[PATCH v9 7/8] libcamera: pipeline: Load images
Cheng-Hao Yang
chenghaoyang at chromium.org
Thu Aug 29 21:58:04 CEST 2024
Hi Jacopo,
On Wed, Aug 28, 2024 at 10:14 AM Jacopo Mondi <jacopo.mondi at ideasonboard.com>
wrote:
> Hi
>
> On Wed, Aug 21, 2024 at 02:05:57AM GMT, Barnabás Pőcze wrote:
> > Hi
> >
> >
> > 2024. augusztus 20., kedd 18:23 keltezéssel, Harvey Yang <
> chenghaoyang at chromium.org> írta:
> >
> > > From: Konami Shu <konamiz at google.com>
> > >
> > > - Refactor ImageFrameGenerator and TestPatternGenerator
> > > - Extend the config file to have section to choose to use images or
> test
> > > pattern
> > > - Extend Parser to parse the configuration
> > > - Add ImageFrameGenerator which uses images to provide frames
>
> Please make a proper commit message out of this
>
Updated.
>
> > >
> > > Patchset1->2
> > > - Extend the parser to accept files with ".jpeg"
> > > - Heap allocate some buffers in ImageFrameGenerator::generateFrame so
> > > the buffers won't cause stack over flow
> > >
> > > Patchset5->6
> > > - Move static factory function from the interface class to the derived
> > > classes
> > >
> > > Patchset6->8
> > > - create FrameGenerator before starting the camera
>
> Changelogs should go below
>
Let's put them in the cover letter, if it's even needed.
>
> ---
>
> So that they don't appear in git history
>
> > >
> > > Signed-off-by: Konami Shu <konamiz at google.com>
> > > Co-developed-by: Harvey Yang <chenghaoyang at chromium.org>
> > > Co-developed-by: Yunke Cao <yunkec at chromium.org>
> > > Co-developed-by: Tomasz Figa <tfiga at chromium.org>
> > > ---
> > > src/libcamera/pipeline/virtual/README.md | 18 +-
> > > .../pipeline/virtual/common_functions.cpp | 27 +++
> > > .../pipeline/virtual/common_functions.h | 18 ++
> > > .../pipeline/virtual/frame_generator.h | 2 +-
> > > .../virtual/image_frame_generator.cpp | 154 ++++++++++++++++++
> > > .../pipeline/virtual/image_frame_generator.h | 65 ++++++++
> > > src/libcamera/pipeline/virtual/meson.build | 6 +-
> > > src/libcamera/pipeline/virtual/parser.cpp | 77 +++++++--
> > > src/libcamera/pipeline/virtual/parser.h | 5 +-
> > > .../virtual/test_pattern_generator.cpp | 4 +-
> > > .../pipeline/virtual/test_pattern_generator.h | 2 +-
> > > src/libcamera/pipeline/virtual/virtual.cpp | 37 +++--
> > > src/libcamera/pipeline/virtual/virtual.h | 20 ++-
> > > 13 files changed, 390 insertions(+), 45 deletions(-)
> > > create mode 100644 src/libcamera/pipeline/virtual/common_functions.cpp
> > > create mode 100644 src/libcamera/pipeline/virtual/common_functions.h
> > > create mode 100644
> src/libcamera/pipeline/virtual/image_frame_generator.cpp
> > > create mode 100644
> src/libcamera/pipeline/virtual/image_frame_generator.h
> > >
> > > diff --git a/src/libcamera/pipeline/virtual/README.md
> b/src/libcamera/pipeline/virtual/README.md
> > > index 27d6283df..5e21ce74a 100644
> > > --- a/src/libcamera/pipeline/virtual/README.md
> > > +++ b/src/libcamera/pipeline/virtual/README.md
> > > @@ -15,8 +15,13 @@ Each camera block is a dictionary, containing the
> following keys:
> > > - `supported_formats` (list of `VirtualCameraData::Resolution`,
> optional) : List of supported resolution and frame rates of the emulated
> camera
> > > - `width` (`unsigned int`, default=1920): Width of the window
> resolution. This needs to be even.
> > > - `height` (`unsigned int`, default=1080): Height of the window
> resolution.
> > > - - `frame_rates` (list of `int`, default=`[30,60]` ): Range of the
> frame rate. The list has to be two values of the lower bound and the upper
> bound of the frame rate.
> > > -- `test_pattern` (`string`, default="bars"): Which test pattern to
> use as frames. The options are "bars", "lines".
> > > + - `frame_rates` (list of `int`, default=`[30,60]` ): Range of the
> frame rate. The list has to be two values of the lower bound and the upper
> bound of the frame rate. This does not affect the frame rate for now.
> > > +- `frames` (dictionary):
> > > + - `path` (`string`, default="bars"): Name of a test pattern, path
> to an image, or path to a directory of a series of images.
> > > + - The test patterns are "bars" which means color bars, and
> "lines" which means diagonal lines.
> > > + - The path to an image has ".jpg" extension.
> > > + - The path to a directory ends with "/". The name of the images
> in the directory are "{n}.jpg" with {n} is the sequence of images starting
> with 0.
> > > + - `scale_mode`(`string`, default="fill"): Scale mode when the
> frames are images. The scale modes are "fill", "contain", and "cover". This
> does not matter when frames is a test pattern. This does not affect the
> scale mode for now.
> > > - `location` (`string`, default="front"): The location of the camera.
> Support "front" and "back". This is displayed in qcam camera selection
> window but this does not change the output.
> > > - `model` (`string`, default="Unknown"): The model name of the
> camera. This is displayed in qcam camera selection window but this does not
> change the output.
> > >
> > > @@ -35,13 +40,16 @@ A sample config file:
> > > frame_rates:
> > > - 70
> > > - 80
> > > - test_pattern: "bars"
> > > + frames:
> > > + path: "lines"
> > > location: "front"
> > > model: "Virtual Video Device"
> > > "Virtual1":
> > > supported_formats:
> > > - width: 800
> > > - test_pattern: "lines"
> > > + frames:
> > > + path: "path/to/directory_of_images/"
> > > + scale_mode: "contain"
> > > location: "back"
> > > model: "Virtual Video Device1"
> > > "Virtual2":
> > > @@ -61,7 +69,7 @@ This is the procedure of the Parser class:
> > > - If the config file contains invalid configuration, this method
> returns nullptr. The camera will be skipped.
> > > 3. Parse each property and register the data.
> > > - `parseSupportedFormats()`: Parses `supported_formats` in the
> config, which contains resolutions and frame rates.
> > > - - `parseTestPattern()`: Parses `test_pattern` in the config.
> > > + - `parseFrame()`: Parses `frames` in the config.
> > > - `parseLocation()`: Parses `location` in the config.
> > > - `parseModel()`: Parses `model` in the config.
> > > 4. Back to `parseConfigFile()` and append the camera configuration.
> > > diff --git a/src/libcamera/pipeline/virtual/common_functions.cpp
> b/src/libcamera/pipeline/virtual/common_functions.cpp
> > > new file mode 100644
> > > index 000000000..207827ee0
> > > --- /dev/null
> > > +++ b/src/libcamera/pipeline/virtual/common_functions.cpp
> > > @@ -0,0 +1,27 @@
> > > +/* SPDX-License-Identifier: LGPL-2.1-or-later */
> > > +/*
> > > + * Copyright (C) 2023, Google Inc.
> > > + *
> > > + * common_functions.cpp - Helper that do not depend on any class
> > > + */
> > > +
> > > +#include "common_functions.h"
> > > +
> > > +namespace libcamera {
> > > +
> > > +std::string getExtension(const std::string &path)
> >
> > This looks very fragile. Why not use
> `std::filesystem::path::extension()`?
> >
> >
> > > +{
> > > + size_t i = path.find(".");
> > > + if (i != std::string::npos) {
> > > + return path.substr(i);
> > > + }
> > > + return "";
> > > +}
> > > +
> > > +std::size_t numberOfFilesInDirectory(std::filesystem::path path)
> > > +{
> > > + using std::filesystem::directory_iterator;
> > > + return std::distance(directory_iterator(path),
> directory_iterator{});
> > > +}
>
> And inline this function so that you can drop common_functions.cpp/h
> completely
>
Done
>
> > > +
> > > +} // namespace libcamera
> > > diff --git a/src/libcamera/pipeline/virtual/common_functions.h
> b/src/libcamera/pipeline/virtual/common_functions.h
> > > new file mode 100644
> > > index 000000000..4203f9505
> > > --- /dev/null
> > > +++ b/src/libcamera/pipeline/virtual/common_functions.h
> > > @@ -0,0 +1,18 @@
> > > +/* SPDX-License-Identifier: LGPL-2.1-or-later */
> > > +/*
> > > + * Copyright (C) 2023, Google Inc.
> > > + *
> > > + * common_functions.h - Helper that do not depend on any class
> > > + */
> > > +
> > > +#pragma once
> > > +
> > > +#include <filesystem>
> > > +
> > > +namespace libcamera {
> > > +
> > > +std::string getExtension(const std::string &path);
> > > +
> > > +std::size_t numberOfFilesInDirectory(std::filesystem::path path);
> > > +
> > > +} // namespace libcamera
> > > diff --git a/src/libcamera/pipeline/virtual/frame_generator.h
> b/src/libcamera/pipeline/virtual/frame_generator.h
> > > index 9699af7a4..f69576b36 100644
> > > --- a/src/libcamera/pipeline/virtual/frame_generator.h
> > > +++ b/src/libcamera/pipeline/virtual/frame_generator.h
> > > @@ -23,7 +23,7 @@ public:
> > > /** Fill the output frame buffer.
> > > * Use the frame at the frameCount of image frames
> > > */
> > > - virtual void generateFrame(const Size &size,
> > > + virtual void generateFrame(unsigned int &frameCount, const Size
> &size,
>
> Are you passing frameCount by reference because you want the function
> to modify the value passed in by caller ? Use a pointer then which
> better conveys this semantic in my opinion.
>
Done. Thanks!
>
> > > const FrameBuffer *buffer) = 0;
> > >
> > > protected:
> > > diff --git a/src/libcamera/pipeline/virtual/image_frame_generator.cpp
> b/src/libcamera/pipeline/virtual/image_frame_generator.cpp
> > > new file mode 100644
> > > index 000000000..d374877ff
> > > --- /dev/null
> > > +++ b/src/libcamera/pipeline/virtual/image_frame_generator.cpp
> > > @@ -0,0 +1,154 @@
> > > +/* SPDX-License-Identifier: LGPL-2.1-or-later */
> > > +/*
> > > + * Copyright (C) 2023, Google Inc.
> > > + *
> > > + * image_frame_generator.cpp - Derived class of FrameGenerator for
> > > + * generating frames from images
> > > + */
> > > +
> > > +#include "image_frame_generator.h"
> > > +
> > > +#include <memory>
> > > +#include <optional>
>
> both already included by image_frame_generator.h
>
Removed. I added memory again as the source file uses
std::make_unique on its own as well, but anyways :)
>
> > > +#include <string>
> > > +
> > > +#include <libcamera/base/file.h>
> > > +#include <libcamera/base/log.h>
> > > +
> > > +#include <libcamera/framebuffer.h>
> > > +
> > > +#include "libcamera/internal/mapped_framebuffer.h"
> > > +
> > > +#include "libyuv/convert.h"
> > > +#include "libyuv/scale.h"
>
> empty line
>
Done
>
> > > +namespace libcamera {
> > > +
> > > +LOG_DECLARE_CATEGORY(Virtual)
> > > +
> > > +std::unique_ptr<ImageFrameGenerator> ImageFrameGenerator::create(
> > > + ImageFrames &imageFrames)
>
> Or rather
>
> std::unique_ptr<ImageFrameGenerator>
> ImageFrameGenerator::create(ImageFrames &imageFrames)
> {
>
Done
>
> > > +{
> > > + std::unique_ptr<ImageFrameGenerator> imageFrameGenerator =
> > > + std::make_unique<ImageFrameGenerator>();
> > > + imageFrameGenerator->imageFrames_ = &imageFrames;
> > > +
> > > + /** For each file in the directory
>
> This is not doxygen. Drop /** and use the conventional commenting
> style
> /*
> * This is for multi
> * line comments
> */
>
> Done. Thanks!
> > > + * load the image, convert it to NV12, and store the pointer
> > > + */
> > > + for (unsigned int i = 0; i < imageFrames.number.value_or(1); i++) {
> > > + std::string path;
> > > + if (!imageFrames.number.has_value()) {
>
> I think you can drop has_value() as it's an alias for operator bool
>
Done
>
> > > + /* If the path is to an image */
> > > + path = imageFrames.path;
> > > + } else {
> > > + /* If the path is to a directory */
> > > + path = constructPath(imageFrames.path, i);
>
> This assumes that if you're given a directory images are named
> 0.jpeg 1.jpeg etc. Up to you, but it would be trivial to enumerate all
> .jpeg files in a directory. True, you won't know in which order to
> display them then...
>
Yes, the original concern is exactly about the order... Do you think we
should use alphabetic order instead?
>
> > > + }
>
> No {} for single liners, please
>
Hmm, even with comments that make them more than one line?
>
> > > +
> > > + File file(path);
> > > + bool isOpen = file.open(File::OpenModeFlag::ReadOnly);
> > > + if (!isOpen) {
> >
> > I think `if (!file.open(...))` is fine.
> >
> >
> > > + LOG(Virtual, Error) << "Failed to open image file:
> " << file.fileName();
> > > + return nullptr;
> > > + }
> > > +
> > > + /* Read the image file to data */
> > > + uint8_t buffer[file.size()];
> >
> > It's probably best to avoid VLAs.
> >
> >
> > > + Span<unsigned char> data{ buffer, (unsigned
> long)file.size() };
>
> You can use uint8_t here as well for consistency
>
> No plain C-style casts. In the version I pulled from fdo you can
> declare fileSize a unsigned long and replace
>
> > > + long dataSize = file.read(data);
>
> with
>
> auto buffer = std::make_unique<uint8_t[]>(fileSize);
> long dataSize = file.read({ buffer.get(), fileSize });
>
> then replace all occurrences of data.data() with buffer.get().
>
> Also 'dataSize' and 'fileSize' should have the same value, so I think
> you can drop dataSize (or if you want to make sure do
>
> if (file.read({ buffer.get(), fileSize }) != fileSize)
> error out
>
> Adopted mostly, except that the return type of file.read is also ssize_t.
End up using one static_cast.
> > > +
> > > + /* Get the width and height of the image */
> > > + int width, height;
>
> unsigned
>
Hmm, libyuv::MJPGSize takes int* though...
>
> > > + if (libyuv::MJPGSize(data.data(), dataSize, &width,
> &height)) {
> > > + LOG(Virtual, Error) << "Failed to get the size of
> the image file: "
> > > + << file.fileName();
> > > + return nullptr;
> > > + }
> > > +
> > > + /* Convert to NV12 and write the data to tmpY and tmpUV */
> > > + int halfWidth = (width + 1) / 2;
> > > + int halfHeight = (height + 1) / 2;
>
> unsigned
>
Done
>
> > > + std::unique_ptr<uint8_t[]> dstY =
> > > + std::make_unique<uint8_t[]>(width * height);
> > > + std::unique_ptr<uint8_t[]> dstUV =
> > > + std::make_unique<uint8_t[]>(halfWidth * halfHeight
> * 2);
> > > + int ret = libyuv::MJPGToNV12(data.data(), dataSize,
> > > + dstY.get(), width,
> dstUV.get(),
> > > + width, width, height, width,
> height);
> > > + if (ret != 0) {
> > > + LOG(Virtual, Error) << "MJPGToNV12() failed with "
> << ret;
> > > + }
>
> no {}
>
> > > +
> > > + imageFrameGenerator->imageFrameDatas_.emplace_back(
> > > + ImageFrameData{ std::move(dstY), std::move(dstUV),
> > > + Size(width, height) });
> > > + }
>
> empty line before return
>
Done
>
> > > + return imageFrameGenerator;
> > > +}
> > > +
> > > +std::string ImageFrameGenerator::constructPath(std::string &name,
> unsigned int &i)
> >
> > Does this need to be in `ImageFrameGenerator`? Is it not enough to have
> it in an
> > anonymous namespace in this file?
>
> In the version I have pulled from gitlab this is now in an anonymous
> namespace. The real question I have is what is the point of a one
> liner function called from a single place. Drop it and inline it in
> the caller please
>
Done
>
> >
> >
> > > +{
> > > + return name + std::to_string(i) + ".jpg";
> > > +}
> > > +
> > > +void ImageFrameGenerator::configure(const Size &size)
> > > +{
> > > + for (unsigned int i = 0; i < imageFrames_->number.value_or(1);
> i++) {
> > > + /* Scale the imageFrameDatas_ to scaledY and scaledUV */
> > > + int halfSizeWidth = (size.width + 1) / 2;
> > > + int halfSizeHeight = (size.height + 1) / 2;
>
> unsigned
>
Done
>
> > > + std::unique_ptr<uint8_t[]> scaledY =
> > > + std::make_unique<uint8_t[]>(size.width *
> size.height);
> > > + std::unique_ptr<uint8_t[]> scaledUV =
> > > + std::make_unique<uint8_t[]>(halfSizeWidth *
> halfSizeHeight * 2);
> > > + auto &src = imageFrameDatas_[i];
> > > +
> > > + /*
> > > + * \todo Implement "contain" & "cover", based on
> > > + * |imageFrames_[i].scaleMode|.
> > > + */
>
> drop them from the class and documentation if not implemented
>
Done
>
> > > +
> > > + /*
> > > + * \todo Some platforms might enforce stride due to GPU,
> like
> > > + * ChromeOS ciri (64). The weight needs to be a
> multiple of
> > > + * the stride to work properly for now.
>
> Indented with spaces
>
Using tabs now.
>
> > > + */
> > > + libyuv::NV12Scale(src.Y.get(), src.size.width,
> > > + src.UV.get(), src.size.width,
> > > + src.size.width, src.size.height,
> > > + scaledY.get(), size.width,
> scaledUV.get(), size.width,
> > > + size.width, size.height,
> libyuv::FilterMode::kFilterBilinear);
> > > +
> > > + /* Store the pointers to member variable */
>
> rather useless comment
>
Removed
>
> > > + scaledFrameDatas_.emplace_back(
> > > + ImageFrameData{ std::move(scaledY),
> std::move(scaledUV), size });
> > > + }
> > > +}
> > > +
> > > +void ImageFrameGenerator::generateFrame(unsigned int &frameCount,
> const Size &size, const FrameBuffer *buffer)
> > > +{
> > > + /* Don't do anything when the list of buffers is empty*/
> > > + if (scaledFrameDatas_.size() == 0)
> > > + return;
>
> can this happen ? If create() fails, you shouln't be able to call
> generateFrames() and configure() doesn't seem to fail anywhere.
>
Hmm, actually I didn't check if a directory could be empty or not...
Added the check in the parser, and made this into an ASSERT instead.
>
> > > +
> > > + MappedFrameBuffer mappedFrameBuffer(buffer,
> MappedFrameBuffer::MapFlag::Write);
> > > +
> > > + auto planes = mappedFrameBuffer.planes();
> > > +
> > > + /* Make sure the frameCount does not over the number of images */
> > > + frameCount %= imageFrames_->number.value_or(1);
>
> am I wrong or if you have a single image the index you want is 0 not 1 ?
>
If the number is std::nullopt, which means that the path is a single image,
we can only divide `frameCount` by 1, instead of 0. (Well, we take the
remainder of the division)
`frameCount` will always be 0 in this case, which is what we want as the
index.
> > > +
> > > + /* Write the scaledY and scaledUV to the mapped frame buffer */
> > > + libyuv::NV12Copy(scaledFrameDatas_[frameCount].Y.get(), size.width,
> > > + scaledFrameDatas_[frameCount].UV.get(),
> size.width, planes[0].begin(),
> > > + size.width, planes[1].begin(), size.width,
> > > + size.width, size.height);
> > > +
> > > + /* proceed an image every 4 frames */
> > > + /* \todo read the parameter_ from the configuration file? */
> > > + parameter_++;
>
> this is never initialized
>
Right, reset it at the beginning of configure().
>
> > > + if (parameter_ % 4 == 0)
> > > + frameCount++;
> > > +}
> > > +
> > > +} // namespace libcamera
> > > diff --git a/src/libcamera/pipeline/virtual/image_frame_generator.h
> b/src/libcamera/pipeline/virtual/image_frame_generator.h
> > > new file mode 100644
> > > index 000000000..74468e075
> > > --- /dev/null
> > > +++ b/src/libcamera/pipeline/virtual/image_frame_generator.h
> > > @@ -0,0 +1,65 @@
> > > +/* SPDX-License-Identifier: LGPL-2.1-or-later */
> > > +/*
> > > + * Copyright (C) 2023, Google Inc.
> > > + *
> > > + * image_frame_generator.h - Derived class of FrameGenerator for
> > > + * generating frames from images
> > > + */
> > > +
> > > +#pragma once
> > > +
> > > +#include <memory>
> > > +#include <optional>
> > > +#include <stdint.h>
> > > +#include <sys/types.h>
> > > +
> > > +#include "frame_generator.h"
> > > +
> > > +namespace libcamera {
> > > +
> > > +enum class ScaleMode : char {
> > > + Fill = 0,
> > > + Contain = 1,
> > > + Cover = 2,
> > > +};
> > > +
> > > +/* Frame configuration provided by the config file */
> > > +struct ImageFrames {
> > > + std::string path;
> >
> > I think `std::filesystem::path` would be preferable.
> >
>
> yes!
>
Done :)
>
> >
> > > + ScaleMode scaleMode;
> > > + std::optional<unsigned int> number;
>
> What's the point of an optional ? The Parser::parseFrame() initializes
> this with either:
>
> data->config_.frame = ImageFrames{ path, scaleMode, std::nullopt };
> or
>
> data->config_.frame = ImageFrames{ path, scaleMode,
> numberOfFilesInDirectory(path) };
>
> and you always get it with value_or(1). Why not set it to 1 directly
> then ?
>
> It's checked in `ImageFrameGenerator::create`, to know if it's a file
or a directory [1].
[1]:
https://chromium-review.git.corp.google.com/c/chromiumos/third_party/libcamera/+/4874478/11/src/libcamera/pipeline/virtual/image_frame_generator.cpp#40
> > > +};
> > > +
> > > +class ImageFrameGenerator : public FrameGenerator
> > > +{
> > > +public:
> > > + /** Factory function to create an ImageFrameGenerator object.
>
> This is not doxygen.
> Use the canonical commenting style used throughout the whole code
> base.
> Move comments to function to the cpp file.
>
Done, please check again.
>
> > > + * Read the images and convert them to buffers in NV12 format.
> > > + * Store the pointers to the buffers to a list (imageFrameDatas)
> > > + */
> > > + static std::unique_ptr<ImageFrameGenerator> create(ImageFrames
> &imageFrames);
> > > +
> > > +private:
> > > + struct ImageFrameData {
> > > + std::unique_ptr<uint8_t[]> Y;
> > > + std::unique_ptr<uint8_t[]> UV;
> > > + Size size;
> > > + };
> > > +
> > > + /* Scale the buffers for image frames. */
> > > + void configure(const Size &size) override;
> > > + void generateFrame(unsigned int &frameCount, const Size &size,
> const FrameBuffer *buffer) override;
>
> You could easily
>
> void generateFrame(unsigned int &frameCount, const Size &size,
> const FrameBuffer *buffer) override;
>
> From coding-style.rst
>
> * 2 "Breaking Long Lines" striving to fit code within 80 columns and
> accepting up to 120 columns when necessary
>
> Done
> > > +
> > > + static std::string constructPath(std::string &name, unsigned int
> &i);
> > > +
> > > + /* List of pointers to the not scaled image buffers */
> > > + std::vector<ImageFrameData> imageFrameDatas_;
> > > + /* List of pointers to the scaled image buffers */
> > > + std::vector<ImageFrameData> scaledFrameDatas_;
> > > + /* Pointer to the imageFrames_ in VirtualCameraData */
> > > + ImageFrames *imageFrames_;
> > > + /* Speed parameter. Change to the next image every parameter_
> frames. */
> > > + int parameter_;
>
> unsigned
>
Done
>
> as long as it's not configurable make it a constexpr
>
It'll be updated in generateFrame().
>
> > > +};
> > > +
> > > +} /* namespace libcamera */
> > > diff --git a/src/libcamera/pipeline/virtual/meson.build
> b/src/libcamera/pipeline/virtual/meson.build
> > > index 2e82e64cb..d56aedec4 100644
> > > --- a/src/libcamera/pipeline/virtual/meson.build
> > > +++ b/src/libcamera/pipeline/virtual/meson.build
> > > @@ -2,11 +2,14 @@
> > >
> > > libcamera_sources += files([
> > > 'virtual.cpp',
> > > - 'test_pattern_generator.cpp',
> > > 'parser.cpp',
> > > + 'test_pattern_generator.cpp',
> > > + 'image_frame_generator.cpp',
> > > + 'common_functions.cpp',
> > > ])
> > >
> > > libyuv_dep = dependency('libyuv', required : false)
> > > +libjpeg = dependency('libjpeg', required : false)
>
> where is this required ?
>
I think it's needed in libyuv. Without it, building will fail with:
```
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::MJpegDecoder()':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:79:(.text+0xa9):
undefined reference to `jpeg_std_error'
/usr/bin/ld:
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:88:(.text+0x129):
undefined reference to `jpeg_resync_to_restart'
/usr/bin/ld:
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:90:(.text+0x15a):
undefined reference to `jpeg_CreateDecompress'
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::~MJpegDecoder()':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:97:(.text+0x1a8):
undefined reference to `jpeg_destroy_decompress'
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::LoadFrame(unsigned char const*, unsigned
long)':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:122:(.text+0x2b6):
undefined reference to `jpeg_read_header'
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::UnloadFrame()':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:242:(.text+0x7d2):
undefined reference to `jpeg_abort_decompress'
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::StartDecode()':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:528:(.text+0x166b):
undefined reference to `jpeg_start_decompress'
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::FinishDecode()':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:538:(.text+0x169e):
undefined reference to `jpeg_abort_decompress'
/usr/bin/ld: subprojects/libyuv/libyuv.a(source_mjpeg_decoder.cc.o): in
function `libyuv::MJpegDecoder::DecodeImcuRow()':
/usr/local/google/home/chenghaoyang/Workspace/libcamera/build/../subprojects/libyuv/source/mjpeg_decoder.cc:554:(.text._ZN6libyuv12MJpegDecoder13DecodeImcuRowEv[_ZN6libyuv12MJpegDecoder13DecodeImcuRowEv]+0x40):
undefined reference to `jpeg_read_raw_data'
collect2: error: ld returned 1 exit status
```
>
> > >
> > > # Fallback to a subproject if libyuv isn't found, as it's typically
> not
> > > # provided by distributions.
> > > @@ -26,3 +29,4 @@ if not libyuv_dep.found()
> > > endif
> > >
> > > libcamera_deps += [libyuv_dep]
> > > +libcamera_deps += [libjpeg]
> > > diff --git a/src/libcamera/pipeline/virtual/parser.cpp
> b/src/libcamera/pipeline/virtual/parser.cpp
> > > index 032c0cd9d..46d1ab181 100644
> > > --- a/src/libcamera/pipeline/virtual/parser.cpp
> > > +++ b/src/libcamera/pipeline/virtual/parser.cpp
> > > @@ -18,6 +18,7 @@
> > > #include "libcamera/internal/pipeline_handler.h"
> > > #include "libcamera/internal/yaml_parser.h"
> > >
> > > +#include "common_functions.h"
>
> drop this
>
Done
>
> > > #include "virtual.h"
> > >
> > > namespace libcamera {
> > > @@ -52,12 +53,12 @@ std::vector<std::unique_ptr<VirtualCameraData>>
> Parser::parseConfigFile(
> > > continue;
> > > }
> > >
> > > - data->id_ = cameraId;
> > > + data->config_.id = cameraId;
> > > ControlInfoMap::Map controls;
> > > /* todo: Check which resolution's frame rate to be
> reported */
> > > controls[&controls::FrameDurationLimits] =
> > > - ControlInfo(int64_t(1000 /
> data->supportedResolutions_[0].frameRates[1]),
> > > - int64_t(1000 /
> data->supportedResolutions_[0].frameRates[0]));
> > > + ControlInfo(int64_t(1000 /
> data->config_.resolutions[0].frameRates[1]),
> > > + int64_t(1000 /
> data->config_.resolutions[0].frameRates[0]));
> > > data->controlInfo_ = ControlInfoMap(std::move(controls),
> controls::controls);
> > > configurations.push_back(std::move(data));
> > > }
> > > @@ -72,7 +73,7 @@ std::unique_ptr<VirtualCameraData>
> Parser::parseCameraConfigData(
> > > if (parseSupportedFormats(cameraConfigData, data.get()))
> > > return nullptr;
> > >
> > > - if (parseTestPattern(cameraConfigData, data.get()))
> > > + if (parseFrame(cameraConfigData, data.get()))
> > > return nullptr;
> > >
> > > if (parseLocation(cameraConfigData, data.get()))
> > > @@ -122,14 +123,14 @@ int Parser::parseSupportedFormats(
> > > frameRates.push_back(60);
> > > }
> > >
> > > - data->supportedResolutions_.emplace_back(
> > > + data->config_.resolutions.emplace_back(
> > > VirtualCameraData::Resolution{ Size{
> width, height },
> > > frameRates
> });
> > >
> > > activeResolution = std::max(activeResolution,
> Size{ width, height });
> > > }
> > > } else {
> > > - data->supportedResolutions_.emplace_back(
> > > + data->config_.resolutions.emplace_back(
> > > VirtualCameraData::Resolution{ Size{ 1920, 1080 },
> > > { 30, 60 } });
> > > activeResolution = Size(1920, 1080);
> > > @@ -141,21 +142,65 @@ int Parser::parseSupportedFormats(
> > > return 0;
> > > }
> > >
> > > -int Parser::parseTestPattern(
> > > +int Parser::parseFrame(
> > > const YamlObject &cameraConfigData, VirtualCameraData *data)
>
> Why ??
>
> int Parser::parseFrame(const YamlObject &cameraConfigData,
> VirtualCameraData *data)
>
> is fine or
>
> int Parser::parseFrame(const YamlObject &cameraConfigData,
> VirtualCameraData *data)
>
> if you really want to
>
> Done
> > > {
> > > - std::string testPattern =
> cameraConfigData["test_pattern"].get<std::string>().value();
> > > + const YamlObject &frames = cameraConfigData["frames"];
>
> empty line please
>
Done
>
> > > + /* When there is no frames provided in the config file, use color
> bar test pattern */
> > > + if (frames.size() == 0) {
> > > + data->config_.frame = TestPattern::ColorBars;
> > > + return 0;
> > > + }
> > > +
> > > + if (!frames.isDictionary()) {
> > > + LOG(Virtual, Error) << "'frames' is not a dictionary.";
> > > + return -EINVAL;
> > > + }
> > > +
> > > + std::string path = frames["path"].get<std::string>().value();
> >
> > What if `path` is not specified? Is it checked somewhere else?
> >
>
> I would really make the "path" and "test_pattern" cases separate,
> otherwise you're here parsing a string that can be either a system
> path (which you have to validate) or "bars"\"lines" that are used for
> TPG.
>
Yeah you're right. Split it into two different arguments.
>
> >
> > > +
> > > + if (auto ext = getExtension(path); ext == ".jpg" || ext ==
> ".jpeg") {
> > > + ScaleMode scaleMode;
>
> declaring a variable inside the if () clause is valid but rather
> unusual (at least to me)
>
Moved it outside.
>
> > > + if (parseScaleMode(frames, &scaleMode))
> > > + return -EINVAL;
> > > + data->config_.frame = ImageFrames{ path, scaleMode,
> std::nullopt };
> > > + } else if (path.back() == '/') {
> >
> > This looks very fragile. Why not use `std::filesystem::symlink_status()`
> to check
> > if it's a file or a directory.
> >
> >
> > > + ScaleMode scaleMode;
> > > + if (parseScaleMode(frames, &scaleMode))
> > > + return -EINVAL;
> > > + data->config_.frame = ImageFrames{ path, scaleMode,
> > > +
> numberOfFilesInDirectory(path) };
> >
> > Admittedly I did not check in detail, but is this necessary? Is it
> necessary
> > to count the number of files in advance?
> >
> >
> > > + } else if (path == "bars" || path == "") {
> > > + /* Default value is "bars" */
> > > + data->config_.frame = TestPattern::ColorBars;
> > > + } else if (path == "lines") {
> > > + data->config_.frame = TestPattern::DiagonalLines;
> > > + } else {
> > > + LOG(Virtual, Error) << "Frame: " << path
> > > + << " is not supported";
> > > + return -EINVAL;
> > > + }
>
> Could you consider reworking the above by changing the schema to
> separate TPG from images ?
>
Done
>
> Empty line
>
Done
>
> > > + return 0;
> > > +}
> > >
> > > - /* Default value is "bars" */
> > > - if (testPattern == "bars" || testPattern == "") {
> > > - data->testPattern_ = TestPattern::ColorBars;
> > > - } else if (testPattern == "lines") {
> > > - data->testPattern_ = TestPattern::DiagonalLines;
> > > +int Parser::parseScaleMode(
> > > + const YamlObject &framesConfigData, ScaleMode *scaleMode)
> > > +{
> > > + std::string mode =
> framesConfigData["scale_mode"].get<std::string>().value();
> >
> > Same here, regarding empty optionals.
> >
> >
> > > +
> > > + /* Default value is fill */
> > > + if (mode == "fill" || mode == "") {
> > > + *scaleMode = ScaleMode::Fill;
> > > + } else if (mode == "contain") {
> > > + *scaleMode = ScaleMode::Contain;
> > > + } else if (mode == "cover") {
> > > + *scaleMode = ScaleMode::Cover;
>
> You don't seem to support these in code, why allow them ? Drop them
> all and assume the only supported one for now
>
Done
>
> > > } else {
> > > - LOG(Virtual, Error) << "Test pattern: " << testPattern
> > > - << "is not supported";
> > > + LOG(Virtual, Error) << "scaleMode: " << mode
> > > + << " is not supported";
> > > return -EINVAL;
> > > }
> > > +
> > > return 0;
> > > }
> > >
> > > @@ -195,4 +240,4 @@ int Parser::parseModel(
> > > return 0;
> > > }
> > >
> > > -} /* namespace libcamera */
> > > +} // namespace libcamera
>
> Why oh why ?
>
Hmm, not sure why I had this change...
Removed.
>
> > > diff --git a/src/libcamera/pipeline/virtual/parser.h
> b/src/libcamera/pipeline/virtual/parser.h
> > > index a377d8aa1..38ea460d5 100644
> > > --- a/src/libcamera/pipeline/virtual/parser.h
> > > +++ b/src/libcamera/pipeline/virtual/parser.h
> > > @@ -34,12 +34,15 @@ private:
> > >
> > > int parseSupportedFormats(
> > > const YamlObject &cameraConfigData, VirtualCameraData
> *data);
> > > - int parseTestPattern(
> > > + int parseFrame(
> > > const YamlObject &cameraConfigData, VirtualCameraData
> *data);
> > > int parseLocation(
> > > const YamlObject &cameraConfigData, VirtualCameraData
> *data);
> > > int parseModel(
> > > const YamlObject &cameraConfigData, VirtualCameraData
> *data);
> > > +
> > > + int parseScaleMode(
> > > + const YamlObject &framesConfigData, ScaleMode *scaleMode);
>
> I've gave up trying to understand these indentation rules
>
Made it one line.
>
> > > };
> > >
> > > } // namespace libcamera
> > > diff --git a/src/libcamera/pipeline/virtual/test_pattern_generator.cpp
> b/src/libcamera/pipeline/virtual/test_pattern_generator.cpp
> > > index 6df9b31e9..844dffd49 100644
> > > --- a/src/libcamera/pipeline/virtual/test_pattern_generator.cpp
> > > +++ b/src/libcamera/pipeline/virtual/test_pattern_generator.cpp
> > > @@ -20,7 +20,7 @@ LOG_DECLARE_CATEGORY(Virtual)
> > > static const unsigned int kARGBSize = 4;
> > >
> > > void TestPatternGenerator::generateFrame(
> > > - const Size &size,
> > > + [[maybe_unused]] unsigned int &frameCount, const Size &size,
> > > const FrameBuffer *buffer)
> > > {
> > > MappedFrameBuffer mappedFrameBuffer(buffer,
> > > @@ -29,7 +29,7 @@ void TestPatternGenerator::generateFrame(
> > > auto planes = mappedFrameBuffer.planes();
> > >
> > > /* TODO: select whether to do shifting or not */
> > > - shiftLeft(size);
> > > + // shiftLeft(size);
>
> Is this related at all ?
>
Removed.
>
> > >
> > > /* Convert the template_ to the frame buffer */
> > > int ret = libyuv::ARGBToNV12(
> > > diff --git a/src/libcamera/pipeline/virtual/test_pattern_generator.h
> b/src/libcamera/pipeline/virtual/test_pattern_generator.h
> > > index ed8d4e43b..cebdd0141 100644
> > > --- a/src/libcamera/pipeline/virtual/test_pattern_generator.h
> > > +++ b/src/libcamera/pipeline/virtual/test_pattern_generator.h
> > > @@ -25,7 +25,7 @@ enum class TestPattern : char {
> > > class TestPatternGenerator : public FrameGenerator
> > > {
> > > private:
> > > - void generateFrame(const Size &size,
> > > + void generateFrame(unsigned int &frameCount, const Size &size,
> > > const FrameBuffer *buffer) override;
> > >
> > > protected:
> > > diff --git a/src/libcamera/pipeline/virtual/virtual.cpp
> b/src/libcamera/pipeline/virtual/virtual.cpp
> > > index 0fe471f00..2412af703 100644
> > > --- a/src/libcamera/pipeline/virtual/virtual.cpp
> > > +++ b/src/libcamera/pipeline/virtual/virtual.cpp
> > > @@ -20,9 +20,13 @@
> > > #include "libcamera/internal/pipeline_handler.h"
> > > #include "libcamera/internal/yaml_parser.h"
> > >
> > > -#include "frame_generator.h"
> > > #include "parser.h"
> > >
> > > +#define ifTestPattern(v) std::holds_alternative<TestPattern>(v)
> > > +#define getTestPattern(v) std::get<TestPattern>(v)
> > > +#define ifImageFrames(v) std::holds_alternative<ImageFrames>(v)
> > > +#define getImageFrames(v) std::get<ImageFrames>(v)
> > > +
> > > namespace libcamera {
> > >
> > > LOG_DEFINE_CATEGORY(Virtual)
> > > @@ -63,12 +67,12 @@ CameraConfiguration::Status
> VirtualCameraConfiguration::validate()
> > > }
> > >
> > > Size maxSize;
> > > - for (const auto &resolution : data_->supportedResolutions_)
> > > + for (const auto &resolution : data_->config_.resolutions)
> > > maxSize = std::max(maxSize, resolution.size);
> > >
> > > for (StreamConfiguration &cfg : config_) {
> > > bool found = false;
> > > - for (const auto &resolution :
> data_->supportedResolutions_) {
> > > + for (const auto &resolution : data_->config_.resolutions) {
> > > if (resolution.size.width == cfg.size.width &&
> > > resolution.size.height == cfg.size.height) {
> > > found = true;
> > > @@ -110,7 +114,7 @@
> PipelineHandlerVirtual::generateConfiguration(Camera *camera,
> > > return config;
> > >
> > > Size minSize, sensorResolution;
> > > - for (const auto &resolution : data->supportedResolutions_) {
> > > + for (const auto &resolution : data->config_.resolutions) {
> > > if (minSize.isNull() || minSize > resolution.size)
> > > minSize = resolution.size;
> > >
> > > @@ -199,7 +203,7 @@ int PipelineHandlerVirtual::exportFrameBuffers(
> > > int PipelineHandlerVirtual::start(Camera *camera,
> > > [[maybe_unused]] const ControlList
> *controls)
> > > {
> > > - /* \todo Start reading the virtual video if any. */
> > > + /* Start reading the images/generating test patterns */
> > > VirtualCameraData *data = cameraData(camera);
> > >
> > >
> data->frameGenerator_->configure(data->stream_.configuration().size);
> > > @@ -219,8 +223,8 @@ int
> PipelineHandlerVirtual::queueRequestDevice([[maybe_unused]] Camera *camera,
> > >
> > > /* \todo Read from the virtual video if any. */
> > > for (auto const &[stream, buffer] : request->buffers()) {
> > > - /* map buffer and fill test patterns */
> > > -
> data->frameGenerator_->generateFrame(stream->configuration().size, buffer);
> > > + /* Map buffer. Fill test patterns or images */
> > > + data->frameGenerator_->generateFrame(data->frameCount_,
> stream->configuration().size, buffer);
>
> I don't see frameCount_ be used anywhere here, but I suspect you want
> it be modified by generateFrame() (answers my first question of the
> email). I don't know what you intend to use
> it for, but if, as the name suggest you want to count frames, you can
> do it here without the frameGenerator being involved ? It's
> one-request-one-frame after all, isn't it ? So you don't have to
> pollute TestPatternGenerator::generateFrame with a
> [[maybe_unused]] unsigned int &frameCount,
>
Yeah right. Migrated to ImageFrameGenerator.
I think the original idea is it allows the pipeline handler to control
which frame
to use in each request, but let's delay the feature when we actually
implement
it.
>
>
> > > completeBuffer(request, buffer);
> > > }
> > >
> > > @@ -250,9 +254,10 @@ bool
> PipelineHandlerVirtual::match([[maybe_unused]] DeviceEnumerator *enumerator
> > > /* Configure and register cameras with configData */
> > > for (auto &data : configData) {
> > > std::set<Stream *> streams{ &data->stream_ };
> > > - std::string id = data->id_;
> > > + std::string id = data->config_.id;
> > > std::shared_ptr<Camera> camera =
> Camera::create(std::move(data), id, streams);
> > >
> > > + /* Initialize FrameGenerator*/
> > > initFrameGenerator(camera.get());
> > >
> > > registerCamera(std::move(camera));
> > > @@ -264,13 +269,19 @@ bool
> PipelineHandlerVirtual::match([[maybe_unused]] DeviceEnumerator *enumerator
> > > void PipelineHandlerVirtual::initFrameGenerator(Camera *camera)
> > > {
> > > auto data = cameraData(camera);
> > > - if (data->testPattern_ == TestPattern::DiagonalLines) {
> > > - data->frameGenerator_ = DiagonalLinesGenerator::create();
> > > - } else {
> > > - data->frameGenerator_ = ColorBarsGenerator::create();
> > > + auto &frame = data->config_.frame;
> > > + if (ifTestPattern(frame)) {
> > > + TestPattern &testPattern = getTestPattern(frame);
> > > + if (testPattern == TestPattern::DiagonalLines) {
> > > + data->frameGenerator_ =
> DiagonalLinesGenerator::create();
> > > + } else {
> > > + data->frameGenerator_ =
> ColorBarsGenerator::create();
> > > + }
> > > + } else if (ifImageFrames(frame)) {
> > > + data->frameGenerator_ =
> ImageFrameGenerator::create(getImageFrames(frame));
> > > }
> >
> > I think I would strongly prefer `std::visit()` here. But even
> `std::get_if()` is better in my opinion.
>
> Regarding the revistied version on fdo
>
> if (testPattern ==
> TestPattern::DiagonalLines) {
> data->frameGenerator_ =
> DiagonalLinesGenerator::create();
> } else {
> data->frameGenerator_ =
> ColorBarsGenerator::create();
> }
>
> again, no {} for single line statements
>
Removed {}.
>
> >
> >
> > > }
> > >
> > > REGISTER_PIPELINE_HANDLER(PipelineHandlerVirtual, "virtual")
> > >
> > > -} /* namespace libcamera */
> > > +} // namespace libcamera
>
> meh
>
Removed.
>
> > > diff --git a/src/libcamera/pipeline/virtual/virtual.h
> b/src/libcamera/pipeline/virtual/virtual.h
> > > index c1ac4eb90..f41a4a906 100644
> > > --- a/src/libcamera/pipeline/virtual/virtual.h
> > > +++ b/src/libcamera/pipeline/virtual/virtual.h
> > > @@ -7,16 +7,22 @@
> > >
> > > #pragma once
> > >
> > > +#include <variant>
> > > +
> > > #include <libcamera/base/file.h>
> > >
> > > #include "libcamera/internal/camera.h"
> > > #include "libcamera/internal/dma_buf_allocator.h"
> > > #include "libcamera/internal/pipeline_handler.h"
> > >
> > > +#include "frame_generator.h"
> > > +#include "image_frame_generator.h"
> > > #include "test_pattern_generator.h"
> > >
> > > namespace libcamera {
> > >
> > > +using VirtualFrame = std::variant<TestPattern, ImageFrames>;
> > > +
> > > class VirtualCameraData : public Camera::Private
> > > {
> > > public:
> > > @@ -24,6 +30,13 @@ public:
> > > Size size;
> > > std::vector<int> frameRates;
> > > };
> > > + /* The config file is parsed to the Configuration struct */
> > > + struct Configuration {
> > > + std::string id;
>
> so you need <string>
>
Done
>
> > > + std::vector<Resolution> resolutions;
>
> and <vector>
>
Done
>
> > > + VirtualFrame frame;
> > > + };
> > > +
> > > VirtualCameraData(PipelineHandler *pipe)
> > > : Camera::Private(pipe)
> > > {
> > > @@ -31,12 +44,9 @@ public:
> > >
> > > ~VirtualCameraData() = default;
> > >
> > > - std::string id_;
> > > - std::vector<Resolution> supportedResolutions_;
>
> well, you did already
>
Sorry, I did what...?
>
> > > - TestPattern testPattern_;
> > > -
> > > + unsigned int frameCount_ = 0;
> > > + Configuration config_;
> > > Stream stream_;
> > > -
> > > std::unique_ptr<FrameGenerator> frameGenerator_;
> > > };
> > >
> > > --
> > > 2.46.0.184.g6999bdac58-goog
> > >
> >
> >
> > Regards,
> > Barnabás Pőcze
>
BR, Harvey
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.libcamera.org/pipermail/libcamera-devel/attachments/20240829/adc37018/attachment.htm>
More information about the libcamera-devel
mailing list