Adsk contrib - Add support for neon intrinsic integration

.github/workflows/ci_workflow.yml

@@ -243,7 +243,7 @@ jobs:
                 -DOCIO_BUILD_DOCS=${{ matrix.build-docs }} \
                 -DOCIO_BUILD_OPENFX=${{ matrix.build-openfx }} \
                 -DOCIO_BUILD_GPU_TESTS=OFF \
-                -DOCIO_USE_SSE=${{ matrix.use-sse }} \
+                -DOCIO_USE_SIMD=${{ matrix.use-sse }} \
                 -DOCIO_USE_OIIO_FOR_APPS=${{ matrix.use-oiio }} \
                 -DOCIO_INSTALL_EXT_PACKAGES=ALL \
                 -DOCIO_WARNING_AS_ERROR=ON \
@@ -384,7 +384,7 @@ jobs:
                 -DOCIO_BUILD_DOCS=${{ matrix.build-docs }} \
                 -DOCIO_BUILD_OPENFX=${{ matrix.build-openfx }} \
                 -DOCIO_BUILD_GPU_TESTS=OFF \
-                -DOCIO_USE_SSE=${{ matrix.use-sse }} \
+                -DOCIO_USE_SIMD=${{ matrix.use-sse }} \
                 -DOCIO_USE_OIIO_FOR_APPS=${{ matrix.use-oiio }} \
                 -DOCIO_INSTALL_EXT_PACKAGES=ALL \
                 -DOCIO_WARNING_AS_ERROR=ON \
@@ -532,7 +532,7 @@ jobs:
                 -DOCIO_BUILD_DOCS=${{ matrix.build-docs }} \
                 -DOCIO_BUILD_OPENFX=${{ matrix.build-openfx }} \
                 -DOCIO_BUILD_GPU_TESTS=OFF \
-                -DOCIO_USE_SSE=${{ matrix.use-sse }} \
+                -DOCIO_USE_SIMD=${{ matrix.use-sse }} \
                 -DOCIO_USE_OIIO_FOR_APPS=${{ matrix.use-oiio }} \
                 -DOCIO_INSTALL_EXT_PACKAGES=ALL \
                 -DOCIO_WARNING_AS_ERROR=ON \

CMakeLists.txt

@@ -25,6 +25,14 @@ if(APPLE AND NOT DEFINED CMAKE_OSX_DEPLOYMENT_TARGET)
 endif()
 
 
+###############################################################################
+# By default, build the library, tests, tools, and Python binding as universal binaries for macOS.
+
+if(APPLE AND (NOT DEFINED CMAKE_OSX_ARCHITECTURES OR CMAKE_OSX_ARCHITECTURES STREQUAL ""))
+    set(CMAKE_OSX_ARCHITECTURES "arm64;x86_64" CACHE STRING "Default OS X architectures" FORCE)
+endif()
+
+
 ###############################################################################
 # Project definition.
 
@@ -173,21 +181,49 @@ endif()
 
 ###############################################################################
 # Optimization / internal linking preferences
-
-option(OCIO_USE_SSE "Specify whether to enable SSE CPU performance optimizations" ON)
+option(OCIO_USE_SIMD "Specify whether to enable SIMD CPU performance optimizations" ON)
 option(OCIO_USE_OIIO_FOR_APPS "Request OIIO to build apps (ociolutimage, ocioconvert and ociodisplay), the default uses OpenEXR." OFF)
 
-if ("${CMAKE_SYSTEM_PROCESSOR}" MATCHES "(AMD64|IA64|EM64T|X86|x86_64|i386|i686)")
-    option(OCIO_USE_SSE2 "Specify whether to enable SSE2 CPU performance optimizations" ON)
-    option(OCIO_USE_SSE3 "Specify whether to enable SSE3 CPU performance optimizations" ON)
-    option(OCIO_USE_SSSE3 "Specify whether to enable SSSE3 CPU performance optimizations" ON)
-    option(OCIO_USE_SSE4 "Specify whether to enable SSE4 CPU performance optimizations" ON)
-    option(OCIO_USE_SSE42 "Specify whether to enable SSE4.2 CPU performance optimizations" ON)
-    option(OCIO_USE_AVX "Specify whether to enable AVX CPU performance optimizations" ON)
-    option(OCIO_USE_AVX2 "Specify whether to enable AVX2 CPU performance optimizations" ON)
-    option(OCIO_USE_AVX512 "Specify whether to enable AVX512 CPU performance optimizations" ON)
-    option(OCIO_USE_F16C  "Specify whether to enable F16C CPU performance optimizations" ON)
+if ("${CMAKE_SYSTEM_PROCESSOR}" MATCHES "(AMD64|IA64|EM64T|X86|x86_64|i386|i686)" OR
+    (APPLE AND "${CMAKE_SYSTEM_PROCESSOR}" MATCHES "arm64" AND "${CMAKE_OSX_ARCHITECTURES}" MATCHES "(arm64;x86_64|x86_64;arm64|x86_64)"))
+
+    # Enable OCIO_ARCH_X86 for any intel-based architecture or Apple Rosetta (x86_64 or universal build).
     set(OCIO_ARCH_X86 1)
+
+    # For a Mac OS universal build, OCIO translates the SSE instructions into ARM Neon with 
+    # the library SSE2NEON for the arm64 side of the binary.
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE ON)
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_AVX ON)
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_F16C ON)
+elseif(APPLE AND "${CMAKE_SYSTEM_PROCESSOR}" MATCHES "arm64" AND "${CMAKE_OSX_ARCHITECTURES}" MATCHES "arm64")
+
+    # Apple ARM only build.
+
+    # OCIO translates the SSE instructions into ARM Neon with the library SSE2NEON.
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE ON)
+    # OCIO does not translate AVX to ARM Neon.
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_AVX OFF)
+    # Turn F16C off since they are x86_64 instructions.
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_F16C OFF)
+else()
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE OFF)
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_AVX OFF)
+    set(OCIO_BUILD_ENABLE_OPTIMIZATIONS_F16C OFF)
+    set(OCIO_ARCH_X86 0)
+endif()
+
+option(OCIO_USE_SSE2 "Specify whether to enable SSE2 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE})
+option(OCIO_USE_SSE3 "Specify whether to enable SSE3 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE})
+option(OCIO_USE_SSSE3 "Specify whether to enable SSSE3 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE})
+option(OCIO_USE_SSE4 "Specify whether to enable SSE4 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE})
+option(OCIO_USE_SSE42 "Specify whether to enable SSE4.2 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_SSE})
+option(OCIO_USE_AVX "Specify whether to enable AVX CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_AVX})
+option(OCIO_USE_AVX2 "Specify whether to enable AVX2 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_AVX})
+option(OCIO_USE_AVX512 "Specify whether to enable AVX512 CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_AVX})
+option(OCIO_USE_F16C  "Specify whether to enable F16C CPU performance optimizations" ${OCIO_BUILD_ENABLE_OPTIMIZATIONS_F16C})
+
+if (APPLE)
+   option(OCIO_USE_SSE2NEON "Specify whether to enable SSE CPU performance optimizations using SSE2NEON for Apple ARM architecture" ON)
 endif()
 
 ###############################################################################
@@ -197,6 +233,60 @@ message(STATUS "Checking for GPU configuration...")
 include(CheckSupportGL)
 
 
+###############################################################################
+# Check for ARM neon here because we need to know if ARM NEON is supported
+# quickly. Once we know that ARM NEON is supported, we can add sse2neon library
+# to the build.
+
+if(OCIO_USE_SIMD)
+    include(CheckSupportARMNeon)
+
+    if (NOT COMPILER_SUPPORTS_ARM_NEON)
+        # Force it to OFF since ARM Neon is not supported.
+        set(OCIO_USE_SSE2NEON OFF)
+    endif()
+endif()
+
+
+###############################################################################
+# Add sse2neon to the build if ARM NEON intrinsics are supported.
+
+if(OCIO_USE_SIMD AND OCIO_USE_SSE2NEON AND COMPILER_SUPPORTS_ARM_NEON)
+    # Install sse2neon. Please note that sse2neon is downloaded during the configure step as it is
+    # needed for CompilerFlags.cmake and CheckSupportSSE2.cmake.
+
+    # Sse2neon is not treated like an imported target. The logic to find sse2neon is here because 
+    # a find module is not suitable for sse2neon's use case.
+    find_path(sse2neon_INCLUDE_DIR
+        NAMES
+            sse2neon.h
+        HINTS
+            ${sse2neon_ROOT}
+        PATH_SUFFIXES
+            sse2neon
+            include
+            sse2neon/include
+    )
+
+    # As per instructions on sse2neon's GitHub page, the following compiler flags should be used:
+    # "-march=armv8-a+fp+simd+crypto+crc". These flags are required for some ARM platforms that do
+    # not enable floating point calculations or SIMD instructions by default. However, for ARM64
+    # (Apple ARM platform) and x86_64 platforms, these features are already enabled by default.
+    # Therefore, no additional compiler flags are needed.
+    if (NOT sse2neon_INCLUDE_DIR)
+        include(Installsse2neon)
+    else()
+        # Any changes to the following lines must be replicated in Installsse2neon.cmake as well.
+        # Create a target for sse2neon (non-imported)
+        add_library(sse2neon INTERFACE)
+        # Add the include directories to the target.
+        target_include_directories(sse2neon INTERFACE "${sse2neon_INCLUDE_DIR}")
+        # Ignore the warnings coming from sse2neon.h as they are false positives.
+        target_compile_options(sse2neon INTERFACE -Wno-unused-parameter)
+    endif()
+endif()
+
+
 ###############################################################################
 # Define compilation and link flags
 

docs/quick_start/installation.rst

@@ -264,6 +264,7 @@ CMake Options
 +++++++++++++
 
 There are many options available in `CMake. 
+
 <https://cmake.org/cmake/help/latest/guide/user-interaction/index.html#guide:User%20Interaction%20Guide>`_
 
 Several of the most common ones are:
@@ -277,14 +278,37 @@ Here are the most common OCIO-specific CMake options (the default values are sho
 - ``-DOCIO_USE_OIIO_FOR_APPS=OFF`` (Set ON to build tools with OpenImageIO rather than OpenEXR)
 - ``-DOCIO_BUILD_PYTHON=ON`` (Set to OFF to not build the Python binding)
 - ``-DOCIO_BUILD_OPENFX=OFF`` (Set to ON to build the OpenFX plug-ins)
-- ``-DOCIO_USE_SSE=ON`` (Set to OFF to turn off SSE CPU performance optimizations)
+- ``-DOCIO_USE_SIMD=ON`` (Set to OFF to turn off SIMD CPU performance optimizations, such as SSE and NEON)
+- ``-DOCIO_USE_SSE2=ON`` (Set to OFF to turn off SSE2 CPU performance optimizations)
+- ``-DOCIO_USE_SSE3=ON`` (Set to OFF to turn off SSE3 CPU performance optimizations)
+- ``-DOCIO_USE_SSSE2=ON`` (Set to OFF to turn off SSSE3 CPU performance optimizations)
+- ``-DOCIO_USE_SSE4=ON`` (Set to OFF to turn off SSE4 CPU performance optimizations)
+- ``-DOCIO_USE_SSE42=ON`` (Set to OFF to turn off SSE4.2 CPU performance optimizations)
+- ``-DOCIO_USE_AVX=ON`` (Set to OFF to turn off AVX CPU performance optimizations)
+- ``-DOCIO_USE_AVX2=ON`` (Set to OFF to turn off AVX2 CPU performance optimizations)
+- ``-DOCIO_USE_AVX512=ON`` (Set to OFF to turn off AVX512 CPU performance optimizations)
+- ``-DOCIO_USE_F16C=ON`` (Set to OFF to turn off F16C CPU performance optimizations)
+- ``-OCIO_USE_SSE2NEON=ON`` (Apple Only; Set to OFF to turn off the SSE2NEON translation performance optimizations on Apple ARM)
 - ``-DOCIO_BUILD_TESTS=ON`` (Set to OFF to not build the unit tests)
 - ``-DOCIO_BUILD_GPU_TESTS=ON`` (Set to OFF to not build the GPU unit tests)
 - ``-DOCIO_USE_HEADLESS=OFF`` (Set to ON to do headless GPU reendering)
 - ``-DOCIO_WARNING_AS_ERROR=ON`` (Set to OFF to turn off warnings as errors)
 - ``-DOCIO_BUILD_DOCS=OFF`` (Set to ON to build the documentation)
 - ``-DOCIO_BUILD_FROZEN_DOCS=OFF`` (Set to ON to update the Python documentation)
 
+Note that OCIO will turn off any specific SIMD CPU performance optimizations if they are not supported 
+by the build target architecture.
+
+On the MacOS, the default is to build universal binaries 
+(natively supporting both the Intel and ARM processors). The ``-DCMAKE_OSX_ARCHITECTURES`` option 
+may be set to just arm64 or x86_64 to override the default value, which is ``arm64;x86_64``.
+
+When doing a universal build, note that the OCIO dependencies must be built as universal libraries 
+too. If you are running in OCIO_INSTALL_EXT_PACKAGES=MISSING or NONE mode, your build will fail if 
+any of your installed libraries are not universal. The easiest way to address this is to set 
+OCIO_INSTALL_EXT_PACKAGES=ALL in order to let OCIO build everything. Alternatively, you may set 
+CMAKE_OSX_ARCHITECTURES to just the platform you are targeting.
+
 Several command-line tools (such as ``ocioconvert``) require reading or writing image files.
 If ``OCIO_USE_OIIO_FOR_APPS=OFF``, these will be built using OpenEXR rather than OpenImageIO
 and therefore you will be limited to using OpenEXR files with these tools rather than the

share/cmake/macros/CheckForOpenEXRCompatibility.cmake

@@ -0,0 +1,66 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright Contributors to the OpenColorIO Project.
+#
+# Check for compatibility between OpenEXR and OpenImageIO since OCIO requires OpenEXR 3+.
+#
+
+message(STATUS "Checking if the OpenImageIO found is built with OpenEXR 3+...")
+
+find_path (OpenImageIO_INCLUDE_DIR
+    NAMES
+        OpenImageIO/imageio.h
+    HINTS
+        ${OpenImageIO_ROOT}
+        # Assuming that OpenImageIO was installed normally, go back a few folders down
+        # to get the equivalent of OpenImageIO_ROOT.
+        ${OpenImageIO_DIR}/../../..
+    PATH_SUFFIXES
+        OpenImageIO/include
+        include
+)
+
+if (NOT OpenImageIO_INCLUDE_DIR)
+    message(STATUS "${ColorWarning}Could not find OpenImageIO header to evaluate the OpenEXR version.")
+    message(STATUS "Please provide the OpenImageIO_DIR variable.")
+    message(STATUS "If your OpenImageIO's files are located in different root directory, \
+please provide the OpenImageIO_ROOT where the include files are located.${ColorReset}")
+endif()
+
+# Try to figure out version number
+set (OIIO_VERSION_HEADER "${OpenImageIO_INCLUDE_DIR}/OpenImageIO/oiioversion.h")
+if (EXISTS "${OIIO_VERSION_HEADER}")
+    file (STRINGS "${OIIO_VERSION_HEADER}" TMP REGEX "^#define OIIO_VERSION_MAJOR .*$")
+    string (REGEX MATCHALL "[0-9]+" OpenImageIO_VERSION_MAJOR ${TMP})
+    file (STRINGS "${OIIO_VERSION_HEADER}" TMP REGEX "^#define OIIO_VERSION_MINOR .*$")
+    string (REGEX MATCHALL "[0-9]+" OpenImageIO_VERSION_MINOR ${TMP})
+    file (STRINGS "${OIIO_VERSION_HEADER}" TMP REGEX "^#define OIIO_VERSION_PATCH .*$")
+    string (REGEX MATCHALL "[0-9]+" OpenImageIO_VERSION_PATCH ${TMP})
+    file (STRINGS "${OIIO_VERSION_HEADER}" TMP REGEX "^#define OIIO_VERSION_TWEAK .*$")
+    if (TMP)
+        string (REGEX MATCHALL "[0-9]+" OpenImageIO_VERSION_TWEAK ${TMP})
+    else ()
+        set (OpenImageIO_VERSION_TWEAK 0)
+    endif ()
+    set (OpenImageIO_VERSION "${OpenImageIO_VERSION_MAJOR}.${OpenImageIO_VERSION_MINOR}.${OpenImageIO_VERSION_PATCH}.${OpenImageIO_VERSION_TWEAK}")
+endif ()
+
+set (OIIO_IMATH_HEADER "${OpenImageIO_INCLUDE_DIR}/OpenImageIO/Imath.h")
+if (EXISTS "${OIIO_IMATH_HEADER}")
+    file(STRINGS "${OIIO_IMATH_HEADER}" TMP REGEX "^#define OIIO_USING_IMATH .*$")
+    string(REGEX MATCHALL "[0-9]" OIIO_IMATH_VERSION ${TMP})
+    if (OIIO_IMATH_VERSION LESS 3)
+        message(STATUS "Skipping OpenImageIO built against OpenEXR 2, please use version 3 or greater.")
+    else()
+        set(is_OpenEXR_VERSION_valid TRUE)
+    endif()
+endif()
+
+# clean up variables
+unset(OpenImageIO_INCLUDE_DIR)
+unset(OIIO_VERSION_HEADER)
+unset(OIIO_VERSION_MAJOR)
+unset(OIIO_VERSION_MINOR)
+unset(OIIO_VERSION_PATCH)
+unset(OIIO_VERSION_TWEAK)
+unset(OIIO_IMATH_HEADER)
+unset(OIIO_IMATH_VERSION)

share/cmake/modules/FindExtPackages.cmake

@@ -197,13 +197,10 @@ if((OCIO_BUILD_APPS AND OCIO_USE_OIIO_FOR_APPS) OR OCIO_BUILD_TESTS)
     # Supported from OIIO 2.4+. Setting this for lower versions doesn't affect anything.
     set(OPENIMAGEIO_CONFIG_DO_NOT_FIND_IMATH 1)
 
-    include(ocio_check_dependency_version)
-    # Since OpenImageIO will try to find OpenEXR through its OpenImageIOConfig.cmake file,
-    # let's try to find OpenEXR first and if the version is too old, OCIO will not try to find
-    # OpenImageIO.
-    ocio_check_dependency_version(  OpenEXR "is_OpenEXR_VERSION_valid" 
-                                    MIN_VERSION ${OpenEXR_MININUM_VERSION} 
-                                    CONFIG)
+    set(is_OpenEXR_VERSION_valid FALSE)
+    # Check for compatibility between OpenEXR and OpenImageIO.
+    # Will set is_OpenEXR_VERSION_valid to TRUE if valid.
+    include(CheckForOpenEXRCompatibility)
 
     # Do not try to find OpenImageIO if the version of OpenEXR is too old.                                    
     if (is_OpenEXR_VERSION_valid)
@@ -227,8 +224,6 @@ if((OCIO_BUILD_APPS AND OCIO_USE_OIIO_FOR_APPS) OR OCIO_BUILD_TESTS)
                                  MIN_VERSION ${OIIO_VERSION}
                                  RECOMMENDED_VERSION ${OIIO_RECOMMENDED_VERSION}
                                  PROMOTE_TARGET OpenImageIO::OpenImageIO)
-    else()
-        message(WARNING "Skipping OpenImageIO because the OpenEXR found by OpenImageIO is too old (under ${OpenEXR_MININUM_VERSION})")
     endif()
 endif()
 

share/cmake/modules/install/InstallOpenEXR.cmake

@@ -201,6 +201,7 @@ if(_OpenEXR_TARGET_CREATE)
         IMPORTED_LOCATION ${IlmThread_LIBRARY}
         INTERFACE_INCLUDE_DIRECTORIES "${OpenEXR_INCLUDE_DIR}"
         INTERFACE_LINK_LIBRARIES "OpenEXR::IlmThreadConfig;OpenEXR::IlmThreadConfig;OpenEXR::Iex;Threads::Threads"
+        STATIC_LIBRARY_OPTIONS "-no_warning_for_no_symbols"
     )
     set_target_properties(OpenEXR::IlmThreadConfig PROPERTIES
         INTERFACE_INCLUDE_DIRECTORIES "${OpenEXR_INCLUDE_DIR};${OpenEXR_INCLUDE_DIR}/OpenEXR"
@@ -217,6 +218,7 @@ if(_OpenEXR_TARGET_CREATE)
         IMPORTED_LOCATION ${OpenEXRCore_LIBRARY}
         INTERFACE_INCLUDE_DIRECTORIES "${OpenEXR_INCLUDE_DIR}"
         INTERFACE_LINK_LIBRARIES "OpenEXR::IlmThreadConfig;ZLIB::ZLIB;\$<LINK_ONLY:Imath::ImathConfig>"
+        STATIC_LIBRARY_OPTIONS "-no_warning_for_no_symbols"
     )
     set_target_properties(OpenEXR::OpenEXRUtil PROPERTIES
         IMPORTED_LOCATION ${OpenEXRUtil_LIBRARY}