[09:46] <hverkuil> mszyprow: ping
[09:47] <mszyprow> hverkuil: pong... I have that task on my todo list...
[09:47] <mszyprow> hverkuil: but I got urgent issue to fix in meantime
[09:51] <hverkuil> mszyprow: just a quick question (might save you from having to test cec):
[09:51] <hverkuil> I got my odroid board to work with the latest kernel and cec patches over the weekend.
[09:55] <hverkuil> But I don't see a cec device. Are there some requirements? I'm using the drm HDMI driver, not the s5p-tv.
[09:55] <hverkuil> I would expect most users to use the drm driver.
[09:57] <hverkuil> I just want to make sure I understand
[10:49] *** WeaselSoup has quit IRC (Quit: WOOPS)
[12:16] <ad-n770> hi
[12:16] <ad-n770> I'm looking into one issue in the gstreamer element v4l2src that seems to lead to a discrepancy in the v4l documentation
[12:17] <ad-n770> the issue is reproduced with vivid and a uvcvideo compatible hardware
[12:18] <ad-n770> when NV12 format is selected for capture the GStreamer element produces wrong colorimetry information that leads black turn to gray
[12:18] <ad-n770> we tracked down to the following facts
[12:20] <ad-n770> 1) for NV12 is reported V4L2_COLORSPACE_SRGB and V4L2_QUANTIZATION_DEFAULT
[12:21] <ad-n770> 2) at https://linuxtv.org/downloads/v4l-dvb-apis/ch02s06.html#col-srgb it's said "The default Y'CbCr quantization is full range."
[12:22] <ad-n770> 3) at https://linuxtv.org/downloads/v4l-dvb-apis/ch02s05.html bottom note for V4L2_QUANTIZATION_DEFAULT it's said "V4L2_QUANTIZATION_DEFAULT
[12:22] <ad-n770> Use the default quantization encoding as defined by the colorspace. This is always full range for R'G'B' (except for the BT.2020 colorspace) and usually limited range for Y'CbCr."
[12:24] <ad-n770> seems that the issue in the colors is due the fact that we followed the rule of full range because V4L2_COLORSPACE_SRGB
[12:25] <ad-n770> but we are not sure if it's either correct for NV12 reporting as SRGB or if we should follow the note in V4L2_QUANTIZATION_DEFAULT "usually limited range for Y'CbCr."
[12:26] <ad-n770> please could you give us some light on this?
[12:27] <hverkuil> ad-n770: I believe that the ENC_SYCC documentation is wrong. I think it is actually the same as ENC_601 and that it is limited range instead of full range.
[12:29] <hverkuil> What is confusing is that SYCC is the same as BT601 within the limited range, but allows values above and below that range.
[12:30] *** dannas has left 
[12:31] <hverkuil> It is best to follow the V4L2_MAP_QUANTIZATION_DEFAULT() macro from videodev2.h: that does the right thing to the best of my knowledge.
[12:32] <hverkuil> The spec regarding SYCC is confusing. I plan on testing this with a signal generator to double check this.
[12:32] <hverkuil> In the meantime I'll prepare a patch to bring the spec in line with the V4L2_MAP_QUANTIZATION_DEFAULT macro.
[13:01] <hverkuil> ad-n770: I refreshed my memory:
[13:02] <ad-n770> tell me
[13:02] <hverkuil> ENC_SYCC is the same as ENC_601 using full quantization range.
[13:03] <ad-n770> actually vivid reports V4L2_YCBCR_ENC_DEFAULT instead of ENC_SYCC for NV12
[13:04] <ad-n770> maybe what is wrong is report V4L2_COLORSPACE_SRGB for NV12?
[13:04] <hverkuil> But I am pretty sure that most devices that report COLORSPACE_SRGB but provide a YUV format actually implement full range YUV, that will use limited range instead.
[13:05] <hverkuil> So the default for RGB in NV12 should be limited range ENC_601.
[13:05] <hverkuil> which is what the V4L2_MAP_YCBCR_ENC_DEFAULT and V4L2_MAP_QUANTIZATION_DEFAULT macros do.
[13:06] <hverkuil> The code is right, the spec isn't.
[13:07] <ad-n770> I've opened a bug in GStreamer for the issue https://bugzilla.gnome.org/show_bug.cgi?id=762529
[13:07] <ad-n770> I'm not sure yet how to implement the fix there
[13:08] <hverkuil> They should use the V4L2 macros to map default values to actual values. Not try to do that themselves.
[13:09] <hverkuil> If there are bugs in that mapping then it only has to be fixed in one place (the macro).
[13:10] <ad-n770> then probably add something at https://cgit.freedesktop.org/gstreamer/gst-plugins-good/tree/sys/v4l2/gstv4l2object.c#n1951 that uses the macro
[13:10] <ad-n770> and just use the switch case to translate the value to gst enums
[13:11] <ad-n770> going to try this
[13:11] <hverkuil> I think I have a nice example, let me look it up
[13:13] <hverkuil> http://git.linuxtv.org/v4l-utils.git/tree/utils/qv4l2/capture-win-gl.cpp
[13:13] <hverkuil> function CaptureWinGLEngine::setColorspace
[13:14] <hverkuil> the end result is a tuple of (colorspace, xfer_func, ycbcr_enc, quantization) that sets these values explicitly (i.e. no more DEFAULTs).
[13:15] <hverkuil> Of course, ycbcr_enc only makes sense for ycbcr formats and is ignored for rgb.
[13:17] <ad-n770> ok, I''m going to write a patch following this pattern and I'll add your pointed code as reference
[18:33] <nohous> hi, does anyone here know the CoaXPress video transmission standard?
[18:43] *** awalls has left 
[18:46] *** dannas has left 
[22:58] *** posciak has quit IRC (*.net *.split)
[22:58] *** aballier has quit IRC (*.net *.split)
[22:58] *** crope has quit IRC (*.net *.split)
[22:58] *** agd5f has quit IRC (*.net *.split)
[22:58] *** harrow has quit IRC (*.net *.split)
[22:58] *** enyc has quit IRC (*.net *.split)
[22:58] *** r0kc4t has quit IRC (*.net *.split)
[22:58] *** ChanServ has quit IRC (*.net *.split)
[23:06] *** leguin.freenode.net sets mode: +o ChanServ