Bitterblue Smith <rtl8821cerfe2@xxxxxxxxx> wrote:
> On 25/09/2024 04:25, Ping-Ke Shih wrote:
> > Bitterblue Smith <rtl8821cerfe2@xxxxxxxxx> wrote:
> >> On 23/09/2024 08:47, Ping-Ke Shih wrote:
> >>> Bitterblue Smith <rtl8821cerfe2@xxxxxxxxx> wrote:
> >>>>>>>>>> +
> >>>>>>>>>> +const struct rtw_chip_info rtw8812a_hw_spec = {
> >>>>>>>>>
> >>>>>>>>> Is it possible moving 8812a to individual file?
> >>>>>>>>> Since you have rtw8812au.c and rtw8821au.c.
> >>>>>>>>>
> >>>>>>>>
> >>>>>>>> I think it is possible. But most of the code is common to both chips.
> >>>>>>>> Only the IQ calibration could be moved.
> >>>>>>>
> >>>>>>> Yep, depend on how much IQK code echo chip has.
> >>>>>>>
> >>>>>>
> >>>>>> The IQ calibration for RTL8812AU is about 700 lines.
> >>>>>
> >>>>> rtw8812au -----> (a) rtw8812a
> >>>>> |
> >>>>> v
> >>>>> (b) rtw8821a_common (hard to give a name)
> >>>>> ^
> >>>>> |
> >>>>> rtw8821au -----> (c) rtw8821a
> >>>>>
> >>>>> Put all common code to (b). IQK code in (a) or (c).
> >>>>>
> >>>>> I feel you have thought above picture already. What are problems we will encounter?
> >>>>> Many export symbols? If so, how about below?
> >>>>>
> >>>>> rtw8812au -----> (1) rtw8812a
> >>>>> +---------+
> >>>>> +-> (2) rtw8821a_common (hard to give a name)
> >>>>> +---------+
> >>>>> rtw8821au -----> (3) rtw8821a
> >>>>>
> >>>>> Put rtw8812a_hw_spec and rtw8821a_hw_spec in (2). Only IQK code in (1) and (3)
> >>>>> respectively, and export IQK entry only. Does it work?
> >>>>>
> >>>>>
> >>>> For the name of the common module, I was thinking rtw88_88xxa.ko.
> >>>>
> >>>> I wonder, what is the goal? To put the code in separate kernel
> >>>> modules, or just separate files?
> >>>
> >>> I would like to reduce runtime memory. As I asked, how many IQK code are different
> >>> from them? If you have separated and compiled them, can you share size by the
> >>> output of 'size' command?
> >>>
> >>
> >> I separated the IQK code into two files (still just one module).
> >> size says:
> >>
> >> text data bss dec hex filename
> >> 7192 32 0 7224 1c38 rtw8821a-iqk.o
> >> 12319 40 0 12359 3047 rtw8812a-iqk.o
> >>
> >> This is x86_64.
> >>
> >>>>
> >>>> I think we can have rtw88xxa.c (common code), rtw8821a.c (IQK code,
> >>>> rtw8821a_hw_spec, bluetooth stuff), and rtw8812a.c (IQK code,
> >>>> rtw8812a_hw_spec, some efuse stuff, channel switching)... if these
> >>>> three files compile into a single module, rtw88_88xxa.ko.
> >>>>
> >>>> If each file compiles into a module of its own, we have circular
> >>>> dependencies: rtw8821a_hw_spec -> common code -> IQK code.
> >>>> If *_hw_spec go in the common module, it still depends on both of
> >>>> the other two modules, so what use is it?
> >>>
> >>> If we have dependency of common code -> IQK code, we can't save runtime
> >>> memory, because common code reference to both IQK code. So I felt
> >>> dependency of IQK code would be rtw8812au --> IQK code as above second
> >>> diagram.
> >>>
> >>> But if the work is complicated and save not few runtime memory, we can
> >>> use simple design as current did.
> >>>
> >>>
> >>
> >> The IQK code can be separated into different modules if I duplicate
> >> rtw8821a_ops and rtw8821a_pwr_track, and rtw8821a_phy_pwrtrack takes
> >> a pointer to the IQK function. Then your first diagram above can work.
> >
> > Not sure the "duplicate" you meant. If it only a struct, that would be fine.
> > Not prefer duplicate of tables.
> >
>
> Yes, it's a struct rtw_chip_ops.
>
> >>
> >> The tables also take up a bit of space:
> >>
> >> text data bss dec hex filename
> >> 16832 0 0 16832 41c0 rtw8821a_table.o
> >> 21552 0 0 21552 5430 rtw8812a_table.o
> >
> > Good point.
> >
> >>
> >> I don't know how many kilobytes is enough to make it worth
> >> creating two more modules.
> >
> > I think we can list all *.o related to rtw8821a/8812a, and check the
> > percentage to make decisions. I mean if it occupies 50%, I will prefer
> > to have separated module. But I don't have an exact number now.
> >
>
> text data bss dec hex filename
> 12319 40 0 12359 3047 rtw8812a-iqk.o
> 21552 0 0 21552 5430 rtw8812a_table.o
> 7192 32 0 7224 1c38 rtw8821a-iqk.o
> 16832 0 0 16832 41c0 rtw8821a_table.o
> 29445 429 0 29874 74b2 rtw8821a.o
> =========
> 87340 total. So it's about 38% for 8812a and 27% for 8821a.
> Maybe a bit more in the final version.
chip separated(a) single one(b) increase rate(c)
----- ------------ ------------- ----------------
8812a 63,785 87,841 38%
8821a 53,930 87,841 63%
* increase rate (c) = (b - a) / a
Since increasing rate of 8821a is 63%, I feel separated case would be better.
