Modifications to support dynamic vector dispatch.

cmake/OpenCVCompilerOptions.cmake

@@ -140,15 +140,15 @@ if(CMAKE_COMPILER_IS_GNUCXX)
   # SSE3 and further should be disabled under MingW because it generates compiler errors
   if(NOT MINGW)
     if(ENABLE_AVX)
-      add_extra_compiler_option(-mavx)
+      ocv_check_flag_support(CXX "-mavx" _varname)
     endif()
 
     if(ENABLE_AVX2)
-      add_extra_compiler_option(-mavx2)
+      ocv_check_flag_support(CXX "-mavx2" _varname)
     endif()
 
     # GCC depresses SSEx instructions when -mavx is used. Instead, it generates new AVX instructions or AVX equivalence for all SSEx instructions when needed.
-    if(NOT OPENCV_EXTRA_CXX_FLAGS MATCHES "-m(avx|avx2)")
+    if(NOT OPENCV_EXTRA_CXX_FLAGS MATCHES "-mavx")
       if(ENABLE_SSE3)
         add_extra_compiler_option(-msse3)
       endif()
@@ -169,7 +169,7 @@ if(CMAKE_COMPILER_IS_GNUCXX)
 
   if(X86 OR X86_64)
     if(NOT APPLE AND CMAKE_SIZEOF_VOID_P EQUAL 4)
-      if(OPENCV_EXTRA_CXX_FLAGS MATCHES "-m(sse2|avx|avx2)")
+      if(OPENCV_EXTRA_CXX_FLAGS MATCHES "-m(sse2|avx)")
         add_extra_compiler_option(-mfpmath=sse)# !! important - be on the same wave with x64 compilers
       else()
         add_extra_compiler_option(-mfpmath=387)

cmake/OpenCVModule.cmake

@@ -526,6 +526,20 @@ macro(ocv_glob_module_sources)
     list(APPEND lib_srcs ${cl_kernels} "${CMAKE_CURRENT_BINARY_DIR}/opencl_kernels.cpp" "${CMAKE_CURRENT_BINARY_DIR}/opencl_kernels.hpp")
   endif()
 
+  if(ENABLE_AVX)
+    file(GLOB avx_srcs "src/avx/*.cpp")
+    foreach(src ${avx_srcs})
+      set_source_files_properties(${src} PROPERTIES COMPILE_FLAGS -mavx)
+    endforeach()
+  endif()
+
+  if(ENABLE_AVX2)
+    file(GLOB avx2_srcs "src/avx2/*.cpp")
+    foreach(src ${avx2_srcs})
+      set_source_files_properties(${src} PROPERTIES COMPILE_FLAGS -mavx2)
+    endforeach()
+  endif()
+
   source_group("Include" FILES ${lib_hdrs})
   source_group("Include\\detail" FILES ${lib_hdrs_detail})
 

modules/core/doc/utility_and_system_functions_and_macros.rst

@@ -317,6 +317,7 @@ Returns true if the specified feature is supported by the host hardware.
                         * ``CV_CPU_SSE4_2`` - SSE 4.2
                         * ``CV_CPU_POPCNT`` - POPCOUNT
                         * ``CV_CPU_AVX`` - AVX
+                        * ``CV_CPU_AVX2`` - AVX2
 
 The function returns true if the host hardware supports the specified feature. When user calls ``setUseOptimized(false)``, the subsequent calls to ``checkHardwareSupport()`` will return false until ``setUseOptimized(true)`` is called. This way user can dynamically switch on and off the optimized code in OpenCV.
 

modules/core/include/opencv2/core/core.hpp

@@ -284,6 +284,7 @@ CV_EXPORTS_W int64 getCPUTickCount();
   - CV_CPU_SSE4_2 - SSE 4.2
   - CV_CPU_POPCNT - POPCOUNT
   - CV_CPU_AVX - AVX
+  - CV_CPU_AVX2 - AVX2
 
   \note {Note that the function output is not static. Once you called cv::useOptimized(false),
   most of the hardware acceleration is disabled and thus the function will returns false,

modules/core/src/system.cpp

@@ -253,7 +253,6 @@ struct HWFeatures
             f.have[CV_CPU_AVX]    = (((cpuid_data[2] & (1<<28)) != 0)&&((cpuid_data[2] & (1<<27)) != 0));//OS uses XSAVE_XRSTORE and CPU support AVX
         }
 
-#if CV_AVX2
     #if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
         __cpuidex(cpuid_data, 7, 0);
     #elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
@@ -286,7 +285,6 @@ struct HWFeatures
         {
             f.have[CV_CPU_AVX2] = (cpuid_data[1] & (1<<5)) != 0;
         }
-#endif
 
         return f;
     }

modules/imgproc/src/avx/imgwarp_avx.cpp

+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                           License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "../precomp.hpp"
+#include "imgwarp_avx.hpp"
+
+#if CV_AVX
+int VResizeLinearVec_32f_avx(const uchar** _src, uchar* _dst, const uchar* _beta, int width )
+{
+    const float** src = (const float**)_src;
+    const float* beta = (const float*)_beta;
+    const float *S0 = src[0], *S1 = src[1];
+    float* dst = (float*)_dst;
+    int x = 0;
+
+    __m256 b0 = _mm256_set1_ps(beta[0]), b1 = _mm256_set1_ps(beta[1]);
+
+    if( (((size_t)S0|(size_t)S1)&31) == 0 )
+        for( ; x <= width - 16; x += 16 )
+        {
+            __m256 x0, x1, y0, y1;
+            x0 = _mm256_load_ps(S0 + x);
+            x1 = _mm256_load_ps(S0 + x + 8);
+            y0 = _mm256_load_ps(S1 + x);
+            y1 = _mm256_load_ps(S1 + x + 8);
+
+            x0 = _mm256_add_ps(_mm256_mul_ps(x0, b0), _mm256_mul_ps(y0, b1));
+            x1 = _mm256_add_ps(_mm256_mul_ps(x1, b0), _mm256_mul_ps(y1, b1));
+
+            _mm256_storeu_ps( dst + x, x0);
+            _mm256_storeu_ps( dst + x + 8, x1);
+        }
+    else
+        for( ; x <= width - 16; x += 16 )
+        {
+            __m256 x0, x1, y0, y1;
+            x0 = _mm256_loadu_ps(S0 + x);
+            x1 = _mm256_loadu_ps(S0 + x + 8);
+            y0 = _mm256_loadu_ps(S1 + x);
+            y1 = _mm256_loadu_ps(S1 + x + 8);
+
+            x0 = _mm256_add_ps(_mm256_mul_ps(x0, b0), _mm256_mul_ps(y0, b1));
+            x1 = _mm256_add_ps(_mm256_mul_ps(x1, b0), _mm256_mul_ps(y1, b1));
+
+            _mm256_storeu_ps( dst + x, x0);
+            _mm256_storeu_ps( dst + x + 8, x1);
+        }
+
+    return x;
+}
+
+int VResizeCubicVec_32f_avx(const uchar** _src, uchar* _dst, const uchar* _beta, int width )
+{
+    const float** src = (const float**)_src;
+    const float* beta = (const float*)_beta;
+    const float *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3];
+    float* dst = (float*)_dst;
+    int x = 0;
+    __m256 b0 = _mm256_set1_ps(beta[0]), b1 = _mm256_set1_ps(beta[1]),
+        b2 = _mm256_set1_ps(beta[2]), b3 = _mm256_set1_ps(beta[3]);
+
+    if( (((size_t)S0|(size_t)S1|(size_t)S2|(size_t)S3)&31) == 0 )
+        for( ; x <= width - 16; x += 16 )
+        {
+            __m256 x0, x1, y0, y1, s0, s1;
+            x0 = _mm256_load_ps(S0 + x);
+            x1 = _mm256_load_ps(S0 + x + 8);
+            y0 = _mm256_load_ps(S1 + x);
+            y1 = _mm256_load_ps(S1 + x + 8);
+
+            s0 = _mm256_mul_ps(x0, b0);
+            s1 = _mm256_mul_ps(x1, b0);
+            y0 = _mm256_mul_ps(y0, b1);
+            y1 = _mm256_mul_ps(y1, b1);
+            s0 = _mm256_add_ps(s0, y0);
+            s1 = _mm256_add_ps(s1, y1);
+
+            x0 = _mm256_load_ps(S2 + x);
+            x1 = _mm256_load_ps(S2 + x + 8);
+            y0 = _mm256_load_ps(S3 + x);
+            y1 = _mm256_load_ps(S3 + x + 8);
+
+            x0 = _mm256_mul_ps(x0, b2);
+            x1 = _mm256_mul_ps(x1, b2);
+            y0 = _mm256_mul_ps(y0, b3);
+            y1 = _mm256_mul_ps(y1, b3);
+            s0 = _mm256_add_ps(s0, x0);
+            s1 = _mm256_add_ps(s1, x1);
+            s0 = _mm256_add_ps(s0, y0);
+            s1 = _mm256_add_ps(s1, y1);
+
+            _mm256_storeu_ps( dst + x, s0);
+            _mm256_storeu_ps( dst + x + 8, s1);
+        }
+    else
+        for( ; x <= width - 16; x += 16 )
+        {
+            __m256 x0, x1, y0, y1, s0, s1;
+            x0 = _mm256_loadu_ps(S0 + x);
+            x1 = _mm256_loadu_ps(S0 + x + 8);
+            y0 = _mm256_loadu_ps(S1 + x);
+            y1 = _mm256_loadu_ps(S1 + x + 8);
+
+            s0 = _mm256_mul_ps(x0, b0);
+            s1 = _mm256_mul_ps(x1, b0);
+            y0 = _mm256_mul_ps(y0, b1);
+            y1 = _mm256_mul_ps(y1, b1);
+            s0 = _mm256_add_ps(s0, y0);
+            s1 = _mm256_add_ps(s1, y1);
+
+            x0 = _mm256_loadu_ps(S2 + x);
+            x1 = _mm256_loadu_ps(S2 + x + 8);
+            y0 = _mm256_loadu_ps(S3 + x);
+            y1 = _mm256_loadu_ps(S3 + x + 8);
+
+            x0 = _mm256_mul_ps(x0, b2);
+            x1 = _mm256_mul_ps(x1, b2);
+            y0 = _mm256_mul_ps(y0, b3);
+            y1 = _mm256_mul_ps(y1, b3);
+            s0 = _mm256_add_ps(s0, x0);
+            s1 = _mm256_add_ps(s1, x1);
+            s0 = _mm256_add_ps(s0, y0);
+            s1 = _mm256_add_ps(s1, y1);
+
+            _mm256_storeu_ps( dst + x, s0);
+            _mm256_storeu_ps( dst + x + 8, s1);
+        }
+
+    return x;
+}
+#else
+int VResizeLinearVec_32f_avx(const uchar**, uchar*, const uchar*, int ) { return 0; }
+
+int VResizeCubicVec_32f_avx(const uchar**, uchar*, const uchar*, int ) { return 0; }
+#endif
+
+/* End of file. */

modules/imgproc/src/avx/imgwarp_avx.hpp

+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                        Intel License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000, Intel Corporation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of Intel Corporation may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#ifndef _CV_IMGWARP_AVX_H_
+#define _CV_IMGWARP_AVX_H_
+
+int VResizeLinearVec_32f_avx(const uchar** _src, uchar* _dst, const uchar* _beta, int width );
+
+int VResizeCubicVec_32f_avx(const uchar** _src, uchar* _dst, const uchar* _beta, int width );
+
+#endif
+
+/* End of file. */

modules/imgproc/src/avx2/imgwarp_avx2.hpp

+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                        Intel License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000, Intel Corporation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of Intel Corporation may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#ifndef _CV_IMGWARP_AVX2_H_
+#define _CV_IMGWARP_AVX2_H_
+
+int VResizeLinearVec_32s8u_avx2(const uchar** _src, uchar* dst, const uchar* _beta, int width );
+
+template<int shiftval>
+int VResizeLinearVec_32f16_avx2(const uchar** _src, uchar* _dst, const uchar* _beta, int width );
+
+int VResizeCubicVec_32s8u_avx2(const uchar** _src, uchar* dst, const uchar* _beta, int width );
+
+template<int shiftval>
+int VResizeCubicVec_32f16_avx2(const uchar** _src, uchar* _dst, const uchar* _beta, int width );
+
+#endif
+
+/* End of file. */

modules/ts/src/ts_func.cpp

@@ -3002,12 +3002,8 @@ void printVersionInfo(bool useStdOut)
 #if CV_SSE4_2
     if (checkHardwareSupport(CV_CPU_SSE4_2)) cpu_features += " sse4.2";
 #endif
-#if CV_AVX
     if (checkHardwareSupport(CV_CPU_AVX)) cpu_features += " avx";
-#endif
-#if CV_AVX2
     if (checkHardwareSupport(CV_CPU_AVX2)) cpu_features += " avx2";
-#endif
 #if CV_NEON
     cpu_features += " neon"; // NEON is currently not checked at runtime
 #endif