[libcamera-devel] [PATCH 2/2] libcamera: v4l2_device: Add methods to get/set format
Jacopo Mondi
jacopo at jmondi.org
Mon Jan 28 17:07:32 CET 2019
Hi Kieran,
On Mon, Jan 28, 2019 at 03:42:27PM +0000, Kieran Bingham wrote:
> Hi Jacopo,
>
> I only have variable/layout concerns here - which is certainly not a
> blocker at the moment.
>
> Also - if the outcome of the discussion below changes this patches
> expectations - then we'll need to adapt other patches too - so that
> would be a global patch fix at that point.
>
> Thus I don't think that discussion blocks this patch.
>
>
> On 28/01/2019 15:11, Jacopo Mondi wrote:
> > Add methods to set and get the image format programmed on a V4L2Device
> > for both the single and multi planar use case.
> >
> > Signed-off-by: Jacopo Mondi <jacopo at jmondi.org>
>
> Reviewed-by: Kieran Bingham <kieran.bingham at ideasonboard.com>
>
>
> > ---
> > src/libcamera/include/v4l2_device.h | 10 +++
> > src/libcamera/v4l2_device.cpp | 128 ++++++++++++++++++++++++++++
> > 2 files changed, 138 insertions(+)
> >
> > diff --git a/src/libcamera/include/v4l2_device.h b/src/libcamera/include/v4l2_device.h
> > index c70959e..fe54220 100644
> > --- a/src/libcamera/include/v4l2_device.h
> > +++ b/src/libcamera/include/v4l2_device.h
> > @@ -86,10 +86,20 @@ public:
> > const char *deviceName() const { return caps_.card(); }
> > const char *busName() const { return caps_.bus_info(); }
> >
> > + int format(V4L2DeviceFormat *fmt);
> > + int setFormat(V4L2DeviceFormat *fmt);
> > +
> > private:
> > std::string deviceNode_;
> > int fd_;
> > V4L2Capability caps_;
> > + enum v4l2_buf_type bufferType_;
> > +
> > + int getFormatSingleplane(V4L2DeviceFormat *fmt);
> > + int setFormatSingleplane(V4L2DeviceFormat *fmt);
> > +
> > + int getFormatMultiplane(V4L2DeviceFormat *fmt);
> > + int setFormatMultiplane(V4L2DeviceFormat *fmt);
>
>
> Can I confirm <or prompt discussion> on our class layouts?
>
> Have we defined a layout anywhere?
>
> Here this creates:
>
>
> class ClassName
> {
> public:
> ClassName();
> ~ClassName();
> void publicFunctions();
>
> int publicData;
> bool publicBool;
>
> private:
> int privateData;
> bool privateBool;
>
> int privateFunctions();
> int morePrivateFunctions();
> }
>
> Which is:
> PublicFunctions
> PublicData
> PrivateData
> PrivateFunctions
>
>
> So the 'data' both public and private is sandwiched in the middle.
> I don't mind this - but in my mind I thought we were doing:
>
>
> class ClassName
> {
> public:
> ClassName();
> ~ClassName();
> void publicFunctions();
>
> int publicData;
> bool publicBool;
>
> private:
> int privateFunctions();
> int morePrivateFunctions();
>
> int privateData;
> bool privateBool;
> }
>
> Which is:
> PublicFunctions
> PublicData
> PrivateFunctions
> PrivateData
>
> so that the public, and private (or protected) sections follow the same
> sequence?
>
> Any comments from the team here?
Quoting the Google's C++ style guide (which I refer to because we
actually started from there when we defined our coding guidelines):
https://google.github.io/styleguide/cppguide.html#Declaration_Order
... generally prefer the following order: types (including typedef, using,
and nested structs and classes), constants, factory functions, constructors,
assignment operators, destructor, all other methods, data members.
So your
PublicFunctions
PublicData
PrivateFunctions
PrivateData
Is accurate.
I fear we would need a library-wide cleanup, but I can start by making
this right at least.
>
> > };
> >
> > } /* namespace libcamera */
> > diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
> > index d6143f2..5c415d0 100644
> > --- a/src/libcamera/v4l2_device.cpp
> > +++ b/src/libcamera/v4l2_device.cpp
> > @@ -227,6 +227,15 @@ int V4L2Device::open()
> > return -EINVAL;
> > }
> >
> > + if (caps_.isCapture())
> > + bufferType_ = caps_.isMultiplanar()
> > + ? V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE
> > + : V4L2_BUF_TYPE_VIDEO_CAPTURE;
> > + else
> > + bufferType_ = caps_.isMultiplanar()
> > + ? V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE
> > + : V4L2_BUF_TYPE_VIDEO_OUTPUT;
> > +
> > return 0;
> > }
> >
> > @@ -269,4 +278,123 @@ void V4L2Device::close()
> > * \return The string containing the device location
> > */
> >
> > +/**
> > + * \brief Retrieve the image format set on the V4L2 device
> > + * \return 0 for success, a negative error code otherwise
> > + */
> > +int V4L2Device::format(V4L2DeviceFormat *fmt)
> > +{> + return caps_.isMultiplanar() ? getFormatMultiplane(fmt) :
>
> I think the precedence is that the : would be below the ? here ?
>
> But I'm doubting myself - so hopefully Laurent will chime in with his
> preferences here.
>
>
> If so perhaps we should set the following change in .clang-format:
>
> -BreakBeforeTernaryOperators: false
> +BreakBeforeTernaryOperators: true
>
> which would produce the following:
>
>
> + return caps_.isMultiplanar() ? getFormatMultiplane(fmt)
> + : getFormatSingleplane(fmt);
>
>
>
> It doesn't however align the ? with the = in the bufferType_ above:
>
> bufferType_ = caps_.isMultiplanar()
> ? V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE
> : V4L2_BUF_TYPE_VIDEO_OUTPUT;
> ? V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE
> : V4L2_BUF_TYPE_VIDEO_OUTPUT;
>
> So again - I think that's a non-blocker currently.
What makes the style checker and most of the team happy, makes me
happy, so I'm open to all this solutions.
>
>
> > + getFormatSingleplane(fmt);
> > +}
> > +
> > +/**
> > + * \brief Configure an image format on the V4L2 device
> > + * \return 0 for success, a negative error code otherwise
> > + */
> > +int V4L2Device::setFormat(V4L2DeviceFormat *fmt)
> > +{
> > + return caps_.isMultiplanar() ? setFormatMultiplane(fmt) :
> > + setFormatSingleplane(fmt);
> > +}
> > +
> > +int V4L2Device::getFormatSingleplane(V4L2DeviceFormat *fmt)
> > +{
> > + struct v4l2_format v4l2Fmt;
> > + struct v4l2_pix_format *pix = &v4l2Fmt.fmt.pix;
> > + int ret;
> > +
> > + v4l2Fmt.type = bufferType_;
> > + ret = ioctl(fd_, VIDIOC_S_FMT, &v4l2Fmt);
> > + if (ret) {
> > + ret = -errno;
> > + LOG(Error) << "Unable to get format: " << strerror(-ret);
> > + return ret;
> > + }
> > +
> > + fmt->width = pix->width;
> > + fmt->height = pix->height;
> > + fmt->fourcc = pix->pixelformat;
> > + fmt->planes = 1;
> > + fmt->planesFmt[0].bpl = pix->bytesperline;
> > + fmt->planesFmt[0].size = pix->sizeimage;
> > +
>
> I think we might need to cache the V4L2DeviceFormat in the V4L2Device
> somewhere...
>
Yes, indeed. I would have had a V4L2DataFormat member initialized at
device 'open()' time that could be cached, and re-freshed at every
s_fmt..
> But perhaps that's not required by this patch - if something needs to
> access this (like the BufferPool creation) it can handle caching the
> format object.
>
I left it out from this two patches to ease integration, but I agree
this is a possible optimization, and if any class needs that, it shall
be done here.
Thanks
j
>
>
> > + return 0;
> > +}
> > +
> > +int V4L2Device::setFormatSingleplane(V4L2DeviceFormat *fmt)
> > +{
> > + struct v4l2_format v4l2Fmt;
> > + struct v4l2_pix_format *pix = &v4l2Fmt.fmt.pix;
> > + int ret;
> > +
> > + v4l2Fmt.type = bufferType_;
> > + pix->width = fmt->width;
> > + pix->height = fmt->height;
> > + pix->pixelformat = fmt->fourcc;
> > +
> > + ret = ioctl(fd_, VIDIOC_S_FMT, &v4l2Fmt);
> > + if (ret) {
> > + ret = -errno;
> > + LOG(Error) << "Unable to set format: " << strerror(-ret);
> > + return ret;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +int V4L2Device::getFormatMultiplane(V4L2DeviceFormat *fmt)
> > +{
> > + struct v4l2_format v4l2Fmt;
> > + struct v4l2_pix_format_mplane *pix = &v4l2Fmt.fmt.pix_mp;
> > + int ret;
> > +
> > + v4l2Fmt.type = bufferType_;
> > + ret = ioctl(fd_, VIDIOC_G_FMT, &v4l2Fmt);
> > + if (ret) {
> > + ret = -errno;
> > + LOG(Error) << "Unable to get format: " << strerror(-ret);
> > + return ret;
> > + }
> > +
> > + fmt->width = pix->width;
> > + fmt->height = pix->height;
> > + fmt->fourcc = pix->pixelformat;
> > + fmt->planes = pix->num_planes;
> > +
> > + for (unsigned int i = 0; i < fmt->planes; ++i) {
> > + fmt->planesFmt[i].bpl = pix->plane_fmt[i].bytesperline;
> > + fmt->planesFmt[i].size = pix->plane_fmt[i].sizeimage;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +int V4L2Device::setFormatMultiplane(V4L2DeviceFormat *fmt)
> > +{
> > + struct v4l2_format v4l2Fmt;
> > + struct v4l2_pix_format_mplane *pix = &v4l2Fmt.fmt.pix_mp;
> > + int ret;
> > +
> > + v4l2Fmt.type = bufferType_;
> > + pix->width = fmt->width;
> > + pix->height = fmt->height;
> > + pix->pixelformat = fmt->fourcc;
> > + pix->num_planes = fmt->planes;
> > +
> > + for (unsigned int i = 0; i < pix->num_planes; ++i) {
> > + pix->plane_fmt[i].bytesperline = fmt->planesFmt[i].bpl;
> > + pix->plane_fmt[i].sizeimage = fmt->planesFmt[i].size;
> > + }
> > +
> > + ret = ioctl(fd_, VIDIOC_S_FMT, &v4l2Fmt);
> > + if (ret) {
> > + ret = -errno;
> > + LOG(Error) << "Unable to set format: " << strerror(-ret);
> > + return ret;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > } /* namespace libcamera */
> >
>
> --
> Regards
> --
> Kieran
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://lists.libcamera.org/pipermail/libcamera-devel/attachments/20190128/6708e603/attachment.sig>
More information about the libcamera-devel
mailing list