Merge pull request #15103 from alalek:simd_intrinsics_in_user_code

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

@@ -0,0 +1,88 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#ifndef OPENCV_CORE_SIMD_INTRINSICS_HPP
+#define OPENCV_CORE_SIMD_INTRINSICS_HPP
+
+/**
+Helper header to support SIMD intrinsics (universal intrinsics) in user code.
+Intrinsics documentation: https://docs.opencv.org/3.4/df/d91/group__core__hal__intrin.html
+
+
+Checks of target CPU instruction set based on compiler definitions don't work well enough.
+More reliable solutions require utilization of configuration systems (like CMake).
+
+So, probably you need to specify your own configuration.
+
+You can do that via CMake in this way:
+    add_definitions(/DOPENCV_SIMD_CONFIG_HEADER=opencv_simd_config_custom.hpp)
+or
+    add_definitions(/DOPENCV_SIMD_CONFIG_INCLUDE_DIR=1)
+
+Additionally you may need to add include directory to your files:
+    include_directories("${CMAKE_CURRENT_LIST_DIR}/opencv_config_${MYTARGET}")
+
+These files can be pre-generated for target configurations of your application
+or generated by CMake on the fly (use CMAKE_BINARY_DIR for that).
+
+Notes:
+- H/W capability checks are still responsibility of your applcation
+- runtime dispatching is not covered by this helper header
+*/
+
+#ifdef __OPENCV_BUILD
+#error "Use core/hal/intrin.hpp during OpenCV build"
+#endif
+
+#ifdef OPENCV_HAL_INTRIN_HPP
+#error "core/simd_intrinsics.hpp must be included before core/hal/intrin.hpp"
+#endif
+
+#include "opencv2/core/cvdef.h"
+#include "opencv2/core/version.hpp"
+
+#ifdef OPENCV_SIMD_CONFIG_HEADER
+#include CVAUX_STR(OPENCV_SIMD_CONFIG_HEADER)
+#elif defined(OPENCV_SIMD_CONFIG_INCLUDE_DIR)
+#include "opencv_simd_config.hpp"  // corresponding directory should be added via -I compiler parameter
+#else  // custom config headers
+
+#if (!defined(CV_AVX_512F) || !CV_AVX_512F) && (defined(__AVX512__) || defined(__AVX512F__))
+#  include <immintrin.h>
+#  undef CV_AVX_512F
+#  define CV_AVX_512F 1
+#  ifndef OPENCV_SIMD_DONT_ASSUME_SKX  // Skylake-X with AVX-512F/CD/BW/DQ/VL
+#    undef CV_AVX512_SKX
+#    define CV_AVX512_SKX 1
+#    undef CV_AVX_512CD
+#    define CV_AVX_512CD 1
+#    undef CV_AVX_512BW
+#    define CV_AVX_512BW 1
+#    undef CV_AVX_512DQ
+#    define CV_AVX_512DQ 1
+#    undef CV_AVX_512VL
+#    define CV_AVX_512VL 1
+#  endif
+#endif // AVX512
+
+// GCC/Clang: -mavx2
+// MSVC: /arch:AVX2
+#if defined __AVX2__
+#  include <immintrin.h>
+#  undef CV_AVX2
+#  define CV_AVX2 1
+#  if defined __F16C__
+#    undef CV_FP16
+#    define CV_FP16 1
+#  endif
+#endif
+
+#endif
+
+// SSE / NEON / VSX is handled by cv_cpu_dispatch.h compatibility block
+#include "cv_cpu_dispatch.h"
+
+#include "hal/intrin.hpp"
+
+#endif // OPENCV_CORE_SIMD_INTRINSICS_HPP

samples/cpp/CMakeLists.txt

@@ -57,6 +57,13 @@ foreach(sample_filename ${cpp_samples})
   if(HAVE_OPENGL AND sample_filename MATCHES "detect_mser")
     target_compile_definitions(${tgt} PRIVATE HAVE_OPENGL)
   endif()
+  if(sample_filename MATCHES "simd_")
+    # disabled intentionally - demonstation purposes only
+    #target_include_directories(${tgt} PRIVATE "${CMAKE_CURRENT_LIST_DIR}")
+    #target_compile_definitions(${tgt} PRIVATE OPENCV_SIMD_CONFIG_HEADER=opencv_simd_config_custom.hpp)
+    #target_compile_definitions(${tgt} PRIVATE OPENCV_SIMD_CONFIG_INCLUDE_DIR=1)
+    #target_compile_options(${tgt} PRIVATE -mavx2)
+  endif()
 endforeach()
 
 include("tutorial_code/calib3d/real_time_pose_estimation/CMakeLists.txt" OPTIONAL)

samples/cpp/simd_basic.cpp

@@ -0,0 +1,49 @@
+#include "opencv2/core.hpp"
+#include "opencv2/core/simd_intrinsics.hpp"
+
+using namespace cv;
+
+int main(int /*argc*/, char** /*argv*/)
+{
+    printf("==================  macro dump  ===================\n");
+#ifdef CV_SIMD
+    printf("CV_SIMD is defined: " CVAUX_STR(CV_SIMD) "\n");
+#ifdef CV_SIMD_WIDTH
+    printf("CV_SIMD_WIDTH is defined: " CVAUX_STR(CV_SIMD_WIDTH) "\n");
+#endif
+#ifdef CV_SIMD128
+    printf("CV_SIMD128 is defined: " CVAUX_STR(CV_SIMD128) "\n");
+#endif
+#ifdef CV_SIMD256
+    printf("CV_SIMD256 is defined: " CVAUX_STR(CV_SIMD256) "\n");
+#endif
+#ifdef CV_SIMD512
+    printf("CV_SIMD512 is defined: " CVAUX_STR(CV_SIMD512) "\n");
+#endif
+#ifdef CV_SIMD_64F
+    printf("CV_SIMD_64F is defined: " CVAUX_STR(CV_SIMD_64F) "\n");
+#endif
+#ifdef CV_SIMD_FP16
+    printf("CV_SIMD_FP16 is defined: " CVAUX_STR(CV_SIMD_FP16) "\n");
+#endif
+#else
+    printf("CV_SIMD is NOT defined\n");
+#endif
+
+#ifdef CV_SIMD
+    printf("=================  sizeof checks  =================\n");
+    printf("sizeof(v_uint8) = %d\n", (int)sizeof(v_uint8));
+    printf("sizeof(v_int32) = %d\n", (int)sizeof(v_int32));
+    printf("sizeof(v_float32) = %d\n", (int)sizeof(v_float32));
+
+    printf("==================  arithm check  =================\n");
+    v_uint8 a = vx_setall_u8(10);
+    v_uint8 c = a + vx_setall_u8(45);
+    printf("(vx_setall_u8(10) + vx_setall_u8(45)).get0() => %d\n", (int)c.get0());
+#else
+    printf("\nSIMD intrinsics are not available. Check compilation target and passed build options.\n");
+#endif
+
+    printf("=====================  done  ======================\n");
+    return 0;
+}