nasm/simd support for libjpeg-turbo

3rdparty/libjpeg-turbo/CMakeLists.txt

@@ -15,8 +15,55 @@ endif()
 
 message(STATUS "libjpeg-turbo: VERSION = ${VERSION}, BUILD = ${BUILD}")
 
+math(EXPR BITS "${CMAKE_SIZEOF_VOID_P} * 8")
+string(TOLOWER ${CMAKE_SYSTEM_PROCESSOR} CMAKE_SYSTEM_PROCESSOR_LC)
+
+if(CMAKE_SYSTEM_PROCESSOR_LC MATCHES "x86_64" OR
+  CMAKE_SYSTEM_PROCESSOR_LC MATCHES "amd64" OR
+  CMAKE_SYSTEM_PROCESSOR_LC MATCHES "i[0-9]86" OR
+  CMAKE_SYSTEM_PROCESSOR_LC MATCHES "x86" OR
+  CMAKE_SYSTEM_PROCESSOR_LC MATCHES "ia32")
+  if(BITS EQUAL 64 OR CMAKE_C_COMPILER_ABI MATCHES "ELF X32")
+    set(CPU_TYPE x86_64)
+  else()
+    set(CPU_TYPE i386)
+  endif()
+  if(NOT CMAKE_SYSTEM_PROCESSOR STREQUAL ${CPU_TYPE})
+    set(CMAKE_SYSTEM_PROCESSOR ${CPU_TYPE})
+  endif()
+elseif(CMAKE_SYSTEM_PROCESSOR_LC STREQUAL "aarch64" OR
+  CMAKE_SYSTEM_PROCESSOR_LC MATCHES "^arm")
+  if(BITS EQUAL 64)
+    set(CPU_TYPE arm64)
+  else()
+    set(CPU_TYPE arm)
+  endif()
+elseif(CMAKE_SYSTEM_PROCESSOR_LC MATCHES "^ppc" OR
+  CMAKE_SYSTEM_PROCESSOR_LC MATCHES "^powerpc")
+  set(CPU_TYPE powerpc)
+else()
+  set(CPU_TYPE ${CMAKE_SYSTEM_PROCESSOR_LC})
+endif()
+if(CMAKE_OSX_ARCHITECTURES MATCHES "x86_64" OR
+  CMAKE_OSX_ARCHITECTURES MATCHES "arm64" OR
+  CMAKE_OSX_ARCHITECTURES MATCHES "i386")
+  set(CPU_TYPE ${CMAKE_OSX_ARCHITECTURES})
+endif()
+if(CMAKE_OSX_ARCHITECTURES MATCHES "ppc")
+  set(CPU_TYPE powerpc)
+endif()
+if(MSVC_IDE AND CMAKE_GENERATOR_PLATFORM MATCHES "arm64")
+  set(CPU_TYPE arm64)
+endif()
+
+message(STATUS "${BITS}-bit build (${CPU_TYPE})")
+
+OCV_OPTION(ENABLE_LIBJPEG_TURBO_SIMD "Include SIMD extensions for libjpeg-turbo, if available for this platform" (NOT CV_DISABLE_OPTIMIZATION)
+  VISIBLE_IF BUILD_JPEG)
 option(WITH_ARITH_ENC "Include arithmetic encoding support when emulating the libjpeg v6b API/ABI" TRUE)
 option(WITH_ARITH_DEC "Include arithmetic decoding support when emulating the libjpeg v6b API/ABI" TRUE)
+set(WITH_SIMD 1)
+set(IS_LIBJPEG_TURBO_SIMD_AVAILABLE 0 PARENT_SCOPE)
 
 include(CheckCSourceCompiles)
 include(CheckIncludeFiles)
@@ -99,12 +146,73 @@ if(WITH_ARITH_DEC)
   set(JPEG_SOURCES ${JPEG_SOURCES} jdarith.c)
 endif()
 
-# No SIMD
+if(MSVC)
-set(JPEG_SOURCES ${JPEG_SOURCES} jsimd_none.c)
+  option(WITH_CRT_DLL
+    "Link all ${CMAKE_PROJECT_NAME} libraries and executables with the C run-time DLL (msvcr*.dll) instead of the static C run-time library (libcmt*.lib.)  The default is to use the C run-time DLL only with the libraries and executables that need it."
+    FALSE)
+  if(NOT WITH_CRT_DLL)
+    # Use the static C library for all build types
+    foreach(var CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
+      CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO)
+      if(${var} MATCHES "/MD")
+        string(REGEX REPLACE "/MD" "/MT" ${var} "${${var}}")
+      endif()
+    endforeach()
+  endif()
+  add_definitions(-D_CRT_NONSTDC_NO_WARNINGS)
+endif()
+
+if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_C_COMPILER_ID STREQUAL "Clang")
+  # Use the maximum optimization level for release builds
+  foreach(var CMAKE_C_FLAGS_RELEASE CMAKE_C_FLAGS_RELWITHDEBINFO)
+    if(${var} MATCHES "-O2")
+      string(REGEX REPLACE "-O2" "-O3" ${var} "${${var}}")
+    endif()
+  endforeach()
+endif()
+
+if(CMAKE_SYSTEM_NAME STREQUAL "SunOS")
+  if(CMAKE_C_COMPILER_ID MATCHES "SunPro")
+    # Use the maximum optimization level for release builds
+    foreach(var CMAKE_C_FLAGS_RELEASE CMAKE_C_FLAGS_RELWITHDEBINFO)
+      if(${var} MATCHES "-xO3")
+        string(REGEX REPLACE "-xO3" "-xO5" ${var} "${${var}}")
+      endif()
+      if(${var} MATCHES "-xO2")
+        string(REGEX REPLACE "-xO2" "-xO5" ${var} "${${var}}")
+      endif()
+    endforeach()
+  endif()
+endif()
+
+message(STATUS ${WITH_LIBJPEG_TURBO_SIMD})
+if(ENABLE_LIBJPEG_TURBO_SIMD)
+  add_subdirectory(src/simd)
+  if(NEON_INTRINSICS)
+    add_definitions(-DNEON_INTRINSICS)
+  endif()
+else()
+  set(WITH_SIMD 0)
+endif()
+
+if(WITH_SIMD)
+  message(STATUS "SIMD extensions: ${CPU_TYPE} (WITH_SIMD = ${WITH_SIMD})")
+  set(IS_LIBJPEG_TURBO_SIMD_AVAILABLE 1 PARENT_SCOPE)
+  if(MSVC_IDE OR XCODE)
+    set_source_files_properties(${SIMD_OBJS} PROPERTIES GENERATED 1)
+  endif()
+else()
+  add_library(simd OBJECT src/jsimd_none.c)
+  if(NOT WIN32 AND (CMAKE_POSITION_INDEPENDENT_CODE OR ENABLE_SHARED))
+    set_target_properties(simd PROPERTIES POSITION_INDEPENDENT_CODE 1)
+  endif()
+endif()
 
 ocv_list_add_prefix(JPEG_SOURCES src/)
 
-add_library(${JPEG_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${JPEG_SOURCES} ${SIMD_OBJS})
+set(JPEG_SOURCES ${JPEG_SOURCES} ${SIMD_OBJS})
+
+add_library(${JPEG_LIBRARY} STATIC ${OPENCV_3RDPARTY_EXCLUDE_FROM_ALL} ${JPEG_SOURCES} $<TARGET_OBJECTS:simd> ${SIMD_OBJS})
 
 set_target_properties(${JPEG_LIBRARY}
   PROPERTIES OUTPUT_NAME ${JPEG_LIBRARY}

3rdparty/libjpeg-turbo/src/simd/CMakeLists.txt

@@ -1,9 +1,13 @@
 macro(simd_fail message)
-  if(REQUIRE_SIMD)
-    message(FATAL_ERROR "${message}.")
-  else()
     message(WARNING "${message}.  Performance will suffer.")
     set(WITH_SIMD 0 PARENT_SCOPE)
+endmacro()
+
+macro(boolean_number var)
+  if(${var})
+    set(${var} 1 ${ARGN})
+  else()
+    set(${var} 0 ${ARGN})
   endif()
 endmacro()
 
@@ -41,14 +45,14 @@ elseif(CPU_TYPE STREQUAL "i386")
   endif()
 endif()
 
-if(NOT REQUIRE_SIMD)
+
-  include(CheckLanguage)
+include(CheckLanguage)
-  check_language(ASM_NASM)
+check_language(ASM_NASM)
-  if(NOT CMAKE_ASM_NASM_COMPILER)
+if(NOT CMAKE_ASM_NASM_COMPILER)
-    simd_fail("SIMD extensions disabled: could not find NASM compiler")
+  simd_fail("SIMD extensions disabled: could not find NASM compiler")
-    return()
+  return()
-  endif()
 endif()
+
 enable_language(ASM_NASM)
 message(STATUS "CMAKE_ASM_NASM_COMPILER = ${CMAKE_ASM_NASM_COMPILER}")
 
@@ -224,6 +228,7 @@ elseif(CPU_TYPE STREQUAL "arm64" OR CPU_TYPE STREQUAL "arm")
 # following test determines whether -mfloat-abi=softfp should be explicitly
 # added to the compile flags for the intrinsics implementation of the Neon SIMD
 # extensions.
+
 if(BITS EQUAL 32)
   check_c_source_compiles("
     #if defined(__ARM_NEON__) || (!defined(__linux__) && !defined(ANDROID) && !defined(__ANDROID__))

CMakeLists.txt

@@ -1341,6 +1341,16 @@ if(WITH_JPEG OR HAVE_JPEG)
     status("    JPEG:" NO)
   elseif(BUILD_JPEG)
     status("    JPEG:" "build-${JPEG_LIBRARY} (ver ${JPEG_LIB_VERSION})")
+    if(ENABLE_LIBJPEG_TURBO_SIMD)
+      status("        SIMD Support Request:" "YES")
+      if(IS_LIBJPEG_TURBO_SIMD_AVAILABLE)
+        status("        SIMD Support:" "YES")
+      else()
+        status("        SIMD Support:" "NO")
+      endif()
+    else()
+      status("        SIMD Support Request:" "NO")
+    endif()
   else()
     status("    JPEG:" "${JPEG_LIBRARY} (ver ${JPEG_LIB_VERSION})")
   endif()

modules/imgcodecs/perf/perf_jpeg.cpp

@@ -0,0 +1,38 @@
+// 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
+#include "perf_precomp.hpp"
+
+namespace opencv_test
+{
+using namespace perf;
+
+PERF_TEST(JPEG, Decode)
+{
+    String filename = getDataPath("stitching/boat1.jpg");
+
+    FILE *f = fopen(filename.c_str(), "rb");
+    fseek(f, 0, SEEK_END);
+    long len = ftell(f);
+    fseek(f, 0, SEEK_SET);
+    vector<uchar> file_buf((size_t)len);
+    EXPECT_EQ(len, (long)fread(&file_buf[0], 1, (size_t)len, f));
+    fclose(f); f = NULL;
+
+    TEST_CYCLE() imdecode(file_buf, IMREAD_UNCHANGED);
+
+    SANITY_CHECK_NOTHING();
+}
+
+PERF_TEST(JPEG, Encode)
+{
+    String filename = getDataPath("stitching/boat1.jpg");
+    cv::Mat src = imread(filename);
+
+    vector<uchar> buf;
+    TEST_CYCLE() imencode(".jpg", src, buf);
+
+    SANITY_CHECK_NOTHING();
+}
+
+} // namespace