bayer_8.* shaders are now used for all unpacked sensor data formats,
regardless of the pixel bit width. Let's rename the "8-bit" shaders to
avoid confusion.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
The GPU processing supports 8-bit sensor formats and 10/12-bit packed
formats. Support for 10/12-bit unpacked formats is missing, let's add
it.
10/12-bit unpacked formats use two adjacent bytes to store the value.
This means the 8-bit shaders can be used if we can modify them for
additional support of 16-bit addressing. This requires the following
modifications:
- Using GL_RG (two bytes per pixel) instead of GL_LUMINANCE (one byte
per pixel) as the texture format for the given input formats.
- Setting the texture width to the number of pixels rather than the
number of bytes.
- Making the definition of `fetch' macro variable, according to the
pixel format.
- Using only `fetch' for accessing the texture.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
When accessing a texture position in a shader, the pixel with the
nearest centre to the specified texture coordinates (as mandated by
specifying GL_NEAREST parameter) is taken. The current vertex shader
determines the positions of the neighbouring pixels by adding the
provided texture steps to the exact centre pixel coordinates. But this
places the computed coordinates, from the point of view of GL_NEAREST,
exactly between the pixels and is thus prone to floating point
inaccuracies. Wrong neighbouring pixel coordinates may be used,
resulting in artefacts in the output image.
Let's fix the problem by shifting the initial coordinates a bit from the
pixel border.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
We can easily facilitate swapping R and B on output. Pivot on an
environment define for this purpose.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Extend out the bayer fragment shaders to take 3 x 256 byte inputs as
textures from the CPU.
We then use an index to the table to recover the colour-corrected values
provided by the SoftIPA thread.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Introduce an eGL base helper class which provides an eGL context based on a
passed width and height.
The initGLContext function could be overloaded to provide an interface to a
real display.
A set of helper functions is provided to compile and link GLSL shaders.
linkShaderProgram currently compiles vertex/fragment pairs but could be
overloaded or passed a parameter to link a compute shader instead.
Breaking the eGL interface away from debayering - allows to use the eGL
context inside of a dma-buf heap cleanly, reuse that context inside of a
debayer layer and conceivably reuse the context in a multi-stage shader
pass.
Small note the image_attrs[] array doesn't pass checkstyle.py however the
elements of the array are in pairs.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
A helper class to interact with GBM. This will allow us to specify the
internal storage format of the CPU when making a texture for the Debayer
vertext/fragment shaders and thus ensure we receive an uncompressed and
untiled output buffer.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Encode the bayer shader files into a header as part of the build process.
Qcam already compiles the shader files down into a QT resource file which
it references internally.
In order to share the debayering shader programs outside of qcam create a
generic header which both qcam and libcamera can operate from.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Moving the GL shaders to src/libcamera/assets/shader to allow for reuse of
these inside of the SoftISP.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Add an element to the DebayerParams structure to capture the calculated
CCM. This allows us to pass the CCM into the eGL Debayer shader later on.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Make the type of object Debayer not DebayerCpu thus allowing us to assign
the object to either DebayerCpu or DebayerEGL.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Add MappedFrameBuffer::getPlaneFD() which takes a plane index and returns
the file descriptor associated with it.
This fd will be used to feed into eglCreateImageKHR for both texture
creation on upload and directly render-to-texture where the texture buffer
comes from the fd given to eglCreateImageKHR.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Take a pointer to the given framebuffer as a private member for further use
in later patches.
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Add a method to the SwstatsCpu class to process a whole Framebuffer in
one go, rather then line by line. This is useful for gathering stats
when debayering is not necessary or is not done on the CPU.
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Move the code for the builtin benchmark to its own small
Benchmark class.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Move the swstats_cpu.h file to include/libcamera/internal/software_isp/
so that it can be used outside the src/libcamera/software_isp/ directory.
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
`appendPOD()` does a single insertion, so if only a single `appendPOD()`
will be called on a vector before returning, then calling `reserve()`
is not that useful, so remove it.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Currently, modifying `controls.py` does not make those build targets dirty
that use a script that includes it (e.g. `gen-controls.py`) because meson
has no knowledge of this dependency. Add `depend_files` to each
`custom_target()` invocation to fix this.
Ideally it would be possible to attach this dependency to `gen_controls`,
`gen_gst_controls`, etc. objects themselves, so that repetition is
avoided, but this does not seem possible at the moment.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Acked-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
It is useful to multiply matrices and vectors of heterogeneous types, for
instance float and double. Extend the multiplication operator to support
this, avoiding the need to convert one of the operations. The type of the
returned vector is selected automatically to avoid loosing precision.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
It is useful to multiply matrices of heterogneous types, for instance
float and double. Extend the multiplication operator to support this,
avoiding the need to convert one of the matrices. The type of the
returned matrix is selected automatically to avoid loosing precision.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
For calculations in upcoming algorithm patches, the inverse of a matrix
is required. Add an implementation of the inverse() function for square
matrices.
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
When one wants to create a Vector from existing data, currently the only
way is via std::array. Add a Span based constructor to allow creation
from std::vectors and alike.
While at it, replace the manual loop with std::copy.
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
When one wants to create a Matrix from existing data, currently the only
way is via std::array. Add a Span based constructor to allow creation
from vectors and alike.
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
By zero-initializing the data_ member we can make most functions
constexpr which will come in handy in upcoming patches. Note that this
is due to C++17. In C++20 we will be able to leave data_ uninitialized
for constexpr. The Matrix(std::array) version of the constructor can
not be constexpr because std::copy only became constexpr in C++20.
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
SFINAE is difficult to read and not needed in these cases. Replace it
with static_asserts. The idea came from [1] where it is stated:
"The use of enable_if seems misguided to me. SFINAE is useful for the
situation where we consider multiple candidates for something (overloads
or class template specializations) and try to choose the correct one,
without causing compilation to fail."
[1]: https://stackoverflow.com/questions/62109526/c-friend-template-that-use-sfinae
Signed-off-by: Stefan Klug <stefan.klug@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Provide a MediaPipeline class to help identifing and managing pipelines across
a MediaDevice graph.
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Stefan Klug <stefan.klug@ideasonboard.com>
Provide a helper on the MediaDevice to return a list of all
available entities which match a given function in the graph.
As a drive by, also fix a whitespace error in the documentation of
MediaDevice::setupLink.
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Stefan Klug <stefan.klug@ideasonboard.com>
Arrays of arrays, even arrays of strings, are not supported by
the current `ControlValue` mechanism, so disable them for now
to trigger compile time errors if attempts are made to use them.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The `signature()` getter can just return a reference to the private vector
member variable, and let the caller make a copy if needed. Since the
return type is const qualified, this was likely the original intention.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Add helper to check if frame start event are supported by subdevice.
Since kernel does not have interface to query supported events
use subscribe interface.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com> # v3
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> # v5
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Use `if constexpr` instead of SFINAE to handle return values of type `void`.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Use `std::forward()` to forward the received arguments to enable the
potential use of move constructors instead of copy constructors.
Commit 0eacde623b ("libcamera: object: Avoid argument copies in invokeMethod()")
added the forwarding references to `invokeMethod()` but it did not add the
appropriate `std::forward()` calls, so copying could still take place
even if not necessary.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Do not force the caller to have an `std::string` object as a
simple string view is sufficient to do the lookup.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Report exposure and gain in metadata.
This is more complicated than it could be expected because the exposure
value should be in microseconds but it's handled using V4L2_CID_EXPOSURE
control, which doesn't specify the unit, see
https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/control.html.
So the unit conversion is done in the way rkisp1 IPA uses.
This requires getting and passing IPACameraSensorInfo around. To avoid
naming confusion and to improve consistency with rkisp1 IPA,
sensorCtrlInfoMap parameter is renamed to sensorControls.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Extend the Simple IPA IPC to support returning a metadata ControlList
when the process call has completed.
A new signal from the IPA is introduced to report the metadata,
similarly to what the hardware pipelines do.
Merge the metadata reported by the ISP into any completing request to
provide to the application. Completion of a request is delayed until
this is done; this doesn't apply to canceled requests.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
This patch applies color correction matrix (CCM) in debayering if the
CCM is specified. Not using CCM must still be supported for performance
reasons.
The CCM is applied as follows:
[r1 g1 b1] [r]
[r2 g2 b2] * [g]
[r3 g3 b3] [b]
The CCM matrix (the left side of the multiplication) is constant during
single frame processing, while the input pixel (the right side) changes.
Because each of the color channels is only 8-bit in software ISP, we can
make 9 lookup tables with 256 input values for multiplications of each
of the r_i, g_i, b_i values. This way we don't have to multiply each
pixel, we can use table lookups and additions instead. Gamma (which is
non-linear and thus cannot be a part of the 9 lookup tables values) is
applied on the final values rounded to integers using another lookup
table.
Because the changing part is the pixel value with three color elements,
only three dynamic table lookups are needed. We use three lookup tables
to represent the multiplied matrix values, each of the tables
corresponding to the given matrix column and pixel color.
We use int16_t to store the precomputed multiplications. This seems to
be noticeably (>10%) faster than `float' for the price of slightly less
accuracy and it covers the range of values that sane CCMs produce. The
selection and structure of data is performance critical, for example
using bytes would add significant (>10%) speedup but would be too short
to cover the value range.
The color lookup tables can be represented either as unions,
accommodating tables for both the CCM and non-CCM cases, or as separate
tables for each of the cases, leaving the tables for the other case
unused. The latter is selected as a matter of preference.
The tables are copied (as before), which is not elegant but also not a
big problem. There are patches posted that use shared buffers for
parameters passing in software ISP (see software ISP TODO #5) and they
can be adjusted for the new parameter format.
Color gains from white balance are supposed not to be a part of the
specified CCM. They are applied on it using matrix multiplication,
which is simple and in correspondence with future additions in the form
of matrix multiplication, like saturation adjustment.
With this patch, the reported per-frame slowdown when applying CCM is
about 45% on Debix Model A and about 75% on TI AM69 SK.
Using std::clamp in debayering adds some performance penalty (a few
percent). The clamping is necessary to eliminate out of range values
possibly produced by the CCM. If it could be avoided by adjusting the
precomputed tables some way then performance could be improved a bit.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Applying color correction matrix (CCM) in software ISP is optional due
to performance reasons. CCM is applied if and only if `Ccm' algorithm
is present in the tuning file.
Software ISP debayering is a performance critical piece of code and we
do not want to use dynamic conditionals there. Therefore we pass
information about CCM application to debayering configuration and let it
select the right versions of debayering functions using templates. This
is a trick similar to the previously used one for adding or not adding
an alpha channel to the output.
Debayering gets this information but it ignores it in this patch.
Actual processing with CCM is added in the followup patch.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Only enums whose sizes match that of `int32_t` can be directly
supported. Otherwise the expected size and the real size would
disagree, leading to an assertion failure in `ControlValue::set()`.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The `dynamic_extent` specialization is currently not trivially copyable
unlike its standard counterpart, `std::span`. This is because the copy
assignment operator is user-defined. Explicitly default it just like
it is done in the main template definition.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The `V4L2BufferCache` type is not thread-safe. Its `lastUsedCounter_`
member is not used in contexts where its atomicity would matter.
So it does not need to be have an atomic type.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Express the ownership more clearly by using a smart pointer type.
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
`MediaEntity::ancillaryEntities()` can just return a const lvalue
reference to the underlying array, a copy need not be made. That
was likely the original intention.
Fixes: 9490c664b5 ("libcamera: Add members to MediaEntity to support ancillary entities")
Signed-off-by: Barnabás Pőcze <barnabas.pocze@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add the Raspberry Pi 5 ISP (PiSP) pipeline handler to libcamera. To
include this pipeline handler in the build, set the following meson
option:
meson configure -Dpipelines=rpi/pisp
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The libcamera public API requires C++17, drop support for older
standards.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
There may be pending messages in SoftwareIsp message queue when
SoftwareIsp stops. The call to IPAProxySoft::stop() will dispatch them
before SoftwareIsp::stop() finishes. But this is dependent on
IPAProxySoft::stop() implementation, let's break this dependency and
dispatch messages to SoftwareIsp explicitly in SoftwareIsp::stop().
This also allows dropping `running_' flag. Since the SoftwareIsp
messages get processed and invoke IPA calls before the IPA proxy is set
to ProxyStopping state and the SoftwareIsp worker thread is no longer
running, it's guaranteed that no new messages come to SoftwareIsp and
attempt to call the stopped IPA proxy.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The Thread::dispatchMessage() function supports filtering messages based
on their type. It can be useful to also dispatch only messages posted
for a specific receiver. Add an optional receiver argument to the
dispatchMessage() function to do so. When set to null (the default
value), the behaviour of the function is not changed.
This facility is actually used in followup patches.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
When SoftwareIsp stops, input and output buffers queued to it may not
yet be fully processed. They will be eventually returned but stop means
stop, there should be no processing related actions invoked afterwards.
Let's stop forwarding processed input buffers from SoftwareIsp slots
when SoftwareIsp is stopped. Let's track the queued input buffers and
return them back for capture in SoftwareIsp::stop().
The returned input buffers are marked as cancelled. This is not
necessary at the moment but it gives the pipeline handlers chance to
deal with this if they need to.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
