kicad-developers team mailing list archive

[jp charras <jp.charras@xxxxxxxxxx>]

Le 25/07/2017 à 15:58, Wayne Stambaugh a écrit :
> On 7/24/2017 7:50 PM, hauptmech wrote:
>>
>> I haven't learned enough about the KIWAY to trace the events from pcbnew
>> to eeschema yet.
>>
>> I'd be curious is this is reproducible in other systems. Easy to test on
>> linux using the perf tool https://perf.wiki.kernel.org/index.php/Main_Page
>>
>> 1) run kicad, open a pcb and schematic of any size.
>> 2) in a terminal run > perf record -g -p $(pidof kicad)
>> 3) pan and zoom the pcb continuously for a statistically significant
>> time (about 15-30 sec works for me).
> 
> This makes even less sense.  Why would the schematic sheet list need to
> be rebuilt while panning and zooming a board in pcbnew?  I'm assuming
> that this doesn't happen in the legacy canvas.

In fact, SCH_EDIT_FRAME::IsAutoSaveRequired() (which calls BuildSheetList() ) is called on each
mouse event, even if the SCH_EDIT_FRAME does not have the focus (which is a bit strange).

So this is not panning and zooming a board which calls BuildSheetList(), but any mouse move event.

I do not think this is the explanation of Performance issues in pcbnew,
however SCH_EDIT_FRAME::IsAutoSaveRequired() needs an optimization:
no need to call BuildSheetList() to know if a screen is modified, especially if the SCH_EDIT_FRAME
does not have the focus.

> 
>> 4) CTRL C stop perf
>> 5) > perf report
>>
>> On a release build SCH_SHEET_LIST() or a sub function shows up at the top.
>>
>>
>>
>> On 25/07/17 10:25, Wayne Stambaugh wrote:
>>> On 7/24/2017 6:12 PM, hauptmech wrote:
>>>> Yeah, it's weird since all I'm doing when collecting perf data is
>>>> zooming and panning the pcb, (with eeschema open).  A root page and 11
>>>> sub pages at the next level. Roughly 800 components and 2000 wires.
>>>>
>>>> Probably something somewhere is instantiating a new SCH_SHEET_LIST
>>>> (which then has to chew through each and every schematic item to find
>>>> the sheets) on every event?
>>> This is even more disturbing.  When did we start building the sheet list
>>> in pcbnew and more importantly why?
>>>
>>>>
>>>>
>>>> On 25/07/17 01:59, Wayne Stambaugh wrote:
>>>>> On 7/23/2017 7:36 PM, hauptmech wrote:
>>>>>> I'm getting a distracting amount of lag when panning and zooming with
>>>>>> GAL. Perf reports TOOL_MANAGER::dispatchInternal and, when eeschema is
>>>>>> open, SCH_SHEET_LIST::BuildSheetList, as the main time sinks.
>>>>> Given that all BuildSheetList does is create a list of unique sheets in
>>>>> a schematic, I find it difficult to believe that you would have that
>>>>> many sheets (hierarchical or otherwise) that would cause a performance
>>>>> hit.  How many sheets does your design have and how deeply are they
>>>>> nested?
>>>>>
>>>>>> Anyone have any hints for me improve this?
>>>>>>
>>>>>> In general I'm getting the impression that the design I'm working
>>>>>> on is
>>>>>> a bit more complex than the developers normally test on. I'll see
>>>>>> if my
>>>>>> client will allow a sanitized version of this board to go into /demos.
>>>>>>
>>>>>> -hauptmech
>>>>>>



-- 
Jean-Pierre CHARRAS