Commit 4ddcccf1 authored by Victor's avatar Victor
Browse files

NEW initial contribution for QCD

parent 8e79193b
PRACE QCD Accelerator Benchmark 1
=================================
This benchmark is part of the QCD section of the Accelerator
Benchmarks Suite developed as part of a PRACE EU funded project
(http://www.prace-ri.eu).
The suite is derived from the Unified European Applications
Benchmark Suite (UEABS) http://www.prace-ri.eu/ueabs/
This specific component is a direct port of "QCD kernel E" from the
UEABS, which is based on the MILC code suite
(http://www.physics.utah.edu/~detar/milc/). The performance-portable
targetDP model has been used to allow the benchmark to utilise NVIDIA
GPUs, Intel Xeon Phi manycore CPUs and traditional multi-core
CPUs. The use of MPI (in conjunction with targetDP) allows multiple
nodes to be used in parallel.
For full details of this benchmark, and for results on NVIDIA GPU and
Intel Knights Corner Xeon Phi architectures (in addition to regular
CPUs), please see:
**********************************************************************
Gray, Alan, and Kevin Stratford. "A lightweight approach to
performance portability with targetDP." The International Journal of
High Performance Computing Applications (2016): 1094342016682071, Also
available at https://arxiv.org/abs/1609.01479
**********************************************************************
To Build
--------
Choose a configuration file from the "config" directory that best
matches your platform, and copy to "config.mk" in this (the
top-level) directory. Then edit this file, if necessary, to properly
set the compilers and paths on your system.
Note that if you are building for a GPU system, and the TARGETCC
variable in the configuration file is set to the NVIDIA compiler nvcc,
then the build process will automatically build the GPU
version. Otherwise, the threaded CPU version will be built which can
run on Xeon Phi manycore CPUs or regular multi-core CPUs.
Then, build the targetDP performance-portable library:
cd targetDP
make clean
make
cd ..
And finally build the benchmark code
cd src
make clean
make
cd ..
To Validate
-----------
After building, an executable "bench" will exist in the src directory.
To run the default validation (64x64x64x8, 1 iteration) case:
cd src
./bench
The code will automatically self-validate by comparing with the
appropriate output reference file for this case which exists in
output_ref, and will print to stdout, e.g.
Validating against output_ref/kernel_E.output.nx64ny64nz64nt8.i1.t1:
VALIDATION PASSED
The benchmark time is also printed to stdout, e.g.
******BENCHMARK TIME 1.6767786769196391e-01 seconds******
(Where this time is as reported on an NVIDIA K40 GPU).
To Run Different Cases
---------------------
You can edit the input file
src/kernel_E.input
if you want to deviate from the default system size, number of
iterations and/or run using more than 1 MPI task. E.g. replacing
totnodes 1 1 1 1
with
totnodes 2 1 1 1
will run with 2 MPI tasks rather than 1, where the domain is decomposed in
the "X" direction.
To Run using a Script
---------------------
The "run" directory contains an example script which
- sets up a temporary scratch directory
- copies in the input file, plus also some reference output files
- sets the number of OpenMP threads (for a multi/many core CPU run)
- runs the code (which will automatically validate if an
appropriate output reference file exists)
So, in the run directory, you should copy "run_example.sh" to
run.sh, which you can customise for your system.
Known Issues
------------
The quantity used for validation (see congrad.C) becomes very small
after a few iterations. Therefore, only a small number of iterations
should be used for validation. This is not an issue specific to this
port of the benchmark, but is also true of the original version (see
above), with which this version is designed to be consistent.
Performance Results for Reference
--------------------------------
Here are some performance timings obtained using this benchmark.
From the paper cited above:
64x64x64x32x8, 1000 iterations, single chip
Chip Time (s)
Intel Ivy-Bridge 12-core CPU 361.55
Intel Haswell 8-core CPU 376.08
AMD Opteron 16-core CPU 618.19
Intel KNC Xeon Phi 139.94
NVIDIA K20X GPU 96.84
NVIDIA K40 GPU 90.90
Multi-node scaling:
Titan GPU (one K20X per node)
Titan CPU (one 16-core Interlagos per node)
ARCHER CPU (two 12-core Ivy-bridge per node)
All times in seconds.
Small Case: 64x64x32x8, 1000 iterations
Nodes Titan GPU Titan CPU ARCHER CPU
1 9.64E+01 6.01E+02 1.86E+02
2 5.53E+01 3.14E+02 9.57E+01
4 3.30E+01 1.65E+02 5.22E+01
8 2.18E+01 8.33E+01 2.60E+01
16 1.35E+01 4.02E+01 1.27E+01
32 8.80E+00 2.06E+01 6.49E+00
64 6.54E+00 9.90E+00 2.36E+00
128 5.13E+00 4.31E+00 1.86E+00
256 4.25E+00 2.95E+00 1.96E+00
Large Case: 64x64x64x192, 1000 iterations
Nodes Titan GPU Titan CPU ARCHER CPU
64 1.36E+02 5.19E+02 1.61E+02
128 8.23E+01 2.75E+02 8.51E+01
256 6.70E+01 1.61E+02 4.38E+01
512 3.79E+01 8.80E+01 2.18E+01
1024 2.41E+01 5.72E+01 1.46E+01
2048 1.81E+01 3.88E+01 7.35E+00
4096 1.56E+01 2.28E+01 6.53E+00
Preliminary results on new Pascal GPU and Intel KNL architectures:
Single chip, 64x64x64x8, 1000 iterations
Chip Time (s)
12-core Intel Ivy-Bridge 7.24E+02
Intel KNL Xeon Phi 9.72E+01
NVIDIA P100 GPU 5.60E+01
\ No newline at end of file
<!DOCTYPE html>
<html>
<head><meta charset="utf-8">
<title></title>
</head>
<body>
<p>###<br />
###&nbsp;&nbsp; &nbsp;README - QCD Accelerator Benchmarksuite Part 2 &nbsp;<br />
###<br />
###&nbsp;&nbsp; 2017 -&nbsp; Jacob Finkenrath - CaSToRC - The Cyprus Institute&nbsp; (j.finkenrath@cyi.ac.cy)<br />
###</p>
<p>The QCD Accelerator Benchmark suite Part 2 consists of two kernels,<br />
the QUDA and the QPhix library. The QUDA library is based on CUDA and<br />
optimized for running on NVIDIA GPUs. The QPhix library consists of<br />
routines which are optimize to use INTEL intrinsic functions of<br />
multiple vector length, including optimized routines for KNC and<br />
KNL. In both QUDA and QPhix, this benchmark uses the Conjugate<br />
Gradient solvers implemented within the libraries.</p>
<p>[1] R. Babbich, M. Clark and B. Joo, &ldquo;Parallelizing the QUDA Library for Multi-GPU Calculations<br />
in Lattice Quantum Chromodynamics&rdquo; SC 10 (Supercomputing 2010)</p>
<p>[2] B. Joo, D. D. Kalamkar, K. Vaidyanathan, M. Smelyanskiy, K. Pamnany, V. W. Lee, P. Dubey,<br />
W. Watson III, &ldquo;Lattice QCD on Intel Xeon Phi&rdquo;, International Supercomputing Conference (ISC&rsquo;13), 2013</p>
<p>###<br />
###&nbsp; Table of Contents<br />
###</p>
<p><br />
GPU - BENCHMARK SUITE (QUDA)<br />
1. Compile and Run the GPU-Benchmark Suite<br />
1.1 Compile<br />
1.2 Run<br />
1.2.1 Main-script: &quot;run_ana.sh&quot;<br />
1.2.2 Main-script: &quot;prepare_submit_job.sh&quot;<br />
1.2.3 Main-script: &quot;submit_job.sh.template&quot;<br />
1.3 Example Benchmark results</p>
<p>XEONPHI - BENCHMARK SUITE (QPHIX)<br />
2. Compile and Run the XeonPhi-Benchmark Suite<br />
2.1 Compile<br />
2.1.1 Example compilation on PRACE machines<br />
2.1.1.1 BSC - Marenostrum III Hybrid partitions<br />
2.1.1.2 CINES - Frioul<br />
2.2 Run<br />
2.2.1 Main-script: &quot;run_ana.sh&quot;<br />
2.2.2 Main-script: &quot;prepare_submit_job.sh&quot;<br />
2.2.3 Main-script: &quot;submit_job.sh.template&quot;<br />
2.3 Example Benchmark Results</p>
<p><br />
###<br />
###<br />
###&nbsp;&nbsp; GPU - BENCHMARK SUITE<br />
###<br />
###<br />
##<br />
## 1. Compile and Run the GPU-Benchmark Suite<br />
##</p>
<p>##<br />
## 1.1 Compile<br />
##</p>
<p>Download Cmake and Quda</p>
<p>General information how to build QUDA with cmake can be found under:<br />
&quot;https://github.com/lattice/quda/wiki/Building-QUDA-with-cmake&quot;. Here<br />
we just give a short overview:</p>
<p>Build Cmake: (./QCD_Accelerator_Benchmarksuite_Part2/GPUs/src/cmake-3.7.0.tar.gz)</p>
<p>Cmake can be downloaded from source at URL:<br />
https://cmake.org/download/. This guide uses version 3.7.0. The build<br />
instruction can be found in the main directory under &quot;README.rst&quot;. Use<br />
the configure file &quot;./configure&quot; .&nbsp; Then run &quot;gmake&quot; to compile.</p>
<p>Build Quda: (./QCD_Accelerator_Benchmarksuite_Part2/GPUs/src/quda.tar.gz)</p>
<p>Download quda for example by using &quot;git clone<br />
https://github.com/lattice/quda.git&quot;.&nbsp; Create a build-folder. Use<br />
&quot;cmake&quot; in the build-folder, which should be under cmake/bin if you<br />
compiled cmake from source. Execute:</p>
<p>./$PATH2CMAKE/cmake $PATH2QUDA -DQUDA_GPU_ARCH=sm_XX -DQUDA_DIRAC_WILSON=ON -DQUDA_DIRAC_TWISTED_MASS=OFF<br />
-DQUDA_DIRACR_DOMAIN_WALL=OFF -DQUDA_HISQ_LINK=OFF -DQUDA_GAUGE_FORCE=OFF -DQUDA_HISQ_FORCE=OFF -DQUDA_MPI=ON</p>
<p>with</p>
<p>&nbsp;&nbsp; &nbsp;PATH2CMAKE=&lt;path to the cmake-executable&gt;<br />
&nbsp;&nbsp; &nbsp;PAT2QUDA=&lt;path to the home dir of QUDA&gt;</p>
<p>Set -DQUDA_GPU_ARCH=sm_XX to the GPU Architecture (sm_60 for Pascal, sm_35 for Kepler)</p>
<p>If cmake or the compilation fails, library paths and options can be<br />
set via the text user interface of cmake by using &quot;ccmake&quot;.&nbsp; Use<br />
&quot;./PATH2CMAKE/ccmake PATH2BUILD_DIR&quot; to see and edit the available<br />
options. After successfully configuring the buil, run &quot;make&quot;.&nbsp; Now in<br />
the folder test/ one can find the needed Quda executables which begin<br />
with &quot;invert_&quot;.</p>
<p>##<br />
##&nbsp;&nbsp; &nbsp;1.2 Run<br />
##</p>
<p><br />
The Accelerator QCD-Benchmarksuite Part 2 provides bash-scripts<br />
located in the folder<br />
./QCD_Accelerator_Benchmarksuite_Part2/GPUs/scripts&quot; to setup the<br />
benchmark runs on the target machines. This bash-scripts are:</p>
<p>run_ana.sh&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; :&nbsp;&nbsp; Main-script, sets up the benchmark mode and submits the jobs (analyse the results)<br />
prepare_submit_job.sh&nbsp;&nbsp; :&nbsp;&nbsp; Generates the job-scripts<br />
submit_job.sh.template&nbsp; :&nbsp;&nbsp; Template for submit script</p>
<p>##<br />
## 1.2.1 Main-script: &quot;run_ana.sh&quot;<br />
##</p>
<p>The path to the executable has to be set by $PATH2EXE .&nbsp; Upon first<br />
run, QUDA automatically tunes the GPU-kernels by sweeping the number<br />
of threads per block. The optimal setup will be saved in the folder<br />
which pointed to in the environment variable &quot;QUDA_RESOURCE_PATH&quot;. You<br />
must set this variable, otherwise the tune data will be lost and<br />
performance will be sub-optimal. Set it to the folder where the tuning<br />
data should be saved. Strong scaling or Weak scaling can be chosen by<br />
using the variable sca_mode (=&quot;Strong&quot; or =&quot;Weak&quot;).&nbsp; The lattice sizes<br />
can be set by &quot;gx&quot; and &quot;gt&quot;.&nbsp; Choose mode=&quot;Run&quot; for run mode or<br />
mode=&quot;Analysis&quot; for extracting the GFLOPS. Note that the script<br />
assumes Slurm is used as the job scheduler. If not, change the line<br />
which includes the &quot;sbatch&quot; command accordingly.</p>
<p>##<br />
## 1.2.2 Main-script: &quot;prepare_submit_job.sh&quot;<br />
##</p>
<p>Add additional options if necessary.</p>
<p>##<br />
## 1.2.3 Main-script: &quot;submit_job.sh.template&quot;<br />
##</p>
<p>The submit-template will be edited by &quot;prepare_submit_job.sh&quot; to<br />
generate the final submit-script. The first lines (beginning with<br />
&quot;#SBATCH&quot;) depend on the queuing system of the target machine, which in<br />
this case is assumed to be Slurm. These should be changed in case of a<br />
different queuing system.</p>
<p>The Accelerator QCD-Benchmarksuite Part 2 provides bash-scripts to<br />
setup the benchmark runs on the target machines. These bash-scripts<br />
are:</p>
<p>##<br />
## 1.3 Example Benchmark results<br />
##</p>
<p>Here are shown the benchmark results on PizDaint located in Switzerland at CSCS<br />
and the GPGPU-partition of Cartesius at Surfsara based in Netherland, Amsterdam. The runs are performed by using the provided bash-scripts. PizDaint has one Pascal-GPU per node and two different testcases are shown,<br />
the &quot;Strong-Scaling mode with a random lattice configuration of size 32^3x96 and<br />
a &quot;Weak-Scaling&quot; mode with a configuration of local lattice size 48^3x24.<br />
The GPGPU nodes of Cartesius has two Kepler-GPU per node and the &quot;Strong-Scaling&quot; test is shown for the case<br />
that one card per node and two cards per node are used.<br />
The benchmark are done by using the Conjugated Gradient solver which<br />
solve a linear equation, D * x = b, for the unknown solution &quot;x&quot; based on the clover improved Wilson Dirac operator<br />
&quot;D&quot; and a known right hand side &quot;b&quot;.</p>
<p>---------------------<br />
&nbsp; PizDaint - Pascal&nbsp; P100<br />
---------------------<br />
Strong - Scaling:<br />
global lattice size (32x32x32x96)</p>
<p>sloppy-precision: single<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: single</p>
<p>GPUs&nbsp;&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
1&nbsp;&nbsp;&nbsp; 786.520000 4.569600<br />
2&nbsp;&nbsp; 1522.410000 3.086040<br />
4&nbsp;&nbsp; 2476.900000 2.447180<br />
8&nbsp;&nbsp; 3426.020000 2.117580<br />
16&nbsp; 5091.330000 1.895790<br />
32&nbsp; 8234.310000 1.860760<br />
64&nbsp; 8276.480000 1.869230</p>
<p>sloppy-precision: double<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: double</p>
<p>GPUs&nbsp;&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
1&nbsp;&nbsp;&nbsp; 385.965000 6.126730<br />
2&nbsp;&nbsp;&nbsp; 751.227000 3.846940<br />
4&nbsp;&nbsp; 1431.570000 2.774470<br />
8&nbsp;&nbsp; 1368.000000 2.367040<br />
16&nbsp; 2304.900000 2.071160<br />
32&nbsp; 4965.480000 2.095180<br />
64&nbsp; 2308.850000 2.005110</p>
<p><br />
Weak - Scaling:<br />
local lattice size (48x48x48x24)</p>
<p>sloppy-precision: single<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: single</p>
<p>GPUs&nbsp;&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
1&nbsp;&nbsp;&nbsp;&nbsp; 765.967000 3.940280<br />
2&nbsp;&nbsp;&nbsp; 1472.980000 4.004630<br />
4&nbsp;&nbsp;&nbsp; 2865.600000 4.044360<br />
8&nbsp;&nbsp;&nbsp; 5421.270000 4.056410<br />
16&nbsp;&nbsp; 9373.760000 7.396590<br />
32&nbsp; 17995.100000 4.243390<br />
64&nbsp; 27219.800000 4.535410</p>
<p>sloppy-precision: double<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: double</p>
<p>GPUs&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
&nbsp;1&nbsp;&nbsp; 376.611000 5.108900<br />
&nbsp;2&nbsp;&nbsp; 728.973000 5.190880<br />
&nbsp;4&nbsp; 1453.500000 5.144160<br />
&nbsp;8&nbsp; 2884.390000 5.207090<br />
16&nbsp; 5004.520000 5.362020<br />
32&nbsp; 8744.090000 5.623290<br />
64&nbsp; 14053.00000 5.910520</p>
<p><br />
---------------------<br />
&nbsp; SurfSara - Kepler&nbsp;&nbsp; K20m<br />
---------------------<br />
##<br />
## 1 GPU per Node<br />
##</p>
<p>Strong - Scaling:<br />
global lattice size (32x32x32x96)</p>
<p>sloppy-precision: single<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: single<br />
GPUs&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
1&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 243.084000 4.030000<br />
2&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 478.179000 2.630000<br />
4&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 939.953000 2.250000<br />
8&nbsp;&nbsp;&nbsp;&nbsp; 1798.240000 1.570000<br />
16&nbsp;&nbsp;&nbsp; 3072.440000 1.730000<br />
32&nbsp;&nbsp;&nbsp; 4365.320000 1.310000</p>
<p>sloppy-precision: double<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: double</p>
<p>GPUs&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
1&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 119.786000 6.060000<br />
2&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 234.179000 3.290000<br />
4&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 463.594000 2.250000<br />
8&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 898.090000 1.960000<br />
16&nbsp;&nbsp;&nbsp; 1604.210000 1.480000<br />
32&nbsp;&nbsp;&nbsp; 2420.130000 1.630000</p>
<p>##<br />
## 2 GPU per Node<br />
##</p>
<p>Strong - Scaling:<br />
global lattice size (32x32x32x96)</p>
<p>sloppy-precision: single<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: single</p>
<p>GPUs&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
2&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 463.041000 2.720000<br />
4&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 896.707000 1.940000<br />
8&nbsp;&nbsp;&nbsp;&nbsp; 1672.080000 1.680000<br />
16&nbsp;&nbsp;&nbsp; 2518.240000 1.420000<br />
32&nbsp;&nbsp;&nbsp; 3800.970000 1.460000<br />
64&nbsp;&nbsp;&nbsp; 4505.440000 1.430000</p>
<p>sloppy-precision: double<br />
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; precision: double</p>
<p>GPUs&nbsp;&nbsp;&nbsp; GFLOPS&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; sec<br />
2&nbsp;&nbsp;&nbsp;&nbsp; 229.579000 3.380000<br />
4&nbsp;&nbsp;&nbsp;&nbsp; 450.425000 2.280000<br />
8&nbsp;&nbsp;&nbsp;&nbsp; 863.117000 1.830000<br />
16&nbsp;&nbsp; 1348.760000 1.510000<br />
32&nbsp;&nbsp; 1842.560000 1.550000<br />
64&nbsp;&nbsp; 2645.590000 1.480000</p>
<p>###<br />
###<br />
###&nbsp;&nbsp; XEONPHI - BENCHMARK SUITE<br />
###<br />
###</p>
<p>##<br />
## 2. Compile and Run the XeonPhi-Benchmark Suite<br />
##</p>
<p>Unpack the provided source tar-file located in<br />
&quot;./QCD_Accelerator_Benchmarksuite_Part2/XeonPhi/src&quot; or clone the<br />
actual git-hub branches of the code packages QMP:</p>
<p>&quot;git clone https://github.com/usqcd-software/qmp&quot;</p>
<p>and for QPhix</p>
<p>&quot;git clone https://github.com/JeffersonLab/qphix&quot;</p>
<p>Note that the AVX512 instructions, which are needed for an optimal run<br />
on KNLs, are not yet part of the main branch. The AVX512 instructions<br />
are available in the avx512-branch (&quot;git checkout avx512). The<br />
provided source file is using the avx512-branch (Status as of 01/2017).</p>
<p>##<br />
## 2.1 Compile<br />
##</p>
<p>The QPhix library must be built upon QMP, a thin communication layer<br />
on top of MPI. Compile QMP first:</p>
<p>./configure --prefix=$QMP_INSTALL_DIR CC=mpiicc CFLAGS=&quot; -mmic/-xAVX512 -std=c99&quot; --with-qmp-comms-type=MPI --host=x86_64-linux-gnu --build=none-none-none</p>
<p>Create the install folder and link with $QMP_INSTALL_DIR to it.&nbsp; Use<br />
the compiler flag &quot;-mmic&quot; for the compilation for KNC while use<br />
&quot;-xAVX512&quot; for the compilation for KNL.&nbsp; Then use &quot;make&quot; to compile<br />
and &quot;make install&quot; to copy the necessary source files in<br />
$QMP_INSTALL_DIR.</p>
<p>The QPhix executable can be compiled by using, for KNC:</p>
<p>./configure --enable-parallel-arch=parscalar --enable-proc=MIC --enable-soalen=8 --enable-clover --enable-openmp --enable-cean --enable-mm-malloc CXXFLAGS=&quot;-openmp -mmic -vec-report -restrict -mGLOB_default_function_attrs=\&quot;use_gather_scatter_hint=off\&quot; -g -O2 -finline-functions -fno-alias -std=c++0x&quot; CFLAGS=&quot;-mmic -vec-report -restrict -mGLOB_default_function_attrs=\&quot;use_gather_scatter_hint=off\&quot; -openmp -g&nbsp; -O2 -fno-alias -std=c9l9&quot; CXX=mpiicpc CC=mpiicc --host=x86_64-linux-gnu --build=none-none-none --with-qmp=$QMP_INSTALL_DIR</p>
<p>or for KNL:</p>
<p>./configure --enable-parallel-arch=parscalar --enable-proc=AVX512 --enable-soalen=8 --enable-clover --enable-openmp --enable-cean --enable-mm-malloc CXXFLAGS=&quot;-qopenmp -xMIC-AVX512 -g -O3 -std=c++14&quot; CFLAGS=&quot;-xMIC-AVX512 -qopenmp -O3 -std=c99&quot; CXX=mpiicpc CC=mpiicc --host=x86_64-linux-gnu --build=none-none-none --with-qmp=$QMP_INSTALL_DIR</p>
<p>by using the previously set variable QMP_INSTALL_DIR which links to<br />
the folder in which the QMP library was copied. The executable<br />
&quot;time_clov_noqdp&quot; should appear in the &quot;./qphix/test&quot; folder. Note<br />
that the avx512-branch will compile an additional executable which has<br />
dependencies on the package QDP (which will generate an error at the<br />
end of the compilation process).</p>
<p>##<br />
## 2.1.1 Example compilation on PRACE machines<br />
##</p>
<p>In the subsection we provide some example compilation on PRACE machines<br />
which where used to develop the QCD Benchmarksuite 2.</p>
<p>##<br />
## 2.1.1.1 BSC - Marenostrum III Hybrid partitions<br />
##</p>
<p>The nodes of the hybrid partition of Marenostrum are equipped with KNC<br />
cards. First load the following modules:</p>
<p>module unload openmpi<br />
module load impi</p>
<p>and then setup the appropriate environment with:</p>
<p>source /opt/intel/impi/4.1.1.036/bin64/mpivars.sh<br />
source /opt/intel/2013.5.192/composer_xe_2013.5.192/bin/compilervars.sh intel64<br />
export I_MPI_MIC=enable<br />
export I_MPI_HYDRA_BOOTSTRAP=ssh</p>
<p>Configure and compile the QMP-library with:</p>
<p>./configure --prefix=$QMP_INSTALL_DIR CC=mpiicc CFLAGS=&quot;-mmic -std=c99&quot; --with-qmp-comms-type=MPI --host=x86_64-linux-gnu --build=none-none-none</p>
<p>make<br />
make install</p>
<p>Configure and compile QPhix with:</p>
<p>./configure --enable-parallel-arch=parscalar --enable-proc=MIC --enable-soalen=8 --enable-clover --enable-openmp --enable-cean --enable-mm-malloc CXXFLAGS=&quot;-openmp -mmic -vec-report -restrict -mGLOB_default_function_attrs=\&quot;use_gather_scatter_hint=off\&quot; -g -O2 -finline-functions -fno-alias -std=c++0x&quot; CFLAGS=&quot;-mmic -vec-report -restrict -mGLOB_default_function_attrs=\&quot;use_gather_scatter_hint=off\&quot; -openmp -g&nbsp; -O2 -fno-alias -std=c9l9&quot; CXX=mpiicpc CC=mpiicc --host=x86_64-linux-gnu --build=none-none-none --with-qmp=$QMP_INSTALL_DIR<br />
make</p>
<p>##<br />
## 2.1.1.2 CINES - Frioul<br />
##</p>
<p>On a test cluster at CINES the Benchmarksuite was tested on KNL cards.<br />
The steps are similar to Marenostrum above. First setup the appropriate environment with:</p>
<p>source /opt/software/intel/composer_xe_2015/bin/compilervars.sh intel64<br />
source /opt/software/intel/impi_5.0.3/bin64/mpivars.sh</p>
<p>Configure and compile QMP with:<br />
&nbsp;<br />
./configure --prefix=$QMP_INSTALL_DIR CC=mpiicc CFLAGS=&quot;-xMIC-AVX512 -mGLOB_default_function_attrs=&quot;use_gather_scatter_hint=off&quot; -openmp -g&nbsp; -O2 -fno-alias -std=c99&quot;&nbsp; --with-qmp-comms-type=MPI --host=x86_64-linux-gnu --build=none-none-none<br />
make<br />
make install<br />
&nbsp;<br />
Configure and compile QPhix with:<br />
&nbsp;<br />
./configure --enable-parallel-arch=parscalar --enable-proc=AVX512 --enable-soalen=8 --enable-clover --enable-openmp --enable-cean --enable-mm-malloc CXXFLAGS=&quot;-qopenmp -xMIC-AVX512 -g -O3 -std=c++14&quot; CFLAGS=&quot;-xMIC-AVX512 -qopenmp -O3 -std=c99&quot; CXX=mpiicpc CC=mpiicc --host=x86_64-linux-gnu --build=none-none-none --with-qmp=/home/finkenrath/benchmark/qmp/install</p>
<p>and</p>
<p>make</p>
<p>##<br />
##&nbsp;&nbsp; &nbsp;2.2 Run<br />
##</p>
<p><br />
The Accelerator QCD-Benchmarksuite Part 2 provides bash-scripts to<br />
setup the benchmark runs on the target machines. These are:</p>
<p>run_ana.sh&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; :&nbsp;&nbsp; Main-script, set up the bechmark mode and submit the jobs (analyse the results)<br />
prepare_submit_job.sh&nbsp;&nbsp; :&nbsp;&nbsp; Generate the job-scripts<br />
submit_job.sh.template&nbsp; :&nbsp;&nbsp; Template for submit script</p>
<p>##<br />
## 2.2.1 Main-script: &quot;run_ana.sh&quot;<br />
##</p>
<p>The path to the executable has to be set by $PATH2EXE .&nbsp; Choose a<br />
scaling mode between Strong scaling or Weak scaling by setting the<br />
variable sca_mode (=&quot;Strong&quot; or =&quot;Weak&quot;). The lattice sizes can be set<br />
by &quot;gx&quot; and &quot;gt&quot;.&nbsp; Choose between mode=&quot;Run&quot; for run mode or<br />
mode=&quot;Analysis&quot; for extracting the GFLOPS. Note that the script<br />
assumes Slurm is used as the job scheduler. If not, change the line<br />
which includes the &quot;sbatch&quot; command accordingly.</p>
<p>##<br />
## 2.2.2 Main-script: &quot;prepare_submit_job.sh&quot;<br />