On 19/12/17 18:01, Tim Harvey wrote:

On Tue, Dec 19, 2017 at 3:12 AM, Hans Verkuil [off-list ref] wrote:

quoted

On 16/12/17 19:00, Tim Harvey wrote:

quoted

+
+static int tda1997x_fill_format(struct tda1997x_state *state,
+                             struct v4l2_mbus_framefmt *format)
+{
+     const struct v4l2_bt_timings *bt;
+     struct v4l2_hdmi_colorimetry c;
+
+     v4l_dbg(1, debug, state->client, "%s\n", __func__);
+
+     if (!state->detected_timings)
+             return -EINVAL;
+     bt = &state->detected_timings->bt;
+     memset(format, 0, sizeof(*format));
+     c = v4l2_hdmi_rx_colorimetry(&state->avi_infoframe, NULL, bt->height);
+     format->width = bt->width;
+     format->height = bt->height;
+     format->field = (bt->interlaced) ?
+             V4L2_FIELD_ALTERNATE : V4L2_FIELD_NONE;
+     format->colorspace = c.colorspace;
+     format->ycbcr_enc = c.ycbcr_enc;
+     format->quantization = c.quantization;
+     format->xfer_func = c.xfer_func;

This is wrong. v4l2_hdmi_rx_colorimetry returns what arrives on the HDMI link,
that's not the same as is output towards the SoC. You need to take limited/full
range conversions and 601/709 conversions into account since that's what ends
up in memory.

Also note: you are still parsing the colorimetry information from avi_infoframe
in the infoframe parse function. There is no need to do that, just call
v4l2_hdmi_rx_colorimetry and let that function parse and interpret all this.

Otherwise we still have two places that try to interpret that information.

Hans,

Ok so v4l2_hdmi_rx_colorimetry() handles parsing the source avi
infoframe and deals with enforcing the detailed rules and returns
'v4l2' enums:

tda1997x_parse_infoframe(...)
...
        case HDMI_INFOFRAME_TYPE_AVI:
                state->avi_infoframe = frame.avi; /* hold on to avi
infoframe for later use in logging etc */
                /* parse avi infoframe colorimetry data for v4l2
colorspace/ycbcr_encoding/quantization/xfer_func */
                state->hdmi_colorimetry = v4l2_hdmi_rx_colorimetry(&frame.avi,
                                                NULL,
                                                state->timings.bt.height);

Also here I still need to override the quant range passed from the
source avi infoframe per the user control (if not auto) and set per
vic if default:

                /* Quantization Range */
                switch (state->rgb_quantization_range) {
                case V4L2_DV_RGB_RANGE_AUTO:
                        state->range = frame.avi.quantization_range;
                        break;
                case V4L2_DV_RGB_RANGE_LIMITED:
                        state->range = HDMI_QUANTIZATION_RANGE_LIMITED;
                        break;
                case V4L2_DV_RGB_RANGE_FULL:
                        state->range = HDMI_QUANTIZATION_RANGE_FULL;
                        break;
                }
                if (state->range == HDMI_QUANTIZATION_RANGE_DEFAULT) {
                        if (frame.avi.video_code <= 1)
                                state->range = HDMI_QUANTIZATION_RANGE_FULL;
                        else
                                state->range = HDMI_QUANTIZATION_RANGE_LIMITED;
                }

No, the vic check is already done in v4l2_hdmi_rx_colorimetry.

Call v4l2_hdmi_rx_colorimetry first, then:

	/* If ycbcr_enc is V4L2_YCBCR_ENC_DEFAULT, then we receive RGB */
	if (c.ycbcr_enc == V4L2_YCBCR_ENC_DEFAULT)
		switch (state->rgb_quantization_range) {
                	case V4L2_DV_RGB_RANGE_LIMITED:
                        	c.quantization = V4L2_QUANTIZATION_FULL_RANGE;
				break;
			case V4L2_DV_RGB_RANGE_FULL:
                        	c.quantization = V4L2_QUANTIZATION_LIM_RANGE;
				break;
		}

(c is of type struct v4l2_hdmi_colorimetry)

Then tda1997x_fill_format() then needs to fill in details of what's on
the bus so I should be filling in only width/height/field/colorspace
and use colorspace based on my csc conversion chosen output
(V4L2_COLORSPACE_SRGB|V4L2_COLORSPACE_SMPTE170M|V4L2_COLORSPACE_REC709)
and I don't need to set ycbcr_enc/quantization/xfer_func.

You don't touch the colorspace and xfer_func fields. The simple matrix
csc can only change quantization range and/or ycbcr encoding.

It doesn't change the underlying colorspace ('chromaticities') or the
used transfer function.

In practice if the output is RGB then ycbcr_enc should be set to
V4L2_YCBCR_ENC_DEFAULT and quantization to FULL_RANGE. For YUV output you
set ycbcr_enc to V4L2_YCBCR_ENC_601 or 709 and quantization to LIM_RANGE.

Regards,

	Hans

does this sound right?

Thanks,

Tim