Help is available by moving the cursor above any 
symbol or by checking MAQAO website.
[ 4 / 4 ] Application profile is long enough (17.82 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 2.05 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 31.75% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (23.16%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (51.26%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.33% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (22.71%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (23.00%)
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.
[ 4 / 4 ] Affinity is good (97.86%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (41.40%). 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 (0.15%) lower than cumulative innermost loop coverage (23.00%)
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% (0.10%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 22 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
[ 4 / 4 ] Application profile is long enough (17.41 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 2.25 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 25.00% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (23.84%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (51.37%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.25% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (23.32%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (23.68%)
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.
[ 4 / 4 ] Affinity is good (97.34%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (41.34%). 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 (0.15%) lower than cumulative innermost loop coverage (23.68%)
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% (0.10%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 23 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
[ 4 / 4 ] Application profile is long enough (16.94 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 2.29 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 23.81% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (25.33%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (51.71%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.45% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (24.81%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (25.17%)
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.
[ 4 / 4 ] Affinity is good (97.31%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (37.51%). 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 (0.16%) lower than cumulative innermost loop coverage (25.17%)
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% (0.11%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 24 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
[ 4 / 4 ] Application profile is long enough (17.92 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 2.03 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 32.26% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (25.19%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (50.02%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.73% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (24.61%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (25.05%)
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.
[ 4 / 4 ] Affinity is good (97.47%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (42.64%). 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 (0.14%) lower than cumulative innermost loop coverage (25.05%)
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% (0.10%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 24 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
[ 4 / 4 ] Application profile is long enough (17.08 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 1.84 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 38.78% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (23.49%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (51.62%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.54% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (22.95%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (23.33%)
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.
[ 4 / 4 ] Affinity is good (97.27%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (39.57%). 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 (0.16%) lower than cumulative innermost loop coverage (23.33%)
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% (0.10%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 22 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
[ 4 / 4 ] Application profile is long enough (17.26 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 1.94 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 35.38% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (24.92%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (51.17%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.70% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (24.35%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (24.77%)
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.
[ 4 / 4 ] Affinity is good (97.16%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (39.46%). 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 (0.15%) lower than cumulative innermost loop coverage (24.77%)
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% (0.10%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 24 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Loop 914 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
[ 4 / 4 ] Application profile is long enough (17.60 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 2.16 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 28.07% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (24.76%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 1 / 4 ] A significant amount of threads are idle (50.90%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.53% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (24.16%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (24.62%)
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.
[ 4 / 4 ] Affinity is good (97.02%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (43.00%). 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 (0.14%) lower than cumulative innermost loop coverage (24.62%)
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% (0.10%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 24 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
[ 4 / 4 ] Application profile is long enough (18.30 s)
To have good quality measurements, it is advised that the application profiling time is greater than 10 seconds.
[ 1.75 / 3 ] Most of time spent in analyzed modules comes from functions without compilation information
Functions without compilation information (typically not compiled with -g and -grecord-gcc-switches) cumulate 41.67% of the time spent in analyzed modules. Check that -g and -grecord-gcc-switches are present. Remark: if -g and -grecord-gcc-switches are indeed used, this can also be due to some compiler built-in functions (typically math) or statically linked libraries. This warning can be ignored in that case.
[ 3 / 3 ] Optimization level option is correctly used
[ 3 / 3 ] Host configuration allows retrieval of all necessary metrics.
[ 0 / 3 ] Compilation of some functions is not optimized for the target processor
Application run on the GRANITE_RAPIDS micro-architecture while the code was specialized for graniterapids. Architecture specific options are needed to produce efficient code for a specific processor ( -x(target) or -ax(target) ).
[ 2 / 2 ] Application is correctly profiled ("Others" category represents 0.00 % 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 / 4 ] Too little time of the experiment time spent in analyzed loops (25.51%)
If the time spent in analyzed loops is less than 30%, standard loop optimizations will have a limited impact on application performances.
[ 2 / 4 ] A significant amount of threads are idle (49.44%)
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.
[ 4 / 4 ] CPU activity is good
CPU cores are active 96.72% of time
[ 4 / 4 ] Loop profile is not flat
At least one loop coverage is greater than 4% (24.89%), representing an hotspot for the application
[ 4 / 4 ] Enough time of the experiment time spent in analyzed innermost loops (25.37%)
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.
[ 4 / 4 ] Affinity is good (98.56%)
Threads are not migrating to CPU cores: probably successfully pinned
[ 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 (44.65%). 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 (0.14%) lower than cumulative innermost loop coverage (25.37%)
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% (0.09%) is spend in Libm/SVML (special functions)
| Loop ID | Analysis | Penalty Score |
|---|---|---|
| ○Loop 548 - libggml-cpu.so | Execution Time: 24 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ○Loop 1232 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Loop 2304 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Vectorization Roadblocks | 2 | |
| ○ | [SA] Several paths (2 paths) - Simplify control structure or force the compiler to use masked instructions. There are 2 issues ( = paths) costing 1 point each. | 2 |
| ►Loop 399 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 90.91 % - Vector Length Use: 38.76 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 68 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 18 | |
| ○ | [SA] Presence of constant non unit stride data access - Use array restructuring, perform loop interchange or use gather instructions to lower a bit the cost. There are 9 issues ( = data accesses) costing 2 point each. | 18 |
| ►Inefficient Vectorization | 48 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, BLEND/MERGE, SHUFFLE/PERM) - Simplify data access and try to get stride 1 access. There are 48 issues (= instructions) costing 1 point each. | 48 |
| ►Loop 713 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 0.00 % | |
| ►Control Flow Issues | 2 | |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1002 | |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Loop 2015 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 20 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 4 issues (= instructions) costing 4 points each. | 16 |
| ○ | [SA] Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA - Reorganize arithmetic expressions to exhibit potential for FMA. This issue costs 4 points. | 4 |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Loop 2294 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 18.59 % - Vector Length Use: 12.30 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Control Flow Issues | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Data Access Issues | 14 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1003 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (4213 paths) - Simplify control structure. There are 4213 issues ( = paths) costing 1 point, limited to 1000. | 1000 |
| ○ | [SA] Non innermost loop (InBetween) - Collapse loop with innermost ones. This issue costs 2 points. | 2 |
| ►Inefficient Vectorization | 12 | |
| ○ | [SA] Presence of special instructions executing on a single port (BROADCAST) - Simplify data access and try to get stride 1 access. There are 12 issues (= instructions) costing 1 point each. | 12 |
| ►Loop 117 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 9.58 % | |
| ►Control Flow Issues | 1 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ►Data Access Issues | 2 | |
| ○ | [SA] More than 20% of the loads are accessing the stack - Perform loop splitting to decrease pressure on registers. This issue costs 2 points. | 2 |
| ►Vectorization Roadblocks | 1001 | |
| ○ | [SA] Presence of calls - Inline either by compiler or by hand and use SVML for libm calls. There are 1 issues (= calls) costing 1 point each. | 1 |
| ○ | [SA] Too many paths (at least 1000 paths) - Simplify control structure. There are at least 1000 issues ( = paths) costing 1 point. | 1000 |
| ►Loop 3150 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 60.70 % - Vector Length Use: 29.66 % | |
| ►Loop Computation Issues | 4 | |
| ○ | [SA] Presence of expensive FP instructions - Perform hoisting, change algorithm, use SVML or proper numerical library or perform value profiling (count the number of distinct input values). There are 1 issues (= instructions) costing 4 points each. | 4 |
| ►Control Flow Issues | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Data Access Issues | 5 | |
| ○ | [SA] Inefficient vectorization: more than 10% of the vector loads instructions are unaligned - When allocating arrays, don’t forget to align them. There are 0 issues ( = arrays) costing 2 points each | 0 |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Vectorization Roadblocks | 3 | |
| ○ | [SA] Several paths (3 paths) - Simplify control structure or force the compiler to use masked instructions. There are 3 issues ( = paths) costing 1 point each. | 3 |
| ►Inefficient Vectorization | 5 | |
| ○ | [SA] Presence of special instructions executing on a single port (INSERT/EXTRACT, SHUFFLE/PERM, BROADCAST) - Simplify data access and try to get stride 1 access. There are 5 issues (= instructions) costing 1 point each. | 5 |
| ►Loop 2292 - libggml-cpu.so | Execution Time: 0 % - Vectorization Ratio: 0.00 % - Vector Length Use: 6.25 % | |
| ►Loop Computation Issues | 2 | |
| ○ | [SA] Presence of a large number of scalar integer instructions - Simplify loop structure, perform loop splitting or perform unroll and jam. This issue costs 2 points. | 2 |
| ►Data Access Issues | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |
| ►Vectorization Roadblocks | 4 | |
| ○ | [SA] Presence of indirect accesses - Use array restructuring or gather instructions to lower the cost. There are 1 issues ( = indirect data accesses) costing 4 point each. | 4 |