guix_nmad.sl b/tools/bench/chameleon_guix_nmad.sl
new file mode 100644
index 0000000000000000000000000000000000000000..b3b952de26f297ff1dbad48c71d25116a6c54335
--- /dev/null
+++ b/tools/bench/chameleon_guix_nmad.sl
@@ -0,0 +1,22 @@
+#!/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
new file mode 100644
index 0000000000000000000000000000000000000000..fd84573afffc28ac15ce54f9c41cd7bd3d6bcf48
--- /dev/null
+++ b/tools/bench/chameleon_guix_openmpi.sl
@@ -0,0 +1,22 @@
+#!/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 f5e4502cbcb6f18d16b299c1213b5ab73aab890d..63a68d6d27dc93ccc39efe81711b6f4ab8c02a0a 100755
--- a/tools/bench/jube/add_result.py
+++ b/tools/bench/jube/add_result.py
@@ -32,24 +32,23 @@ def open_csv(filename: str) -> List[Dict[str, str]]:
return csv_rows
-def format_entry(row: Row, commit_chameleon: Repo, commit_guix: str, commit_guix_hpc: str, commit_guix_hpcnonfree: str) -> Dict[str, Any]:
+def format_entry(row: Row, mpivendor: str, commit_chameleon: Repo, commit_guix: str, commit_guix_hpc: str, commit_guix_hpcnonfree: str) -> Dict[str, Any]:
""""format a result"""
commit_date_chameleon = str(time.strftime("%Y-%m-%d %H:%M:%S", time.gmtime(commit_chameleon.committed_date)))
commit_sha_chameleon = str(commit_chameleon.hexsha)
hostname = str(row.pop('hostname'))
algorithm = str(row.pop('algorithm'))
precision = str(row.pop('precision'))
- nmpi = int(row.pop('NMPI'))
- tdbc_p = int(row.pop('P'))
- tdbc_q = int(row.pop('Q'))
- nthread = int(row.pop('NTHREAD'))
- ngpu = int(row.pop('NGPU'))
- m = int(row.pop('M'))
- n = int(row.pop('N'))
- k = int(row.pop('K'))
- cputime = float(row.pop('CPUTIME'))
- gflops = float(row.pop('GFLOPS'))
- stddev = float(row.pop('STDDEV'))
+ nmpi = int(row.pop('nmpi'))
+ p = int(row.pop('p'))
+ q = int(row.pop('q'))
+ nthread = int(row.pop('nthr'))
+ ngpu = int(row.pop('ngpu'))
+ m = int(row.pop('m'))
+ n = int(row.pop('n'))
+ k = int(row.pop('k'))
+ cputime = float(row.pop('cputime'))
+ gflops = float(row.pop('gflops'))
result = {
"Commit_date_chameleon": commit_date_chameleon,
"Commit_sha_chameleon": commit_sha_chameleon,
@@ -57,19 +56,19 @@ def format_entry(row: Row, commit_chameleon: Repo, commit_guix: str, commit_guix
"Commit_sha_guix_hpc": commit_guix_hpc,
"Commit_sha_guix_hpcnonfree": commit_guix_hpcnonfree,
"Hostname": hostname,
+ "MPIvendor": mpivendor,
"Algorithm": algorithm,
"Precision": precision,
"Nmpi": nmpi,
- "P": tdbc_p,
- "Q": tdbc_q,
+ "P": p,
+ "Q": q,
"Nthread": nthread,
"Ngpu": ngpu,
"M": m,
"N": n,
"K": k,
"Cputime": cputime,
- "Gflops": gflops,
- "Stddev": stddev
+ "Gflops": gflops
}
return result
@@ -80,6 +79,7 @@ def format_entry(row: Row, commit_chameleon: Repo, commit_guix: str, commit_guix
@click.option("-t", "--team", required=True, help="team name")
@click.option("-p", "--project", required=True, help="project name")
@click.option("-h", "--host", required=True, help="host name")
+@click.option("-m", "--mpi", required=True, help="MPI vendor (openmpi, nmad)")
@click.argument("csv-files", nargs=-1)
def main(
directory: str,
@@ -87,6 +87,7 @@ def main(
team: str,
project: str,
host: str,
+ mpi: str,
csv_files: str,
):
"""Add a result to an elasticsearch database."""
@@ -104,6 +105,7 @@ def main(
"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"},
@@ -115,8 +117,7 @@ def main(
"N": {"type": "integer"},
"K": {"type": "integer"},
"Cputime": {"type": "float"},
- "Gflops": {"type": "float"},
- "Stddev": {"type": "float"}
+ "Gflops": {"type": "float"}
}
}
}
@@ -141,7 +142,7 @@ def main(
request
for file in csv_files
for request in map(
- lambda row: format_entry(row, commit_chameleon, commit_guix, commit_guix_hpc, commit_guix_hpcnonfree),
+ lambda row: format_entry(row, mpi, commit_chameleon, commit_guix, commit_guix_hpc, commit_guix_hpcnonfree),
open_csv(file)
)
]
diff --git a/tools/bench/jube/paths.xml b/tools/bench/jube/paths.xml
deleted file mode 100644
index b92cad88a8aa3cb975df3c4784c2b2cf73f910d1..0000000000000000000000000000000000000000
--- a/tools/bench/jube/paths.xml
+++ /dev/null
@@ -1,6 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<jube>
- <parameterset name="paths">
- <parameter name="BIN_DIR" type="string">{{CHAMELEON_DIR}}</parameter>
- </parameterset>
-</jube>
diff --git a/tools/bench/jube/patterns.xml b/tools/bench/jube/patterns.xml
index 05c8f976db026952d3aa8e6da79aa2dc51e62191..6770537d74c7b50042ee672913da399409532104 100644
--- a/tools/bench/jube/patterns.xml
+++ b/tools/bench/jube/patterns.xml
@@ -2,16 +2,7 @@
<jube>
<!-- Regex pattern -->
<patternset name="chameleon">
- <pattern name="NMPI" type="int"># Nb mpi: *(\d+)</pattern>
- <pattern name="P" type="int"># PxQ: *(\d+)x\d+</pattern>
- <pattern name="Q" type="int"># PxQ: *\d+x(\d+)</pattern>
- <pattern name="NTHREAD" type="int"># Nb threads: *(\d+)</pattern>
- <pattern name="NGPU" type="int"># Nb GPUs: *(\d+)</pattern>
- <pattern name="M" type="int">^ *(\d+) *\d+ *\d+ *\d+\.?\d+ *\d+\.?\d+ \+\- *\d+\.?\d+ *</pattern>
- <pattern name="N" type="int">^ *\d+ *(\d+) *\d+ *\d+\.?\d+ *\d+\.?\d+ \+\- *\d+\.?\d+ *</pattern>
- <pattern name="K" type="int">^ *\d+ *\d+ *(\d+) *\d+\.?\d+ *\d+\.?\d+ \+\- *\d+\.?\d+ *</pattern>
- <pattern name="CPUTIME" type="float">^ *\d+ *\d+ *\d+ *(\d+\.?\d+) *\d+\.?\d+ \+\- *\d+\.?\d+ *</pattern>
- <pattern name="GFLOPS" type="float">^ *\d+ *\d+ *\d+ *\d+\.?\d+ *(\d+\.?\d+) \+\- *\d+\.?\d+ *</pattern>
- <pattern name="STDDEV" type="float">^ *\d+ *\d+ *\d+ *\d+\.?\d+ *\d+\.?\d+ \+\- *(\d+\.?\d+) *</pattern>
+ <pattern name="cputime" type="float">^$jube_pat_nint;$jube_pat_nwrd;.*;$jube_pat_fp;$jube_pat_nfp$</pattern>
+ <pattern name="gflops" type="float">^$jube_pat_nint;$jube_pat_nwrd;.*;$jube_pat_nfp;$jube_pat_fp$</pattern>
</patternset>
</jube>
diff --git a/tools/bench/plafrim/bora/chameleon_nmad.xml b/tools/bench/plafrim/bora/chameleon_nmad.xml
new file mode 100644
index 0000000000000000000000000000000000000000..522da430c413789cda55d4e0cc8445a651d41eb0
--- /dev/null
+++ b/tools/bench/plafrim/bora/chameleon_nmad.xml
@@ -0,0 +1,108 @@
+<?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 $b</do>
+ </step>
+
+ <!-- Analyse -->
+ <analyser name="analyse">
+ <!-- use a pattern set -->
+ <use from="../../jube/patterns.xml">chameleon</use>
+ <analyse step="run_gemm" tag="gemm">
+ <file>stdout</file> <!-- file which should be scanned -->
+ </analyse>
+ <analyse step="run_potrf" tag="potrf">
+ <file>stdout</file> <!-- file which should be scanned -->
+ </analyse>
+ <analyse step="run_geqrf_hqr" tag="geqrf">
+ <file>stdout</file> <!-- file which should be scanned -->
+ </analyse>
+ </analyser>
+
+
+ <!-- Create result table -->
+ <result>
+ <use>analyse</use> <!-- use existing analyser -->
+ <!--<table name="result" style="csv" sort="number">-->
+ <table name="result" style="csv">
+ <column>hostname</column>
+ <column>algorithm</column>
+ <column>precision</column>
+ <column>nmpi</column>
+ <column>p</column>
+ <column>q</column>
+ <column>nthr</column>
+ <column>ngpu</column>
+ <column>m</column>
+ <column>n</column>
+ <column>k</column>
+ <column>cputime</column>
+ <column>gflops</column>
+ </table>
+ </result>
+ </benchmark>
+</jube>
diff --git a/tools/bench/plafrim/bora/chameleon_openmpi.xml b/tools/bench/plafrim/bora/chameleon_openmpi.xml
new file mode 100644
index 0000000000000000000000000000000000000000..fe80ac54b44ce1078f140c6f9beb72cd6b41fa92
--- /dev/null
+++ b/tools/bench/plafrim/bora/chameleon_openmpi.xml
@@ -0,0 +1,108 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<jube>
+ <benchmark name="bora" outpath="results">
+ <comment>benchmark chameleon on host plafrim bora</comment>
+
+ <parameterset name="param_gemm">
+ <parameter name="hostname" type="string">bora</parameter>
+ <parameter name="algorithm" type="string">gemm</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$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
new file mode 100755
index 0000000000000000000000000000000000000000..baf9e742435029acc751944c4dee659e08fd5fab
--- /dev/null
+++ b/tools/bench/plafrim/bora/run.sh
@@ -0,0 +1,70 @@
+#!/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/miriel/chameleon_guix.sh b/tools/bench/plafrim/miriel/chameleon_guix.sh
deleted file mode 100755
index a31d935ee83b87dbe533306febd66a62b3f0c4aa..0000000000000000000000000000000000000000
--- a/tools/bench/plafrim/miriel/chameleon_guix.sh
+++ /dev/null
@@ -1,25 +0,0 @@
-#!/bin/bash
-
-set -x
-
-# Configure and Build Chameleon
-echo $VERSION
-cd ../../../../build-$VERSION
-CHAMELEON_DIR=`pwd`
-cmake $BUILD_OPTIONS ..
-make -j5
-cd -
-
-# Define where to find the build directory for jube
-sed 's@{{CHAMELEON_DIR}}@'"${CHAMELEON_DIR}"'@g' -i ../../jube/paths.xml
-
-# Execute jube benchmarks
-jube run chameleon.xml --tag gemm potrf geqrf
-# jube analysis
-jube analyse results/$VERSION/
-# jube report
-jube result results/$VERSION/ -i last > chameleon.csv
-
-# send results to the elasticsearch server
-export PYTHONPATH=$GUIX_ENVIRONMENT/lib/python3.7/site-packages
-python3 ../../jube/add_result.py -e https://elasticsearch.bordeaux.inria.fr -t hiepacs -p "chameleon" -h $VERSION chameleon.csv
diff --git a/tools/bench/plafrim/miriel/chameleon_guix.sl b/tools/bench/plafrim/miriel/chameleon_guix.sl
deleted file mode 100644
index 8c3dd9d6e85e5238d4fcd10ad329fd10bd33ef2b..0000000000000000000000000000000000000000
--- a/tools/bench/plafrim/miriel/chameleon_guix.sl
+++ /dev/null
@@ -1,25 +0,0 @@
-#!/usr/bin/env bash
-#SBATCH --exclusive
-#SBATCH --ntasks-per-node=1
-#SBATCH --threads-per-core=1
-
-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"
-
-# to avoid a lock during fetching chameleon branch in parallel
-export XDG_CACHE_HOME=/tmp/guix-$$
-
-# save guix commits
-guix describe --format=json > guix.json
-
-# Submit jobs
-exec guix environment --pure --preserve=SLURM --preserve=VERSION --preserve=BUILD_OPTIONS chameleon --with-input=openblas=mkl --ad-hoc slurm jube python python-click python-gitpython python-elasticsearch python-certifi sed coreutils grep gawk openssh perl hwloc openmpi starpu mkl -- /bin/bash --norc chameleon_guix.sh
-
-echo "####################### End Chameleon benchmarks #######################"
-
-# clean tmp
-rm -rf /tmp/guix-$$
diff --git a/tools/bench/plafrim/miriel/chameleon.xml b/tools/bench/plafrim/miriel/chameleon_nmad.xml
similarity index 75%
rename from tools/bench/plafrim/miriel/chameleon.xml
rename to tools/bench/plafrim/miriel/chameleon_nmad.xml
index 98d369ff2fd038eea344e1a350d8ee6fa5994103..28be9ef0b711e22132deefe24d17f83e7909c0c8 100644
--- a/tools/bench/plafrim/miriel/chameleon.xml
+++ b/tools/bench/plafrim/miriel/chameleon_nmad.xml
@@ -15,9 +15,9 @@
<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" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
+ <parameter name="m" mode="python" type="int" >[${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}][$i_mn]</parameter>
</parameterset>
<parameterset name="param_potrf">
@@ -32,8 +32,8 @@
<parameter name="ngpu" type="int" >0</parameter>
<parameter name="b" type="int" >320</parameter>
<parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
+ <parameter name="m" mode="python" type="int" >[${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}][$i_mn]</parameter>
<parameter name="k" type="int" >1</parameter>
</parameterset>
@@ -49,26 +49,23 @@
<parameter name="ngpu" type="int" >0</parameter>
<parameter name="b" type="int" >320</parameter>
<parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
+ <parameter name="m" mode="python" type="int" >[${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>
- <use from="../../jube/paths.xml">paths</use>
- <do>STARPU_SILENT=1 mpiexec -np $nmpi $BIN_DIR/timing/time_${precision}${algorithm}_tile -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b --niter=3</do>
+ <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>
- <use from="../../jube/paths.xml">paths</use>
- <do>STARPU_SILENT=1 mpiexec -np $nmpi $BIN_DIR/timing/time_${precision}${algorithm}_tile -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b --niter=3</do>
+ <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>
- <use from="../../jube/paths.xml">paths</use>
- <do>STARPU_SILENT=1 mpiexec -np $nmpi $BIN_DIR/timing/time_${precision}${algorithm}_tile -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b --niter=3</do>
+ <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 $b</do>
</step>
<!-- Analyse -->
@@ -95,17 +92,16 @@
<column>hostname</column>
<column>algorithm</column>
<column>precision</column>
- <column>NMPI</column>
- <column>P</column>
- <column>Q</column>
- <column>NTHREAD</column>
- <column>NGPU</column>
- <column>M</column>
- <column>N</column>
- <column>K</column>
- <column>CPUTIME</column>
- <column>GFLOPS</column>
- <column>STDDEV</column>
+ <column>nmpi</column>
+ <column>p</column>
+ <column>q</column>
+ <column>nthr</column>
+ <column>ngpu</column>
+ <column>m</column>
+ <column>n</column>
+ <column>k</column>
+ <column>cputime</column>
+ <column>gflops</column>
</table>
</result>
</benchmark>
diff --git a/tools/bench/plafrim/miriel/chameleon_openmpi.xml b/tools/bench/plafrim/miriel/chameleon_openmpi.xml
new file mode 100644
index 0000000000000000000000000000000000000000..d6b70cc8417353a590740a17fae67ed07fd5a040
--- /dev/null
+++ b/tools/bench/plafrim/miriel/chameleon_openmpi.xml
@@ -0,0 +1,108 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<jube>
+ <benchmark name="miriel" outpath="results/miriel">
+ <comment>benchmark chameleon on host plafrim miriel</comment>
+
+ <parameterset name="param_gemm">
+ <parameter name="hostname" type="string">miriel</parameter>
+ <parameter name="algorithm" type="string">gemm</parameter>
+ <parameter name="precision" type="string">s, d</parameter>
+ <parameter name="i_pq" type="int" >0, 1, 2</parameter>
+ <parameter name="p" mode="python" type="int" >[1, 2, 3][$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
index c5e21926e3aa1a00f5389fd017e9f4814645a97d..d37974b0381fb086db07d620188eefe9c3e99fd5 100755
--- a/tools/bench/plafrim/miriel/run.sh
+++ b/tools/bench/plafrim/miriel/run.sh
@@ -6,26 +6,21 @@ 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"
-# Parameters of the Slurm jobs
-TIME=01:00:00
-PART=court
-CONS=MirielIB
-EXCL=
-NP=9
-JOBSLIM=1
+set -x
function wait_completion {
# Wait for completion of jobs
echo "JOB_LIST $JOB_LIST"
- while [ "$ITER" -ge "$JOBSLIM" ]
+ while [ "$NJOB" -gt 0 ]
do
for JOB in $JOB_LIST
do
IS_JOB_IN_QUEUE=`squeue |grep "$JOB"`
if [[ -z "$IS_JOB_IN_QUEUE" ]]
then
- ITER=$[ITER-1]
+ NJOB=$[NJOB-1]
JOB_LIST=`echo $JOB_LIST | sed "s#$JOB##"`
echo "JOB $JOB finished"
else
@@ -36,22 +31,40 @@ function wait_completion {
done
}
+# Parameters for scripts
+export PLATFORM=plafrim
+export NODE=miriel
+export BUILD_OPTIONS="-DCHAMELEON_USE_MPI=ON -DCMAKE_BUILD_TYPE=Release"
+export STARPU_SILENT=1
+export STARPU_LIMIT_CPU_MEM=120000
+export STARPU_LIMIT_MAX_SUBMITTED_TASKS=16000
+export STARPU_LIMIT_MIN_SUBMITTED_TASKS=15000
+
+# Parameters of the Slurm jobs
+TIME=01:00:00
+PART=routage
+CONS="miriel,omnipath"
+EXCL=
+NP=9
# Submit jobs
-ITER=0
-JOB_ID=`JOB_NAME=chameleon_bench\_$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 $JOB_ID"
- ITER=$[ITER+1]
-fi
+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
-# Print results
-cat chameleon_bench\_$NP.out
-
echo "####################### End Chameleon benchmarks #######################"
exit 0
diff --git a/tools/bench/plafrim/sirocco/chameleon.xml b/tools/bench/plafrim/sirocco/chameleon.xml
index 341d035fb0dbdecfd4dcf8a3460cb43eb013ff28..fda2c21c7db4a7c204dc1ab1da99003b4194181d 100644
--- a/tools/bench/plafrim/sirocco/chameleon.xml
+++ b/tools/bench/plafrim/sirocco/chameleon.xml
@@ -15,9 +15,9 @@
<parameter name="ngpu" type="int" >2</parameter>
<parameter name="b" type="int" >1600</parameter>
<parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="k" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
+ <parameter name="m" mode="python" type="int" >[${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}][$i_mn]</parameter>
</parameterset>
<parameterset name="param_potrf">
@@ -32,8 +32,8 @@
<parameter name="ngpu" type="int" >2</parameter>
<parameter name="b" type="int" >1600</parameter>
<parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
+ <parameter name="m" mode="python" type="int" >[${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}][$i_mn]</parameter>
<parameter name="k" type="int" >1</parameter>
</parameterset>
@@ -46,29 +46,26 @@
<parameter name="q" mode="python" type="int" >[1, 2, 3][$i_pq]</parameter>
<parameter name="nmpi" mode="python" type="int" >[1, 4, 9][$i_pq]</parameter>
<parameter name="nthr" type="int" >29</parameter>
- <parameter name="ngpu" type="int" >0</parameter>
+ <parameter name="ngpu" type="int" >2</parameter>
<parameter name="b" type="int" >1600</parameter>
<parameter name="i_mn" type="int" >0, 1, 2, 3, 4</parameter>
- <parameter name="m" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
- <parameter name="n" mode="python" type="int" >[${p}*${b}, ${p}*5*${b}, ${p}*10*${b}, ${p}*20*${b}, ${p}*50*${b}][$i_mn]</parameter>
+ <parameter name="m" mode="python" type="int" >[${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>
- <use from="../../jube/paths.xml">paths</use>
- <do>STARPU_SILENT=1 mpiexec -np $nmpi $BIN_DIR/timing/time_${precision}${algorithm}_tile -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b --niter=3</do>
+ <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>
- <use from="../../jube/paths.xml">paths</use>
- <do>STARPU_SILENT=1 mpiexec -np $nmpi $BIN_DIR/timing/time_${precision}${algorithm}_tile -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b --niter=3</do>
+ <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>
- <use from="../../jube/paths.xml">paths</use>
- <do>STARPU_SILENT=1 mpiexec -np $nmpi $BIN_DIR/timing/time_${precision}${algorithm}_tile -P $p -t $nthr -g $ngpu -m $m -n $n -k $k -b $b --niter=3</do>
+ <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 -->
@@ -95,17 +92,16 @@
<column>hostname</column>
<column>algorithm</column>
<column>precision</column>
- <column>NMPI</column>
- <column>P</column>
- <column>Q</column>
- <column>NTHREAD</column>
- <column>NGPU</column>
- <column>M</column>
- <column>N</column>
- <column>K</column>
- <column>CPUTIME</column>
- <column>GFLOPS</column>
- <column>STDDEV</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>
diff --git a/tools/bench/plafrim/sirocco/chameleon_guix.sh b/tools/bench/plafrim/sirocco/chameleon_guix.sh
deleted file mode 100755
index a31d935ee83b87dbe533306febd66a62b3f0c4aa..0000000000000000000000000000000000000000
--- a/tools/bench/plafrim/sirocco/chameleon_guix.sh
+++ /dev/null
@@ -1,25 +0,0 @@
-#!/bin/bash
-
-set -x
-
-# Configure and Build Chameleon
-echo $VERSION
-cd ../../../../build-$VERSION
-CHAMELEON_DIR=`pwd`
-cmake $BUILD_OPTIONS ..
-make -j5
-cd -
-
-# Define where to find the build directory for jube
-sed 's@{{CHAMELEON_DIR}}@'"${CHAMELEON_DIR}"'@g' -i ../../jube/paths.xml
-
-# Execute jube benchmarks
-jube run chameleon.xml --tag gemm potrf geqrf
-# jube analysis
-jube analyse results/$VERSION/
-# jube report
-jube result results/$VERSION/ -i last > chameleon.csv
-
-# send results to the elasticsearch server
-export PYTHONPATH=$GUIX_ENVIRONMENT/lib/python3.7/site-packages
-python3 ../../jube/add_result.py -e https://elasticsearch.bordeaux.inria.fr -t hiepacs -p "chameleon" -h $VERSION chameleon.csv
diff --git a/tools/bench/plafrim/sirocco/chameleon_guix.sl b/tools/bench/plafrim/sirocco/chameleon_guix.sl
deleted file mode 100644
index 8c3dd9d6e85e5238d4fcd10ad329fd10bd33ef2b..0000000000000000000000000000000000000000
--- a/tools/bench/plafrim/sirocco/chameleon_guix.sl
+++ /dev/null
@@ -1,25 +0,0 @@
-#!/usr/bin/env bash
-#SBATCH --exclusive
-#SBATCH --ntasks-per-node=1
-#SBATCH --threads-per-core=1
-
-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"
-
-# to avoid a lock during fetching chameleon branch in parallel
-export XDG_CACHE_HOME=/tmp/guix-$$
-
-# save guix commits
-guix describe --format=json > guix.json
-
-# Submit jobs
-exec guix environment --pure --preserve=SLURM --preserve=VERSION --preserve=BUILD_OPTIONS chameleon --with-input=openblas=mkl --ad-hoc slurm jube python python-click python-gitpython python-elasticsearch python-certifi sed coreutils grep gawk openssh perl hwloc openmpi starpu mkl -- /bin/bash --norc chameleon_guix.sh
-
-echo "####################### End Chameleon benchmarks #######################"
-
-# clean tmp
-rm -rf /tmp/guix-$$
diff --git a/tools/bench/plafrim/sirocco/run.sh b/tools/bench/plafrim/sirocco/run.sh
index 5c3f5e3c2de3fb4e019ad99c237f66ba64991e3a..e082b5a7958626374e66f5826c796fbe156554ec 100755
--- a/tools/bench/plafrim/sirocco/run.sh
+++ b/tools/bench/plafrim/sirocco/run.sh
@@ -6,52 +6,65 @@ 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"
-# Parameters of the Slurm jobs
-TIME=01:00:00
-PART=court_sirocco
-CONS=Skylake
-EXCL=
-NP=1
-JOBSLIM=1
+set -x
function wait_completion {
# Wait for completion of jobs
echo "JOB_LIST $JOB_LIST"
- while [ "$ITER" -ge "$JOBSLIM" ]
+ while [ "$NJOB" -gt 0 ]
do
- for JOB in $JOB_LIST
- do
- IS_JOB_IN_QUEUE=`squeue |grep "$JOB"`
- if [[ -z "$IS_JOB_IN_QUEUE" ]]
- then
- ITER=$[ITER-1]
- JOB_LIST=`echo $JOB_LIST | sed "s#$JOB##"`
- echo "JOB $JOB finished"
- else
- echo "$IS_JOB_IN_QUEUE"
- fi
- done
- sleep 30
+ 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
-ITER=0
-JOB_ID=`JOB_NAME=chameleon_bench\_$NP && sbatch --job-name="$JOB_NAME" --output="$JOB_NAME.out" --error="$JOB_NAME.err" --nodes=$NP --time=$TIME --partition=$PART --constraint=$CONS chameleon_guix.sl | sed "s#Submitted batch job ##"`
-if [[ -n "$JOB_ID" ]]
-then
- JOB_LIST="$JOB_LIST $JOB_ID"
- ITER=$[ITER+1]
-fi
+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
-# Print results
-cat chameleon_bench\_$NP.out
-
echo "####################### End Chameleon benchmarks #######################"
exit 0