Re: [hatari-devel] Bogus profiler CPU cycles values

[ Thread Index | Date Index | More lists.tuxfamily.org/hatari-devel Archives ]


Great stuff. I'll soon be looking at cycle-level information again - not quite yet, but soon. Will revisit all of this at that point and will report back.

D.

On 15 February 2013 13:51, Eero Tamminen <oak@xxxxxxxxxxxxxx> wrote:
Hi,

On perjantai 15 helmikuu 2013, Douglas Little wrote:
> > Currently I'm showing time in seconds, with 3 decimals, is
> > that enough?  (It's assumed to be used with longer operations)
>
> If it is accumulated time over the duration of the profiling, then yes I
> think that's fine.
>
> (For instantaneous / 'per visit' timings, that would need to be
> represented as cycles only).

All values profiler shows, are accumulated ones.

I eventually decided on 5 decimals, it doesn't look too bad:
-----------------
Normal RAM (0-0x400000):
- no activity
ROM TOS (0xE00000-0xE80000):
- active address range:
  0xe00034-0xe49422
- active instruction addresses:
  16370 (100.00% of all)
- executed instructions:
  30785602 (100.00% of all)
- instruction cache misses:
  5085856 (100.00% of all)
- used cycles:
  167483888 (100.00% of all)
  = 10.44002s
Cartridge ROM (0xFA0000-0xFC0000):
- no activity

= 10.44002s

Cache misses per instruction, number of occurrences:
- 0: 26289475
- 1: 4121892
- 2: 158741
- 3: 215494
-----------------

(5 decimals might be useful for timing when breakpoints are close together.)


This is commit for profiler itself:
http://hg.tuxfamily.org/mercurialroot/hatari/hatari/rev/9a9c44adb88f

Data header and post-processor changes come later.


        - Eero





Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/