options

gmx_mpi - 2023-08-08 09:56:51 - MAQAO 2.17.7

Help is available by moving the cursor above any symbol or by checking MAQAO website.

Global Metrics

Total Time (s)56.97
Profiled Time (s)48.48
Time in analyzed loops (%)80.8
Time in analyzed innermost loops (%)64.4
Time in user code (%)82.4
Compilation Options Score (%)75.0
Perfect Flow Complexity1.05
Array Access Efficiency (%)58.3
GFLOPS1.26 E3
Perfect OpenMP + MPI + Pthread1.04
Perfect OpenMP + MPI + Pthread + Perfect Load Distribution1.17
No Scalar IntegerPotential Speedup1.06
Nb Loops to get 80%10
FP VectorisedPotential Speedup1.07
Nb Loops to get 80%10
Fully VectorisedPotential Speedup1.39
Nb Loops to get 80%21
FP Arithmetic OnlyPotential Speedup1.26
Nb Loops to get 80%21

CQA Potential Speedups Summary

Loop Based Profile⏎

Innermost Loop Based Profile⏎

Application Categorization⏎

Compilation Options⏎

Source ObjectIssue
▼libgromacs_mpi.so.7–
○pairlist_simd_4xm.h-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○fft5d.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○threaded_force_buffer.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pme_gather.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○listed_forces.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○kernel_outer.h-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○manage_threading.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○kernel_prune.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pairs.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pairlist.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○update.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○md_support.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pme.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○kernel_common.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○mdatoms.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○lincs.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pbc.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○constr.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pme_grid.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○localtopology.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○vector.tcc-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pme_solve.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○pme_spread.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○calc_verletbuf.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○bonded.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○vec.h-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○sim_util.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○grid.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○partition.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○arrayref.h-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○domdec_constraints.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○settle.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○kerneldispatch.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.
○atomdata.cpp-march=x86-64 is used but it should be replaced by a more architecture specific option or -march=native.

Loop Path Count Profile⏎

Cumulated Speedup If No Scalar Integer⏎

Cumulated Speedup If FP Vectorized⏎

Cumulated Speedup If Fully Vectorized⏎

Cumulated Speedup If FP Arithmetic Only⏎

Experiment Summary

Application/ccc/work/cont001/ocre/oserete/gromacs-2022.4-install-gcc-ompi/bin/gmx_mpi
Timestamp2023-08-08 09:56:51 Universal Timestamp1691481411
Number of processes observed1 Number of threads observed52
Experiment TypeMPI; OpenMP;
Machineinti6206
Model NameAMD EPYC 7763 64-Core Processor
Architecturex86_64 Micro ArchitectureZEN_V3
Cache Size512 KB Number of Cores64
OS VersionLinux 4.18.0-305.88.1.el8_4.x86_64 #1 SMP Thu Apr 6 10:22:46 EDT 2023
Architecture used during static analysisx86_64 Micro Architecture used during static analysisZEN_V3
Frequency Driveracpi-cpufreq Frequency Governorperformance
Huge Pagesalways Hyperthreadingon
Number of sockets2 Number of cores per socket64
Compilation Options
libgromacs_mpi.so.7: GNU C++17 12.2.0 -mavx2 -mfma -mtune=generic -march=x86-64 -g -g -O2 -std=c++17 -fno-omit-frame-pointer -fcf-protection=none -fPIC -fexcess-precision=fast -funroll-all-loops -fopenmp -fexceptions
CommentsGROMACS compiled with gcc 12.2.0 + OpenMPI, Zen 3, OV1, 10000 steps, 52 cores

Configuration Summary

Dataset
Run Command<executable> mdrun -s ion_channel.tpr -nsteps 10000 -pin on -deffnm gcc
MPI Commandccc_mprun -p milan-bxi -T 1200 -n <number_processes> -x -E --enable_perf -c 128
Number Processes1
Number Nodes1
Number Processes per Nodes1
FilterNot Used
Profile StartNot Used
Maximal Path Number4
×