Re: [hatari-devel] Hatari profiler updates and CPU cycle questions

[ Thread Index | Date Index | More Archives ]

The good news is - the DSP profiler output looks much better and symbol table conversion and import is working properly. Things do appear to map back to the source.

The only problem I noticed involved a few cases where the expected cycle count (?cyc) did not match (total cycles / count), (evaluated to 4cyc instead of 5cyc).... possibly an issue assigning cycles to the previous instruction in some cases? Anyway for the most part it looks right and is usable for performance analysis - with just that exception...

One 'request' I would have is for more decimal precision on the global %age values, since it represents time on individual instructions of which there are many - as opposed to whole functions - this means every opcode needs much finer granularity for differentiation - 2 digits isn't nearly enough. Most opcodes come out at 0.00%, with the highest being 7.xx% or so for cases with long blocking waits... probably 4 to 6 digits is needed to make this bit useful even on a small DSP program... and t's much more severe on large CPU programs.

I'll look more closely at the CPU/devpac symbols later today if I can.


Mail converted by MHonArc 2.6.19+