From f643964c6863dcbcdc83029c611b76cf9224a49b Mon Sep 17 00:00:00 2001
From: Alexander Smorkalov <alexander.smorkalov@xperience.ai>
Date: Thu, 11 Apr 2024 15:11:02 +0300
Subject: [PATCH] Added HAL function for popular bit-exact branch of
 GaussianBlur with sigma=0.

---
 modules/imgproc/src/hal_replacement.hpp | 25 +++++++++++++++++++++-
 modules/imgproc/src/smooth.dispatch.cpp | 28 +++++++++++++++++++++++++
 2 files changed, 52 insertions(+), 1 deletion(-)

diff --git a/modules/imgproc/src/hal_replacement.hpp b/modules/imgproc/src/hal_replacement.hpp
index 5c6497bd80..6d1f2a4272 100644
--- a/modules/imgproc/src/hal_replacement.hpp
+++ b/modules/imgproc/src/hal_replacement.hpp
@@ -855,7 +855,7 @@ inline int hal_ni_boxFilter(const uchar* src_data, size_t src_step, uchar* dst_d
 //! @endcond
 
 /**
-   @brief Blurs an image using a Gaussian filter.
+   @brief Blurs an image using a generic Gaussian filter.
    @param src_data Source image data
    @param src_step Source image step
    @param dst_data Destination image data
@@ -880,6 +880,29 @@ inline int hal_ni_gaussianBlur(const uchar* src_data, size_t src_step, uchar* ds
 #define cv_hal_gaussianBlur hal_ni_gaussianBlur
 //! @endcond
 
+/**
+   @brief Blurs an image using a symmetric Gaussian filter with square kernel and sigma=0.
+   @param src_data Source image data
+   @param src_step Source image step
+   @param dst_data Destination image data
+   @param dst_step Destination image step
+   @param width Source image width
+   @param height Source image height
+   @param depth Depth of source and destination image
+   @param cn Number of channels
+   @param margin_left Left margins for source image
+   @param margin_top Top margins for source image
+   @param margin_right Right margins for source image
+   @param margin_bottom Bottom margins for source image
+   @param ksize Width of kernel
+   @param border_type Border type
+*/
+inline int hal_ni_gaussianBlurBinomial(const uchar* src_data, size_t src_step, uchar* dst_data, size_t dst_step, int width, int height, int depth, int cn, size_t margin_left, size_t margin_top, size_t margin_right, size_t margin_bottom, size_t ksize, int border_type) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
+
+//! @cond IGNORED
+#define cv_hal_gaussianBlurBinomial hal_ni_gaussianBlurBinomial
+//! @endcond
+
 /**
    @brief Computes Sobel derivatives
    @param src_depth Depth of source image
diff --git a/modules/imgproc/src/smooth.dispatch.cpp b/modules/imgproc/src/smooth.dispatch.cpp
index 26566c7e03..d0f50a73bb 100644
--- a/modules/imgproc/src/smooth.dispatch.cpp
+++ b/modules/imgproc/src/smooth.dispatch.cpp
@@ -683,8 +683,22 @@ void GaussianBlur(InputArray _src, OutputArray _dst, Size ksize,
 
             if (src.data == dst.data)
                 src = src.clone();
+
+            if ((sigma1 == 0.0) && (sigma2 == 0.0) && (ksize.height == ksize.width))
+            {
+                Point ofs;
+                Size wsz(src.cols, src.rows);
+                Mat src2 = src;
+                if(!(borderType & BORDER_ISOLATED))
+                    src2.locateROI( wsz, ofs );
+
+                CALL_HAL(gaussianBlurBinomial, cv_hal_gaussianBlurBinomial, src2.ptr(), src2.step, dst.ptr(), dst.step, src2.cols, src2.rows, sdepth, cn,
+                         ofs.x, ofs.y, wsz.width - src2.cols - ofs.x,  wsz.height - src2.rows - ofs.y, ksize.width, borderType&~BORDER_ISOLATED);
+            }
+
             CV_CPU_DISPATCH(GaussianBlurFixedPoint, (src, dst, (const uint16_t*)&fkx[0], (int)fkx.size(), (const uint16_t*)&fky[0], (int)fky.size(), borderType),
                 CV_CPU_DISPATCH_MODES_ALL);
+
             return;
         }
     }
@@ -720,8 +734,22 @@ void GaussianBlur(InputArray _src, OutputArray _dst, Size ksize,
 
             if (src.data == dst.data)
                 src = src.clone();
+
+            if ((sigma1 == 0.0) && (sigma2 == 0.0) && (ksize.height == ksize.width))
+            {
+                Point ofs;
+                Size wsz(src.cols, src.rows);
+                Mat src2 = src;
+                if(!(borderType & BORDER_ISOLATED))
+                    src2.locateROI( wsz, ofs );
+
+                CALL_HAL(gaussianBlurBinomial, cv_hal_gaussianBlurBinomial, src2.ptr(), src2.step, dst.ptr(), dst.step, src2.cols, src2.rows, sdepth, cn,
+                         ofs.x, ofs.y, wsz.width - src2.cols - ofs.x,  wsz.height - src2.rows - ofs.y, ksize.width, borderType&~BORDER_ISOLATED);
+            }
+
             CV_CPU_DISPATCH(GaussianBlurFixedPoint, (src, dst, (const uint32_t*)&fkx[0], (int)fkx.size(), (const uint32_t*)&fky[0], (int)fky.size(), borderType),
                 CV_CPU_DISPATCH_MODES_ALL);
+
             return;
         }
     }