Re: [PATCH v2 05/10] drm/fourcc: Add DRM_FORMAT_C[124]

[Pekka Paalanen <ppaalanen@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 10709 bytes --]

On Tue, 15 Mar 2022 09:57:23 +0100
Geert Uytterhoeven <geert@linux-m68k.org> wrote:

> Hi Pekka,
> 
> On Tue, Mar 15, 2022 at 9:46 AM Pekka Paalanen <ppaalanen@gmail.com> wrote:
> > On Tue, 15 Mar 2022 08:51:31 +0100
> > Geert Uytterhoeven <geert@linux-m68k.org> wrote:  
> > > On Tue, Mar 15, 2022 at 8:33 AM Pekka Paalanen <ppaalanen@gmail.com> wrote:  
> > > > On Tue, 15 Mar 2022 09:15:08 +1100 (AEDT)
> > > > Finn Thain <fthain@linux-m68k.org> wrote:  
> > > > > On Mon, 14 Mar 2022, Geert Uytterhoeven wrote:  
> > > > > > On Mon, Mar 14, 2022 at 4:05 PM Pekka Paalanen <ppaalanen@gmail.com> wrote:  
> > > > > > > On Mon, 14 Mar 2022 14:30:18 +0100
> > > > > > > Geert Uytterhoeven <geert@linux-m68k.org> wrote:  
> > > > > > > > On Mon, Mar 7, 2022 at 9:53 PM Geert Uytterhoeven <geert@linux-m68k.org> wrote:  
> > > > > > > > > Introduce fourcc codes for color-indexed frame buffer formats with
> > > > > > > > > two, four, and sixteen colors, and provide a mapping from bit per
> > > > > > > > > pixel and depth to fourcc codes.
> > > > > > > > >
> > > > > > > > > As the number of bits per pixel is less than eight, these rely on
> > > > > > > > > proper block handling for the calculation of bits per pixel and
> > > > > > > > > pitch.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>  
> > > > > > > >  
> > > > > > > > > --- a/include/uapi/drm/drm_fourcc.h
> > > > > > > > > +++ b/include/uapi/drm/drm_fourcc.h
> > > > > > > > > @@ -99,7 +99,10 @@ extern "C" {
> > > > > > > > >  #define DRM_FORMAT_INVALID     0
> > > > > > > > >
> > > > > > > > >  /* color index */
> > > > > > > > > -#define DRM_FORMAT_C8          fourcc_code('C', '8', ' ', ' ') /* [7:0] C */
> > > > > > > > > +#define DRM_FORMAT_C1          fourcc_code('C', '1', ' ', ' ') /* [7:0] C0:C1:C2:C3:C4:C5:C6:C7 1:1:1:1:1:1:1:1 eight pixels/byte */
> > > > > > > > > +#define DRM_FORMAT_C2          fourcc_code('C', '2', ' ', ' ') /* [7:0] C0:C1:C2:C3 2:2:2:2 four pixels/byte */
> > > > > > > > > +#define DRM_FORMAT_C4          fourcc_code('C', '4', ' ', ' ') /* [7:0] C0:C1 4:4 two pixels/byte */
> > > > > > > > > +#define DRM_FORMAT_C8          fourcc_code('C', '8', ' ', ' ') /* [7:0] C 8 one pixel/byte */
> > > > > > > > >
> > > > > > > > >  /* 8 bpp Red */
> > > > > > > > >  #define DRM_FORMAT_R8          fourcc_code('R', '8', ' ', ' ') /* [7:0] R */  
> > > > > > > >
> > > > > > > > After replying to Ilia's comment[1], I realized the CFB drawing
> > > > > > > > operations use native byte and bit ordering, unless
> > > > > > > > FBINFO_FOREIGN_ENDIAN is set.
> > > > > > > > While Amiga, Atari, and Sun-3 use big-endian bit ordering,
> > > > > > > > e.g. Acorn VIDC[2] uses little endian, and SH7760[3] is configurable
> > > > > > > > (sh7760fb configures ordering to match host order).
> > > > > > > > BTW, ssd130{7fb,x}_update_rect() both assume little-endian, so I
> > > > > > > > guess they are broken on big-endian.
> > > > > > > > Fbtest uses big-endian bit ordering, so < 8 bpp is probably broken
> > > > > > > > on little-endian.
> > > > > > > >
> > > > > > > > Hence the above should become:
> > > > > > > >
> > > > > > > >     #define DRM_FORMAT_C1          fourcc_code('C', '1', ' ', ' ') /*
> > > > > > > > [7:0] C7:C6:C5:C4:C3:C2:C1:C0 1:1:1:1:1:1:1:1 eight pixels/byte */
> > > > > > > >     #define DRM_FORMAT_C2          fourcc_code('C', '2', ' ', ' ') /*
> > > > > > > > [7:0] C3:C2:C1:C0 2:2:2:2 four pixels/byte */
> > > > > > > >     #define DRM_FORMAT_C4          fourcc_code('C', '4', ' ', ' ') /*
> > > > > > > > [7:0] C1:C0 4:4 two pixels/byte */
> > > > > > > >
> > > > > > > > The same changes should be made for DRM_FORMAT_[RD][124].
> > > > > > > >
> > > > > > > > The fbdev emulation code should gain support for these with and without
> > > > > > > > DRM_FORMAT_BIG_ENDIAN, the latter perhaps only on big-endian platforms?
> > > > > > > >
> > > > > > > > [1] https://lore.kernel.org/r/CAKb7UvgEdm9U=+RyRwL0TGRfA_Qc7NbhCWoZOft2DKdXggtKYw@mail.gmail.com/
> > > > > > > > [2] See p.30 of the VIDC datasheet
> > > > > > > >     http://chrisacorns.computinghistory.org.uk/docs/Acorn/Misc/Acorn_VIDC_Datasheet.pdf
> > > > > > > > [3] See p.1178 of the SH7660 datasheet
> > > > > > > >     https://datasheet.octopart.com/HD6417760BL200AV-Renesas-datasheet-14105759.pdf  
> > > > > > >
> > > > > > > why would CPU endianess affect the order of bits in a byte?  
> > > > > >
> > > > > > It doesn't, but see below.
> > > > > >  
> > > > > > > Do you mean that bit 0 one machine is (1 << 0), and on another machine
> > > > > > > bit 0 is (1 << 7)?  
> > > > > >
> > > > > > No, I mean that in case of multiple pixels per byte, the display
> > > > > > hardware pumps out pixels to the CRTC starting from either the MSB
> > > > > > or the LSB of the first display byte.  Which order depends on the
> > > > > > display hardware, not on the CPU.
> > > > > >  
> > > > > > > In C, we have only one way to address bits of a byte and that is with
> > > > > > > arithmetic. You cannot take the address of a bit any other way, can you?
> > > > > > >
> > > > > > > Can we standardise on "bit n of a byte is addressed as (1 << n)"?  
> > > > > >
> > > > > > BIT(n) in Linux works the same for little- and big-endian CPUs.
> > > > > > But display hardware may use a different bit order.  
> > > > >
> > > > > Perhaps some of this confusion could be avoided if you describe the
> > > > > problem in terms of the sequence of scan-out of pixels, rather than in
> > > > > terms of the serialization of bits. The significance of bits within each
> > > > > pixel and the ordering of pixels within each memory word are independent,
> > > > > right?  
> > > >
> > > > Yes, that might help.  
> > >
> > > Display:
> > >
> > >      P0  P1  P2  P3  P4  P5  P6  P7  P8  P9 P10 P11 P12 P13 P14 P15
> > >
> > >     P15 P14 P13 P12 P11 P10  P9  P8  P7  P6  P5  P4  P3  P2  P1  P0  
> >
> > Hi Geert,
> >
> > does this mean the display hardware emits even rows from left to right
> > and odd rows from right to left?  
> 
> No, it means I should have my morning coffee first, and remove all
> temporary cruft before pressing send :-(
> 
> The above paragraph should have read:
> 
>     Display (16 pixels):
> 
>         P0  P1  P2  P3  P4  P5  P6  P7  P8  P9 P10 P11 P12 P13 P14 P15
> 
> > I'm guessing P stands for "pixel".  
> 
> Exactly.
> 
> > > Memory:
> > >
> > >   1 bpp (MSB first):
> > >
> > >               bit7 bit6 bit5 bit4 bit3 bit2 bit1 bit0
> > >               ---- ---- ---- ---- ---- ---- ---- ----
> > >       byte 0:   P0   P1   P2   P3   P4   P5   P6   P7
> > >       byte 1:   P8   P9  P10  P11  P12  P13  P14  P15
> > >
> > >   1 bpp (LSB first):
> > >
> > >               bit7 bit6 bit5 bit4 bit3 bit2 bit1 bit0
> > >               ---- ---- ---- ---- ---- ---- ---- ----
> > >       byte 0:   P7   P6   P5   P4   P3   P2   P1   P0
> > >       byte 1:  P15  P14  P13  P12  P11  P10   P9   P8
> > >
> > >   2 bpp (MSB first):
> > >
> > >               bits7-6 bits5-4 bits3-2 bits1-0
> > >               ------- ------- ------- -------
> > >       byte 0:    P0      P1      P2      P3
> > >       byte 1:    P4      P5      P6      P7
> > >       byte 2:    P8      P9     P10     P11
> > >       byte 3:   P12     P13     P14     P15
> > >
> > >   2 bpp (LSB first):
> > >
> > >               bits7-6 bits5-4 bits3-2 bits1-0
> > >               ------- ------- ------- -------
> > >       byte 0:    P3      P2      P1      P0
> > >       byte 1:    P7      P6      P5      P4
> > >       byte 2:   P11     P10      P9      P8
> > >       byte 3:   P15     P14     P13     P12
> > >
> > >   4 bpp (MSB first):
> > >
> > >               bits7-4 bits3-0
> > >               ------- -------
> > >       byte 0:    P0      P1
> > >       byte 1:    P2      P3
> > >       byte 2:    P4      P5
> > >       byte 3:    P6      P7
> > >       byte 4:    P8      P9
> > >       byte 5:   P10     P11
> > >       byte 6:   P12     P13
> > >       byte 7:   P14     P15
> > >
> > >   4 bpp (LSB first):
> > >
> > >               bits7-4 bits3-0
> > >               ------- -------
> > >       byte 0:    P1      P0
> > >       byte 1:    P3      P2
> > >       byte 2:    P5      P4
> > >       byte 3:    P7      P6
> > >       byte 4:    P9      P8
> > >       byte 5:   P11     P10
> > >       byte 6:   P13     P12
> > >       byte 7:   P15     P14  
> >
> > I think I can guess what you meant there, and it looks understandable
> > to me. These tables are actually very clear, and leave only one thing
> > undefined: when multiple bits form a pixel, in which order do the bits
> > form the value. I recall you said fbdev allows for both orderings but
> > only one order is ever used if I understood right.  
> 
> Indeed.  The third ordering is the ordering of the bits in a pixel.
> As fb_bitfield.msb_right is always false, no hardware ever supported by
> fbdev used the other ordering, so we only have to care about:
> 
>    1 bpp: P = [ bitN ]
>    2 bpp: P = [ bitN bitN-1 ]
>    4 bpp: P = [ bitN bitN-1 bitN-2 bitN-3 ]

Excellent!

> > > > Also, when drm_fourcc.h is describing pixel formats, it needs to
> > > > consider only how a little-endian CPU accesses them. That's how pixel
> > > > data in memory is described. Display hardware plays no part in that.
> > > > It is the driver's job to expose the pixel formats that match display
> > > > hardware behaviour.  
> > >
> > > But if the "CPU format" does not match the "display support",
> > > all pixel data must be converted?  
> >
> > Of course. If the driver author does not want to convert pixel data in
> > flight, then the author should not let the driver expose a format that
> > needs conversion.  
> 
> ... in which case we need a DRM fourcc code for the format?

Yes. You can define any new formats you need as long as the format
definition does not depend on (is not affected/modified by) CPU
endianess or any other CPU or display hardware property. I believe this
is the convention used with drm_fourcc.

If the format wanted by display hardware depends on something, then you
need all relevant pixel formats defined and choose at build or driver
initialisation time which ones to expose.

> BTW, Atari and Amiga use bitplanes for bpp <= 8, so they need
> conversion anyway.

Right, that's probably the most reasonable approach. If you really
wanted to expose bitplanes, I could imagine that some new format
modifiers could achieve that.


Thanks,
pq

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]