From: Hugues FRUCHET <hugues.fruchet@st.com>
To: Hans Verkuil <hverkuil@xs4all.nl>,
Maxime Coquelin <mcoquelin.stm32@gmail.com>,
Alexandre TORGUE <alexandre.torgue@st.com>,
Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: "devicetree@vger.kernel.org" <devicetree@vger.kernel.org>,
"linux-arm-kernel@lists.infradead.org"
<linux-arm-kernel@lists.infradead.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"linux-media@vger.kernel.org" <linux-media@vger.kernel.org>,
Benjamin Gaignard <benjamin.gaignard@linaro.org>,
Yannick FERTRE <yannick.fertre@st.com>
Subject: Re: [PATCH v1 3/5] [media] stm32-dcmi: crop sensor image to match user resolution
Date: Mon, 26 Jun 2017 14:41:58 +0000 [thread overview]
Message-ID: <3f04ca80-28cd-c793-72da-e0ea8efdb8a9@st.com> (raw)
In-Reply-To: <2844ed1a-d310-4381-c4a7-88ab4633f32f@xs4all.nl>
On 06/26/2017 12:07 PM, Hans Verkuil wrote:
> On 26/06/17 11:53, Hugues FRUCHET wrote:
>> Hi Hans, thanks for review.
>>
>> Reply below.
>>
>> BR
>> Hugues.
>>
>> On 06/22/2017 05:19 PM, Hans Verkuil wrote:
>>> On 06/22/2017 05:12 PM, Hugues Fruchet wrote:
>>>> Add flexibility on supported resolutions by cropping sensor
>>>> image to fit user resolution format request.
>>>>
>>>> Signed-off-by: Hugues Fruchet <hugues.fruchet@st.com>
>>>> ---
>>>> drivers/media/platform/stm32/stm32-dcmi.c | 54 ++++++++++++++++++++++++++++++-
>>>> 1 file changed, 53 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/media/platform/stm32/stm32-dcmi.c b/drivers/media/platform/stm32/stm32-dcmi.c
>>>> index 75d53aa..bc5e052 100644
>>>> --- a/drivers/media/platform/stm32/stm32-dcmi.c
>>>> +++ b/drivers/media/platform/stm32/stm32-dcmi.c
>>>> @@ -131,6 +131,8 @@ struct stm32_dcmi {
>>>> struct v4l2_async_notifier notifier;
>>>> struct dcmi_graph_entity entity;
>>>> struct v4l2_format fmt;
>>>> + struct v4l2_rect crop;
>>>> + bool do_crop;
>>>>
>>>> const struct dcmi_format **user_formats;
>>>> unsigned int num_user_formats;
>>>> @@ -538,6 +540,27 @@ static int dcmi_start_streaming(struct vb2_queue *vq, unsigned int count)
>>>> if (dcmi->bus.flags & V4L2_MBUS_PCLK_SAMPLE_RISING)
>>>> val |= CR_PCKPOL;
>>>>
>>>> + if (dcmi->do_crop) {
>>>> + u32 size, start;
>>>> +
>>>> + /* Crop resolution */
>>>> + size = ((dcmi->crop.height - 1) << 16) |
>>>> + ((dcmi->crop.width << 1) - 1);
>>>> + reg_write(dcmi->regs, DCMI_CWSIZE, size);
>>>> +
>>>> + /* Crop start point */
>>>> + start = ((dcmi->crop.top) << 16) |
>>>> + ((dcmi->crop.left << 1));
>>>> + reg_write(dcmi->regs, DCMI_CWSTRT, start);
>>>> +
>>>> + dev_dbg(dcmi->dev, "Cropping to %ux%u@%u:%u\n",
>>>> + dcmi->crop.width, dcmi->crop.height,
>>>> + dcmi->crop.left, dcmi->crop.top);
>>>> +
>>>> + /* Enable crop */
>>>> + val |= CR_CROP;
>>>> + };
>>>> +
>>>> reg_write(dcmi->regs, DCMI_CR, val);
>>>>
>>>> /* Enable dcmi */
>>>> @@ -707,6 +730,8 @@ static int dcmi_try_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f,
>>>> .which = V4L2_SUBDEV_FORMAT_TRY,
>>>> };
>>>> int ret;
>>>> + __u32 width, height;
>>>> + struct v4l2_mbus_framefmt *mf = &format.format;
>>>>
>>>> dcmi_fmt = find_format_by_fourcc(dcmi, pixfmt->pixelformat);
>>>> if (!dcmi_fmt) {
>>>> @@ -724,8 +749,18 @@ static int dcmi_try_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f,
>>>> if (ret < 0)
>>>> return ret;
>>>>
>>>> + /* Align format on what sensor can do */
>>>> + width = pixfmt->width;
>>>> + height = pixfmt->height;
>>>> v4l2_fill_pix_format(pixfmt, &format.format);
>>>>
>>>> + /* We can do any resolution thanks to crop */
>>>> + if ((mf->width > width) || (mf->height > height)) {
>>>> + /* Restore width/height */
>>>> + pixfmt->width = width;
>>>> + pixfmt->height = height;
>>>> + };
>>>> +
>>>> pixfmt->field = V4L2_FIELD_NONE;
>>>> pixfmt->bytesperline = pixfmt->width * dcmi_fmt->bpp;
>>>> pixfmt->sizeimage = pixfmt->bytesperline * pixfmt->height;
>>>> @@ -741,6 +776,8 @@ static int dcmi_set_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f)
>>>> struct v4l2_subdev_format format = {
>>>> .which = V4L2_SUBDEV_FORMAT_ACTIVE,
>>>> };
>>>> + struct v4l2_mbus_framefmt *mf = &format.format;
>>>> + struct v4l2_pix_format *pixfmt = &f->fmt.pix;
>>>> const struct dcmi_format *current_fmt;
>>>> int ret;
>>>>
>>>> @@ -748,13 +785,28 @@ static int dcmi_set_fmt(struct stm32_dcmi *dcmi, struct v4l2_format *f)
>>>> if (ret)
>>>> return ret;
>>>>
>>>> - v4l2_fill_mbus_format(&format.format, &f->fmt.pix,
>>>> + v4l2_fill_mbus_format(&format.format, pixfmt,
>>>> current_fmt->mbus_code);
>>>> ret = v4l2_subdev_call(dcmi->entity.subdev, pad,
>>>> set_fmt, NULL, &format);
>>>> if (ret < 0)
>>>> return ret;
>>>>
>>>> + /* Enable crop if sensor resolution is larger than request */
>>>> + dcmi->do_crop = false;
>>>> + if ((mf->width > pixfmt->width) || (mf->height > pixfmt->height)) {
>>>> + dcmi->crop.width = pixfmt->width;
>>>> + dcmi->crop.height = pixfmt->height;
>>>> + dcmi->crop.left = (mf->width - pixfmt->width) / 2;
>>>> + dcmi->crop.top = (mf->height - pixfmt->height) / 2;
>>>> + dcmi->do_crop = true;
>>>
>>> Why not implement the selection API instead? I assume that you can crop from any
>>> region of the sensor, not just the center part.
>>
>> The point here was to add some flexibility for user in term of
>> resolution and also less memory consumption.
>> For example here I want to make a 480x272 preview:
>> - without this change: S_FMT(480x272) returns VGA (the OV9655 larger
>> discrete resolution), then app has to capture VGA frames then crop to
>> fit 480x272 frame buffer.
>> - with this change: S_FMT(480x272) returns 480x272 (crop done by
>> hardware), app can directly capture 480x272 then copy to framebuffer
>> without any conversion.
>>
>> Implementation of V4L2 crop using SELECTION API could also be used,
>> but I need to change app.
>>
>> More generally, with a given couple ISP+sensor, will S_FMT()
>> return the sensor only supported resolutions ? or the supported
>> resolutions of the couple ISP+sensor (ISP will downscale/upscale/crop
>> the sensor discrete resolution to fit user request) ?
>>
>> Hans, what are your recommendations ?
>
> This is not the way the V4L2 API works.
>
> Sensors report their supported resolutions through the ENUM_FRAMESIZES
> (and the related ENUM_FRAMEINTERVALS) ioctls.
>
> With S_FMT you pick the resolution you want.
>
> If you then want to crop the driver should implement the selection API
> (this will also automatically enable the G/S_CROP/CROPCAP ioctls) so the
> application can select which part of the image should be cropped. Assuming
> there is no scaler then after cropping the format size will be reduced
> to the size of the cropped image.
>
> That's in a nutshell how these ioctls relate to one another.
>
> Regards,
>
> Hans
>
Ok, crystal clear, I'll implement selection API so.
BR,
Hugues.
>>
>>>
>>> Regards,
>>>
>>> Hans
>>>
>>>> +
>>>> + dev_dbg(dcmi->dev, "%ux%u cropped to %ux%u@(%u,%u)\n",
>>>> + mf->width, mf->height,
>>>> + dcmi->crop.width, dcmi->crop.height,
>>>> + dcmi->crop.left, dcmi->crop.top);
>>>> + };
>>>> +
>>>> dcmi->fmt = *f;
>>>> dcmi->current_fmt = current_fmt;
>>>>
>>>>
>
next prev parent reply other threads:[~2017-06-26 14:42 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-06-22 15:12 [PATCH v1 0/5] Camera support on STM32F746G-DISCO board Hugues Fruchet
2017-06-22 15:12 ` [PATCH v1 1/5] [media] stm32-dcmi: catch dma submission error Hugues Fruchet
2017-06-22 15:12 ` [PATCH v1 2/5] [media] stm32-dcmi: revisit control register handling Hugues Fruchet
2017-06-22 15:12 ` [PATCH v1 3/5] [media] stm32-dcmi: crop sensor image to match user resolution Hugues Fruchet
2017-06-22 15:19 ` Hans Verkuil
2017-06-26 9:53 ` Hugues FRUCHET
2017-06-26 10:07 ` Hans Verkuil
2017-06-26 14:41 ` Hugues FRUCHET [this message]
2017-06-25 15:02 ` [PATCH] stm32-dcmi: fix semicolon.cocci warnings kbuild test robot
2017-06-25 15:02 ` [PATCH v1 3/5] [media] stm32-dcmi: crop sensor image to match user resolution kbuild test robot
2017-06-22 15:12 ` [PATCH v1 4/5] ARM: dts: stm32: Enable DCMI support on STM32F746 MCU Hugues Fruchet
2017-07-02 0:40 ` kbuild test robot
2017-06-22 15:12 ` [PATCH v1 5/5] ARM: configs: stm32: DCMI + OV9655 camera support Hugues Fruchet
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3f04ca80-28cd-c793-72da-e0ea8efdb8a9@st.com \
--to=hugues.fruchet@st.com \
--cc=alexandre.torgue@st.com \
--cc=benjamin.gaignard@linaro.org \
--cc=devicetree@vger.kernel.org \
--cc=hverkuil@xs4all.nl \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=mchehab@kernel.org \
--cc=mcoquelin.stm32@gmail.com \
--cc=yannick.fertre@st.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®