[hatari-devel] Re: hatari profiler crash?

[Douglas Little <doug694@xxxxxxxxxxxxxx>]

....one more bit of supporting information - although I don't know if it is relevant to this. I notice every time I use the profiler I see this 'Counters OVERFLOW' warning. This happens regardless of how long the profiling session is (a few seconds is enough).

Allocated CPU profile buffer (87 MB).

Allocated CPU profile address buffer (9 KB).

Normal RAM (0-0xE00000):

- active address range:

  0x01ddb6-0x025364

- active instruction addresses:

  1792 (99.56% of all)

- executed instructions:

  15237464 (100.00% of all)

- used cycles:

  3650737951796 (99.65% of all)

- instruction cache misses:

  513112 (99.96% of all)

- Counters OVERFLOW!

Cartridge ROM (0xFA0000-0xFC0000):

- no activity

ROM TOS (0xE00000-0xE80000):

- active address range:

  0xe032ac-0xe032ea

- active instruction addresses:

  8 (0.44% of all)

- executed instructions:

  256 (0.00% of all)

- used cycles:

  12884902445 (0.35% of all)

- instruction cache misses:

  224 (0.04% of all)

- Counters OVERFLOW!

Cache misses per instruction, number of occurrences:

- 0: 14745409

- 1: 478591

- 2: 6415

- 3: 7305

CPU=$23fbc, VBL=6563, FrameCycles=70, HBL=0, LineCycles=70, DSP=$d6

> profile addresses 0 bbbb.txt

Disassembled 1800 (of active 1800) CPU addresses.

On 12 February 2013 10:38, Douglas Little <doug694@xxxxxxxxxxxxxx> wrote:

...one small update - while the profiler crash is now gone, the instruction cycle tally information is still peppered with -1 values, so this will need diagnosed too at some point... (4294967295 == 0xffffffff)

$023ea8 :             move.w    #6,d0                      0.00% (41, 0, 41)

$023eac :             move.w    (a1)+,d1                   0.00% (287, 0, 41)

$023eae :             move.w    d1,$ffffa206.w             0.00% (287, 574, 0)

$023eb2 :             dbra      d0,$23eac                  0.00% (287, 4294967295, 0)

$023eb6 :             move.w    #$1554,d0                  0.00% (41, 4294967295, 0)

$023eba :             clr.w     d1                         0.00% (41, 0, 0)

$023ebc :             move.b    (a1)+,d1                   1.47% (223901, 4294967295, 50)

$023ebe :             move.w    d1,$ffffa206.w             1.47% (223901, 4294967295, 44)

$023ec2 :             dbra      d0,$23ebc                  1..47% (223901, 4294967295, 13)

$023ec6 :             movem.l   (sp)+,d0-d7/a0-a6          0.00% (41, 2542, 0)

$023eca :             tst.b     $c5444                     0..00% (41, 4294967295, 0)

$023ed0 :             bne       $24158                     0.00% (41, 4294967295, 41)

$023ed4 :             lea       $ffffa206.w,a4             0.00% (41, 0, 41)

$023ed8 :             lea       $ffffa202.w,a5             0.00% (41, 0, 41)

$023edc :             moveq     #8,d0                      0.00% (41, 4294967295, 41)

$023ede :             btst      #1,(a5)                    0.00% (41, 164, 0)

$023ee2 :             beq.s     $23ede                     0..00% (41, 4294967295, 0)

On 12 February 2013 10:31, Douglas Little <doug694@xxxxxxxxxxxxxxx> wrote:

Hi just a quick note to let you know the profiler crash is now gone - managed to extract profile data from an optimized build using the latest code without problems.

I'm still not quite geared up to use the profiler in anger - reworking quite a lot of code at the moment - but I will return to it before long when I start on the final optimization pass for both CPU and DSP.

Thanks,

D.

On 11 February 2013 18:44, Eero Tamminen <oak@xxxxxxxxxxxxxx> wrote:

Hi,

On maanantai 28 tammikuu 2013, Douglas Little wrote:
> Conclusion - it works with the patch applied.

I'm pretty sure I found the reason & fix for your crash:
http://hg.tuxfamily.org/mercurialroot/hatari/hatari/rev/3aaa293359f0

Please test.


        - Eero

Ps. Valgrind actually found the bug, but as it seems to propagate undefined
value assignment from one variable to another and report an issue only
when the final variable's value is used for branching, I first looked
into several wrong places...