[PATCH 09/16] libcamera: software_isp: Avoid copying particular color tables

[Milan Zamazal]

The color lookup tables in software CPU debayering are currently copied
from the provided parameters.  The preceding work avoided copying the
parameters when passed to debayering, this patch avoids the copying
inside debayering.

Signed-off-by: Milan Zamazal <mzamazal at redhat.com>
---
 src/libcamera/software_isp/debayer_cpu.cpp | 60 +++++++++++-----------
 src/libcamera/software_isp/debayer_cpu.h   |  7 +--
 2 files changed, 34 insertions(+), 33 deletions(-)

diff --git a/src/libcamera/software_isp/debayer_cpu.cpp b/src/libcamera/software_isp/debayer_cpu.cpp
index af08ab10..d3831474 100644
--- a/src/libcamera/software_isp/debayer_cpu.cpp
+++ b/src/libcamera/software_isp/debayer_cpu.cpp
@@ -51,7 +51,8 @@ DebayerCpu::DebayerCpu(std::unique_ptr<SwStatsCpu> stats,
 
 	/* Initialize color lookup tables */
 	for (unsigned int i = 0; i < DebayerParams::kRGBLookupSize; i++)
-		red_[i] = green_[i] = blue_[i] = i;
+		initColorLookupTable_[i] = i;
+	red_ = green_ = blue_ = &initColorLookupTable_;
 
 	paramsBuffers_ = std::map<unsigned int, DebayerParams *>();
 
@@ -80,12 +81,12 @@ DebayerCpu::~DebayerCpu() = default;
  * 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))]; \
-	if constexpr (addAlphaByte)                                                           \
-		*dst++ = 255;                                                                 \
+#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))]; \
+	if constexpr (addAlphaByte)                                                              \
+		*dst++ = 255;                                                                    \
 	x++;
 
 /*
@@ -93,12 +94,12 @@ DebayerCpu::~DebayerCpu() = default;
  * 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))]; \
-	if constexpr (addAlphaByte)                               \
-		*dst++ = 255;                                     \
+#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))]; \
+	if constexpr (addAlphaByte)                                  \
+		*dst++ = 255;                                        \
 	x++;
 
 /*
@@ -106,12 +107,12 @@ DebayerCpu::~DebayerCpu() = default;
  * 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))];          \
-	if constexpr (addAlphaByte)                                \
-		*dst++ = 255;                                      \
+#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))];          \
+	if constexpr (addAlphaByte)                                   \
+		*dst++ = 255;                                         \
 	x++;
 
 /*
@@ -119,12 +120,12 @@ DebayerCpu::~DebayerCpu() = default;
  * 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)];                                                        \
-	if constexpr (addAlphaByte)                                                            \
-		*dst++ = 255;                                                                  \
+#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)];                                                        \
+	if constexpr (addAlphaByte)                                                               \
+		*dst++ = 255;                                                                     \
 	x++;
 
 template<bool addAlphaByte>
@@ -752,12 +753,10 @@ void DebayerCpu::process(uint32_t frame,
 	}
 
 	DebayerParams *params = paramsBuffers_.at(paramsBufferId);
-	releaseIspParams.emit(paramsBufferId);
 
-	/* \todo Avoid copying here. */
-	green_ = params->green;
-	red_ = swapRedBlueGains_ ? params->blue : params->red;
-	blue_ = swapRedBlueGains_ ? params->red : params->blue;
+	green_ = &params->green;
+	red_ = swapRedBlueGains_ ? &params->blue : &params->red;
+	blue_ = swapRedBlueGains_ ? &params->red : &params->blue;
 
 	/* Copy metadata from the input buffer */
 	FrameMetadata &metadata = output->_d()->metadata();
@@ -779,6 +778,7 @@ void DebayerCpu::process(uint32_t frame,
 		process2(in.planes()[0].data(), out.planes()[0].data());
 	else
 		process4(in.planes()[0].data(), out.planes()[0].data());
+	releaseIspParams.emit(paramsBufferId);
 
 	metadata.planes()[0].bytesused = out.planes()[0].size();
 
diff --git a/src/libcamera/software_isp/debayer_cpu.h b/src/libcamera/software_isp/debayer_cpu.h
index 324fc6c0..f25520be 100644
--- a/src/libcamera/software_isp/debayer_cpu.h
+++ b/src/libcamera/software_isp/debayer_cpu.h
@@ -139,9 +139,10 @@ private:
 	/* Max. supported Bayer pattern height is 4, debayering this requires 5 lines */
 	static constexpr unsigned int kMaxLineBuffers = 5;
 
-	DebayerParams::ColorLookupTable red_;
-	DebayerParams::ColorLookupTable green_;
-	DebayerParams::ColorLookupTable blue_;
+	DebayerParams::ColorLookupTable initColorLookupTable_;
+	DebayerParams::ColorLookupTable *red_;
+	DebayerParams::ColorLookupTable *green_;
+	DebayerParams::ColorLookupTable *blue_;
 	debayerFn debayer0_;
 	debayerFn debayer1_;
 	debayerFn debayer2_;
-- 
2.44.1