Re: [PATCH v2 1/7] v4l: Correct the ordering of LSBs of the 10-bit raw packed formats

[Hans Verkuil <hverkuil@xs4all.nl>]

On 06/20/2016 10:38 PM, Dave Stevenson wrote:
> Hi Hans.
> 
> On 20/06/16 18:03, Hans Verkuil wrote:
>> On 06/20/2016 06:20 PM, Sakari Ailus wrote:
>>> The 10-bit packed raw bayer format documented that the data of the first
>>> pixel of a four-pixel group was found in the first byte and the two
>>> highest bits of the fifth byte. This was not entirely correct. The two
>>> bits in the fifth byte are the two lowest bits. The second pixel occupies
>>> the second byte and third and fourth least significant bits and so on.
>>>
>>> Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
>> As mentioned, this needs confirmation. I wonder, isn't this specified in the UVC
>> spec?

I can't find anything in the uvc spec. This format was apparently added for an
Intel R200 3D camera, so I suspect it followed the SMIA/CSI standard and that
the doc is indeed wrong.

So:

Acked-by: Hans Verkuil <hans.verkuil@cisco.com>

Regards,

	Hans

>>
>> Regards,
>>
>> 	Hans
> I'm assuming this is intended to be the same format as generated by many 
> Bayer sensors.
> Those are defined in both the SMIA CCP2 (section 7.9), and MIPI CSI2 
> (section 11.4.4) specs. Whilst nominally restricted, they are both 
> available via unofficial websites if you Google for them (I'm happy to 
> send links, but didn't want to break mailing list rules by just posting 
> them).
> CSI2 draft spec Figure 98 "RAW10 Data Transmission on CSI-2 Bus Bitwise 
> Illustration" is probably the clearest confirmation of the bit ordering.
> 
> dcraw from http://www.cybercom.net/~dcoffin/dcraw/ can consume Raw10 via 
> nokia_load_raw
>      for (dp=data, col=0; col < raw_width; dp+=5, col+=4)
>        FORC4 RAW(row,col+c) = (dp[c] << 2) | (dp[4] >> (c << 1) & 3);
> 
> And checking against the Raspberry Pi hardware simulator, the RAW10 
> parser code has
>              for (i = 0; i < width; i++) {
>                 switch ((i + tile_x) & 3) {
>                    case 0:  val = (buf[0] << 2) | (buf[4] & 3); break;
>                    case 1:  val = (buf[1] << 2) | ((buf[4] >> 2) & 3); 
> break;
>                    case 2:  val = (buf[2] << 2) | ((buf[4] >> 4) & 3); 
> break;
>                    default: val = (buf[3] << 2) | ((buf[4] >> 6) & 3);
> 
> 
> All of those agree with Sakari's update that the first pixel's LSBits 
> are in bits 1..0 of byte 5, 2nd pixel in bits 3..2, etc.
> 
> Regards,
>    Dave
> (working on the Pi CSI2 receiver V4L2 driver as there is now sufficient 
> data in the public domain to do it. I'll be wanting these formats!)
> --
> To unsubscribe from this list: send the line "unsubscribe linux-media" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>