[ 4 / 4 ] Application profile is long enough (154.78 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

[ 3.00 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 11.38 % 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 (69.51%)

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

[ 3 / 4 ] CPU activity is below 90% (78.95%)

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.

[ 3 / 4 ] A significant amount of threads are idle (21.05%)

On average, more than 10% of observed threads are idle. Such threads are probably IO/sync waiting. Some hints: use faster filesystems to read/write data, improve parallel load balancing and/or scheduling.

[ 3 / 4 ] Affinity stability is lower than 90% (78.08%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (10.45%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (37.71%)

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 / 3 ] Cumulative Outermost/In between loops coverage (31.80%) lower than cumulative innermost loop coverage (37.71%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (82.17 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

[ 3.00 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 11.00 % 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 (71.58%)

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

[ 3 / 4 ] CPU activity is below 90% (73.06%)

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 ] Threads activity is good

On average, more than 126.97% of observed threads are actually active

[ 3 / 4 ] Affinity stability is lower than 90% (74.00%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (10.55%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (38.04%)

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 / 3 ] Cumulative Outermost/In between loops coverage (33.54%) lower than cumulative innermost loop coverage (38.04%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (46.46 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

[ 3.00 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 11.45 % 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 (70.88%)

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

[ 2 / 4 ] CPU activity is below 90% (64.27%)

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 ] Threads activity is good

On average, more than 184.95% of observed threads are actually active

[ 3 / 4 ] Affinity stability is lower than 90% (71.58%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (9.92%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (37.09%)

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 / 3 ] Cumulative Outermost/In between loops coverage (33.79%) lower than cumulative innermost loop coverage (37.09%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (29.07 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

[ 3.00 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 11.91 % 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 (71.31%)

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

[ 2 / 4 ] CPU activity is below 90% (50.92%)

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 ] Threads activity is good

On average, more than 240.64% of observed threads are actually active

[ 2 / 4 ] Affinity stability is lower than 90% (66.39%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (9.96%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (36.87%)

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 / 3 ] Cumulative Outermost/In between loops coverage (34.44%) lower than cumulative innermost loop coverage (36.87%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (19.40 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

[ 2.99 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 12.18 % 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 (70.67%)

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

[ 1 / 4 ] CPU activity is below 90% (36.30%)

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 ] Threads activity is good

On average, more than 283.88% of observed threads are actually active

[ 2 / 4 ] Affinity stability is lower than 90% (66.10%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (9.88%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (36.45%)

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 / 3 ] Cumulative Outermost/In between loops coverage (34.22%) lower than cumulative innermost loop coverage (36.45%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (15.29 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

[ 2.99 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 2 / 2 ] Application is correctly profiled ("Others" category represents 14.84 % 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 (68.06%)

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

[ 0 / 4 ] CPU activity is below 90% (23.77%)

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 ] Threads activity is good

On average, more than 323.89% of observed threads are actually active

[ 2 / 4 ] Affinity stability is lower than 90% (52.75%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (9.29%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (34.58%)

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 / 3 ] Cumulative Outermost/In between loops coverage (33.47%) lower than cumulative innermost loop coverage (34.58%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (14.30 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

[ 2.99 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 0 / 2 ] Too much execution time spent in category "Others" (21.82 %)

If the category "Others" represents more than 20% of the execution time, it means that the application profile misses a representative part of the application. Examine functions details to properly identify “Others” category components. Rerun after adding most represented library names (e.g. more than 20% of coverage) to external_libraries (the names can be directly provided by ONE View)

[ 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 (59.39%)

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

[ 0 / 4 ] CPU activity is below 90% (18.53%)

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 ] Threads activity is good

On average, more than 475.63% of observed threads are actually active

[ 2 / 4 ] Affinity stability is lower than 90% (63.91%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (10.73%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (30.93%)

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 / 3 ] Cumulative Outermost/In between loops coverage (28.46%) lower than cumulative innermost loop coverage (30.93%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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 (14.68 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

[ 2.98 / 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.

[ 2 / 3 ] Security settings from the host restrict profiling. Some metrics will be missing or incomplete.

Current value for kernel.perf_event_paranoid is 2. If possible, set it to 1 or check with your system administrator which flag can be used to achieve this.

[ 0 / 3 ] Compilation of some functions is not optimized for the target processor

Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ). Application run on the ZEN_V4 micro-architecture while the code was specialized for znver3.

[ 0 / 2 ] Too much execution time spent in category "Others" (41.53 %)

If the category "Others" represents more than 20% of the execution time, it means that the application profile misses a representative part of the application. Examine functions details to properly identify “Others” category components. Rerun after adding most represented library names (e.g. more than 20% of coverage) to external_libraries (the names can be directly provided by ONE View)

[ 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 (41.10%)

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

[ 0 / 4 ] CPU activity is below 90% (16.46%)

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 ] Threads activity is good

On average, more than 855.71% of observed threads are actually active

[ 3 / 4 ] Affinity stability is lower than 90% (82.59%)

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.

[ 4 / 4 ] Loop profile is not flat

At least one loop coverage is greater than 4% (8.73%), representing an hotspot for the application

[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (20.91%)

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 / 3 ] Cumulative Outermost/In between loops coverage (20.19%) lower than cumulative innermost loop coverage (20.91%)

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

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

It could be more efficient to inline by hand BLAS1 operations

[ 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)