From b654564713dfdba38ccd82d72b129b58d06a85d0 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Stefan=20Br=C3=BCns?= Date: Mon, 15 Jul 2019 20:04:01 +0200 Subject: [PATCH] Fix OpenMP 4.0 issues with GCC9 Backport of pull request #2550. --- CMakeLists.txt | 72 ++++++++---------------- ConfigureChecks.cmake | 30 ++++++++++ DefineOptions.cmake | 44 +++++++++++++++ src/CMakeLists.txt | 2 - src/chart/cairo.c | 5 +- src/chart/main.c | 16 ++++-- src/common/bilateral.c | 18 ++++-- src/common/color_picker.c | 9 ++- src/common/darktable.h | 24 ++++++-- src/common/dwt.c | 20 +++++-- src/common/focus.h | 10 +++- src/common/gaussian.c | 22 ++++++-- src/common/heal.c | 22 ++++++-- src/common/histogram.c | 10 +++- src/common/imageio.c | 34 ++++++++--- src/common/imageio_rawspeed.cc | 10 +++- src/common/interpolation.c | 16 ++++-- src/common/locallaplacian.c | 99 ++++++++++++++++++++++++++------- src/common/opencl.c | 8 ++- src/config.cmake.h | 2 + src/control/jobs/control_jobs.c | 5 +- src/develop/blend.c | 36 ++++++++---- src/develop/imageop_math.c | 55 ++++++++++++++---- src/develop/imageop_math.h | 5 +- src/develop/masks/circle.c | 12 +++- src/develop/masks/ellipse.c | 12 +++- src/develop/masks/gradient.c | 24 ++++++-- src/develop/masks/group.c | 24 ++++++-- src/develop/pixelpipe_hb.c | 15 ++++- src/develop/tiling.c | 35 +++++++++--- src/gui/draw.h | 11 +++- src/iop/ashift.c | 50 +++++++++++++---- src/iop/atrous.c | 33 ++++++++--- src/iop/basecurve.c | 48 ++++++++++++---- src/iop/bilateral.cc | 6 +- src/iop/bloom.c | 20 +++++-- src/iop/channelmixer.c | 5 +- src/iop/clahe.c | 11 +++- src/iop/clipping.c | 11 +++- src/iop/colisa.c | 15 ++++- src/iop/colorbalance.c | 39 +++++++++++-- src/iop/colorchecker.c | 5 +- src/iop/colorcontrast.c | 14 ++++- src/iop/colorin.c | 50 +++++++++++++---- src/iop/colorize.c | 5 +- src/iop/colormapping.c | 15 ++++- src/iop/colorout.c | 16 ++++-- src/iop/colorreconstruction.c | 11 +++- src/iop/colortransfer.c | 15 ++++- src/iop/colorzones.c | 5 +- src/iop/defringe.c | 14 +++-- src/iop/demosaic.c | 71 ++++++++++++++++++----- src/iop/denoiseprofile.c | 64 ++++++++++++++++----- src/iop/dither.c | 12 +++- src/iop/equalizer_eaw.h | 22 ++++++-- src/iop/exposure.c | 8 ++- src/iop/filmic.c | 16 +++++- src/iop/gamma.c | 16 ++++-- src/iop/globaltonemap.c | 15 ++++- src/iop/graduatednd.c | 20 +++++-- src/iop/grain.c | 5 +- src/iop/hazeremoval.c | 41 ++++++++++---- src/iop/highlights.c | 44 +++++++++++---- src/iop/highpass.c | 10 +++- src/iop/hotpixels.c | 13 ++++- src/iop/invert.c | 27 +++++++-- src/iop/lens.c | 46 ++++++++++++--- src/iop/levels.c | 4 +- src/iop/liquify.c | 4 +- src/iop/lowlight.c | 5 +- src/iop/lowpass.c | 15 ++++- src/iop/monochrome.c | 10 +++- src/iop/nlmeans.c | 22 ++++++-- src/iop/overexposed.c | 5 +- src/iop/profile_gamma.c | 20 +++++-- src/iop/rawdenoise.c | 48 ++++++++++++---- src/iop/rawoverexposed.c | 12 +++- src/iop/rawprepare.c | 26 +++++++-- src/iop/relight.c | 5 +- src/iop/retouch.c | 65 +++++++++++++++++----- src/iop/rotatepixels.c | 5 +- src/iop/scalepixels.c | 5 +- src/iop/shadhi.c | 13 ++++- src/iop/sharpen.c | 41 ++++++++++---- src/iop/soften.c | 38 ++++++++++--- src/iop/splittoning.c | 6 +- src/iop/spots.c | 4 +- src/iop/temperature.c | 30 ++++++++-- src/iop/tonecurve.c | 6 +- src/iop/velvia.c | 9 ++- src/iop/vibrance.c | 5 +- src/iop/vignette.c | 6 +- src/iop/zonesystem.c | 29 ++++++++-- 93 files changed, 1519 insertions(+), 444 deletions(-) create mode 100644 ConfigureChecks.cmake create mode 100644 DefineOptions.cmake diff --git a/CMakeLists.txt b/CMakeLists.txt index 5b783d1..ce4047b 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -33,15 +33,7 @@ IF(DEFINED DT_FORCE_CXX_COMPILER) set(CMAKE_CXX_COMPILER ${DT_FORCE_CXX_COMPILER}) endif() -include(CheckCCompilerFlag) -include(TestBigEndian) - -# Check if this is source package build -if(NOT IS_DIRECTORY ${CMAKE_SOURCE_DIR}/.git) - set(SOURCE_PACKAGE 1) -else() - set(SOURCE_PACKAGE 0) -endif() +include(DefineOptions.cmake) # Include GNUInstallDirs, which sets sensible defaults for install directories. # See https://cmake.org/cmake/help/v3.0/module/GNUInstallDirs.html for further information. @@ -50,41 +42,31 @@ endif() include(GNUInstallDirs) -option(USE_CAMERA_SUPPORT "Detect and use camera support if available." ON) -option(USE_NLS "Build Native Language Support (using gettext)" ON) -option(USE_COLORD "Enable colord support" ON) -option(USE_MAP "Build Map View parts" ON) -option(USE_LUA "Build lua scripting support" ON) -option(DONT_USE_INTERNAL_LUA "Never fall back to the intree copy of lua" ON) -option(USE_FLICKR "Enable Flickr support" ON) -option(USE_KWALLET "Build kwallet password storage back-end" ON) -option(USE_LIBSECRET "Build libsecret password storage back-end" ON) -option(USE_UNITY "Use libunity to report progress in the launcher" OFF) -option(USE_OPENMP "Use openmp threading support." ON) -option(USE_OPENCL "Use OpenCL support." ON) -option(USE_GRAPHICSMAGICK "Use GraphicsMagick library for image import." ON) -option(USE_DARKTABLE_PROFILING OFF) -option(CUSTOM_CFLAGS "Don't override compiler optimization flags." OFF) -option(BUILD_USERMANUAL "Build all the versions of the usermanual." OFF) -option(BINARY_PACKAGE_BUILD "Sets march optimization to generic" OFF) -option(USE_XMLLINT "Run xmllint to test if darktableconfig.xml is valid" ON) -option(USE_OPENJPEG "Enable JPEG 2000 support" ON) -option(USE_WEBP "Enable WebP export support" ON) -option(BUILD_CMSTEST "Build a test program to check your system's color management setup" ON) -option(USE_OPENEXR "Enable OpenEXR support" ON) -option(BUILD_PRINT "Build the print module" ON) -option(BUILD_RS_IDENTIFY "Build the darktable-rs-identify debug aid" ON) -option(BUILD_SSE2_CODEPATHS "(EXPERIMENTAL OPTION, DO NOT DISABLE) Building SSE2-optimized codepaths" ON) -option(VALIDATE_APPDATA_FILE "Use appstream-util (if found) to validate the .appdata file" OFF) -option(BUILD_TESTS "Build tests in src/tests/, runnable from the build/ directory" OFF) -option(BUILD_BATTERY_INDICATOR "Add an icon to the top toolbar showing the state of a laptop battery" OFF) -option(BUILD_MSYS2_INSTALL "Build an MSYS2 version of the install, aka for Windows platform, but without dependency installs" OFF) -option(BUILD_NOISE_TOOLS "Build tools for generating noise proifles" OFF) +if (USE_OPENMP) + if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR + CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang") -if(USE_OPENCL) - option(TESTBUILD_OPENCL_PROGRAMS "Test-compile opencl programs (needs llvm and clang 3.9+)" ON) + # Clang has an option to specify the OpenMP standard to use. Specify it. + # FIXME: Implement this in FindOpenMP.cmake + set(OPENMP_VERSION_SPECIFIER "-fopenmp-version=40") + + set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OPENMP_VERSION_SPECIFIER}") + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OPENMP_VERSION_SPECIFIER}") + endif() + + find_package(OpenMP 4.0 REQUIRED) +endif() + +include(ConfigureChecks.cmake) + +include(CheckCCompilerFlag) +include(TestBigEndian) + +# Check if this is source package build +if(NOT IS_DIRECTORY ${CMAKE_SOURCE_DIR}/.git) + set(SOURCE_PACKAGE 1) else() - set(TESTBUILD_OPENCL_PROGRAMS OFF) + set(SOURCE_PACKAGE 0) endif() if(BUILD_SSE2_CODEPATHS) @@ -122,12 +104,6 @@ MESSAGE(STATUS "Is the target platform supported: ${IS_SUPPORTED_PLATFORM}") set(CMAKE_REQUIRED_INCLUDES "${CMAKE_REQUIRED_INCLUDES_OLD}") unset(CMAKE_REQUIRED_INCLUDES_OLD) -if(APPLE) - option(USE_MAC_INTEGRATION "Enable OS X integration" ON) -else(APPLE) - set(USE_MAC_INTEGRATION OFF) -endif(APPLE) - # When cross compiling, CMAKE_INSTALL_PREFIX will point to something like "/opt/darktable", but that's not useful when using the path to load # modules on runtime. Then we need something like "C:\Program Files\darktable". Doesn't need to be set when doing regular compiles. if(NOT DEFINED RUNTIME_INSTALL_PREFIX) diff --git a/DefineOptions.cmake b/DefineOptions.cmake new file mode 100644 index 0000000..b49c21b --- /dev/null +++ b/DefineOptions.cmake @@ -0,0 +1,44 @@ +option(USE_CAMERA_SUPPORT "Detect and use camera support if available." ON) +option(USE_NLS "Build Native Language Support (using gettext)" ON) +option(USE_COLORD "Enable colord support" ON) +option(USE_MAP "Build Map View parts" ON) +option(USE_LUA "Build lua scripting support" ON) +option(DONT_USE_INTERNAL_LUA "Never fall back to the intree copy of lua" ON) +option(USE_FLICKR "Enable Flickr support" ON) +option(USE_KWALLET "Build kwallet password storage back-end" ON) +option(USE_LIBSECRET "Build libsecret password storage back-end" ON) +option(USE_UNITY "Use libunity to report progress in the launcher" OFF) +option(USE_OPENMP "Use openmp threading support." ON) +option(USE_OPENCL "Use OpenCL support." ON) +option(USE_GRAPHICSMAGICK "Use GraphicsMagick library for image import." ON) +option(USE_DARKTABLE_PROFILING OFF) +option(CUSTOM_CFLAGS "Don't override compiler optimization flags." OFF) +option(BUILD_USERMANUAL "Build all the versions of the usermanual." OFF) +option(BINARY_PACKAGE_BUILD "Sets march optimization to generic" OFF) +option(USE_XMLLINT "Run xmllint to test if darktableconfig.xml is valid" ON) +option(USE_OPENJPEG "Enable JPEG 2000 support" ON) +option(USE_WEBP "Enable WebP export support" ON) +option(BUILD_CMSTEST "Build a test program to check your system's color management setup" ON) +option(USE_OPENEXR "Enable OpenEXR support" ON) +option(BUILD_PRINT "Build the print module" ON) +option(BUILD_RS_IDENTIFY "Build the darktable-rs-identify debug aid" ON) +option(BUILD_SSE2_CODEPATHS "(EXPERIMENTAL OPTION, DO NOT DISABLE) Building SSE2-optimized codepaths" ON) +option(VALIDATE_APPDATA_FILE "Use appstream-util (if found) to validate the .appdata file" OFF) +option(BUILD_TESTS "Build tests in src/tests/, runnable from the build/ directory" OFF) +option(BUILD_BATTERY_INDICATOR "Add an icon to the top toolbar showing the state of a laptop battery" OFF) +option(BUILD_MSYS2_INSTALL "Build an MSYS2 version of the install, aka for Windows platform, but without dependency installs" OFF) +option(BUILD_NOISE_TOOLS "Build tools for generating noise proifles" OFF) +option(BUILD_CURVE_TOOLS "Build tools for generating base and tone curves" OFF) +option(USE_LENSFUN "Enable LensFun support" ON) + +if (USE_OPENCL) + option(TESTBUILD_OPENCL_PROGRAMS "Test-compile opencl programs (needs llvm and clang 3.9+)" ON) +else () + set(TESTBUILD_OPENCL_PROGRAMS OFF) +endif () + +if(APPLE) + option(USE_MAC_INTEGRATION "Enable OS X integration" ON) +else(APPLE) + set(USE_MAC_INTEGRATION OFF) +endif(APPLE) diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index 0ed087d..d01dc6b 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -225,8 +225,6 @@ list(APPEND LIBS "${CMAKE_THREAD_LIBS_INIT}") # Need to explicitly link against math library. list(APPEND LIBS "-lm") -# Check for openmp -find_package(OpenMP) if(USE_OPENMP) if(OpenMP_C_FLAGS) set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}") diff --git a/ConfigureChecks.cmake b/ConfigureChecks.cmake new file mode 100644 index 0000000..58d229a --- /dev/null +++ b/ConfigureChecks.cmake @@ -0,0 +1,30 @@ +include(CheckCSourceCompiles) + +if (OpenMP_FOUND) + +set(CMAKE_REQUIRED_FLAGS ${OpenMP_C_FLAGS}) +set(CMAKE_REQUIRED_LIBRARIES ${OpenMP_C_LIBRARIES}) +check_c_source_compiles(" +#include + +static void sink(const int x, int a[]) +{ +#pragma omp parallel for default(none) firstprivate(x) shared(a) + for(int i = 0; i < 3; i++) { + a[i] = x + i; + } +} + +int main(void) +{ + int x = 42; + int a[3] = {0}; + + sink(x, a); + + return 0; +}" HAVE_OMP_FIRSTPRIVATE_WITH_CONST) + +set(CMAKE_REQUIRED_FLAGS) +set(CMAKE_REQUIRED_LIBRARIES) +endif() diff --git a/src/config.cmake.h b/src/config.cmake.h index 715ea48..bdd72c9 100644 --- a/src/config.cmake.h +++ b/src/config.cmake.h @@ -66,6 +66,8 @@ static const char *dt_supported_extensions[] __attribute__((unused)) = {"@DT_SUP #define ASAN_UNPOISON_MEMORY_REGION(addr, size) ((void)(addr), (void)(size)) #endif +#cmakedefine HAVE_OMP_FIRSTPRIVATE_WITH_CONST 1 + // modelines: These editor modelines have been set for all relevant files by tools/update_modelines.sh // vim: shiftwidth=2 expandtab tabstop=2 cindent // kate: tab-indents: off; indent-width 2; replace-tabs on; indent-mode cstyle; remove-trailing-spaces modified; diff --git a/src/common/darktable.h b/src/common/darktable.h index 25e8d46..5bdcbf8 100644 --- a/src/common/darktable.h +++ b/src/common/darktable.h @@ -75,11 +75,23 @@ typedef unsigned int u_int; #endif #ifdef _OPENMP -#include -#else -#define omp_get_max_threads() 1 -#define omp_get_thread_num() 0 -#endif +# include + +/* See https://redmine.darktable.org/issues/12568#note-14 */ +# ifdef HAVE_OMP_FIRSTPRIVATE_WITH_CONST + /* If the compiler correctly supports firstprivate, use it. */ +# define dt_omp_firstprivate(...) firstprivate(__VA_ARGS__) +# else /* HAVE_OMP_FIRSTPRIVATE_WITH_CONST */ + /* This is needed for clang < 7.0 */ +# define dt_omp_firstprivate(...) +# endif/* HAVE_OMP_FIRSTPRIVATE_WITH_CONST */ + +#else /* _OPENMP */ + +# define omp_get_max_threads() 1 +# define omp_get_thread_num() 0 + +#endif /* _OPENMP */ #ifndef _RELEASE #include "common/poison.h" @@ -542,7 +554,7 @@ int dt_load_from_string(const gchar *image_to_load, gboolean open_image_in_dr, g #define dt_unreachable_codepath_with_desc(D) \ dt_unreachable_codepath_with_caller(D, __FILE__, __LINE__, __FUNCTION__) -#define dt_unreachable_codepath() dt_unreachable_codepath_with_caller(NULL, __FILE__, __LINE__, __FUNCTION__) +#define dt_unreachable_codepath() dt_unreachable_codepath_with_caller("unreachable", __FILE__, __LINE__, __FUNCTION__) static inline void dt_unreachable_codepath_with_caller(const char *description, const char *file, const int line, const char *function) { diff --git a/src/common/bilateral.c b/src/common/bilateral.c index 2708eea..ac67210 100644 --- a/src/common/bilateral.c +++ b/src/common/bilateral.c @@ -127,7 +127,7 @@ void dt_bilateral_splat(dt_bilateral_t *b, const float *const in) const int oz = b->size_y * b->size_x; // splat into downsampled grid #ifdef _OPENMP -#pragma omp parallel for default(none) shared(b) +#pragma omp parallel for default(none) dt_omp_firstprivate(in, oy, oz) shared(b) #endif for(int j = 0; j < b->height; j++) { @@ -170,7 +170,9 @@ static void blur_line_z(float *buf, const int offset1, const int offset2, const const float w1 = 4.f / 16.f; const float w2 = 2.f / 16.f; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(buf) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(size1, size2, size3, offset1, offset2, offset3) \ + shared(buf) #endif for(int k = 0; k < size1; k++) { @@ -208,7 +210,9 @@ static void blur_line(float *buf, const int offset1, const int offset2, const in const float w1 = 4.f / 16.f; const float w2 = 1.f / 16.f; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(buf) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(size1, size2, size3, offset1, offset2, offset3) \ + shared(buf) #endif for(int k = 0; k < size1; k++) { @@ -260,7 +264,9 @@ void dt_bilateral_slice(const dt_bilateral_t *const b, const float *const in, fl const int oy = b->size_x; const int oz = b->size_y * b->size_x; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(b, in, norm, oy, oz) \ + shared(out) #endif for(int j = 0; j < b->height; j++) { @@ -306,7 +312,9 @@ void dt_bilateral_slice_to_output(const dt_bilateral_t *const b, const float *co const int oy = b->size_x; const int oz = b->size_y * b->size_x; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(b, in, norm, oy, oz) \ + shared(out) #endif for(int j = 0; j < b->height; j++) { diff --git a/src/common/color_picker.c b/src/common/color_picker.c index 76875d4..45c49a2 100644 --- a/src/common/color_picker.c +++ b/src/common/color_picker.c @@ -79,7 +79,8 @@ static void color_picker_helper_4ch_parallel(const dt_iop_buffer_dsc_t *dsc, con } #ifdef _OPENMP -#pragma omp parallel default(none) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(w, pixel, width, box, mean, mmin, mmax) #endif { const int tnum = dt_get_thread_num(); @@ -199,7 +200,8 @@ static void color_picker_helper_bayer_parallel(const dt_iop_buffer_dsc_t *const } #ifdef _OPENMP -#pragma omp parallel default(none) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(pixel, width, roi, filters, box, msum, mmin, mmax, cnt) #endif { const int tnum = dt_get_thread_num(); @@ -326,7 +328,8 @@ static void color_picker_helper_xtrans_parallel(const dt_iop_buffer_dsc_t *const } #ifdef _OPENMP -#pragma omp parallel default(none) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(pixel, width, roi, xtrans, box, cnt, msum, mmin, mmax) #endif { const int tnum = dt_get_thread_num(); diff --git a/src/common/dwt.c b/src/common/dwt.c index eb37224..0fedbe8 100644 --- a/src/common/dwt.c +++ b/src/common/dwt.c @@ -194,7 +194,10 @@ static void dwt_add_layer_sse(float *const img, float *layers, dwt_params_t *con const int i_size = p->width * p->height * 4; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(layers) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(img, i_size) \ + shared(layers) \ + schedule(static) #endif for(int i = 0; i < i_size; i += 4) { @@ -216,7 +219,10 @@ static void dwt_add_layer(float *const img, float *layers, dwt_params_t *const p const int i_size = p->width * p->height * p->ch; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(layers) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(img, i_size) \ + shared(layers) \ + schedule(static) #endif for(int i = 0; i < i_size; i++) layers[i] += img[i]; } @@ -233,7 +239,10 @@ static void dwt_subtract_layer_sse(float *bl, float *bh, dwt_params_t *const p) const int size = p->width * p->height * 4; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(bl, bh) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(v4_lpass_mult, size) \ + shared(bl, bh) \ + schedule(static) #endif for(int i = 0; i < size; i += 4) { @@ -258,7 +267,10 @@ static void dwt_subtract_layer(float *bl, float *bh, dwt_params_t *const p) const int size = p->width * p->height * p->ch; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(bl, bh) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(size) \ + shared(bl, bh) \ + schedule(static) #endif for(int i = 0; i < size; i++) { diff --git a/src/common/gaussian.c b/src/common/gaussian.c index 545377e..0798060 100644 --- a/src/common/gaussian.c +++ b/src/common/gaussian.c @@ -173,8 +173,10 @@ void dt_gaussian_blur(dt_gaussian_t *g, const float *const in, float *const out) // vertical blur column by column #ifdef _OPENMP -#pragma omp parallel for default(none) shared(temp, Labmin, Labmax, a0, a1, a2, a3, b1, b2, coefp, \ - coefn) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, width, height, ch) \ + shared(temp, Labmin, Labmax, a0, a1, a2, a3, b1, b2, coefp, coefn) \ + schedule(static) #endif for(int i = 0; i < width; i++) { @@ -245,8 +247,10 @@ void dt_gaussian_blur(dt_gaussian_t *g, const float *const in, float *const out) // horizontal blur line by line #ifdef _OPENMP -#pragma omp parallel for default(none) shared(temp, Labmin, Labmax, a0, a1, a2, a3, b1, b2, coefp, \ - coefn) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(out, ch, width, height) \ + shared(temp, Labmin, Labmax, a0, a1, a2, a3, b1, b2, coefp, coefn) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -340,7 +344,10 @@ static void dt_gaussian_blur_4c_sse(dt_gaussian_t *g, const float *const in, flo // vertical blur column by column #ifdef _OPENMP -#pragma omp parallel for default(none) shared(temp, a0, a1, a2, a3, b1, b2, coefp, coefn) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, Labmin, Labmax, width, height) \ + shared(temp, a0, a1, a2, a3, b1, b2, coefp, coefn) \ + schedule(static) #endif for(int i = 0; i < width; i++) { @@ -408,7 +415,10 @@ static void dt_gaussian_blur_4c_sse(dt_gaussian_t *g, const float *const in, flo // horizontal blur line by line #ifdef _OPENMP -#pragma omp parallel for default(none) shared(temp, a0, a1, a2, a3, b1, b2, coefp, coefn) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(out, Labmin, Labmax, width, height) \ + shared(temp, a0, a1, a2, a3, b1, b2, coefp, coefn) \ + schedule(static) #endif for(size_t j = 0; j < height; j++) { diff --git a/src/common/heal.c b/src/common/heal.c index 37e53eb..56ae231 100644 --- a/src/common/heal.c +++ b/src/common/heal.c @@ -54,7 +54,10 @@ static void dt_heal_sub(const float *const top_buffer, const float *const bottom const int i_size = width * height * ch; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(result_buffer) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(top_buffer, bottom_buffer, i_size) \ + shared(result_buffer) \ + schedule(static) #endif for(int i = 0; i < i_size; i++) result_buffer[i] = top_buffer[i] - bottom_buffer[i]; } @@ -66,7 +69,10 @@ static void dt_heal_add(const float *const first_buffer, const float *const seco const int i_size = width * height * ch; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(result_buffer) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(first_buffer, second_buffer, i_size) \ + shared(result_buffer) \ + schedule(static) #endif for(int i = 0; i < i_size; i++) result_buffer[i] = first_buffer[i] + second_buffer[i]; } @@ -78,7 +84,11 @@ static float dt_heal_laplace_iteration_sse(float *pixels, const float *const Adi float err = 0.f; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(pixels) schedule(static) reduction(+ : err) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(Adiag, Aidx, w, nmask_from, nmask_to) \ + shared(pixels) \ + schedule(static) \ + reduction(+ : err) #endif for(int i = nmask_from; i < nmask_to; i++) { @@ -135,7 +145,11 @@ static float dt_heal_laplace_iteration(float *pixels, const float *const Adiag, const int ch1 = (ch == 4) ? ch - 1 : ch; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(pixels) schedule(static) reduction(+ : err) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(Adiag, Aidx, w, nmask_from, nmask_to, ch1) \ + shared(pixels) \ + schedule(static) \ + reduction(+ : err) #endif for(int i = nmask_from; i < nmask_to; i++) { diff --git a/src/common/histogram.c b/src/common/histogram.c index 2fccdd3..b7bbfd7 100644 --- a/src/common/histogram.c +++ b/src/common/histogram.c @@ -217,7 +217,10 @@ void dt_histogram_worker(dt_dev_histogram_collection_params_t *const histogram_p const dt_histogram_roi_t *const roi = histogram_params->roi; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(partial_hists) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(histogram_params, pixel, Worker, bins_total, roi) \ + shared(partial_hists) \ + schedule(static) #endif for(int j = roi->crop_y; j < roi->height - roi->crop_height; j++) { @@ -230,7 +233,10 @@ void dt_histogram_worker(dt_dev_histogram_collection_params_t *const histogram_p memset(*histogram, 0, buf_size); uint32_t *hist = *histogram; -#pragma omp parallel for schedule(static) default(none) shared(hist, partial_hists) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(nthreads, bins_total) \ + shared(hist, partial_hists) \ + schedule(static) for(size_t k = 0; k < bins_total; k++) { for(size_t n = 0; n < nthreads; n++) diff --git a/src/common/imageio.c b/src/common/imageio.c index c82cdb5..f1ac149 100644 --- a/src/common/imageio.c +++ b/src/common/imageio.c @@ -182,7 +182,10 @@ void dt_imageio_flip_buffers(char *out, const char *in, const size_t bpp, const if(!orientation) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ht, wd, bpp, stride) \ + shared(in, out) \ + schedule(static) #endif for(int j = 0; j < ht; j++) memcpy(out + (size_t)j * bpp * wd, in + (size_t)j * stride, bpp * wd); return; @@ -205,7 +208,10 @@ void dt_imageio_flip_buffers(char *out, const char *in, const size_t bpp, const si = -si; } #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out, jj, ii, sj, si) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(wd, bpp, ht, stride) \ + shared(in, out, jj, ii, sj, si) \ + schedule(static) #endif for(int j = 0; j < ht; j++) { @@ -229,7 +235,10 @@ void dt_imageio_flip_buffers_ui16_to_float(float *out, const uint16_t *in, const if(!orientation) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(wd, ch, scale, stride, black, ht) \ + shared(in, out) \ + schedule(static) #endif for(int j = 0; j < ht; j++) for(int i = 0; i < wd; i++) @@ -255,7 +264,10 @@ void dt_imageio_flip_buffers_ui16_to_float(float *out, const uint16_t *in, const si = -si; } #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out, jj, ii, sj, si) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(wd, ch, black, scale, stride, ht) \ + shared(in, out, jj, ii, sj, si) \ + schedule(static) #endif for(int j = 0; j < ht; j++) { @@ -279,7 +291,10 @@ void dt_imageio_flip_buffers_ui8_to_float(float *out, const uint8_t *in, const f if(!orientation) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(wd, scale, black, ht, ch, stride) \ + shared(in, out) \ + schedule(static) #endif for(int j = 0; j < ht; j++) for(int i = 0; i < wd; i++) @@ -305,7 +320,10 @@ void dt_imageio_flip_buffers_ui8_to_float(float *out, const uint8_t *in, const f si = -si; } #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out, jj, ii, sj, si) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(wd, ch, scale, black, stride, ht) \ + shared(in, out, jj, ii, sj, si) \ + schedule(static) #endif for(int j = 0; j < ht; j++) { @@ -865,7 +883,9 @@ int dt_imageio_export_with_flags(const uint32_t imgid, const char *filename, { // !display_byteorder, need to swap: uint8_t *const buf8 = pipe.backbuf; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(processed_width, processed_height, buf8) \ + schedule(static) #endif // just flip byte order for(size_t k = 0; k < (size_t)processed_width * processed_height; k++) diff --git a/src/common/imageio_rawspeed.cc b/src/common/imageio_rawspeed.cc index 0dcdc81..a6fc74d 100644 --- a/src/common/imageio_rawspeed.cc +++ b/src/common/imageio_rawspeed.cc @@ -383,7 +383,10 @@ dt_imageio_retval_t dt_imageio_open_rawspeed_sraw(dt_image_t *img, RawImage r, d */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(r, img, buf) +#pragma omp parallel for default(none) \ + schedule(static) \ + dt_omp_firstprivate(cpp) \ + shared(r, img, buf) #endif for(int j = 0; j < img->height; j++) { @@ -407,7 +410,10 @@ dt_imageio_retval_t dt_imageio_open_rawspeed_sraw(dt_image_t *img, RawImage r, d */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(r, img, buf) +#pragma omp parallel for default(none) \ + schedule(static) \ + dt_omp_firstprivate(cpp) \ + shared(r, img, buf) #endif for(int j = 0; j < img->height; j++) { diff --git a/src/common/interpolation.c b/src/common/interpolation.c index f627d0d..154715a 100644 --- a/src/common/interpolation.c +++ b/src/common/interpolation.c @@ -1378,7 +1378,9 @@ static void dt_interpolation_resample_plain(const struct dt_interpolation *itor, int64_t ts_resampling = getts(); #endif #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, roi_out, x0, l) \ + shared(out) #endif for(int y = 0; y < roi_out->height; y++) { @@ -1424,7 +1426,9 @@ static void dt_interpolation_resample_plain(const struct dt_interpolation *itor, // Process each output line #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out, hindex, hlength, hkernel, vindex, vlength, vkernel, vmeta) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, roi_out) \ + shared(out, hindex, hlength, hkernel, vindex, vlength, vkernel, vmeta) #endif for(int oy = 0; oy < roi_out->height; oy++) { @@ -1536,7 +1540,9 @@ static void dt_interpolation_resample_sse(const struct dt_interpolation *itor, f int64_t ts_resampling = getts(); #endif #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, roi_out, x0, l) \ + shared(out) #endif for(int y = 0; y < roi_out->height; y++) { @@ -1582,7 +1588,9 @@ static void dt_interpolation_resample_sse(const struct dt_interpolation *itor, f // Process each output line #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out, hindex, hlength, hkernel, vindex, vlength, vkernel, vmeta) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, roi_out) \ + shared(out, hindex, hlength, hkernel, vindex, vlength, vkernel, vmeta) #endif for(int oy = 0; oy < roi_out->height; oy++) { diff --git a/src/common/locallaplacian.c b/src/common/locallaplacian.c index 5deb3c1..bfe7f4c 100644 --- a/src/common/locallaplacian.c +++ b/src/common/locallaplacian.c @@ -114,7 +114,10 @@ static inline void gauss_expand( const int ht) { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) collapse(2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fine, input, wd, ht) \ + schedule(static) \ + collapse(2) #endif for(int j=1;j<((ht-1)&~1);j++) // even ht: two px boundary. odd ht: one px. for(int i=1;i<((wd-1)&~1);i++) @@ -151,7 +154,9 @@ static inline void gauss_reduce_sse2( float *const row = ringbuf + (rowj % 5)*stride; const float *const in = input + rowj*wd; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cw, in, row) \ + schedule(static) #endif for(int i=1;imode == 2) { // pad by preview buffer #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) shared(wd2, ht2) collapse(2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ht, input, max_supp, out, wd) \ + shared(wd2, ht2) \ + schedule(dynamic) \ + collapse(2) #endif // fill regular pixels: for(int j=0;j [0,1] @@ -276,22 +290,38 @@ static inline float *ll_pad_input( out[*wd2*j+i] = b->pad0[b->pwd*py+px];\ } } while(0) #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) shared(wd2, ht2, b) collapse(2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(input, max_supp, out, wd) \ + shared(wd2, ht2, b) \ + schedule(dynamic) \ + collapse(2) #endif // left border for(int j=max_supp;j<*ht2-max_supp;j++) for(int i=0;i [0,1] } #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) shared(wd2, ht2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(max_supp, out) \ + shared(wd2, ht2) \ + schedule(dynamic) #endif for(int j=0;jwhitelevel = fmaxf(d->whitelevel, saturation * cal); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(d, saturation) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ivoid, cal, photoncnt) \ + shared(d, saturation) \ + schedule(static) #endif for(int y = 0; y < d->ht; y++) for(int x = 0; x < d->wd; x++) diff --git a/src/develop/blend.c b/src/develop/blend.c index 044880b..ad24cb5 100644 --- a/src/develop/blend.c +++ b/src/develop/blend.c @@ -2901,7 +2901,8 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp // blend uniformly (no drawn or parametric mask) #ifdef _OPENMP -#pragma omp parallel for default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask, opacity) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = opacity; } @@ -2920,7 +2921,8 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp { // if we have a mask and this flag is set -> invert the mask #ifdef _OPENMP -#pragma omp parallel for default(none) + #pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = 1.0f - mask[i]; } @@ -2931,7 +2933,8 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp // we fill the buffer with 1.0f or 0.0f depending on mask_combine const float fill = (d->mask_combine & DEVELOP_COMBINE_MASKS_POS) ? 0.0f : 1.0f; #ifdef _OPENMP -#pragma omp parallel for default(none) + #pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask, fill) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = fill; } @@ -2940,14 +2943,17 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp // we fill the buffer with 1.0f or 0.0f depending on mask_combine const float fill = (d->mask_combine & DEVELOP_COMBINE_INCL) ? 0.0f : 1.0f; #ifdef _OPENMP -#pragma omp parallel for default(none) + #pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask, fill) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = fill; } // get parametric mask (if any) and apply global opacity #ifdef _OPENMP -#pragma omp parallel for default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(bch, ch, cst, d, oheight, opacity, ivoid, iwidth, \ + mask, owidth, ovoid, xoffs, yoffs) #endif for(size_t y = 0; y < oheight; y++) { @@ -2985,7 +2991,8 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp { float *const guide_tmp = dt_alloc_align(64, sizeof(*guide_tmp) * buffsize * ch); #ifdef _OPENMP -#pragma omp parallel for default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, guide_tmp, ivoid, iwidth, oheight, owidth, xoffs, yoffs) #endif for(size_t y = 0; y < oheight; y++) { @@ -3018,7 +3025,8 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp const float e = expf(3.f * d->contrast); const float brightness = d->brightness; #ifdef _OPENMP -#pragma omp parallel for default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(brightness, buffsize, e, mask, opacity) #endif for(size_t k = 0; k < buffsize; k++) { @@ -3047,7 +3055,10 @@ void dt_develop_blend_process(struct dt_iop_module_t *self, struct dt_dev_pixelp // select the blend operator _blend_row_func *const blend = dt_develop_choose_blend_func(d->blend_mode); #ifdef _OPENMP -#pragma omp parallel for default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(bch, blend, blendflag, ch, cst, ivoid, iwidth, mask, \ + mask_display, oheight, ovoid, owidth, \ + request_mask_display, xoffs, yoffs) #endif for(size_t y = 0; y < oheight; y++) { @@ -3224,7 +3235,8 @@ int dt_develop_blend_process_cl(struct dt_iop_module_t *self, struct dt_dev_pixe { // if we have a mask and this flag is set -> invert the mask #ifdef _OPENMP -#pragma omp parallel for default(none) + #pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = 1.0f - mask[i]; } @@ -3235,7 +3247,8 @@ int dt_develop_blend_process_cl(struct dt_iop_module_t *self, struct dt_dev_pixe // we fill the buffer with 1.0f or 0.0f depending on mask_combine const float fill = (d->mask_combine & DEVELOP_COMBINE_MASKS_POS) ? 0.0f : 1.0f; #ifdef _OPENMP -#pragma omp parallel for default(none) + #pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask, fill) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = fill; } @@ -3244,7 +3257,8 @@ int dt_develop_blend_process_cl(struct dt_iop_module_t *self, struct dt_dev_pixe // we fill the buffer with 1.0f or 0.0f depending on mask_combine const float fill = (d->mask_combine & DEVELOP_COMBINE_INCL) ? 0.0f : 1.0f; #ifdef _OPENMP -#pragma omp parallel for default(none) + #pragma omp parallel for default(none) \ + dt_omp_firstprivate(buffsize, mask, fill) #endif for(size_t i = 0; i < buffsize; i++) mask[i] = fill; } diff --git a/src/develop/imageop_math.c b/src/develop/imageop_math.c index 3159e14..a10492c 100644 --- a/src/develop/imageop_math.c +++ b/src/develop/imageop_math.c @@ -67,7 +67,10 @@ void dt_iop_flip_and_zoom_8(const uint8_t *in, int32_t iw, int32_t ih, uint8_t * const int32_t offm = half_pixel * bpp * MIN(MIN(0, si), MIN(sj, si + sj)); const int32_t offM = half_pixel * bpp * MAX(MAX(0, si), MAX(sj, si + sj)); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out, jj, ii, sj, si, iw, ih) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(bpp, half_pixel, ht, offM, offm, scale, wd) \ + shared(in, out, jj, ii, sj, si, iw, ih) \ + schedule(static) #endif for(uint32_t j = 0; j < ht; j++) { @@ -211,7 +214,9 @@ void dt_iop_clip_and_zoom_mosaic_half_size_plain(uint16_t *const out, const uint } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(clut) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(filters, in, in_stride, out, out_stride, px_footprint, rggbx, rggby, roi_in, roi_out) \ + shared(clut) schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -270,7 +275,9 @@ void dt_iop_clip_and_zoom_mosaic_half_size_sse2(uint16_t *const out, const uint1 const int rggbx = trggbx, rggby = trggby; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out, out_stride, px_footprint, rggbx, rggby, roi_in, roi_out, samples) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -479,7 +486,9 @@ void dt_iop_clip_and_zoom_mosaic_half_size_f_plain(float *const out, const float const int rggbx = trggbx, rggby = trggby; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out, out_stride, px_footprint, rggbx, rggby, roi_in, roi_out, samples) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -666,7 +675,10 @@ void dt_iop_clip_and_zoom_mosaic_half_size_f_sse2(float *const out, const float const int rggbx = trggbx, rggby = trggby; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out, out_stride, px_footprint, rggbx, \ + rggby, roi_in, roi_out, samples) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -867,7 +879,9 @@ void dt_iop_clip_and_zoom_mosaic_third_size_xtrans(uint16_t *const out, const ui // sample (rounded to nearest input pixel) to anti-alias. Higher MP // images need larger filters to avoid artifacts. #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out, out_stride, px_footprint, roi_in, roi_out, xtrans) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -906,7 +920,9 @@ void dt_iop_clip_and_zoom_mosaic_third_size_xtrans_f(float *const out, const flo { const float px_footprint = 1.f / roi_out->scale; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out, out_stride, px_footprint, roi_in, roi_out, xtrans) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -951,7 +967,10 @@ void dt_iop_clip_and_zoom_demosaic_passthrough_monochrome_f_plain(float *out, co const int samples = round(px_footprint); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, px_footprint, roi_in, roi_out, samples) \ + shared(out) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -1093,7 +1112,10 @@ void dt_iop_clip_and_zoom_demosaic_passthrough_monochrome_f_sse2(float *out, con const int samples = round(px_footprint); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, px_footprint, roi_in, roi_out, samples) \ + shared(out) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -1347,7 +1369,10 @@ void dt_iop_clip_and_zoom_demosaic_half_size_f_plain(float *out, const float *co const int rggbx = trggbx, rggby = trggby; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, px_footprint, rggbx, rggby, roi_in, roi_out, samples) \ + shared(out) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -1523,7 +1548,10 @@ void dt_iop_clip_and_zoom_demosaic_half_size_f_sse2(float *out, const float *con const int rggbx = trggbx, rggby = trggby; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, px_footprint, rggbx, rggby, out_stride, roi_in, roi_out, samples) \ + shared(out) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -1712,7 +1740,10 @@ void dt_iop_clip_and_zoom_demosaic_third_size_xtrans_f(float *out, const float * // by non-integer number of samples. #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, in_stride, out_stride, px_footprint, roi_in, roi_out, samples, xtrans) \ + shared(out) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { diff --git a/src/develop/masks/ellipse.c b/src/develop/masks/ellipse.c index f498c84..f422b9c 100644 --- a/src/develop/masks/ellipse.c +++ b/src/develop/masks/ellipse.c @@ -1631,7 +1631,9 @@ static int dt_ellipse_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(iscale, mh, mw, py, px) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -1689,7 +1691,9 @@ static int dt_ellipse_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points, a, b, ta, tb, alpha) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(center, mh, mw) \ + shared(points, a, b, ta, tb, alpha) #else #pragma omp parallel for shared(points, a, b, ta, tb, alpha) #endif @@ -1721,7 +1725,9 @@ static int dt_ellipse_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop // we fill the output buffer by interpolation #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(h, mw, w) \ + shared(points, buffer) #else #pragma omp parallel for shared(points, buffer) #endif diff --git a/src/develop/masks/gradient.c b/src/develop/masks/gradient.c index 5a22673..f3e5f56 100644 --- a/src/develop/masks/gradient.c +++ b/src/develop/masks/gradient.c @@ -805,7 +805,9 @@ static int dt_gradient_get_mask(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(mh, mw, px, py) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -849,7 +851,9 @@ static int dt_gradient_get_mask(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cosv, hwscale, normf, offset, mh, mw, steepness, sinv) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -879,7 +883,9 @@ static int dt_gradient_get_mask(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t // we fill the mask buffer by interpolation #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(h, mw, w) \ + shared(points, buffer) #else #pragma omp parallel for shared(points, buffer) #endif @@ -933,7 +939,9 @@ static int dt_gradient_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_io #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(iscale, mh, mw, py, px) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -979,7 +987,9 @@ static int dt_gradient_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_io #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cosv, hwscale, mh, mw, normf, offset, sinv, steepness) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -1002,7 +1012,9 @@ static int dt_gradient_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_io // we fill the mask buffer by interpolation #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(h, mw, w) \ + shared(points, buffer) #else #pragma omp parallel for shared(points, buffer) #endif diff --git a/src/develop/masks/group.c b/src/develop/masks/group.c index f5b1ec8..df9c434 100644 --- a/src/develop/masks/group.c +++ b/src/develop/masks/group.c @@ -549,7 +549,9 @@ static int dt_group_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t { #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(bufs) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(bufs) #else #pragma omp parallel for shared(bufs) #endif @@ -566,7 +568,9 @@ static int dt_group_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t { #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(bufs, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, op, width) \ + shared(bufs, buffer) #else #pragma omp parallel for shared(bufs, buffer) #endif @@ -582,7 +586,9 @@ static int dt_group_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t { #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(bufs, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, op, width) \ + shared(bufs, buffer) #else #pragma omp parallel for shared(bufs, buffer) #endif @@ -603,7 +609,9 @@ static int dt_group_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t { #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(bufs, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, op, width) \ + shared(bufs, buffer) #else #pragma omp parallel for shared(bufs, buffer) #endif @@ -621,7 +629,9 @@ static int dt_group_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t { #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(bufs, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, op, width) \ + shared(bufs, buffer) #else #pragma omp parallel for shared(bufs, buffer) #endif @@ -642,7 +652,9 @@ static int dt_group_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_t { #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(bufs, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, op, width) \ + shared(bufs, buffer) #else #pragma omp parallel for shared(bufs, buffer) #endif diff --git a/src/develop/masks/circle.c b/src/develop/masks/circle.c index 7ba7edb..424d890 100644 --- a/src/develop/masks/circle.c +++ b/src/develop/masks/circle.c @@ -937,7 +937,9 @@ static int dt_circle_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_ #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(iscale, mh, mw, px, py) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -976,7 +978,9 @@ static int dt_circle_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_ #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(center, mh, mw, radius2, total2) \ + shared(points) #else #pragma omp parallel for shared(points) #endif @@ -1003,7 +1007,9 @@ static int dt_circle_get_mask_roi(dt_iop_module_t *module, dt_dev_pixelpipe_iop_ // we fill the output buffer by interpolation #ifdef _OPENMP #if !defined(__SUNOS__) && !defined(__NetBSD__) -#pragma omp parallel for default(none) shared(points, buffer) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(h, mw, w) \ + shared(points, buffer) #else #pragma omp parallel for shared(points, buffer) #endif diff --git a/src/develop/pixelpipe_hb.c b/src/develop/pixelpipe_hb.c index 290c4cd..97057b3 100644 --- a/src/develop/pixelpipe_hb.c +++ b/src/develop/pixelpipe_hb.c @@ -689,7 +689,10 @@ static int dt_dev_pixelpipe_process_rec(dt_dev_pixelpipe_t *pipe, dt_develop_t * const int cp_height = MIN(roi_out->height, pipe->iheight - in_y); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(pipe, roi_out, roi_in, output) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(bpp, cp_height, cp_width, in_x, in_y) \ + shared(pipe, roi_out, roi_in, output) \ + schedule(static) #endif for(int j = 0; j < cp_height; j++) memcpy(((char *)*output) + (size_t)bpp * j * roi_out->width, @@ -791,7 +794,10 @@ static int dt_dev_pixelpipe_process_rec(dt_dev_pixelpipe_t *pipe, dt_develop_t * else { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(roi_out, roi_in, output, input) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in_bpp, out_bpp) \ + shared(roi_out, roi_in, output, input) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) memcpy(((char *)*output) + (size_t)out_bpp * j * roi_out->width, @@ -800,7 +806,10 @@ static int dt_dev_pixelpipe_process_rec(dt_dev_pixelpipe_t *pipe, dt_develop_t * } #else // don't HAVE_OPENCL #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(roi_out, roi_in, output, input) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in_bpp, out_bpp) \ + shared(roi_out, roi_in, output, input) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) memcpy(((char *)*output) + (size_t)out_bpp * j * roi_out->width, diff --git a/src/develop/tiling.c b/src/develop/tiling.c index fdb5261..c8b363a 100644 --- a/src/develop/tiling.c +++ b/src/develop/tiling.c @@ -753,7 +753,10 @@ static void _default_process_tiling_ptp(struct dt_iop_module_t *self, struct dt_ /* prepare input tile buffer */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(input, width, ioffs) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ht, in_bpp, ipitch, ivoid, wd) \ + shared(input, width, ioffs) \ + schedule(static) #endif for(size_t j = 0; j < ht; j++) memcpy((char *)input + j * wd * in_bpp, (char *)ivoid + ioffs + j * ipitch, (size_t)wd * in_bpp); @@ -794,7 +797,10 @@ static void _default_process_tiling_ptp(struct dt_iop_module_t *self, struct dt_ /* copy "good" part of tile to output buffer */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(ooffs, output, width, origin, region) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(opitch, out_bpp, ovoid, wd) \ + shared(ooffs, output, width, origin, region) \ + schedule(static) #endif for(size_t j = 0; j < region[1]; j++) memcpy((char *)ovoid + ooffs + j * opitch, @@ -1096,7 +1102,10 @@ static void _default_process_tiling_roi(struct dt_iop_module_t *self, struct dt_ } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(input, ioffs, iroi_full) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in_bpp, ipitch, ivoid) \ + shared(input, ioffs, iroi_full) \ + schedule(static) #endif for(size_t j = 0; j < iroi_full.height; j++) memcpy((char *)input + j * iroi_full.width * in_bpp, (char *)ivoid + ioffs + j * ipitch, @@ -1125,7 +1134,10 @@ static void _default_process_tiling_roi(struct dt_iop_module_t *self, struct dt_ const int origin_x = oroi_good.x - oroi_full.x; const int origin_y = oroi_good.y - oroi_full.y; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(ooffs, output, oroi_good, oroi_full) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(opitch, origin_x, origin_y, out_bpp, ovoid) \ + shared(ooffs, output, oroi_good, oroi_full) \ + schedule(static) #endif for(size_t j = 0; j < oroi_good.height; j++) memcpy((char *)ovoid + ooffs + j * opitch, @@ -1408,7 +1420,10 @@ static int _default_process_tiling_cl_ptp(struct dt_iop_module_t *self, struct d { /* prepare pinned input tile buffer: copy part of input image */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(input_buffer, width, ioffs, wd, ht) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in_bpp, ipitch, ivoid) \ + shared(input_buffer, width, ioffs, wd, ht) \ + schedule(static) #endif for(size_t j = 0; j < ht; j++) memcpy((char *)input_buffer + j * wd * in_bpp, (char *)ivoid + ioffs + j * ipitch, @@ -1863,7 +1878,9 @@ static int _default_process_tiling_cl_roi(struct dt_iop_module_t *self, struct d { /* prepare pinned input tile buffer: copy part of input image */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(input_buffer, width, ioffs, iroi_full) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in_bpp, ipitch, ivoid) \ + shared(input_buffer, width, ioffs, iroi_full) schedule(static) #endif for(size_t j = 0; j < iroi_full.height; j++) memcpy((char *)input_buffer + j * iroi_full.width * in_bpp, (char *)ivoid + ioffs + j * ipitch, @@ -1910,8 +1927,10 @@ static int _default_process_tiling_cl_roi(struct dt_iop_module_t *self, struct d /* copy "good" part of tile from pinned output buffer to output image */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(ooffs, output_buffer, oroi_full, oorigin, \ - oregion) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ipitch, opitch, ovoid, out_bpp) \ + shared(ooffs, output_buffer, oroi_full, oorigin, oregion) \ + schedule(static) #endif for(size_t j = 0; j < oregion[1]; j++) memcpy((char *)ovoid + ooffs + j * opitch, diff --git a/src/gui/draw.h b/src/gui/draw.h index 50d4550..dcaaf21 100644 --- a/src/gui/draw.h +++ b/src/gui/draw.h @@ -24,6 +24,7 @@ #include "config.h" #endif +#include "common/darktable.h" #include "common/curve_tools.h" #include #include @@ -229,14 +230,20 @@ static inline void dt_draw_curve_calc_values(dt_draw_curve_t *c, const float min if(x) { #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(x) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(res) \ + shared(x) \ + schedule(static) #endif for(int k = 0; k < res; k++) x[k] = k * (1.0f / res); } if(y) { #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(y, c) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(min, max, res) \ + shared(y, c) \ + schedule(static) #endif for(int k = 0; k < res; k++) y[k] = min + (max - min) * c->csample.m_Samples[k] * (1.0f / 0x10000); } diff --git a/src/iop/hotpixels.c b/src/iop/hotpixels.c index 0c07f7d..74bc0db 100644 --- a/src/iop/hotpixels.c +++ b/src/iop/hotpixels.c @@ -117,7 +117,11 @@ static int process_bayer(const dt_iop_hotpixels_data_t *data, int fixed = 0; #ifdef _OPENMP -#pragma omp parallel for default(none) reduction(+ : fixed) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ivoid, markfixed, min_neighbours, multiplier, ovoid, \ + roi_out, threshold, width, widthx2) \ + reduction(+ : fixed) \ + schedule(static) #endif for(int row = 2; row < roi_out->height - 2; row++) { @@ -214,7 +218,12 @@ static int process_xtrans(const dt_iop_hotpixels_data_t *data, int fixed = 0; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(offsets) reduction(+ : fixed) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ivoid, markfixed, min_neighbours, multiplier, ovoid, \ + roi_out, threshold, xtrans, width) \ + shared(offsets) \ + reduction(+ : fixed) \ + schedule(static) #endif for(int row = 2; row < roi_out->height - 2; row++) { diff --git a/src/iop/lens.c b/src/iop/lens.c index ff0a72a..d9fea09 100644 --- a/src/iop/lens.c +++ b/src/iop/lens.c @@ -370,7 +370,11 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c void *buf = dt_alloc_align(16, bufsize * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(buf, modifier) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(bufsize, ch, ch_width, d, interpolation, ivoid, \ + mask_display, ovoid, roi_in, roi_out) \ + shared(buf, modifier) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -424,7 +428,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c if(modflags & LF_MODIFY_VIGNETTING) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(modifier) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, pixelformat, roi_out, ovoid) \ + shared(modifier) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -446,7 +453,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c if(modflags & LF_MODIFY_VIGNETTING) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(buf, modifier) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, pixelformat, roi_in) \ + shared(buf, modifier) \ + schedule(static) #endif for(int y = 0; y < roi_in->height; y++) { @@ -465,7 +475,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c void *buf2 = dt_alloc_align(16, buf2size * sizeof(float) * dt_get_num_threads()); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(buf2, buf, modifier) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(buf2size, ch, ch_width, d, interpolation, mask_display, ovoid, roi_in, roi_out) \ + shared(buf2, buf, modifier) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -616,7 +629,10 @@ int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_m if(modflags & (LF_MODIFY_TCA | LF_MODIFY_DISTORTION | LF_MODIFY_GEOMETRY | LF_MODIFY_SCALE)) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmpbuf, d, modifier) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(tmpbufwidth, roi_out) \ + shared(tmpbuf, d, modifier) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -652,7 +668,10 @@ int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_m if(modflags & LF_MODIFY_VIGNETTING) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmpbuf, modifier, d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, pixelformat, roi_out) \ + shared(tmpbuf, modifier, d) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -691,7 +710,10 @@ int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_m if(modflags & LF_MODIFY_VIGNETTING) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmpbuf, modifier, d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, pixelformat, roi_in) \ + shared(tmpbuf, modifier, d) \ + schedule(static) #endif for(int y = 0; y < roi_in->height; y++) { @@ -725,7 +747,10 @@ int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_m if(modflags & (LF_MODIFY_TCA | LF_MODIFY_DISTORTION | LF_MODIFY_GEOMETRY | LF_MODIFY_SCALE)) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmpbuf, d, modifier) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(tmpbufwidth, roi_out) \ + shared(tmpbuf, d, modifier) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -888,7 +913,10 @@ void modify_roi_in(struct dt_iop_module_t *self, struct dt_dev_pixelpipe_iop_t * float *const buf = dt_alloc_align(16, nbpoints * 2 * 3 * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel default(none) shared(modifier) reduction(min : xm, ym) reduction(max : xM, yM) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(aheight, awidth, buf, height, nbpoints, width, xoff, \ + xstep, yoff, ystep) \ + shared(modifier) reduction(min : xm, ym) reduction(max : xM, yM) #endif { #ifdef _OPENMP diff --git a/src/develop/imageop_math.h b/src/develop/imageop_math.h index 7203fa7..0b4acf8 100644 --- a/src/develop/imageop_math.h +++ b/src/develop/imageop_math.h @@ -158,7 +158,10 @@ static inline float dt_iop_eval_exp(const float *const coeff, const float x) static inline void dt_iop_alpha_copy(const void *ivoid, void *ovoid, const int width, const int height) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(ovoid, ivoid) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(ovoid, ivoid) \ + schedule(static) #endif for(int j = 0; j < height; j++) { diff --git a/src/iop/hazeremoval.c b/src/iop/hazeremoval.c index 409e8d2..772a9ed 100644 --- a/src/iop/hazeremoval.c +++ b/src/iop/hazeremoval.c @@ -369,7 +369,9 @@ static void box_max(const gray_image img1, const gray_image img2, const int w) if(img1.data == img2.data) { #ifdef _OPENMP -#pragma omp parallel default(none) private(img2_bak) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(img1, img2, w) \ + private(img2_bak) #endif { img2_bak = new_gray_image(img2.width, 1); @@ -387,7 +389,9 @@ static void box_max(const gray_image img1, const gray_image img2, const int w) else { #ifdef _OPENMP -#pragma omp parallel default(none) private(img2_bak) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(img1, img2, w) \ + private(img2_bak) #endif { #ifdef _OPENMP @@ -398,7 +402,9 @@ static void box_max(const gray_image img1, const gray_image img2, const int w) } } #ifdef _OPENMP -#pragma omp parallel default(none) private(img2_bak) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(img1, img2, w) \ + private(img2_bak) #endif { img2_bak = new_gray_image(1, img2.height); @@ -440,7 +446,9 @@ static void box_min(const gray_image img1, const gray_image img2, const int w) if(img1.data == img2.data) { #ifdef _OPENMP -#pragma omp parallel default(none) private(img2_bak) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(img1, img2, w) \ + private(img2_bak) #endif { img2_bak = new_gray_image(img2.width, 1); @@ -458,7 +466,9 @@ static void box_min(const gray_image img1, const gray_image img2, const int w) else { #ifdef _OPENMP -#pragma omp parallel default(none) private(img2_bak) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(img1, img2, w) \ + private(img2_bak) #endif { #ifdef _OPENMP @@ -469,7 +479,9 @@ static void box_min(const gray_image img1, const gray_image img2, const int w) } } #ifdef _OPENMP -#pragma omp parallel default(none) private(img2_bak) +#pragma omp parallel default(none) \ + dt_omp_firstprivate(img1, img2, w) \ + private(img2_bak) #endif { img2_bak = new_gray_image(1, img2.height); @@ -490,7 +502,9 @@ static void dark_channel(const const_rgb_image img1, const gray_image img2, cons { const size_t size = (size_t)img1.height * img1.width; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(img1, img2, size) \ + schedule(static) #endif for(size_t i = 0; i < size; i++) { @@ -509,7 +523,9 @@ static void transition_map(const const_rgb_image img1, const gray_image img2, co { const size_t size = (size_t)img1.height * img1.width; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(A0, img1, img2, size, strength) \ + schedule(static) #endif for(size_t i = 0; i < size; i++) { @@ -782,7 +798,10 @@ static float ambient_light(const const_rgb_image img, int w1, rgb_pixel *pA0) size_t N_bright_hazy = 0; const float *const data = dark_ch.data; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) reduction(+ : N_bright_hazy, A0_r, A0_g, A0_b) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(crit_brightness, crit_haze_level, data, img, size) \ + schedule(static) \ + reduction(+ : N_bright_hazy, A0_r, A0_g, A0_b) #endif for(size_t i = 0; i < size; i++) { @@ -898,7 +917,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const gray_image c_trans_map_filtered = trans_map_filtered; #ifdef _OPENMP // use dynamic load ballancing as tiles may have varying size -#pragma omp parallel for default(none) schedule(dynamic) collapse(2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(c_trans_map, c_trans_map_filtered, height, img_in, w2, eps, tile_width, width) \ + schedule(dynamic) \ + collapse(2) #endif for(int j = 0; j < height; j += tile_width) { @@ -914,7 +936,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const = fmaxf(expf(-distance * distance_max), 1.f / 1024); // minimum allowed value for transition map const float *const c_A0 = A0; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(c_A0, c_trans_map_filtered, img_in, img_out, size, t_min) \ + schedule(static) #endif for(size_t i = 0; i < size; i++) { diff --git a/src/iop/ashift.c b/src/iop/ashift.c index 1245178..8650cf9 100644 --- a/src/iop/ashift.c +++ b/src/iop/ashift.c @@ -878,7 +878,10 @@ int distort_transform(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, floa const float cy = fullheight * data->ct; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(points, points_count, homograph) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cx, cy) \ + shared(points, points_count, homograph) \ + schedule(static) #endif for(size_t i = 0; i < points_count * 2; i += 2) { @@ -912,7 +915,10 @@ int distort_backtransform(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float cy = fullheight * data->ct; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(points, points_count, ihomograph) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cx, cy) \ + shared(points, points_count, ihomograph) \ + schedule(static) #endif for(size_t i = 0; i < points_count * 2; i += 2) { @@ -1062,7 +1068,10 @@ static void rgb2grey256(const float *in, double *out, const int width, const int const int ch = 4; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(in, out) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -1089,7 +1098,10 @@ static void edge_enhance_1d(const double *in, double *out, const int width, cons const double *kernel = (dir == ASHIFT_ENHANCE_HORIZONTAL) ? (const double *)hkernel : (const double *)vkernel; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out, kernel) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(in, out, kernel) \ + schedule(static) #endif // loop over image pixels and perform sobel convolution for(int j = khwidth; j < height - khwidth; j++) @@ -1113,7 +1125,10 @@ static void edge_enhance_1d(const double *in, double *out, const int width, cons } #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(out) \ + schedule(static) #endif // border fill in output buffer, so we don't get pseudo lines at image frame for(int j = 0; j < height; j++) @@ -1155,7 +1170,10 @@ static int edge_enhance(const double *in, double *out, const int width, const in // calculate absolute values #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(Gx, Gy, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(Gx, Gy, out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) { @@ -1202,7 +1220,10 @@ static int detail_enhance(const float *in, float *out, const int width, const in // convert RGB input to Lab, use output buffer for intermediate storage #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(in, out) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -1231,7 +1252,10 @@ static int detail_enhance(const float *in, float *out, const int width, const in // convert resulting Lab to RGB output #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(out) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -1251,7 +1275,10 @@ static int detail_enhance(const float *in, float *out, const int width, const in static void gamma_correct(const float *in, float *out, const int width, const int height) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(in, out) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -2835,7 +2862,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(ihomograph, interpolation) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ch_width, cx, cy, ivoid, ovoid, roi_in, roi_out) \ + shared(ihomograph, interpolation) \ + schedule(static) #endif // go over all pixels of output image for(int j = 0; j < roi_out->height; j++) diff --git a/src/iop/globaltonemap.c b/src/iop/globaltonemap.c index 45493e3..e71e744 100644 --- a/src/iop/globaltonemap.c +++ b/src/iop/globaltonemap.c @@ -139,7 +139,10 @@ static inline void process_reinhard(struct dt_iop_module_t *self, dt_dev_pixelpi const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, roi_out) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -216,7 +219,10 @@ static inline void process_drago(struct dt_iop_module_t *self, dt_dev_pixelpipe_ const float bl = logf(fmaxf(eps, data->drago.bias)) / logf(0.5); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, lwmax) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, bl, ldc, roi_out) \ + shared(in, out, lwmax) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -240,7 +246,10 @@ static inline void process_filmic(struct dt_iop_module_t *self, dt_dev_pixelpipe const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, roi_out) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/colorcontrast.c b/src/iop/colorcontrast.c index f88dd15..b7de39d 100644 --- a/src/iop/colorcontrast.c +++ b/src/iop/colorcontrast.c @@ -153,7 +153,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const if(d->unbound) { #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, d, in, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -166,7 +168,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const else { #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, d, in, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -200,7 +204,11 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c // iterate over all output pixels (same coordinates as input) #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(d) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, max, min, offset, ovoid, roi_in, roi_out, \ + scale, unbound) \ + shared(d) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/colorize.c b/src/iop/colorize.c index 23f0340..5fb0db7 100644 --- a/src/iop/colorize.c +++ b/src/iop/colorize.c @@ -144,7 +144,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float Lmlmix = L - (mix * 100.0f) / 2.0f; #ifdef _OPENMP -#pragma omp parallel for default(none) private(in, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(a, b, ch, ivoid, Lmlmix, mix, ovoid, roi_out) \ + private(in, out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/nlmeans.c b/src/iop/nlmeans.c index a9c1e5d..d3e6cd7 100644 --- a/src/iop/nlmeans.c +++ b/src/iop/nlmeans.c @@ -410,7 +410,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // do this in parallel with a little threading overhead. could parallelize the outer loops with a bit more // memory #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) firstprivate(inited_slide) shared(kj, ki, Sa) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ivoid, norm2, ovoid, P, roi_in, roi_out, sharpness) \ + firstprivate(inited_slide) \ + shared(kj, ki, Sa) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -496,7 +500,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const out = ((float *const)ovoid); #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, in, invert, out, roi_out, weight) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -554,7 +561,11 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c // do this in parallel with a little threading overhead. could parallelize the outer loops with a bit more // memory #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) firstprivate(inited_slide) shared(kj, ki, Sa) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ivoid, norm2, ovoid, P, roi_in, roi_out, sharpness) \ + firstprivate(inited_slide) \ + shared(kj, ki, Sa) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -686,7 +697,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 weight = _mm_set_ps(1.0f, d->chroma, d->chroma, d->luma); const __m128 invert = _mm_sub_ps(_mm_set1_ps(1.0f), weight); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(d) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(invert, ivoid, ovoid, roi_out, weight) \ + shared(d) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/lowlight.c b/src/iop/lowlight.c index 59b56e9..24cf8a6 100644 --- a/src/iop/lowlight.c +++ b/src/iop/lowlight.c @@ -132,7 +132,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const dt_Lab_to_XYZ(Lab_sw, XYZ_sw); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(d, XYZ_sw) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, i, o, roi_out) \ + shared(d, XYZ_sw) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/overexposed.c b/src/iop/overexposed.c index 2969f2b..a417ec7 100644 --- a/src/iop/overexposed.c +++ b/src/iop/overexposed.c @@ -125,7 +125,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const out = (float *const)ovoid; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, lower, lower_color, out, roi_out, \ + upper, upper_color) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -174,7 +177,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 lower_color = _mm_load_ps(dt_iop_overexposed_colors[colorscheme][1]); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, lower, lower_color, ovoid, roi_out, \ + mupper, mlower, upper, upper_color) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/clipping.c b/src/iop/clipping.c index 3e72040..42bb0ca 100644 --- a/src/iop/clipping.c +++ b/src/iop/clipping.c @@ -844,7 +844,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const && roi_in->height == roi_out->height) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(d) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, ovoid, roi_out) \ + shared(d) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -870,8 +873,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const keystone_get_matrix(k_space, kxa, kxb, kxc, kxd, kya, kyb, kyc, kyd, &ma, &mb, &md, &me, &mg, &mh); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(d, interpolation, k_space, ma, mb, md, me, \ - mg, mh) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ch_width, ivoid, kxa, kya, ovoid, roi_in, roi_out) \ + shared(d, interpolation, k_space, ma, mb, md, me, mg, mh) \ + schedule(static) #endif // (slow) point-by-point transformation. // TODO: optimize with scanlines and linear steps between? diff --git a/src/iop/rawdenoise.c b/src/iop/rawdenoise.c index 0a4773f..86173b0 100644 --- a/src/iop/rawdenoise.c +++ b/src/iop/rawdenoise.c @@ -235,7 +235,10 @@ static void wavelet_denoise(const float *const in, float *const out, const dt_io const int halfheight = roi->height / 2 + (roi->height & (~c) & 1); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(c) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, fimg, roi, size, halfwidth) \ + shared(c) \ + schedule(static) #endif for(int row = c & 1; row < roi->height; row += 2) { @@ -255,7 +258,10 @@ static void wavelet_denoise(const float *const in, float *const out, const dt_io // filter horizontally and transpose #ifdef _OPENMP -#pragma omp parallel for default(none) shared(lev) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, halfheight, halfwidth, pass1, pass2) \ + shared(lev) \ + schedule(static) #endif for(int col = 0; col < halfwidth; col++) { @@ -264,7 +270,10 @@ static void wavelet_denoise(const float *const in, float *const out, const dt_io } // filter vertically and transpose back #ifdef _OPENMP -#pragma omp parallel for default(none) shared(lev) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, halfheight, halfwidth, pass2, pass3) \ + shared(lev) \ + schedule(static) #endif for(int row = 0; row < halfheight; row++) { @@ -274,7 +283,9 @@ static void wavelet_denoise(const float *const in, float *const out, const dt_io const float thold = threshold * noise[lev]; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(lev) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, halfheight, halfwidth, pass1, pass3, thold) \ + shared(lev) #endif for(size_t i = 0; i < (size_t)halfwidth * halfheight; i++) { @@ -286,7 +297,10 @@ static void wavelet_denoise(const float *const in, float *const out, const dt_io lastpass = pass3; } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(c, lastpass) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, halfwidth, out, roi) \ + shared(c, lastpass) \ + schedule(static) #endif for(int row = c & 1; row < roi->height; row += 2) { @@ -390,7 +404,10 @@ static void wavelet_denoise_xtrans(const float *const in, float *out, const dt_i memset(fimg, 0, size * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(c) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, height, in, roi, size, width, xtrans) \ + shared(c) \ + schedule(static) #endif for(int row = (c != 1); row < height - 1; row++) { @@ -423,20 +440,28 @@ static void wavelet_denoise_xtrans(const float *const in, float *out, const dt_i // filter horizontally and transpose #ifdef _OPENMP -#pragma omp parallel for default(none) shared(lev) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, height, pass1, pass2, width) \ + shared(lev) \ + schedule(static) #endif for(int col = 0; col < width; col++) hat_transform(fimg + pass2 + (size_t)col * height, fimg + pass1 + col, width, height, 1 << lev); // filter vertically and transpose back #ifdef _OPENMP -#pragma omp parallel for default(none) shared(lev) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, height, pass2, pass3, width) \ + shared(lev) \ + schedule(static) #endif for(int row = 0; row < height; row++) hat_transform(fimg + pass3 + (size_t)row * width, fimg + pass2 + row, height, width, 1 << lev); const float thold = threshold * noise[lev]; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(lev) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(fimg, pass1, pass3, size, thold) \ + shared(lev) #endif for(size_t i = 0; i < size; i++) { @@ -449,7 +474,10 @@ static void wavelet_denoise_xtrans(const float *const in, float *out, const dt_i } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(c, lastpass, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, fimg, roi, width, xtrans) \ + shared(c, lastpass, out) \ + schedule(static) #endif for(int row = 0; row < height; row++) { diff --git a/src/iop/colorchecker.c b/src/iop/colorchecker.c index 1218a59..a32557d 100644 --- a/src/iop/colorchecker.c +++ b/src/iop/colorchecker.c @@ -380,7 +380,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const dt_iop_colorchecker_data_t *const data = (dt_iop_colorchecker_data_t *)piece->data; const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) collapse(2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, data, ivoid, ovoid, roi_in, roi_out) \ + schedule(static) \ + collapse(2) #endif for(int j=0;jheight;j++) { diff --git a/src/iop/colorzones.c b/src/iop/colorzones.c index d798cf4..604b583 100644 --- a/src/iop/colorzones.c +++ b/src/iop/colorzones.c @@ -203,7 +203,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const dt_iop_colorzones_data_t *d = (dt_iop_colorzones_data_t *)(piece->data); const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(d) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, i, o, roi_out) \ + shared(d) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/colorout.c b/src/iop/colorout.c index e9108a0..be9995b 100644 --- a/src/iop/colorout.c +++ b/src/iop/colorout.c @@ -292,7 +292,9 @@ static void process_fastpath_apply_tonecurves(struct dt_iop_module_t *self, dt_d { // apply profile float *const out = (float *const)ovoid; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -307,7 +309,9 @@ static void process_fastpath_apply_tonecurves(struct dt_iop_module_t *self, dt_d { // apply profile float *const out = (float *const)ovoid; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -340,7 +344,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // fprintf(stderr,"Using cmatrix codepath\n"); // convert to rgb using matrix #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -366,7 +372,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const { // fprintf(stderr,"Using xform codepath\n"); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, gamutcheck, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -410,7 +418,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c // fprintf(stderr,"Using cmatrix codepath\n"); // convert to rgb using matrix #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(d, ch, ivoid, ovoid, roi_in, roi_out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -441,7 +451,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c // fprintf(stderr,"Using xform codepath\n"); const __m128 outofgamutpixel = _mm_set_ps(0.0f, 1.0f, 1.0f, 0.0f); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, gamutcheck, outofgamutpixel, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/bilateral.cc b/src/iop/bilateral.cc index 79cf150..2a0cac7 100644 --- a/src/iop/bilateral.cc +++ b/src/iop/bilateral.cc @@ -144,7 +144,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const weights_buf = (float *)malloc(weights_size * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(m, mat, isig2col) private(in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, ovoid, rad, roi_in, roi_out, wd, weights_buf) \ + shared(m, mat, isig2col) \ + private(in, out) \ + schedule(static) #endif for(int j = rad; j < roi_out->height - rad; j++) { diff --git a/src/iop/colorbalance.c b/src/iop/colorbalance.c index 19270e4..a966442 100644 --- a/src/iop/colorbalance.c +++ b/src/iop/colorbalance.c @@ -334,7 +334,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const (gamma[2] != 0.0) ? 1.0 / gamma[2] : 1000000.0 }; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(d) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, gain, gamma_inv, lift, ivoid, ovoid, roi_in, \ + roi_out) \ + shared(d) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_in->width * roi_out->height; k += ch) { @@ -381,7 +385,12 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const (gamma[2] != 0.0) ? 1.0 / gamma[2] : 1000000.0 }; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(d) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, contrast, gain, gamma_inv, grey, ivoid, lift, \ + ovoid, roi_in, roi_out, run_contrast, \ + run_saturation, run_saturation_out) \ + shared(d) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_in->width * roi_out->height; k += ch) { @@ -445,7 +454,12 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const (2.0f - d->gamma[CHANNEL_BLUE]) * (2.0f - d->gamma[CHANNEL_FACTOR])}; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(d) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, contrast, gain, gamma, grey, ivoid, lift, ovoid, \ + roi_in, roi_out, run_contrast, run_saturation, \ + run_saturation_out) \ + shared(d) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_in->width * roi_out->height; k += ch) { @@ -543,7 +557,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c 0.0f); #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, gain, gamma_inv, ivoid, lift, one, ovoid, roi_in, roi_out, zero) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_in->width * roi_out->height; k += ch) { @@ -591,7 +607,13 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 gamma_inv_RGB = _mm_set1_ps(1.0f/2.2f); #ifdef _OPENMP -#pragma omp parallel for SIMD()default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, contrast, gain, gamma_inv, gamma_inv_RGB, \ + gamma_RGB, grey, ivoid, lift, one, ovoid, roi_in, \ + roi_out, run_contrast, run_saturation, \ + run_saturation_out, saturation, saturation_out, \ + zero) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_in->width * roi_out->height; k += ch) { @@ -659,7 +681,12 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c 0.0f); #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, contrast, gain, gamma, grey, ivoid, lift, ovoid, \ + roi_in, roi_out, run_contrast, run_saturation, \ + run_saturation_out, saturation, saturation_out, \ + zero) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_in->width * roi_out->height; k += ch) { diff --git a/src/iop/equalizer_eaw.h b/src/iop/equalizer_eaw.h index 1d58e24..e2afc51 100644 --- a/src/iop/equalizer_eaw.h +++ b/src/iop/equalizer_eaw.h @@ -43,7 +43,11 @@ static void dt_iop_equalizer_wtf(float *buf, float **weight_a, const int l, cons float *const tmp_width_buf = (float *)malloc(width * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(weight_a, buf) private(ch) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, l, st, step, tmp_width_buf, wd, width) \ + shared(weight_a, buf) \ + private(ch) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -73,7 +77,11 @@ static void dt_iop_equalizer_wtf(float *buf, float **weight_a, const int l, cons float *const tmp_height_buf = (float *)malloc(height * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(weight_a, buf) private(ch) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, l, st, step, tmp_height_buf, wd, width) \ + shared(weight_a, buf) \ + private(ch) \ + schedule(static) #endif for(int i = 0; i < width; i++) { @@ -110,7 +118,10 @@ static void dt_iop_equalizer_iwtf(float *buf, float **weight_a, const int l, con float *const tmp_height_buf = (float *)malloc(height * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(weight_a, buf) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, l, st, step, tmp_height_buf, wd, width) \ + shared(weight_a, buf) \ + schedule(static) #endif for(int i = 0; i < width; i++) { @@ -139,7 +150,10 @@ static void dt_iop_equalizer_iwtf(float *buf, float **weight_a, const int l, con float *const tmp_width_buf = (float *)malloc(width * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(weight_a, buf) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, l, st, step, tmp_width_buf, wd, width) \ + shared(weight_a, buf) \ + schedule(static) #endif for(int j = 0; j < height; j++) { diff --git a/src/iop/rotatepixels.c b/src/iop/rotatepixels.c index 519180a..eacc681 100644 --- a/src/iop/rotatepixels.c +++ b/src/iop/rotatepixels.c @@ -257,7 +257,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c const struct dt_interpolation *interpolation = dt_interpolation_new(DT_INTERPOLATION_USERPREF); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(piece, interpolation) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ch_width, ivoid, ovoid, roi_in, roi_out, scale) \ + shared(piece, interpolation) \ + schedule(static) #endif // (slow) point-by-point transformation. // TODO: optimize with scanlines and linear steps between? diff --git a/src/iop/vibrance.c b/src/iop/vibrance.c index c9eaafd..f3fe554 100644 --- a/src/iop/vibrance.c +++ b/src/iop/vibrance.c @@ -101,7 +101,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float amount = (d->amount * 0.01); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(amount, ch, roi_out) \ + shared(in, out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/colorreconstruction.c b/src/iop/colorreconstruction.c index a5e3ffe..ac7e98a 100644 --- a/src/iop/colorreconstruction.c +++ b/src/iop/colorreconstruction.c @@ -365,7 +365,9 @@ static void dt_iop_colorreconstruct_bilateral_splat(dt_iop_colorreconstruct_bila // splat into downsampled grid #ifdef _OPENMP -#pragma omp parallel for default(none) shared(b, precedence, params) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, threshold) \ + shared(b, precedence, params) #endif for(int j = 0; j < b->height; j++) { @@ -439,7 +441,9 @@ static void blur_line(dt_iop_colorreconstruct_Lab_t *buf, const int offset1, con const float w1 = 4.f / 16.f; const float w2 = 1.f / 16.f; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(buf) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(offset1, offset2, offset3, size1, size2, size3) \ + shared(buf) #endif for(int k = 0; k < size1; k++) { @@ -515,7 +519,8 @@ static void dt_iop_colorreconstruct_bilateral_slice(const dt_iop_colorreconstruc const int oy = b->size_x; const int oz = b->size_y * b->size_x; #ifdef _OPENMP -#pragma omp parallel for default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(b, in, out, oy, oz, rescale, roi, threshold) #endif for(int j = 0; j < roi->height; j++) { diff --git a/src/iop/bloom.c b/src/iop/bloom.c index d76ce73..2e03ce0 100644 --- a/src/iop/bloom.c +++ b/src/iop/bloom.c @@ -126,7 +126,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const /* get the thresholded lights into buffer */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data, blurlightness) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, roi_out, scale) \ + shared(data, blurlightness) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -146,7 +149,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const for(int iteration = 0; iteration < BOX_ITERATIONS; iteration++) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(blurlightness) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(hr, roi_out, scanline_buf, size) \ + shared(blurlightness) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -180,7 +186,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const #ifdef _OPENMP -#pragma omp parallel for default(none) shared(blurlightness) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(hr, npoffs, opoffs, roi_out, size, scanline_buf) \ + shared(blurlightness) \ + schedule(static) #endif for(int x = 0; x < roi_out->width; x++) { @@ -214,7 +223,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const /* screen blend lightness with original */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data, blurlightness) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, roi_out) \ + shared(in, out, data, blurlightness) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/rawprepare.c b/src/iop/rawprepare.c index 030fe5b..d985637 100644 --- a/src/iop/rawprepare.c +++ b/src/iop/rawprepare.c @@ -301,7 +301,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const out = (float *const)ovoid; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(csx, csy, d, in, out, roi_in, roi_out) \ + schedule(static) \ + collapse(2) #endif for(int j = 0; j < roi_out->height; j++) { @@ -325,7 +328,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const out = (float *const)ovoid; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(csx, csy, d, in, out, roi_in, roi_out) \ + schedule(static) \ + collapse(2) #endif for(int j = 0; j < roi_out->height; j++) { @@ -353,7 +359,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(3) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, csx, csy, div, in, out, roi_in, roi_out, sub) \ + schedule(static) collapse(3) #endif for(int j = 0; j < roi_out->height; j++) { @@ -387,7 +395,9 @@ void process_sse2(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const vo if(piece->pipe->dsc.filters && piece->dsc_in.channels == 1 && piece->dsc_in.datatype == TYPE_UINT16) { // raw mosaic #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(csx, csy, d, ivoid, ovoid, roi_in, roi_out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -446,7 +456,9 @@ void process_sse2(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const vo else if(piece->pipe->dsc.filters && piece->dsc_in.channels == 1 && piece->dsc_in.datatype == TYPE_FLOAT) { // raw mosaic, fp, unnormalized #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(csx, csy, d, ivoid, ovoid, roi_in, roi_out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -498,7 +510,9 @@ void process_sse2(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const vo const __m128 sub = _mm_load_ps(d->sub), div = _mm_load_ps(d->div); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(csx, csy, div, ivoid, ovoid, roi_in, roi_out, sub) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/filmic.c b/src/iop/filmic.c index 91afe02..9739a62 100644 --- a/src/iop/filmic.c +++ b/src/iop/filmic.c @@ -431,7 +431,9 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c const float saturation = data->global_saturation / 100.0f; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, data, desaturate, ivoid, ovoid, preserve_color, roi_out, saturation) \ + schedule(static) #endif for(size_t k = 0; k < roi_out->height * roi_out->width * ch; k += ch) { @@ -556,7 +558,12 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 one = _mm_set1_ps(1.0f); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(black, black_sse, ch, data, desaturate, dynamic_range, \ + dynamic_range_sse, EPS, grey, grey_sse, ivoid, one, \ + ovoid, power, preserve_color, roi_out, saturation_sse, \ + zero) \ + schedule(static) #endif for(size_t k = 0; k < roi_out->height * roi_out->width * ch; k += ch) { @@ -1421,7 +1428,10 @@ void commit_params(dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pixelpipe_ const float sigma = saturation * saturation * latitude * latitude; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(d) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(center, sigma) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < 65536; k++) { diff --git a/src/iop/monochrome.c b/src/iop/monochrome.c index ba98c59..e8ca0c5 100644 --- a/src/iop/monochrome.c +++ b/src/iop/monochrome.c @@ -153,7 +153,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float sigma2 = (d->size * 128.0) * (d->size * 128.0f); // first pass: evaluate color filter: #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(i, o, roi_out, sigma2) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -180,7 +183,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const dt_bilateral_free(b); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(i, o, roi_out) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/colorin.c b/src/iop/colorin.c index 107b4e3..59914ab 100644 --- a/src/iop/colorin.c +++ b/src/iop/colorin.c @@ -483,7 +483,9 @@ static void process_cmatrix_bm(struct dt_iop_module_t *self, dt_dev_pixelpipe_io // fprintf(stderr, "Using cmatrix codepath\n"); // only color matrix. use our optimized fast path! #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, clipping, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -562,7 +564,9 @@ static void process_cmatrix_fastpath_simple(struct dt_iop_module_t *self, dt_dev // fprintf(stderr, "Using cmatrix codepath\n"); // only color matrix. use our optimized fast path! #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -594,7 +598,9 @@ static void process_cmatrix_fastpath_clipping(struct dt_iop_module_t *self, dt_d // fprintf(stderr, "Using cmatrix codepath\n"); // only color matrix. use our optimized fast path! #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -659,7 +665,9 @@ static void process_cmatrix_proper(struct dt_iop_module_t *self, dt_dev_pixelpip // fprintf(stderr, "Using cmatrix codepath\n"); // only color matrix. use our optimized fast path! #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, clipping, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -755,7 +763,9 @@ static void process_lcms2_bm(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_ // use general lcms2 fallback #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -800,7 +810,9 @@ static void process_lcms2_proper(struct dt_iop_module_t *self, dt_dev_pixelpipe_ // use general lcms2 fallback #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -884,7 +896,10 @@ static void process_sse2_cmatrix_bm(struct dt_iop_module_t *self, dt_dev_pixelpi float *in = (float *)ivoid; float *out = (float *)ovoid; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out, in) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, clipping, cmat, d, lmat, nmat, roi_in, roi_out) \ + shared(out, in) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -956,7 +971,9 @@ static void process_sse2_cmatrix_fastpath_simple(struct dt_iop_module_t *self, d const __m128 cm2 = _mm_set_ps(0.0f, cmat[8], cmat[5], cmat[2]); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, cm0, cm1, cm2, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -994,7 +1011,9 @@ static void process_sse2_cmatrix_fastpath_clipping(struct dt_iop_module_t *self, const __m128 lm2 = _mm_set_ps(0.0f, lmat[8], lmat[5], lmat[2]); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, lm0, lm1, lm2, nm0, nm1, nm2, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -1047,7 +1066,10 @@ static void process_sse2_cmatrix_proper(struct dt_iop_module_t *self, dt_dev_pix float *in = (float *)ivoid; float *out = (float *)ovoid; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out, in) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, clipping, cmat, d, lmat, nmat, roi_in, roi_out) \ + shared(out, in) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -1131,7 +1153,9 @@ static void process_sse2_lcms2_bm(struct dt_iop_module_t *self, dt_dev_pixelpipe // use general lcms2 fallback #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -1179,7 +1203,9 @@ static void process_sse2_lcms2_proper(struct dt_iop_module_t *self, dt_dev_pixel // use general lcms2 fallback #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/basecurve.c b/src/iop/basecurve.c index 72dc2a8..44a3b05 100644 --- a/src/iop/basecurve.c +++ b/src/iop/basecurve.c @@ -823,7 +823,9 @@ static inline void apply_ev_and_curve( const float *const unbounded_coeffs) { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, in, mul, out, table, unbounded_coeffs, width) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) { @@ -852,7 +854,10 @@ static inline void compute_features( // 2) saturation // 3) local contrast (handled in laplacian form later) #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) collapse(2) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(col, ht, wd) \ + schedule(static) \ + collapse(2) #endif for(int j=0;jdata); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, in, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/shadhi.c b/src/iop/shadhi.c index b1ed33b..930b066 100644 --- a/src/iop/shadhi.c +++ b/src/iop/shadhi.c @@ -373,7 +373,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // invert and desaturate #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(roi_out) \ + shared(out) \ + schedule(static) #endif for(size_t j = 0; j < (size_t)roi_out->width * roi_out->height * 4; j += 4) { @@ -391,7 +394,13 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, compress, doublemax, flags, halfmax, height, \ + highlights, highlights_ccorrect, lmax, lmin, \ + low_approximation, max, min, shadows, \ + shadows_ccorrect, unbound_mask, whitepoint, width) \ + shared(in, out) \ + schedule(static) #endif for(size_t j = 0; j < (size_t)width * height * ch; j += ch) { diff --git a/src/iop/sharpen.c b/src/iop/sharpen.c index 06f7e71..ffd00fd 100644 --- a/src/iop/sharpen.c +++ b/src/iop/sharpen.c @@ -318,7 +318,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // gauss blur the image horizontally #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, mat, rad, roi_in, roi_out, tmp, wd4) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -358,7 +360,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // gauss blur the image vertically #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, mat, ovoid, rad, roi_in, roi_out, tmp, wd4) \ + schedule(static) #endif for(int j = rad; j < roi_out->height - wd4 * 4 + rad; j++) { @@ -385,7 +389,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const } } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, mat, ovoid, rad, roi_in, roi_out, tmp, wd4) \ + schedule(static) #endif for(int j = roi_out->height - wd4 * 4 + rad; j < roi_out->height - rad; j++) { @@ -418,7 +424,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const dt_free_align(tmp); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, ovoid, rad, roi_out) \ + schedule(static) #endif for(int j = rad; j < roi_out->height - rad; j++) { @@ -429,7 +437,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, data, ivoid, ovoid, roi_out) \ + schedule(static) #endif // subtract blurred image, if diff > thrs, add *amount to original image for(int j = 0; j < roi_out->height; j++) @@ -502,7 +512,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c // gauss blur the image horizontally #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, mat, rad, roi_in, roi_out, tmp, wd4) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -543,7 +555,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c // gauss blur the image vertically #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, mat, ovoid, rad, roi_in, roi_out, tmp, wd4) \ + schedule(static) #endif for(int j = rad; j < roi_out->height - wd4 * 4 + rad; j++) { @@ -571,7 +585,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c } } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, mat, ovoid, rad, roi_in, roi_out, tmp, wd4) \ + schedule(static) #endif for(int j = roi_out->height - wd4 * 4 + rad; j < roi_out->height - rad; j++) { @@ -606,7 +622,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c dt_free_align(tmp); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, ovoid, rad, roi_out) \ + schedule(static) #endif for(int j = rad; j < roi_out->height - rad; j++) { @@ -617,7 +635,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, ovoid, roi_out) \ + shared(data) \ + schedule(static) #endif // subtract blurred image, if diff > thrs, add *amount to original image for(int j = 0; j < roi_out->height; j++) diff --git a/src/iop/highpass.c b/src/iop/highpass.c index 7920c7e..e765b59 100644 --- a/src/iop/highpass.c +++ b/src/iop/highpass.c @@ -288,7 +288,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const /* create inverted image and then blur */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, roi_out) \ + shared(in, out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) out[ch * k] = 100.0f - LCLIP(in[ch * k]); // only L in Lab space @@ -365,7 +368,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float contrast_scale = ((data->contrast / 100.0) * 7.5); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, contrast_scale, roi_out) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/invert.c b/src/iop/invert.c index f5ceae4..6106906 100644 --- a/src/iop/invert.c +++ b/src/iop/invert.c @@ -250,7 +250,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const if(filters == 9u) { // xtrans float mosaiced #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(film_rgb_f, in, out, roi_out, xtrans) \ + schedule(static) \ + collapse(2) #endif for(int j = 0; j < roi_out->height; j++) { @@ -267,7 +270,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const { // bayer float mosaiced #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(film_rgb_f, filters, in, out, roi_out) \ + schedule(static) \ + collapse(2) #endif for(int j = 0; j < roi_out->height; j++) { @@ -285,7 +291,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, d, in, out, roi_out) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -326,7 +335,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 val_max = _mm_set1_ps(1.0f); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(film_rgb_f, ivoid, ovoid, roi_out, val_max, val_min, xtrans) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -381,7 +392,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 val_max = _mm_set1_ps(1.0f); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(film_rgb_f, filters, ivoid, ovoid, roi_out, val_max, val_min) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -423,7 +436,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 film = _mm_set_ps(1.0f, d->color[2], d->color[1], d->color[0]); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, film, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/gamma.c b/src/iop/gamma.c index 7cbd8a7..a9fbaad 100644 --- a/src/iop/gamma.c +++ b/src/iop/gamma.c @@ -199,7 +199,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const { const float yellow[3] = { 1.0f, 1.0f, 0.0f }; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, i, mask_display, o, roi_out, yellow) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -222,7 +224,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const { const float yellow[3] = { 1.0f, 1.0f, 0.0f }; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, i, mask_display, o, roi_out, yellow) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -243,7 +247,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const { const float yellow[3] = { 1.0f, 1.0f, 0.0f }; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, i, o, roi_out, yellow) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -264,7 +270,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const else { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, i, o, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/defringe.c b/src/iop/defringe.c index fdab4dc..950c101 100644 --- a/src/iop/defringe.c +++ b/src/iop/defringe.c @@ -251,8 +251,11 @@ void process(struct dt_iop_module_t *module, dt_dev_pixelpipe_iop_t *piece, cons } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(width, height, \ - d) reduction(+ : avg_edge_chroma) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, out) \ + shared(width, height, d) \ + reduction(+ : avg_edge_chroma) \ + schedule(static) #endif for(int v = 0; v < height; v++) { @@ -288,8 +291,11 @@ void process(struct dt_iop_module_t *module, dt_dev_pixelpipe_iop_t *piece, cons #ifdef _OPENMP // dynamically/guided scheduled due to possible uneven edge-chroma distribution (thanks to rawtherapee code // for this hint!) -#pragma omp parallel for default(none) shared(width, height, d, xy_small, xy_avg, xy_artifact) \ - firstprivate(thresh, avg_edge_chroma) schedule(guided, 32) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, out, samples_avg, samples_small) \ + shared(width, height, d, xy_small, xy_avg, xy_artifact) \ + firstprivate(thresh, avg_edge_chroma) \ + schedule(guided, 32) #endif for(int v = 0; v < height; v++) { diff --git a/src/iop/atrous.c b/src/iop/atrous.c index d915429..16a4395 100644 --- a/src/iop/atrous.c +++ b/src/iop/atrous.c @@ -324,7 +324,9 @@ static void eaw_decompose(float *const out, const float *const in, float *const /* The first "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, mult, out, sharpen, width) \ + schedule(static) #endif for(int j = 0; j < 2 * mult; j++) { @@ -345,7 +347,9 @@ static void eaw_decompose(float *const out, const float *const in, float *const } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, mult, out, sharpen, width) \ + schedule(static) #endif for(int j = 2 * mult; j < height - 2 * mult; j++) { @@ -402,7 +406,9 @@ static void eaw_decompose(float *const out, const float *const in, float *const /* The last "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, mult, out, sharpen, width) \ + schedule(static) #endif for(int j = height - 2 * mult; j < height; j++) { @@ -439,7 +445,9 @@ static void eaw_decompose_sse2(float *const out, const float *const in, float *c /* The first "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, mult, out, sharpen, width) \ + schedule(static) #endif for(int j = 0; j < 2 * mult; j++) { @@ -460,7 +468,9 @@ static void eaw_decompose_sse2(float *const out, const float *const in, float *c } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, mult, out, sharpen, width) \ + schedule(static) #endif for(int j = 2 * mult; j < height - 2 * mult; j++) { @@ -517,7 +527,9 @@ static void eaw_decompose_sse2(float *const out, const float *const in, float *c /* The last "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, mult, out, sharpen, width) \ + schedule(static) #endif for(int j = height - 2 * mult; j < height; j++) { @@ -558,7 +570,10 @@ static void eaw_synthesize(float *const out, const float *const in, const float const float boost[4] = { boostf[0], boostf[1], boostf[2], boostf[3] }; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(boost, detail, height, in, out, width, threshold) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)4 * width * height; k += 4) { @@ -580,7 +595,9 @@ static void eaw_synthesize_sse2(float *const out, const float *const in, const f const __m128 boost = _mm_set_ps(boostf[3], boostf[2], boostf[1], boostf[0]); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(boost, detail, height, in, out, threshold, width) \ + schedule(static) #endif for(int j = 0; j < height; j++) { diff --git a/src/iop/colisa.c b/src/iop/colisa.c index 99acf04..777306b 100644 --- a/src/iop/colisa.c +++ b/src/iop/colisa.c @@ -180,7 +180,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, width) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) { @@ -250,7 +253,10 @@ void commit_params(struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pix const float contrastm1sq = boost * (d->contrast - 1.0f) * (d->contrast - 1.0f); const float contrastscale = sqrt(1.0f + contrastm1sq); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(contrastm1sq, contrastscale) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < 0x10000; k++) { @@ -272,7 +278,10 @@ void commit_params(struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pix const float gamma = (d->brightness >= 0.0f) ? 1.0f / (1.0f + d->brightness) : (1.0f - d->brightness); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(gamma) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < 0x10000; k++) { diff --git a/src/iop/profile_gamma.c b/src/iop/profile_gamma.c index 33975bf..c7e9049 100644 --- a/src/iop/profile_gamma.c +++ b/src/iop/profile_gamma.c @@ -333,7 +333,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c const float noise = powf(2.0f, -16.0f); #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(data) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, grey, ivoid, ovoid, roi_out) \ + shared(data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k++) { @@ -356,7 +359,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c case PROFILEGAMMA_GAMMA: { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, ovoid, roi_out) \ + shared(data) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -688,7 +694,10 @@ void commit_params(dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pixelpipe_ if(linear == 0.0) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(gamma) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < 0x10000; k++) d->table[k] = powf(1.00 * k / 0x10000, gamma); } @@ -707,7 +716,10 @@ void commit_params(dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pixelpipe_ c = 1.0; } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d, a, b, c, g) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(linear) \ + shared(d, a, b, c, g) \ + schedule(static) #endif for(int k = 0; k < 0x10000; k++) { diff --git a/src/iop/rawoverexposed.c b/src/iop/rawoverexposed.c index 0016762..aafd4de 100644 --- a/src/iop/rawoverexposed.c +++ b/src/iop/rawoverexposed.c @@ -188,7 +188,12 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c void *coordbuf = dt_alloc_align(16, coordbufsize * sizeof(float) * dt_get_num_threads()); #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(self, coordbuf, buf) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, color, coordbufsize, d, \ + dt_iop_rawoverexposed_colors, filters, priority, mode, \ + out, raw, roi_in, roi_out, xtrans) \ + shared(self, coordbuf, buf) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -311,7 +316,10 @@ int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_m if(coordbuf == NULL) goto error; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) shared(self, coordbuf, buf) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(height, roi_in, roi_out, width) \ + shared(self, coordbuf, buf) \ + schedule(static) #endif for(int j = 0; j < height; j++) { diff --git a/src/iop/soften.c b/src/iop/soften.c index cee9f87..b20752d 100644 --- a/src/iop/soften.c +++ b/src/iop/soften.c @@ -129,7 +129,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const /* create overexpose image and then blur */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(brightness, ch, in, out, roi_out, saturation) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -154,7 +156,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const for(int iteration = 0; iteration < BOX_ITERATIONS; iteration++) { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, radius, out, roi_out, scanline_buf, scanline_size) \ + schedule(static) #endif /* horizontal blur out into out */ for(int y = 0; y < roi_out->height; y++) @@ -206,7 +210,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int opoffs = -(radius + 1) * roi_out->width; const int npoffs = (radius)*roi_out->width; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, npoffs, opoffs, radius, out, roi_out, \ + scanline_buf, scanline_size) \ + schedule(static) #endif for(int x = 0; x < roi_out->width; x++) { @@ -262,7 +269,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float amount_1 = (1 - (d->amount) / 100.0); #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(amount, amount_1, ch, in, out, roi_out) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -286,7 +296,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const float saturation = data->saturation / 100.0; /* create overexpose image and then blur */ #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, brightness, roi_out, saturation) \ + shared(in, out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -311,7 +324,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c for(int iteration = 0; iteration < BOX_ITERATIONS; iteration++) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, radius, roi_out, scanline_buf, size) \ + shared(out) \ + schedule(static) #endif /* horizontal blur out into out */ for(int y = 0; y < roi_out->height; y++) @@ -344,7 +360,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const int opoffs = -(radius + 1) * roi_out->width; const int npoffs = (radius)*roi_out->width; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, npoffs, opoffs, radius, roi_out, scanline_buf, size) \ + shared(out) \ + schedule(static) #endif for(int x = 0; x < roi_out->width; x++) { @@ -381,7 +400,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 amount = _mm_set1_ps(data->amount / 100.0); const __m128 amount_1 = _mm_set1_ps(1 - (data->amount) / 100.0); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, amount, amount_1, roi_out) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/denoiseprofile.c b/src/iop/denoiseprofile.c index 8eaa05e..5901eee 100644 --- a/src/iop/denoiseprofile.c +++ b/src/iop/denoiseprofile.c @@ -377,7 +377,10 @@ static inline void precondition(const float *const in, float *const buf, const i (b[2] / a[2]) * (b[2] / a[2]) + 3.f / 8.f }; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(a) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(buf, ht, in, sigma2_plus_3_8, wd) \ + shared(a) \ + schedule(static) #endif for(int j = 0; j < ht; j++) { @@ -405,7 +408,10 @@ static inline void backtransform(float *const buf, const int wd, const int ht, c (b[2] / a[2]) * (b[2] / a[2]) + 1.f / 8.f }; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(a) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(buf, ht, sigma2_plus_1_8, wd) \ + shared(a) \ + schedule(static) #endif for(int j = 0; j < ht; j++) { @@ -590,7 +596,9 @@ static void eaw_decompose(float *const out, const float *const in, float *const /* The first "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, inv_sigma2, mult, out, width) \ + schedule(static) #endif for(int j = 0; j < 2 * mult; j++) { @@ -611,7 +619,9 @@ static void eaw_decompose(float *const out, const float *const in, float *const } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, inv_sigma2, mult, out, width) \ + schedule(static) #endif for(int j = 2 * mult; j < height - 2 * mult; j++) { @@ -668,7 +678,9 @@ static void eaw_decompose(float *const out, const float *const in, float *const /* The last "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, inv_sigma2, mult, out, width) \ + schedule(static) #endif for(int j = height - 2 * mult; j < height; j++) { @@ -705,7 +717,9 @@ static void eaw_decompose_sse(float *const out, const float *const in, float *co /* The first "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, inv_sigma2, mult, out, width) \ + schedule(static) #endif for(int j = 0; j < 2 * mult; j++) { @@ -726,7 +740,9 @@ static void eaw_decompose_sse(float *const out, const float *const in, float *co } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, inv_sigma2, mult, out, width) \ + schedule(static) #endif for(int j = 2 * mult; j < height - 2 * mult; j++) { @@ -783,7 +799,9 @@ static void eaw_decompose_sse(float *const out, const float *const in, float *co /* The last "2*mult" lines use the macro with tests because the 5x5 kernel * requires nearest pixel interpolation for at least a pixel in the sum */ #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(detail, filter, height, in, inv_sigma2, mult, out, width) \ + schedule(static) #endif for(int j = height - 2 * mult; j < height; j++) { @@ -824,7 +842,10 @@ static void eaw_synthesize(float *const out, const float *const in, const float const float boost[4] = { boostf[0], boostf[1], boostf[2], boostf[3] }; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(boost, detail, height, in, out, threshold, width) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)4 * width * height; k += 4) { @@ -846,7 +867,9 @@ static void eaw_synthesize_sse2(float *const out, const float *const in, const f const __m128 boost = _mm_set_ps(boostf[3], boostf[2], boostf[1], boostf[0]); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(boost, detail, height, in, out, threshold, width) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -1102,7 +1125,11 @@ static void process_nlmeans(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t // do this in parallel with a little threading overhead. could parallelize the outer loops with a bit more // memory #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) firstprivate(inited_slide) shared(kj, ki, in, Sa) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(d, ovoid, P, roi_in, roi_out) \ + firstprivate(inited_slide) \ + shared(kj, ki, in, Sa) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -1188,7 +1215,9 @@ static void process_nlmeans(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t // normalize #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -1246,7 +1275,11 @@ static void process_nlmeans_sse(struct dt_iop_module_t *self, dt_dev_pixelpipe_i // do this in parallel with a little threading overhead. could parallelize the outer loops with a bit more // memory #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) firstprivate(inited_slide) shared(kj, ki, in, Sa) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ovoid, P, roi_in, roi_out) \ + firstprivate(inited_slide) \ + shared(kj, ki, in, Sa) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -1379,7 +1412,10 @@ static void process_nlmeans_sse(struct dt_iop_module_t *self, dt_dev_pixelpipe_i } // normalize #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(d) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ovoid, roi_out) \ + shared(d) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/spots.c b/src/iop/spots.c index 80601de..f4fa791 100644 --- a/src/iop/spots.c +++ b/src/iop/spots.c @@ -372,7 +372,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // we don't modify most of the image: #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(out, in) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, out, roi_in, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/lowpass.c b/src/iop/lowpass.c index 6076de7..c97f126 100644 --- a/src/iop/lowpass.c +++ b/src/iop/lowpass.c @@ -420,7 +420,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float *const Labminf = (float *)&Labmin; const float *const Labmaxf = (float *)&Labmax; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, Labmaxf, Labminf, roi_out) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -529,7 +532,10 @@ void commit_params(struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pix const float contrastm1sq = boost * (fabs(d->contrast) - 1.0f) * (fabs(d->contrast) - 1.0f); const float contrastscale = copysign(sqrt(1.0f + contrastm1sq), d->contrast); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(contrastm1sq, contrastscale) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < 0x10000; k++) { @@ -551,7 +557,10 @@ void commit_params(struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pix const float gamma = (d->brightness >= 0.0f) ? 1.0f / (1.0f + d->brightness) : (1.0f - d->brightness); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(gamma) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < 0x10000; k++) { diff --git a/src/iop/temperature.c b/src/iop/temperature.c index 2bdf8d8..927980e 100644 --- a/src/iop/temperature.c +++ b/src/iop/temperature.c @@ -454,7 +454,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const if(filters == 9u) { // xtrans float mosaiced #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(d, in, out, roi_out, xtrans) \ + schedule(static) \ + collapse(2) #endif for(int j = 0; j < roi_out->height; j++) { @@ -468,7 +471,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const else if(filters) { // bayer float mosaiced #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(d, filters, in, out, roi_out) \ + schedule(static) \ + collapse(2) #endif for(int j = 0; j < roi_out->height; j++) { @@ -484,7 +490,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, d, in, out, roi_out) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -516,7 +525,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c if(filters == 9u) { // xtrans float mosaiced #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ivoid, ovoid, roi_out, xtrans) \ + shared(d) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -557,7 +569,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c else if(filters) { // bayer float mosaiced #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(filters, ivoid, ovoid, roi_out) \ + shared(d) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -599,7 +614,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 coeffs = _mm_set_ps(1.0f, d->coeffs[2], d->coeffs[1], d->coeffs[0]); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, coeffs, ivoid, ovoid, roi_out) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/scalepixels.c b/src/iop/scalepixels.c index 9db887c..037fb21 100644 --- a/src/iop/scalepixels.c +++ b/src/iop/scalepixels.c @@ -185,7 +185,10 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c const dt_iop_scalepixels_data_t * const d = piece->data; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(interpolation) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch_width, d, ivoid, ovoid, roi_in, roi_out) \ + shared(interpolation) \ + schedule(static) #endif // (slow) point-by-point transformation. // TODO: optimize with scanlines and linear steps between? diff --git a/src/iop/velvia.c b/src/iop/velvia.c index 99c97e9..1701d8a 100644 --- a/src/iop/velvia.c +++ b/src/iop/velvia.c @@ -140,7 +140,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const else { #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, data, ivoid, ovoid, roi_out, strength) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -186,7 +188,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c else { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(in, out, data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, roi_out, strength) \ + shared(in, out, data) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { diff --git a/src/iop/vignette.c b/src/iop/vignette.c index cdf529f..4df43e5 100644 --- a/src/iop/vignette.c +++ b/src/iop/vignette.c @@ -740,7 +740,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const unsigned int *const tea_states = calloc(2 * dt_get_num_threads(), sizeof(unsigned int)); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data, yscale, xscale, dither) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, dscale, exp1, exp2, fscale, ivoid, ovoid, \ + roi_center_scaled, roi_out, tea_states, unbound) \ + shared(data, yscale, xscale, dither) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/levels.c b/src/iop/levels.c index 8bef474..f183d15 100644 --- a/src/iop/levels.c +++ b/src/iop/levels.c @@ -301,7 +301,9 @@ void process(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *c } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/liquify.c b/src/iop/liquify.c index 70193c5..d91248b 100644 --- a/src/iop/liquify.c +++ b/src/iop/liquify.c @@ -1337,7 +1337,9 @@ void process(struct dt_iop_module_t *module, dt_dev_pixelpipe_iop_t *piece, cons assert (ch == 4); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, out, roi_in, roi_out) \ + schedule(static) #endif for (int i = 0; i < roi_out->height; i++) { diff --git a/src/iop/tonecurve.c b/src/iop/tonecurve.c index b8a6e93..ba051a3 100644 --- a/src/iop/tonecurve.c +++ b/src/iop/tonecurve.c @@ -307,7 +307,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int unbound_ab = d->unbound_ab; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(d) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(autoscale_ab, ch, height, i, low_approximation, o, \ + xm_al, xm_ar, xm_bl, xm_br, xm_L, unbound_ab, width) \ + shared(d) \ + schedule(static) #endif for(int k = 0; k < height; k++) { diff --git a/src/iop/splittoning.c b/src/iop/splittoning.c index 7259a30..9cb8375 100644 --- a/src/iop/splittoning.c +++ b/src/iop/splittoning.c @@ -165,7 +165,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float compress = (data->compress / 110.0) / 2.0; // Don't allow 100% compression.. #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data) private(in, out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, compress, ivoid, ovoid, roi_out) \ + shared(data) \ + private(in, out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/relight.c b/src/iop/relight.c index ed20436..6269386 100644 --- a/src/iop/relight.c +++ b/src/iop/relight.c @@ -128,7 +128,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float c = (data->width / 10.0) / 2.0; // Width #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(b, c, ch, ivoid, ovoid, roi_out) \ + shared(data) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/grain.c b/src/iop/grain.c index 0cabbad..d610348 100644 --- a/src/iop/grain.c +++ b/src/iop/grain.c @@ -477,7 +477,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const float fib1div2 = fib1 / fib2; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data, hash) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, filter, filtermul, ivoid, ovoid, roi_out, strength, \ + wd, zoom) \ + shared(data, hash) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/dither.c b/src/iop/dither.c index 00fecb3..ebb5b0e 100644 --- a/src/iop/dither.c +++ b/src/iop/dither.c @@ -308,7 +308,9 @@ static void process_floyd_steinberg(struct dt_iop_module_t *self, dt_dev_pixelpi } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, ivoid, ovoid, width) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -474,7 +476,9 @@ static void process_floyd_steinberg_sse2(struct dt_iop_module_t *self, dt_dev_pi } #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, ivoid, ovoid, width) \ + schedule(static) #endif for(int j = 0; j < height; j++) { @@ -604,7 +608,9 @@ static void process_random(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t unsigned int *const tea_states = calloc(2 * dt_get_num_threads(), sizeof(unsigned int)); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, dither, height, ivoid, ovoid, tea_states, width) \ + schedule(static) #endif for(int j = 0; j < height; j++) { diff --git a/src/iop/clahe.c b/src/iop/clahe.c index 1978862..72b0a9f 100644 --- a/src/iop/clahe.c +++ b/src/iop/clahe.c @@ -86,7 +86,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *luminance = (float *)malloc(((size_t)roi_out->width * roi_out->height) * sizeof(float)); // double lsmax=0.0,lsmin=1.0; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(luminance) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ivoid, roi_out) \ + shared(luminance) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -115,7 +118,11 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // CLAHE #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(luminance) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, dest_buf, destbuf_size, ivoid, ovoid, rad, roi_in, \ + roi_out, slope) \ + shared(luminance) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/exposure.c b/src/iop/exposure.c index 79a9131..5b24b21 100644 --- a/src/iop/exposure.c +++ b/src/iop/exposure.c @@ -400,7 +400,9 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, d, i, o, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k++) { @@ -425,7 +427,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const __m128 scalev = _mm_set1_ps(d->scale); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(blackv, ch, i, o, roi_out, scalev) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/colortransfer.c b/src/iop/colortransfer.c index a3ed68c..8c41b42 100644 --- a/src/iop/colortransfer.c +++ b/src/iop/colortransfer.c @@ -260,7 +260,10 @@ static void kmeans(const float *col, const dt_iop_roi_t *const roi, const int n, for(int k = 0; k < n; k++) cnt[k] = 0; // randomly sample col positions inside roi #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(col, mean_out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cnt, mean, n, roi, samples, var) \ + shared(col, mean_out) \ + schedule(static) #endif for(int s = 0; s < samples; s++) { @@ -358,7 +361,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const int hist[HISTN]; capture_histogram(in, roi_in, hist); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(data, in, out, hist) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, roi_out) \ + shared(data, in, out, hist) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { @@ -385,7 +391,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // for all pixels: find input cluster, transfer to mapped target cluster #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(data, in, out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, mapio, mean, roi_out, var) \ + shared(data, in, out) \ + schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { diff --git a/src/iop/colormapping.c b/src/iop/colormapping.c index 7de02c9..1495087 100644 --- a/src/iop/colormapping.c +++ b/src/iop/colormapping.c @@ -333,7 +333,10 @@ static void kmeans(const float *col, const int width, const int height, const in for(int k = 0; k < n; k++) cnt[k] = 0; // randomly sample col positions inside roi #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(col, mean_out) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(cnt, height, mean, n, samples, var, width) \ + shared(col, mean_out) \ + schedule(static) #endif for(int s = 0; s < samples; s++) { @@ -493,7 +496,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // first get delta L of equalized L minus original image L, scaled to fit into [0 .. 100] #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(data, in, out, equalization) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, width) \ + shared(data, in, out, equalization) \ + schedule(static) #endif for(int k = 0; k < height; k++) { @@ -528,7 +534,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const weight_buf = malloc(data->n * dt_get_num_threads() * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) shared(data, in, out, equalization) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, mapio, var_ratio, weight_buf, width) \ + shared(data, in, out, equalization) \ + schedule(static) #endif for(int k = 0; k < height; k++) { diff --git a/src/iop/highlights.c b/src/iop/highlights.c index 1a5e192..bced8ba 100644 --- a/src/iop/highlights.c +++ b/src/iop/highlights.c @@ -526,7 +526,9 @@ static void process_lch_bayer(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *pie const uint32_t filters = piece->pipe->dsc.filters; #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clip, filters, ivoid, ovoid, roi_out) \ + schedule(dynamic) #endif for(int j = 0; j < roi_out->height; j++) { @@ -625,7 +627,9 @@ static void process_lch_xtrans(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *pi const uint8_t(*const xtrans)[6] = (const uint8_t(*const)[6])piece->pipe->dsc.xtrans; #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clip, ivoid, ovoid, roi_in, roi_out, xtrans) \ + schedule(dynamic) #endif for(int j = 0; j < roi_out->height; j++) { @@ -760,7 +764,9 @@ static void process_clip_plain(dt_dev_pixelpipe_iop_t *piece, const void *const if(piece->pipe->dsc.filters) { // raw mosaic #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(clip, in, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)roi_out->width * roi_out->height; k++) { @@ -772,7 +778,9 @@ static void process_clip_plain(dt_dev_pixelpipe_iop_t *piece, const void *const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, clip, in, out, roi_out) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k++) { @@ -793,7 +801,9 @@ static void process_clip_sse2(dt_dev_pixelpipe_iop_t *piece, const void *const i float *const out = (float *)ovoid; float *const in = (float *)ivoid; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clipm, in, n, out) \ + schedule(static) #endif for(size_t j = 0; j < (n & ~3u); j += 4) _mm_stream_ps(out + j, _mm_min_ps(clipm, _mm_load_ps(in + j))); _mm_sfence(); @@ -807,7 +817,9 @@ static void process_clip_sse2(dt_dev_pixelpipe_iop_t *piece, const void *const i const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, clipm, ivoid, ovoid, roi_in, roi_out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -868,7 +880,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const { const uint8_t(*const xtrans)[6] = (const uint8_t(*const)[6])piece->pipe->dsc.xtrans; #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clips, filters, ivoid, ovoid, roi_in, roi_out, \ + xtrans) \ + schedule(dynamic) #endif for(int j = 0; j < roi_out->height; j++) { @@ -876,7 +891,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 0, -1, j, clips, xtrans, 1); } #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clips, filters, ivoid, ovoid, roi_in, roi_out, \ + xtrans) \ + schedule(dynamic) #endif for(int i = 0; i < roi_out->width; i++) { @@ -887,7 +905,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const else { #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) shared(data, piece) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clips, filters, ivoid, ovoid, roi_out) \ + shared(data, piece) \ + schedule(dynamic) #endif for(int j = 0; j < roi_out->height; j++) { @@ -897,7 +918,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const // up/down directions #ifdef _OPENMP -#pragma omp parallel for schedule(dynamic) default(none) shared(data, piece) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(clips, filters, ivoid, ovoid, roi_out) \ + shared(data, piece) \ + schedule(dynamic) #endif for(int i = 0; i < roi_out->width; i++) { diff --git a/src/iop/channelmixer.c b/src/iop/channelmixer.c index 5f3a03f..1bb3f19 100644 --- a/src/iop/channelmixer.c +++ b/src/iop/channelmixer.c @@ -155,7 +155,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const const int ch = piece->colors; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(data) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, gray_mix_mode, ivoid, ovoid, roi_out) \ + shared(data) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/retouch.c b/src/iop/retouch.c index 68a36c7..7c966b1 100644 --- a/src/iop/retouch.c +++ b/src/iop/retouch.c @@ -3525,7 +3525,10 @@ static void image_rgb2lab(float *img_src, const int width, const int height, con if(ch == 4 && use_sse) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_src) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, stride) \ + shared(img_src) \ + schedule(static) #endif for(int i = 0; i < stride; i += ch) { @@ -3541,7 +3544,10 @@ static void image_rgb2lab(float *img_src, const int width, const int height, con #endif #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_src) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, stride) \ + shared(img_src) \ + schedule(static) #endif for(int i = 0; i < stride; i += ch) { @@ -3560,7 +3566,10 @@ static void image_lab2rgb(float *img_src, const int width, const int height, con if(ch == 4 && use_sse) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_src) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, stride) \ + shared(img_src) \ + schedule(static) #endif for(int i = 0; i < stride; i += ch) { @@ -3576,7 +3585,10 @@ static void image_lab2rgb(float *img_src, const int width, const int height, con #endif #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_src) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, stride) \ + shared(img_src) \ + schedule(static) #endif for(int i = 0; i < stride; i += ch) { @@ -3597,8 +3609,12 @@ static void rt_process_stats(const float *const img_src, const int width, const int count = 0; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) reduction(+ : count, l_sum) reduction(max : l_max) \ - reduction(min : l_min) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, img_src, size) \ + schedule(static) \ + reduction(+ : count, l_sum) \ + reduction(max : l_max) \ + reduction(min : l_min) #endif for(int i = 0; i < size; i += ch) { @@ -3636,7 +3652,10 @@ static void rt_adjust_levels(float *img_src, const int width, const int height, const float in_inv_gamma = pow(10, tmp); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_src) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in_inv_gamma, left, right, size) \ + shared(img_src) \ + schedule(static) #endif for(int i = 0; i < size; i += ch) { @@ -3701,7 +3720,10 @@ static void rt_copy_in_to_out(const float *const in, const struct dt_iop_roi_t * const int y_to = MIN(roi_out->height, roi_in->height); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, out, roi_in, roi_out, rowsize, xoffs, yoffs, \ + y_to) \ + schedule(static) #endif for(int y = 0; y < y_to; y++) { @@ -3744,7 +3766,10 @@ static void rt_build_scaled_mask(float *const mask, dt_iop_roi_t *const roi_mask } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(mask_tmp, roi_mask_scaled) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(mask, roi_in, roi_mask, x_to, y_to) \ + shared(mask_tmp, roi_mask_scaled) \ + schedule(static) #endif for(int yy = roi_mask_scaled->y; yy < y_to; yy++) { @@ -3778,7 +3803,10 @@ static void rt_copy_image_masked(float *const img_src, float *img_dest, dt_iop_r if(ch == 4 && use_sse) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_dest) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, img_src, mask_scaled, opacity, roi_dest, roi_mask_scaled) \ + shared(img_dest) \ + schedule(static) #endif for(int yy = 0; yy < roi_mask_scaled->height; yy++) { @@ -3809,7 +3837,10 @@ static void rt_copy_image_masked(float *const img_src, float *img_dest, dt_iop_r const int ch1 = (ch == 4) ? ch - 1 : ch; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(img_dest) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ch1, img_src, mask_scaled, opacity, roi_dest, roi_mask_scaled) \ + shared(img_dest) \ + schedule(static) #endif for(int yy = 0; yy < roi_mask_scaled->height; yy++) { @@ -3841,7 +3872,9 @@ static void rt_copy_mask_to_alpha(float *const img, dt_iop_roi_t *const roi_img, const float opacity) { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, img, mask_scaled, opacity, roi_img, roi_mask_scaled) \ + schedule(static) #endif for(int yy = 0; yy < roi_mask_scaled->height; yy++) { @@ -3871,7 +3904,9 @@ static void retouch_fill_sse(float *const in, dt_iop_roi_t *const roi_in, float const __m128 val_fill = _mm_load_ps(valf4_fill); #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, in, mask_scaled, opacity, roi_in, roi_mask_scaled, val_fill) \ + schedule(static) #endif for(int yy = 0; yy < roi_mask_scaled->height; yy++) { @@ -3909,7 +3944,9 @@ static void retouch_fill(float *const in, dt_iop_roi_t *const roi_in, const int const int ch1 = (ch == 4) ? ch - 1 : ch; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, ch1, fill_color, in, mask_scaled, opacity, roi_in, roi_mask_scaled) \ + schedule(static) #endif for(int yy = 0; yy < roi_mask_scaled->height; yy++) { diff --git a/src/iop/graduatednd.c b/src/iop/graduatednd.c index e48b4ab..6473b4a 100644 --- a/src/iop/graduatednd.c +++ b/src/iop/graduatednd.c @@ -705,7 +705,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const if(data->density > 0) { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, cosv, data, filter_compression, hh_inv, hw_inv, \ + ivoid, ix, iy, offset, ovoid, roi_out, sinv) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -751,7 +754,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const else { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, cosv, data, filter_compression, hh_inv, hw_inv, \ + ivoid, ix, iy, offset, ovoid, roi_out, sinv) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -830,7 +836,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c if(data->density > 0) { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, cosv, data, filter_compression, hh_inv, hw_inv, \ + ivoid, ix, iy, offset, ovoid, roi_out, sinv) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { @@ -878,7 +887,10 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c else { #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, cosv, data, filter_compression, hh_inv, hw_inv, \ + ivoid, ix, iy, offset, ovoid, roi_out, sinv) \ + schedule(static) #endif for(int y = 0; y < roi_out->height; y++) { diff --git a/src/iop/zonesystem.c b/src/iop/zonesystem.c index b8a4437..a4be9ed 100644 --- a/src/iop/zonesystem.c +++ b/src/iop/zonesystem.c @@ -224,7 +224,10 @@ static void process_common_cleanup(struct dt_iop_module_t *self, dt_dev_pixelpip if(gauss && tmp) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmp) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, width, ivoid) \ + shared(tmp) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) tmp[k] = ((float *)ivoid)[ch * k]; @@ -233,7 +236,10 @@ static void process_common_cleanup(struct dt_iop_module_t *self, dt_dev_pixelpip /* create zonemap preview for input */ dt_pthread_mutex_lock(&g->lock); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmp, g) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, size, width) \ + shared(tmp, g) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) { @@ -243,7 +249,10 @@ static void process_common_cleanup(struct dt_iop_module_t *self, dt_dev_pixelpip #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmp) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, height, ovoid, width) \ + shared(tmp) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) tmp[k] = ((float *)ovoid)[ch * k]; @@ -253,7 +262,10 @@ static void process_common_cleanup(struct dt_iop_module_t *self, dt_dev_pixelpip /* create zonemap preview for output */ dt_pthread_mutex_lock(&g->lock); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(tmp, g) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, size, width) \ + shared(tmp, g) \ + schedule(static) #endif for(size_t k = 0; k < (size_t)width * height; k++) { @@ -281,7 +293,10 @@ void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const float *const out = (float *const)ovoid; #ifdef _OPENMP -#pragma omp parallel for SIMD() default(none) schedule(static) collapse(2) +#pragma omp parallel for SIMD() default(none) \ + dt_omp_firstprivate(ch, d, in, out, roi_out, size) \ + schedule(static) \ + collapse(2) #endif for(size_t k = 0; k < (size_t)ch * roi_out->width * roi_out->height; k += ch) { @@ -311,7 +326,9 @@ void process_sse2(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, c const int size = d->params.size; #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(ch, d, ivoid, ovoid, roi_out, size) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { diff --git a/src/iop/demosaic.c b/src/iop/demosaic.c index f4a943d..57f0141 100644 --- a/src/iop/demosaic.c +++ b/src/iop/demosaic.c @@ -310,7 +310,10 @@ static void pre_median_b(float *out, const float *const in, const dt_iop_roi_t * for(int pass = 0; pass < num_passes; pass++) { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(filters, in, lim, roi, threshold) \ + shared(out) \ + schedule(static) #endif for(int row = 3; row < roi->height - 3; row++) { @@ -370,7 +373,10 @@ static void color_smoothing(float *out, const dt_iop_roi_t *const roi_out, const for(int i = 0; i < roi_out->width; i++, outp += 4) outp[3] = outp[c]; } #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(out, c) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(roi_out, width4) \ + shared(out, c) \ + schedule(static) #endif for(int j = 1; j < roi_out->height - 1; j++) { @@ -425,7 +431,10 @@ static void green_equilibration_lavg(float *out, const float *const in, const in memcpy(out, in, height * width * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(out, oi, oj) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, in, thr, width) \ + shared(out, oi, oj) \ + schedule(static) #endif for(size_t j = oj; j < height - 2; j += 2) { @@ -471,7 +480,11 @@ static void green_equilibration_favg(float *out, const float *const in, const in const int g2_offset = oi ? -1 : 1; memcpy(out, in, (size_t)height * width * sizeof(float)); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) reduction(+ : sum1, sum2) shared(oi, oj) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(g2_offset, height, in, width) \ + reduction(+ : sum1, sum2) \ + shared(oi, oj) \ + schedule(static) #endif for(size_t j = oj; j < (height - 1); j += 2) { @@ -488,7 +501,10 @@ static void green_equilibration_favg(float *out, const float *const in, const in return; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(out, oi, oj, gr_ratio) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(g2_offset, height, in, width) \ + shared(out, oi, oj, gr_ratio) \ + schedule(static) #endif for(int j = oj; j < (height - 1); j += 2) { @@ -584,7 +600,10 @@ static void xtrans_markesteijn_interpolate(float *out, const float *const in, // extra passes propagates out errors at edges, hence need more padding const int pad_tile = (passes == 1) ? 12 : 17; #ifdef _OPENMP -#pragma omp parallel for default(none) shared(sgrow, sgcol, allhex, out) schedule(dynamic) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(all_buffers, buffer_size, dir, height, in, ndir, pad_tile, passes, roi_in, width, xtrans) \ + shared(sgrow, sgcol, allhex, out) \ + schedule(dynamic) #endif // step through TSxTS cells of image, each tile overlapping the // prior as interpolation needs a substantial border @@ -1597,7 +1616,10 @@ static void xtrans_fdc_interpolate(float *out, const float *const in, const dt_i } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(sgrow, sgcol, allhex, out, rowoffset, coloffset) schedule(dynamic) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(all_buffers, buffer_size, dir, height, in, ndir, pad_tile, roi_in, width, xtrans, directionality, harr, Minv, modarr) \ + shared(sgrow, sgcol, allhex, out, rowoffset, coloffset) \ + schedule(dynamic) #endif // step through TSxTS cells of image, each tile overlapping the // prior as interpolation needs a substantial border @@ -2089,7 +2111,10 @@ static void lin_interpolate(float *out, const float *const in, const dt_iop_roi_ // border interpolate #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(colors, filters, in, roi_in, roi_out, xtrans) \ + shared(out) \ + schedule(static) #endif for(int row = 0; row < roi_out->height; row++) for(int col = 0; col < roi_out->width; col++) @@ -2163,7 +2188,10 @@ static void lin_interpolate(float *out, const float *const in, const dt_iop_roi_ } #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(colors, in, lookup, roi_in, roi_out, size) \ + shared(out) \ + schedule(static) #endif for(int row = 1; row < roi_out->height - 1; row++) { @@ -2297,7 +2325,11 @@ static void vng_interpolate(float *out, const float *const in, for(int row = 2; row < height - 2; row++) /* Do VNG interpolation */ { #ifdef _OPENMP -#pragma omp parallel for default(none) shared(row, code, brow, out, filters4) private(ip) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(colors, pcol, prow, roi_in, width, xtrans) \ + shared(row, code, brow, out, filters4) \ + private(ip) \ + schedule(static) #endif for(int col = 2; col < width - 2; col++) { @@ -2362,7 +2394,10 @@ static void vng_interpolate(float *out, const float *const in, if(filters != 9 && !FILTERS_ARE_4BAYER(filters)) // x-trans or CYGM/RGBE // for Bayer mix the two greens to make VNG4 #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(out) \ + schedule(static) #endif for(int i = 0; i < height * width; i++) out[i * 4 + 1] = (out[i * 4 + 1] + out[i * 4 + 3]) / 2.0f; } @@ -2376,7 +2411,10 @@ static void passthrough_monochrome(float *out, const float *const in, dt_iop_roi assert(roi_in->height >= roi_out->height); #ifdef _OPENMP -#pragma omp parallel for default(none) shared(out) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(in, roi_out, roi_in) \ + shared(out) \ + schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { @@ -2445,7 +2483,10 @@ static void demosaic_ppg(float *const out, const float *const in, const dt_iop_r } // for all pixels: interpolate green into float array, or copy color. #ifdef _OPENMP -#pragma omp parallel for default(none) shared(input) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(filters, out, roi_in, roi_out) \ + shared(input) \ + schedule(static) #endif for(int j = offy; j < roi_out->height - offY; j++) { @@ -2525,7 +2566,9 @@ static void demosaic_ppg(float *const out, const float *const in, const dt_iop_r // for all pixels: interpolate colors into float array #ifdef _OPENMP -#pragma omp parallel for default(none) schedule(static) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(filters, out, roi_out) \ + schedule(static) #endif for(int j = 1; j < roi_out->height - 1; j++) { diff --git a/src/chart/main.c b/src/chart/main.c index 2df0c83..450758c 100644 --- a/src/chart/main.c +++ b/src/chart/main.c @@ -1424,10 +1424,13 @@ static void get_xyz_sample_from_image(const image_t *const image, float shrink, double sample_x = 0.0, sample_y = 0.0, sample_z = 0.0; size_t n_samples = 0; #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) \ - shared(corners, x_start, y_start, x_end, y_end, delta_x_top, delta_y_top, delta_x_bottom, delta_y_bottom, \ - delta_x_left, delta_y_left, delta_x_right, \ - delta_y_right) reduction(+ : n_samples, sample_x, sample_y, sample_z) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(image) \ + shared(corners, x_start, y_start, x_end, y_end, delta_x_top, delta_y_top, \ + delta_x_bottom, delta_y_bottom, delta_x_left, delta_y_left, \ + delta_x_right, delta_y_right) \ + reduction(+ : n_samples, sample_x, sample_y, sample_z) \ + schedule(static) #endif for(int y = y_start; y < y_end; y++) for(int x = x_start; x < x_end; x++) @@ -1542,7 +1545,10 @@ static void free_image(image_t *image) static void image_lab_to_xyz(float *image, const int width, const int height) { #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(image) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, width) \ + shared(image) \ + schedule(static) #endif for(int y = 0; y < height; y++) for(int x = 0; x < width; x++) diff --git a/src/chart/cairo.c b/src/chart/cairo.c index acc2717..e5b4f12 100644 --- a/src/chart/cairo.c +++ b/src/chart/cairo.c @@ -187,7 +187,10 @@ cairo_surface_t *cairo_surface_create_from_xyz_data(const float *const image, co unsigned char *rgbbuf = (unsigned char *)malloc(sizeof(unsigned char) * height * width * 4); #ifdef _OPENMP -#pragma omp parallel for schedule(static) default(none) shared(rgbbuf) +#pragma omp parallel for default(none) \ + dt_omp_firstprivate(height, image, width) \ + shared(rgbbuf) \ + schedule(static) #endif for(int y = 0; y < height; y++) { -- 2.22.0