On 1/28/19 11:03 AM, Tim Harvey wrote:
On Fri, Jan 25, 2019 at 3:57 PM Steve Longerbeam <slongerbeam@xxxxxxxxx> wrote:
Hi Tim, cc: Philipp,
On 1/23/19 3:21 PM, Tim Harvey wrote:
Steve,
I'm testing IMX6 capture again with the tda1997x HDMI receiver found
on Gateworks GW54xx and GW551x boards. This is hooked up to the IMX6
CSI in a way where it can be configured for 8bit BT656 mode
(UYVY8_2X8) or 16bit YUV mode (UYVY8_1X16). Also I have a Marshall
Electronics V-SG4K-HDI HDMI signal generator here so I can test a wide
variety of resolutions, bus formats, interlaced modes, and
colorspaces.
I'm using the following for testing:
- Linux 4.20 +
- media: imx-csi: Input connections to CSI should be optional
- imx-media: Fixes for interlaced capture
- media: imx: Stop stream before disabling IDMA channels
- v4l-utils 1.16.1
- gstreamer-1.14.1 (with the ability to test master as well)
I have a script that generates the media-ctl/v4l2-ctl commands based
on providing a sensor ('adv7180' analog decoder or 'tda1997x' HDMI
decoder) and a 'mode' which I've defined as:
mode0: sensor -> mux -> csi -> /dev/videoN
mode1: sensor -> mux -> csi -> ic_prp -> ic_prpenc -> /dev/videoN (provides a
mode2: sensor -> mux -> csi -> ic_prp -> ic_prpvf -> /dev/videoN
mode3: sensor -> mux -> csi -> vdic -> ic_prp -> ic_prpvf -> /dev/videoN
The media-ctl topologies for each cpu/board combo are at
http://dev.gateworks.com/docs/linux/media/
I'm trying to test out simple v4l2-ctl based capture of a single frame
as well as capture and stream via both software JPEG encode and H264
hardware encode via CODA.
Please let me know of any changes that should be made to the commands
below even if only to purely help document things through clarity.
One of the issues I run into right away is image size: The imx.rst
docs state that the ic has a 'resize' limit of 1024x1024 but I think
I'm mislead by the word 'resize' and this also means you can't push
say 1080x720 through it (not resizing, just putting that on the src
pad) as it clips it to 1024x720.
The IC is limited to a resized *output* frame of 1024x1024, no higher.
Ok so this means for anything >1024 I need to either use the csi to
downscale it by factors of 2 first, or use mem2mem solution which
means I can't use the VDIC (so no interlaced).
Actually that isn't true. You can still use a VDIC pipeline for the
1280x720 interlaced tda19971 source. It's just that you would need to
downscale to <=1024 (either using the CSI /2, and/or using the IC prpvf
scaler). Then use the mem2mem gstreamer element to upscale the prpvf
capture output to something >1024.
I will try your patch
that adds registration of the mem2mem driver to the media-ctl api.
Ok. That patch doesn't have anything functional to it, but it's nice to
see the mem2mem device as part of the media graph.
I believe some of Phillip's pending
patches may be aimed at rectifying this limitation?
If you mean the mem2mem driver, yes. The mem2mem driver makes use the
ipu-image-convert APIs which includes tiled resizing, so it will support
greater-than 1024x1024 output frames. Latest status I've heard is that
there is some misunderstanding of the behavior of rotation, that will
require some rework of the mem2mem driver and ipu-image-convert (and
sigh, the ic-prpencvf sub-device which also supports h/w rotation).
But you could actually grab the last posted version of mem2mem driver
(v7). It works fine except for the rotation misunderstanding (but even
that works but doesn't conform to v4l2 expected behavior). You will then
have a gstreamer element v4l2convertN for >1024 resizing support.
Ok, so I've applied that to my kernel and now have a new /dev/video*
device for ipu_ic_pp. I'll start testing with it.
Cool.
Also recently merged from Philipp is the work to relax capture width
alignment, by DMA'ing to padded frames. But that is not related to mem2mem.
Ok - I'll merge this one too so I'm testing with it.
Ok.
Another issue I'm running into is colorspace conversion, specifically
colorimetry
Example: imx6q-gw54xx tda19971 720p60Hz YUV via BT656 IPU1_CSI0
MODE1:sensor->mux->csi->ic_prp->ic_prpenc
# set sensor output pad to sensor source format
v4l2-ctl -d /dev/v4l-subdev15 --set-dv-bt-timings=query
# sensor format
media-ctl --get-v4l2 '"tda19971 2-0048":0' # fmt:UYVY8_2X8/1280x720
field:none colorspace:rec709
# reset all links
media-ctl --reset
# setup links
media-ctl -l "'tda19971 2-0048':0 -> 'ipu1_csi0_mux':1[1]"
media-ctl -l "'ipu1_csi0_mux':2 -> 'ipu1_csi0':0[1]"
media-ctl -l "'ipu1_csi0':1 -> 'ipu1_ic_prp':0[1]"
media-ctl -l "'ipu1_ic_prp':1 -> 'ipu1_ic_prpenc':0[1]"
media-ctl -l "'ipu1_ic_prpenc':1 -> 'ipu1_ic_prpenc capture':0[1]"
# configure pads
media-ctl -V "'tda19971 2-0048':0 [fmt:UYVY8_2X8/1280x720 field:none]"
media-ctl -V "'ipu1_csi0_mux':2 [fmt:UYVY8_2X8/1280x720 field:none]"
media-ctl -V "'ipu1_csi0':1 [fmt:AYUV32/1280x720]"
media-ctl -V "'ipu1_ic_prp':1 [fmt:AYUV32/1280x720]"
media-ctl -V "'ipu1_ic_prpenc':1 [fmt:AYUV32/1280x720]"
# capture device
media-ctl -e 'ipu1_ic_prpenc capture' # /dev/video0
v4l2-ctl --device /dev/video0 --get-fmt-video
Format Video Capture:
Width/Height : 1024/720
Pixel Format : 'UYVY' (UYVY 4:2:2)
Field : None
Bytes per Line : 2048
Size Image : 1474560
Colorspace : Rec. 709
Transfer Function : Rec. 709
YCbCr/HSV Encoding: ITU-R 601
Quantization : Limited Range
Flags :
^^^ Note 1080x720 has been reduced to 1024x720 - ouch!
Yes as discussed above that is correct behavior, you'll need mem2mem
device to support >1024 tiled scaling.
# capture 1 frame
v4l2-ctl --device /dev/video0 --stream-mmap --stream-to=x.raw --stream-count=1
^^^ works
# stream JPEG/RTP/UDP
gst-launch-1.0 v4l2src device=/dev/video0 ! \
video/x-raw,format=UYVY ! \
jpegenc ! rtpjpegpay ! udpsink host=172.24.20.19 port=5000
^^^ works (but of course squashed horizontally because of x=1024)
# stream H264/RTP/UDP (via coda)
# need to make sure we feed coda NV12/rec709
media-ctl --get-v4l2 '"ipu1_ic_prpenc":0' #
fmt:AYUV8_1X32/1280x720@1/25 field:none colorspace:rec709 xfer:709
ycbcr:601 quantization:lim-range
^^^ input colorspace is rec709 but colorimetry needs to be changed to 709
media-ctl --set-v4l2 '"ipu1_ic_prpenc":0[fmt:AYUV32/1280x720
colorspace:rec709 ycbcr:709]'
^^^ no error but doing a get-v4l2 shows no change - is this all
because of trying to deal with 1080 instead of 1024?
No. The Image Converter can only Y`CbCr encode to ITU-R 601
(V4L2_YCBCR_ENC_601). So the driver won't accept Rec. 709 Y`CbCr
encoding when using an IC pipeline.
See imx_media_fill_default_mbus_fields().
gst-launch-1.0 v4l2src device=/dev/video0 ! \
v4l2h264enc output-io-mode=dmabuf-import ! \
rtph264pay ! udpsink host=172.24.20.19 port=5001
v4l2src0: Device '/dev/video0' does not support 2:0:0:0 colorimetry
^^^ fails because of colorimetry not getting set
Right, that is correct behavior also. See above.
Philipp, can the coda driver accept V4L2_YCBCR_ENC_601 colorspace?
I will have to go back and look what the old downstream Freescale
kernel drivers did here... I could swear I could easily de-interlace
via VDIC and encode 1080i with that and gstreamer-imx.
<snip>
Let's try an interlaced format...
Example: imx6q-gw54xx tda19971 480i60Hz YUV via BT656 IPU1_CSI0
MODE1:sensor->mux->csi->ic_prp->ic_prpenc
# set sensor output pad to sensor source format
v4l2-ctl -d /dev/v4l-subdev15 --set-dv-bt-timings=query
# sensor format
media-ctl --get-v4l2 '"tda19971 2-0048":0' # fmt:UYVY8_2X8/720x480
field:seq-tb colorspace:rec709
# reset all links
media-ctl --reset
# setup links
media-ctl -l "'tda19971 2-0048':0 -> 'ipu1_csi0_mux':1[1]"
media-ctl -l "'ipu1_csi0_mux':2 -> 'ipu1_csi0':0[1]"
media-ctl -l "'ipu1_csi0':1 -> 'ipu1_vdic':0[1]"
media-ctl -l "'ipu1_vdic':2 -> 'ipu1_ic_prp':0[1]"
media-ctl -l "'ipu1_ic_prp':2 -> 'ipu1_ic_prpvf':0[1]"
media-ctl -l "'ipu1_ic_prpvf':1 -> 'ipu1_ic_prpvf capture':0[1]"
# configure pads
media-ctl -V "'tda19971 2-0048':0 [fmt:UYVY8_2X8/720x480 field:seq-tb]"
media-ctl -V "'ipu1_csi0_mux':2 [fmt:UYVY8_2X8/720x480 field:seq-tb]"
media-ctl -V "'ipu1_csi0':1 [fmt:AYUV32/720x480]"
media-ctl -V "'ipu1_vdic':2 [fmt:AYUV32/720x480]"
media-ctl -V "'ipu1_ic_prp':2 [fmt:AYUV32/720x480]"
media-ctl -V "'ipu1_ic_prpvf':1 [fmt:AYUV32/720x480]"
# capture device
media-ctl -e 'ipu1_ic_prpvf capture' # /dev/video1
v4l2-ctl --device /dev/video1 --get-fmt-video
Format Video Capture:
Width/Height : 720/480
Pixel Format : 'UYVY' (UYVY 4:2:2)
Field : None
Bytes per Line : 1440
Size Image : 691200
Colorspace : Rec. 709
Transfer Function : Rec. 709
YCbCr/HSV Encoding: ITU-R 601
Quantization : Limited Range
Flags :
# capture 1 frame
v4l2-ctl --device /dev/video1 --stream-mmap --stream-to=x.raw --stream-count=1
^^^ fails with ipu1_ic_prpvf: EOF timeout
^^^ this is almost identical to the adv7180 case which works... the
only exception I see is field:seq-tb vs field:alternate from the
source which should be fine (more specific than 'alternate')
Hmm, can you send the complete pad formats from --get-v4l2. I don't see
anything obviously wrong here.
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'tda19971 2-0048':0"
[fmt:UYVY8_2X8/720x480 field:seq-tb colorspace:rec709]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_csi0_mux':2"
[fmt:UYVY8_2X8/720x480 field:seq-tb colorspace:rec709]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_csi0':1"
[fmt:AYUV8_1X32/720x480@1/30 field:seq-tb
colorspace:rec709 xfer:709 ycbcr:601 quantization:lim-range]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_vdic':2"
[fmt:AYUV8_1X32/720x480@1/60 field:none
colorspace:rec709 xfer:709 ycbcr:601 quantization:lim-range]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_ic_prp':2"
[fmt:AYUV8_1X32/720x480@1/30 field:none
colorspace:rec709 xfer:709 ycbcr:601 quantization:lim-range]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_ic_prpvf':1"
[fmt:AYUV8_1X32/720x480@1/30 field:none
colorspace:rec709 xfer:709 ycbcr:601 quantization:lim-range]
root@imx6q-gw5404:~# v4l2-ctl --device /dev/video1 --get-fmt-video
Format Video Capture:
Width/Height : 720/480
Pixel Format : 'UYVY' (UYVY 4:2:2)
Field : None
Bytes per Line : 1440
Size Image : 691200
Colorspace : Rec. 709
Transfer Function : Rec. 709
YCbCr/HSV Encoding: ITU-R 601
Quantization : Limited Range
Flags :
root@imx6q-gw5404:~# v4l2-ctl --device /dev/video1 --stream-mmap
--stream-to=x.raw --stream-count=1
[ 159.828654] ipu1_ic_prpvf: EOF timeout
VIDIOC_DQBUF: failed: Input/output error
Sorry everything looks fine, I don't know why that pipeline is failing
to generate any data.
Now lets go back to a 480p60 source but this time include the vdic
(which isn't necessary but should still work right?)
No. First, the CSI will only capture in bt.656 mode if it sees
interlaced fields (bt.656 interlaced sync codes). Well, let me rephrase,
the CSI does support progressive BT.656 but I have never tested that myself.
But more importantly, the VDIC must be given interlaced fields at its input.
ok, I understand now that VDIC can only accept interlaced fields, so
progressive should skip the VDIC.
Example: imx6q-gw54xx tda19971 480p60Hz YUV via BT656 IPU1_CSI0
MODE1:sensor->mux->csi->vdic->ic_prp->ic_prpvf
# set sensor output pad to sensor source format
v4l2-ctl -d /dev/v4l-subdev15 --set-dv-bt-timings=query
# sensor format
media-ctl --get-v4l2 '"tda19971 2-0048":0' # fmt:UYVY8_2X8/720x480
field:none colorspace:rec709
# reset all links
media-ctl --reset
# setup links
media-ctl -l "'tda19971 2-0048':0 -> 'ipu1_csi0_mux':1[1]"
media-ctl -l "'ipu1_csi0_mux':2 -> 'ipu1_csi0':0[1]"
media-ctl -l "'ipu1_csi0':1 -> 'ipu1_vdic':0[1]"
media-ctl -l "'ipu1_vdic':2 -> 'ipu1_ic_prp':0[1]"
media-ctl -l "'ipu1_ic_prp':2 -> 'ipu1_ic_prpvf':0[1]"
media-ctl -l "'ipu1_ic_prpvf':1 -> 'ipu1_ic_prpvf capture':0[1]"
# configure pads
media-ctl -V "'tda19971 2-0048':0 [fmt:UYVY8_2X8/720x480 field:none]"
media-ctl -V "'ipu1_csi0_mux':2 [fmt:UYVY8_2X8/720x480 field:none]"
media-ctl -V "'ipu1_csi0':1 [fmt:AYUV32/720x480]"
media-ctl -V "'ipu1_vdic':2 [fmt:AYUV32/720x480]"
media-ctl -V "'ipu1_ic_prp':2 [fmt:AYUV32/720x480]"
media-ctl -V "'ipu1_ic_prpvf':1 [fmt:AYUV32/720x480]"
# capture device
media-ctl -e 'ipu1_ic_prpvf capture' # /dev/video1
v4l2-ctl --device /dev/video1 --get-fmt-video
Format Video Capture:
Width/Height : 720/480
Pixel Format : 'UYVY' (UYVY 4:2:2)
Field : None
Bytes per Line : 1440
Size Image : 691200
Colorspace : Rec. 709
Transfer Function : Rec. 709
YCbCr/HSV Encoding: ITU-R 601
Quantization : Limited Range
Flags :
# capture 1 frame
v4l2-ctl --device /dev/video1 --stream-mmap --stream-to=x.raw --stream-count=1
ipu1_ic_prpvf: pipeline start failed with -32
^^^ maybe because we are feeding progressive to vdic? vdic isn't
needed here but it seems it should not fail to me
this is a pipeline src->sink mismatch error, because the VDIC enforces
fields at its input pad.
right... as you said above vdic requires interlaced fields so
progressive must skip the vdic.
ok now I'm going to boot with a different device-tree that connects
the tda1997x to IMX6 via 16bit YUV. Advantages of this mode should be
1) better pixel bpp 2) colorimetry is rec709 so no conversion needed
for coda 3) ability to capture 1080p60 as I have confirmed tyring to
do so via bt656 exceeds the IMX pixel clock rate and causes CSI
corruption. Downsides to this mode is that I currently have a bug in
the tda1997x driver regarding this mode and interlaced capture.
Example: imx6q-gw54xx tda19971 480p60Hz YUV via 16bit YUV IPU1_CSI0
MODE0:sensor->mux->csi
# set sensor output pad to sensor source format
v4l2-ctl -d /dev/v4l-subdev15 --set-dv-bt-timings=query
# sensor format
media-ctl --get-v4l2 '"tda19971 2-0048":0' # fmt:UYVY8_1X16/720x480
field:none colorspace:rec709
# reset all links
media-ctl --reset
# setup links
media-ctl -l "'tda19971 2-0048':0 -> 'ipu1_csi0_mux':1[1]"
media-ctl -l "'ipu1_csi0_mux':2 -> 'ipu1_csi0':0[1]"
media-ctl -l "'ipu1_csi0':2 -> 'ipu1_csi0 capture':0[1]"
# configure pads
media-ctl -V "'tda19971 2-0048':0 [fmt:UYVY8_1X16/720x480 field:none]"
media-ctl -V "'ipu1_csi0_mux':2 [fmt:UYVY8_1X16/720x480 field:none]"
media-ctl -V "'ipu1_csi0':2 [fmt:AYUV32/720x480]"
media-ctl --get-v4l2 '"ipu1_csi0":2'
[fmt:UYVY8_1X16/720x480@1/30 field:none
This is passthrough, the CSI can only pass through input pixel data
unmodified as "generic" data, when the input bus is 16-bit parallel,
which it is in this case (UYVY8_1X16). So the pixel code at the CSI
output pad is forced to same as input pad (UYVY8_1X16).
But aside from passthrough, that shouldn't be a problem...
colorspace:rec709 xfer:709 ycbcr:709 quantization:lim-range]
# capture device
media-ctl -e 'ipu1_csi0 capture' # /dev/video4
v4l2-ctl --device /dev/video4 --get-fmt-video
Format Video Capture:
Width/Height : 720/480
Pixel Format : 'UYVY' (UYVY 4:2:2)
Field : None
Bytes per Line : 1440
Size Image : 691200
Colorspace : Rec. 709
Transfer Function : Rec. 709
YCbCr/HSV Encoding: Rec. 709
Quantization : Limited Range
Flags : premultiplied-alpha
# capture 1 frame
v4l2-ctl --device /dev/video4 --stream-mmap --stream-to=x.raw --stream-count=1
convert -size 720x480 -depth 16 uyvy:x.raw /var/www/html/files/frame.png
So capture of 1 frame worked fine?
yes this worked fine
# stream JPEG/RTP/UDP
gst-launch-1.0 v4l2src device=/dev/video4 ! \
video/x-raw,format=UYVY ! \
jpegenc ! rtpjpegpay ! udpsink host=172.24.20.19 port=5000
and this worked?
yes this worked fine
# stream H264/RTP/UDP
media-ctl --get-v4l2 '"ipu1_csi0":0'
[fmt:UYVY8_1X16/720x480@1/30 field:none
colorspace:rec709 xfer:709 ycbcr:709 quantization:lim-range
crop.bounds:(0,0)/720x480
crop:(0,0)/720x480
compose.bounds:(0,0)/720x480
compose:(0,0)/720x480]
^^^ no conv needed
media-ctl --get-v4l2 '"ipu1_csi0":0'
huh?
typo'd here and cut-and-pasted another occurrence of the media-ctl
--get-v4l2 command but my point here is that we have the correct
colorspace/colorimetry for coda so the H264 below should work fine
gst-launch-1.0 v4l2src device=/dev/video4 ! \
v4l2h264enc output-io-mode=dmabuf-import ! \
rtph264pay ! udpsink host=172.24.20.19 port=5001
v4l2src0: Internal data stream error.
Sorry, I don't have enough info to help here.
Here's the pad details:
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'tda19971 2-0048':0"
[fmt:UYVY8_1X16/720x480 field:none colorspace:rec709]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_csi0_mux':2"
[fmt:UYVY8_1X16/720x480 field:none colorspace:rec709]
root@imx6q-gw5404:~# media-ctl --get-v4l2 "'ipu1_csi0':2"
[fmt:UYVY8_1X16/720x480@1/30 field:none
colorspace:rec709 xfer:709 ycbcr:709 quantization:lim-range]
root@imx6q-gw5404:~# v4l2-ctl --device /dev/video4 --get-fmt-video
Format Video Capture:
Width/Height : 720/480
Pixel Format : 'UYVY' (UYVY 4:2:2)
Field : None
Bytes per Line : 1440
Size Image : 691200
Colorspace : Rec. 709
Transfer Function : Rec. 709
YCbCr/HSV Encoding: Rec. 709
Quantization : Limited Range
Flags :
So in this passthrough case v4l2-ctl stream capture works, gstreamer
capture with sw jpeg encode works, but gstreamer capture with h264
encode fails:
root@imx6q-gw5404:~# gst-launch-1.0 v4l2src device=/dev/video4 ! \
v4l2h264enc output-io-mode=dmabuf-import ! \
rtph264pay ! udpsink host=172.24.20.19 port=5001
Setting pipeline to PAUSED ...
Pipeline is live and does not need PREROLL ...
Setting pipeline to PLAYING ...
ERROR: from element /GstPipeline:pipeline0/GstV4l2Src:v4l2src0:
Internal data stream error.
Additional debug info:
gstbasesrc.c(3055): gst_base_src_loop ():
/GstPipeline:pipeline0/GstV4l2Src:v4l2src0:
streaming stopped, reason not-negotiated (-4)
Execution ended after 0:00:00.000416672
Setting pipeline to PAUSED ...
Setting pipeline to READY ...
Setting pipeline to NULL ...
Freeing pipeline ...
It would appear that this must have something to do with gstreamer.
The only difference here between this case and the adv7180 is that
field=none and colorimetry is itu601... I'm guessing the field=none is
causing an issue. I'll post a question to gstreamer-devel.
Ok. Given that v4l2-ctl stream capture works, gstreamer capture with sw
jpeg encode works, but gstreamer capture with h264does not, this does
sound like some issue with gstreamer.
Steve
