Merge pull request #27194 from asmorkalov:as/ipp_transpose

asmorkalov · web-flow · commit 6ef574639152 · 2025-04-10T21:04:46.000+03:00
Migrated IPP impl for flip and transpose to HAL
diff --git a/3rdparty/ipphal/CMakeLists.txt b/3rdparty/ipphal/CMakeLists.txt
@@ -12,13 +12,23 @@ add_library(ipphal STATIC
     "${CMAKE_CURRENT_SOURCE_DIR}/src/minmax_ipp.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/norm_ipp.cpp"
     "${CMAKE_CURRENT_SOURCE_DIR}/src/cart_polar_ipp.cpp"
+    "${CMAKE_CURRENT_SOURCE_DIR}/src/transforms_ipp.cpp"
 )
 
+#TODO: HAVE_IPP_ICV and HAVE_IPP_IW added as private macro till OpenCV itself is
+#      source of IPP and public definitions lead to redefinition warning
+#      The macro should be redefined as PUBLIC when IPP part is removed from core
+#      to make HAL the source of IPP integration
 if(HAVE_IPP_ICV)
   target_compile_definitions(ipphal PRIVATE HAVE_IPP_ICV)
 endif()
 
+if(HAVE_IPP_IW)
+  target_compile_definitions(ipphal PRIVATE HAVE_IPP_IW)
+endif()
+
 target_include_directories(ipphal PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/include")
+ocv_warnings_disable(CMAKE_CXX_FLAGS -Wno-suggest-override)
 
 target_include_directories(ipphal PRIVATE
   "${CMAKE_CURRENT_SOURCE_DIR}/src"
diff --git a/3rdparty/ipphal/include/ipp_hal_core.hpp b/3rdparty/ipphal/include/ipp_hal_core.hpp
@@ -42,4 +42,18 @@ int ipp_hal_polarToCart64f(const double* mag, const double* angle, double* x, do
 #undef cv_hal_polarToCart64f
 #define cv_hal_polarToCart64f ipp_hal_polarToCart64f
 
+#ifdef HAVE_IPP_IW
+int ipp_hal_flip(int src_type, const uchar* src_data, size_t src_step, int src_width, int src_height,
+                 uchar* dst_data, size_t dst_step, int flip_mode);
+
+#undef cv_hal_flip
+#define cv_hal_flip ipp_hal_flip
+#endif
+
+int ipp_hal_transpose2d(const uchar* src_data, size_t src_step, uchar* dst_data, size_t dst_step, int src_width,
+                        int src_height, int element_size);
+
+#undef cv_hal_transpose2d
+#define cv_hal_transpose2d ipp_hal_transpose2d
+
 #endif
diff --git a/3rdparty/ipphal/src/mean_ipp.cpp b/3rdparty/ipphal/src/mean_ipp.cpp
@@ -1,6 +1,6 @@
 #include "ipp_hal_core.hpp"
 
-#include <opencv2/core/core.hpp>
+#include <opencv2/core.hpp>
 #include <opencv2/core/base.hpp>
 
 #if IPP_VERSION_X100 >= 700
diff --git a/3rdparty/ipphal/src/minmax_ipp.cpp b/3rdparty/ipphal/src/minmax_ipp.cpp
@@ -1,6 +1,6 @@
 #include "ipp_hal_core.hpp"
 
-#include <opencv2/core/core.hpp>
+#include <opencv2/core.hpp>
 #include <opencv2/core/base.hpp>
 
 #define IPP_DISABLE_MINMAXIDX_MANY_ROWS 1  // see Core_MinMaxIdx.rows_overflow test
diff --git a/3rdparty/ipphal/src/norm_ipp.cpp b/3rdparty/ipphal/src/norm_ipp.cpp
@@ -1,6 +1,6 @@
 #include "ipp_hal_core.hpp"
 
-#include <opencv2/core/core.hpp>
+#include <opencv2/core.hpp>
 #include <opencv2/core/base.hpp>
 
 #if IPP_VERSION_X100 >= 700
diff --git a/3rdparty/ipphal/src/transforms_ipp.cpp b/3rdparty/ipphal/src/transforms_ipp.cpp
@@ -0,0 +1,142 @@
+#include "ipp_hal_core.hpp"
+
+#include <opencv2/core.hpp>
+#include <opencv2/core/base.hpp>
+#ifdef HAVE_IPP_IW
+#include "iw++/iw.hpp"
+#endif
+
+// HACK: Should be removed, when IPP management moved to HAL
+namespace cv
+{
+    namespace ipp
+    {
+        unsigned long long getIppTopFeatures(); // Returns top major enabled IPP feature flag
+    }
+}
+
+//bool ipp_transpose( Mat &src, Mat &dst )
+int ipp_hal_transpose2d(const uchar* src_data, size_t src_step, uchar* dst_data, size_t dst_step, int src_width,
+                        int src_height, int element_size)
+{
+    typedef IppStatus (CV_STDCALL * IppiTranspose)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize);
+    typedef IppStatus (CV_STDCALL * IppiTransposeI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize);
+    IppiTranspose ippiTranspose = nullptr;
+    IppiTransposeI ippiTranspose_I = nullptr;
+
+    if (dst_data == src_data && src_width == src_height)
+    {
+        CV_SUPPRESS_DEPRECATED_START
+        ippiTranspose_I =
+            element_size == 1*sizeof(char) ? (IppiTransposeI)ippiTranspose_8u_C1IR :
+            element_size == 3*sizeof(char) ? (IppiTransposeI)ippiTranspose_8u_C3IR :
+            element_size == 1*sizeof(short) ? (IppiTransposeI)ippiTranspose_16u_C1IR :
+            element_size == 4*sizeof(char) ? (IppiTransposeI)ippiTranspose_8u_C4IR :
+            element_size == 3*sizeof(short) ? (IppiTransposeI)ippiTranspose_16u_C3IR :
+            element_size == 4*sizeof(short) ? (IppiTransposeI)ippiTranspose_16u_C4IR :
+            element_size == 3*sizeof(int) ? (IppiTransposeI)ippiTranspose_32s_C3IR :
+            element_size == 4*sizeof(int) ? (IppiTransposeI)ippiTranspose_32s_C4IR : 0;
+        CV_SUPPRESS_DEPRECATED_END
+    }
+    else
+    {
+        ippiTranspose =
+            element_size == 1*sizeof(char) ? (IppiTranspose)ippiTranspose_8u_C1R :
+            element_size == 3*sizeof(char) ? (IppiTranspose)ippiTranspose_8u_C3R :
+            element_size == 4*sizeof(char) ? (IppiTranspose)ippiTranspose_8u_C4R :
+            element_size == 1*sizeof(short) ? (IppiTranspose)ippiTranspose_16u_C1R :
+            element_size == 3*sizeof(short) ? (IppiTranspose)ippiTranspose_16u_C3R :
+            element_size == 4*sizeof(short) ? (IppiTranspose)ippiTranspose_16u_C4R :
+            element_size == 1*sizeof(int) ? (IppiTranspose)ippiTranspose_32s_C1R :
+            element_size == 3*sizeof(int) ? (IppiTranspose)ippiTranspose_32s_C3R :
+            element_size == 4*sizeof(int) ? (IppiTranspose)ippiTranspose_32s_C4R : 0;
+    }
+
+    IppiSize roiSize = { src_width, src_height };
+    if (ippiTranspose != 0)
+    {
+        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose, src_data, (int)src_step, dst_data, (int)dst_step, roiSize) >= 0)
+            return CV_HAL_ERROR_OK;
+    }
+    else if (ippiTranspose_I != 0)
+    {
+        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose_I, dst_data, (int)dst_step, roiSize) >= 0)
+            return CV_HAL_ERROR_OK;
+    }
+    return CV_HAL_ERROR_NOT_IMPLEMENTED;
+}
+
+#ifdef HAVE_IPP_IW
+
+static inline IppDataType ippiGetDataType(int depth)
+{
+    depth = CV_MAT_DEPTH(depth);
+    return depth == CV_8U ? ipp8u :
+    depth == CV_8S ? ipp8s :
+    depth == CV_16U ? ipp16u :
+    depth == CV_16S ? ipp16s :
+    depth == CV_32S ? ipp32s :
+    depth == CV_32F ? ipp32f :
+    depth == CV_64F ? ipp64f :
+    (IppDataType)-1;
+}
+
+static inline ::ipp::IwiImage ippiGetImage(int src_type, const uchar* src_data, size_t src_step, int src_width, int src_height)
+{
+    ::ipp::IwiImage dst;
+    ::ipp::IwiBorderSize inMemBorder;
+//     if(src.isSubmatrix()) // already have physical border
+//     {
+//         cv::Size  origSize;
+//         cv::Point offset;
+//         src.locateROI(origSize, offset);
+//
+//         inMemBorder.left   = (IwSize)offset.x;
+//         inMemBorder.top    = (IwSize)offset.y;
+//         inMemBorder.right  = (IwSize)(origSize.width - src.cols - offset.x);
+//         inMemBorder.bottom = (IwSize)(origSize.height - src.rows - offset.y);
+//     }
+
+    dst.Init({src_width, src_height}, ippiGetDataType(CV_MAT_DEPTH(src_type)),
+             CV_MAT_CN(src_type), inMemBorder, (void*)src_data, src_step);
+
+    return dst;
+}
+
+int ipp_hal_flip(int src_type, const uchar* src_data, size_t src_step, int src_width, int src_height,
+                 uchar* dst_data, size_t dst_step, int flip_mode)
+
+{
+    int64_t total = src_step*src_height*CV_ELEM_SIZE(src_type);
+    // Details: https://github.com/opencv/opencv/issues/12943
+    if (flip_mode <= 0 /* swap rows */
+        && total > 0 && total >= CV_BIG_INT(0x80000000)/*2Gb*/
+        && cv::ipp::getIppTopFeatures() != ippCPUID_SSE42)
+    {
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+    }
+
+    IppiAxis ippMode;
+    if(flip_mode < 0)
+        ippMode = ippAxsBoth;
+    else if(flip_mode == 0)
+        ippMode = ippAxsHorizontal;
+    else
+        ippMode = ippAxsVertical;
+
+    try
+    {
+        ::ipp::IwiImage iwSrc = ippiGetImage(src_type, src_data, src_step, src_width, src_height);
+        ::ipp::IwiImage iwDst = ippiGetImage(src_type, dst_data, dst_step, src_width, src_height);
+
+        CV_INSTRUMENT_FUN_IPP(::ipp::iwiMirror, iwSrc, iwDst, ippMode);
+    }
+    catch(const ::ipp::IwException &)
+    {
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+    }
+
+    return CV_HAL_ERROR_OK;
+}
+
+#endif
diff --git a/modules/core/src/matrix_transform.cpp b/modules/core/src/matrix_transform.cpp
@@ -173,74 +173,6 @@ static bool ocl_transpose( InputArray _src, OutputArray _dst )
 
 #endif
 
-#ifdef HAVE_IPP
-static bool ipp_transpose( Mat &src, Mat &dst )
-{
-    CV_INSTRUMENT_REGION_IPP();
-
-    int type = src.type();
-    typedef IppStatus (CV_STDCALL * IppiTranspose)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize);
-    typedef IppStatus (CV_STDCALL * IppiTransposeI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize);
-    IppiTranspose ippiTranspose = 0;
-    IppiTransposeI ippiTranspose_I = 0;
-
-    if (dst.data == src.data && dst.cols == dst.rows)
-    {
-        CV_SUPPRESS_DEPRECATED_START
-        ippiTranspose_I =
-            type == CV_8UC1 ? (IppiTransposeI)ippiTranspose_8u_C1IR :
-            type == CV_8UC3 ? (IppiTransposeI)ippiTranspose_8u_C3IR :
-            type == CV_8UC4 ? (IppiTransposeI)ippiTranspose_8u_C4IR :
-            type == CV_16UC1 ? (IppiTransposeI)ippiTranspose_16u_C1IR :
-            type == CV_16UC3 ? (IppiTransposeI)ippiTranspose_16u_C3IR :
-            type == CV_16UC4 ? (IppiTransposeI)ippiTranspose_16u_C4IR :
-            type == CV_16SC1 ? (IppiTransposeI)ippiTranspose_16s_C1IR :
-            type == CV_16SC3 ? (IppiTransposeI)ippiTranspose_16s_C3IR :
-            type == CV_16SC4 ? (IppiTransposeI)ippiTranspose_16s_C4IR :
-            type == CV_32SC1 ? (IppiTransposeI)ippiTranspose_32s_C1IR :
-            type == CV_32SC3 ? (IppiTransposeI)ippiTranspose_32s_C3IR :
-            type == CV_32SC4 ? (IppiTransposeI)ippiTranspose_32s_C4IR :
-            type == CV_32FC1 ? (IppiTransposeI)ippiTranspose_32f_C1IR :
-            type == CV_32FC3 ? (IppiTransposeI)ippiTranspose_32f_C3IR :
-            type == CV_32FC4 ? (IppiTransposeI)ippiTranspose_32f_C4IR : 0;
-        CV_SUPPRESS_DEPRECATED_END
-    }
-    else
-    {
-        ippiTranspose =
-            type == CV_8UC1 ? (IppiTranspose)ippiTranspose_8u_C1R :
-            type == CV_8UC3 ? (IppiTranspose)ippiTranspose_8u_C3R :
-            type == CV_8UC4 ? (IppiTranspose)ippiTranspose_8u_C4R :
-            type == CV_16UC1 ? (IppiTranspose)ippiTranspose_16u_C1R :
-            type == CV_16UC3 ? (IppiTranspose)ippiTranspose_16u_C3R :
-            type == CV_16UC4 ? (IppiTranspose)ippiTranspose_16u_C4R :
-            type == CV_16SC1 ? (IppiTranspose)ippiTranspose_16s_C1R :
-            type == CV_16SC3 ? (IppiTranspose)ippiTranspose_16s_C3R :
-            type == CV_16SC4 ? (IppiTranspose)ippiTranspose_16s_C4R :
-            type == CV_32SC1 ? (IppiTranspose)ippiTranspose_32s_C1R :
-            type == CV_32SC3 ? (IppiTranspose)ippiTranspose_32s_C3R :
-            type == CV_32SC4 ? (IppiTranspose)ippiTranspose_32s_C4R :
-            type == CV_32FC1 ? (IppiTranspose)ippiTranspose_32f_C1R :
-            type == CV_32FC3 ? (IppiTranspose)ippiTranspose_32f_C3R :
-            type == CV_32FC4 ? (IppiTranspose)ippiTranspose_32f_C4R : 0;
-    }
-
-    IppiSize roiSize = { src.cols, src.rows };
-    if (ippiTranspose != 0)
-    {
-        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose, src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, roiSize) >= 0)
-            return true;
-    }
-    else if (ippiTranspose_I != 0)
-    {
-        if (CV_INSTRUMENT_FUN_IPP(ippiTranspose_I, dst.ptr(), (int)dst.step, roiSize) >= 0)
-            return true;
-    }
-    return false;
-}
-#endif
-
-
 void transpose( InputArray _src, OutputArray _dst )
 {
     CV_INSTRUMENT_REGION();
@@ -271,8 +203,6 @@ void transpose( InputArray _src, OutputArray _dst )
 
     CALL_HAL(transpose2d, cv_hal_transpose2d, src.data, src.step, dst.data, dst.step, src.cols, src.rows, esz);
 
-    CV_IPP_RUN_FAST(ipp_transpose(src, dst))
-
     if( dst.data == src.data )
     {
         TransposeInplaceFunc func = transposeInplaceTab[esz];
@@ -735,48 +665,6 @@ static bool ocl_flip(InputArray _src, OutputArray _dst, int flipCode )
 
 #endif
 
-#if defined HAVE_IPP
-static bool ipp_flip(Mat &src, Mat &dst, int flip_mode)
-{
-#ifdef HAVE_IPP_IW
-    CV_INSTRUMENT_REGION_IPP();
-
-    // Details: https://github.com/opencv/opencv/issues/12943
-    if (flip_mode <= 0 /* swap rows */
-        && cv::ipp::getIppTopFeatures() != ippCPUID_SSE42
-        && (int64_t)(src.total()) * src.elemSize() >= CV_BIG_INT(0x80000000)/*2Gb*/
-    )
-        return false;
-
-    IppiAxis ippMode;
-    if(flip_mode < 0)
-        ippMode = ippAxsBoth;
-    else if(flip_mode == 0)
-        ippMode = ippAxsHorizontal;
-    else
-        ippMode = ippAxsVertical;
-
-    try
-    {
-        ::ipp::IwiImage iwSrc = ippiGetImage(src);
-        ::ipp::IwiImage iwDst = ippiGetImage(dst);
-
-        CV_INSTRUMENT_FUN_IPP(::ipp::iwiMirror, iwSrc, iwDst, ippMode);
-    }
-    catch(const ::ipp::IwException &)
-    {
-        return false;
-    }
-
-    return true;
-#else
-    CV_UNUSED(src); CV_UNUSED(dst); CV_UNUSED(flip_mode);
-    return false;
-#endif
-}
-#endif
-
-
 void flip( InputArray _src, OutputArray _dst, int flip_mode )
 {
     CV_INSTRUMENT_REGION();
@@ -808,8 +696,6 @@ void flip( InputArray _src, OutputArray _dst, int flip_mode )
     CALL_HAL(flip, cv_hal_flip, type, src.ptr(), src.step, src.cols, src.rows,
              dst.ptr(), dst.step, flip_mode);
 
-    CV_IPP_RUN_FAST(ipp_flip(src, dst, flip_mode));
-
     size_t esz = CV_ELEM_SIZE(type);
 
     if( flip_mode <= 0 )
