Ideapad

Ideas and suggestions how to improve and extend LIKWID

Assembler code analyzer

It is helpful to be able to analyze the instructions used in a assembler code. This means the instruction types (vectorized/ float / integer) and the instruction count. Other useful information is the loop structure and the size of a loop body instruction code. This would be especially helpful if combined with a profiling run in order to focus on the hot loops.

Performance Top

Implement a performance top based on likwid-perfctr. This can be as top but with all relevant performance parameters for all cores.

Using automatic instrumentation

Use automatic instrumentation to gain a quick overview about runtime distribution, call chains and performance properties.

Latency benchmarks for likwid-bench

Integrate a memory latency benchmark and maybe instruction benchmarks.

Use interrupt feature of HPM

This would enable LIKWID to perform instruction-based, cycle-based or even UOPs based sampling. PROBLEM: Not working without kernel module. Either perf_event or own module.

Use software events provided by perf_event

Hardware events are already supported using perf_event. Some of the software events might be interesting.