RE: omap3isp: cropping bug in previewer?

RE: omap3isp: cropping bug in previewer?

[Florian Neuhaus]

Hi Laurent,

Laurent Pinchart wrote on 2012-07-05:

> There's really an issue, which was introduced in v3.5-rc1. Could you please try
> the following patch instead of yours ?
> 
> diff --git a/drivers/media/video/omap3isp/isppreview.c
> b/drivers/media/video/omap3isp/isppreview.c index 9c6dd44..614752a
> 100644 --- a/drivers/media/video/omap3isp/isppreview.c +++
> b/drivers/media/video/omap3isp/isppreview.c @@ -1116,7 +1116,7 @@ static
> void preview_config_input_size(struct isp_prev_device *prev, u32 active)
>  	unsigned int elv = prev->crop.top + prev->crop.height - 1;
>  	u32 features;
> -	if (format->code == V4L2_MBUS_FMT_Y10_1X10) {
> +	if (format->code != V4L2_MBUS_FMT_Y10_1X10) {
>  		sph -= 2;
>  		eph += 2;
>  		slv -= 2;

This patch resolves the issue for me. I get a good picture at both the previewer and the resizer output. Thanks for your help!

--
Best regards,
Florian Neuhaus

Re: omap3isp: cropping bug in previewer?

[Laurent Pinchart]

Hi Florian,

On Friday 06 July 2012 11:55:40 Florian Neuhaus wrote:
> Hi Laurent,
> 
> Laurent Pinchart wrote on 2012-07-05:
> > There's really an issue, which was introduced in v3.5-rc1. Could you
> > please try the following patch instead of yours ?
> > 
> > diff --git a/drivers/media/video/omap3isp/isppreview.c
> > b/drivers/media/video/omap3isp/isppreview.c index 9c6dd44..614752a
> > 100644 --- a/drivers/media/video/omap3isp/isppreview.c +++
> > b/drivers/media/video/omap3isp/isppreview.c @@ -1116,7 +1116,7 @@ static
> > void preview_config_input_size(struct isp_prev_device *prev, u32 active)
> > 
> >  	unsigned int elv = prev->crop.top + prev->crop.height - 1;
> >  	u32 features;
> > 
> > -	if (format->code == V4L2_MBUS_FMT_Y10_1X10) {
> > +	if (format->code != V4L2_MBUS_FMT_Y10_1X10) {
> > 
> >  		sph -= 2;
> >  		eph += 2;
> >  		slv -= 2;
> 
> This patch resolves the issue for me. I get a good picture at both the
> previewer and the resizer output. Thanks for your help!

You're welcome. Can I include your

Tested-by: Florian Neuhaus <florian.neuhaus@reberinformatik.ch>

in the patch ?

-- 
Regards,

Laurent Pinchart

AW: omap3isp: cropping bug in previewer?

[Florian Neuhaus]

Hi Laurent,

Laurent Pinchart wrote on 2012-07-06:

>> This patch resolves the issue for me. I get a good picture at both the
>> previewer and the resizer output. Thanks for your help!
> 
> You're welcome. Can I include your
> 
> Tested-by: Florian Neuhaus <florian.neuhaus@reberinformatik.ch>
> 
> in the patch ?

Yes, of course!

I tested it with the following commands:

For the previewer out:
# media-ctl -v -r -l '"mt9p031 2-0048":0->"OMAP3 ISP CCDC":0[1], "OMAP3 ISP CCDC":2->"OMAP3 ISP preview":0[1], "OMAP3 ISP preview":1->"OMAP3 ISP preview output":0[1]'
# media-ctl -v -f '"mt9p031 2-0048":0 [SGRBG12 800x600], "OMAP3 ISP CCDC":2 [SGRBG10 800x600], "OMAP3 ISP preview":1 [UYVY 800x600]'

# yavta -c1000 -p --stdout --skip 3 -f UYVY -s 782x592 /dev/video4 | mplayer - -demuxer rawvideo -rawvideo w=784:h=592:format=uyvy -vo fbdev -fs

(Note the 784 width in the mplayer command)

For the resizer out:
# media-ctl -v -r -l '"mt9p031 2-0048":0->"OMAP3 ISP CCDC":0[1], "OMAP3 ISP CCDC":2->"OMAP3 ISP preview":0[1], "OMAP3 ISP preview":1->"OMAP3 ISP resizer":0[1], "OMAP3 ISP resizer":1->"OMAP3 ISP resizer output": 0[1]' 
# media-ctl -v -f '"mt9p031 2-0048":0 [SGRBG12 800x600], "OMAP3 ISP CCDC":2 [SGRBG10 800x600], "OMAP3 ISP preview":1 [UYVY 800x600], "OMAP3 ISP resizer":1 [UYVY 800x600]' 
# yavta -f UYVY -s 800x600 -n 8 --skip 3 --capture=1000 --stdout /dev/video6 | mplayer - -demuxer rawvideo -rawvideo w=800:h=600:format=uyvy -vo fbdev

--
Best regards,
Florian Neuhaus

Re: omap3isp: cropping bug in previewer?

[Laurent Pinchart]

Hi Florian,

On Thursday 05 July 2012 16:06:03 Florian Neuhaus wrote:
> Laurent Pinchart wrote on 2012-07-05:
> >> When I now capture a frame with yavta (see [3] for details), I must use
> >> 846x639 as frame size (as this size is reported by the driver). But it
> >> seems that the outputted image is 2px wider (that means 848x639). This
> >> results in a "scrambled"/unusable image on screen when streaming (see
> >> [6] bad-frame-846x639_on_display.bmp for an example how it looks like
> >> on screen). Also the file size too big for a 846x639 image: The frame
> >> size is 1083744 bytes, which is exactly 848*639*2 (NOT 846*639*2)!
> > 
> > The OMAP3 ISP pads lines to multiples of 32 or 64 bytes when reading
> > from/writing to memory. 846 pixels * 2 bytes per pixel is not a multiple
> > of 32 bytes, so the line length gets padded to the next multiple, 848
> > pixels in this case. The information is reported by the bytesperline field
> > of the v4l2_pix_format structure returned by VIDIOC_G_FMT and VIDIOC_S_FMT
> > on the preview engine output video node. You need to take the padding into
> > account in your application, that should solve your issue. raw2rgbpnm
> > tries to detect padding at the end of lines, and skips it automatically.
> 
> Thanks for your fast answer and the explanation!
> So you're saying that yavta doesn't check that the image is coming from the
> previewer-output and has maybe a padding? So yavta needs a patch to extend
> the line width when not aligned on 32 bytes or strip out the padding?

yavta takes the bytesperline field into account, and writes full frames to
disk, including padding. We could patch it to (optionally) discard the padding
bytes, although that kind of feature might belong to higher-level component
(such as raw2rgbpnm or custom applications).

> >> If you look in the isp-datasheet [7] in table 6-40 (page
> >> 1201) you see, that the CFA interpolation block for bayer-mode crops 4
> >> px per line and 4 lines. So shouldn't we respect this in the
> >> preview_config_input_size function? My RFC is:
> >> 
> >> Index: git/drivers/media/video/omap3isp/isppreview.c
> >> 
> >> =========================================================
> >> --- git.orig/drivers/media/video/omap3isp/isppreview.c	2012-07-05
> >> 10:59:33.675358396 +0200 +++
> >> git/drivers/media/video/omap3isp/isppreview.c	2012-07-05
> >> 12:14:33.723223514 +0200 @@ -1140,6 +1140,12 @@
> >> 
> >>  	}
> >>  	if (features & (OMAP3ISP_PREV_CHROMA_SUPP |
> >>  OMAP3ISP_PREV_LUMAENH))
> >>  		sph -= 2;
> >> 
> >> +	if (features & OMAP3ISP_PREV_CFA) {
> >> +		sph -= 2;
> >> +		eph += 2;
> >> +		slv -= 2;
> >> +		elv += 2;
> >> +	}
> >> 
> >>  	isp_reg_writel(isp, (sph << ISPPRV_HORZ_INFO_SPH_SHIFT) | eph,
> >>  		       OMAP3_ISP_IOMEM_PREV, ISPPRV_HORZ_INFO);
> >> 
> >> =========================================================
> >> NOTE: This still gives an unusable picture at the previewer output BUT if
> >> I extend the pipeline to the resizer output, the picture is good. So I
> >> must be missing something...
> 
> After reading your explanation about the padding, I understand why the image
> is broken on the previewer out. But if I configure the pipeline to output
> on the resizer-out, the image is still broken (without my patch). I used a
> resolution of 800x600 for the resizer-out, so the alignment should be fine:
> 
> # media-ctl -v -r -l '"mt9p031 2-0048":0->"OMAP3 ISP CCDC":0[1], "OMAP3 ISP
> CCDC":2->"OMAP3 ISP preview":0[1], "OMAP3 ISP preview":1->"OMAP3 ISP
> resizer":0[1], "OMAP3 ISP resizer":1->"OMAP3 ISP resizer output": 0[1]'
> # media-ctl -v -f '"mt9p031 2-0048":0 [SGRBG12 800x600], "OMAP3 ISP CCDC":2
> [SGRBG10 800x600], "OMAP3 ISP preview":1 [UYVY 800x600], "OMAP3 ISP
> resizer":1 [UYVY 800x600]'
> # yavta -f UYVY -s 800x600 -n 8 --skip 3 --capture=1000 --stdout /dev/video6
> | mplayer - -demuxer rawvideo -rawvideo w=800:h=600:format=uyvy -vo fbdev
> 
> Does my patch just output a good picture by chance, or is there really an
> issue?

There's really an issue, which was introduced in v3.5-rc1. Could you please
try the following patch instead of yours ?

diff --git a/drivers/media/video/omap3isp/isppreview.c b/drivers/media/video/omap3isp/isppreview.c
index 9c6dd44..614752a 100644
--- a/drivers/media/video/omap3isp/isppreview.c
+++ b/drivers/media/video/omap3isp/isppreview.c
@@ -1116,7 +1116,7 @@ static void preview_config_input_size(struct isp_prev_device *prev, u32 active)
 	unsigned int elv = prev->crop.top + prev->crop.height - 1;
 	u32 features;
 
-	if (format->code == V4L2_MBUS_FMT_Y10_1X10) {
+	if (format->code != V4L2_MBUS_FMT_Y10_1X10) {
 		sph -= 2;
 		eph += 2;
 		slv -= 2;

-- 
Regards,

Laurent Pinchart

RE: omap3isp: cropping bug in previewer?

[Florian Neuhaus]

Hi Laurent,

Laurent Pinchart wrote on 2012-07-05:
>> When I now capture a frame with yavta (see [3] for details), I must use
>> 846x639 as frame size (as this size is reported by the driver). But it
>> seems that the outputted image is 2px wider (that means 848x639). This
>> results in a "scrambled"/unusable image on screen when streaming (see
>> [6] bad-frame-846x639_on_display.bmp for an example how it looks like
>> on screen). Also the file size too big for a 846x639 image: The frame
>> size is 1083744 bytes, which is exactly 848*639*2 (NOT 846*639*2)!
> 
> The OMAP3 ISP pads lines to multiples of 32 or 64 bytes when reading
> from/writing to memory. 846 pixels * 2 bytes per pixel is not a multiple of 32
> bytes, so the line length gets padded to the next multiple, 848 pixels in this
> case. The information is reported by the bytesperline field of the
> v4l2_pix_format structure returned by VIDIOC_G_FMT and VIDIOC_S_FMT
> on the
> preview engine output video node. You need to take the padding into
> account in
> your application, that should solve your issue. raw2rgbpnm tries to detect
> padding at the end of lines, and skips it automatically.

Thanks for your fast answer and the explanation!
So you're saying that yavta doesn't check that the image is coming from the previewer-output and has maybe a padding? So yavta needs a patch to extend the line width when not aligned on 32 bytes or strip out the padding?

>> If you look in the isp-datasheet [7] in table 6-40 (page
>> 1201) you see, that the CFA interpolation block for bayer-mode crops 4
>> px per line and 4 lines. So shouldn't we respect this in the
>> preview_config_input_size function? My RFC is:
>> 
>> Index: git/drivers/media/video/omap3isp/isppreview.c
>> 
>> =========================================================
>> --- git.orig/drivers/media/video/omap3isp/isppreview.c	2012-07-05
>> 10:59:33.675358396 +0200 +++
>> git/drivers/media/video/omap3isp/isppreview.c	2012-07-05
>> 12:14:33.723223514 +0200 @@ -1140,6 +1140,12 @@
>>  	} 	if (features & (OMAP3ISP_PREV_CHROMA_SUPP |
>>  OMAP3ISP_PREV_LUMAENH)) 		sph -= 2;
>> +	if (features & OMAP3ISP_PREV_CFA) {
>> +		sph -= 2;
>> +		eph += 2;
>> +		slv -= 2;
>> +		elv += 2;
>> +	}
>> 
>>  	isp_reg_writel(isp, (sph << ISPPRV_HORZ_INFO_SPH_SHIFT) | eph,
>>  		       OMAP3_ISP_IOMEM_PREV, ISPPRV_HORZ_INFO);
>> =========================================================
>> NOTE: This still gives an unusable picture at the previewer output BUT if I
>> extend the pipeline to the resizer output, the picture is good. So I must
>> be missing something...

After reading your explanation about the padding, I understand why the image is broken on the previewer out. But if I configure the pipeline to output on the resizer-out, the image is still broken (without my patch). I used a resolution of 800x600 for the resizer-out, so the alignment should be fine:

# media-ctl -v -r -l '"mt9p031 2-0048":0->"OMAP3 ISP CCDC":0[1], "OMAP3 ISP CCDC":2->"OMAP3 ISP preview":0[1], "OMAP3 ISP preview":1->"OMAP3 ISP resizer":0[1], "OMAP3 ISP resizer":1->"OMAP3 ISP resizer output": 0[1]' 
# media-ctl -v -f '"mt9p031 2-0048":0 [SGRBG12 800x600], "OMAP3 ISP CCDC":2 [SGRBG10 800x600], "OMAP3 ISP preview":1 [UYVY 800x600], "OMAP3 ISP resizer":1 [UYVY 800x600]' 
# yavta -f UYVY -s 800x600 -n 8 --skip 3 --capture=1000 --stdout /dev/video6 | mplayer - -demuxer rawvideo -rawvideo w=800:h=600:format=uyvy -vo fbdev

Does my patch just output a good picture by chance, or is there really an issue?

-- 
Best regards,
Florian Neuhaus

Re: omap3isp: cropping bug in previewer?

[Laurent Pinchart]

Hi Florian,

On Thursday 05 July 2012 12:28:04 Florian Neuhaus wrote:
> Dear all,
> I am trying to get a mt9p031 sensor running on a beagleboard-xm with the
> following configuration:
> 
> Hardware:
> - beagleboard-xm, rev c1
> - Leopard Imaging cam module LI-5M03 with a mt9p031 5MP sensor
> 
> Software:
> - Angstrom-distro built with bitbake using the setup-scripts from [4]
> (commit da56a56b690bcc07a50716f1071e90e2b3a4fb47). - own bitbake recipe to
> build a linux-omap kernel 3.5-rc1 from the tmdlind branch (this source:
> [5], tag omap-fixes-for-v3.5-rc1) - some patches to update the 3.5-rc1
> omap3-isp module (not the whole kernel) to the latest omap3isp-sensors-next
> branch from Laurent Pinchart [1] - yavta with an extension to output data
> to stdout
> - mediactl to configure the omap3isp pipeline
> 
> Problem:
> 
> I configure a pipe with mediactl from OMAP3 ISP CCDC input to the previewer
> output (see [3] for a detailed log) with an example resolution of 800x600.
> This resolution is adapted by the omap3isp driver to 846x639 at the
> previewer output. In my understanding the adjustment of the resolution
> (from 800x600 to 846x639) is a result of the following process: 1) The
> closest possible windowing of the mt9p031 sensor is 864x648. 2) The
> ccdc-source pad crops the height by one line (see function ccdc_try_format
> in ispccdc.c) - we are now on 864x647 3) The previewer (isppreview.c) crops
> a left margin of 8px and a right margin of 6px (see the PREV_MARGIN_*
> defines) plus 4px if the input is from ccdc (see preview_try_crop) - we are
> now on 846x639. As there are no filters activated, the input size will not
> be modified by the preview_config_input_size function.
> 
> When I now capture a frame with yavta (see [3] for details), I must use
> 846x639 as frame size (as this size is reported by the driver). But it
> seems that the outputted image is 2px wider (that means 848x639). This
> results in a "scrambled"/unusable image on screen when streaming (see [6]
> bad-frame-846x639_on_display.bmp for an example how it looks like on
> screen). Also the file size too big for a 846x639 image: The frame size is
> 1083744 bytes, which is exactly 848*639*2 (NOT 846*639*2)!

The OMAP3 ISP pads lines to multiples of 32 or 64 bytes when reading 
from/writing to memory. 846 pixels * 2 bytes per pixel is not a multiple of 32 
bytes, so the line length gets padded to the next multiple, 848 pixels in this 
case. The information is reported by the bytesperline field of the 
v4l2_pix_format structure returned by VIDIOC_G_FMT and VIDIOC_S_FMT on the 
preview engine output video node. You need to take the padding into account in 
your application, that should solve your issue. raw2rgbpnm tries to detect 
padding at the end of lines, and skips it automatically.

> Then I transformed the "bad" yuv-picture with raw2rgbpnm which gives me a
> good picture with both frame-sizes (see bad-frame-846x639.pnm and
> bad-frame-848x639_on_display.bmp in [6]). So the picture-information seems
> to be good, but I guess that the input-size is configured badly by the
> driver. If you look in the isp-datasheet [7] in table 6-40 (page 1201) you
> see, that the CFA interpolation block for bayer-mode crops 4 px per line
> and 4 lines. So shouldn't we respect this in the preview_config_input_size
> function? My RFC is:
> 
> Index: git/drivers/media/video/omap3isp/isppreview.c
> ===================================================================
> --- git.orig/drivers/media/video/omap3isp/isppreview.c	2012-07-05
> 10:59:33.675358396 +0200 +++
> git/drivers/media/video/omap3isp/isppreview.c	2012-07-05 
12:14:33.723223514
> +0200 @@ -1140,6 +1140,12 @@
>  	}
>  	if (features & (OMAP3ISP_PREV_CHROMA_SUPP | OMAP3ISP_PREV_LUMAENH))
>  		sph -= 2;
> +	if (features & OMAP3ISP_PREV_CFA) {
> +		sph -= 2;
> +		eph += 2;
> +		slv -= 2;
> +		elv += 2;
> +	}
> 
>  	isp_reg_writel(isp, (sph << ISPPRV_HORZ_INFO_SPH_SHIFT) | eph,
>  		       OMAP3_ISP_IOMEM_PREV, ISPPRV_HORZ_INFO);
> ===================================================================
> NOTE: This still gives an unusable picture at the previewer output BUT if I
> extend the pipeline to the resizer output, the picture is good. So I must
> be missing something...
> 
> It would be nice, if someone could tell me, if my assumptions are right and
> point me the right direction.
> 
> Further information:
> - Bootup dmesg: [2]
> - Configuration of the pipe with mediactl, capturing of an image with yavta
> and analyze of the image with raw2rgbpnm: [3]
> 
> [1]
> http://git.linuxtv.org/pinchartl/media.git/commit/019214973ee4f03c8f2d58246
> 8b914fcf3385089 [2] http://pastebin.com/7PQnzcmx
> [3] http://pastebin.com/ChEaYHMy
> [4] https://github.com/Angstrom-distribution/setup-scripts
> [5] git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap.git
> [6] https://www.dropbox.com/sh/p2fy5u4i71c3vy8/Fyya25YqK-
> [7] http://www.ti.com/lit/ug/sprugn4q/sprugn4q.pdf

-- 
Regards,

Laurent Pinchart

omap3isp: cropping bug in previewer?

[Florian Neuhaus]

Dear all,
I am trying to get a mt9p031 sensor running on a beagleboard-xm with the following configuration:

Hardware:
- beagleboard-xm, rev c1
- Leopard Imaging cam module LI-5M03 with a mt9p031 5MP sensor

Software:
- Angstrom-distro built with bitbake using the setup-scripts from [4] (commit da56a56b690bcc07a50716f1071e90e2b3a4fb47).
- own bitbake recipe to build a linux-omap kernel 3.5-rc1 from the tmdlind branch (this source: [5], tag omap-fixes-for-v3.5-rc1)
- some patches to update the 3.5-rc1 omap3-isp module (not the whole kernel) to the latest omap3isp-sensors-next branch from Laurent Pinchart [1]
- yavta with an extension to output data to stdout
- mediactl to configure the omap3isp pipeline

Problem:

I configure a pipe with mediactl from OMAP3 ISP CCDC input to the previewer output (see [3] for a detailed log) with an example resolution of 800x600. This resolution is adapted by the omap3isp driver to 846x639 at the previewer output. In my understanding the adjustment of the resolution (from 800x600 to 846x639) is a result of the following process:
1) The closest possible windowing of the mt9p031 sensor is 864x648.
2) The ccdc-source pad crops the height by one line (see function ccdc_try_format in ispccdc.c) - we are now on 864x647
3) The previewer (isppreview.c) crops a left margin of 8px and a right margin of 6px (see the PREV_MARGIN_* defines) plus 4px if the input is from ccdc (see preview_try_crop) - we are now on 846x639.
As there are no filters activated, the input size will not be modified by the preview_config_input_size function.

When I now capture a frame with yavta (see [3] for details), I must use 846x639 as frame size (as this size is reported by the driver). But it seems that the outputted image is 2px wider (that means 848x639). This results in a "scrambled"/unusable image on screen when streaming (see [6] bad-frame-846x639_on_display.bmp for an example how it looks like on screen). Also the file size too big for a 846x639 image: 
The frame size is 1083744 bytes, which is exactly 848*639*2 (NOT 846*639*2)!

Then I transformed the "bad" yuv-picture with raw2rgbpnm which gives me a good picture with both frame-sizes (see bad-frame-846x639.pnm and bad-frame-848x639_on_display.bmp in [6]).
So the picture-information seems to be good, but I guess that the input-size is configured badly by the driver.
If you look in the isp-datasheet [7] in table 6-40 (page 1201) you see, that the CFA interpolation block for bayer-mode crops 4 px per line and 4 lines.
So shouldn't we respect this in the preview_config_input_size function?
My RFC is:

Index: git/drivers/media/video/omap3isp/isppreview.c
===================================================================
--- git.orig/drivers/media/video/omap3isp/isppreview.c	2012-07-05 10:59:33.675358396 +0200
+++ git/drivers/media/video/omap3isp/isppreview.c	2012-07-05 12:14:33.723223514 +0200
@@ -1140,6 +1140,12 @@
 	}
 	if (features & (OMAP3ISP_PREV_CHROMA_SUPP | OMAP3ISP_PREV_LUMAENH))
 		sph -= 2;
+	if (features & OMAP3ISP_PREV_CFA) {
+		sph -= 2;
+		eph += 2;
+		slv -= 2;
+		elv += 2;
+	}
 
 	isp_reg_writel(isp, (sph << ISPPRV_HORZ_INFO_SPH_SHIFT) | eph,
 		       OMAP3_ISP_IOMEM_PREV, ISPPRV_HORZ_INFO);
===================================================================
NOTE: This still gives an unusable picture at the previewer output BUT if I extend the pipeline to the resizer output, the picture is good. So I must be missing something... 

It would be nice, if someone could tell me, if my assumptions are right and point me the right direction. 

Further information:
- Bootup dmesg: [2]
- Configuration of the pipe with mediactl, capturing of an image with yavta and analyze of the image with raw2rgbpnm: [3]

[1] http://git.linuxtv.org/pinchartl/media.git/commit/019214973ee4f03c8f2d582468b914fcf3385089
[2] http://pastebin.com/7PQnzcmx
[3] http://pastebin.com/ChEaYHMy
[4] https://github.com/Angstrom-distribution/setup-scripts
[5] git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap.git
[6] https://www.dropbox.com/sh/p2fy5u4i71c3vy8/Fyya25YqK-
[7] http://www.ti.com/lit/ug/sprugn4q/sprugn4q.pdf

Greetings,
Florian Neuhaus