/* SPDX-License-Identifier: LGPL-2.1-or-later */ /* * Copyright (C) 2020, Google Inc. * * ipu3.cpp - IPU3 Image Processing Algorithms */ #include <algorithm> #include <array> #include <cmath> #include <limits> #include <map> #include <memory> #include <stdint.h> #include <utility> #include <vector> #include <linux/intel-ipu3.h> #include <linux/v4l2-controls.h> #include <libcamera/base/file.h> #include <libcamera/base/log.h> #include <libcamera/base/utils.h> #include <libcamera/control_ids.h> #include <libcamera/framebuffer.h> #include <libcamera/ipa/ipa_interface.h> #include <libcamera/ipa/ipa_module_info.h> #include <libcamera/ipa/ipu3_ipa_interface.h> #include <libcamera/request.h> #include "libcamera/internal/mapped_framebuffer.h" #include "libcamera/internal/yaml_parser.h" #include "algorithms/af.h" #include "algorithms/agc.h" #include "algorithms/algorithm.h" #include "algorithms/awb.h" #include "algorithms/blc.h" #include "algorithms/tone_mapping.h" #include "libipa/camera_sensor_helper.h" /* Minimum grid width, expressed as a number of cells */ static constexpr uint32_t kMinGridWidth = 16; /* Maximum grid width, expressed as a number of cells */ static constexpr uint32_t kMaxGridWidth = 80; /* Minimum grid height, expressed as a number of cells */ static constexpr uint32_t kMinGridHeight = 16; /* Maximum grid height, expressed as a number of cells */ static constexpr uint32_t kMaxGridHeight = 60; /* log2 of the minimum grid cell width and height, in pixels */ static constexpr uint32_t kMinCellSizeLog2 = 3; /* log2 of the maximum grid cell width and height, in pixels */ static constexpr uint32_t kMaxCellSizeLog2 = 6; namespace libcamera { LOG_DEFINE_CATEGORY(IPAIPU3) using namespace std::literals::chrono_literals; namespace ipa::ipu3 { /** * \brief The IPU3 IPA implementation * * The IPU3 Pipeline defines an IPU3-specific interface for communication * between the PipelineHandler and the IPA module. * * We extend the IPAIPU3Interface to implement our algorithms and handle * calls from the IPU3 PipelineHandler to satisfy requests from the * application. * * At initialisation time, a CameraSensorHelper is instantiated to support * camera-specific calculations, while the default controls are computed, and * the algorithms are instantiated from the tuning data file. * * The IPU3 ImgU operates with a grid layout to divide the overall frame into * rectangular cells of pixels. When the IPA is configured, we determine the * best grid for the statistics based on the pipeline handler Bayer Down Scaler * output size. * * Two main events are then handled to operate the IPU3 ImgU by populating its * parameter buffer, and adapting the settings of the sensor attached to the * IPU3 CIO2 through sensor-specific V4L2 controls. * * In fillParamsBuffer(), we populate the ImgU parameter buffer with * settings to configure the device in preparation for handling the frame * queued in the Request. * * When the frame has completed processing, the ImgU will generate a statistics * buffer which is given to the IPA with processStatsBuffer(). In this we run the * algorithms to parse the statistics and cache any results for the next * fillParamsBuffer() call. * * The individual algorithms are split into modular components that are called * iteratively to allow them to process statistics from the ImgU in the order * defined in the tuning data file. * * The current implementation supports five core algorithms: * * - Auto focus (AF) * - Automatic gain and exposure control (AGC) * - Automatic white balance (AWB) * - Black level correction (BLC) * - Tone mapping (Gamma) * * AWB is implemented using a Greyworld algorithm, and calculates the red and * blue gains to apply to generate a neutral grey frame overall. * * AGC is handled by calculating a histogram of the green channel to estimate an * analogue gain and shutter time which will provide a well exposed frame. A * low-pass IIR filter is used to smooth the changes to the sensor to reduce * perceivable steps. * * The tone mapping algorithm provides a gamma correction table to improve the * contrast of the scene. * * The black level compensation algorithm subtracts a hardcoded black level from * all pixels. * * The IPU3 ImgU has further processing blocks to support image quality * improvements through bayer and temporal noise reductions, however those are * not supported in the current implementation, and will use default settings as * provided by the kernel driver. * * Demosaicing is operating with the default parameters and could be further * optimised to provide improved sharpening coefficients, checker artifact * removal, and false color correction. * * Additional image enhancements can be made by providing lens and * sensor-specific tuning to adapt for Black Level compensation (BLC), Lens * shading correction (SHD) and Color correction (CCM). */ class IPAIPU3 : public IPAIPU3Interface, public Module { public: int init(const IPASettings &settings, const IPACameraSensorInfo &sensorInfo, const ControlInfoMap &sensorControls, ControlInfoMap *ipaControls) override; int start() override; void stop() override; int configure(const IPAConfigInfo &configInfo, ControlInfoMap *ipaControls) override; void mapBuffers(const std::vector<IPABuffer> &buffers) override; void unmapBuffers(const std::vector<unsigned int> &ids) override; void queueRequest(const uint32_t frame, const ControlList &controls) override; void fillParamsBuffer(const uint32_t frame, const uint32_t bufferId) override; void processStatsBuffer(const uint32_t frame, const int64_t frameTimestamp, const uint32_t bufferId, const ControlList &sensorControls) override; protected: std::string logPrefix() const override; private: void updateControls(const IPACameraSensorInfo &sensorInfo, const ControlInfoMap &sensorControls, ControlInfoMap *ipaControls); void updateSessionConfiguration(const ControlInfoMap &sensorControls); bool validateSensorControls(); void setControls(unsigned int frame); void calculateBdsGrid(const Size &bdsOutputSize); std::map<unsigned int, MappedFrameBuffer> buffers_; ControlInfoMap sensorCtrls_; ControlInfoMap lensCtrls_; IPACameraSensorInfo sensorInfo_; /* Interface to the Camera Helper */ std::unique_ptr<CameraSensorHelper> camHelper_; /* Local parameter storage */ struct IPAContext context_; }; std::string IPAIPU3::logPrefix() const { return "ipu3"; } /** * \brief Compute IPASessionConfiguration using the sensor information and the * sensor V4L2 controls */ void IPAIPU3::updateSessionConfiguration(const ControlInfoMap &sensorControls) { const ControlInfo vBlank = sensorControls.find(V4L2_CID_VBLANK)->second; context_.configuration.sensor.defVBlank = vBlank.def().get<int32_t>(); const ControlInfo &v4l2Exposure = sensorControls.find(V4L2_CID_EXPOSURE)->second; int32_t minExposure = v4l2Exposure.min().get<int32_t>(); int32_t maxExposure = v4l2Exposure.max().get<int32_t>(); const ControlInfo &v4l2Gain = sensorControls.find(V4L2_CID_ANALOGUE_GAIN)->second; int32_t minGain = v4l2Gain.min().get<int32_t>(); int32_t maxGain = v4l2Gain.max().get<int32_t>(); /* * When the AGC computes the new exposure values for a frame, it needs * to know the limits for shutter speed and analogue gain. * As it depends on the sensor, update it with the controls. * * \todo take VBLANK into account for maximum shutter speed */ context_.configuration.agc.minShutterSpeed = minExposure * context_.configuration.sensor.lineDuration; context_.configuration.agc.maxShutterSpeed = maxExposure * context_.configuration.sensor.lineDuration; context_.configuration.agc.minAnalogueGain = camHelper_->gain(minGain); context_.configuration.agc.maxAnalogueGain = camHelper_->gain(maxGain); } /** * \brief Compute camera controls using the sensor information and the sensor * V4L2 controls * * Some of the camera controls are computed by the pipeline handler, some others * by the IPA module which is in charge of handling, for example, the exposure * time and the frame duration. * * This function computes: * - controls::ExposureTime * - controls::FrameDurationLimits */ void IPAIPU3::updateControls(const IPACameraSensorInfo &sensorInfo, const ControlInfoMap &sensorControls, ControlInfoMap *ipaControls) { ControlInfoMap::Map controls{}; double lineDuration = context_.configuration.sensor.lineDuration.get<std::micro>(); /* * Compute exposure time limits by using line length and pixel rate * converted to microseconds. Use the V4L2_CID_EXPOSURE control to get * exposure min, max and default and convert it from lines to * microseconds. */ const ControlInfo &v4l2Exposure = sensorControls.find(V4L2_CID_EXPOSURE)->second; int32_t minExposure = v4l2Exposure.min().get<int32_t>() * lineDuration; int32_t maxExposure = v4l2Exposure.max().get<int32_t>() * lineDuration; int32_t defExposure = v4l2Exposure.def().get<int32_t>() * lineDuration; controls[&controls::ExposureTime] = ControlInfo(minExposure, maxExposure, defExposure); /* * Compute the frame duration limits. * * The frame length is computed assuming a fixed line length combined * with the vertical frame sizes. */ const ControlInfo &v4l2HBlank = sensorControls.find(V4L2_CID_HBLANK)->second; uint32_t hblank = v4l2HBlank.def().get<int32_t>(); uint32_t lineLength = sensorInfo.outputSize.width + hblank; const ControlInfo &v4l2VBlank = sensorControls.find(V4L2_CID_VBLANK)->second; std::array<uint32_t, 3> frameHeights{ v4l2VBlank.min().get<int32_t>() + sensorInfo.outputSize.height, v4l2VBlank.max().get<int32_t>() + sensorInfo.outputSize.height, v4l2VBlank.def().get<int32_t>() + sensorInfo.outputSize.height, }; std::array<int64_t, 3> frameDurations; for (unsigned int i = 0; i < frameHeights.size(); ++i) { uint64_t frameSize = lineLength * frameHeights[i]; frameDurations[i] = frameSize / (sensorInfo.pixelRate / 1000000U); } controls[&controls::FrameDurationLimits] = ControlInfo(frameDurations[0], frameDurations[1], frameDurations[2]); *ipaControls = ControlInfoMap(std::move(controls), controls::controls); } /** * \brief Validate that the sensor controls mandatory for the IPA exists */ bool IPAIPU3::validateSensorControls() { static const uint32_t ctrls[] = { V4L2_CID_ANALOGUE_GAIN, V4L2_CID_EXPOSURE, V4L2_CID_VBLANK, }; for (auto c : ctrls) { if (sensorCtrls_.find(c) == sensorCtrls_.end()) { LOG(IPAIPU3, Error) << "Unable to find sensor control " << utils::hex(c); return false; } } return true; } /** * \brief Initialize the IPA module and its controls * * This function receives the camera sensor information from the pipeline * handler, computes the limits of the controls it handles and returns * them in the \a ipaControls output parameter. */ int IPAIPU3::init(const IPASettings &settings, const IPACameraSensorInfo &sensorInfo, const ControlInfoMap &sensorControls, ControlInfoMap *ipaControls) { camHelper_ = CameraSensorHelperFactory::create(settings.sensorModel); if (camHelper_ == nullptr) { LOG(IPAIPU3, Error) << "Failed to create camera sensor helper for " << settings.sensorModel; return -ENODEV; } /* Clean context */ context_.configuration = {}; context_.configuration.sensor.lineDuration = sensorInfo.lineLength * 1.0s / sensorInfo.pixelRate; /* Load the tuning data file. */ File file(settings.configurationFile); if (!file.open(File::OpenModeFlag::ReadOnly)) { int ret = file.error(); LOG(IPAIPU3, Error) << "Failed to open configuration file " << settings.configurationFile << ": " << strerror(-ret); return ret; } std::unique_ptr<libcamera::YamlObject> data = YamlParser::parse(file); if (!data) return -EINVAL; unsigned int version = (*data)["version"].get<uint32_t>(0); if (version != 1) { LOG(IPAIPU3, Error) << "Invalid tuning file version " << version; return -EINVAL; } if (!data->contains("algorithms")) { LOG(IPAIPU3, Error) << "Tuning file doesn't contain any algorithm"; return -EINVAL; } int ret = createAlgorithms(context_, (*data)["algorithms"]); if (ret) return ret; /* Initialize controls. */ updateControls(sensorInfo, sensorControls, ipaControls); return 0; } /** * \brief Perform any processing required before the first frame */ int IPAIPU3::start() { /* * Set the sensors V4L2 controls before the first frame to ensure that * we have an expected and known configuration from the start. */ setControls(0); return 0; } /** * \brief Ensure that all processing has completed */ void IPAIPU3::stop() { } /** * \brief Calculate a grid for the AWB statistics * * This function calculates a grid for the AWB algorithm in the IPU3 firmware. * Its input is the BDS output size calculated in the ImgU. * It is limited for now to the simplest method: find the lesser error * with the width/height and respective log2 width/height of the cells. * * \todo The frame is divided into cells which can be 8x8 => 64x64. * As a smaller cell improves the algorithm precision, adapting the * x_start and y_start parameters of the grid would provoke a loss of * some pixels but would also result in more accurate algorithms. */ void IPAIPU3::calculateBdsGrid(const Size &bdsOutputSize) { Size best; Size bestLog2; /* Set the BDS output size in the IPAConfiguration structure */ context_.configuration.grid.bdsOutputSize = bdsOutputSize; uint32_t minError = std::numeric_limits<uint32_t>::max(); for (uint32_t shift = kMinCellSizeLog2; shift <= kMaxCellSizeLog2; ++shift) { uint32_t width = std::clamp(bdsOutputSize.width >> shift, kMinGridWidth, kMaxGridWidth); width = width << shift; uint32_t error = utils::abs_diff(width, bdsOutputSize.width); if (error >= minError) continue; minError = error; best.width = width; bestLog2.width = shift; } minError = std::numeric_limits<uint32_t>::max(); for (uint32_t shift = kMinCellSizeLog2; shift <= kMaxCellSizeLog2; ++shift) { uint32_t height = std::clamp(bdsOutputSize.height >> shift, kMinGridHeight, kMaxGridHeight); height = height << shift; uint32_t error = utils::abs_diff(height, bdsOutputSize.height); if (error >= minError) continue; minError = error; best.height = height; bestLog2.height = shift; } struct ipu3_uapi_grid_config &bdsGrid = context_.configuration.grid.bdsGrid; bdsGrid.x_start = 0; bdsGrid.y_start = 0; bdsGrid.width = best.width >> bestLog2.width; bdsGrid.block_width_log2 = bestLog2.width; bdsGrid.height = best.height >> bestLog2.height; bdsGrid.block_height_log2 = bestLog2.height; /* The ImgU pads the lines to a multiple of 4 cells. */ context_.configuration.grid.stride = utils::alignUp(bdsGrid.width, 4); LOG(IPAIPU3, Debug) << "Best grid found is: (" << (int)bdsGrid.width << " << " << (int)bdsGrid.block_width_log2 << ") x (" << (int)bdsGrid.height << " << " << (int)bdsGrid.block_height_log2 << ")"; } /** * \brief Configure the IPU3 IPA * \param[in] configInfo The IPA configuration data, received from the pipeline * handler * \param[in] ipaControls The IPA controls to update * * Calculate the best grid for the statistics based on the pipeline handler BDS * output, and parse the minimum and maximum exposure and analogue gain control * values. * * \todo Document what the BDS is, ideally in a block diagram of the ImgU. * * All algorithm modules are called to allow them to prepare the * \a IPASessionConfiguration structure for the \a IPAContext. */ int IPAIPU3::configure(const IPAConfigInfo &configInfo, ControlInfoMap *ipaControls) { if (configInfo.sensorControls.empty()) { LOG(IPAIPU3, Error) << "No sensor controls provided"; return -ENODATA; } sensorInfo_ = configInfo.sensorInfo; lensCtrls_ = configInfo.lensControls; /* * Compute the sensor V4L2 controls to be used by the algorithms and * to be set on the sensor. */ sensorCtrls_ = configInfo.sensorControls; calculateBdsGrid(configInfo.bdsOutputSize); /* Clean IPAActiveState at each reconfiguration. */ context_.activeState = {}; IPAFrameContext initFrameContext; context_.frameContexts.fill(initFrameContext); if (!validateSensorControls()) { LOG(IPAIPU3, Error) << "Sensor control validation failed."; return -EINVAL; } /* Update the camera controls using the new sensor settings. */ updateControls(sensorInfo_, sensorCtrls_, ipaControls); /* Update the IPASessionConfiguration using the sensor settings. */ updateSessionConfiguration(sensorCtrls_); for (auto const &algo : algorithms()) { int ret = algo->configure(context_, configInfo); if (ret) return ret; } return 0; } /** * \brief Map the parameters and stats buffers allocated in the pipeline handler * \param[in] buffers The buffers to map */ void IPAIPU3::mapBuffers(const std::vector<IPABuffer> &buffers) { for (const IPABuffer &buffer : buffers) { const FrameBuffer fb(buffer.planes); buffers_.emplace(buffer.id, MappedFrameBuffer(&fb, MappedFrameBuffer::MapFlag::ReadWrite)); } } /** * \brief Unmap the parameters and stats buffers * \param[in] ids The IDs of the buffers to unmap */ void IPAIPU3::unmapBuffers(const std::vector<unsigned int> &ids) { for (unsigned int id : ids) { auto it = buffers_.find(id); if (it == buffers_.end()) continue; buffers_.erase(it); } } /** * \brief Fill and return a buffer with ISP processing parameters for a frame * \param[in] frame The frame number * \param[in] bufferId ID of the parameter buffer to fill * * Algorithms are expected to fill the IPU3 parameter buffer for the next * frame given their most recent processing of the ImgU statistics. */ void IPAIPU3::fillParamsBuffer(const uint32_t frame, const uint32_t bufferId) { auto it = buffers_.find(bufferId); if (it == buffers_.end()) { LOG(IPAIPU3, Error) << "Could not find param buffer!"; return; } Span<uint8_t> mem = it->second.planes()[0]; ipu3_uapi_params *params = reinterpret_cast<ipu3_uapi_params *>(mem.data()); /* * The incoming params buffer may contain uninitialised data, or the * parameters of previously queued frames. Clearing the entire buffer * may be an expensive operation, and the kernel will only read from * structures which have their associated use-flag set. * * It is the responsibility of the algorithms to set the use flags * accordingly for any data structure they update during prepare(). */ params->use = {}; IPAFrameContext &frameContext = context_.frameContexts[frame % kMaxFrameContexts]; for (auto const &algo : algorithms()) algo->prepare(context_, frame, frameContext, params); paramsBufferReady.emit(frame); } /** * \brief Process the statistics generated by the ImgU * \param[in] frame The frame number * \param[in] frameTimestamp Timestamp of the frame * \param[in] bufferId ID of the statistics buffer * \param[in] sensorControls Sensor controls * * Parse the most recently processed image statistics from the ImgU. The * statistics are passed to each algorithm module to run their calculations and * update their state accordingly. */ void IPAIPU3::processStatsBuffer(const uint32_t frame, [[maybe_unused]] const int64_t frameTimestamp, const uint32_t bufferId, const ControlList &sensorControls) { auto it = buffers_.find(bufferId); if (it == buffers_.end()) { LOG(IPAIPU3, Error) << "Could not find stats buffer!"; return; } Span<uint8_t> mem = it->second.planes()[0]; const ipu3_uapi_stats_3a *stats = reinterpret_cast<ipu3_uapi_stats_3a *>(mem.data()); IPAFrameContext &frameContext = context_.frameContexts[frame % kMaxFrameContexts]; frameContext.sensor.exposure = sensorControls.get(V4L2_CID_EXPOSURE).get<int32_t>(); frameContext.sensor.gain = camHelper_->gain(sensorControls.get(V4L2_CID_ANALOGUE_GAIN).get<int32_t>()); double lineDuration = context_.configuration.sensor.lineDuration.get<std::micro>(); int32_t vBlank = context_.configuration.sensor.defVBlank; ControlList ctrls(controls::controls); for (auto const &algo : algorithms()) algo->process(context_, frame, frameContext, stats); setControls(frame); /* \todo Use VBlank value calculated from each frame exposure. */ int64_t frameDuration = (vBlank + sensorInfo_.outputSize.height) * lineDuration; ctrls.set(controls::FrameDuration, frameDuration); ctrls.set(controls::AnalogueGain, frameContext.sensor.gain); ctrls.set(controls::ColourTemperature, context_.activeState.awb.temperatureK); ctrls.set(controls::ExposureTime, frameContext.sensor.exposure * lineDuration); /* * \todo The Metadata provides a path to getting extended data * out to the application. Further data such as a simplifed Histogram * might have value to be exposed, however such data may be * difficult to report in a generically parsable way and we * likely want to avoid putting platform specific metadata in. */ metadataReady.emit(frame, ctrls); } /** * \brief Queue a request and process the control list from the application * \param[in] frame The number of the frame which will be processed next * \param[in] controls The controls for the \a frame * * Parse the request to handle any IPA-managed controls that were set from the * application such as manual sensor settings. */ void IPAIPU3::queueRequest(const uint32_t frame, const ControlList &controls) { /* \todo Start processing for 'frame' based on 'controls'. */ context_.frameContexts[frame % kMaxFrameContexts] = { controls }; } /** * \brief Handle sensor controls for a given \a frame number * \param[in] frame The frame on which the sensor controls should be set * * Send the desired sensor control values to the pipeline handler to request * that they are applied on the camera sensor. */ void IPAIPU3::setControls(unsigned int frame) { int32_t exposure = context_.activeState.agc.exposure; int32_t gain = camHelper_->gainCode(context_.activeState.agc.gain); ControlList ctrls(sensorCtrls_); ctrls.set(V4L2_CID_EXPOSURE, exposure); ctrls.set(V4L2_CID_ANALOGUE_GAIN, gain); ControlList lensCtrls(lensCtrls_); lensCtrls.set(V4L2_CID_FOCUS_ABSOLUTE, static_cast<int32_t>(context_.activeState.af.focus)); setSensorControls.emit(frame, ctrls, lensCtrls); } } /* namespace ipa::ipu3 */ /** * \brief External IPA module interface * * The IPAModuleInfo is required to match an IPA module construction against the * intented pipeline handler with the module. The API and pipeline handler * versions must match the corresponding IPA interface and pipeline handler. * * \sa struct IPAModuleInfo */ extern "C" { const struct IPAModuleInfo ipaModuleInfo = { IPA_MODULE_API_VERSION, 1, "PipelineHandlerIPU3", "ipu3", }; /** * \brief Create an instance of the IPA interface * * This function is the entry point of the IPA module. It is called by the IPA * manager to create an instance of the IPA interface for each camera. When * matched against with a pipeline handler, the IPAManager will construct an IPA * instance for each associated Camera. */ IPAInterface *ipaCreate() { return new ipa::ipu3::IPAIPU3(); } } } /* namespace libcamera */