diff options
Diffstat (limited to 'src/libcamera/software_isp/debayer_cpu.cpp')
| -rw-r--r-- | src/libcamera/software_isp/debayer_cpu.cpp | 297 |
1 files changed, 202 insertions, 95 deletions
diff --git a/src/libcamera/software_isp/debayer_cpu.cpp b/src/libcamera/software_isp/debayer_cpu.cpp index 82163458..66f6038c 100644 --- a/src/libcamera/software_isp/debayer_cpu.cpp +++ b/src/libcamera/software_isp/debayer_cpu.cpp @@ -1,23 +1,28 @@ /* SPDX-License-Identifier: LGPL-2.1-or-later */ /* * Copyright (C) 2023, Linaro Ltd - * Copyright (C) 2023, Red Hat Inc. + * Copyright (C) 2023-2025 Red Hat Inc. * * Authors: * Hans de Goede <hdegoede@redhat.com> * - * debayer_cpu.cpp - CPU based debayering class + * CPU based debayering class */ #include "debayer_cpu.h" -#include <math.h> +#include <algorithm> #include <stdlib.h> +#include <sys/ioctl.h> #include <time.h> +#include <utility> + +#include <linux/dma-buf.h> #include <libcamera/formats.h> #include "libcamera/internal/bayer_format.h" +#include "libcamera/internal/dma_buf_allocator.h" #include "libcamera/internal/framebuffer.h" #include "libcamera/internal/mapped_framebuffer.h" @@ -35,7 +40,7 @@ namespace libcamera { * \param[in] stats Pointer to the stats object to use */ DebayerCpu::DebayerCpu(std::unique_ptr<SwStatsCpu> stats) - : stats_(std::move(stats)), gammaCorrection_(1.0) + : stats_(std::move(stats)) { /* * Reading from uncached buffers may be very slow. @@ -47,69 +52,87 @@ DebayerCpu::DebayerCpu(std::unique_ptr<SwStatsCpu> stats) */ enableInputMemcpy_ = true; - /* Initialize gamma to 1.0 curve */ - for (unsigned int i = 0; i < kGammaLookupSize; i++) - gamma_[i] = i / (kGammaLookupSize / kRGBLookupSize); - - for (unsigned int i = 0; i < kMaxLineBuffers; i++) - lineBuffers_[i] = nullptr; + /* Initialize color lookup tables */ + for (unsigned int i = 0; i < DebayerParams::kRGBLookupSize; i++) { + red_[i] = green_[i] = blue_[i] = i; + redCcm_[i] = { static_cast<int16_t>(i), 0, 0 }; + greenCcm_[i] = { 0, static_cast<int16_t>(i), 0 }; + blueCcm_[i] = { 0, 0, static_cast<int16_t>(i) }; + } } -DebayerCpu::~DebayerCpu() -{ - for (unsigned int i = 0; i < kMaxLineBuffers; i++) - free(lineBuffers_[i]); -} +DebayerCpu::~DebayerCpu() = default; #define DECLARE_SRC_POINTERS(pixel_t) \ const pixel_t *prev = (const pixel_t *)src[0] + xShift_; \ const pixel_t *curr = (const pixel_t *)src[1] + xShift_; \ const pixel_t *next = (const pixel_t *)src[2] + xShift_; +#define GAMMA(value) \ + *dst++ = gammaLut_[std::clamp(value, 0, static_cast<int>(gammaLut_.size()) - 1)] + +#define STORE_PIXEL(b_, g_, r_) \ + if constexpr (ccmEnabled) { \ + const DebayerParams::CcmColumn &blue = blueCcm_[b_]; \ + const DebayerParams::CcmColumn &green = greenCcm_[g_]; \ + const DebayerParams::CcmColumn &red = redCcm_[r_]; \ + GAMMA(blue.b + green.b + red.b); \ + GAMMA(blue.g + green.g + red.g); \ + GAMMA(blue.r + green.r + red.r); \ + } else { \ + *dst++ = blue_[b_]; \ + *dst++ = green_[g_]; \ + *dst++ = red_[r_]; \ + } \ + if constexpr (addAlphaByte) \ + *dst++ = 255; \ + x++; + /* * RGR * GBG * RGR */ -#define BGGR_BGR888(p, n, div) \ - *dst++ = blue_[curr[x] / (div)]; \ - *dst++ = green_[(prev[x] + curr[x - p] + curr[x + n] + next[x]) / (4 * (div))]; \ - *dst++ = red_[(prev[x - p] + prev[x + n] + next[x - p] + next[x + n]) / (4 * (div))]; \ - x++; +#define BGGR_BGR888(p, n, div) \ + STORE_PIXEL( \ + curr[x] / (div), \ + (prev[x] + curr[x - p] + curr[x + n] + next[x]) / (4 * (div)), \ + (prev[x - p] + prev[x + n] + next[x - p] + next[x + n]) / (4 * (div))) /* * GBG * RGR * GBG */ -#define GRBG_BGR888(p, n, div) \ - *dst++ = blue_[(prev[x] + next[x]) / (2 * (div))]; \ - *dst++ = green_[curr[x] / (div)]; \ - *dst++ = red_[(curr[x - p] + curr[x + n]) / (2 * (div))]; \ - x++; +#define GRBG_BGR888(p, n, div) \ + STORE_PIXEL( \ + (prev[x] + next[x]) / (2 * (div)), \ + curr[x] / (div), \ + (curr[x - p] + curr[x + n]) / (2 * (div))) /* * GRG * BGB * GRG */ -#define GBRG_BGR888(p, n, div) \ - *dst++ = blue_[(curr[x - p] + curr[x + n]) / (2 * (div))]; \ - *dst++ = green_[curr[x] / (div)]; \ - *dst++ = red_[(prev[x] + next[x]) / (2 * (div))]; \ - x++; +#define GBRG_BGR888(p, n, div) \ + STORE_PIXEL( \ + (curr[x - p] + curr[x + n]) / (2 * (div)), \ + curr[x] / (div), \ + (prev[x] + next[x]) / (2 * (div))) /* * BGB * GRG * BGB */ -#define RGGB_BGR888(p, n, div) \ - *dst++ = blue_[(prev[x - p] + prev[x + n] + next[x - p] + next[x + n]) / (4 * (div))]; \ - *dst++ = green_[(prev[x] + curr[x - p] + curr[x + n] + next[x]) / (4 * (div))]; \ - *dst++ = red_[curr[x] / (div)]; \ - x++; +#define RGGB_BGR888(p, n, div) \ + STORE_PIXEL( \ + (prev[x - p] + prev[x + n] + next[x - p] + next[x + n]) / (4 * (div)), \ + (prev[x] + curr[x - p] + curr[x + n] + next[x]) / (4 * (div)), \ + curr[x] / (div)) +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer8_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) { DECLARE_SRC_POINTERS(uint8_t) @@ -120,6 +143,7 @@ void DebayerCpu::debayer8_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer8_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) { DECLARE_SRC_POINTERS(uint8_t) @@ -130,6 +154,7 @@ void DebayerCpu::debayer8_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer10_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) { DECLARE_SRC_POINTERS(uint16_t) @@ -141,6 +166,7 @@ void DebayerCpu::debayer10_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer10_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) { DECLARE_SRC_POINTERS(uint16_t) @@ -152,6 +178,7 @@ void DebayerCpu::debayer10_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer12_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) { DECLARE_SRC_POINTERS(uint16_t) @@ -163,6 +190,7 @@ void DebayerCpu::debayer12_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer12_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) { DECLARE_SRC_POINTERS(uint16_t) @@ -174,6 +202,7 @@ void DebayerCpu::debayer12_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer10P_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) { const int widthInBytes = window_.width * 5 / 4; @@ -199,6 +228,7 @@ void DebayerCpu::debayer10P_BGBG_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer10P_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) { const int widthInBytes = window_.width * 5 / 4; @@ -219,6 +249,7 @@ void DebayerCpu::debayer10P_GRGR_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer10P_GBGB_BGR888(uint8_t *dst, const uint8_t *src[]) { const int widthInBytes = window_.width * 5 / 4; @@ -239,6 +270,7 @@ void DebayerCpu::debayer10P_GBGB_BGR888(uint8_t *dst, const uint8_t *src[]) } } +template<bool addAlphaByte, bool ccmEnabled> void DebayerCpu::debayer10P_RGRG_BGR888(uint8_t *dst, const uint8_t *src[]) { const int widthInBytes = window_.width * 5 / 4; @@ -281,7 +313,12 @@ int DebayerCpu::getInputConfig(PixelFormat inputFormat, DebayerInputConfig &conf config.bpp = (bayerFormat.bitDepth + 7) & ~7; config.patternSize.width = 2; config.patternSize.height = 2; - config.outputFormats = std::vector<PixelFormat>({ formats::RGB888 }); + config.outputFormats = std::vector<PixelFormat>({ formats::RGB888, + formats::XRGB8888, + formats::ARGB8888, + formats::BGR888, + formats::XBGR8888, + formats::ABGR8888 }); return 0; } @@ -291,7 +328,12 @@ int DebayerCpu::getInputConfig(PixelFormat inputFormat, DebayerInputConfig &conf config.bpp = 10; config.patternSize.width = 4; /* 5 bytes per *4* pixels */ config.patternSize.height = 2; - config.outputFormats = std::vector<PixelFormat>({ formats::RGB888 }); + config.outputFormats = std::vector<PixelFormat>({ formats::RGB888, + formats::XRGB8888, + formats::ARGB8888, + formats::BGR888, + formats::XBGR8888, + formats::ABGR8888 }); return 0; } @@ -302,11 +344,17 @@ int DebayerCpu::getInputConfig(PixelFormat inputFormat, DebayerInputConfig &conf int DebayerCpu::getOutputConfig(PixelFormat outputFormat, DebayerOutputConfig &config) { - if (outputFormat == formats::RGB888) { + if (outputFormat == formats::RGB888 || outputFormat == formats::BGR888) { config.bpp = 24; return 0; } + if (outputFormat == formats::XRGB8888 || outputFormat == formats::ARGB8888 || + outputFormat == formats::XBGR8888 || outputFormat == formats::ABGR8888) { + config.bpp = 32; + return 0; + } + LOG(Debayer, Info) << "Unsupported output format " << outputFormat.toString(); return -EINVAL; @@ -338,30 +386,78 @@ int DebayerCpu::setupStandardBayerOrder(BayerFormat::Order order) return 0; } -/* \todo This ignores outputFormat since there is only 1 supported outputFormat - for now */ -int DebayerCpu::setDebayerFunctions(PixelFormat inputFormat, [[maybe_unused]] PixelFormat outputFormat) +#define SET_DEBAYER_METHODS(method0, method1) \ + debayer0_ = addAlphaByte \ + ? (ccmEnabled ? &DebayerCpu::method0<true, true> : &DebayerCpu::method0<true, false>) \ + : (ccmEnabled ? &DebayerCpu::method0<false, true> : &DebayerCpu::method0<false, false>); \ + debayer1_ = addAlphaByte \ + ? (ccmEnabled ? &DebayerCpu::method1<true, true> : &DebayerCpu::method1<true, false>) \ + : (ccmEnabled ? &DebayerCpu::method1<false, true> : &DebayerCpu::method1<false, false>); + +int DebayerCpu::setDebayerFunctions(PixelFormat inputFormat, + PixelFormat outputFormat, + bool ccmEnabled) { BayerFormat bayerFormat = BayerFormat::fromPixelFormat(inputFormat); + bool addAlphaByte = false; xShift_ = 0; + swapRedBlueGains_ = false; + + auto invalidFmt = []() -> int { + LOG(Debayer, Error) << "Unsupported input output format combination"; + return -EINVAL; + }; + + switch (outputFormat) { + case formats::XRGB8888: + case formats::ARGB8888: + addAlphaByte = true; + [[fallthrough]]; + case formats::RGB888: + break; + case formats::XBGR8888: + case formats::ABGR8888: + addAlphaByte = true; + [[fallthrough]]; + case formats::BGR888: + /* Swap R and B in bayer order to generate BGR888 instead of RGB888 */ + swapRedBlueGains_ = true; + + switch (bayerFormat.order) { + case BayerFormat::BGGR: + bayerFormat.order = BayerFormat::RGGB; + break; + case BayerFormat::GBRG: + bayerFormat.order = BayerFormat::GRBG; + break; + case BayerFormat::GRBG: + bayerFormat.order = BayerFormat::GBRG; + break; + case BayerFormat::RGGB: + bayerFormat.order = BayerFormat::BGGR; + break; + default: + return invalidFmt(); + } + break; + default: + return invalidFmt(); + } if ((bayerFormat.bitDepth == 8 || bayerFormat.bitDepth == 10 || bayerFormat.bitDepth == 12) && bayerFormat.packing == BayerFormat::Packing::None && isStandardBayerOrder(bayerFormat.order)) { switch (bayerFormat.bitDepth) { case 8: - debayer0_ = &DebayerCpu::debayer8_BGBG_BGR888; - debayer1_ = &DebayerCpu::debayer8_GRGR_BGR888; + SET_DEBAYER_METHODS(debayer8_BGBG_BGR888, debayer8_GRGR_BGR888) break; case 10: - debayer0_ = &DebayerCpu::debayer10_BGBG_BGR888; - debayer1_ = &DebayerCpu::debayer10_GRGR_BGR888; + SET_DEBAYER_METHODS(debayer10_BGBG_BGR888, debayer10_GRGR_BGR888) break; case 12: - debayer0_ = &DebayerCpu::debayer12_BGBG_BGR888; - debayer1_ = &DebayerCpu::debayer12_GRGR_BGR888; + SET_DEBAYER_METHODS(debayer12_BGBG_BGR888, debayer12_GRGR_BGR888) break; } setupStandardBayerOrder(bayerFormat.order); @@ -372,32 +468,28 @@ int DebayerCpu::setDebayerFunctions(PixelFormat inputFormat, [[maybe_unused]] Pi bayerFormat.packing == BayerFormat::Packing::CSI2) { switch (bayerFormat.order) { case BayerFormat::BGGR: - debayer0_ = &DebayerCpu::debayer10P_BGBG_BGR888; - debayer1_ = &DebayerCpu::debayer10P_GRGR_BGR888; + SET_DEBAYER_METHODS(debayer10P_BGBG_BGR888, debayer10P_GRGR_BGR888) return 0; case BayerFormat::GBRG: - debayer0_ = &DebayerCpu::debayer10P_GBGB_BGR888; - debayer1_ = &DebayerCpu::debayer10P_RGRG_BGR888; + SET_DEBAYER_METHODS(debayer10P_GBGB_BGR888, debayer10P_RGRG_BGR888) return 0; case BayerFormat::GRBG: - debayer0_ = &DebayerCpu::debayer10P_GRGR_BGR888; - debayer1_ = &DebayerCpu::debayer10P_BGBG_BGR888; + SET_DEBAYER_METHODS(debayer10P_GRGR_BGR888, debayer10P_BGBG_BGR888) return 0; case BayerFormat::RGGB: - debayer0_ = &DebayerCpu::debayer10P_RGRG_BGR888; - debayer1_ = &DebayerCpu::debayer10P_GBGB_BGR888; + SET_DEBAYER_METHODS(debayer10P_RGRG_BGR888, debayer10P_GBGB_BGR888) return 0; default: break; } } - LOG(Debayer, Error) << "Unsupported input output format combination"; - return -EINVAL; + return invalidFmt(); } int DebayerCpu::configure(const StreamConfiguration &inputCfg, - const std::vector<std::reference_wrapper<StreamConfiguration>> &outputCfgs) + const std::vector<std::reference_wrapper<StreamConfiguration>> &outputCfgs, + bool ccmEnabled) { if (getInputConfig(inputCfg.pixelFormat, inputConfig_) != 0) return -EINVAL; @@ -436,7 +528,10 @@ int DebayerCpu::configure(const StreamConfiguration &inputCfg, return -EINVAL; } - if (setDebayerFunctions(inputCfg.pixelFormat, outputCfg.pixelFormat) != 0) + int ret = setDebayerFunctions(inputCfg.pixelFormat, + outputCfg.pixelFormat, + ccmEnabled); + if (ret != 0) return -EINVAL; window_.x = ((inputCfg.size.width - outputCfg.size.width) / 2) & @@ -453,13 +548,10 @@ int DebayerCpu::configure(const StreamConfiguration &inputCfg, lineBufferPadding_ = inputConfig_.patternSize.width * inputConfig_.bpp / 8; lineBufferLength_ = window_.width * inputConfig_.bpp / 8 + 2 * lineBufferPadding_; - for (unsigned int i = 0; - i < (inputConfig_.patternSize.height + 1) && enableInputMemcpy_; - i++) { - free(lineBuffers_[i]); - lineBuffers_[i] = (uint8_t *)malloc(lineBufferLength_); - if (!lineBuffers_[i]) - return -ENOMEM; + + if (enableInputMemcpy_) { + for (unsigned int i = 0; i <= inputConfig_.patternSize.height; i++) + lineBuffers_[i].resize(lineBufferLength_); } measuredFrames_ = 0; @@ -514,9 +606,10 @@ void DebayerCpu::setupInputMemcpy(const uint8_t *linePointers[]) return; for (unsigned int i = 0; i < patternHeight; i++) { - memcpy(lineBuffers_[i], linePointers[i + 1] - lineBufferPadding_, + memcpy(lineBuffers_[i].data(), + linePointers[i + 1] - lineBufferPadding_, lineBufferLength_); - linePointers[i + 1] = lineBuffers_[i] + lineBufferPadding_; + linePointers[i + 1] = lineBuffers_[i].data() + lineBufferPadding_; } /* Point lineBufferIndex_ to first unused lineBuffer */ @@ -541,9 +634,10 @@ void DebayerCpu::memcpyNextLine(const uint8_t *linePointers[]) if (!enableInputMemcpy_) return; - memcpy(lineBuffers_[lineBufferIndex_], linePointers[patternHeight] - lineBufferPadding_, + memcpy(lineBuffers_[lineBufferIndex_].data(), + linePointers[patternHeight] - lineBufferPadding_, lineBufferLength_); - linePointers[patternHeight] = lineBuffers_[lineBufferIndex_] + lineBufferPadding_; + linePointers[patternHeight] = lineBuffers_[lineBufferIndex_].data() + lineBufferPadding_; lineBufferIndex_ = (lineBufferIndex_ + 1) % (patternHeight + 1); } @@ -652,13 +746,17 @@ void DebayerCpu::process4(const uint8_t *src, uint8_t *dst) } } -static inline int64_t timeDiff(timespec &after, timespec &before) +namespace { + +inline int64_t timeDiff(timespec &after, timespec &before) { return (after.tv_sec - before.tv_sec) * 1000000000LL + (int64_t)after.tv_nsec - (int64_t)before.tv_nsec; } -void DebayerCpu::process(FrameBuffer *input, FrameBuffer *output, DebayerParams params) +} /* namespace */ + +void DebayerCpu::process(uint32_t frame, FrameBuffer *input, FrameBuffer *output, DebayerParams params) { timespec frameStartTime; @@ -667,29 +765,31 @@ void DebayerCpu::process(FrameBuffer *input, FrameBuffer *output, DebayerParams clock_gettime(CLOCK_MONOTONIC_RAW, &frameStartTime); } - /* Apply DebayerParams */ - if (params.gamma != gammaCorrection_) { - for (unsigned int i = 0; i < kGammaLookupSize; i++) - gamma_[i] = UINT8_MAX * powf(i / (kGammaLookupSize - 1.0), params.gamma); - - gammaCorrection_ = params.gamma; - } - - for (unsigned int i = 0; i < kRGBLookupSize; i++) { - constexpr unsigned int div = - kRGBLookupSize * DebayerParams::kGain10 / kGammaLookupSize; - unsigned int idx; - - /* Apply gamma after gain! */ - idx = std::min({ i * params.gainR / div, (kGammaLookupSize - 1) }); - red_[i] = gamma_[idx]; - - idx = std::min({ i * params.gainG / div, (kGammaLookupSize - 1) }); - green_[i] = gamma_[idx]; - - idx = std::min({ i * params.gainB / div, (kGammaLookupSize - 1) }); - blue_[i] = gamma_[idx]; + std::vector<DmaSyncer> dmaSyncers; + for (const FrameBuffer::Plane &plane : input->planes()) + dmaSyncers.emplace_back(plane.fd, DmaSyncer::SyncType::Read); + + for (const FrameBuffer::Plane &plane : output->planes()) + dmaSyncers.emplace_back(plane.fd, DmaSyncer::SyncType::Write); + + green_ = params.green; + greenCcm_ = params.greenCcm; + if (swapRedBlueGains_) { + red_ = params.blue; + blue_ = params.red; + redCcm_ = params.blueCcm; + blueCcm_ = params.redCcm; + for (unsigned int i = 0; i < 256; i++) { + std::swap(redCcm_[i].r, redCcm_[i].b); + std::swap(blueCcm_[i].r, blueCcm_[i].b); + } + } else { + red_ = params.red; + blue_ = params.blue; + redCcm_ = params.redCcm; + blueCcm_ = params.blueCcm; } + gammaLut_ = params.gammaLut; /* Copy metadata from the input buffer */ FrameMetadata &metadata = output->_d()->metadata(); @@ -714,6 +814,8 @@ void DebayerCpu::process(FrameBuffer *input, FrameBuffer *output, DebayerParams metadata.planes()[0].bytesused = out.planes()[0].size(); + dmaSyncers.clear(); + /* Measure before emitting signals */ if (measuredFrames_ < DebayerCpu::kLastFrameToMeasure && ++measuredFrames_ > DebayerCpu::kFramesToSkip) { @@ -731,7 +833,12 @@ void DebayerCpu::process(FrameBuffer *input, FrameBuffer *output, DebayerParams } } - stats_->finishFrame(); + /* + * Buffer ids are currently not used, so pass zeros as its parameter. + * + * \todo Pass real bufferId once stats buffer passing is changed. + */ + stats_->finishFrame(frame, 0); outputBufferReady.emit(output); inputBufferReady.emit(input); } |