[libcamera-devel] [PATCH 00/11] libcamera: introduce Software ISP and Software IPA

[Bryan O'Donoghue]

On 14/12/2023 16:25, Hans de Goede wrote:
> Hi,
> 
> On 12/14/23 17:02, Hans de Goede wrote:
>> Hi Bryan,
>>
>> On 12/14/23 16:16, Bryan O'Donoghue wrote:
>>> On 14/12/2023 14:47, Hans de Goede wrote:
>>
>> <snip>
>>
>>>> BTW I'm currently working on adding support for 10bpp unpacked
>>>> bayer and while working on this I noticed a small bug in the swstats
>>>> code, you may want to squash in this fix:
>>>
>>> Applied that change.
>>>
>>> I'm comparing to your earlier branch
>>>
>>> commit ae92fa44991ded151c63d5d202efdacc9d640aff (HEAD -> SoftwareISP-v01-hans1, softisp/SoftwareISP-v01-hans1)
>>> Author: Hans de Goede <hdegoede at redhat.com>
>>> Date:   Thu Nov 30 20:13:29 2023 +0100
>>>
>>>
>>> On the earlier branch I get 30fps @ 68% CPU usage. On this branch I'm getting 18fps @ 100%.
>>> Seems to be eating alot more cycles
>>
>> Ok, that is no good.
>>
>> Can you try:
>>
>> https://gitlab.freedesktop.org/camera/libcamera-softisp/-/commits/SoftwareISP-v02-hans3
>>
>> And then start with checking out:
>>
>> 64c9bc0f55f459b9722fe78dfd35a07fb35d2a7b ("libcamera: software_isp: Add debayer-line callbacks for bayer patterns repeating every 4 lines")
>>
>> that is basically softisp/SoftwareISP-v01-hans1 rebased on top of SoftwareISP-v02,
>> so that should give you to same performance as before.
>>
>> First please check this indeed restores performance ?
>>
>> And then after that try newer commits from that branch in this order
>> (I'm skipping commits which should not have a performance impact here):
>>
>> 7136d4d59aafb2564a24a2d4be773ca220257fdc
>> e3c2a5931dd825c58f626da8c12429b70d20219b (not really expecting a performance impact from this one but maybe)
>> b3aa4e4f781e881746953177dacbce8c943cb5a3 (bugfix but one which is expected to have some performance impact)
>> ec75339a5bff8b8e9a0031b9408bd47020905a4f
>> f62ed5df54850a24406bf3a762271a5fa3c0303d
>>
>> The reason why I'm asking you to test these instead of do a bisect
>> is since there might be a number of smaller performance regressions
>> all adding up ...
>>
>> If you can let me know the results of this then I'll see if I can
>> restore the old performance for you.
>>
>> Note that as mentioned above b3aa4e4f781e is a bug fix which
>> some potentially big performance implications but the old code
>> really was wrong there ...
>>
>>> and producing a more pinkish result on my hw.
>>
>> As part of my work I more or less rewrote white-balancing, but I thought I fixed
>> the pinkish thing. It seems the new white-balancing is somewhat sensitive to
>> over-exposure and your image does look a bit overexposed.
> 
> p.s.
> 
> It would also be good to compare:
> 
> https://gitlab.freedesktop.org/camera/libcamera-softisp/-/commits/SoftwareISP-v02-hans2
> https://gitlab.freedesktop.org/camera/libcamera-softisp/-/commits/SoftwareISP-v02-hans3

Pavel pointed out libcamera is -Os not -O3

meson setup -Dprefix=/home/deckard/Development/FireFox/libcamera-install 
-Dpipelines=simple/simple -Dipas=simple/simple --buildtype=release 
build.swisp

Switching to release brings the cpu occupancy down to 60% on the topic 
branch for this series.

We should probably shouldn't be too concerned about performance of -Os 
code for this case.

Still looks washed out / red in comparsion to 
64c9bc0f55f459b9722fe78dfd35a07fb35d2a7b

I think that warrants fixing.

---
bod