[ 4 / 4 ] Application profile is long enough (185.86 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improves the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.54 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Application profile is long enough (119.58 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improves the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.63 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (70.38%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 151.45% of observed threads are actually active

[ 3 / 4 ] CPU activity is below 90% (75.72%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (61.62%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (64.31%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 3 / 4 ] Affinity stability is lower than 90% (76.91%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (45.15%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (6.07%) lower than cumulative innermost loop coverage (64.31%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)

[ 4 / 4 ] Application profile is long enough (87.58 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improve the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 13.13 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (70.27%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 205.28% of observed threads are actually active

[ 2 / 4 ] CPU activity is below 90% (51.32%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (61.72%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (64.50%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 2 / 4 ] Affinity stability is lower than 90% (52.92%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (61.51%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (5.78%) lower than cumulative innermost loop coverage (64.50%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)

[ 4 / 4 ] Application profile is long enough (71.32 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improve the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 13.00 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (70.38%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 251.95% of observed threads are actually active

[ 1 / 4 ] CPU activity is below 90% (31.50%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (61.70%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (64.48%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 1 / 4 ] Affinity stability is lower than 90% (32.92%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (75.25%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (5.91%) lower than cumulative innermost loop coverage (64.48%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)

[ 4 / 4 ] Application profile is long enough (63.26 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improve the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.82 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (70.37%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 285.32% of observed threads are actually active

[ 0 / 4 ] CPU activity is below 90% (17.83%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (61.61%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (64.41%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 0 / 4 ] Affinity stability is lower than 90% (18.82%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (85.13%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (5.96%) lower than cumulative innermost loop coverage (64.41%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)

[ 4 / 4 ] Application profile is long enough (59.12 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improves the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.47 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (70.69%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 308.11% of observed threads are actually active

[ 0 / 4 ] CPU activity is below 90% (9.63%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (61.03%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (64.68%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 0 / 4 ] Affinity stability is lower than 90% (10.55%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (90.58%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (6.01%) lower than cumulative innermost loop coverage (64.68%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)

[ 4 / 4 ] Application profile is long enough (57.74 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improves the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.02 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (71.18%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 321.05% of observed threads are actually active

[ 0 / 4 ] CPU activity is below 90% (6.69%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (60.48%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (65.22%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 0 / 4 ] Affinity stability is lower than 90% (7.50%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (92.68%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (5.95%) lower than cumulative innermost loop coverage (65.22%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)

[ 4 / 4 ] Application profile is long enough (56.97 s)

To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.

[ 3 / 3 ] Optimization level option is correctly used

[ 3 / 3 ] Most of time spent in analyzed modules comes from functions compiled with -g and -fno-omit-frame-pointer

-g option gives access to debugging informations, such are source locations. -fno-omit-frame-pointer improves the accuracy of callchains found during the application profiling.

[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.

[ 3 / 3 ] Architecture specific option -mcpu is used

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.17 % of the execution time)

To have a representative profiling, it is advised that the category "Others" represents less than 20% of the execution time in order to analyze as much as possible of the user code

[ 1 / 1 ] Lstopo present. The Topology lstopo report will be generated.

[ 0 / 0 ] Fastmath not used

Consider to add ffast-math to compilation flags (or replace -O3 with -Ofast) to unlock potential extra speedup by relaxing floating-point computation consistency. Warning: floating-point accuracy may be reduced and the compliance to IEEE/ISO rules/specifications for math functions will be relaxed, typically 'errno' will no longer be set after calling some math functions.

[ 4 / 4 ] Enough time of the experiment time spent in analyzed loops (71.49%)

If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.

[ 4 / 4 ] Threads activity is good

On average, more than 329.00% of observed threads are actually active

[ 0 / 4 ] CPU activity is below 90% (5.14%)

CPU cores are idle more than 10% of time. Threads supposed to run on these cores are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (59.71%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (65.70%)

If the time spent in analyzed innermost loops is less than 15%, standard innermost loop optimizations such as vectorisation will have a limited impact on application performances.

[ 0 / 4 ] Affinity stability is lower than 90% (5.84%)

Threads are often migrating to other CPU cores/threads. For OpenMP, typically set (OMP_PLACES=cores OMP_PROC_BIND=close) or (OMP_PLACES=threads OMP_PROC_BIND=spread). With OpenMPI + OpenMP, use --bind-to core --map-by node:PE=$OMP_NUM_THREADS --report-bindings. With IntelMPI + OpenMP, set I_MPI_PIN_DOMAIN=omp:compact or I_MPI_PIN_DOMAIN=omp:scatter and use -print-rank-map.

[ 3 / 3 ] Less than 10% (0.00%) is spend in BLAS1 operations

It could be more efficient to inline by hand BLAS1 operations

[ 0 / 3 ] Too many functions do not use all threads

Functions running on a reduced number of threads (typically sequential code) cover at least 10% of application walltime (93.53%). Check both "Max Inclusive Time Over Threads" and "Nb Threads" in Functions or Loops tabs and consider parallelizing sequential regions or improving parallelization of regions running on a reduced number of threads

[ 3 / 3 ] Cumulative Outermost/In between loops coverage (5.80%) lower than cumulative innermost loop coverage (65.70%)

Having cumulative Outermost/In between loops coverage greater than cumulative innermost loop coverage will make loop optimization more complex

[ 2 / 2 ] Less than 10% (0.00%) is spend in BLAS2 operations

BLAS2 calls usually could make a poor cache usage and could benefit from inlining.

[ 2 / 2 ] Less than 10% (0.00%) is spend in Libm/SVML (special functions)